• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_dir2.h"
31 #include "xfs_inode.h"
32 #include "xfs_btree.h"
33 #include "xfs_trans.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_extfree_item.h"
36 #include "xfs_alloc.h"
37 #include "xfs_bmap.h"
38 #include "xfs_bmap_util.h"
39 #include "xfs_bmap_btree.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_trans_space.h"
44 #include "xfs_buf_item.h"
45 #include "xfs_trace.h"
46 #include "xfs_symlink.h"
47 #include "xfs_attr_leaf.h"
48 #include "xfs_filestream.h"
49 #include "xfs_rmap.h"
50 #include "xfs_ag_resv.h"
51 #include "xfs_refcount.h"
52 #include "xfs_rmap_btree.h"
53 #include "xfs_icache.h"
54 
55 
56 kmem_zone_t		*xfs_bmap_free_item_zone;
57 
58 /*
59  * Miscellaneous helper functions
60  */
61 
62 /*
63  * Compute and fill in the value of the maximum depth of a bmap btree
64  * in this filesystem.  Done once, during mount.
65  */
66 void
xfs_bmap_compute_maxlevels(xfs_mount_t * mp,int whichfork)67 xfs_bmap_compute_maxlevels(
68 	xfs_mount_t	*mp,		/* file system mount structure */
69 	int		whichfork)	/* data or attr fork */
70 {
71 	int		level;		/* btree level */
72 	uint		maxblocks;	/* max blocks at this level */
73 	uint		maxleafents;	/* max leaf entries possible */
74 	int		maxrootrecs;	/* max records in root block */
75 	int		minleafrecs;	/* min records in leaf block */
76 	int		minnoderecs;	/* min records in node block */
77 	int		sz;		/* root block size */
78 
79 	/*
80 	 * The maximum number of extents in a file, hence the maximum
81 	 * number of leaf entries, is controlled by the type of di_nextents
82 	 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
83 	 * (a signed 16-bit number, xfs_aextnum_t).
84 	 *
85 	 * Note that we can no longer assume that if we are in ATTR1 that
86 	 * the fork offset of all the inodes will be
87 	 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
88 	 * with ATTR2 and then mounted back with ATTR1, keeping the
89 	 * di_forkoff's fixed but probably at various positions. Therefore,
90 	 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
91 	 * of a minimum size available.
92 	 */
93 	if (whichfork == XFS_DATA_FORK) {
94 		maxleafents = MAXEXTNUM;
95 		sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
96 	} else {
97 		maxleafents = MAXAEXTNUM;
98 		sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
99 	}
100 	maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
101 	minleafrecs = mp->m_bmap_dmnr[0];
102 	minnoderecs = mp->m_bmap_dmnr[1];
103 	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
104 	for (level = 1; maxblocks > 1; level++) {
105 		if (maxblocks <= maxrootrecs)
106 			maxblocks = 1;
107 		else
108 			maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
109 	}
110 	mp->m_bm_maxlevels[whichfork] = level;
111 }
112 
113 STATIC int				/* error */
xfs_bmbt_lookup_eq(struct xfs_btree_cur * cur,xfs_fileoff_t off,xfs_fsblock_t bno,xfs_filblks_t len,int * stat)114 xfs_bmbt_lookup_eq(
115 	struct xfs_btree_cur	*cur,
116 	xfs_fileoff_t		off,
117 	xfs_fsblock_t		bno,
118 	xfs_filblks_t		len,
119 	int			*stat)	/* success/failure */
120 {
121 	cur->bc_rec.b.br_startoff = off;
122 	cur->bc_rec.b.br_startblock = bno;
123 	cur->bc_rec.b.br_blockcount = len;
124 	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
125 }
126 
127 STATIC int				/* error */
xfs_bmbt_lookup_ge(struct xfs_btree_cur * cur,xfs_fileoff_t off,xfs_fsblock_t bno,xfs_filblks_t len,int * stat)128 xfs_bmbt_lookup_ge(
129 	struct xfs_btree_cur	*cur,
130 	xfs_fileoff_t		off,
131 	xfs_fsblock_t		bno,
132 	xfs_filblks_t		len,
133 	int			*stat)	/* success/failure */
134 {
135 	cur->bc_rec.b.br_startoff = off;
136 	cur->bc_rec.b.br_startblock = bno;
137 	cur->bc_rec.b.br_blockcount = len;
138 	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
139 }
140 
141 /*
142  * Check if the inode needs to be converted to btree format.
143  */
xfs_bmap_needs_btree(struct xfs_inode * ip,int whichfork)144 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
145 {
146 	return whichfork != XFS_COW_FORK &&
147 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
148 		XFS_IFORK_NEXTENTS(ip, whichfork) >
149 			XFS_IFORK_MAXEXT(ip, whichfork);
150 }
151 
152 /*
153  * Check if the inode should be converted to extent format.
154  */
xfs_bmap_wants_extents(struct xfs_inode * ip,int whichfork)155 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
156 {
157 	return whichfork != XFS_COW_FORK &&
158 		XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
159 		XFS_IFORK_NEXTENTS(ip, whichfork) <=
160 			XFS_IFORK_MAXEXT(ip, whichfork);
161 }
162 
163 /*
164  * Update the record referred to by cur to the value given
165  * by [off, bno, len, state].
166  * This either works (return 0) or gets an EFSCORRUPTED error.
167  */
168 STATIC int
xfs_bmbt_update(struct xfs_btree_cur * cur,xfs_fileoff_t off,xfs_fsblock_t bno,xfs_filblks_t len,xfs_exntst_t state)169 xfs_bmbt_update(
170 	struct xfs_btree_cur	*cur,
171 	xfs_fileoff_t		off,
172 	xfs_fsblock_t		bno,
173 	xfs_filblks_t		len,
174 	xfs_exntst_t		state)
175 {
176 	union xfs_btree_rec	rec;
177 
178 	xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
179 	return xfs_btree_update(cur, &rec);
180 }
181 
182 /*
183  * Compute the worst-case number of indirect blocks that will be used
184  * for ip's delayed extent of length "len".
185  */
186 STATIC xfs_filblks_t
xfs_bmap_worst_indlen(xfs_inode_t * ip,xfs_filblks_t len)187 xfs_bmap_worst_indlen(
188 	xfs_inode_t	*ip,		/* incore inode pointer */
189 	xfs_filblks_t	len)		/* delayed extent length */
190 {
191 	int		level;		/* btree level number */
192 	int		maxrecs;	/* maximum record count at this level */
193 	xfs_mount_t	*mp;		/* mount structure */
194 	xfs_filblks_t	rval;		/* return value */
195 	xfs_filblks_t   orig_len;
196 
197 	mp = ip->i_mount;
198 
199 	/* Calculate the worst-case size of the bmbt. */
200 	orig_len = len;
201 	maxrecs = mp->m_bmap_dmxr[0];
202 	for (level = 0, rval = 0;
203 	     level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
204 	     level++) {
205 		len += maxrecs - 1;
206 		do_div(len, maxrecs);
207 		rval += len;
208 		if (len == 1) {
209 			rval += XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
210 				level - 1;
211 			break;
212 		}
213 		if (level == 0)
214 			maxrecs = mp->m_bmap_dmxr[1];
215 	}
216 
217 	/* Calculate the worst-case size of the rmapbt. */
218 	if (xfs_sb_version_hasrmapbt(&mp->m_sb))
219 		rval += 1 + xfs_rmapbt_calc_size(mp, orig_len) +
220 				mp->m_rmap_maxlevels;
221 
222 	return rval;
223 }
224 
225 /*
226  * Calculate the default attribute fork offset for newly created inodes.
227  */
228 uint
xfs_default_attroffset(struct xfs_inode * ip)229 xfs_default_attroffset(
230 	struct xfs_inode	*ip)
231 {
232 	struct xfs_mount	*mp = ip->i_mount;
233 	uint			offset;
234 
235 	if (mp->m_sb.sb_inodesize == 256) {
236 		offset = XFS_LITINO(mp, ip->i_d.di_version) -
237 				XFS_BMDR_SPACE_CALC(MINABTPTRS);
238 	} else {
239 		offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
240 	}
241 
242 	ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
243 	return offset;
244 }
245 
246 /*
247  * Helper routine to reset inode di_forkoff field when switching
248  * attribute fork from local to extent format - we reset it where
249  * possible to make space available for inline data fork extents.
250  */
251 STATIC void
xfs_bmap_forkoff_reset(xfs_inode_t * ip,int whichfork)252 xfs_bmap_forkoff_reset(
253 	xfs_inode_t	*ip,
254 	int		whichfork)
255 {
256 	if (whichfork == XFS_ATTR_FORK &&
257 	    ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
258 	    ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
259 	    ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
260 		uint	dfl_forkoff = xfs_default_attroffset(ip) >> 3;
261 
262 		if (dfl_forkoff > ip->i_d.di_forkoff)
263 			ip->i_d.di_forkoff = dfl_forkoff;
264 	}
265 }
266 
267 #ifdef DEBUG
268 STATIC struct xfs_buf *
xfs_bmap_get_bp(struct xfs_btree_cur * cur,xfs_fsblock_t bno)269 xfs_bmap_get_bp(
270 	struct xfs_btree_cur	*cur,
271 	xfs_fsblock_t		bno)
272 {
273 	struct xfs_log_item_desc *lidp;
274 	int			i;
275 
276 	if (!cur)
277 		return NULL;
278 
279 	for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
280 		if (!cur->bc_bufs[i])
281 			break;
282 		if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
283 			return cur->bc_bufs[i];
284 	}
285 
286 	/* Chase down all the log items to see if the bp is there */
287 	list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
288 		struct xfs_buf_log_item	*bip;
289 		bip = (struct xfs_buf_log_item *)lidp->lid_item;
290 		if (bip->bli_item.li_type == XFS_LI_BUF &&
291 		    XFS_BUF_ADDR(bip->bli_buf) == bno)
292 			return bip->bli_buf;
293 	}
294 
295 	return NULL;
296 }
297 
298 STATIC void
xfs_check_block(struct xfs_btree_block * block,xfs_mount_t * mp,int root,short sz)299 xfs_check_block(
300 	struct xfs_btree_block	*block,
301 	xfs_mount_t		*mp,
302 	int			root,
303 	short			sz)
304 {
305 	int			i, j, dmxr;
306 	__be64			*pp, *thispa;	/* pointer to block address */
307 	xfs_bmbt_key_t		*prevp, *keyp;
308 
309 	ASSERT(be16_to_cpu(block->bb_level) > 0);
310 
311 	prevp = NULL;
312 	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
313 		dmxr = mp->m_bmap_dmxr[0];
314 		keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
315 
316 		if (prevp) {
317 			ASSERT(be64_to_cpu(prevp->br_startoff) <
318 			       be64_to_cpu(keyp->br_startoff));
319 		}
320 		prevp = keyp;
321 
322 		/*
323 		 * Compare the block numbers to see if there are dups.
324 		 */
325 		if (root)
326 			pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
327 		else
328 			pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
329 
330 		for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
331 			if (root)
332 				thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
333 			else
334 				thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
335 			if (*thispa == *pp) {
336 				xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
337 					__func__, j, i,
338 					(unsigned long long)be64_to_cpu(*thispa));
339 				panic("%s: ptrs are equal in node\n",
340 					__func__);
341 			}
342 		}
343 	}
344 }
345 
346 /*
347  * Check that the extents for the inode ip are in the right order in all
348  * btree leaves. THis becomes prohibitively expensive for large extent count
349  * files, so don't bother with inodes that have more than 10,000 extents in
350  * them. The btree record ordering checks will still be done, so for such large
351  * bmapbt constructs that is going to catch most corruptions.
352  */
353 STATIC void
xfs_bmap_check_leaf_extents(xfs_btree_cur_t * cur,xfs_inode_t * ip,int whichfork)354 xfs_bmap_check_leaf_extents(
355 	xfs_btree_cur_t		*cur,	/* btree cursor or null */
356 	xfs_inode_t		*ip,		/* incore inode pointer */
357 	int			whichfork)	/* data or attr fork */
358 {
359 	struct xfs_btree_block	*block;	/* current btree block */
360 	xfs_fsblock_t		bno;	/* block # of "block" */
361 	xfs_buf_t		*bp;	/* buffer for "block" */
362 	int			error;	/* error return value */
363 	xfs_extnum_t		i=0, j;	/* index into the extents list */
364 	xfs_ifork_t		*ifp;	/* fork structure */
365 	int			level;	/* btree level, for checking */
366 	xfs_mount_t		*mp;	/* file system mount structure */
367 	__be64			*pp;	/* pointer to block address */
368 	xfs_bmbt_rec_t		*ep;	/* pointer to current extent */
369 	xfs_bmbt_rec_t		last = {0, 0}; /* last extent in prev block */
370 	xfs_bmbt_rec_t		*nextp;	/* pointer to next extent */
371 	int			bp_release = 0;
372 
373 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
374 		return;
375 	}
376 
377 	/* skip large extent count inodes */
378 	if (ip->i_d.di_nextents > 10000)
379 		return;
380 
381 	bno = NULLFSBLOCK;
382 	mp = ip->i_mount;
383 	ifp = XFS_IFORK_PTR(ip, whichfork);
384 	block = ifp->if_broot;
385 	/*
386 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
387 	 */
388 	level = be16_to_cpu(block->bb_level);
389 	ASSERT(level > 0);
390 	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
391 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
392 	bno = be64_to_cpu(*pp);
393 
394 	ASSERT(bno != NULLFSBLOCK);
395 	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
396 	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
397 
398 	/*
399 	 * Go down the tree until leaf level is reached, following the first
400 	 * pointer (leftmost) at each level.
401 	 */
402 	while (level-- > 0) {
403 		/* See if buf is in cur first */
404 		bp_release = 0;
405 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
406 		if (!bp) {
407 			bp_release = 1;
408 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
409 						XFS_BMAP_BTREE_REF,
410 						&xfs_bmbt_buf_ops);
411 			if (error)
412 				goto error_norelse;
413 		}
414 		block = XFS_BUF_TO_BLOCK(bp);
415 		if (level == 0)
416 			break;
417 
418 		/*
419 		 * Check this block for basic sanity (increasing keys and
420 		 * no duplicate blocks).
421 		 */
422 
423 		xfs_check_block(block, mp, 0, 0);
424 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
425 		bno = be64_to_cpu(*pp);
426 		XFS_WANT_CORRUPTED_GOTO(mp,
427 					XFS_FSB_SANITY_CHECK(mp, bno), error0);
428 		if (bp_release) {
429 			bp_release = 0;
430 			xfs_trans_brelse(NULL, bp);
431 		}
432 	}
433 
434 	/*
435 	 * Here with bp and block set to the leftmost leaf node in the tree.
436 	 */
437 	i = 0;
438 
439 	/*
440 	 * Loop over all leaf nodes checking that all extents are in the right order.
441 	 */
442 	for (;;) {
443 		xfs_fsblock_t	nextbno;
444 		xfs_extnum_t	num_recs;
445 
446 
447 		num_recs = xfs_btree_get_numrecs(block);
448 
449 		/*
450 		 * Read-ahead the next leaf block, if any.
451 		 */
452 
453 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
454 
455 		/*
456 		 * Check all the extents to make sure they are OK.
457 		 * If we had a previous block, the last entry should
458 		 * conform with the first entry in this one.
459 		 */
460 
461 		ep = XFS_BMBT_REC_ADDR(mp, block, 1);
462 		if (i) {
463 			ASSERT(xfs_bmbt_disk_get_startoff(&last) +
464 			       xfs_bmbt_disk_get_blockcount(&last) <=
465 			       xfs_bmbt_disk_get_startoff(ep));
466 		}
467 		for (j = 1; j < num_recs; j++) {
468 			nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
469 			ASSERT(xfs_bmbt_disk_get_startoff(ep) +
470 			       xfs_bmbt_disk_get_blockcount(ep) <=
471 			       xfs_bmbt_disk_get_startoff(nextp));
472 			ep = nextp;
473 		}
474 
475 		last = *ep;
476 		i += num_recs;
477 		if (bp_release) {
478 			bp_release = 0;
479 			xfs_trans_brelse(NULL, bp);
480 		}
481 		bno = nextbno;
482 		/*
483 		 * If we've reached the end, stop.
484 		 */
485 		if (bno == NULLFSBLOCK)
486 			break;
487 
488 		bp_release = 0;
489 		bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
490 		if (!bp) {
491 			bp_release = 1;
492 			error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
493 						XFS_BMAP_BTREE_REF,
494 						&xfs_bmbt_buf_ops);
495 			if (error)
496 				goto error_norelse;
497 		}
498 		block = XFS_BUF_TO_BLOCK(bp);
499 	}
500 
501 	return;
502 
503 error0:
504 	xfs_warn(mp, "%s: at error0", __func__);
505 	if (bp_release)
506 		xfs_trans_brelse(NULL, bp);
507 error_norelse:
508 	xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
509 		__func__, i);
510 	panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
511 	return;
512 }
513 
514 /*
515  * Add bmap trace insert entries for all the contents of the extent records.
516  */
517 void
xfs_bmap_trace_exlist(xfs_inode_t * ip,xfs_extnum_t cnt,int whichfork,unsigned long caller_ip)518 xfs_bmap_trace_exlist(
519 	xfs_inode_t	*ip,		/* incore inode pointer */
520 	xfs_extnum_t	cnt,		/* count of entries in the list */
521 	int		whichfork,	/* data or attr or cow fork */
522 	unsigned long	caller_ip)
523 {
524 	xfs_extnum_t	idx;		/* extent record index */
525 	xfs_ifork_t	*ifp;		/* inode fork pointer */
526 	int		state = 0;
527 
528 	if (whichfork == XFS_ATTR_FORK)
529 		state |= BMAP_ATTRFORK;
530 	else if (whichfork == XFS_COW_FORK)
531 		state |= BMAP_COWFORK;
532 
533 	ifp = XFS_IFORK_PTR(ip, whichfork);
534 	ASSERT(cnt == xfs_iext_count(ifp));
535 	for (idx = 0; idx < cnt; idx++)
536 		trace_xfs_extlist(ip, idx, state, caller_ip);
537 }
538 
539 /*
540  * Validate that the bmbt_irecs being returned from bmapi are valid
541  * given the caller's original parameters.  Specifically check the
542  * ranges of the returned irecs to ensure that they only extend beyond
543  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
544  */
545 STATIC void
xfs_bmap_validate_ret(xfs_fileoff_t bno,xfs_filblks_t len,int flags,xfs_bmbt_irec_t * mval,int nmap,int ret_nmap)546 xfs_bmap_validate_ret(
547 	xfs_fileoff_t		bno,
548 	xfs_filblks_t		len,
549 	int			flags,
550 	xfs_bmbt_irec_t		*mval,
551 	int			nmap,
552 	int			ret_nmap)
553 {
554 	int			i;		/* index to map values */
555 
556 	ASSERT(ret_nmap <= nmap);
557 
558 	for (i = 0; i < ret_nmap; i++) {
559 		ASSERT(mval[i].br_blockcount > 0);
560 		if (!(flags & XFS_BMAPI_ENTIRE)) {
561 			ASSERT(mval[i].br_startoff >= bno);
562 			ASSERT(mval[i].br_blockcount <= len);
563 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
564 			       bno + len);
565 		} else {
566 			ASSERT(mval[i].br_startoff < bno + len);
567 			ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
568 			       bno);
569 		}
570 		ASSERT(i == 0 ||
571 		       mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
572 		       mval[i].br_startoff);
573 		ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
574 		       mval[i].br_startblock != HOLESTARTBLOCK);
575 		ASSERT(mval[i].br_state == XFS_EXT_NORM ||
576 		       mval[i].br_state == XFS_EXT_UNWRITTEN);
577 	}
578 }
579 
580 #else
581 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)		do { } while (0)
582 #define	xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)	do { } while (0)
583 #endif /* DEBUG */
584 
585 /*
586  * bmap free list manipulation functions
587  */
588 
589 /*
590  * Add the extent to the list of extents to be free at transaction end.
591  * The list is maintained sorted (by block number).
592  */
593 void
xfs_bmap_add_free(struct xfs_mount * mp,struct xfs_defer_ops * dfops,xfs_fsblock_t bno,xfs_filblks_t len,struct xfs_owner_info * oinfo)594 xfs_bmap_add_free(
595 	struct xfs_mount		*mp,
596 	struct xfs_defer_ops		*dfops,
597 	xfs_fsblock_t			bno,
598 	xfs_filblks_t			len,
599 	struct xfs_owner_info		*oinfo)
600 {
601 	struct xfs_extent_free_item	*new;		/* new element */
602 #ifdef DEBUG
603 	xfs_agnumber_t		agno;
604 	xfs_agblock_t		agbno;
605 
606 	ASSERT(bno != NULLFSBLOCK);
607 	ASSERT(len > 0);
608 	ASSERT(len <= MAXEXTLEN);
609 	ASSERT(!isnullstartblock(bno));
610 	agno = XFS_FSB_TO_AGNO(mp, bno);
611 	agbno = XFS_FSB_TO_AGBNO(mp, bno);
612 	ASSERT(agno < mp->m_sb.sb_agcount);
613 	ASSERT(agbno < mp->m_sb.sb_agblocks);
614 	ASSERT(len < mp->m_sb.sb_agblocks);
615 	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
616 #endif
617 	ASSERT(xfs_bmap_free_item_zone != NULL);
618 
619 	new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
620 	new->xefi_startblock = bno;
621 	new->xefi_blockcount = (xfs_extlen_t)len;
622 	if (oinfo)
623 		new->xefi_oinfo = *oinfo;
624 	else
625 		xfs_rmap_skip_owner_update(&new->xefi_oinfo);
626 	trace_xfs_bmap_free_defer(mp, XFS_FSB_TO_AGNO(mp, bno), 0,
627 			XFS_FSB_TO_AGBNO(mp, bno), len);
628 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_FREE, &new->xefi_list);
629 }
630 
631 /*
632  * Inode fork format manipulation functions
633  */
634 
635 /*
636  * Transform a btree format file with only one leaf node, where the
637  * extents list will fit in the inode, into an extents format file.
638  * Since the file extents are already in-core, all we have to do is
639  * give up the space for the btree root and pitch the leaf block.
640  */
641 STATIC int				/* error */
xfs_bmap_btree_to_extents(xfs_trans_t * tp,xfs_inode_t * ip,xfs_btree_cur_t * cur,int * logflagsp,int whichfork)642 xfs_bmap_btree_to_extents(
643 	xfs_trans_t		*tp,	/* transaction pointer */
644 	xfs_inode_t		*ip,	/* incore inode pointer */
645 	xfs_btree_cur_t		*cur,	/* btree cursor */
646 	int			*logflagsp, /* inode logging flags */
647 	int			whichfork)  /* data or attr fork */
648 {
649 	/* REFERENCED */
650 	struct xfs_btree_block	*cblock;/* child btree block */
651 	xfs_fsblock_t		cbno;	/* child block number */
652 	xfs_buf_t		*cbp;	/* child block's buffer */
653 	int			error;	/* error return value */
654 	xfs_ifork_t		*ifp;	/* inode fork data */
655 	xfs_mount_t		*mp;	/* mount point structure */
656 	__be64			*pp;	/* ptr to block address */
657 	struct xfs_btree_block	*rblock;/* root btree block */
658 	struct xfs_owner_info	oinfo;
659 
660 	mp = ip->i_mount;
661 	ifp = XFS_IFORK_PTR(ip, whichfork);
662 	ASSERT(whichfork != XFS_COW_FORK);
663 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
664 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
665 	rblock = ifp->if_broot;
666 	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
667 	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
668 	ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
669 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
670 	cbno = be64_to_cpu(*pp);
671 	*logflagsp = 0;
672 #ifdef DEBUG
673 	if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
674 		return error;
675 #endif
676 	error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
677 				&xfs_bmbt_buf_ops);
678 	if (error)
679 		return error;
680 	cblock = XFS_BUF_TO_BLOCK(cbp);
681 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
682 		return error;
683 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
684 	xfs_bmap_add_free(mp, cur->bc_private.b.dfops, cbno, 1, &oinfo);
685 	ip->i_d.di_nblocks--;
686 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
687 	xfs_trans_binval(tp, cbp);
688 	if (cur->bc_bufs[0] == cbp)
689 		cur->bc_bufs[0] = NULL;
690 	xfs_iroot_realloc(ip, -1, whichfork);
691 	ASSERT(ifp->if_broot == NULL);
692 	ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
693 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
694 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
695 	return 0;
696 }
697 
698 /*
699  * Convert an extents-format file into a btree-format file.
700  * The new file will have a root block (in the inode) and a single child block.
701  */
702 STATIC int					/* error */
xfs_bmap_extents_to_btree(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,xfs_btree_cur_t ** curp,int wasdel,int * logflagsp,int whichfork)703 xfs_bmap_extents_to_btree(
704 	xfs_trans_t		*tp,		/* transaction pointer */
705 	xfs_inode_t		*ip,		/* incore inode pointer */
706 	xfs_fsblock_t		*firstblock,	/* first-block-allocated */
707 	struct xfs_defer_ops	*dfops,		/* blocks freed in xaction */
708 	xfs_btree_cur_t		**curp,		/* cursor returned to caller */
709 	int			wasdel,		/* converting a delayed alloc */
710 	int			*logflagsp,	/* inode logging flags */
711 	int			whichfork)	/* data or attr fork */
712 {
713 	struct xfs_btree_block	*ablock;	/* allocated (child) bt block */
714 	xfs_buf_t		*abp;		/* buffer for ablock */
715 	xfs_alloc_arg_t		args;		/* allocation arguments */
716 	xfs_bmbt_rec_t		*arp;		/* child record pointer */
717 	struct xfs_btree_block	*block;		/* btree root block */
718 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
719 	xfs_bmbt_rec_host_t	*ep;		/* extent record pointer */
720 	int			error;		/* error return value */
721 	xfs_extnum_t		i, cnt;		/* extent record index */
722 	xfs_ifork_t		*ifp;		/* inode fork pointer */
723 	xfs_bmbt_key_t		*kp;		/* root block key pointer */
724 	xfs_mount_t		*mp;		/* mount structure */
725 	xfs_extnum_t		nextents;	/* number of file extents */
726 	xfs_bmbt_ptr_t		*pp;		/* root block address pointer */
727 
728 	mp = ip->i_mount;
729 	ASSERT(whichfork != XFS_COW_FORK);
730 	ifp = XFS_IFORK_PTR(ip, whichfork);
731 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
732 
733 	/*
734 	 * Make space in the inode incore.
735 	 */
736 	xfs_iroot_realloc(ip, 1, whichfork);
737 	ifp->if_flags |= XFS_IFBROOT;
738 
739 	/*
740 	 * Fill in the root.
741 	 */
742 	block = ifp->if_broot;
743 	if (xfs_sb_version_hascrc(&mp->m_sb))
744 		xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
745 				 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
746 				 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
747 	else
748 		xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
749 				 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
750 				 XFS_BTREE_LONG_PTRS);
751 
752 	/*
753 	 * Need a cursor.  Can't allocate until bb_level is filled in.
754 	 */
755 	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
756 	cur->bc_private.b.firstblock = *firstblock;
757 	cur->bc_private.b.dfops = dfops;
758 	cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
759 	/*
760 	 * Convert to a btree with two levels, one record in root.
761 	 */
762 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
763 	memset(&args, 0, sizeof(args));
764 	args.tp = tp;
765 	args.mp = mp;
766 	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
767 	args.firstblock = *firstblock;
768 	if (*firstblock == NULLFSBLOCK) {
769 		args.type = XFS_ALLOCTYPE_START_BNO;
770 		args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
771 	} else if (dfops->dop_low) {
772 		args.type = XFS_ALLOCTYPE_START_BNO;
773 try_another_ag:
774 		args.fsbno = *firstblock;
775 	} else {
776 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
777 		args.fsbno = *firstblock;
778 	}
779 	args.minlen = args.maxlen = args.prod = 1;
780 	args.wasdel = wasdel;
781 	*logflagsp = 0;
782 	if ((error = xfs_alloc_vextent(&args))) {
783 		xfs_iroot_realloc(ip, -1, whichfork);
784 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
785 		return error;
786 	}
787 
788 	/*
789 	 * During a CoW operation, the allocation and bmbt updates occur in
790 	 * different transactions.  The mapping code tries to put new bmbt
791 	 * blocks near extents being mapped, but the only way to guarantee this
792 	 * is if the alloc and the mapping happen in a single transaction that
793 	 * has a block reservation.  That isn't the case here, so if we run out
794 	 * of space we'll try again with another AG.
795 	 */
796 	if (xfs_sb_version_hasreflink(&cur->bc_mp->m_sb) &&
797 	    args.fsbno == NULLFSBLOCK &&
798 	    args.type == XFS_ALLOCTYPE_NEAR_BNO) {
799 		args.type = XFS_ALLOCTYPE_FIRST_AG;
800 		goto try_another_ag;
801 	}
802 	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
803 		xfs_iroot_realloc(ip, -1, whichfork);
804 		xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
805 		return -ENOSPC;
806 	}
807 	/*
808 	 * Allocation can't fail, the space was reserved.
809 	 */
810 	ASSERT(*firstblock == NULLFSBLOCK ||
811 	       args.agno >= XFS_FSB_TO_AGNO(mp, *firstblock));
812 	*firstblock = cur->bc_private.b.firstblock = args.fsbno;
813 	cur->bc_private.b.allocated++;
814 	ip->i_d.di_nblocks++;
815 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
816 	abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
817 	/*
818 	 * Fill in the child block.
819 	 */
820 	abp->b_ops = &xfs_bmbt_buf_ops;
821 	ablock = XFS_BUF_TO_BLOCK(abp);
822 	if (xfs_sb_version_hascrc(&mp->m_sb))
823 		xfs_btree_init_block_int(mp, ablock, abp->b_bn,
824 				XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
825 				XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
826 	else
827 		xfs_btree_init_block_int(mp, ablock, abp->b_bn,
828 				XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
829 				XFS_BTREE_LONG_PTRS);
830 
831 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
832 	nextents =  xfs_iext_count(ifp);
833 	for (cnt = i = 0; i < nextents; i++) {
834 		ep = xfs_iext_get_ext(ifp, i);
835 		if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
836 			arp->l0 = cpu_to_be64(ep->l0);
837 			arp->l1 = cpu_to_be64(ep->l1);
838 			arp++; cnt++;
839 		}
840 	}
841 	ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
842 	xfs_btree_set_numrecs(ablock, cnt);
843 
844 	/*
845 	 * Fill in the root key and pointer.
846 	 */
847 	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
848 	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
849 	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
850 	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
851 						be16_to_cpu(block->bb_level)));
852 	*pp = cpu_to_be64(args.fsbno);
853 
854 	/*
855 	 * Do all this logging at the end so that
856 	 * the root is at the right level.
857 	 */
858 	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
859 	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
860 	ASSERT(*curp == NULL);
861 	*curp = cur;
862 	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
863 	return 0;
864 }
865 
866 /*
867  * Convert a local file to an extents file.
868  * This code is out of bounds for data forks of regular files,
869  * since the file data needs to get logged so things will stay consistent.
870  * (The bmap-level manipulations are ok, though).
871  */
872 void
xfs_bmap_local_to_extents_empty(struct xfs_inode * ip,int whichfork)873 xfs_bmap_local_to_extents_empty(
874 	struct xfs_inode	*ip,
875 	int			whichfork)
876 {
877 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
878 
879 	ASSERT(whichfork != XFS_COW_FORK);
880 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
881 	ASSERT(ifp->if_bytes == 0);
882 	ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
883 
884 	xfs_bmap_forkoff_reset(ip, whichfork);
885 	ifp->if_flags &= ~XFS_IFINLINE;
886 	ifp->if_flags |= XFS_IFEXTENTS;
887 	XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
888 }
889 
890 
891 STATIC int				/* error */
xfs_bmap_local_to_extents(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,xfs_extlen_t total,int * logflagsp,int whichfork,void (* init_fn)(struct xfs_trans * tp,struct xfs_buf * bp,struct xfs_inode * ip,struct xfs_ifork * ifp))892 xfs_bmap_local_to_extents(
893 	xfs_trans_t	*tp,		/* transaction pointer */
894 	xfs_inode_t	*ip,		/* incore inode pointer */
895 	xfs_fsblock_t	*firstblock,	/* first block allocated in xaction */
896 	xfs_extlen_t	total,		/* total blocks needed by transaction */
897 	int		*logflagsp,	/* inode logging flags */
898 	int		whichfork,
899 	void		(*init_fn)(struct xfs_trans *tp,
900 				   struct xfs_buf *bp,
901 				   struct xfs_inode *ip,
902 				   struct xfs_ifork *ifp))
903 {
904 	int		error = 0;
905 	int		flags;		/* logging flags returned */
906 	xfs_ifork_t	*ifp;		/* inode fork pointer */
907 	xfs_alloc_arg_t	args;		/* allocation arguments */
908 	xfs_buf_t	*bp;		/* buffer for extent block */
909 	xfs_bmbt_rec_host_t *ep;	/* extent record pointer */
910 
911 	/*
912 	 * We don't want to deal with the case of keeping inode data inline yet.
913 	 * So sending the data fork of a regular inode is invalid.
914 	 */
915 	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
916 	ifp = XFS_IFORK_PTR(ip, whichfork);
917 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
918 
919 	if (!ifp->if_bytes) {
920 		xfs_bmap_local_to_extents_empty(ip, whichfork);
921 		flags = XFS_ILOG_CORE;
922 		goto done;
923 	}
924 
925 	flags = 0;
926 	error = 0;
927 	ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
928 								XFS_IFINLINE);
929 	memset(&args, 0, sizeof(args));
930 	args.tp = tp;
931 	args.mp = ip->i_mount;
932 	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
933 	args.firstblock = *firstblock;
934 	/*
935 	 * Allocate a block.  We know we need only one, since the
936 	 * file currently fits in an inode.
937 	 */
938 	if (*firstblock == NULLFSBLOCK) {
939 try_another_ag:
940 		args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
941 		args.type = XFS_ALLOCTYPE_START_BNO;
942 	} else {
943 		args.fsbno = *firstblock;
944 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
945 	}
946 	args.total = total;
947 	args.minlen = args.maxlen = args.prod = 1;
948 	error = xfs_alloc_vextent(&args);
949 	if (error)
950 		goto done;
951 
952 	/*
953 	 * During a CoW operation, the allocation and bmbt updates occur in
954 	 * different transactions.  The mapping code tries to put new bmbt
955 	 * blocks near extents being mapped, but the only way to guarantee this
956 	 * is if the alloc and the mapping happen in a single transaction that
957 	 * has a block reservation.  That isn't the case here, so if we run out
958 	 * of space we'll try again with another AG.
959 	 */
960 	if (xfs_sb_version_hasreflink(&ip->i_mount->m_sb) &&
961 	    args.fsbno == NULLFSBLOCK &&
962 	    args.type == XFS_ALLOCTYPE_NEAR_BNO) {
963 		goto try_another_ag;
964 	}
965 	/* Can't fail, the space was reserved. */
966 	ASSERT(args.fsbno != NULLFSBLOCK);
967 	ASSERT(args.len == 1);
968 	*firstblock = args.fsbno;
969 	bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
970 
971 	/*
972 	 * Initialize the block, copy the data and log the remote buffer.
973 	 *
974 	 * The callout is responsible for logging because the remote format
975 	 * might differ from the local format and thus we don't know how much to
976 	 * log here. Note that init_fn must also set the buffer log item type
977 	 * correctly.
978 	 */
979 	init_fn(tp, bp, ip, ifp);
980 
981 	/* account for the change in fork size */
982 	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
983 	xfs_bmap_local_to_extents_empty(ip, whichfork);
984 	flags |= XFS_ILOG_CORE;
985 
986 	xfs_iext_add(ifp, 0, 1);
987 	ep = xfs_iext_get_ext(ifp, 0);
988 	xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
989 	trace_xfs_bmap_post_update(ip, 0,
990 			whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
991 			_THIS_IP_);
992 	XFS_IFORK_NEXT_SET(ip, whichfork, 1);
993 	ip->i_d.di_nblocks = 1;
994 	xfs_trans_mod_dquot_byino(tp, ip,
995 		XFS_TRANS_DQ_BCOUNT, 1L);
996 	flags |= xfs_ilog_fext(whichfork);
997 
998 done:
999 	*logflagsp = flags;
1000 	return error;
1001 }
1002 
1003 /*
1004  * Called from xfs_bmap_add_attrfork to handle btree format files.
1005  */
1006 STATIC int					/* error */
xfs_bmap_add_attrfork_btree(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,int * flags)1007 xfs_bmap_add_attrfork_btree(
1008 	xfs_trans_t		*tp,		/* transaction pointer */
1009 	xfs_inode_t		*ip,		/* incore inode pointer */
1010 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1011 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1012 	int			*flags)		/* inode logging flags */
1013 {
1014 	xfs_btree_cur_t		*cur;		/* btree cursor */
1015 	int			error;		/* error return value */
1016 	xfs_mount_t		*mp;		/* file system mount struct */
1017 	int			stat;		/* newroot status */
1018 
1019 	mp = ip->i_mount;
1020 	if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1021 		*flags |= XFS_ILOG_DBROOT;
1022 	else {
1023 		cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1024 		cur->bc_private.b.dfops = dfops;
1025 		cur->bc_private.b.firstblock = *firstblock;
1026 		if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1027 			goto error0;
1028 		/* must be at least one entry */
1029 		XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
1030 		if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1031 			goto error0;
1032 		if (stat == 0) {
1033 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1034 			return -ENOSPC;
1035 		}
1036 		*firstblock = cur->bc_private.b.firstblock;
1037 		cur->bc_private.b.allocated = 0;
1038 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1039 	}
1040 	return 0;
1041 error0:
1042 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1043 	return error;
1044 }
1045 
1046 /*
1047  * Called from xfs_bmap_add_attrfork to handle extents format files.
1048  */
1049 STATIC int					/* error */
xfs_bmap_add_attrfork_extents(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,int * flags)1050 xfs_bmap_add_attrfork_extents(
1051 	xfs_trans_t		*tp,		/* transaction pointer */
1052 	xfs_inode_t		*ip,		/* incore inode pointer */
1053 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1054 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1055 	int			*flags)		/* inode logging flags */
1056 {
1057 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
1058 	int			error;		/* error return value */
1059 
1060 	if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1061 		return 0;
1062 	cur = NULL;
1063 	error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops, &cur, 0,
1064 		flags, XFS_DATA_FORK);
1065 	if (cur) {
1066 		cur->bc_private.b.allocated = 0;
1067 		xfs_btree_del_cursor(cur,
1068 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1069 	}
1070 	return error;
1071 }
1072 
1073 /*
1074  * Called from xfs_bmap_add_attrfork to handle local format files. Each
1075  * different data fork content type needs a different callout to do the
1076  * conversion. Some are basic and only require special block initialisation
1077  * callouts for the data formating, others (directories) are so specialised they
1078  * handle everything themselves.
1079  *
1080  * XXX (dgc): investigate whether directory conversion can use the generic
1081  * formatting callout. It should be possible - it's just a very complex
1082  * formatter.
1083  */
1084 STATIC int					/* error */
xfs_bmap_add_attrfork_local(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,int * flags)1085 xfs_bmap_add_attrfork_local(
1086 	xfs_trans_t		*tp,		/* transaction pointer */
1087 	xfs_inode_t		*ip,		/* incore inode pointer */
1088 	xfs_fsblock_t		*firstblock,	/* first block allocated */
1089 	struct xfs_defer_ops	*dfops,		/* blocks to free at commit */
1090 	int			*flags)		/* inode logging flags */
1091 {
1092 	xfs_da_args_t		dargs;		/* args for dir/attr code */
1093 
1094 	if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1095 		return 0;
1096 
1097 	if (S_ISDIR(VFS_I(ip)->i_mode)) {
1098 		memset(&dargs, 0, sizeof(dargs));
1099 		dargs.geo = ip->i_mount->m_dir_geo;
1100 		dargs.dp = ip;
1101 		dargs.firstblock = firstblock;
1102 		dargs.dfops = dfops;
1103 		dargs.total = dargs.geo->fsbcount;
1104 		dargs.whichfork = XFS_DATA_FORK;
1105 		dargs.trans = tp;
1106 		return xfs_dir2_sf_to_block(&dargs);
1107 	}
1108 
1109 	if (S_ISLNK(VFS_I(ip)->i_mode))
1110 		return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1111 						 flags, XFS_DATA_FORK,
1112 						 xfs_symlink_local_to_remote);
1113 
1114 	/* should only be called for types that support local format data */
1115 	ASSERT(0);
1116 	return -EFSCORRUPTED;
1117 }
1118 
1119 /*
1120  * Convert inode from non-attributed to attributed.
1121  * Must not be in a transaction, ip must not be locked.
1122  */
1123 int						/* error code */
xfs_bmap_add_attrfork(xfs_inode_t * ip,int size,int rsvd)1124 xfs_bmap_add_attrfork(
1125 	xfs_inode_t		*ip,		/* incore inode pointer */
1126 	int			size,		/* space new attribute needs */
1127 	int			rsvd)		/* xact may use reserved blks */
1128 {
1129 	xfs_fsblock_t		firstblock;	/* 1st block/ag allocated */
1130 	struct xfs_defer_ops	dfops;		/* freed extent records */
1131 	xfs_mount_t		*mp;		/* mount structure */
1132 	xfs_trans_t		*tp;		/* transaction pointer */
1133 	int			blks;		/* space reservation */
1134 	int			version = 1;	/* superblock attr version */
1135 	int			logflags;	/* logging flags */
1136 	int			error;		/* error return value */
1137 
1138 	ASSERT(XFS_IFORK_Q(ip) == 0);
1139 
1140 	mp = ip->i_mount;
1141 	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1142 
1143 	blks = XFS_ADDAFORK_SPACE_RES(mp);
1144 
1145 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_addafork, blks, 0,
1146 			rsvd ? XFS_TRANS_RESERVE : 0, &tp);
1147 	if (error)
1148 		return error;
1149 
1150 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1151 	error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1152 			XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1153 			XFS_QMOPT_RES_REGBLKS);
1154 	if (error)
1155 		goto trans_cancel;
1156 	if (XFS_IFORK_Q(ip))
1157 		goto trans_cancel;
1158 	if (ip->i_d.di_anextents != 0) {
1159 		error = -EFSCORRUPTED;
1160 		goto trans_cancel;
1161 	}
1162 	if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1163 		/*
1164 		 * For inodes coming from pre-6.2 filesystems.
1165 		 */
1166 		ASSERT(ip->i_d.di_aformat == 0);
1167 		ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1168 	}
1169 
1170 	xfs_trans_ijoin(tp, ip, 0);
1171 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1172 
1173 	switch (ip->i_d.di_format) {
1174 	case XFS_DINODE_FMT_DEV:
1175 		ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1176 		break;
1177 	case XFS_DINODE_FMT_UUID:
1178 		ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1179 		break;
1180 	case XFS_DINODE_FMT_LOCAL:
1181 	case XFS_DINODE_FMT_EXTENTS:
1182 	case XFS_DINODE_FMT_BTREE:
1183 		ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1184 		if (!ip->i_d.di_forkoff)
1185 			ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1186 		else if (mp->m_flags & XFS_MOUNT_ATTR2)
1187 			version = 2;
1188 		break;
1189 	default:
1190 		ASSERT(0);
1191 		error = -EINVAL;
1192 		goto trans_cancel;
1193 	}
1194 
1195 	ASSERT(ip->i_afp == NULL);
1196 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1197 	ip->i_afp->if_flags = XFS_IFEXTENTS;
1198 	logflags = 0;
1199 	xfs_defer_init(&dfops, &firstblock);
1200 	switch (ip->i_d.di_format) {
1201 	case XFS_DINODE_FMT_LOCAL:
1202 		error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &dfops,
1203 			&logflags);
1204 		break;
1205 	case XFS_DINODE_FMT_EXTENTS:
1206 		error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1207 			&dfops, &logflags);
1208 		break;
1209 	case XFS_DINODE_FMT_BTREE:
1210 		error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &dfops,
1211 			&logflags);
1212 		break;
1213 	default:
1214 		error = 0;
1215 		break;
1216 	}
1217 	if (logflags)
1218 		xfs_trans_log_inode(tp, ip, logflags);
1219 	if (error)
1220 		goto bmap_cancel;
1221 	if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1222 	   (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1223 		bool log_sb = false;
1224 
1225 		spin_lock(&mp->m_sb_lock);
1226 		if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1227 			xfs_sb_version_addattr(&mp->m_sb);
1228 			log_sb = true;
1229 		}
1230 		if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1231 			xfs_sb_version_addattr2(&mp->m_sb);
1232 			log_sb = true;
1233 		}
1234 		spin_unlock(&mp->m_sb_lock);
1235 		if (log_sb)
1236 			xfs_log_sb(tp);
1237 	}
1238 
1239 	error = xfs_defer_finish(&tp, &dfops, NULL);
1240 	if (error)
1241 		goto bmap_cancel;
1242 	error = xfs_trans_commit(tp);
1243 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1244 	return error;
1245 
1246 bmap_cancel:
1247 	xfs_defer_cancel(&dfops);
1248 trans_cancel:
1249 	xfs_trans_cancel(tp);
1250 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1251 	return error;
1252 }
1253 
1254 /*
1255  * Internal and external extent tree search functions.
1256  */
1257 
1258 /*
1259  * Read in the extents to if_extents.
1260  * All inode fields are set up by caller, we just traverse the btree
1261  * and copy the records in. If the file system cannot contain unwritten
1262  * extents, the records are checked for no "state" flags.
1263  */
1264 int					/* error */
xfs_bmap_read_extents(xfs_trans_t * tp,xfs_inode_t * ip,int whichfork)1265 xfs_bmap_read_extents(
1266 	xfs_trans_t		*tp,	/* transaction pointer */
1267 	xfs_inode_t		*ip,	/* incore inode */
1268 	int			whichfork) /* data or attr fork */
1269 {
1270 	struct xfs_btree_block	*block;	/* current btree block */
1271 	xfs_fsblock_t		bno;	/* block # of "block" */
1272 	xfs_buf_t		*bp;	/* buffer for "block" */
1273 	int			error;	/* error return value */
1274 	xfs_exntfmt_t		exntf;	/* XFS_EXTFMT_NOSTATE, if checking */
1275 	xfs_extnum_t		i, j;	/* index into the extents list */
1276 	xfs_ifork_t		*ifp;	/* fork structure */
1277 	int			level;	/* btree level, for checking */
1278 	xfs_mount_t		*mp;	/* file system mount structure */
1279 	__be64			*pp;	/* pointer to block address */
1280 	/* REFERENCED */
1281 	xfs_extnum_t		room;	/* number of entries there's room for */
1282 
1283 	mp = ip->i_mount;
1284 	ifp = XFS_IFORK_PTR(ip, whichfork);
1285 	exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1286 					XFS_EXTFMT_INODE(ip);
1287 	block = ifp->if_broot;
1288 	/*
1289 	 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1290 	 */
1291 	level = be16_to_cpu(block->bb_level);
1292 	ASSERT(level > 0);
1293 	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1294 	bno = be64_to_cpu(*pp);
1295 
1296 	/*
1297 	 * Go down the tree until leaf level is reached, following the first
1298 	 * pointer (leftmost) at each level.
1299 	 */
1300 	while (level-- > 0) {
1301 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1302 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1303 		if (error)
1304 			return error;
1305 		block = XFS_BUF_TO_BLOCK(bp);
1306 		if (level == 0)
1307 			break;
1308 		pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1309 		bno = be64_to_cpu(*pp);
1310 		XFS_WANT_CORRUPTED_GOTO(mp,
1311 			XFS_FSB_SANITY_CHECK(mp, bno), error0);
1312 		xfs_trans_brelse(tp, bp);
1313 	}
1314 	/*
1315 	 * Here with bp and block set to the leftmost leaf node in the tree.
1316 	 */
1317 	room = xfs_iext_count(ifp);
1318 	i = 0;
1319 	/*
1320 	 * Loop over all leaf nodes.  Copy information to the extent records.
1321 	 */
1322 	for (;;) {
1323 		xfs_bmbt_rec_t	*frp;
1324 		xfs_fsblock_t	nextbno;
1325 		xfs_extnum_t	num_recs;
1326 		xfs_extnum_t	start;
1327 
1328 		num_recs = xfs_btree_get_numrecs(block);
1329 		if (unlikely(i + num_recs > room)) {
1330 			ASSERT(i + num_recs <= room);
1331 			xfs_warn(ip->i_mount,
1332 				"corrupt dinode %Lu, (btree extents).",
1333 				(unsigned long long) ip->i_ino);
1334 			XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1335 				XFS_ERRLEVEL_LOW, ip->i_mount, block);
1336 			goto error0;
1337 		}
1338 		/*
1339 		 * Read-ahead the next leaf block, if any.
1340 		 */
1341 		nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1342 		if (nextbno != NULLFSBLOCK)
1343 			xfs_btree_reada_bufl(mp, nextbno, 1,
1344 					     &xfs_bmbt_buf_ops);
1345 		/*
1346 		 * Copy records into the extent records.
1347 		 */
1348 		frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1349 		start = i;
1350 		for (j = 0; j < num_recs; j++, i++, frp++) {
1351 			xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1352 			trp->l0 = be64_to_cpu(frp->l0);
1353 			trp->l1 = be64_to_cpu(frp->l1);
1354 		}
1355 		if (exntf == XFS_EXTFMT_NOSTATE) {
1356 			/*
1357 			 * Check all attribute bmap btree records and
1358 			 * any "older" data bmap btree records for a
1359 			 * set bit in the "extent flag" position.
1360 			 */
1361 			if (unlikely(xfs_check_nostate_extents(ifp,
1362 					start, num_recs))) {
1363 				XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1364 						 XFS_ERRLEVEL_LOW,
1365 						 ip->i_mount);
1366 				goto error0;
1367 			}
1368 		}
1369 		xfs_trans_brelse(tp, bp);
1370 		bno = nextbno;
1371 		/*
1372 		 * If we've reached the end, stop.
1373 		 */
1374 		if (bno == NULLFSBLOCK)
1375 			break;
1376 		error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1377 				XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1378 		if (error)
1379 			return error;
1380 		block = XFS_BUF_TO_BLOCK(bp);
1381 	}
1382 	if (i != XFS_IFORK_NEXTENTS(ip, whichfork))
1383 		return -EFSCORRUPTED;
1384 	ASSERT(i == xfs_iext_count(ifp));
1385 	XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1386 	return 0;
1387 error0:
1388 	xfs_trans_brelse(tp, bp);
1389 	return -EFSCORRUPTED;
1390 }
1391 
1392 
1393 /*
1394  * Search the extent records for the entry containing block bno.
1395  * If bno lies in a hole, point to the next entry.  If bno lies
1396  * past eof, *eofp will be set, and *prevp will contain the last
1397  * entry (null if none).  Else, *lastxp will be set to the index
1398  * of the found entry; *gotp will contain the entry.
1399  */
1400 STATIC xfs_bmbt_rec_host_t *		/* pointer to found extent entry */
xfs_bmap_search_multi_extents(xfs_ifork_t * ifp,xfs_fileoff_t bno,int * eofp,xfs_extnum_t * lastxp,xfs_bmbt_irec_t * gotp,xfs_bmbt_irec_t * prevp)1401 xfs_bmap_search_multi_extents(
1402 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1403 	xfs_fileoff_t	bno,		/* block number searched for */
1404 	int		*eofp,		/* out: end of file found */
1405 	xfs_extnum_t	*lastxp,	/* out: last extent index */
1406 	xfs_bmbt_irec_t	*gotp,		/* out: extent entry found */
1407 	xfs_bmbt_irec_t	*prevp)		/* out: previous extent entry found */
1408 {
1409 	xfs_bmbt_rec_host_t *ep;		/* extent record pointer */
1410 	xfs_extnum_t	lastx;		/* last extent index */
1411 
1412 	/*
1413 	 * Initialize the extent entry structure to catch access to
1414 	 * uninitialized br_startblock field.
1415 	 */
1416 	gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1417 	gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1418 	gotp->br_state = XFS_EXT_INVALID;
1419 	gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1420 	prevp->br_startoff = NULLFILEOFF;
1421 
1422 	ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1423 	if (lastx > 0) {
1424 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1425 	}
1426 	if (lastx < xfs_iext_count(ifp)) {
1427 		xfs_bmbt_get_all(ep, gotp);
1428 		*eofp = 0;
1429 	} else {
1430 		if (lastx > 0) {
1431 			*gotp = *prevp;
1432 		}
1433 		*eofp = 1;
1434 		ep = NULL;
1435 	}
1436 	*lastxp = lastx;
1437 	return ep;
1438 }
1439 
1440 /*
1441  * Search the extents list for the inode, for the extent containing bno.
1442  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
1443  * *eofp will be set, and *prevp will contain the last entry (null if none).
1444  * Else, *lastxp will be set to the index of the found
1445  * entry; *gotp will contain the entry.
1446  */
1447 xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
xfs_bmap_search_extents(xfs_inode_t * ip,xfs_fileoff_t bno,int fork,int * eofp,xfs_extnum_t * lastxp,xfs_bmbt_irec_t * gotp,xfs_bmbt_irec_t * prevp)1448 xfs_bmap_search_extents(
1449 	xfs_inode_t     *ip,            /* incore inode pointer */
1450 	xfs_fileoff_t   bno,            /* block number searched for */
1451 	int             fork,      	/* data or attr fork */
1452 	int             *eofp,          /* out: end of file found */
1453 	xfs_extnum_t    *lastxp,        /* out: last extent index */
1454 	xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
1455 	xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
1456 {
1457 	xfs_ifork_t	*ifp;		/* inode fork pointer */
1458 	xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
1459 
1460 	XFS_STATS_INC(ip->i_mount, xs_look_exlist);
1461 	ifp = XFS_IFORK_PTR(ip, fork);
1462 
1463 	ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1464 
1465 	if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1466 		     !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1467 		xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1468 				"Access to block zero in inode %llu "
1469 				"start_block: %llx start_off: %llx "
1470 				"blkcnt: %llx extent-state: %x lastx: %x",
1471 			(unsigned long long)ip->i_ino,
1472 			(unsigned long long)gotp->br_startblock,
1473 			(unsigned long long)gotp->br_startoff,
1474 			(unsigned long long)gotp->br_blockcount,
1475 			gotp->br_state, *lastxp);
1476 		*lastxp = NULLEXTNUM;
1477 		*eofp = 1;
1478 		return NULL;
1479 	}
1480 	return ep;
1481 }
1482 
1483 /*
1484  * Returns the file-relative block number of the first unused block(s)
1485  * in the file with at least "len" logically contiguous blocks free.
1486  * This is the lowest-address hole if the file has holes, else the first block
1487  * past the end of file.
1488  * Return 0 if the file is currently local (in-inode).
1489  */
1490 int						/* error */
xfs_bmap_first_unused(xfs_trans_t * tp,xfs_inode_t * ip,xfs_extlen_t len,xfs_fileoff_t * first_unused,int whichfork)1491 xfs_bmap_first_unused(
1492 	xfs_trans_t	*tp,			/* transaction pointer */
1493 	xfs_inode_t	*ip,			/* incore inode */
1494 	xfs_extlen_t	len,			/* size of hole to find */
1495 	xfs_fileoff_t	*first_unused,		/* unused block */
1496 	int		whichfork)		/* data or attr fork */
1497 {
1498 	int		error;			/* error return value */
1499 	int		idx;			/* extent record index */
1500 	xfs_ifork_t	*ifp;			/* inode fork pointer */
1501 	xfs_fileoff_t	lastaddr;		/* last block number seen */
1502 	xfs_fileoff_t	lowest;			/* lowest useful block */
1503 	xfs_fileoff_t	max;			/* starting useful block */
1504 	xfs_fileoff_t	off;			/* offset for this block */
1505 	xfs_extnum_t	nextents;		/* number of extent entries */
1506 
1507 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1508 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1509 	       XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1510 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1511 		*first_unused = 0;
1512 		return 0;
1513 	}
1514 	ifp = XFS_IFORK_PTR(ip, whichfork);
1515 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1516 	    (error = xfs_iread_extents(tp, ip, whichfork)))
1517 		return error;
1518 	lowest = *first_unused;
1519 	nextents = xfs_iext_count(ifp);
1520 	for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1521 		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1522 		off = xfs_bmbt_get_startoff(ep);
1523 		/*
1524 		 * See if the hole before this extent will work.
1525 		 */
1526 		if (off >= lowest + len && off - max >= len) {
1527 			*first_unused = max;
1528 			return 0;
1529 		}
1530 		lastaddr = off + xfs_bmbt_get_blockcount(ep);
1531 		max = XFS_FILEOFF_MAX(lastaddr, lowest);
1532 	}
1533 	*first_unused = max;
1534 	return 0;
1535 }
1536 
1537 /*
1538  * Returns the file-relative block number of the last block - 1 before
1539  * last_block (input value) in the file.
1540  * This is not based on i_size, it is based on the extent records.
1541  * Returns 0 for local files, as they do not have extent records.
1542  */
1543 int						/* error */
xfs_bmap_last_before(xfs_trans_t * tp,xfs_inode_t * ip,xfs_fileoff_t * last_block,int whichfork)1544 xfs_bmap_last_before(
1545 	xfs_trans_t	*tp,			/* transaction pointer */
1546 	xfs_inode_t	*ip,			/* incore inode */
1547 	xfs_fileoff_t	*last_block,		/* last block */
1548 	int		whichfork)		/* data or attr fork */
1549 {
1550 	xfs_fileoff_t	bno;			/* input file offset */
1551 	int		eof;			/* hit end of file */
1552 	xfs_bmbt_rec_host_t *ep;		/* pointer to last extent */
1553 	int		error;			/* error return value */
1554 	xfs_bmbt_irec_t	got;			/* current extent value */
1555 	xfs_ifork_t	*ifp;			/* inode fork pointer */
1556 	xfs_extnum_t	lastx;			/* last extent used */
1557 	xfs_bmbt_irec_t	prev;			/* previous extent value */
1558 
1559 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1560 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1561 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1562 	       return -EIO;
1563 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1564 		*last_block = 0;
1565 		return 0;
1566 	}
1567 	ifp = XFS_IFORK_PTR(ip, whichfork);
1568 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1569 	    (error = xfs_iread_extents(tp, ip, whichfork)))
1570 		return error;
1571 	bno = *last_block - 1;
1572 	ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1573 		&prev);
1574 	if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1575 		if (prev.br_startoff == NULLFILEOFF)
1576 			*last_block = 0;
1577 		else
1578 			*last_block = prev.br_startoff + prev.br_blockcount;
1579 	}
1580 	/*
1581 	 * Otherwise *last_block is already the right answer.
1582 	 */
1583 	return 0;
1584 }
1585 
1586 int
xfs_bmap_last_extent(struct xfs_trans * tp,struct xfs_inode * ip,int whichfork,struct xfs_bmbt_irec * rec,int * is_empty)1587 xfs_bmap_last_extent(
1588 	struct xfs_trans	*tp,
1589 	struct xfs_inode	*ip,
1590 	int			whichfork,
1591 	struct xfs_bmbt_irec	*rec,
1592 	int			*is_empty)
1593 {
1594 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
1595 	int			error;
1596 	int			nextents;
1597 
1598 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1599 		error = xfs_iread_extents(tp, ip, whichfork);
1600 		if (error)
1601 			return error;
1602 	}
1603 
1604 	nextents = xfs_iext_count(ifp);
1605 	if (nextents == 0) {
1606 		*is_empty = 1;
1607 		return 0;
1608 	}
1609 
1610 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1611 	*is_empty = 0;
1612 	return 0;
1613 }
1614 
1615 /*
1616  * Check the last inode extent to determine whether this allocation will result
1617  * in blocks being allocated at the end of the file. When we allocate new data
1618  * blocks at the end of the file which do not start at the previous data block,
1619  * we will try to align the new blocks at stripe unit boundaries.
1620  *
1621  * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1622  * at, or past the EOF.
1623  */
1624 STATIC int
xfs_bmap_isaeof(struct xfs_bmalloca * bma,int whichfork)1625 xfs_bmap_isaeof(
1626 	struct xfs_bmalloca	*bma,
1627 	int			whichfork)
1628 {
1629 	struct xfs_bmbt_irec	rec;
1630 	int			is_empty;
1631 	int			error;
1632 
1633 	bma->aeof = 0;
1634 	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1635 				     &is_empty);
1636 	if (error)
1637 		return error;
1638 
1639 	if (is_empty) {
1640 		bma->aeof = 1;
1641 		return 0;
1642 	}
1643 
1644 	/*
1645 	 * Check if we are allocation or past the last extent, or at least into
1646 	 * the last delayed allocated extent.
1647 	 */
1648 	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1649 		(bma->offset >= rec.br_startoff &&
1650 		 isnullstartblock(rec.br_startblock));
1651 	return 0;
1652 }
1653 
1654 /*
1655  * Returns the file-relative block number of the first block past eof in
1656  * the file.  This is not based on i_size, it is based on the extent records.
1657  * Returns 0 for local files, as they do not have extent records.
1658  */
1659 int
xfs_bmap_last_offset(struct xfs_inode * ip,xfs_fileoff_t * last_block,int whichfork)1660 xfs_bmap_last_offset(
1661 	struct xfs_inode	*ip,
1662 	xfs_fileoff_t		*last_block,
1663 	int			whichfork)
1664 {
1665 	struct xfs_bmbt_irec	rec;
1666 	int			is_empty;
1667 	int			error;
1668 
1669 	*last_block = 0;
1670 
1671 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1672 		return 0;
1673 
1674 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1675 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1676 	       return -EIO;
1677 
1678 	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1679 	if (error || is_empty)
1680 		return error;
1681 
1682 	*last_block = rec.br_startoff + rec.br_blockcount;
1683 	return 0;
1684 }
1685 
1686 /*
1687  * Returns whether the selected fork of the inode has exactly one
1688  * block or not.  For the data fork we check this matches di_size,
1689  * implying the file's range is 0..bsize-1.
1690  */
1691 int					/* 1=>1 block, 0=>otherwise */
xfs_bmap_one_block(xfs_inode_t * ip,int whichfork)1692 xfs_bmap_one_block(
1693 	xfs_inode_t	*ip,		/* incore inode */
1694 	int		whichfork)	/* data or attr fork */
1695 {
1696 	xfs_bmbt_rec_host_t *ep;	/* ptr to fork's extent */
1697 	xfs_ifork_t	*ifp;		/* inode fork pointer */
1698 	int		rval;		/* return value */
1699 	xfs_bmbt_irec_t	s;		/* internal version of extent */
1700 
1701 #ifndef DEBUG
1702 	if (whichfork == XFS_DATA_FORK)
1703 		return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1704 #endif	/* !DEBUG */
1705 	if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1706 		return 0;
1707 	if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1708 		return 0;
1709 	ifp = XFS_IFORK_PTR(ip, whichfork);
1710 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1711 	ep = xfs_iext_get_ext(ifp, 0);
1712 	xfs_bmbt_get_all(ep, &s);
1713 	rval = s.br_startoff == 0 && s.br_blockcount == 1;
1714 	if (rval && whichfork == XFS_DATA_FORK)
1715 		ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1716 	return rval;
1717 }
1718 
1719 /*
1720  * Extent tree manipulation functions used during allocation.
1721  */
1722 
1723 /*
1724  * Convert a delayed allocation to a real allocation.
1725  */
1726 STATIC int				/* error */
xfs_bmap_add_extent_delay_real(struct xfs_bmalloca * bma,int whichfork)1727 xfs_bmap_add_extent_delay_real(
1728 	struct xfs_bmalloca	*bma,
1729 	int			whichfork)
1730 {
1731 	struct xfs_bmbt_irec	*new = &bma->got;
1732 	int			diff;	/* temp value */
1733 	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
1734 	int			error;	/* error return value */
1735 	int			i;	/* temp state */
1736 	xfs_ifork_t		*ifp;	/* inode fork pointer */
1737 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
1738 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
1739 					/* left is 0, right is 1, prev is 2 */
1740 	int			rval=0;	/* return value (logging flags) */
1741 	int			state = 0;/* state bits, accessed thru macros */
1742 	xfs_filblks_t		da_new; /* new count del alloc blocks used */
1743 	xfs_filblks_t		da_old; /* old count del alloc blocks used */
1744 	xfs_filblks_t		temp=0;	/* value for da_new calculations */
1745 	xfs_filblks_t		temp2=0;/* value for da_new calculations */
1746 	int			tmp_rval;	/* partial logging flags */
1747 	struct xfs_mount	*mp;
1748 	xfs_extnum_t		*nextents;
1749 
1750 	mp = bma->ip->i_mount;
1751 	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1752 	ASSERT(whichfork != XFS_ATTR_FORK);
1753 	nextents = (whichfork == XFS_COW_FORK ? &bma->ip->i_cnextents :
1754 						&bma->ip->i_d.di_nextents);
1755 
1756 	ASSERT(bma->idx >= 0);
1757 	ASSERT(bma->idx <= xfs_iext_count(ifp));
1758 	ASSERT(!isnullstartblock(new->br_startblock));
1759 	ASSERT(!bma->cur ||
1760 	       (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1761 
1762 	XFS_STATS_INC(mp, xs_add_exlist);
1763 
1764 #define	LEFT		r[0]
1765 #define	RIGHT		r[1]
1766 #define	PREV		r[2]
1767 
1768 	if (whichfork == XFS_COW_FORK)
1769 		state |= BMAP_COWFORK;
1770 
1771 	/*
1772 	 * Set up a bunch of variables to make the tests simpler.
1773 	 */
1774 	ep = xfs_iext_get_ext(ifp, bma->idx);
1775 	xfs_bmbt_get_all(ep, &PREV);
1776 	new_endoff = new->br_startoff + new->br_blockcount;
1777 	ASSERT(PREV.br_startoff <= new->br_startoff);
1778 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1779 
1780 	da_old = startblockval(PREV.br_startblock);
1781 	da_new = 0;
1782 
1783 	/*
1784 	 * Set flags determining what part of the previous delayed allocation
1785 	 * extent is being replaced by a real allocation.
1786 	 */
1787 	if (PREV.br_startoff == new->br_startoff)
1788 		state |= BMAP_LEFT_FILLING;
1789 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1790 		state |= BMAP_RIGHT_FILLING;
1791 
1792 	/*
1793 	 * Check and set flags if this segment has a left neighbor.
1794 	 * Don't set contiguous if the combined extent would be too large.
1795 	 */
1796 	if (bma->idx > 0) {
1797 		state |= BMAP_LEFT_VALID;
1798 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1799 
1800 		if (isnullstartblock(LEFT.br_startblock))
1801 			state |= BMAP_LEFT_DELAY;
1802 	}
1803 
1804 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1805 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1806 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1807 	    LEFT.br_state == new->br_state &&
1808 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1809 		state |= BMAP_LEFT_CONTIG;
1810 
1811 	/*
1812 	 * Check and set flags if this segment has a right neighbor.
1813 	 * Don't set contiguous if the combined extent would be too large.
1814 	 * Also check for all-three-contiguous being too large.
1815 	 */
1816 	if (bma->idx < xfs_iext_count(ifp) - 1) {
1817 		state |= BMAP_RIGHT_VALID;
1818 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1819 
1820 		if (isnullstartblock(RIGHT.br_startblock))
1821 			state |= BMAP_RIGHT_DELAY;
1822 	}
1823 
1824 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1825 	    new_endoff == RIGHT.br_startoff &&
1826 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1827 	    new->br_state == RIGHT.br_state &&
1828 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1829 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1830 		       BMAP_RIGHT_FILLING)) !=
1831 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1832 		       BMAP_RIGHT_FILLING) ||
1833 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1834 			<= MAXEXTLEN))
1835 		state |= BMAP_RIGHT_CONTIG;
1836 
1837 	error = 0;
1838 	/*
1839 	 * Switch out based on the FILLING and CONTIG state bits.
1840 	 */
1841 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1842 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1843 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1844 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1845 		/*
1846 		 * Filling in all of a previously delayed allocation extent.
1847 		 * The left and right neighbors are both contiguous with new.
1848 		 */
1849 		bma->idx--;
1850 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1851 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1852 			LEFT.br_blockcount + PREV.br_blockcount +
1853 			RIGHT.br_blockcount);
1854 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1855 
1856 		xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1857 		(*nextents)--;
1858 		if (bma->cur == NULL)
1859 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1860 		else {
1861 			rval = XFS_ILOG_CORE;
1862 			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1863 					RIGHT.br_startblock,
1864 					RIGHT.br_blockcount, &i);
1865 			if (error)
1866 				goto done;
1867 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1868 			error = xfs_btree_delete(bma->cur, &i);
1869 			if (error)
1870 				goto done;
1871 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1872 			error = xfs_btree_decrement(bma->cur, 0, &i);
1873 			if (error)
1874 				goto done;
1875 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1876 			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1877 					LEFT.br_startblock,
1878 					LEFT.br_blockcount +
1879 					PREV.br_blockcount +
1880 					RIGHT.br_blockcount, LEFT.br_state);
1881 			if (error)
1882 				goto done;
1883 		}
1884 		break;
1885 
1886 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1887 		/*
1888 		 * Filling in all of a previously delayed allocation extent.
1889 		 * The left neighbor is contiguous, the right is not.
1890 		 */
1891 		bma->idx--;
1892 
1893 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1894 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1895 			LEFT.br_blockcount + PREV.br_blockcount);
1896 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1897 
1898 		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1899 		if (bma->cur == NULL)
1900 			rval = XFS_ILOG_DEXT;
1901 		else {
1902 			rval = 0;
1903 			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1904 					LEFT.br_startblock, LEFT.br_blockcount,
1905 					&i);
1906 			if (error)
1907 				goto done;
1908 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1909 			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1910 					LEFT.br_startblock,
1911 					LEFT.br_blockcount +
1912 					PREV.br_blockcount, LEFT.br_state);
1913 			if (error)
1914 				goto done;
1915 		}
1916 		break;
1917 
1918 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1919 		/*
1920 		 * Filling in all of a previously delayed allocation extent.
1921 		 * The right neighbor is contiguous, the left is not.
1922 		 */
1923 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1924 		xfs_bmbt_set_startblock(ep, new->br_startblock);
1925 		xfs_bmbt_set_blockcount(ep,
1926 			PREV.br_blockcount + RIGHT.br_blockcount);
1927 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1928 
1929 		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1930 		if (bma->cur == NULL)
1931 			rval = XFS_ILOG_DEXT;
1932 		else {
1933 			rval = 0;
1934 			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1935 					RIGHT.br_startblock,
1936 					RIGHT.br_blockcount, &i);
1937 			if (error)
1938 				goto done;
1939 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1940 			error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1941 					new->br_startblock,
1942 					PREV.br_blockcount +
1943 					RIGHT.br_blockcount, PREV.br_state);
1944 			if (error)
1945 				goto done;
1946 		}
1947 		break;
1948 
1949 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1950 		/*
1951 		 * Filling in all of a previously delayed allocation extent.
1952 		 * Neither the left nor right neighbors are contiguous with
1953 		 * the new one.
1954 		 */
1955 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1956 		xfs_bmbt_set_startblock(ep, new->br_startblock);
1957 		xfs_bmbt_set_state(ep, new->br_state);
1958 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1959 
1960 		(*nextents)++;
1961 		if (bma->cur == NULL)
1962 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1963 		else {
1964 			rval = XFS_ILOG_CORE;
1965 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1966 					new->br_startblock, new->br_blockcount,
1967 					&i);
1968 			if (error)
1969 				goto done;
1970 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1971 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1972 			error = xfs_btree_insert(bma->cur, &i);
1973 			if (error)
1974 				goto done;
1975 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1976 		}
1977 		break;
1978 
1979 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1980 		/*
1981 		 * Filling in the first part of a previous delayed allocation.
1982 		 * The left neighbor is contiguous.
1983 		 */
1984 		trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1985 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1986 			LEFT.br_blockcount + new->br_blockcount);
1987 		xfs_bmbt_set_startoff(ep,
1988 			PREV.br_startoff + new->br_blockcount);
1989 		trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1990 
1991 		temp = PREV.br_blockcount - new->br_blockcount;
1992 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1993 		xfs_bmbt_set_blockcount(ep, temp);
1994 		if (bma->cur == NULL)
1995 			rval = XFS_ILOG_DEXT;
1996 		else {
1997 			rval = 0;
1998 			error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1999 					LEFT.br_startblock, LEFT.br_blockcount,
2000 					&i);
2001 			if (error)
2002 				goto done;
2003 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2004 			error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2005 					LEFT.br_startblock,
2006 					LEFT.br_blockcount +
2007 					new->br_blockcount,
2008 					LEFT.br_state);
2009 			if (error)
2010 				goto done;
2011 		}
2012 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2013 			startblockval(PREV.br_startblock));
2014 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2015 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2016 
2017 		bma->idx--;
2018 		break;
2019 
2020 	case BMAP_LEFT_FILLING:
2021 		/*
2022 		 * Filling in the first part of a previous delayed allocation.
2023 		 * The left neighbor is not contiguous.
2024 		 */
2025 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2026 		xfs_bmbt_set_startoff(ep, new_endoff);
2027 		temp = PREV.br_blockcount - new->br_blockcount;
2028 		xfs_bmbt_set_blockcount(ep, temp);
2029 		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2030 		(*nextents)++;
2031 		if (bma->cur == NULL)
2032 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2033 		else {
2034 			rval = XFS_ILOG_CORE;
2035 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2036 					new->br_startblock, new->br_blockcount,
2037 					&i);
2038 			if (error)
2039 				goto done;
2040 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2041 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2042 			error = xfs_btree_insert(bma->cur, &i);
2043 			if (error)
2044 				goto done;
2045 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2046 		}
2047 
2048 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2049 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2050 					bma->firstblock, bma->dfops,
2051 					&bma->cur, 1, &tmp_rval, whichfork);
2052 			rval |= tmp_rval;
2053 			if (error)
2054 				goto done;
2055 		}
2056 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2057 			startblockval(PREV.br_startblock) -
2058 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
2059 		ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2060 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2061 		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2062 		break;
2063 
2064 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2065 		/*
2066 		 * Filling in the last part of a previous delayed allocation.
2067 		 * The right neighbor is contiguous with the new allocation.
2068 		 */
2069 		temp = PREV.br_blockcount - new->br_blockcount;
2070 		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2071 		xfs_bmbt_set_blockcount(ep, temp);
2072 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2073 			new->br_startoff, new->br_startblock,
2074 			new->br_blockcount + RIGHT.br_blockcount,
2075 			RIGHT.br_state);
2076 		trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2077 		if (bma->cur == NULL)
2078 			rval = XFS_ILOG_DEXT;
2079 		else {
2080 			rval = 0;
2081 			error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2082 					RIGHT.br_startblock,
2083 					RIGHT.br_blockcount, &i);
2084 			if (error)
2085 				goto done;
2086 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2087 			error = xfs_bmbt_update(bma->cur, new->br_startoff,
2088 					new->br_startblock,
2089 					new->br_blockcount +
2090 					RIGHT.br_blockcount,
2091 					RIGHT.br_state);
2092 			if (error)
2093 				goto done;
2094 		}
2095 
2096 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2097 			startblockval(PREV.br_startblock));
2098 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2099 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2100 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2101 
2102 		bma->idx++;
2103 		break;
2104 
2105 	case BMAP_RIGHT_FILLING:
2106 		/*
2107 		 * Filling in the last part of a previous delayed allocation.
2108 		 * The right neighbor is not contiguous.
2109 		 */
2110 		temp = PREV.br_blockcount - new->br_blockcount;
2111 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2112 		xfs_bmbt_set_blockcount(ep, temp);
2113 		xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2114 		(*nextents)++;
2115 		if (bma->cur == NULL)
2116 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2117 		else {
2118 			rval = XFS_ILOG_CORE;
2119 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2120 					new->br_startblock, new->br_blockcount,
2121 					&i);
2122 			if (error)
2123 				goto done;
2124 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2125 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2126 			error = xfs_btree_insert(bma->cur, &i);
2127 			if (error)
2128 				goto done;
2129 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2130 		}
2131 
2132 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2133 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2134 				bma->firstblock, bma->dfops, &bma->cur, 1,
2135 				&tmp_rval, whichfork);
2136 			rval |= tmp_rval;
2137 			if (error)
2138 				goto done;
2139 		}
2140 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2141 			startblockval(PREV.br_startblock) -
2142 			(bma->cur ? bma->cur->bc_private.b.allocated : 0));
2143 		ep = xfs_iext_get_ext(ifp, bma->idx);
2144 		xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2145 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2146 
2147 		bma->idx++;
2148 		break;
2149 
2150 	case 0:
2151 		/*
2152 		 * Filling in the middle part of a previous delayed allocation.
2153 		 * Contiguity is impossible here.
2154 		 * This case is avoided almost all the time.
2155 		 *
2156 		 * We start with a delayed allocation:
2157 		 *
2158 		 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2159 		 *  PREV @ idx
2160 		 *
2161 	         * and we are allocating:
2162 		 *                     +rrrrrrrrrrrrrrrrr+
2163 		 *			      new
2164 		 *
2165 		 * and we set it up for insertion as:
2166 		 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2167 		 *                            new
2168 		 *  PREV @ idx          LEFT              RIGHT
2169 		 *                      inserted at idx + 1
2170 		 */
2171 		temp = new->br_startoff - PREV.br_startoff;
2172 		temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2173 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2174 		xfs_bmbt_set_blockcount(ep, temp);	/* truncate PREV */
2175 		LEFT = *new;
2176 		RIGHT.br_state = PREV.br_state;
2177 		RIGHT.br_startblock = nullstartblock(
2178 				(int)xfs_bmap_worst_indlen(bma->ip, temp2));
2179 		RIGHT.br_startoff = new_endoff;
2180 		RIGHT.br_blockcount = temp2;
2181 		/* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2182 		xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2183 		(*nextents)++;
2184 		if (bma->cur == NULL)
2185 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2186 		else {
2187 			rval = XFS_ILOG_CORE;
2188 			error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2189 					new->br_startblock, new->br_blockcount,
2190 					&i);
2191 			if (error)
2192 				goto done;
2193 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2194 			bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2195 			error = xfs_btree_insert(bma->cur, &i);
2196 			if (error)
2197 				goto done;
2198 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2199 		}
2200 
2201 		if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2202 			error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2203 					bma->firstblock, bma->dfops, &bma->cur,
2204 					1, &tmp_rval, whichfork);
2205 			rval |= tmp_rval;
2206 			if (error)
2207 				goto done;
2208 		}
2209 		temp = xfs_bmap_worst_indlen(bma->ip, temp);
2210 		temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2211 		diff = (int)(temp + temp2 -
2212 			     (startblockval(PREV.br_startblock) -
2213 			      (bma->cur ?
2214 			       bma->cur->bc_private.b.allocated : 0)));
2215 		if (diff > 0) {
2216 			error = xfs_mod_fdblocks(bma->ip->i_mount,
2217 						 -((int64_t)diff), false);
2218 			ASSERT(!error);
2219 			if (error)
2220 				goto done;
2221 		}
2222 
2223 		ep = xfs_iext_get_ext(ifp, bma->idx);
2224 		xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2225 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2226 		trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2227 		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2228 			nullstartblock((int)temp2));
2229 		trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2230 
2231 		bma->idx++;
2232 		da_new = temp + temp2;
2233 		break;
2234 
2235 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2236 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2237 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2238 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2239 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2240 	case BMAP_LEFT_CONTIG:
2241 	case BMAP_RIGHT_CONTIG:
2242 		/*
2243 		 * These cases are all impossible.
2244 		 */
2245 		ASSERT(0);
2246 	}
2247 
2248 	/* add reverse mapping */
2249 	error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
2250 	if (error)
2251 		goto done;
2252 
2253 	/* convert to a btree if necessary */
2254 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
2255 		int	tmp_logflags;	/* partial log flag return val */
2256 
2257 		ASSERT(bma->cur == NULL);
2258 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2259 				bma->firstblock, bma->dfops, &bma->cur,
2260 				da_old > 0, &tmp_logflags, whichfork);
2261 		bma->logflags |= tmp_logflags;
2262 		if (error)
2263 			goto done;
2264 	}
2265 
2266 	/* adjust for changes in reserved delayed indirect blocks */
2267 	if (da_old || da_new) {
2268 		temp = da_new;
2269 		if (bma->cur)
2270 			temp += bma->cur->bc_private.b.allocated;
2271 		if (temp < da_old)
2272 			xfs_mod_fdblocks(bma->ip->i_mount,
2273 					(int64_t)(da_old - temp), false);
2274 	}
2275 
2276 	/* clear out the allocated field, done with it now in any case. */
2277 	if (bma->cur)
2278 		bma->cur->bc_private.b.allocated = 0;
2279 
2280 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
2281 done:
2282 	if (whichfork != XFS_COW_FORK)
2283 		bma->logflags |= rval;
2284 	return error;
2285 #undef	LEFT
2286 #undef	RIGHT
2287 #undef	PREV
2288 }
2289 
2290 /*
2291  * Convert an unwritten allocation to a real allocation or vice versa.
2292  */
2293 STATIC int				/* error */
xfs_bmap_add_extent_unwritten_real(struct xfs_trans * tp,xfs_inode_t * ip,int whichfork,xfs_extnum_t * idx,xfs_btree_cur_t ** curp,xfs_bmbt_irec_t * new,xfs_fsblock_t * first,struct xfs_defer_ops * dfops,int * logflagsp)2294 xfs_bmap_add_extent_unwritten_real(
2295 	struct xfs_trans	*tp,
2296 	xfs_inode_t		*ip,	/* incore inode pointer */
2297 	int			whichfork,
2298 	xfs_extnum_t		*idx,	/* extent number to update/insert */
2299 	xfs_btree_cur_t		**curp,	/* if *curp is null, not a btree */
2300 	xfs_bmbt_irec_t		*new,	/* new data to add to file extents */
2301 	xfs_fsblock_t		*first,	/* pointer to firstblock variable */
2302 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
2303 	int			*logflagsp) /* inode logging flags */
2304 {
2305 	xfs_btree_cur_t		*cur;	/* btree cursor */
2306 	xfs_bmbt_rec_host_t	*ep;	/* extent entry for idx */
2307 	int			error;	/* error return value */
2308 	int			i;	/* temp state */
2309 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2310 	xfs_fileoff_t		new_endoff;	/* end offset of new entry */
2311 	xfs_exntst_t		newext;	/* new extent state */
2312 	xfs_exntst_t		oldext;	/* old extent state */
2313 	xfs_bmbt_irec_t		r[3];	/* neighbor extent entries */
2314 					/* left is 0, right is 1, prev is 2 */
2315 	int			rval=0;	/* return value (logging flags) */
2316 	int			state = 0;/* state bits, accessed thru macros */
2317 	struct xfs_mount	*mp = ip->i_mount;
2318 
2319 	*logflagsp = 0;
2320 
2321 	cur = *curp;
2322 	ifp = XFS_IFORK_PTR(ip, whichfork);
2323 	if (whichfork == XFS_COW_FORK)
2324 		state |= BMAP_COWFORK;
2325 
2326 	ASSERT(*idx >= 0);
2327 	ASSERT(*idx <= xfs_iext_count(ifp));
2328 	ASSERT(!isnullstartblock(new->br_startblock));
2329 
2330 	XFS_STATS_INC(mp, xs_add_exlist);
2331 
2332 #define	LEFT		r[0]
2333 #define	RIGHT		r[1]
2334 #define	PREV		r[2]
2335 
2336 	/*
2337 	 * Set up a bunch of variables to make the tests simpler.
2338 	 */
2339 	error = 0;
2340 	ep = xfs_iext_get_ext(ifp, *idx);
2341 	xfs_bmbt_get_all(ep, &PREV);
2342 	newext = new->br_state;
2343 	oldext = (newext == XFS_EXT_UNWRITTEN) ?
2344 		XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2345 	ASSERT(PREV.br_state == oldext);
2346 	new_endoff = new->br_startoff + new->br_blockcount;
2347 	ASSERT(PREV.br_startoff <= new->br_startoff);
2348 	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2349 
2350 	/*
2351 	 * Set flags determining what part of the previous oldext allocation
2352 	 * extent is being replaced by a newext allocation.
2353 	 */
2354 	if (PREV.br_startoff == new->br_startoff)
2355 		state |= BMAP_LEFT_FILLING;
2356 	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2357 		state |= BMAP_RIGHT_FILLING;
2358 
2359 	/*
2360 	 * Check and set flags if this segment has a left neighbor.
2361 	 * Don't set contiguous if the combined extent would be too large.
2362 	 */
2363 	if (*idx > 0) {
2364 		state |= BMAP_LEFT_VALID;
2365 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2366 
2367 		if (isnullstartblock(LEFT.br_startblock))
2368 			state |= BMAP_LEFT_DELAY;
2369 	}
2370 
2371 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2372 	    LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2373 	    LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2374 	    LEFT.br_state == newext &&
2375 	    LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2376 		state |= BMAP_LEFT_CONTIG;
2377 
2378 	/*
2379 	 * Check and set flags if this segment has a right neighbor.
2380 	 * Don't set contiguous if the combined extent would be too large.
2381 	 * Also check for all-three-contiguous being too large.
2382 	 */
2383 	if (*idx < xfs_iext_count(ifp) - 1) {
2384 		state |= BMAP_RIGHT_VALID;
2385 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2386 		if (isnullstartblock(RIGHT.br_startblock))
2387 			state |= BMAP_RIGHT_DELAY;
2388 	}
2389 
2390 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2391 	    new_endoff == RIGHT.br_startoff &&
2392 	    new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2393 	    newext == RIGHT.br_state &&
2394 	    new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2395 	    ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2396 		       BMAP_RIGHT_FILLING)) !=
2397 		      (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2398 		       BMAP_RIGHT_FILLING) ||
2399 	     LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2400 			<= MAXEXTLEN))
2401 		state |= BMAP_RIGHT_CONTIG;
2402 
2403 	/*
2404 	 * Switch out based on the FILLING and CONTIG state bits.
2405 	 */
2406 	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2407 			 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2408 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2409 	     BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2410 		/*
2411 		 * Setting all of a previous oldext extent to newext.
2412 		 * The left and right neighbors are both contiguous with new.
2413 		 */
2414 		--*idx;
2415 
2416 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2417 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2418 			LEFT.br_blockcount + PREV.br_blockcount +
2419 			RIGHT.br_blockcount);
2420 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2421 
2422 		xfs_iext_remove(ip, *idx + 1, 2, state);
2423 		XFS_IFORK_NEXT_SET(ip, whichfork,
2424 				XFS_IFORK_NEXTENTS(ip, whichfork) - 2);
2425 		if (cur == NULL)
2426 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2427 		else {
2428 			rval = XFS_ILOG_CORE;
2429 			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2430 					RIGHT.br_startblock,
2431 					RIGHT.br_blockcount, &i)))
2432 				goto done;
2433 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2434 			if ((error = xfs_btree_delete(cur, &i)))
2435 				goto done;
2436 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2437 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2438 				goto done;
2439 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2440 			if ((error = xfs_btree_delete(cur, &i)))
2441 				goto done;
2442 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2443 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2444 				goto done;
2445 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2446 			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2447 				LEFT.br_startblock,
2448 				LEFT.br_blockcount + PREV.br_blockcount +
2449 				RIGHT.br_blockcount, LEFT.br_state)))
2450 				goto done;
2451 		}
2452 		break;
2453 
2454 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2455 		/*
2456 		 * Setting all of a previous oldext extent to newext.
2457 		 * The left neighbor is contiguous, the right is not.
2458 		 */
2459 		--*idx;
2460 
2461 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2462 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2463 			LEFT.br_blockcount + PREV.br_blockcount);
2464 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2465 
2466 		xfs_iext_remove(ip, *idx + 1, 1, state);
2467 		XFS_IFORK_NEXT_SET(ip, whichfork,
2468 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2469 		if (cur == NULL)
2470 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2471 		else {
2472 			rval = XFS_ILOG_CORE;
2473 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2474 					PREV.br_startblock, PREV.br_blockcount,
2475 					&i)))
2476 				goto done;
2477 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2478 			if ((error = xfs_btree_delete(cur, &i)))
2479 				goto done;
2480 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2481 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2482 				goto done;
2483 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2484 			if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2485 				LEFT.br_startblock,
2486 				LEFT.br_blockcount + PREV.br_blockcount,
2487 				LEFT.br_state)))
2488 				goto done;
2489 		}
2490 		break;
2491 
2492 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2493 		/*
2494 		 * Setting all of a previous oldext extent to newext.
2495 		 * The right neighbor is contiguous, the left is not.
2496 		 */
2497 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2498 		xfs_bmbt_set_blockcount(ep,
2499 			PREV.br_blockcount + RIGHT.br_blockcount);
2500 		xfs_bmbt_set_state(ep, newext);
2501 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2502 		xfs_iext_remove(ip, *idx + 1, 1, state);
2503 		XFS_IFORK_NEXT_SET(ip, whichfork,
2504 				XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2505 		if (cur == NULL)
2506 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2507 		else {
2508 			rval = XFS_ILOG_CORE;
2509 			if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2510 					RIGHT.br_startblock,
2511 					RIGHT.br_blockcount, &i)))
2512 				goto done;
2513 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2514 			if ((error = xfs_btree_delete(cur, &i)))
2515 				goto done;
2516 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2517 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2518 				goto done;
2519 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2520 			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2521 				new->br_startblock,
2522 				new->br_blockcount + RIGHT.br_blockcount,
2523 				newext)))
2524 				goto done;
2525 		}
2526 		break;
2527 
2528 	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2529 		/*
2530 		 * Setting all of a previous oldext extent to newext.
2531 		 * Neither the left nor right neighbors are contiguous with
2532 		 * the new one.
2533 		 */
2534 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2535 		xfs_bmbt_set_state(ep, newext);
2536 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2537 
2538 		if (cur == NULL)
2539 			rval = XFS_ILOG_DEXT;
2540 		else {
2541 			rval = 0;
2542 			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2543 					new->br_startblock, new->br_blockcount,
2544 					&i)))
2545 				goto done;
2546 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2547 			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2548 				new->br_startblock, new->br_blockcount,
2549 				newext)))
2550 				goto done;
2551 		}
2552 		break;
2553 
2554 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2555 		/*
2556 		 * Setting the first part of a previous oldext extent to newext.
2557 		 * The left neighbor is contiguous.
2558 		 */
2559 		trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2560 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2561 			LEFT.br_blockcount + new->br_blockcount);
2562 		xfs_bmbt_set_startoff(ep,
2563 			PREV.br_startoff + new->br_blockcount);
2564 		trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2565 
2566 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2567 		xfs_bmbt_set_startblock(ep,
2568 			new->br_startblock + new->br_blockcount);
2569 		xfs_bmbt_set_blockcount(ep,
2570 			PREV.br_blockcount - new->br_blockcount);
2571 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2572 
2573 		--*idx;
2574 
2575 		if (cur == NULL)
2576 			rval = XFS_ILOG_DEXT;
2577 		else {
2578 			rval = 0;
2579 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2580 					PREV.br_startblock, PREV.br_blockcount,
2581 					&i)))
2582 				goto done;
2583 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2584 			if ((error = xfs_bmbt_update(cur,
2585 				PREV.br_startoff + new->br_blockcount,
2586 				PREV.br_startblock + new->br_blockcount,
2587 				PREV.br_blockcount - new->br_blockcount,
2588 				oldext)))
2589 				goto done;
2590 			if ((error = xfs_btree_decrement(cur, 0, &i)))
2591 				goto done;
2592 			error = xfs_bmbt_update(cur, LEFT.br_startoff,
2593 				LEFT.br_startblock,
2594 				LEFT.br_blockcount + new->br_blockcount,
2595 				LEFT.br_state);
2596 			if (error)
2597 				goto done;
2598 		}
2599 		break;
2600 
2601 	case BMAP_LEFT_FILLING:
2602 		/*
2603 		 * Setting the first part of a previous oldext extent to newext.
2604 		 * The left neighbor is not contiguous.
2605 		 */
2606 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2607 		ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2608 		xfs_bmbt_set_startoff(ep, new_endoff);
2609 		xfs_bmbt_set_blockcount(ep,
2610 			PREV.br_blockcount - new->br_blockcount);
2611 		xfs_bmbt_set_startblock(ep,
2612 			new->br_startblock + new->br_blockcount);
2613 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2614 
2615 		xfs_iext_insert(ip, *idx, 1, new, state);
2616 		XFS_IFORK_NEXT_SET(ip, whichfork,
2617 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2618 		if (cur == NULL)
2619 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2620 		else {
2621 			rval = XFS_ILOG_CORE;
2622 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2623 					PREV.br_startblock, PREV.br_blockcount,
2624 					&i)))
2625 				goto done;
2626 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2627 			if ((error = xfs_bmbt_update(cur,
2628 				PREV.br_startoff + new->br_blockcount,
2629 				PREV.br_startblock + new->br_blockcount,
2630 				PREV.br_blockcount - new->br_blockcount,
2631 				oldext)))
2632 				goto done;
2633 			cur->bc_rec.b = *new;
2634 			if ((error = xfs_btree_insert(cur, &i)))
2635 				goto done;
2636 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2637 		}
2638 		break;
2639 
2640 	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2641 		/*
2642 		 * Setting the last part of a previous oldext extent to newext.
2643 		 * The right neighbor is contiguous with the new allocation.
2644 		 */
2645 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2646 		xfs_bmbt_set_blockcount(ep,
2647 			PREV.br_blockcount - new->br_blockcount);
2648 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2649 
2650 		++*idx;
2651 
2652 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2653 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2654 			new->br_startoff, new->br_startblock,
2655 			new->br_blockcount + RIGHT.br_blockcount, newext);
2656 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2657 
2658 		if (cur == NULL)
2659 			rval = XFS_ILOG_DEXT;
2660 		else {
2661 			rval = 0;
2662 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2663 					PREV.br_startblock,
2664 					PREV.br_blockcount, &i)))
2665 				goto done;
2666 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2667 			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2668 				PREV.br_startblock,
2669 				PREV.br_blockcount - new->br_blockcount,
2670 				oldext)))
2671 				goto done;
2672 			if ((error = xfs_btree_increment(cur, 0, &i)))
2673 				goto done;
2674 			if ((error = xfs_bmbt_update(cur, new->br_startoff,
2675 				new->br_startblock,
2676 				new->br_blockcount + RIGHT.br_blockcount,
2677 				newext)))
2678 				goto done;
2679 		}
2680 		break;
2681 
2682 	case BMAP_RIGHT_FILLING:
2683 		/*
2684 		 * Setting the last part of a previous oldext extent to newext.
2685 		 * The right neighbor is not contiguous.
2686 		 */
2687 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2688 		xfs_bmbt_set_blockcount(ep,
2689 			PREV.br_blockcount - new->br_blockcount);
2690 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2691 
2692 		++*idx;
2693 		xfs_iext_insert(ip, *idx, 1, new, state);
2694 
2695 		XFS_IFORK_NEXT_SET(ip, whichfork,
2696 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2697 		if (cur == NULL)
2698 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2699 		else {
2700 			rval = XFS_ILOG_CORE;
2701 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2702 					PREV.br_startblock, PREV.br_blockcount,
2703 					&i)))
2704 				goto done;
2705 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2706 			if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2707 				PREV.br_startblock,
2708 				PREV.br_blockcount - new->br_blockcount,
2709 				oldext)))
2710 				goto done;
2711 			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2712 					new->br_startblock, new->br_blockcount,
2713 					&i)))
2714 				goto done;
2715 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2716 			cur->bc_rec.b.br_state = new->br_state;
2717 			if ((error = xfs_btree_insert(cur, &i)))
2718 				goto done;
2719 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2720 		}
2721 		break;
2722 
2723 	case 0:
2724 		/*
2725 		 * Setting the middle part of a previous oldext extent to
2726 		 * newext.  Contiguity is impossible here.
2727 		 * One extent becomes three extents.
2728 		 */
2729 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2730 		xfs_bmbt_set_blockcount(ep,
2731 			new->br_startoff - PREV.br_startoff);
2732 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2733 
2734 		r[0] = *new;
2735 		r[1].br_startoff = new_endoff;
2736 		r[1].br_blockcount =
2737 			PREV.br_startoff + PREV.br_blockcount - new_endoff;
2738 		r[1].br_startblock = new->br_startblock + new->br_blockcount;
2739 		r[1].br_state = oldext;
2740 
2741 		++*idx;
2742 		xfs_iext_insert(ip, *idx, 2, &r[0], state);
2743 
2744 		XFS_IFORK_NEXT_SET(ip, whichfork,
2745 				XFS_IFORK_NEXTENTS(ip, whichfork) + 2);
2746 		if (cur == NULL)
2747 			rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2748 		else {
2749 			rval = XFS_ILOG_CORE;
2750 			if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2751 					PREV.br_startblock, PREV.br_blockcount,
2752 					&i)))
2753 				goto done;
2754 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2755 			/* new right extent - oldext */
2756 			if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2757 				r[1].br_startblock, r[1].br_blockcount,
2758 				r[1].br_state)))
2759 				goto done;
2760 			/* new left extent - oldext */
2761 			cur->bc_rec.b = PREV;
2762 			cur->bc_rec.b.br_blockcount =
2763 				new->br_startoff - PREV.br_startoff;
2764 			if ((error = xfs_btree_insert(cur, &i)))
2765 				goto done;
2766 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2767 			/*
2768 			 * Reset the cursor to the position of the new extent
2769 			 * we are about to insert as we can't trust it after
2770 			 * the previous insert.
2771 			 */
2772 			if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2773 					new->br_startblock, new->br_blockcount,
2774 					&i)))
2775 				goto done;
2776 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2777 			/* new middle extent - newext */
2778 			cur->bc_rec.b.br_state = new->br_state;
2779 			if ((error = xfs_btree_insert(cur, &i)))
2780 				goto done;
2781 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2782 		}
2783 		break;
2784 
2785 	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2786 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2787 	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2788 	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2789 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2790 	case BMAP_LEFT_CONTIG:
2791 	case BMAP_RIGHT_CONTIG:
2792 		/*
2793 		 * These cases are all impossible.
2794 		 */
2795 		ASSERT(0);
2796 	}
2797 
2798 	/* update reverse mappings */
2799 	error = xfs_rmap_convert_extent(mp, dfops, ip, whichfork, new);
2800 	if (error)
2801 		goto done;
2802 
2803 	/* convert to a btree if necessary */
2804 	if (xfs_bmap_needs_btree(ip, whichfork)) {
2805 		int	tmp_logflags;	/* partial log flag return val */
2806 
2807 		ASSERT(cur == NULL);
2808 		error = xfs_bmap_extents_to_btree(tp, ip, first, dfops, &cur,
2809 				0, &tmp_logflags, whichfork);
2810 		*logflagsp |= tmp_logflags;
2811 		if (error)
2812 			goto done;
2813 	}
2814 
2815 	/* clear out the allocated field, done with it now in any case. */
2816 	if (cur) {
2817 		cur->bc_private.b.allocated = 0;
2818 		*curp = cur;
2819 	}
2820 
2821 	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2822 done:
2823 	*logflagsp |= rval;
2824 	return error;
2825 #undef	LEFT
2826 #undef	RIGHT
2827 #undef	PREV
2828 }
2829 
2830 /*
2831  * Convert a hole to a delayed allocation.
2832  */
2833 STATIC void
xfs_bmap_add_extent_hole_delay(xfs_inode_t * ip,int whichfork,xfs_extnum_t * idx,xfs_bmbt_irec_t * new)2834 xfs_bmap_add_extent_hole_delay(
2835 	xfs_inode_t		*ip,	/* incore inode pointer */
2836 	int			whichfork,
2837 	xfs_extnum_t		*idx,	/* extent number to update/insert */
2838 	xfs_bmbt_irec_t		*new)	/* new data to add to file extents */
2839 {
2840 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2841 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2842 	xfs_filblks_t		newlen=0;	/* new indirect size */
2843 	xfs_filblks_t		oldlen=0;	/* old indirect size */
2844 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2845 	int			state;  /* state bits, accessed thru macros */
2846 	xfs_filblks_t		temp=0;	/* temp for indirect calculations */
2847 
2848 	ifp = XFS_IFORK_PTR(ip, whichfork);
2849 	state = 0;
2850 	if (whichfork == XFS_COW_FORK)
2851 		state |= BMAP_COWFORK;
2852 	ASSERT(isnullstartblock(new->br_startblock));
2853 
2854 	/*
2855 	 * Check and set flags if this segment has a left neighbor
2856 	 */
2857 	if (*idx > 0) {
2858 		state |= BMAP_LEFT_VALID;
2859 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2860 
2861 		if (isnullstartblock(left.br_startblock))
2862 			state |= BMAP_LEFT_DELAY;
2863 	}
2864 
2865 	/*
2866 	 * Check and set flags if the current (right) segment exists.
2867 	 * If it doesn't exist, we're converting the hole at end-of-file.
2868 	 */
2869 	if (*idx < xfs_iext_count(ifp)) {
2870 		state |= BMAP_RIGHT_VALID;
2871 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2872 
2873 		if (isnullstartblock(right.br_startblock))
2874 			state |= BMAP_RIGHT_DELAY;
2875 	}
2876 
2877 	/*
2878 	 * Set contiguity flags on the left and right neighbors.
2879 	 * Don't let extents get too large, even if the pieces are contiguous.
2880 	 */
2881 	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2882 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
2883 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2884 		state |= BMAP_LEFT_CONTIG;
2885 
2886 	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2887 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
2888 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2889 	    (!(state & BMAP_LEFT_CONTIG) ||
2890 	     (left.br_blockcount + new->br_blockcount +
2891 	      right.br_blockcount <= MAXEXTLEN)))
2892 		state |= BMAP_RIGHT_CONTIG;
2893 
2894 	/*
2895 	 * Switch out based on the contiguity flags.
2896 	 */
2897 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2898 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2899 		/*
2900 		 * New allocation is contiguous with delayed allocations
2901 		 * on the left and on the right.
2902 		 * Merge all three into a single extent record.
2903 		 */
2904 		--*idx;
2905 		temp = left.br_blockcount + new->br_blockcount +
2906 			right.br_blockcount;
2907 
2908 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2909 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2910 		oldlen = startblockval(left.br_startblock) +
2911 			startblockval(new->br_startblock) +
2912 			startblockval(right.br_startblock);
2913 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2914 					 oldlen);
2915 		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2916 			nullstartblock((int)newlen));
2917 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2918 
2919 		xfs_iext_remove(ip, *idx + 1, 1, state);
2920 		break;
2921 
2922 	case BMAP_LEFT_CONTIG:
2923 		/*
2924 		 * New allocation is contiguous with a delayed allocation
2925 		 * on the left.
2926 		 * Merge the new allocation with the left neighbor.
2927 		 */
2928 		--*idx;
2929 		temp = left.br_blockcount + new->br_blockcount;
2930 
2931 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2932 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2933 		oldlen = startblockval(left.br_startblock) +
2934 			startblockval(new->br_startblock);
2935 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2936 					 oldlen);
2937 		xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2938 			nullstartblock((int)newlen));
2939 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2940 		break;
2941 
2942 	case BMAP_RIGHT_CONTIG:
2943 		/*
2944 		 * New allocation is contiguous with a delayed allocation
2945 		 * on the right.
2946 		 * Merge the new allocation with the right neighbor.
2947 		 */
2948 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2949 		temp = new->br_blockcount + right.br_blockcount;
2950 		oldlen = startblockval(new->br_startblock) +
2951 			startblockval(right.br_startblock);
2952 		newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2953 					 oldlen);
2954 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2955 			new->br_startoff,
2956 			nullstartblock((int)newlen), temp, right.br_state);
2957 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2958 		break;
2959 
2960 	case 0:
2961 		/*
2962 		 * New allocation is not contiguous with another
2963 		 * delayed allocation.
2964 		 * Insert a new entry.
2965 		 */
2966 		oldlen = newlen = 0;
2967 		xfs_iext_insert(ip, *idx, 1, new, state);
2968 		break;
2969 	}
2970 	if (oldlen != newlen) {
2971 		ASSERT(oldlen > newlen);
2972 		xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2973 				 false);
2974 		/*
2975 		 * Nothing to do for disk quota accounting here.
2976 		 */
2977 	}
2978 }
2979 
2980 /*
2981  * Convert a hole to a real allocation.
2982  */
2983 STATIC int				/* error */
xfs_bmap_add_extent_hole_real(struct xfs_bmalloca * bma,int whichfork)2984 xfs_bmap_add_extent_hole_real(
2985 	struct xfs_bmalloca	*bma,
2986 	int			whichfork)
2987 {
2988 	struct xfs_bmbt_irec	*new = &bma->got;
2989 	int			error;	/* error return value */
2990 	int			i;	/* temp state */
2991 	xfs_ifork_t		*ifp;	/* inode fork pointer */
2992 	xfs_bmbt_irec_t		left;	/* left neighbor extent entry */
2993 	xfs_bmbt_irec_t		right;	/* right neighbor extent entry */
2994 	int			rval=0;	/* return value (logging flags) */
2995 	int			state;	/* state bits, accessed thru macros */
2996 	struct xfs_mount	*mp;
2997 
2998 	mp = bma->ip->i_mount;
2999 	ifp = XFS_IFORK_PTR(bma->ip, whichfork);
3000 
3001 	ASSERT(bma->idx >= 0);
3002 	ASSERT(bma->idx <= xfs_iext_count(ifp));
3003 	ASSERT(!isnullstartblock(new->br_startblock));
3004 	ASSERT(!bma->cur ||
3005 	       !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
3006 	ASSERT(whichfork != XFS_COW_FORK);
3007 
3008 	XFS_STATS_INC(mp, xs_add_exlist);
3009 
3010 	state = 0;
3011 	if (whichfork == XFS_ATTR_FORK)
3012 		state |= BMAP_ATTRFORK;
3013 
3014 	/*
3015 	 * Check and set flags if this segment has a left neighbor.
3016 	 */
3017 	if (bma->idx > 0) {
3018 		state |= BMAP_LEFT_VALID;
3019 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
3020 		if (isnullstartblock(left.br_startblock))
3021 			state |= BMAP_LEFT_DELAY;
3022 	}
3023 
3024 	/*
3025 	 * Check and set flags if this segment has a current value.
3026 	 * Not true if we're inserting into the "hole" at eof.
3027 	 */
3028 	if (bma->idx < xfs_iext_count(ifp)) {
3029 		state |= BMAP_RIGHT_VALID;
3030 		xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3031 		if (isnullstartblock(right.br_startblock))
3032 			state |= BMAP_RIGHT_DELAY;
3033 	}
3034 
3035 	/*
3036 	 * We're inserting a real allocation between "left" and "right".
3037 	 * Set the contiguity flags.  Don't let extents get too large.
3038 	 */
3039 	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3040 	    left.br_startoff + left.br_blockcount == new->br_startoff &&
3041 	    left.br_startblock + left.br_blockcount == new->br_startblock &&
3042 	    left.br_state == new->br_state &&
3043 	    left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3044 		state |= BMAP_LEFT_CONTIG;
3045 
3046 	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3047 	    new->br_startoff + new->br_blockcount == right.br_startoff &&
3048 	    new->br_startblock + new->br_blockcount == right.br_startblock &&
3049 	    new->br_state == right.br_state &&
3050 	    new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3051 	    (!(state & BMAP_LEFT_CONTIG) ||
3052 	     left.br_blockcount + new->br_blockcount +
3053 	     right.br_blockcount <= MAXEXTLEN))
3054 		state |= BMAP_RIGHT_CONTIG;
3055 
3056 	error = 0;
3057 	/*
3058 	 * Select which case we're in here, and implement it.
3059 	 */
3060 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3061 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3062 		/*
3063 		 * New allocation is contiguous with real allocations on the
3064 		 * left and on the right.
3065 		 * Merge all three into a single extent record.
3066 		 */
3067 		--bma->idx;
3068 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3069 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3070 			left.br_blockcount + new->br_blockcount +
3071 			right.br_blockcount);
3072 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3073 
3074 		xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3075 
3076 		XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3077 			XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3078 		if (bma->cur == NULL) {
3079 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3080 		} else {
3081 			rval = XFS_ILOG_CORE;
3082 			error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3083 					right.br_startblock, right.br_blockcount,
3084 					&i);
3085 			if (error)
3086 				goto done;
3087 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3088 			error = xfs_btree_delete(bma->cur, &i);
3089 			if (error)
3090 				goto done;
3091 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3092 			error = xfs_btree_decrement(bma->cur, 0, &i);
3093 			if (error)
3094 				goto done;
3095 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3096 			error = xfs_bmbt_update(bma->cur, left.br_startoff,
3097 					left.br_startblock,
3098 					left.br_blockcount +
3099 						new->br_blockcount +
3100 						right.br_blockcount,
3101 					left.br_state);
3102 			if (error)
3103 				goto done;
3104 		}
3105 		break;
3106 
3107 	case BMAP_LEFT_CONTIG:
3108 		/*
3109 		 * New allocation is contiguous with a real allocation
3110 		 * on the left.
3111 		 * Merge the new allocation with the left neighbor.
3112 		 */
3113 		--bma->idx;
3114 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3115 		xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3116 			left.br_blockcount + new->br_blockcount);
3117 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3118 
3119 		if (bma->cur == NULL) {
3120 			rval = xfs_ilog_fext(whichfork);
3121 		} else {
3122 			rval = 0;
3123 			error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3124 					left.br_startblock, left.br_blockcount,
3125 					&i);
3126 			if (error)
3127 				goto done;
3128 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3129 			error = xfs_bmbt_update(bma->cur, left.br_startoff,
3130 					left.br_startblock,
3131 					left.br_blockcount +
3132 						new->br_blockcount,
3133 					left.br_state);
3134 			if (error)
3135 				goto done;
3136 		}
3137 		break;
3138 
3139 	case BMAP_RIGHT_CONTIG:
3140 		/*
3141 		 * New allocation is contiguous with a real allocation
3142 		 * on the right.
3143 		 * Merge the new allocation with the right neighbor.
3144 		 */
3145 		trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3146 		xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3147 			new->br_startoff, new->br_startblock,
3148 			new->br_blockcount + right.br_blockcount,
3149 			right.br_state);
3150 		trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3151 
3152 		if (bma->cur == NULL) {
3153 			rval = xfs_ilog_fext(whichfork);
3154 		} else {
3155 			rval = 0;
3156 			error = xfs_bmbt_lookup_eq(bma->cur,
3157 					right.br_startoff,
3158 					right.br_startblock,
3159 					right.br_blockcount, &i);
3160 			if (error)
3161 				goto done;
3162 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3163 			error = xfs_bmbt_update(bma->cur, new->br_startoff,
3164 					new->br_startblock,
3165 					new->br_blockcount +
3166 						right.br_blockcount,
3167 					right.br_state);
3168 			if (error)
3169 				goto done;
3170 		}
3171 		break;
3172 
3173 	case 0:
3174 		/*
3175 		 * New allocation is not contiguous with another
3176 		 * real allocation.
3177 		 * Insert a new entry.
3178 		 */
3179 		xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3180 		XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3181 			XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3182 		if (bma->cur == NULL) {
3183 			rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3184 		} else {
3185 			rval = XFS_ILOG_CORE;
3186 			error = xfs_bmbt_lookup_eq(bma->cur,
3187 					new->br_startoff,
3188 					new->br_startblock,
3189 					new->br_blockcount, &i);
3190 			if (error)
3191 				goto done;
3192 			XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3193 			bma->cur->bc_rec.b.br_state = new->br_state;
3194 			error = xfs_btree_insert(bma->cur, &i);
3195 			if (error)
3196 				goto done;
3197 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3198 		}
3199 		break;
3200 	}
3201 
3202 	/* add reverse mapping */
3203 	error = xfs_rmap_map_extent(mp, bma->dfops, bma->ip, whichfork, new);
3204 	if (error)
3205 		goto done;
3206 
3207 	/* convert to a btree if necessary */
3208 	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3209 		int	tmp_logflags;	/* partial log flag return val */
3210 
3211 		ASSERT(bma->cur == NULL);
3212 		error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3213 				bma->firstblock, bma->dfops, &bma->cur,
3214 				0, &tmp_logflags, whichfork);
3215 		bma->logflags |= tmp_logflags;
3216 		if (error)
3217 			goto done;
3218 	}
3219 
3220 	/* clear out the allocated field, done with it now in any case. */
3221 	if (bma->cur)
3222 		bma->cur->bc_private.b.allocated = 0;
3223 
3224 	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3225 done:
3226 	bma->logflags |= rval;
3227 	return error;
3228 }
3229 
3230 /*
3231  * Functions used in the extent read, allocate and remove paths
3232  */
3233 
3234 /*
3235  * Adjust the size of the new extent based on di_extsize and rt extsize.
3236  */
3237 int
xfs_bmap_extsize_align(xfs_mount_t * mp,xfs_bmbt_irec_t * gotp,xfs_bmbt_irec_t * prevp,xfs_extlen_t extsz,int rt,int eof,int delay,int convert,xfs_fileoff_t * offp,xfs_extlen_t * lenp)3238 xfs_bmap_extsize_align(
3239 	xfs_mount_t	*mp,
3240 	xfs_bmbt_irec_t	*gotp,		/* next extent pointer */
3241 	xfs_bmbt_irec_t	*prevp,		/* previous extent pointer */
3242 	xfs_extlen_t	extsz,		/* align to this extent size */
3243 	int		rt,		/* is this a realtime inode? */
3244 	int		eof,		/* is extent at end-of-file? */
3245 	int		delay,		/* creating delalloc extent? */
3246 	int		convert,	/* overwriting unwritten extent? */
3247 	xfs_fileoff_t	*offp,		/* in/out: aligned offset */
3248 	xfs_extlen_t	*lenp)		/* in/out: aligned length */
3249 {
3250 	xfs_fileoff_t	orig_off;	/* original offset */
3251 	xfs_extlen_t	orig_alen;	/* original length */
3252 	xfs_fileoff_t	orig_end;	/* original off+len */
3253 	xfs_fileoff_t	nexto;		/* next file offset */
3254 	xfs_fileoff_t	prevo;		/* previous file offset */
3255 	xfs_fileoff_t	align_off;	/* temp for offset */
3256 	xfs_extlen_t	align_alen;	/* temp for length */
3257 	xfs_extlen_t	temp;		/* temp for calculations */
3258 
3259 	if (convert)
3260 		return 0;
3261 
3262 	orig_off = align_off = *offp;
3263 	orig_alen = align_alen = *lenp;
3264 	orig_end = orig_off + orig_alen;
3265 
3266 	/*
3267 	 * If this request overlaps an existing extent, then don't
3268 	 * attempt to perform any additional alignment.
3269 	 */
3270 	if (!delay && !eof &&
3271 	    (orig_off >= gotp->br_startoff) &&
3272 	    (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3273 		return 0;
3274 	}
3275 
3276 	/*
3277 	 * If the file offset is unaligned vs. the extent size
3278 	 * we need to align it.  This will be possible unless
3279 	 * the file was previously written with a kernel that didn't
3280 	 * perform this alignment, or if a truncate shot us in the
3281 	 * foot.
3282 	 */
3283 	temp = do_mod(orig_off, extsz);
3284 	if (temp) {
3285 		align_alen += temp;
3286 		align_off -= temp;
3287 	}
3288 
3289 	/* Same adjustment for the end of the requested area. */
3290 	temp = (align_alen % extsz);
3291 	if (temp)
3292 		align_alen += extsz - temp;
3293 
3294 	/*
3295 	 * For large extent hint sizes, the aligned extent might be larger than
3296 	 * MAXEXTLEN. In that case, reduce the size by an extsz so that it pulls
3297 	 * the length back under MAXEXTLEN. The outer allocation loops handle
3298 	 * short allocation just fine, so it is safe to do this. We only want to
3299 	 * do it when we are forced to, though, because it means more allocation
3300 	 * operations are required.
3301 	 */
3302 	while (align_alen > MAXEXTLEN)
3303 		align_alen -= extsz;
3304 	ASSERT(align_alen <= MAXEXTLEN);
3305 
3306 	/*
3307 	 * If the previous block overlaps with this proposed allocation
3308 	 * then move the start forward without adjusting the length.
3309 	 */
3310 	if (prevp->br_startoff != NULLFILEOFF) {
3311 		if (prevp->br_startblock == HOLESTARTBLOCK)
3312 			prevo = prevp->br_startoff;
3313 		else
3314 			prevo = prevp->br_startoff + prevp->br_blockcount;
3315 	} else
3316 		prevo = 0;
3317 	if (align_off != orig_off && align_off < prevo)
3318 		align_off = prevo;
3319 	/*
3320 	 * If the next block overlaps with this proposed allocation
3321 	 * then move the start back without adjusting the length,
3322 	 * but not before offset 0.
3323 	 * This may of course make the start overlap previous block,
3324 	 * and if we hit the offset 0 limit then the next block
3325 	 * can still overlap too.
3326 	 */
3327 	if (!eof && gotp->br_startoff != NULLFILEOFF) {
3328 		if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3329 		    (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3330 			nexto = gotp->br_startoff + gotp->br_blockcount;
3331 		else
3332 			nexto = gotp->br_startoff;
3333 	} else
3334 		nexto = NULLFILEOFF;
3335 	if (!eof &&
3336 	    align_off + align_alen != orig_end &&
3337 	    align_off + align_alen > nexto)
3338 		align_off = nexto > align_alen ? nexto - align_alen : 0;
3339 	/*
3340 	 * If we're now overlapping the next or previous extent that
3341 	 * means we can't fit an extsz piece in this hole.  Just move
3342 	 * the start forward to the first valid spot and set
3343 	 * the length so we hit the end.
3344 	 */
3345 	if (align_off != orig_off && align_off < prevo)
3346 		align_off = prevo;
3347 	if (align_off + align_alen != orig_end &&
3348 	    align_off + align_alen > nexto &&
3349 	    nexto != NULLFILEOFF) {
3350 		ASSERT(nexto > prevo);
3351 		align_alen = nexto - align_off;
3352 	}
3353 
3354 	/*
3355 	 * If realtime, and the result isn't a multiple of the realtime
3356 	 * extent size we need to remove blocks until it is.
3357 	 */
3358 	if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3359 		/*
3360 		 * We're not covering the original request, or
3361 		 * we won't be able to once we fix the length.
3362 		 */
3363 		if (orig_off < align_off ||
3364 		    orig_end > align_off + align_alen ||
3365 		    align_alen - temp < orig_alen)
3366 			return -EINVAL;
3367 		/*
3368 		 * Try to fix it by moving the start up.
3369 		 */
3370 		if (align_off + temp <= orig_off) {
3371 			align_alen -= temp;
3372 			align_off += temp;
3373 		}
3374 		/*
3375 		 * Try to fix it by moving the end in.
3376 		 */
3377 		else if (align_off + align_alen - temp >= orig_end)
3378 			align_alen -= temp;
3379 		/*
3380 		 * Set the start to the minimum then trim the length.
3381 		 */
3382 		else {
3383 			align_alen -= orig_off - align_off;
3384 			align_off = orig_off;
3385 			align_alen -= align_alen % mp->m_sb.sb_rextsize;
3386 		}
3387 		/*
3388 		 * Result doesn't cover the request, fail it.
3389 		 */
3390 		if (orig_off < align_off || orig_end > align_off + align_alen)
3391 			return -EINVAL;
3392 	} else {
3393 		ASSERT(orig_off >= align_off);
3394 		/* see MAXEXTLEN handling above */
3395 		ASSERT(orig_end <= align_off + align_alen ||
3396 		       align_alen + extsz > MAXEXTLEN);
3397 	}
3398 
3399 #ifdef DEBUG
3400 	if (!eof && gotp->br_startoff != NULLFILEOFF)
3401 		ASSERT(align_off + align_alen <= gotp->br_startoff);
3402 	if (prevp->br_startoff != NULLFILEOFF)
3403 		ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3404 #endif
3405 
3406 	*lenp = align_alen;
3407 	*offp = align_off;
3408 	return 0;
3409 }
3410 
3411 #define XFS_ALLOC_GAP_UNITS	4
3412 
3413 void
xfs_bmap_adjacent(struct xfs_bmalloca * ap)3414 xfs_bmap_adjacent(
3415 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3416 {
3417 	xfs_fsblock_t	adjust;		/* adjustment to block numbers */
3418 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3419 	xfs_mount_t	*mp;		/* mount point structure */
3420 	int		nullfb;		/* true if ap->firstblock isn't set */
3421 	int		rt;		/* true if inode is realtime */
3422 
3423 #define	ISVALID(x,y)	\
3424 	(rt ? \
3425 		(x) < mp->m_sb.sb_rblocks : \
3426 		XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3427 		XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3428 		XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3429 
3430 	mp = ap->ip->i_mount;
3431 	nullfb = *ap->firstblock == NULLFSBLOCK;
3432 	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3433 		xfs_alloc_is_userdata(ap->datatype);
3434 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3435 	/*
3436 	 * If allocating at eof, and there's a previous real block,
3437 	 * try to use its last block as our starting point.
3438 	 */
3439 	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3440 	    !isnullstartblock(ap->prev.br_startblock) &&
3441 	    ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3442 		    ap->prev.br_startblock)) {
3443 		ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3444 		/*
3445 		 * Adjust for the gap between prevp and us.
3446 		 */
3447 		adjust = ap->offset -
3448 			(ap->prev.br_startoff + ap->prev.br_blockcount);
3449 		if (adjust &&
3450 		    ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3451 			ap->blkno += adjust;
3452 	}
3453 	/*
3454 	 * If not at eof, then compare the two neighbor blocks.
3455 	 * Figure out whether either one gives us a good starting point,
3456 	 * and pick the better one.
3457 	 */
3458 	else if (!ap->eof) {
3459 		xfs_fsblock_t	gotbno;		/* right side block number */
3460 		xfs_fsblock_t	gotdiff=0;	/* right side difference */
3461 		xfs_fsblock_t	prevbno;	/* left side block number */
3462 		xfs_fsblock_t	prevdiff=0;	/* left side difference */
3463 
3464 		/*
3465 		 * If there's a previous (left) block, select a requested
3466 		 * start block based on it.
3467 		 */
3468 		if (ap->prev.br_startoff != NULLFILEOFF &&
3469 		    !isnullstartblock(ap->prev.br_startblock) &&
3470 		    (prevbno = ap->prev.br_startblock +
3471 			       ap->prev.br_blockcount) &&
3472 		    ISVALID(prevbno, ap->prev.br_startblock)) {
3473 			/*
3474 			 * Calculate gap to end of previous block.
3475 			 */
3476 			adjust = prevdiff = ap->offset -
3477 				(ap->prev.br_startoff +
3478 				 ap->prev.br_blockcount);
3479 			/*
3480 			 * Figure the startblock based on the previous block's
3481 			 * end and the gap size.
3482 			 * Heuristic!
3483 			 * If the gap is large relative to the piece we're
3484 			 * allocating, or using it gives us an invalid block
3485 			 * number, then just use the end of the previous block.
3486 			 */
3487 			if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3488 			    ISVALID(prevbno + prevdiff,
3489 				    ap->prev.br_startblock))
3490 				prevbno += adjust;
3491 			else
3492 				prevdiff += 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, prevbno) != fb_agno)
3499 				prevbno = NULLFSBLOCK;
3500 		}
3501 		/*
3502 		 * No previous block or can't follow it, just default.
3503 		 */
3504 		else
3505 			prevbno = NULLFSBLOCK;
3506 		/*
3507 		 * If there's a following (right) block, select a requested
3508 		 * start block based on it.
3509 		 */
3510 		if (!isnullstartblock(ap->got.br_startblock)) {
3511 			/*
3512 			 * Calculate gap to start of next block.
3513 			 */
3514 			adjust = gotdiff = ap->got.br_startoff - ap->offset;
3515 			/*
3516 			 * Figure the startblock based on the next block's
3517 			 * start and the gap size.
3518 			 */
3519 			gotbno = ap->got.br_startblock;
3520 			/*
3521 			 * Heuristic!
3522 			 * If the gap is large relative to the piece we're
3523 			 * allocating, or using it gives us an invalid block
3524 			 * number, then just use the start of the next block
3525 			 * offset by our length.
3526 			 */
3527 			if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3528 			    ISVALID(gotbno - gotdiff, gotbno))
3529 				gotbno -= adjust;
3530 			else if (ISVALID(gotbno - ap->length, gotbno)) {
3531 				gotbno -= ap->length;
3532 				gotdiff += adjust - ap->length;
3533 			} else
3534 				gotdiff += adjust;
3535 			/*
3536 			 * If the firstblock forbids it, can't use it,
3537 			 * must use default.
3538 			 */
3539 			if (!rt && !nullfb &&
3540 			    XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3541 				gotbno = NULLFSBLOCK;
3542 		}
3543 		/*
3544 		 * No next block, just default.
3545 		 */
3546 		else
3547 			gotbno = NULLFSBLOCK;
3548 		/*
3549 		 * If both valid, pick the better one, else the only good
3550 		 * one, else ap->blkno is already set (to 0 or the inode block).
3551 		 */
3552 		if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3553 			ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3554 		else if (prevbno != NULLFSBLOCK)
3555 			ap->blkno = prevbno;
3556 		else if (gotbno != NULLFSBLOCK)
3557 			ap->blkno = gotbno;
3558 	}
3559 #undef ISVALID
3560 }
3561 
3562 static int
xfs_bmap_longest_free_extent(struct xfs_trans * tp,xfs_agnumber_t ag,xfs_extlen_t * blen,int * notinit)3563 xfs_bmap_longest_free_extent(
3564 	struct xfs_trans	*tp,
3565 	xfs_agnumber_t		ag,
3566 	xfs_extlen_t		*blen,
3567 	int			*notinit)
3568 {
3569 	struct xfs_mount	*mp = tp->t_mountp;
3570 	struct xfs_perag	*pag;
3571 	xfs_extlen_t		longest;
3572 	int			error = 0;
3573 
3574 	pag = xfs_perag_get(mp, ag);
3575 	if (!pag->pagf_init) {
3576 		error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3577 		if (error)
3578 			goto out;
3579 
3580 		if (!pag->pagf_init) {
3581 			*notinit = 1;
3582 			goto out;
3583 		}
3584 	}
3585 
3586 	longest = xfs_alloc_longest_free_extent(mp, pag,
3587 				xfs_alloc_min_freelist(mp, pag),
3588 				xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3589 	if (*blen < longest)
3590 		*blen = longest;
3591 
3592 out:
3593 	xfs_perag_put(pag);
3594 	return error;
3595 }
3596 
3597 static void
xfs_bmap_select_minlen(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_extlen_t * blen,int notinit)3598 xfs_bmap_select_minlen(
3599 	struct xfs_bmalloca	*ap,
3600 	struct xfs_alloc_arg	*args,
3601 	xfs_extlen_t		*blen,
3602 	int			notinit)
3603 {
3604 	if (notinit || *blen < ap->minlen) {
3605 		/*
3606 		 * Since we did a BUF_TRYLOCK above, it is possible that
3607 		 * there is space for this request.
3608 		 */
3609 		args->minlen = ap->minlen;
3610 	} else if (*blen < args->maxlen) {
3611 		/*
3612 		 * If the best seen length is less than the request length,
3613 		 * use the best as the minimum.
3614 		 */
3615 		args->minlen = *blen;
3616 	} else {
3617 		/*
3618 		 * Otherwise we've seen an extent as big as maxlen, use that
3619 		 * as the minimum.
3620 		 */
3621 		args->minlen = args->maxlen;
3622 	}
3623 }
3624 
3625 STATIC int
xfs_bmap_btalloc_nullfb(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_extlen_t * blen)3626 xfs_bmap_btalloc_nullfb(
3627 	struct xfs_bmalloca	*ap,
3628 	struct xfs_alloc_arg	*args,
3629 	xfs_extlen_t		*blen)
3630 {
3631 	struct xfs_mount	*mp = ap->ip->i_mount;
3632 	xfs_agnumber_t		ag, startag;
3633 	int			notinit = 0;
3634 	int			error;
3635 
3636 	args->type = XFS_ALLOCTYPE_START_BNO;
3637 	args->total = ap->total;
3638 
3639 	startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3640 	if (startag == NULLAGNUMBER)
3641 		startag = ag = 0;
3642 
3643 	while (*blen < args->maxlen) {
3644 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3645 						     &notinit);
3646 		if (error)
3647 			return error;
3648 
3649 		if (++ag == mp->m_sb.sb_agcount)
3650 			ag = 0;
3651 		if (ag == startag)
3652 			break;
3653 	}
3654 
3655 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3656 	return 0;
3657 }
3658 
3659 STATIC int
xfs_bmap_btalloc_filestreams(struct xfs_bmalloca * ap,struct xfs_alloc_arg * args,xfs_extlen_t * blen)3660 xfs_bmap_btalloc_filestreams(
3661 	struct xfs_bmalloca	*ap,
3662 	struct xfs_alloc_arg	*args,
3663 	xfs_extlen_t		*blen)
3664 {
3665 	struct xfs_mount	*mp = ap->ip->i_mount;
3666 	xfs_agnumber_t		ag;
3667 	int			notinit = 0;
3668 	int			error;
3669 
3670 	args->type = XFS_ALLOCTYPE_NEAR_BNO;
3671 	args->total = ap->total;
3672 
3673 	ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3674 	if (ag == NULLAGNUMBER)
3675 		ag = 0;
3676 
3677 	error = xfs_bmap_longest_free_extent(args->tp, ag, blen, &notinit);
3678 	if (error)
3679 		return error;
3680 
3681 	if (*blen < args->maxlen) {
3682 		error = xfs_filestream_new_ag(ap, &ag);
3683 		if (error)
3684 			return error;
3685 
3686 		error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3687 						     &notinit);
3688 		if (error)
3689 			return error;
3690 
3691 	}
3692 
3693 	xfs_bmap_select_minlen(ap, args, blen, notinit);
3694 
3695 	/*
3696 	 * Set the failure fallback case to look in the selected AG as stream
3697 	 * may have moved.
3698 	 */
3699 	ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3700 	return 0;
3701 }
3702 
3703 STATIC int
xfs_bmap_btalloc(struct xfs_bmalloca * ap)3704 xfs_bmap_btalloc(
3705 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
3706 {
3707 	xfs_mount_t	*mp;		/* mount point structure */
3708 	xfs_alloctype_t	atype = 0;	/* type for allocation routines */
3709 	xfs_extlen_t	align = 0;	/* minimum allocation alignment */
3710 	xfs_agnumber_t	fb_agno;	/* ag number of ap->firstblock */
3711 	xfs_agnumber_t	ag;
3712 	xfs_alloc_arg_t	args;
3713 	xfs_extlen_t	blen;
3714 	xfs_extlen_t	nextminlen = 0;
3715 	int		nullfb;		/* true if ap->firstblock isn't set */
3716 	int		isaligned;
3717 	int		tryagain;
3718 	int		error;
3719 	int		stripe_align;
3720 
3721 	ASSERT(ap->length);
3722 
3723 	mp = ap->ip->i_mount;
3724 
3725 	/* stripe alignment for allocation is determined by mount parameters */
3726 	stripe_align = 0;
3727 	if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3728 		stripe_align = mp->m_swidth;
3729 	else if (mp->m_dalign)
3730 		stripe_align = mp->m_dalign;
3731 
3732 	if (ap->flags & XFS_BMAPI_COWFORK)
3733 		align = xfs_get_cowextsz_hint(ap->ip);
3734 	else if (xfs_alloc_is_userdata(ap->datatype))
3735 		align = xfs_get_extsz_hint(ap->ip);
3736 	if (align) {
3737 		error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3738 						align, 0, ap->eof, 0, ap->conv,
3739 						&ap->offset, &ap->length);
3740 		ASSERT(!error);
3741 		ASSERT(ap->length);
3742 	}
3743 
3744 
3745 	nullfb = *ap->firstblock == NULLFSBLOCK;
3746 	fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3747 	if (nullfb) {
3748 		if (xfs_alloc_is_userdata(ap->datatype) &&
3749 		    xfs_inode_is_filestream(ap->ip)) {
3750 			ag = xfs_filestream_lookup_ag(ap->ip);
3751 			ag = (ag != NULLAGNUMBER) ? ag : 0;
3752 			ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3753 		} else {
3754 			ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3755 		}
3756 	} else
3757 		ap->blkno = *ap->firstblock;
3758 
3759 	xfs_bmap_adjacent(ap);
3760 
3761 	/*
3762 	 * If allowed, use ap->blkno; otherwise must use firstblock since
3763 	 * it's in the right allocation group.
3764 	 */
3765 	if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3766 		;
3767 	else
3768 		ap->blkno = *ap->firstblock;
3769 	/*
3770 	 * Normal allocation, done through xfs_alloc_vextent.
3771 	 */
3772 	tryagain = isaligned = 0;
3773 	memset(&args, 0, sizeof(args));
3774 	args.tp = ap->tp;
3775 	args.mp = mp;
3776 	args.fsbno = ap->blkno;
3777 	xfs_rmap_skip_owner_update(&args.oinfo);
3778 
3779 	/* Trim the allocation back to the maximum an AG can fit. */
3780 	args.maxlen = MIN(ap->length, mp->m_ag_max_usable);
3781 	args.firstblock = *ap->firstblock;
3782 	blen = 0;
3783 	if (nullfb) {
3784 		/*
3785 		 * Search for an allocation group with a single extent large
3786 		 * enough for the request.  If one isn't found, then adjust
3787 		 * the minimum allocation size to the largest space found.
3788 		 */
3789 		if (xfs_alloc_is_userdata(ap->datatype) &&
3790 		    xfs_inode_is_filestream(ap->ip))
3791 			error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3792 		else
3793 			error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3794 		if (error)
3795 			return error;
3796 	} else if (ap->dfops->dop_low) {
3797 		if (xfs_inode_is_filestream(ap->ip))
3798 			args.type = XFS_ALLOCTYPE_FIRST_AG;
3799 		else
3800 			args.type = XFS_ALLOCTYPE_START_BNO;
3801 		args.total = args.minlen = ap->minlen;
3802 	} else {
3803 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
3804 		args.total = ap->total;
3805 		args.minlen = ap->minlen;
3806 	}
3807 	/* apply extent size hints if obtained earlier */
3808 	if (align) {
3809 		args.prod = align;
3810 		if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3811 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3812 	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3813 		args.prod = 1;
3814 		args.mod = 0;
3815 	} else {
3816 		args.prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3817 		if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3818 			args.mod = (xfs_extlen_t)(args.prod - args.mod);
3819 	}
3820 	/*
3821 	 * If we are not low on available data blocks, and the
3822 	 * underlying logical volume manager is a stripe, and
3823 	 * the file offset is zero then try to allocate data
3824 	 * blocks on stripe unit boundary.
3825 	 * NOTE: ap->aeof is only set if the allocation length
3826 	 * is >= the stripe unit and the allocation offset is
3827 	 * at the end of file.
3828 	 */
3829 	if (!ap->dfops->dop_low && ap->aeof) {
3830 		if (!ap->offset) {
3831 			args.alignment = stripe_align;
3832 			atype = args.type;
3833 			isaligned = 1;
3834 			/*
3835 			 * Adjust for alignment
3836 			 */
3837 			if (blen > args.alignment && blen <= args.maxlen)
3838 				args.minlen = blen - args.alignment;
3839 			args.minalignslop = 0;
3840 		} else {
3841 			/*
3842 			 * First try an exact bno allocation.
3843 			 * If it fails then do a near or start bno
3844 			 * allocation with alignment turned on.
3845 			 */
3846 			atype = args.type;
3847 			tryagain = 1;
3848 			args.type = XFS_ALLOCTYPE_THIS_BNO;
3849 			args.alignment = 1;
3850 			/*
3851 			 * Compute the minlen+alignment for the
3852 			 * next case.  Set slop so that the value
3853 			 * of minlen+alignment+slop doesn't go up
3854 			 * between the calls.
3855 			 */
3856 			if (blen > stripe_align && blen <= args.maxlen)
3857 				nextminlen = blen - stripe_align;
3858 			else
3859 				nextminlen = args.minlen;
3860 			if (nextminlen + stripe_align > args.minlen + 1)
3861 				args.minalignslop =
3862 					nextminlen + stripe_align -
3863 					args.minlen - 1;
3864 			else
3865 				args.minalignslop = 0;
3866 		}
3867 	} else {
3868 		args.alignment = 1;
3869 		args.minalignslop = 0;
3870 	}
3871 	args.minleft = ap->minleft;
3872 	args.wasdel = ap->wasdel;
3873 	args.resv = XFS_AG_RESV_NONE;
3874 	args.datatype = ap->datatype;
3875 	if (ap->datatype & XFS_ALLOC_USERDATA_ZERO)
3876 		args.ip = ap->ip;
3877 
3878 	error = xfs_alloc_vextent(&args);
3879 	if (error)
3880 		return error;
3881 
3882 	if (tryagain && args.fsbno == NULLFSBLOCK) {
3883 		/*
3884 		 * Exact allocation failed. Now try with alignment
3885 		 * turned on.
3886 		 */
3887 		args.type = atype;
3888 		args.fsbno = ap->blkno;
3889 		args.alignment = stripe_align;
3890 		args.minlen = nextminlen;
3891 		args.minalignslop = 0;
3892 		isaligned = 1;
3893 		if ((error = xfs_alloc_vextent(&args)))
3894 			return error;
3895 	}
3896 	if (isaligned && args.fsbno == NULLFSBLOCK) {
3897 		/*
3898 		 * allocation failed, so turn off alignment and
3899 		 * try again.
3900 		 */
3901 		args.type = atype;
3902 		args.fsbno = ap->blkno;
3903 		args.alignment = 0;
3904 		if ((error = xfs_alloc_vextent(&args)))
3905 			return error;
3906 	}
3907 	if (args.fsbno == NULLFSBLOCK && nullfb &&
3908 	    args.minlen > ap->minlen) {
3909 		args.minlen = ap->minlen;
3910 		args.type = XFS_ALLOCTYPE_START_BNO;
3911 		args.fsbno = ap->blkno;
3912 		if ((error = xfs_alloc_vextent(&args)))
3913 			return error;
3914 	}
3915 	if (args.fsbno == NULLFSBLOCK && nullfb) {
3916 		args.fsbno = 0;
3917 		args.type = XFS_ALLOCTYPE_FIRST_AG;
3918 		args.total = ap->minlen;
3919 		if ((error = xfs_alloc_vextent(&args)))
3920 			return error;
3921 		ap->dfops->dop_low = true;
3922 	}
3923 	if (args.fsbno != NULLFSBLOCK) {
3924 		/*
3925 		 * check the allocation happened at the same or higher AG than
3926 		 * the first block that was allocated.
3927 		 */
3928 		ASSERT(*ap->firstblock == NULLFSBLOCK ||
3929 		       XFS_FSB_TO_AGNO(mp, *ap->firstblock) <=
3930 		       XFS_FSB_TO_AGNO(mp, args.fsbno));
3931 
3932 		ap->blkno = args.fsbno;
3933 		if (*ap->firstblock == NULLFSBLOCK)
3934 			*ap->firstblock = args.fsbno;
3935 		ASSERT(nullfb || fb_agno <= args.agno);
3936 		ap->length = args.len;
3937 		if (!(ap->flags & XFS_BMAPI_COWFORK))
3938 			ap->ip->i_d.di_nblocks += args.len;
3939 		xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3940 		if (ap->wasdel)
3941 			ap->ip->i_delayed_blks -= args.len;
3942 		/*
3943 		 * Adjust the disk quota also. This was reserved
3944 		 * earlier.
3945 		 */
3946 		xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3947 			ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3948 					XFS_TRANS_DQ_BCOUNT,
3949 			(long) args.len);
3950 	} else {
3951 		ap->blkno = NULLFSBLOCK;
3952 		ap->length = 0;
3953 	}
3954 	return 0;
3955 }
3956 
3957 /*
3958  * For a remap operation, just "allocate" an extent at the address that the
3959  * caller passed in, and ensure that the AGFL is the right size.  The caller
3960  * will then map the "allocated" extent into the file somewhere.
3961  */
3962 STATIC int
xfs_bmap_remap_alloc(struct xfs_bmalloca * ap)3963 xfs_bmap_remap_alloc(
3964 	struct xfs_bmalloca	*ap)
3965 {
3966 	struct xfs_trans	*tp = ap->tp;
3967 	struct xfs_mount	*mp = tp->t_mountp;
3968 	xfs_fsblock_t		bno;
3969 	struct xfs_alloc_arg	args;
3970 	int			error;
3971 
3972 	/*
3973 	 * validate that the block number is legal - the enables us to detect
3974 	 * and handle a silent filesystem corruption rather than crashing.
3975 	 */
3976 	memset(&args, 0, sizeof(struct xfs_alloc_arg));
3977 	args.tp = ap->tp;
3978 	args.mp = ap->tp->t_mountp;
3979 	bno = *ap->firstblock;
3980 	args.agno = XFS_FSB_TO_AGNO(mp, bno);
3981 	args.agbno = XFS_FSB_TO_AGBNO(mp, bno);
3982 	if (args.agno >= mp->m_sb.sb_agcount ||
3983 	    args.agbno >= mp->m_sb.sb_agblocks)
3984 		return -EFSCORRUPTED;
3985 
3986 	/* "Allocate" the extent from the range we passed in. */
3987 	trace_xfs_bmap_remap_alloc(ap->ip, *ap->firstblock, ap->length);
3988 	ap->blkno = bno;
3989 	ap->ip->i_d.di_nblocks += ap->length;
3990 	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3991 
3992 	/* Fix the freelist, like a real allocator does. */
3993 	args.datatype = ap->datatype;
3994 	args.pag = xfs_perag_get(args.mp, args.agno);
3995 	ASSERT(args.pag);
3996 
3997 	/*
3998 	 * The freelist fixing code will decline the allocation if
3999 	 * the size and shape of the free space doesn't allow for
4000 	 * allocating the extent and updating all the metadata that
4001 	 * happens during an allocation.  We're remapping, not
4002 	 * allocating, so skip that check by pretending to be freeing.
4003 	 */
4004 	error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING);
4005 	xfs_perag_put(args.pag);
4006 	if (error)
4007 		trace_xfs_bmap_remap_alloc_error(ap->ip, error, _RET_IP_);
4008 	return error;
4009 }
4010 
4011 /*
4012  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
4013  * It figures out where to ask the underlying allocator to put the new extent.
4014  */
4015 STATIC int
xfs_bmap_alloc(struct xfs_bmalloca * ap)4016 xfs_bmap_alloc(
4017 	struct xfs_bmalloca	*ap)	/* bmap alloc argument struct */
4018 {
4019 	if (ap->flags & XFS_BMAPI_REMAP)
4020 		return xfs_bmap_remap_alloc(ap);
4021 	if (XFS_IS_REALTIME_INODE(ap->ip) &&
4022 	    xfs_alloc_is_userdata(ap->datatype))
4023 		return xfs_bmap_rtalloc(ap);
4024 	return xfs_bmap_btalloc(ap);
4025 }
4026 
4027 /* Trim extent to fit a logical block range. */
4028 void
xfs_trim_extent(struct xfs_bmbt_irec * irec,xfs_fileoff_t bno,xfs_filblks_t len)4029 xfs_trim_extent(
4030 	struct xfs_bmbt_irec	*irec,
4031 	xfs_fileoff_t		bno,
4032 	xfs_filblks_t		len)
4033 {
4034 	xfs_fileoff_t		distance;
4035 	xfs_fileoff_t		end = bno + len;
4036 
4037 	if (irec->br_startoff + irec->br_blockcount <= bno ||
4038 	    irec->br_startoff >= end) {
4039 		irec->br_blockcount = 0;
4040 		return;
4041 	}
4042 
4043 	if (irec->br_startoff < bno) {
4044 		distance = bno - irec->br_startoff;
4045 		if (isnullstartblock(irec->br_startblock))
4046 			irec->br_startblock = DELAYSTARTBLOCK;
4047 		if (irec->br_startblock != DELAYSTARTBLOCK &&
4048 		    irec->br_startblock != HOLESTARTBLOCK)
4049 			irec->br_startblock += distance;
4050 		irec->br_startoff += distance;
4051 		irec->br_blockcount -= distance;
4052 	}
4053 
4054 	if (end < irec->br_startoff + irec->br_blockcount) {
4055 		distance = irec->br_startoff + irec->br_blockcount - end;
4056 		irec->br_blockcount -= distance;
4057 	}
4058 }
4059 
4060 /* trim extent to within eof */
4061 void
xfs_trim_extent_eof(struct xfs_bmbt_irec * irec,struct xfs_inode * ip)4062 xfs_trim_extent_eof(
4063 	struct xfs_bmbt_irec	*irec,
4064 	struct xfs_inode	*ip)
4065 
4066 {
4067 	xfs_trim_extent(irec, 0, XFS_B_TO_FSB(ip->i_mount,
4068 					      i_size_read(VFS_I(ip))));
4069 }
4070 
4071 /*
4072  * Trim the returned map to the required bounds
4073  */
4074 STATIC void
xfs_bmapi_trim_map(struct xfs_bmbt_irec * mval,struct xfs_bmbt_irec * got,xfs_fileoff_t * bno,xfs_filblks_t len,xfs_fileoff_t obno,xfs_fileoff_t end,int n,int flags)4075 xfs_bmapi_trim_map(
4076 	struct xfs_bmbt_irec	*mval,
4077 	struct xfs_bmbt_irec	*got,
4078 	xfs_fileoff_t		*bno,
4079 	xfs_filblks_t		len,
4080 	xfs_fileoff_t		obno,
4081 	xfs_fileoff_t		end,
4082 	int			n,
4083 	int			flags)
4084 {
4085 	if ((flags & XFS_BMAPI_ENTIRE) ||
4086 	    got->br_startoff + got->br_blockcount <= obno) {
4087 		*mval = *got;
4088 		if (isnullstartblock(got->br_startblock))
4089 			mval->br_startblock = DELAYSTARTBLOCK;
4090 		return;
4091 	}
4092 
4093 	if (obno > *bno)
4094 		*bno = obno;
4095 	ASSERT((*bno >= obno) || (n == 0));
4096 	ASSERT(*bno < end);
4097 	mval->br_startoff = *bno;
4098 	if (isnullstartblock(got->br_startblock))
4099 		mval->br_startblock = DELAYSTARTBLOCK;
4100 	else
4101 		mval->br_startblock = got->br_startblock +
4102 					(*bno - got->br_startoff);
4103 	/*
4104 	 * Return the minimum of what we got and what we asked for for
4105 	 * the length.  We can use the len variable here because it is
4106 	 * modified below and we could have been there before coming
4107 	 * here if the first part of the allocation didn't overlap what
4108 	 * was asked for.
4109 	 */
4110 	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4111 			got->br_blockcount - (*bno - got->br_startoff));
4112 	mval->br_state = got->br_state;
4113 	ASSERT(mval->br_blockcount <= len);
4114 	return;
4115 }
4116 
4117 /*
4118  * Update and validate the extent map to return
4119  */
4120 STATIC void
xfs_bmapi_update_map(struct xfs_bmbt_irec ** map,xfs_fileoff_t * bno,xfs_filblks_t * len,xfs_fileoff_t obno,xfs_fileoff_t end,int * n,int flags)4121 xfs_bmapi_update_map(
4122 	struct xfs_bmbt_irec	**map,
4123 	xfs_fileoff_t		*bno,
4124 	xfs_filblks_t		*len,
4125 	xfs_fileoff_t		obno,
4126 	xfs_fileoff_t		end,
4127 	int			*n,
4128 	int			flags)
4129 {
4130 	xfs_bmbt_irec_t	*mval = *map;
4131 
4132 	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4133 	       ((mval->br_startoff + mval->br_blockcount) <= end));
4134 	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4135 	       (mval->br_startoff < obno));
4136 
4137 	*bno = mval->br_startoff + mval->br_blockcount;
4138 	*len = end - *bno;
4139 	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4140 		/* update previous map with new information */
4141 		ASSERT(mval->br_startblock == mval[-1].br_startblock);
4142 		ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4143 		ASSERT(mval->br_state == mval[-1].br_state);
4144 		mval[-1].br_blockcount = mval->br_blockcount;
4145 		mval[-1].br_state = mval->br_state;
4146 	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4147 		   mval[-1].br_startblock != DELAYSTARTBLOCK &&
4148 		   mval[-1].br_startblock != HOLESTARTBLOCK &&
4149 		   mval->br_startblock == mval[-1].br_startblock +
4150 					  mval[-1].br_blockcount &&
4151 		   ((flags & XFS_BMAPI_IGSTATE) ||
4152 			mval[-1].br_state == mval->br_state)) {
4153 		ASSERT(mval->br_startoff ==
4154 		       mval[-1].br_startoff + mval[-1].br_blockcount);
4155 		mval[-1].br_blockcount += mval->br_blockcount;
4156 	} else if (*n > 0 &&
4157 		   mval->br_startblock == DELAYSTARTBLOCK &&
4158 		   mval[-1].br_startblock == DELAYSTARTBLOCK &&
4159 		   mval->br_startoff ==
4160 		   mval[-1].br_startoff + mval[-1].br_blockcount) {
4161 		mval[-1].br_blockcount += mval->br_blockcount;
4162 		mval[-1].br_state = mval->br_state;
4163 	} else if (!((*n == 0) &&
4164 		     ((mval->br_startoff + mval->br_blockcount) <=
4165 		      obno))) {
4166 		mval++;
4167 		(*n)++;
4168 	}
4169 	*map = mval;
4170 }
4171 
4172 /*
4173  * Map file blocks to filesystem blocks without allocation.
4174  */
4175 int
xfs_bmapi_read(struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,struct xfs_bmbt_irec * mval,int * nmap,int flags)4176 xfs_bmapi_read(
4177 	struct xfs_inode	*ip,
4178 	xfs_fileoff_t		bno,
4179 	xfs_filblks_t		len,
4180 	struct xfs_bmbt_irec	*mval,
4181 	int			*nmap,
4182 	int			flags)
4183 {
4184 	struct xfs_mount	*mp = ip->i_mount;
4185 	struct xfs_ifork	*ifp;
4186 	struct xfs_bmbt_irec	got;
4187 	struct xfs_bmbt_irec	prev;
4188 	xfs_fileoff_t		obno;
4189 	xfs_fileoff_t		end;
4190 	xfs_extnum_t		lastx;
4191 	int			error;
4192 	int			eof;
4193 	int			n = 0;
4194 	int			whichfork = xfs_bmapi_whichfork(flags);
4195 
4196 	ASSERT(*nmap >= 1);
4197 	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4198 			   XFS_BMAPI_IGSTATE|XFS_BMAPI_COWFORK)));
4199 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4200 
4201 	if (unlikely(XFS_TEST_ERROR(
4202 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4203 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4204 	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4205 		XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4206 		return -EFSCORRUPTED;
4207 	}
4208 
4209 	if (XFS_FORCED_SHUTDOWN(mp))
4210 		return -EIO;
4211 
4212 	XFS_STATS_INC(mp, xs_blk_mapr);
4213 
4214 	ifp = XFS_IFORK_PTR(ip, whichfork);
4215 
4216 	/* No CoW fork?  Return a hole. */
4217 	if (whichfork == XFS_COW_FORK && !ifp) {
4218 		mval->br_startoff = bno;
4219 		mval->br_startblock = HOLESTARTBLOCK;
4220 		mval->br_blockcount = len;
4221 		mval->br_state = XFS_EXT_NORM;
4222 		*nmap = 1;
4223 		return 0;
4224 	}
4225 
4226 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4227 		error = xfs_iread_extents(NULL, ip, whichfork);
4228 		if (error)
4229 			return error;
4230 	}
4231 
4232 	xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4233 	end = bno + len;
4234 	obno = bno;
4235 
4236 	while (bno < end && n < *nmap) {
4237 		/* Reading past eof, act as though there's a hole up to end. */
4238 		if (eof)
4239 			got.br_startoff = end;
4240 		if (got.br_startoff > bno) {
4241 			/* Reading in a hole.  */
4242 			mval->br_startoff = bno;
4243 			mval->br_startblock = HOLESTARTBLOCK;
4244 			mval->br_blockcount =
4245 				XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4246 			mval->br_state = XFS_EXT_NORM;
4247 			bno += mval->br_blockcount;
4248 			len -= mval->br_blockcount;
4249 			mval++;
4250 			n++;
4251 			continue;
4252 		}
4253 
4254 		/* set up the extent map to return. */
4255 		xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4256 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4257 
4258 		/* If we're done, stop now. */
4259 		if (bno >= end || n >= *nmap)
4260 			break;
4261 
4262 		/* Else go on to the next record. */
4263 		if (++lastx < xfs_iext_count(ifp))
4264 			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4265 		else
4266 			eof = 1;
4267 	}
4268 	*nmap = n;
4269 	return 0;
4270 }
4271 
4272 /*
4273  * Add a delayed allocation extent to an inode. Blocks are reserved from the
4274  * global pool and the extent inserted into the inode in-core extent tree.
4275  *
4276  * On entry, got refers to the first extent beyond the offset of the extent to
4277  * allocate or eof is specified if no such extent exists. On return, got refers
4278  * to the extent record that was inserted to the inode fork.
4279  *
4280  * Note that the allocated extent may have been merged with contiguous extents
4281  * during insertion into the inode fork. Thus, got does not reflect the current
4282  * state of the inode fork on return. If necessary, the caller can use lastx to
4283  * look up the updated record in the inode fork.
4284  */
4285 int
xfs_bmapi_reserve_delalloc(struct xfs_inode * ip,int whichfork,xfs_fileoff_t off,xfs_filblks_t len,xfs_filblks_t prealloc,struct xfs_bmbt_irec * got,xfs_extnum_t * lastx,int eof)4286 xfs_bmapi_reserve_delalloc(
4287 	struct xfs_inode	*ip,
4288 	int			whichfork,
4289 	xfs_fileoff_t		off,
4290 	xfs_filblks_t		len,
4291 	xfs_filblks_t		prealloc,
4292 	struct xfs_bmbt_irec	*got,
4293 	xfs_extnum_t		*lastx,
4294 	int			eof)
4295 {
4296 	struct xfs_mount	*mp = ip->i_mount;
4297 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4298 	xfs_extlen_t		alen;
4299 	xfs_extlen_t		indlen;
4300 	char			rt = XFS_IS_REALTIME_INODE(ip);
4301 	xfs_extlen_t		extsz;
4302 	int			error;
4303 	xfs_fileoff_t		aoff = off;
4304 
4305 	/*
4306 	 * Cap the alloc length. Keep track of prealloc so we know whether to
4307 	 * tag the inode before we return.
4308 	 */
4309 	alen = XFS_FILBLKS_MIN(len + prealloc, MAXEXTLEN);
4310 	if (!eof)
4311 		alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4312 	if (prealloc && alen >= len)
4313 		prealloc = alen - len;
4314 
4315 	/* Figure out the extent size, adjust alen */
4316 	if (whichfork == XFS_COW_FORK)
4317 		extsz = xfs_get_cowextsz_hint(ip);
4318 	else
4319 		extsz = xfs_get_extsz_hint(ip);
4320 	if (extsz) {
4321 		struct xfs_bmbt_irec	prev;
4322 
4323 		if (!xfs_iext_get_extent(ifp, *lastx - 1, &prev))
4324 			prev.br_startoff = NULLFILEOFF;
4325 
4326 		error = xfs_bmap_extsize_align(mp, got, &prev, extsz, rt, eof,
4327 					       1, 0, &aoff, &alen);
4328 		ASSERT(!error);
4329 	}
4330 
4331 	if (rt)
4332 		extsz = alen / mp->m_sb.sb_rextsize;
4333 
4334 	/*
4335 	 * Make a transaction-less quota reservation for delayed allocation
4336 	 * blocks.  This number gets adjusted later.  We return if we haven't
4337 	 * allocated blocks already inside this loop.
4338 	 */
4339 	error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4340 			rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4341 	if (error)
4342 		return error;
4343 
4344 	/*
4345 	 * Split changing sb for alen and indlen since they could be coming
4346 	 * from different places.
4347 	 */
4348 	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4349 	ASSERT(indlen > 0);
4350 
4351 	if (rt) {
4352 		error = xfs_mod_frextents(mp, -((int64_t)extsz));
4353 	} else {
4354 		error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4355 	}
4356 
4357 	if (error)
4358 		goto out_unreserve_quota;
4359 
4360 	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4361 	if (error)
4362 		goto out_unreserve_blocks;
4363 
4364 
4365 	ip->i_delayed_blks += alen;
4366 
4367 	got->br_startoff = aoff;
4368 	got->br_startblock = nullstartblock(indlen);
4369 	got->br_blockcount = alen;
4370 	got->br_state = XFS_EXT_NORM;
4371 
4372 	xfs_bmap_add_extent_hole_delay(ip, whichfork, lastx, got);
4373 
4374 	/*
4375 	 * Tag the inode if blocks were preallocated. Note that COW fork
4376 	 * preallocation can occur at the start or end of the extent, even when
4377 	 * prealloc == 0, so we must also check the aligned offset and length.
4378 	 */
4379 	if (whichfork == XFS_DATA_FORK && prealloc)
4380 		xfs_inode_set_eofblocks_tag(ip);
4381 	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4382 		xfs_inode_set_cowblocks_tag(ip);
4383 
4384 	return 0;
4385 
4386 out_unreserve_blocks:
4387 	if (rt)
4388 		xfs_mod_frextents(mp, extsz);
4389 	else
4390 		xfs_mod_fdblocks(mp, alen, false);
4391 out_unreserve_quota:
4392 	if (XFS_IS_QUOTA_ON(mp))
4393 		xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4394 				XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4395 	return error;
4396 }
4397 
4398 static int
xfs_bmapi_allocate(struct xfs_bmalloca * bma)4399 xfs_bmapi_allocate(
4400 	struct xfs_bmalloca	*bma)
4401 {
4402 	struct xfs_mount	*mp = bma->ip->i_mount;
4403 	int			whichfork = xfs_bmapi_whichfork(bma->flags);
4404 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4405 	int			tmp_logflags = 0;
4406 	int			error;
4407 
4408 	ASSERT(bma->length > 0);
4409 
4410 	/*
4411 	 * For the wasdelay case, we could also just allocate the stuff asked
4412 	 * for in this bmap call but that wouldn't be as good.
4413 	 */
4414 	if (bma->wasdel) {
4415 		bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4416 		bma->offset = bma->got.br_startoff;
4417 		if (bma->idx != NULLEXTNUM && bma->idx) {
4418 			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4419 					 &bma->prev);
4420 		}
4421 	} else {
4422 		bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4423 		if (!bma->eof)
4424 			bma->length = XFS_FILBLKS_MIN(bma->length,
4425 					bma->got.br_startoff - bma->offset);
4426 	}
4427 
4428 	/*
4429 	 * Set the data type being allocated. For the data fork, the first data
4430 	 * in the file is treated differently to all other allocations. For the
4431 	 * attribute fork, we only need to ensure the allocated range is not on
4432 	 * the busy list.
4433 	 */
4434 	if (!(bma->flags & XFS_BMAPI_METADATA)) {
4435 		bma->datatype = XFS_ALLOC_NOBUSY;
4436 		if (whichfork == XFS_DATA_FORK) {
4437 			if (bma->offset == 0)
4438 				bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4439 			else
4440 				bma->datatype |= XFS_ALLOC_USERDATA;
4441 		}
4442 		if (bma->flags & XFS_BMAPI_ZERO)
4443 			bma->datatype |= XFS_ALLOC_USERDATA_ZERO;
4444 	}
4445 
4446 	bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4447 
4448 	/*
4449 	 * Only want to do the alignment at the eof if it is userdata and
4450 	 * allocation length is larger than a stripe unit.
4451 	 */
4452 	if (mp->m_dalign && bma->length >= mp->m_dalign &&
4453 	    !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4454 		error = xfs_bmap_isaeof(bma, whichfork);
4455 		if (error)
4456 			return error;
4457 	}
4458 
4459 	error = xfs_bmap_alloc(bma);
4460 	if (error)
4461 		return error;
4462 
4463 	if (bma->cur)
4464 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4465 	if (bma->blkno == NULLFSBLOCK)
4466 		return 0;
4467 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4468 		bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4469 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4470 		bma->cur->bc_private.b.dfops = bma->dfops;
4471 	}
4472 	/*
4473 	 * Bump the number of extents we've allocated
4474 	 * in this call.
4475 	 */
4476 	bma->nallocs++;
4477 
4478 	if (bma->cur)
4479 		bma->cur->bc_private.b.flags =
4480 			bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4481 
4482 	bma->got.br_startoff = bma->offset;
4483 	bma->got.br_startblock = bma->blkno;
4484 	bma->got.br_blockcount = bma->length;
4485 	bma->got.br_state = XFS_EXT_NORM;
4486 
4487 	/*
4488 	 * In the data fork, a wasdelay extent has been initialized, so
4489 	 * shouldn't be flagged as unwritten.
4490 	 *
4491 	 * For the cow fork, however, we convert delalloc reservations
4492 	 * (extents allocated for speculative preallocation) to
4493 	 * allocated unwritten extents, and only convert the unwritten
4494 	 * extents to real extents when we're about to write the data.
4495 	 */
4496 	if ((!bma->wasdel || (bma->flags & XFS_BMAPI_COWFORK)) &&
4497 	    (bma->flags & XFS_BMAPI_PREALLOC) &&
4498 	    xfs_sb_version_hasextflgbit(&mp->m_sb))
4499 		bma->got.br_state = XFS_EXT_UNWRITTEN;
4500 
4501 	if (bma->wasdel)
4502 		error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4503 	else
4504 		error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4505 
4506 	bma->logflags |= tmp_logflags;
4507 	if (error)
4508 		return error;
4509 
4510 	/*
4511 	 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4512 	 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4513 	 * the neighbouring ones.
4514 	 */
4515 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4516 
4517 	ASSERT(bma->got.br_startoff <= bma->offset);
4518 	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4519 	       bma->offset + bma->length);
4520 	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4521 	       bma->got.br_state == XFS_EXT_UNWRITTEN);
4522 	return 0;
4523 }
4524 
4525 STATIC int
xfs_bmapi_convert_unwritten(struct xfs_bmalloca * bma,struct xfs_bmbt_irec * mval,xfs_filblks_t len,int flags)4526 xfs_bmapi_convert_unwritten(
4527 	struct xfs_bmalloca	*bma,
4528 	struct xfs_bmbt_irec	*mval,
4529 	xfs_filblks_t		len,
4530 	int			flags)
4531 {
4532 	int			whichfork = xfs_bmapi_whichfork(flags);
4533 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4534 	int			tmp_logflags = 0;
4535 	int			error;
4536 
4537 	/* check if we need to do unwritten->real conversion */
4538 	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4539 	    (flags & XFS_BMAPI_PREALLOC))
4540 		return 0;
4541 
4542 	/* check if we need to do real->unwritten conversion */
4543 	if (mval->br_state == XFS_EXT_NORM &&
4544 	    (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4545 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4546 		return 0;
4547 
4548 	/*
4549 	 * Modify (by adding) the state flag, if writing.
4550 	 */
4551 	ASSERT(mval->br_blockcount <= len);
4552 	if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4553 		bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4554 					bma->ip, whichfork);
4555 		bma->cur->bc_private.b.firstblock = *bma->firstblock;
4556 		bma->cur->bc_private.b.dfops = bma->dfops;
4557 	}
4558 	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4559 				? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4560 
4561 	/*
4562 	 * Before insertion into the bmbt, zero the range being converted
4563 	 * if required.
4564 	 */
4565 	if (flags & XFS_BMAPI_ZERO) {
4566 		error = xfs_zero_extent(bma->ip, mval->br_startblock,
4567 					mval->br_blockcount);
4568 		if (error)
4569 			return error;
4570 	}
4571 
4572 	error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, whichfork,
4573 			&bma->idx, &bma->cur, mval, bma->firstblock, bma->dfops,
4574 			&tmp_logflags);
4575 	/*
4576 	 * Log the inode core unconditionally in the unwritten extent conversion
4577 	 * path because the conversion might not have done so (e.g., if the
4578 	 * extent count hasn't changed). We need to make sure the inode is dirty
4579 	 * in the transaction for the sake of fsync(), even if nothing has
4580 	 * changed, because fsync() will not force the log for this transaction
4581 	 * unless it sees the inode pinned.
4582 	 *
4583 	 * Note: If we're only converting cow fork extents, there aren't
4584 	 * any on-disk updates to make, so we don't need to log anything.
4585 	 */
4586 	if (whichfork != XFS_COW_FORK)
4587 		bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4588 	if (error)
4589 		return error;
4590 
4591 	/*
4592 	 * Update our extent pointer, given that
4593 	 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4594 	 * of the neighbouring ones.
4595 	 */
4596 	xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4597 
4598 	/*
4599 	 * We may have combined previously unwritten space with written space,
4600 	 * so generate another request.
4601 	 */
4602 	if (mval->br_blockcount < len)
4603 		return -EAGAIN;
4604 	return 0;
4605 }
4606 
4607 /*
4608  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4609  * extent state if necessary.  Details behaviour is controlled by the flags
4610  * parameter.  Only allocates blocks from a single allocation group, to avoid
4611  * locking problems.
4612  *
4613  * The returned value in "firstblock" from the first call in a transaction
4614  * must be remembered and presented to subsequent calls in "firstblock".
4615  * An upper bound for the number of blocks to be allocated is supplied to
4616  * the first call in "total"; if no allocation group has that many free
4617  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4618  */
4619 int
xfs_bmapi_write(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,int flags,xfs_fsblock_t * firstblock,xfs_extlen_t total,struct xfs_bmbt_irec * mval,int * nmap,struct xfs_defer_ops * dfops)4620 xfs_bmapi_write(
4621 	struct xfs_trans	*tp,		/* transaction pointer */
4622 	struct xfs_inode	*ip,		/* incore inode */
4623 	xfs_fileoff_t		bno,		/* starting file offs. mapped */
4624 	xfs_filblks_t		len,		/* length to map in file */
4625 	int			flags,		/* XFS_BMAPI_... */
4626 	xfs_fsblock_t		*firstblock,	/* first allocated block
4627 						   controls a.g. for allocs */
4628 	xfs_extlen_t		total,		/* total blocks needed */
4629 	struct xfs_bmbt_irec	*mval,		/* output: map values */
4630 	int			*nmap,		/* i/o: mval size/count */
4631 	struct xfs_defer_ops	*dfops)		/* i/o: list extents to free */
4632 {
4633 	struct xfs_mount	*mp = ip->i_mount;
4634 	struct xfs_ifork	*ifp;
4635 	struct xfs_bmalloca	bma = { NULL };	/* args for xfs_bmap_alloc */
4636 	xfs_fileoff_t		end;		/* end of mapped file region */
4637 	int			eof;		/* after the end of extents */
4638 	int			error;		/* error return */
4639 	int			n;		/* current extent index */
4640 	xfs_fileoff_t		obno;		/* old block number (offset) */
4641 	int			whichfork;	/* data or attr fork */
4642 
4643 #ifdef DEBUG
4644 	xfs_fileoff_t		orig_bno;	/* original block number value */
4645 	int			orig_flags;	/* original flags arg value */
4646 	xfs_filblks_t		orig_len;	/* original value of len arg */
4647 	struct xfs_bmbt_irec	*orig_mval;	/* original value of mval */
4648 	int			orig_nmap;	/* original value of *nmap */
4649 
4650 	orig_bno = bno;
4651 	orig_len = len;
4652 	orig_flags = flags;
4653 	orig_mval = mval;
4654 	orig_nmap = *nmap;
4655 #endif
4656 	whichfork = xfs_bmapi_whichfork(flags);
4657 
4658 	ASSERT(*nmap >= 1);
4659 	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4660 	ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4661 	ASSERT(tp != NULL ||
4662 	       (flags & (XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK)) ==
4663 			(XFS_BMAPI_CONVERT | XFS_BMAPI_COWFORK));
4664 	ASSERT(len > 0);
4665 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4666 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4667 	ASSERT(!(flags & XFS_BMAPI_REMAP) || whichfork == XFS_DATA_FORK);
4668 	ASSERT(!(flags & XFS_BMAPI_PREALLOC) || !(flags & XFS_BMAPI_REMAP));
4669 	ASSERT(!(flags & XFS_BMAPI_CONVERT) || !(flags & XFS_BMAPI_REMAP));
4670 
4671 	/* zeroing is for currently only for data extents, not metadata */
4672 	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4673 			(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4674 	/*
4675 	 * we can allocate unwritten extents or pre-zero allocated blocks,
4676 	 * but it makes no sense to do both at once. This would result in
4677 	 * zeroing the unwritten extent twice, but it still being an
4678 	 * unwritten extent....
4679 	 */
4680 	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4681 			(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4682 
4683 	if (unlikely(XFS_TEST_ERROR(
4684 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4685 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4686 	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4687 		XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4688 		return -EFSCORRUPTED;
4689 	}
4690 
4691 	if (XFS_FORCED_SHUTDOWN(mp))
4692 		return -EIO;
4693 
4694 	ifp = XFS_IFORK_PTR(ip, whichfork);
4695 
4696 	XFS_STATS_INC(mp, xs_blk_mapw);
4697 
4698 	if (*firstblock == NULLFSBLOCK) {
4699 		if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4700 			bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4701 		else
4702 			bma.minleft = 1;
4703 	} else {
4704 		bma.minleft = 0;
4705 	}
4706 
4707 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4708 		error = xfs_iread_extents(tp, ip, whichfork);
4709 		if (error)
4710 			goto error0;
4711 	}
4712 
4713 	xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4714 				&bma.prev);
4715 	n = 0;
4716 	end = bno + len;
4717 	obno = bno;
4718 
4719 	bma.tp = tp;
4720 	bma.ip = ip;
4721 	bma.total = total;
4722 	bma.datatype = 0;
4723 	bma.dfops = dfops;
4724 	bma.firstblock = firstblock;
4725 
4726 	while (bno < end && n < *nmap) {
4727 		bool			need_alloc = false, wasdelay = false;
4728 
4729 		/* in hole or beyoned EOF? */
4730 		if (eof || bma.got.br_startoff > bno) {
4731 			if (flags & XFS_BMAPI_DELALLOC) {
4732 				/*
4733 				 * For the COW fork we can reasonably get a
4734 				 * request for converting an extent that races
4735 				 * with other threads already having converted
4736 				 * part of it, as there converting COW to
4737 				 * regular blocks is not protected using the
4738 				 * IOLOCK.
4739 				 */
4740 				ASSERT(flags & XFS_BMAPI_COWFORK);
4741 				if (!(flags & XFS_BMAPI_COWFORK)) {
4742 					error = -EIO;
4743 					goto error0;
4744 				}
4745 
4746 				if (eof || bno >= end)
4747 					break;
4748 			} else {
4749 				need_alloc = true;
4750 			}
4751 		} else {
4752 			/*
4753 			 * Make sure we only reflink into a hole.
4754 			 */
4755 			ASSERT(!(flags & XFS_BMAPI_REMAP));
4756 			if (isnullstartblock(bma.got.br_startblock))
4757 				wasdelay = true;
4758 		}
4759 
4760 		/*
4761 		 * First, deal with the hole before the allocated space
4762 		 * that we found, if any.
4763 		 */
4764 		if (need_alloc || wasdelay) {
4765 			bma.eof = eof;
4766 			bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4767 			bma.wasdel = wasdelay;
4768 			bma.offset = bno;
4769 			bma.flags = flags;
4770 
4771 			/*
4772 			 * There's a 32/64 bit type mismatch between the
4773 			 * allocation length request (which can be 64 bits in
4774 			 * length) and the bma length request, which is
4775 			 * xfs_extlen_t and therefore 32 bits. Hence we have to
4776 			 * check for 32-bit overflows and handle them here.
4777 			 */
4778 			if (len > (xfs_filblks_t)MAXEXTLEN)
4779 				bma.length = MAXEXTLEN;
4780 			else
4781 				bma.length = len;
4782 
4783 			ASSERT(len > 0);
4784 			ASSERT(bma.length > 0);
4785 			error = xfs_bmapi_allocate(&bma);
4786 			if (error)
4787 				goto error0;
4788 			if (bma.blkno == NULLFSBLOCK)
4789 				break;
4790 
4791 			/*
4792 			 * If this is a CoW allocation, record the data in
4793 			 * the refcount btree for orphan recovery.
4794 			 */
4795 			if (whichfork == XFS_COW_FORK) {
4796 				error = xfs_refcount_alloc_cow_extent(mp, dfops,
4797 						bma.blkno, bma.length);
4798 				if (error)
4799 					goto error0;
4800 			}
4801 		}
4802 
4803 		/* Deal with the allocated space we found.  */
4804 		xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4805 							end, n, flags);
4806 
4807 		/* Execute unwritten extent conversion if necessary */
4808 		error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4809 		if (error == -EAGAIN)
4810 			continue;
4811 		if (error)
4812 			goto error0;
4813 
4814 		/* update the extent map to return */
4815 		xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4816 
4817 		/*
4818 		 * If we're done, stop now.  Stop when we've allocated
4819 		 * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4820 		 * the transaction may get too big.
4821 		 */
4822 		if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4823 			break;
4824 
4825 		/* Else go on to the next record. */
4826 		bma.prev = bma.got;
4827 		if (++bma.idx < xfs_iext_count(ifp)) {
4828 			xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4829 					 &bma.got);
4830 		} else
4831 			eof = 1;
4832 	}
4833 	*nmap = n;
4834 
4835 	/*
4836 	 * Transform from btree to extents, give it cur.
4837 	 */
4838 	if (xfs_bmap_wants_extents(ip, whichfork)) {
4839 		int		tmp_logflags = 0;
4840 
4841 		ASSERT(bma.cur);
4842 		error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4843 			&tmp_logflags, whichfork);
4844 		bma.logflags |= tmp_logflags;
4845 		if (error)
4846 			goto error0;
4847 	}
4848 
4849 	ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4850 	       XFS_IFORK_NEXTENTS(ip, whichfork) >
4851 		XFS_IFORK_MAXEXT(ip, whichfork));
4852 	error = 0;
4853 error0:
4854 	/*
4855 	 * Log everything.  Do this after conversion, there's no point in
4856 	 * logging the extent records if we've converted to btree format.
4857 	 */
4858 	if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4859 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4860 		bma.logflags &= ~xfs_ilog_fext(whichfork);
4861 	else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4862 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4863 		bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4864 	/*
4865 	 * Log whatever the flags say, even if error.  Otherwise we might miss
4866 	 * detecting a case where the data is changed, there's an error,
4867 	 * and it's not logged so we don't shutdown when we should.
4868 	 */
4869 	if (bma.logflags)
4870 		xfs_trans_log_inode(tp, ip, bma.logflags);
4871 
4872 	if (bma.cur) {
4873 		if (!error) {
4874 			ASSERT(*firstblock == NULLFSBLOCK ||
4875 			       XFS_FSB_TO_AGNO(mp, *firstblock) <=
4876 			       XFS_FSB_TO_AGNO(mp,
4877 				       bma.cur->bc_private.b.firstblock));
4878 			*firstblock = bma.cur->bc_private.b.firstblock;
4879 		}
4880 		xfs_btree_del_cursor(bma.cur,
4881 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4882 	}
4883 	if (!error)
4884 		xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4885 			orig_nmap, *nmap);
4886 	return error;
4887 }
4888 
4889 /*
4890  * When a delalloc extent is split (e.g., due to a hole punch), the original
4891  * indlen reservation must be shared across the two new extents that are left
4892  * behind.
4893  *
4894  * Given the original reservation and the worst case indlen for the two new
4895  * extents (as calculated by xfs_bmap_worst_indlen()), split the original
4896  * reservation fairly across the two new extents. If necessary, steal available
4897  * blocks from a deleted extent to make up a reservation deficiency (e.g., if
4898  * ores == 1). The number of stolen blocks is returned. The availability and
4899  * subsequent accounting of stolen blocks is the responsibility of the caller.
4900  */
4901 static xfs_filblks_t
xfs_bmap_split_indlen(xfs_filblks_t ores,xfs_filblks_t * indlen1,xfs_filblks_t * indlen2,xfs_filblks_t avail)4902 xfs_bmap_split_indlen(
4903 	xfs_filblks_t			ores,		/* original res. */
4904 	xfs_filblks_t			*indlen1,	/* ext1 worst indlen */
4905 	xfs_filblks_t			*indlen2,	/* ext2 worst indlen */
4906 	xfs_filblks_t			avail)		/* stealable blocks */
4907 {
4908 	xfs_filblks_t			len1 = *indlen1;
4909 	xfs_filblks_t			len2 = *indlen2;
4910 	xfs_filblks_t			nres = len1 + len2; /* new total res. */
4911 	xfs_filblks_t			stolen = 0;
4912 	xfs_filblks_t			resfactor;
4913 
4914 	/*
4915 	 * Steal as many blocks as we can to try and satisfy the worst case
4916 	 * indlen for both new extents.
4917 	 */
4918 	if (ores < nres && avail)
4919 		stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4920 	ores += stolen;
4921 
4922 	 /* nothing else to do if we've satisfied the new reservation */
4923 	if (ores >= nres)
4924 		return stolen;
4925 
4926 	/*
4927 	 * We can't meet the total required reservation for the two extents.
4928 	 * Calculate the percent of the overall shortage between both extents
4929 	 * and apply this percentage to each of the requested indlen values.
4930 	 * This distributes the shortage fairly and reduces the chances that one
4931 	 * of the two extents is left with nothing when extents are repeatedly
4932 	 * split.
4933 	 */
4934 	resfactor = (ores * 100);
4935 	do_div(resfactor, nres);
4936 	len1 *= resfactor;
4937 	do_div(len1, 100);
4938 	len2 *= resfactor;
4939 	do_div(len2, 100);
4940 	ASSERT(len1 + len2 <= ores);
4941 	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4942 
4943 	/*
4944 	 * Hand out the remainder to each extent. If one of the two reservations
4945 	 * is zero, we want to make sure that one gets a block first. The loop
4946 	 * below starts with len1, so hand len2 a block right off the bat if it
4947 	 * is zero.
4948 	 */
4949 	ores -= (len1 + len2);
4950 	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4951 	if (ores && !len2 && *indlen2) {
4952 		len2++;
4953 		ores--;
4954 	}
4955 	while (ores) {
4956 		if (len1 < *indlen1) {
4957 			len1++;
4958 			ores--;
4959 		}
4960 		if (!ores)
4961 			break;
4962 		if (len2 < *indlen2) {
4963 			len2++;
4964 			ores--;
4965 		}
4966 	}
4967 
4968 	*indlen1 = len1;
4969 	*indlen2 = len2;
4970 
4971 	return stolen;
4972 }
4973 
4974 int
xfs_bmap_del_extent_delay(struct xfs_inode * ip,int whichfork,xfs_extnum_t * idx,struct xfs_bmbt_irec * got,struct xfs_bmbt_irec * del)4975 xfs_bmap_del_extent_delay(
4976 	struct xfs_inode	*ip,
4977 	int			whichfork,
4978 	xfs_extnum_t		*idx,
4979 	struct xfs_bmbt_irec	*got,
4980 	struct xfs_bmbt_irec	*del)
4981 {
4982 	struct xfs_mount	*mp = ip->i_mount;
4983 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
4984 	struct xfs_bmbt_irec	new;
4985 	int64_t			da_old, da_new, da_diff = 0;
4986 	xfs_fileoff_t		del_endoff, got_endoff;
4987 	xfs_filblks_t		got_indlen, new_indlen, stolen;
4988 	int			error = 0, state = 0;
4989 	bool			isrt;
4990 
4991 	XFS_STATS_INC(mp, xs_del_exlist);
4992 
4993 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4994 	del_endoff = del->br_startoff + del->br_blockcount;
4995 	got_endoff = got->br_startoff + got->br_blockcount;
4996 	da_old = startblockval(got->br_startblock);
4997 	da_new = 0;
4998 
4999 	ASSERT(*idx >= 0);
5000 	ASSERT(*idx <= xfs_iext_count(ifp));
5001 	ASSERT(del->br_blockcount > 0);
5002 	ASSERT(got->br_startoff <= del->br_startoff);
5003 	ASSERT(got_endoff >= del_endoff);
5004 
5005 	if (isrt) {
5006 		int64_t rtexts = XFS_FSB_TO_B(mp, del->br_blockcount);
5007 
5008 		do_div(rtexts, mp->m_sb.sb_rextsize);
5009 		xfs_mod_frextents(mp, rtexts);
5010 	}
5011 
5012 	/*
5013 	 * Update the inode delalloc counter now and wait to update the
5014 	 * sb counters as we might have to borrow some blocks for the
5015 	 * indirect block accounting.
5016 	 */
5017 	error = xfs_trans_reserve_quota_nblks(NULL, ip,
5018 			-((long)del->br_blockcount), 0,
5019 			isrt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
5020 	if (error)
5021 		return error;
5022 	ip->i_delayed_blks -= del->br_blockcount;
5023 
5024 	if (whichfork == XFS_COW_FORK)
5025 		state |= BMAP_COWFORK;
5026 
5027 	if (got->br_startoff == del->br_startoff)
5028 		state |= BMAP_LEFT_CONTIG;
5029 	if (got_endoff == del_endoff)
5030 		state |= BMAP_RIGHT_CONTIG;
5031 
5032 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
5033 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
5034 		/*
5035 		 * Matches the whole extent.  Delete the entry.
5036 		 */
5037 		xfs_iext_remove(ip, *idx, 1, state);
5038 		--*idx;
5039 		break;
5040 	case BMAP_LEFT_CONTIG:
5041 		/*
5042 		 * Deleting the first part of the extent.
5043 		 */
5044 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5045 		got->br_startoff = del_endoff;
5046 		got->br_blockcount -= del->br_blockcount;
5047 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
5048 				got->br_blockcount), da_old);
5049 		got->br_startblock = nullstartblock((int)da_new);
5050 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5051 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5052 		break;
5053 	case BMAP_RIGHT_CONTIG:
5054 		/*
5055 		 * Deleting the last part of the extent.
5056 		 */
5057 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5058 		got->br_blockcount = got->br_blockcount - del->br_blockcount;
5059 		da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
5060 				got->br_blockcount), da_old);
5061 		got->br_startblock = nullstartblock((int)da_new);
5062 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5063 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5064 		break;
5065 	case 0:
5066 		/*
5067 		 * Deleting the middle of the extent.
5068 		 *
5069 		 * Distribute the original indlen reservation across the two new
5070 		 * extents.  Steal blocks from the deleted extent if necessary.
5071 		 * Stealing blocks simply fudges the fdblocks accounting below.
5072 		 * Warn if either of the new indlen reservations is zero as this
5073 		 * can lead to delalloc problems.
5074 		 */
5075 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5076 
5077 		got->br_blockcount = del->br_startoff - got->br_startoff;
5078 		got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
5079 
5080 		new.br_blockcount = got_endoff - del_endoff;
5081 		new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5082 
5083 		WARN_ON_ONCE(!got_indlen || !new_indlen);
5084 		stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
5085 						       del->br_blockcount);
5086 
5087 		got->br_startblock = nullstartblock((int)got_indlen);
5088 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5089 		trace_xfs_bmap_post_update(ip, *idx, 0, _THIS_IP_);
5090 
5091 		new.br_startoff = del_endoff;
5092 		new.br_state = got->br_state;
5093 		new.br_startblock = nullstartblock((int)new_indlen);
5094 
5095 		++*idx;
5096 		xfs_iext_insert(ip, *idx, 1, &new, state);
5097 
5098 		da_new = got_indlen + new_indlen - stolen;
5099 		del->br_blockcount -= stolen;
5100 		break;
5101 	}
5102 
5103 	ASSERT(da_old >= da_new);
5104 	da_diff = da_old - da_new;
5105 	if (!isrt)
5106 		da_diff += del->br_blockcount;
5107 	if (da_diff)
5108 		xfs_mod_fdblocks(mp, da_diff, false);
5109 	return error;
5110 }
5111 
5112 void
xfs_bmap_del_extent_cow(struct xfs_inode * ip,xfs_extnum_t * idx,struct xfs_bmbt_irec * got,struct xfs_bmbt_irec * del)5113 xfs_bmap_del_extent_cow(
5114 	struct xfs_inode	*ip,
5115 	xfs_extnum_t		*idx,
5116 	struct xfs_bmbt_irec	*got,
5117 	struct xfs_bmbt_irec	*del)
5118 {
5119 	struct xfs_mount	*mp = ip->i_mount;
5120 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
5121 	struct xfs_bmbt_irec	new;
5122 	xfs_fileoff_t		del_endoff, got_endoff;
5123 	int			state = BMAP_COWFORK;
5124 
5125 	XFS_STATS_INC(mp, xs_del_exlist);
5126 
5127 	del_endoff = del->br_startoff + del->br_blockcount;
5128 	got_endoff = got->br_startoff + got->br_blockcount;
5129 
5130 	ASSERT(*idx >= 0);
5131 	ASSERT(*idx <= xfs_iext_count(ifp));
5132 	ASSERT(del->br_blockcount > 0);
5133 	ASSERT(got->br_startoff <= del->br_startoff);
5134 	ASSERT(got_endoff >= del_endoff);
5135 	ASSERT(!isnullstartblock(got->br_startblock));
5136 
5137 	if (got->br_startoff == del->br_startoff)
5138 		state |= BMAP_LEFT_CONTIG;
5139 	if (got_endoff == del_endoff)
5140 		state |= BMAP_RIGHT_CONTIG;
5141 
5142 	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
5143 	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
5144 		/*
5145 		 * Matches the whole extent.  Delete the entry.
5146 		 */
5147 		xfs_iext_remove(ip, *idx, 1, state);
5148 		--*idx;
5149 		break;
5150 	case BMAP_LEFT_CONTIG:
5151 		/*
5152 		 * Deleting the first part of the extent.
5153 		 */
5154 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5155 		got->br_startoff = del_endoff;
5156 		got->br_blockcount -= del->br_blockcount;
5157 		got->br_startblock = del->br_startblock + del->br_blockcount;
5158 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5159 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5160 		break;
5161 	case BMAP_RIGHT_CONTIG:
5162 		/*
5163 		 * Deleting the last part of the extent.
5164 		 */
5165 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5166 		got->br_blockcount -= del->br_blockcount;
5167 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5168 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5169 		break;
5170 	case 0:
5171 		/*
5172 		 * Deleting the middle of the extent.
5173 		 */
5174 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5175 		got->br_blockcount = del->br_startoff - got->br_startoff;
5176 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, *idx), got);
5177 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5178 
5179 		new.br_startoff = del_endoff;
5180 		new.br_blockcount = got_endoff - del_endoff;
5181 		new.br_state = got->br_state;
5182 		new.br_startblock = del->br_startblock + del->br_blockcount;
5183 
5184 		++*idx;
5185 		xfs_iext_insert(ip, *idx, 1, &new, state);
5186 		break;
5187 	}
5188 }
5189 
5190 /*
5191  * Called by xfs_bmapi to update file extent records and the btree
5192  * after removing space (or undoing a delayed allocation).
5193  */
5194 STATIC int				/* error */
xfs_bmap_del_extent(xfs_inode_t * ip,xfs_trans_t * tp,xfs_extnum_t * idx,struct xfs_defer_ops * dfops,xfs_btree_cur_t * cur,xfs_bmbt_irec_t * del,int * logflagsp,int whichfork,int bflags)5195 xfs_bmap_del_extent(
5196 	xfs_inode_t		*ip,	/* incore inode pointer */
5197 	xfs_trans_t		*tp,	/* current transaction pointer */
5198 	xfs_extnum_t		*idx,	/* extent number to update/delete */
5199 	struct xfs_defer_ops	*dfops,	/* list of extents to be freed */
5200 	xfs_btree_cur_t		*cur,	/* if null, not a btree */
5201 	xfs_bmbt_irec_t		*del,	/* data to remove from extents */
5202 	int			*logflagsp, /* inode logging flags */
5203 	int			whichfork, /* data or attr fork */
5204 	int			bflags)	/* bmapi flags */
5205 {
5206 	xfs_filblks_t		da_new;	/* new delay-alloc indirect blocks */
5207 	xfs_filblks_t		da_old;	/* old delay-alloc indirect blocks */
5208 	xfs_fsblock_t		del_endblock=0;	/* first block past del */
5209 	xfs_fileoff_t		del_endoff;	/* first offset past del */
5210 	int			delay;	/* current block is delayed allocated */
5211 	int			do_fx;	/* free extent at end of routine */
5212 	xfs_bmbt_rec_host_t	*ep;	/* current extent entry pointer */
5213 	int			error;	/* error return value */
5214 	int			flags;	/* inode logging flags */
5215 	xfs_bmbt_irec_t		got;	/* current extent entry */
5216 	xfs_fileoff_t		got_endoff;	/* first offset past got */
5217 	int			i;	/* temp state */
5218 	xfs_ifork_t		*ifp;	/* inode fork pointer */
5219 	xfs_mount_t		*mp;	/* mount structure */
5220 	xfs_filblks_t		nblks;	/* quota/sb block count */
5221 	xfs_bmbt_irec_t		new;	/* new record to be inserted */
5222 	/* REFERENCED */
5223 	uint			qfield;	/* quota field to update */
5224 	xfs_filblks_t		temp;	/* for indirect length calculations */
5225 	xfs_filblks_t		temp2;	/* for indirect length calculations */
5226 	int			state = 0;
5227 
5228 	mp = ip->i_mount;
5229 	XFS_STATS_INC(mp, xs_del_exlist);
5230 
5231 	if (whichfork == XFS_ATTR_FORK)
5232 		state |= BMAP_ATTRFORK;
5233 	else if (whichfork == XFS_COW_FORK)
5234 		state |= BMAP_COWFORK;
5235 
5236 	ifp = XFS_IFORK_PTR(ip, whichfork);
5237 	ASSERT((*idx >= 0) && (*idx < xfs_iext_count(ifp)));
5238 	ASSERT(del->br_blockcount > 0);
5239 	ep = xfs_iext_get_ext(ifp, *idx);
5240 	xfs_bmbt_get_all(ep, &got);
5241 	ASSERT(got.br_startoff <= del->br_startoff);
5242 	del_endoff = del->br_startoff + del->br_blockcount;
5243 	got_endoff = got.br_startoff + got.br_blockcount;
5244 	ASSERT(got_endoff >= del_endoff);
5245 	delay = isnullstartblock(got.br_startblock);
5246 	ASSERT(isnullstartblock(del->br_startblock) == delay);
5247 	flags = 0;
5248 	qfield = 0;
5249 	error = 0;
5250 	/*
5251 	 * If deleting a real allocation, must free up the disk space.
5252 	 */
5253 	if (!delay) {
5254 		flags = XFS_ILOG_CORE;
5255 		/*
5256 		 * Realtime allocation.  Free it and record di_nblocks update.
5257 		 */
5258 		if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
5259 			xfs_fsblock_t	bno;
5260 			xfs_filblks_t	len;
5261 
5262 			ASSERT(do_mod(del->br_blockcount,
5263 				      mp->m_sb.sb_rextsize) == 0);
5264 			ASSERT(do_mod(del->br_startblock,
5265 				      mp->m_sb.sb_rextsize) == 0);
5266 			bno = del->br_startblock;
5267 			len = del->br_blockcount;
5268 			do_div(bno, mp->m_sb.sb_rextsize);
5269 			do_div(len, mp->m_sb.sb_rextsize);
5270 			error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
5271 			if (error)
5272 				goto done;
5273 			do_fx = 0;
5274 			nblks = len * mp->m_sb.sb_rextsize;
5275 			qfield = XFS_TRANS_DQ_RTBCOUNT;
5276 		}
5277 		/*
5278 		 * Ordinary allocation.
5279 		 */
5280 		else {
5281 			do_fx = 1;
5282 			nblks = del->br_blockcount;
5283 			qfield = XFS_TRANS_DQ_BCOUNT;
5284 		}
5285 		/*
5286 		 * Set up del_endblock and cur for later.
5287 		 */
5288 		del_endblock = del->br_startblock + del->br_blockcount;
5289 		if (cur) {
5290 			if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
5291 					got.br_startblock, got.br_blockcount,
5292 					&i)))
5293 				goto done;
5294 			XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5295 		}
5296 		da_old = da_new = 0;
5297 	} else {
5298 		da_old = startblockval(got.br_startblock);
5299 		da_new = 0;
5300 		nblks = 0;
5301 		do_fx = 0;
5302 	}
5303 
5304 	/*
5305 	 * Set flag value to use in switch statement.
5306 	 * Left-contig is 2, right-contig is 1.
5307 	 */
5308 	switch (((got.br_startoff == del->br_startoff) << 1) |
5309 		(got_endoff == del_endoff)) {
5310 	case 3:
5311 		/*
5312 		 * Matches the whole extent.  Delete the entry.
5313 		 */
5314 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5315 		xfs_iext_remove(ip, *idx, 1,
5316 				whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
5317 		--*idx;
5318 		if (delay)
5319 			break;
5320 
5321 		XFS_IFORK_NEXT_SET(ip, whichfork,
5322 			XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5323 		flags |= XFS_ILOG_CORE;
5324 		if (!cur) {
5325 			flags |= xfs_ilog_fext(whichfork);
5326 			break;
5327 		}
5328 		if ((error = xfs_btree_delete(cur, &i)))
5329 			goto done;
5330 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5331 		break;
5332 
5333 	case 2:
5334 		/*
5335 		 * Deleting the first part of the extent.
5336 		 */
5337 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5338 		xfs_bmbt_set_startoff(ep, del_endoff);
5339 		temp = got.br_blockcount - del->br_blockcount;
5340 		xfs_bmbt_set_blockcount(ep, temp);
5341 		if (delay) {
5342 			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5343 				da_old);
5344 			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5345 			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5346 			da_new = temp;
5347 			break;
5348 		}
5349 		xfs_bmbt_set_startblock(ep, del_endblock);
5350 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5351 		if (!cur) {
5352 			flags |= xfs_ilog_fext(whichfork);
5353 			break;
5354 		}
5355 		if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
5356 				got.br_blockcount - del->br_blockcount,
5357 				got.br_state)))
5358 			goto done;
5359 		break;
5360 
5361 	case 1:
5362 		/*
5363 		 * Deleting the last part of the extent.
5364 		 */
5365 		temp = got.br_blockcount - del->br_blockcount;
5366 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5367 		xfs_bmbt_set_blockcount(ep, temp);
5368 		if (delay) {
5369 			temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
5370 				da_old);
5371 			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5372 			trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5373 			da_new = temp;
5374 			break;
5375 		}
5376 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5377 		if (!cur) {
5378 			flags |= xfs_ilog_fext(whichfork);
5379 			break;
5380 		}
5381 		if ((error = xfs_bmbt_update(cur, got.br_startoff,
5382 				got.br_startblock,
5383 				got.br_blockcount - del->br_blockcount,
5384 				got.br_state)))
5385 			goto done;
5386 		break;
5387 
5388 	case 0:
5389 		/*
5390 		 * Deleting the middle of the extent.
5391 		 */
5392 		temp = del->br_startoff - got.br_startoff;
5393 		trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
5394 		xfs_bmbt_set_blockcount(ep, temp);
5395 		new.br_startoff = del_endoff;
5396 		temp2 = got_endoff - del_endoff;
5397 		new.br_blockcount = temp2;
5398 		new.br_state = got.br_state;
5399 		if (!delay) {
5400 			new.br_startblock = del_endblock;
5401 			flags |= XFS_ILOG_CORE;
5402 			if (cur) {
5403 				if ((error = xfs_bmbt_update(cur,
5404 						got.br_startoff,
5405 						got.br_startblock, temp,
5406 						got.br_state)))
5407 					goto done;
5408 				if ((error = xfs_btree_increment(cur, 0, &i)))
5409 					goto done;
5410 				cur->bc_rec.b = new;
5411 				error = xfs_btree_insert(cur, &i);
5412 				if (error && error != -ENOSPC)
5413 					goto done;
5414 				/*
5415 				 * If get no-space back from btree insert,
5416 				 * it tried a split, and we have a zero
5417 				 * block reservation.
5418 				 * Fix up our state and return the error.
5419 				 */
5420 				if (error == -ENOSPC) {
5421 					/*
5422 					 * Reset the cursor, don't trust
5423 					 * it after any insert operation.
5424 					 */
5425 					if ((error = xfs_bmbt_lookup_eq(cur,
5426 							got.br_startoff,
5427 							got.br_startblock,
5428 							temp, &i)))
5429 						goto done;
5430 					XFS_WANT_CORRUPTED_GOTO(mp,
5431 								i == 1, done);
5432 					/*
5433 					 * Update the btree record back
5434 					 * to the original value.
5435 					 */
5436 					if ((error = xfs_bmbt_update(cur,
5437 							got.br_startoff,
5438 							got.br_startblock,
5439 							got.br_blockcount,
5440 							got.br_state)))
5441 						goto done;
5442 					/*
5443 					 * Reset the extent record back
5444 					 * to the original value.
5445 					 */
5446 					xfs_bmbt_set_blockcount(ep,
5447 						got.br_blockcount);
5448 					flags = 0;
5449 					error = -ENOSPC;
5450 					goto done;
5451 				}
5452 				XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
5453 			} else
5454 				flags |= xfs_ilog_fext(whichfork);
5455 			XFS_IFORK_NEXT_SET(ip, whichfork,
5456 				XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
5457 		} else {
5458 			xfs_filblks_t	stolen;
5459 			ASSERT(whichfork == XFS_DATA_FORK);
5460 
5461 			/*
5462 			 * Distribute the original indlen reservation across the
5463 			 * two new extents. Steal blocks from the deleted extent
5464 			 * if necessary. Stealing blocks simply fudges the
5465 			 * fdblocks accounting in xfs_bunmapi().
5466 			 */
5467 			temp = xfs_bmap_worst_indlen(ip, got.br_blockcount);
5468 			temp2 = xfs_bmap_worst_indlen(ip, new.br_blockcount);
5469 			stolen = xfs_bmap_split_indlen(da_old, &temp, &temp2,
5470 						       del->br_blockcount);
5471 			da_new = temp + temp2 - stolen;
5472 			del->br_blockcount -= stolen;
5473 
5474 			/*
5475 			 * Set the reservation for each extent. Warn if either
5476 			 * is zero as this can lead to delalloc problems.
5477 			 */
5478 			WARN_ON_ONCE(!temp || !temp2);
5479 			xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
5480 			new.br_startblock = nullstartblock((int)temp2);
5481 		}
5482 		trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
5483 		xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5484 		++*idx;
5485 		break;
5486 	}
5487 
5488 	/* remove reverse mapping */
5489 	if (!delay) {
5490 		error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, del);
5491 		if (error)
5492 			goto done;
5493 	}
5494 
5495 	/*
5496 	 * If we need to, add to list of extents to delete.
5497 	 */
5498 	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5499 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5500 			error = xfs_refcount_decrease_extent(mp, dfops, del);
5501 			if (error)
5502 				goto done;
5503 		} else
5504 			xfs_bmap_add_free(mp, dfops, del->br_startblock,
5505 					del->br_blockcount, NULL);
5506 	}
5507 
5508 	/*
5509 	 * Adjust inode # blocks in the file.
5510 	 */
5511 	if (nblks)
5512 		ip->i_d.di_nblocks -= nblks;
5513 	/*
5514 	 * Adjust quota data.
5515 	 */
5516 	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5517 		xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5518 
5519 	/*
5520 	 * Account for change in delayed indirect blocks.
5521 	 * Nothing to do for disk quota accounting here.
5522 	 */
5523 	ASSERT(da_old >= da_new);
5524 	if (da_old > da_new)
5525 		xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new), false);
5526 done:
5527 	*logflagsp = flags;
5528 	return error;
5529 }
5530 
5531 /*
5532  * Unmap (remove) blocks from a file.
5533  * If nexts is nonzero then the number of extents to remove is limited to
5534  * that value.  If not all extents in the block range can be removed then
5535  * *done is set.
5536  */
5537 int						/* error */
__xfs_bunmapi(xfs_trans_t * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t * rlen,int flags,xfs_extnum_t nexts,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops)5538 __xfs_bunmapi(
5539 	xfs_trans_t		*tp,		/* transaction pointer */
5540 	struct xfs_inode	*ip,		/* incore inode */
5541 	xfs_fileoff_t		bno,		/* starting offset to unmap */
5542 	xfs_filblks_t		*rlen,		/* i/o: amount remaining */
5543 	int			flags,		/* misc flags */
5544 	xfs_extnum_t		nexts,		/* number of extents max */
5545 	xfs_fsblock_t		*firstblock,	/* first allocated block
5546 						   controls a.g. for allocs */
5547 	struct xfs_defer_ops	*dfops)		/* i/o: deferred updates */
5548 {
5549 	xfs_btree_cur_t		*cur;		/* bmap btree cursor */
5550 	xfs_bmbt_irec_t		del;		/* extent being deleted */
5551 	int			eof;		/* is deleting at eof */
5552 	xfs_bmbt_rec_host_t	*ep;		/* extent record pointer */
5553 	int			error;		/* error return value */
5554 	xfs_extnum_t		extno;		/* extent number in list */
5555 	xfs_bmbt_irec_t		got;		/* current extent record */
5556 	xfs_ifork_t		*ifp;		/* inode fork pointer */
5557 	int			isrt;		/* freeing in rt area */
5558 	xfs_extnum_t		lastx;		/* last extent index used */
5559 	int			logflags;	/* transaction logging flags */
5560 	xfs_extlen_t		mod;		/* rt extent offset */
5561 	xfs_mount_t		*mp;		/* mount structure */
5562 	xfs_bmbt_irec_t		prev;		/* previous extent record */
5563 	xfs_fileoff_t		start;		/* first file offset deleted */
5564 	int			tmp_logflags;	/* partial logging flags */
5565 	int			wasdel;		/* was a delayed alloc extent */
5566 	int			whichfork;	/* data or attribute fork */
5567 	xfs_fsblock_t		sum;
5568 	xfs_filblks_t		len = *rlen;	/* length to unmap in file */
5569 	xfs_fileoff_t		max_len;
5570 	xfs_agnumber_t		prev_agno = NULLAGNUMBER, agno;
5571 
5572 	trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5573 
5574 	whichfork = xfs_bmapi_whichfork(flags);
5575 	ASSERT(whichfork != XFS_COW_FORK);
5576 	ifp = XFS_IFORK_PTR(ip, whichfork);
5577 	if (unlikely(
5578 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5579 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5580 		XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5581 				 ip->i_mount);
5582 		return -EFSCORRUPTED;
5583 	}
5584 	mp = ip->i_mount;
5585 	if (XFS_FORCED_SHUTDOWN(mp))
5586 		return -EIO;
5587 
5588 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5589 	ASSERT(len > 0);
5590 	ASSERT(nexts >= 0);
5591 
5592 	/*
5593 	 * Guesstimate how many blocks we can unmap without running the risk of
5594 	 * blowing out the transaction with a mix of EFIs and reflink
5595 	 * adjustments.
5596 	 */
5597 	if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK)
5598 		max_len = min(len, xfs_refcount_max_unmap(tp->t_log_res));
5599 	else
5600 		max_len = len;
5601 
5602 	if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5603 	    (error = xfs_iread_extents(tp, ip, whichfork)))
5604 		return error;
5605 	if (xfs_iext_count(ifp) == 0) {
5606 		*rlen = 0;
5607 		return 0;
5608 	}
5609 	XFS_STATS_INC(mp, xs_blk_unmap);
5610 	isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5611 	start = bno;
5612 	bno = start + len - 1;
5613 	ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5614 		&prev);
5615 
5616 	/*
5617 	 * Check to see if the given block number is past the end of the
5618 	 * file, back up to the last block if so...
5619 	 */
5620 	if (eof) {
5621 		ep = xfs_iext_get_ext(ifp, --lastx);
5622 		xfs_bmbt_get_all(ep, &got);
5623 		bno = got.br_startoff + got.br_blockcount - 1;
5624 	}
5625 	logflags = 0;
5626 	if (ifp->if_flags & XFS_IFBROOT) {
5627 		ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5628 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5629 		cur->bc_private.b.firstblock = *firstblock;
5630 		cur->bc_private.b.dfops = dfops;
5631 		cur->bc_private.b.flags = 0;
5632 	} else
5633 		cur = NULL;
5634 
5635 	if (isrt) {
5636 		/*
5637 		 * Synchronize by locking the bitmap inode.
5638 		 */
5639 		xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5640 		xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5641 		xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5642 		xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5643 	}
5644 
5645 	extno = 0;
5646 	while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5647 	       (nexts == 0 || extno < nexts) && max_len > 0) {
5648 		/*
5649 		 * Is the found extent after a hole in which bno lives?
5650 		 * Just back up to the previous extent, if so.
5651 		 */
5652 		if (got.br_startoff > bno) {
5653 			if (--lastx < 0)
5654 				break;
5655 			ep = xfs_iext_get_ext(ifp, lastx);
5656 			xfs_bmbt_get_all(ep, &got);
5657 		}
5658 		/*
5659 		 * Is the last block of this extent before the range
5660 		 * we're supposed to delete?  If so, we're done.
5661 		 */
5662 		bno = XFS_FILEOFF_MIN(bno,
5663 			got.br_startoff + got.br_blockcount - 1);
5664 		if (bno < start)
5665 			break;
5666 		/*
5667 		 * Then deal with the (possibly delayed) allocated space
5668 		 * we found.
5669 		 */
5670 		ASSERT(ep != NULL);
5671 		del = got;
5672 		wasdel = isnullstartblock(del.br_startblock);
5673 
5674 		/*
5675 		 * Make sure we don't touch multiple AGF headers out of order
5676 		 * in a single transaction, as that could cause AB-BA deadlocks.
5677 		 */
5678 		if (!wasdel) {
5679 			agno = XFS_FSB_TO_AGNO(mp, del.br_startblock);
5680 			if (prev_agno != NULLAGNUMBER && prev_agno > agno)
5681 				break;
5682 			prev_agno = agno;
5683 		}
5684 		if (got.br_startoff < start) {
5685 			del.br_startoff = start;
5686 			del.br_blockcount -= start - got.br_startoff;
5687 			if (!wasdel)
5688 				del.br_startblock += start - got.br_startoff;
5689 		}
5690 		if (del.br_startoff + del.br_blockcount > bno + 1)
5691 			del.br_blockcount = bno + 1 - del.br_startoff;
5692 
5693 		/* How much can we safely unmap? */
5694 		if (max_len < del.br_blockcount) {
5695 			del.br_startoff += del.br_blockcount - max_len;
5696 			if (!wasdel)
5697 				del.br_startblock += del.br_blockcount - max_len;
5698 			del.br_blockcount = max_len;
5699 		}
5700 
5701 		sum = del.br_startblock + del.br_blockcount;
5702 		if (isrt &&
5703 		    (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5704 			/*
5705 			 * Realtime extent not lined up at the end.
5706 			 * The extent could have been split into written
5707 			 * and unwritten pieces, or we could just be
5708 			 * unmapping part of it.  But we can't really
5709 			 * get rid of part of a realtime extent.
5710 			 */
5711 			if (del.br_state == XFS_EXT_UNWRITTEN ||
5712 			    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5713 				/*
5714 				 * This piece is unwritten, or we're not
5715 				 * using unwritten extents.  Skip over it.
5716 				 */
5717 				ASSERT(bno >= mod);
5718 				bno -= mod > del.br_blockcount ?
5719 					del.br_blockcount : mod;
5720 				if (bno < got.br_startoff) {
5721 					if (--lastx >= 0)
5722 						xfs_bmbt_get_all(xfs_iext_get_ext(
5723 							ifp, lastx), &got);
5724 				}
5725 				continue;
5726 			}
5727 			/*
5728 			 * It's written, turn it unwritten.
5729 			 * This is better than zeroing it.
5730 			 */
5731 			ASSERT(del.br_state == XFS_EXT_NORM);
5732 			ASSERT(tp->t_blk_res > 0);
5733 			/*
5734 			 * If this spans a realtime extent boundary,
5735 			 * chop it back to the start of the one we end at.
5736 			 */
5737 			if (del.br_blockcount > mod) {
5738 				del.br_startoff += del.br_blockcount - mod;
5739 				del.br_startblock += del.br_blockcount - mod;
5740 				del.br_blockcount = mod;
5741 			}
5742 			del.br_state = XFS_EXT_UNWRITTEN;
5743 			error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5744 					whichfork, &lastx, &cur, &del,
5745 					firstblock, dfops, &logflags);
5746 			if (error)
5747 				goto error0;
5748 			goto nodelete;
5749 		}
5750 		if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5751 			/*
5752 			 * Realtime extent is lined up at the end but not
5753 			 * at the front.  We'll get rid of full extents if
5754 			 * we can.
5755 			 */
5756 			mod = mp->m_sb.sb_rextsize - mod;
5757 			if (del.br_blockcount > mod) {
5758 				del.br_blockcount -= mod;
5759 				del.br_startoff += mod;
5760 				del.br_startblock += mod;
5761 			} else if ((del.br_startoff == start &&
5762 				    (del.br_state == XFS_EXT_UNWRITTEN ||
5763 				     tp->t_blk_res == 0)) ||
5764 				   !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5765 				/*
5766 				 * Can't make it unwritten.  There isn't
5767 				 * a full extent here so just skip it.
5768 				 */
5769 				ASSERT(bno >= del.br_blockcount);
5770 				bno -= del.br_blockcount;
5771 				if (got.br_startoff > bno) {
5772 					if (--lastx >= 0) {
5773 						ep = xfs_iext_get_ext(ifp,
5774 								      lastx);
5775 						xfs_bmbt_get_all(ep, &got);
5776 					}
5777 				}
5778 				continue;
5779 			} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5780 				/*
5781 				 * This one is already unwritten.
5782 				 * It must have a written left neighbor.
5783 				 * Unwrite the killed part of that one and
5784 				 * try again.
5785 				 */
5786 				ASSERT(lastx > 0);
5787 				xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5788 						lastx - 1), &prev);
5789 				ASSERT(prev.br_state == XFS_EXT_NORM);
5790 				ASSERT(!isnullstartblock(prev.br_startblock));
5791 				ASSERT(del.br_startblock ==
5792 				       prev.br_startblock + prev.br_blockcount);
5793 				if (prev.br_startoff < start) {
5794 					mod = start - prev.br_startoff;
5795 					prev.br_blockcount -= mod;
5796 					prev.br_startblock += mod;
5797 					prev.br_startoff = start;
5798 				}
5799 				prev.br_state = XFS_EXT_UNWRITTEN;
5800 				lastx--;
5801 				error = xfs_bmap_add_extent_unwritten_real(tp,
5802 						ip, whichfork, &lastx, &cur,
5803 						&prev, firstblock, dfops,
5804 						&logflags);
5805 				if (error)
5806 					goto error0;
5807 				goto nodelete;
5808 			} else {
5809 				ASSERT(del.br_state == XFS_EXT_NORM);
5810 				del.br_state = XFS_EXT_UNWRITTEN;
5811 				error = xfs_bmap_add_extent_unwritten_real(tp,
5812 						ip, whichfork, &lastx, &cur,
5813 						&del, firstblock, dfops,
5814 						&logflags);
5815 				if (error)
5816 					goto error0;
5817 				goto nodelete;
5818 			}
5819 		}
5820 
5821 		/*
5822 		 * If it's the case where the directory code is running
5823 		 * with no block reservation, and the deleted block is in
5824 		 * the middle of its extent, and the resulting insert
5825 		 * of an extent would cause transformation to btree format,
5826 		 * then reject it.  The calling code will then swap
5827 		 * blocks around instead.
5828 		 * We have to do this now, rather than waiting for the
5829 		 * conversion to btree format, since the transaction
5830 		 * will be dirty.
5831 		 */
5832 		if (!wasdel && tp->t_blk_res == 0 &&
5833 		    XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5834 		    XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5835 			XFS_IFORK_MAXEXT(ip, whichfork) &&
5836 		    del.br_startoff > got.br_startoff &&
5837 		    del.br_startoff + del.br_blockcount <
5838 		    got.br_startoff + got.br_blockcount) {
5839 			error = -ENOSPC;
5840 			goto error0;
5841 		}
5842 
5843 		/*
5844 		 * Unreserve quota and update realtime free space, if
5845 		 * appropriate. If delayed allocation, update the inode delalloc
5846 		 * counter now and wait to update the sb counters as
5847 		 * xfs_bmap_del_extent() might need to borrow some blocks.
5848 		 */
5849 		if (wasdel) {
5850 			ASSERT(startblockval(del.br_startblock) > 0);
5851 			if (isrt) {
5852 				xfs_filblks_t rtexts;
5853 
5854 				rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5855 				do_div(rtexts, mp->m_sb.sb_rextsize);
5856 				xfs_mod_frextents(mp, (int64_t)rtexts);
5857 				(void)xfs_trans_reserve_quota_nblks(NULL,
5858 					ip, -((long)del.br_blockcount), 0,
5859 					XFS_QMOPT_RES_RTBLKS);
5860 			} else {
5861 				(void)xfs_trans_reserve_quota_nblks(NULL,
5862 					ip, -((long)del.br_blockcount), 0,
5863 					XFS_QMOPT_RES_REGBLKS);
5864 			}
5865 			ip->i_delayed_blks -= del.br_blockcount;
5866 			if (cur)
5867 				cur->bc_private.b.flags |=
5868 					XFS_BTCUR_BPRV_WASDEL;
5869 		} else if (cur)
5870 			cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5871 
5872 		error = xfs_bmap_del_extent(ip, tp, &lastx, dfops, cur, &del,
5873 				&tmp_logflags, whichfork, flags);
5874 		logflags |= tmp_logflags;
5875 		if (error)
5876 			goto error0;
5877 
5878 		if (!isrt && wasdel)
5879 			xfs_mod_fdblocks(mp, (int64_t)del.br_blockcount, false);
5880 
5881 		max_len -= del.br_blockcount;
5882 		bno = del.br_startoff - 1;
5883 nodelete:
5884 		/*
5885 		 * If not done go on to the next (previous) record.
5886 		 */
5887 		if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5888 			if (lastx >= 0) {
5889 				ep = xfs_iext_get_ext(ifp, lastx);
5890 				if (xfs_bmbt_get_startoff(ep) > bno) {
5891 					if (--lastx >= 0)
5892 						ep = xfs_iext_get_ext(ifp,
5893 								      lastx);
5894 				}
5895 				xfs_bmbt_get_all(ep, &got);
5896 			}
5897 			extno++;
5898 		}
5899 	}
5900 	if (bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0)
5901 		*rlen = 0;
5902 	else
5903 		*rlen = bno - start + 1;
5904 
5905 	/*
5906 	 * Convert to a btree if necessary.
5907 	 */
5908 	if (xfs_bmap_needs_btree(ip, whichfork)) {
5909 		ASSERT(cur == NULL);
5910 		error = xfs_bmap_extents_to_btree(tp, ip, firstblock, dfops,
5911 			&cur, 0, &tmp_logflags, whichfork);
5912 		logflags |= tmp_logflags;
5913 		if (error)
5914 			goto error0;
5915 	}
5916 	/*
5917 	 * transform from btree to extents, give it cur
5918 	 */
5919 	else if (xfs_bmap_wants_extents(ip, whichfork)) {
5920 		ASSERT(cur != NULL);
5921 		error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5922 			whichfork);
5923 		logflags |= tmp_logflags;
5924 		if (error)
5925 			goto error0;
5926 	}
5927 	/*
5928 	 * transform from extents to local?
5929 	 */
5930 	error = 0;
5931 error0:
5932 	/*
5933 	 * Log everything.  Do this after conversion, there's no point in
5934 	 * logging the extent records if we've converted to btree format.
5935 	 */
5936 	if ((logflags & xfs_ilog_fext(whichfork)) &&
5937 	    XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5938 		logflags &= ~xfs_ilog_fext(whichfork);
5939 	else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5940 		 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5941 		logflags &= ~xfs_ilog_fbroot(whichfork);
5942 	/*
5943 	 * Log inode even in the error case, if the transaction
5944 	 * is dirty we'll need to shut down the filesystem.
5945 	 */
5946 	if (logflags)
5947 		xfs_trans_log_inode(tp, ip, logflags);
5948 	if (cur) {
5949 		if (!error) {
5950 			*firstblock = cur->bc_private.b.firstblock;
5951 			cur->bc_private.b.allocated = 0;
5952 		}
5953 		xfs_btree_del_cursor(cur,
5954 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5955 	}
5956 	return error;
5957 }
5958 
5959 /* Unmap a range of a file. */
5960 int
xfs_bunmapi(xfs_trans_t * tp,struct xfs_inode * ip,xfs_fileoff_t bno,xfs_filblks_t len,int flags,xfs_extnum_t nexts,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,int * done)5961 xfs_bunmapi(
5962 	xfs_trans_t		*tp,
5963 	struct xfs_inode	*ip,
5964 	xfs_fileoff_t		bno,
5965 	xfs_filblks_t		len,
5966 	int			flags,
5967 	xfs_extnum_t		nexts,
5968 	xfs_fsblock_t		*firstblock,
5969 	struct xfs_defer_ops	*dfops,
5970 	int			*done)
5971 {
5972 	int			error;
5973 
5974 	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts, firstblock,
5975 			dfops);
5976 	*done = (len == 0);
5977 	return error;
5978 }
5979 
5980 /*
5981  * Determine whether an extent shift can be accomplished by a merge with the
5982  * extent that precedes the target hole of the shift.
5983  */
5984 STATIC bool
xfs_bmse_can_merge(struct xfs_bmbt_irec * left,struct xfs_bmbt_irec * got,xfs_fileoff_t shift)5985 xfs_bmse_can_merge(
5986 	struct xfs_bmbt_irec	*left,	/* preceding extent */
5987 	struct xfs_bmbt_irec	*got,	/* current extent to shift */
5988 	xfs_fileoff_t		shift)	/* shift fsb */
5989 {
5990 	xfs_fileoff_t		startoff;
5991 
5992 	startoff = got->br_startoff - shift;
5993 
5994 	/*
5995 	 * The extent, once shifted, must be adjacent in-file and on-disk with
5996 	 * the preceding extent.
5997 	 */
5998 	if ((left->br_startoff + left->br_blockcount != startoff) ||
5999 	    (left->br_startblock + left->br_blockcount != got->br_startblock) ||
6000 	    (left->br_state != got->br_state) ||
6001 	    (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
6002 		return false;
6003 
6004 	return true;
6005 }
6006 
6007 /*
6008  * A bmap extent shift adjusts the file offset of an extent to fill a preceding
6009  * hole in the file. If an extent shift would result in the extent being fully
6010  * adjacent to the extent that currently precedes the hole, we can merge with
6011  * the preceding extent rather than do the shift.
6012  *
6013  * This function assumes the caller has verified a shift-by-merge is possible
6014  * with the provided extents via xfs_bmse_can_merge().
6015  */
6016 STATIC int
xfs_bmse_merge(struct xfs_inode * ip,int whichfork,xfs_fileoff_t shift,int current_ext,struct xfs_bmbt_rec_host * gotp,struct xfs_bmbt_rec_host * leftp,struct xfs_btree_cur * cur,int * logflags)6017 xfs_bmse_merge(
6018 	struct xfs_inode		*ip,
6019 	int				whichfork,
6020 	xfs_fileoff_t			shift,		/* shift fsb */
6021 	int				current_ext,	/* idx of gotp */
6022 	struct xfs_bmbt_rec_host	*gotp,		/* extent to shift */
6023 	struct xfs_bmbt_rec_host	*leftp,		/* preceding extent */
6024 	struct xfs_btree_cur		*cur,
6025 	int				*logflags)	/* output */
6026 {
6027 	struct xfs_bmbt_irec		got;
6028 	struct xfs_bmbt_irec		left;
6029 	xfs_filblks_t			blockcount;
6030 	int				error, i;
6031 	struct xfs_mount		*mp = ip->i_mount;
6032 
6033 	xfs_bmbt_get_all(gotp, &got);
6034 	xfs_bmbt_get_all(leftp, &left);
6035 	blockcount = left.br_blockcount + got.br_blockcount;
6036 
6037 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
6038 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6039 	ASSERT(xfs_bmse_can_merge(&left, &got, shift));
6040 
6041 	/*
6042 	 * Merge the in-core extents. Note that the host record pointers and
6043 	 * current_ext index are invalid once the extent has been removed via
6044 	 * xfs_iext_remove().
6045 	 */
6046 	xfs_bmbt_set_blockcount(leftp, blockcount);
6047 	xfs_iext_remove(ip, current_ext, 1, 0);
6048 
6049 	/*
6050 	 * Update the on-disk extent count, the btree if necessary and log the
6051 	 * inode.
6052 	 */
6053 	XFS_IFORK_NEXT_SET(ip, whichfork,
6054 			   XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
6055 	*logflags |= XFS_ILOG_CORE;
6056 	if (!cur) {
6057 		*logflags |= XFS_ILOG_DEXT;
6058 		return 0;
6059 	}
6060 
6061 	/* lookup and remove the extent to merge */
6062 	error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
6063 				   got.br_blockcount, &i);
6064 	if (error)
6065 		return error;
6066 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
6067 
6068 	error = xfs_btree_delete(cur, &i);
6069 	if (error)
6070 		return error;
6071 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
6072 
6073 	/* lookup and update size of the previous extent */
6074 	error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
6075 				   left.br_blockcount, &i);
6076 	if (error)
6077 		return error;
6078 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
6079 
6080 	left.br_blockcount = blockcount;
6081 
6082 	return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
6083 			       left.br_blockcount, left.br_state);
6084 }
6085 
6086 /*
6087  * Shift a single extent.
6088  */
6089 STATIC int
xfs_bmse_shift_one(struct xfs_inode * ip,int whichfork,xfs_fileoff_t offset_shift_fsb,int * current_ext,struct xfs_bmbt_rec_host * gotp,struct xfs_btree_cur * cur,int * logflags,enum shift_direction direction,struct xfs_defer_ops * dfops)6090 xfs_bmse_shift_one(
6091 	struct xfs_inode		*ip,
6092 	int				whichfork,
6093 	xfs_fileoff_t			offset_shift_fsb,
6094 	int				*current_ext,
6095 	struct xfs_bmbt_rec_host	*gotp,
6096 	struct xfs_btree_cur		*cur,
6097 	int				*logflags,
6098 	enum shift_direction		direction,
6099 	struct xfs_defer_ops		*dfops)
6100 {
6101 	struct xfs_ifork		*ifp;
6102 	struct xfs_mount		*mp;
6103 	xfs_fileoff_t			startoff;
6104 	struct xfs_bmbt_rec_host	*adj_irecp;
6105 	struct xfs_bmbt_irec		got;
6106 	struct xfs_bmbt_irec		adj_irec;
6107 	int				error;
6108 	int				i;
6109 	int				total_extents;
6110 
6111 	mp = ip->i_mount;
6112 	ifp = XFS_IFORK_PTR(ip, whichfork);
6113 	total_extents = xfs_iext_count(ifp);
6114 
6115 	xfs_bmbt_get_all(gotp, &got);
6116 
6117 	/* delalloc extents should be prevented by caller */
6118 	XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
6119 
6120 	if (direction == SHIFT_LEFT) {
6121 		startoff = got.br_startoff - offset_shift_fsb;
6122 
6123 		/*
6124 		 * Check for merge if we've got an extent to the left,
6125 		 * otherwise make sure there's enough room at the start
6126 		 * of the file for the shift.
6127 		 */
6128 		if (!*current_ext) {
6129 			if (got.br_startoff < offset_shift_fsb)
6130 				return -EINVAL;
6131 			goto update_current_ext;
6132 		}
6133 		/*
6134 		 * grab the left extent and check for a large
6135 		 * enough hole.
6136 		 */
6137 		adj_irecp = xfs_iext_get_ext(ifp, *current_ext - 1);
6138 		xfs_bmbt_get_all(adj_irecp, &adj_irec);
6139 
6140 		if (startoff <
6141 		    adj_irec.br_startoff + adj_irec.br_blockcount)
6142 			return -EINVAL;
6143 
6144 		/* check whether to merge the extent or shift it down */
6145 		if (xfs_bmse_can_merge(&adj_irec, &got,
6146 				       offset_shift_fsb)) {
6147 			error = xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
6148 					       *current_ext, gotp, adj_irecp,
6149 					       cur, logflags);
6150 			if (error)
6151 				return error;
6152 			adj_irec = got;
6153 			goto update_rmap;
6154 		}
6155 	} else {
6156 		startoff = got.br_startoff + offset_shift_fsb;
6157 		/* nothing to move if this is the last extent */
6158 		if (*current_ext >= (total_extents - 1))
6159 			goto update_current_ext;
6160 		/*
6161 		 * If this is not the last extent in the file, make sure there
6162 		 * is enough room between current extent and next extent for
6163 		 * accommodating the shift.
6164 		 */
6165 		adj_irecp = xfs_iext_get_ext(ifp, *current_ext + 1);
6166 		xfs_bmbt_get_all(adj_irecp, &adj_irec);
6167 		if (startoff + got.br_blockcount > adj_irec.br_startoff)
6168 			return -EINVAL;
6169 		/*
6170 		 * Unlike a left shift (which involves a hole punch),
6171 		 * a right shift does not modify extent neighbors
6172 		 * in any way. We should never find mergeable extents
6173 		 * in this scenario. Check anyways and warn if we
6174 		 * encounter two extents that could be one.
6175 		 */
6176 		if (xfs_bmse_can_merge(&got, &adj_irec, offset_shift_fsb))
6177 			WARN_ON_ONCE(1);
6178 	}
6179 	/*
6180 	 * Increment the extent index for the next iteration, update the start
6181 	 * offset of the in-core extent and update the btree if applicable.
6182 	 */
6183 update_current_ext:
6184 	if (direction == SHIFT_LEFT)
6185 		(*current_ext)++;
6186 	else
6187 		(*current_ext)--;
6188 	xfs_bmbt_set_startoff(gotp, startoff);
6189 	*logflags |= XFS_ILOG_CORE;
6190 	adj_irec = got;
6191 	if (!cur) {
6192 		*logflags |= XFS_ILOG_DEXT;
6193 		goto update_rmap;
6194 	}
6195 
6196 	error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
6197 				   got.br_blockcount, &i);
6198 	if (error)
6199 		return error;
6200 	XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
6201 
6202 	got.br_startoff = startoff;
6203 	error = xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
6204 			got.br_blockcount, got.br_state);
6205 	if (error)
6206 		return error;
6207 
6208 update_rmap:
6209 	/* update reverse mapping */
6210 	error = xfs_rmap_unmap_extent(mp, dfops, ip, whichfork, &adj_irec);
6211 	if (error)
6212 		return error;
6213 	adj_irec.br_startoff = startoff;
6214 	return xfs_rmap_map_extent(mp, dfops, ip, whichfork, &adj_irec);
6215 }
6216 
6217 /*
6218  * Shift extent records to the left/right to cover/create a hole.
6219  *
6220  * The maximum number of extents to be shifted in a single operation is
6221  * @num_exts. @stop_fsb specifies the file offset at which to stop shift and the
6222  * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
6223  * is the length by which each extent is shifted. If there is no hole to shift
6224  * the extents into, this will be considered invalid operation and we abort
6225  * immediately.
6226  */
6227 int
xfs_bmap_shift_extents(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t * next_fsb,xfs_fileoff_t offset_shift_fsb,int * done,xfs_fileoff_t stop_fsb,xfs_fsblock_t * firstblock,struct xfs_defer_ops * dfops,enum shift_direction direction,int num_exts)6228 xfs_bmap_shift_extents(
6229 	struct xfs_trans	*tp,
6230 	struct xfs_inode	*ip,
6231 	xfs_fileoff_t		*next_fsb,
6232 	xfs_fileoff_t		offset_shift_fsb,
6233 	int			*done,
6234 	xfs_fileoff_t		stop_fsb,
6235 	xfs_fsblock_t		*firstblock,
6236 	struct xfs_defer_ops	*dfops,
6237 	enum shift_direction	direction,
6238 	int			num_exts)
6239 {
6240 	struct xfs_btree_cur		*cur = NULL;
6241 	struct xfs_bmbt_rec_host	*gotp;
6242 	struct xfs_bmbt_irec            got;
6243 	struct xfs_mount		*mp = ip->i_mount;
6244 	struct xfs_ifork		*ifp;
6245 	xfs_extnum_t			nexts = 0;
6246 	xfs_extnum_t			current_ext;
6247 	xfs_extnum_t			total_extents;
6248 	xfs_extnum_t			stop_extent;
6249 	int				error = 0;
6250 	int				whichfork = XFS_DATA_FORK;
6251 	int				logflags = 0;
6252 
6253 	if (unlikely(XFS_TEST_ERROR(
6254 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6255 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6256 	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
6257 		XFS_ERROR_REPORT("xfs_bmap_shift_extents",
6258 				 XFS_ERRLEVEL_LOW, mp);
6259 		return -EFSCORRUPTED;
6260 	}
6261 
6262 	if (XFS_FORCED_SHUTDOWN(mp))
6263 		return -EIO;
6264 
6265 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
6266 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6267 	ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT);
6268 	ASSERT(*next_fsb != NULLFSBLOCK || direction == SHIFT_RIGHT);
6269 
6270 	ifp = XFS_IFORK_PTR(ip, whichfork);
6271 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6272 		/* Read in all the extents */
6273 		error = xfs_iread_extents(tp, ip, whichfork);
6274 		if (error)
6275 			return error;
6276 	}
6277 
6278 	if (ifp->if_flags & XFS_IFBROOT) {
6279 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6280 		cur->bc_private.b.firstblock = *firstblock;
6281 		cur->bc_private.b.dfops = dfops;
6282 		cur->bc_private.b.flags = 0;
6283 	}
6284 
6285 	/*
6286 	 * There may be delalloc extents in the data fork before the range we
6287 	 * are collapsing out, so we cannot use the count of real extents here.
6288 	 * Instead we have to calculate it from the incore fork.
6289 	 */
6290 	total_extents = xfs_iext_count(ifp);
6291 	if (total_extents == 0) {
6292 		*done = 1;
6293 		goto del_cursor;
6294 	}
6295 
6296 	/*
6297 	 * In case of first right shift, we need to initialize next_fsb
6298 	 */
6299 	if (*next_fsb == NULLFSBLOCK) {
6300 		gotp = xfs_iext_get_ext(ifp, total_extents - 1);
6301 		xfs_bmbt_get_all(gotp, &got);
6302 		*next_fsb = got.br_startoff;
6303 		if (stop_fsb > *next_fsb) {
6304 			*done = 1;
6305 			goto del_cursor;
6306 		}
6307 	}
6308 
6309 	/* Lookup the extent index at which we have to stop */
6310 	if (direction == SHIFT_RIGHT) {
6311 		gotp = xfs_iext_bno_to_ext(ifp, stop_fsb, &stop_extent);
6312 		/* Make stop_extent exclusive of shift range */
6313 		stop_extent--;
6314 	} else
6315 		stop_extent = total_extents;
6316 
6317 	/*
6318 	 * Look up the extent index for the fsb where we start shifting. We can
6319 	 * henceforth iterate with current_ext as extent list changes are locked
6320 	 * out via ilock.
6321 	 *
6322 	 * gotp can be null in 2 cases: 1) if there are no extents or 2)
6323 	 * *next_fsb lies in a hole beyond which there are no extents. Either
6324 	 * way, we are done.
6325 	 */
6326 	gotp = xfs_iext_bno_to_ext(ifp, *next_fsb, &current_ext);
6327 	if (!gotp) {
6328 		*done = 1;
6329 		goto del_cursor;
6330 	}
6331 
6332 	/* some sanity checking before we finally start shifting extents */
6333 	if ((direction == SHIFT_LEFT && current_ext >= stop_extent) ||
6334 	     (direction == SHIFT_RIGHT && current_ext <= stop_extent)) {
6335 		error = -EIO;
6336 		goto del_cursor;
6337 	}
6338 
6339 	while (nexts++ < num_exts) {
6340 		error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
6341 					   &current_ext, gotp, cur, &logflags,
6342 					   direction, dfops);
6343 		if (error)
6344 			goto del_cursor;
6345 		/*
6346 		 * If there was an extent merge during the shift, the extent
6347 		 * count can change. Update the total and grade the next record.
6348 		 */
6349 		if (direction == SHIFT_LEFT) {
6350 			total_extents = xfs_iext_count(ifp);
6351 			stop_extent = total_extents;
6352 		}
6353 
6354 		if (current_ext == stop_extent) {
6355 			*done = 1;
6356 			*next_fsb = NULLFSBLOCK;
6357 			break;
6358 		}
6359 		gotp = xfs_iext_get_ext(ifp, current_ext);
6360 	}
6361 
6362 	if (!*done) {
6363 		xfs_bmbt_get_all(gotp, &got);
6364 		*next_fsb = got.br_startoff;
6365 	}
6366 
6367 del_cursor:
6368 	if (cur)
6369 		xfs_btree_del_cursor(cur,
6370 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6371 
6372 	if (logflags)
6373 		xfs_trans_log_inode(tp, ip, logflags);
6374 
6375 	return error;
6376 }
6377 
6378 /*
6379  * Splits an extent into two extents at split_fsb block such that it is
6380  * the first block of the current_ext. @current_ext is a target extent
6381  * to be split. @split_fsb is a block where the extents is split.
6382  * If split_fsb lies in a hole or the first block of extents, just return 0.
6383  */
6384 STATIC int
xfs_bmap_split_extent_at(struct xfs_trans * tp,struct xfs_inode * ip,xfs_fileoff_t split_fsb,xfs_fsblock_t * firstfsb,struct xfs_defer_ops * dfops)6385 xfs_bmap_split_extent_at(
6386 	struct xfs_trans	*tp,
6387 	struct xfs_inode	*ip,
6388 	xfs_fileoff_t		split_fsb,
6389 	xfs_fsblock_t		*firstfsb,
6390 	struct xfs_defer_ops	*dfops)
6391 {
6392 	int				whichfork = XFS_DATA_FORK;
6393 	struct xfs_btree_cur		*cur = NULL;
6394 	struct xfs_bmbt_rec_host	*gotp;
6395 	struct xfs_bmbt_irec		got;
6396 	struct xfs_bmbt_irec		new; /* split extent */
6397 	struct xfs_mount		*mp = ip->i_mount;
6398 	struct xfs_ifork		*ifp;
6399 	xfs_fsblock_t			gotblkcnt; /* new block count for got */
6400 	xfs_extnum_t			current_ext;
6401 	int				error = 0;
6402 	int				logflags = 0;
6403 	int				i = 0;
6404 
6405 	if (unlikely(XFS_TEST_ERROR(
6406 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
6407 	     XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
6408 	     mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
6409 		XFS_ERROR_REPORT("xfs_bmap_split_extent_at",
6410 				 XFS_ERRLEVEL_LOW, mp);
6411 		return -EFSCORRUPTED;
6412 	}
6413 
6414 	if (XFS_FORCED_SHUTDOWN(mp))
6415 		return -EIO;
6416 
6417 	ifp = XFS_IFORK_PTR(ip, whichfork);
6418 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
6419 		/* Read in all the extents */
6420 		error = xfs_iread_extents(tp, ip, whichfork);
6421 		if (error)
6422 			return error;
6423 	}
6424 
6425 	/*
6426 	 * gotp can be null in 2 cases: 1) if there are no extents
6427 	 * or 2) split_fsb lies in a hole beyond which there are
6428 	 * no extents. Either way, we are done.
6429 	 */
6430 	gotp = xfs_iext_bno_to_ext(ifp, split_fsb, &current_ext);
6431 	if (!gotp)
6432 		return 0;
6433 
6434 	xfs_bmbt_get_all(gotp, &got);
6435 
6436 	/*
6437 	 * Check split_fsb lies in a hole or the start boundary offset
6438 	 * of the extent.
6439 	 */
6440 	if (got.br_startoff >= split_fsb)
6441 		return 0;
6442 
6443 	gotblkcnt = split_fsb - got.br_startoff;
6444 	new.br_startoff = split_fsb;
6445 	new.br_startblock = got.br_startblock + gotblkcnt;
6446 	new.br_blockcount = got.br_blockcount - gotblkcnt;
6447 	new.br_state = got.br_state;
6448 
6449 	if (ifp->if_flags & XFS_IFBROOT) {
6450 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6451 		cur->bc_private.b.firstblock = *firstfsb;
6452 		cur->bc_private.b.dfops = dfops;
6453 		cur->bc_private.b.flags = 0;
6454 		error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
6455 				got.br_startblock,
6456 				got.br_blockcount,
6457 				&i);
6458 		if (error)
6459 			goto del_cursor;
6460 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6461 	}
6462 
6463 	xfs_bmbt_set_blockcount(gotp, gotblkcnt);
6464 	got.br_blockcount = gotblkcnt;
6465 
6466 	logflags = XFS_ILOG_CORE;
6467 	if (cur) {
6468 		error = xfs_bmbt_update(cur, got.br_startoff,
6469 				got.br_startblock,
6470 				got.br_blockcount,
6471 				got.br_state);
6472 		if (error)
6473 			goto del_cursor;
6474 	} else
6475 		logflags |= XFS_ILOG_DEXT;
6476 
6477 	/* Add new extent */
6478 	current_ext++;
6479 	xfs_iext_insert(ip, current_ext, 1, &new, 0);
6480 	XFS_IFORK_NEXT_SET(ip, whichfork,
6481 			   XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
6482 
6483 	if (cur) {
6484 		error = xfs_bmbt_lookup_eq(cur, new.br_startoff,
6485 				new.br_startblock, new.br_blockcount,
6486 				&i);
6487 		if (error)
6488 			goto del_cursor;
6489 		XFS_WANT_CORRUPTED_GOTO(mp, i == 0, del_cursor);
6490 		cur->bc_rec.b.br_state = new.br_state;
6491 
6492 		error = xfs_btree_insert(cur, &i);
6493 		if (error)
6494 			goto del_cursor;
6495 		XFS_WANT_CORRUPTED_GOTO(mp, i == 1, del_cursor);
6496 	}
6497 
6498 	/*
6499 	 * Convert to a btree if necessary.
6500 	 */
6501 	if (xfs_bmap_needs_btree(ip, whichfork)) {
6502 		int tmp_logflags; /* partial log flag return val */
6503 
6504 		ASSERT(cur == NULL);
6505 		error = xfs_bmap_extents_to_btree(tp, ip, firstfsb, dfops,
6506 				&cur, 0, &tmp_logflags, whichfork);
6507 		logflags |= tmp_logflags;
6508 	}
6509 
6510 del_cursor:
6511 	if (cur) {
6512 		cur->bc_private.b.allocated = 0;
6513 		xfs_btree_del_cursor(cur,
6514 				error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
6515 	}
6516 
6517 	if (logflags)
6518 		xfs_trans_log_inode(tp, ip, logflags);
6519 	return error;
6520 }
6521 
6522 int
xfs_bmap_split_extent(struct xfs_inode * ip,xfs_fileoff_t split_fsb)6523 xfs_bmap_split_extent(
6524 	struct xfs_inode        *ip,
6525 	xfs_fileoff_t           split_fsb)
6526 {
6527 	struct xfs_mount        *mp = ip->i_mount;
6528 	struct xfs_trans        *tp;
6529 	struct xfs_defer_ops    dfops;
6530 	xfs_fsblock_t           firstfsb;
6531 	int                     error;
6532 
6533 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
6534 			XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
6535 	if (error)
6536 		return error;
6537 
6538 	xfs_ilock(ip, XFS_ILOCK_EXCL);
6539 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
6540 
6541 	xfs_defer_init(&dfops, &firstfsb);
6542 
6543 	error = xfs_bmap_split_extent_at(tp, ip, split_fsb,
6544 			&firstfsb, &dfops);
6545 	if (error)
6546 		goto out;
6547 
6548 	error = xfs_defer_finish(&tp, &dfops, NULL);
6549 	if (error)
6550 		goto out;
6551 
6552 	return xfs_trans_commit(tp);
6553 
6554 out:
6555 	xfs_defer_cancel(&dfops);
6556 	xfs_trans_cancel(tp);
6557 	return error;
6558 }
6559 
6560 /* Deferred mapping is only for real extents in the data fork. */
6561 static bool
xfs_bmap_is_update_needed(struct xfs_bmbt_irec * bmap)6562 xfs_bmap_is_update_needed(
6563 	struct xfs_bmbt_irec	*bmap)
6564 {
6565 	return  bmap->br_startblock != HOLESTARTBLOCK &&
6566 		bmap->br_startblock != DELAYSTARTBLOCK;
6567 }
6568 
6569 /* Record a bmap intent. */
6570 static int
__xfs_bmap_add(struct xfs_mount * mp,struct xfs_defer_ops * dfops,enum xfs_bmap_intent_type type,struct xfs_inode * ip,int whichfork,struct xfs_bmbt_irec * bmap)6571 __xfs_bmap_add(
6572 	struct xfs_mount		*mp,
6573 	struct xfs_defer_ops		*dfops,
6574 	enum xfs_bmap_intent_type	type,
6575 	struct xfs_inode		*ip,
6576 	int				whichfork,
6577 	struct xfs_bmbt_irec		*bmap)
6578 {
6579 	int				error;
6580 	struct xfs_bmap_intent		*bi;
6581 
6582 	trace_xfs_bmap_defer(mp,
6583 			XFS_FSB_TO_AGNO(mp, bmap->br_startblock),
6584 			type,
6585 			XFS_FSB_TO_AGBNO(mp, bmap->br_startblock),
6586 			ip->i_ino, whichfork,
6587 			bmap->br_startoff,
6588 			bmap->br_blockcount,
6589 			bmap->br_state);
6590 
6591 	bi = kmem_alloc(sizeof(struct xfs_bmap_intent), KM_SLEEP | KM_NOFS);
6592 	INIT_LIST_HEAD(&bi->bi_list);
6593 	bi->bi_type = type;
6594 	bi->bi_owner = ip;
6595 	bi->bi_whichfork = whichfork;
6596 	bi->bi_bmap = *bmap;
6597 
6598 	error = xfs_defer_join(dfops, bi->bi_owner);
6599 	if (error) {
6600 		kmem_free(bi);
6601 		return error;
6602 	}
6603 
6604 	xfs_defer_add(dfops, XFS_DEFER_OPS_TYPE_BMAP, &bi->bi_list);
6605 	return 0;
6606 }
6607 
6608 /* Map an extent into a file. */
6609 int
xfs_bmap_map_extent(struct xfs_mount * mp,struct xfs_defer_ops * dfops,struct xfs_inode * ip,struct xfs_bmbt_irec * PREV)6610 xfs_bmap_map_extent(
6611 	struct xfs_mount	*mp,
6612 	struct xfs_defer_ops	*dfops,
6613 	struct xfs_inode	*ip,
6614 	struct xfs_bmbt_irec	*PREV)
6615 {
6616 	if (!xfs_bmap_is_update_needed(PREV))
6617 		return 0;
6618 
6619 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_MAP, ip,
6620 			XFS_DATA_FORK, PREV);
6621 }
6622 
6623 /* Unmap an extent out of a file. */
6624 int
xfs_bmap_unmap_extent(struct xfs_mount * mp,struct xfs_defer_ops * dfops,struct xfs_inode * ip,struct xfs_bmbt_irec * PREV)6625 xfs_bmap_unmap_extent(
6626 	struct xfs_mount	*mp,
6627 	struct xfs_defer_ops	*dfops,
6628 	struct xfs_inode	*ip,
6629 	struct xfs_bmbt_irec	*PREV)
6630 {
6631 	if (!xfs_bmap_is_update_needed(PREV))
6632 		return 0;
6633 
6634 	return __xfs_bmap_add(mp, dfops, XFS_BMAP_UNMAP, ip,
6635 			XFS_DATA_FORK, PREV);
6636 }
6637 
6638 /*
6639  * Process one of the deferred bmap operations.  We pass back the
6640  * btree cursor to maintain our lock on the bmapbt between calls.
6641  */
6642 int
xfs_bmap_finish_one(struct xfs_trans * tp,struct xfs_defer_ops * dfops,struct xfs_inode * ip,enum xfs_bmap_intent_type type,int whichfork,xfs_fileoff_t startoff,xfs_fsblock_t startblock,xfs_filblks_t * blockcount,xfs_exntst_t state)6643 xfs_bmap_finish_one(
6644 	struct xfs_trans		*tp,
6645 	struct xfs_defer_ops		*dfops,
6646 	struct xfs_inode		*ip,
6647 	enum xfs_bmap_intent_type	type,
6648 	int				whichfork,
6649 	xfs_fileoff_t			startoff,
6650 	xfs_fsblock_t			startblock,
6651 	xfs_filblks_t			*blockcount,
6652 	xfs_exntst_t			state)
6653 {
6654 	struct xfs_bmbt_irec		bmap;
6655 	int				nimaps = 1;
6656 	xfs_fsblock_t			firstfsb;
6657 	int				flags = XFS_BMAPI_REMAP;
6658 	int				error = 0;
6659 
6660 	bmap.br_startblock = startblock;
6661 	bmap.br_startoff = startoff;
6662 	bmap.br_blockcount = *blockcount;
6663 	bmap.br_state = state;
6664 
6665 	/*
6666 	 * firstfsb is tied to the transaction lifetime and is used to
6667 	 * ensure correct AG locking order and schedule work item
6668 	 * continuations.  XFS_BUI_MAX_FAST_EXTENTS (== 1) restricts us
6669 	 * to only making one bmap call per transaction, so it should
6670 	 * be safe to have it as a local variable here.
6671 	 */
6672 	firstfsb = NULLFSBLOCK;
6673 
6674 	trace_xfs_bmap_deferred(tp->t_mountp,
6675 			XFS_FSB_TO_AGNO(tp->t_mountp, startblock), type,
6676 			XFS_FSB_TO_AGBNO(tp->t_mountp, startblock),
6677 			ip->i_ino, whichfork, startoff, *blockcount, state);
6678 
6679 	if (whichfork != XFS_DATA_FORK && whichfork != XFS_ATTR_FORK)
6680 		return -EFSCORRUPTED;
6681 	if (whichfork == XFS_ATTR_FORK)
6682 		flags |= XFS_BMAPI_ATTRFORK;
6683 
6684 	if (XFS_TEST_ERROR(false, tp->t_mountp,
6685 			XFS_ERRTAG_BMAP_FINISH_ONE,
6686 			XFS_RANDOM_BMAP_FINISH_ONE))
6687 		return -EIO;
6688 
6689 	switch (type) {
6690 	case XFS_BMAP_MAP:
6691 		firstfsb = bmap.br_startblock;
6692 		error = xfs_bmapi_write(tp, ip, bmap.br_startoff,
6693 					bmap.br_blockcount, flags, &firstfsb,
6694 					bmap.br_blockcount, &bmap, &nimaps,
6695 					dfops);
6696 		*blockcount = 0;
6697 		break;
6698 	case XFS_BMAP_UNMAP:
6699 		error = __xfs_bunmapi(tp, ip, startoff, blockcount,
6700 				XFS_BMAPI_REMAP, 1, &firstfsb, dfops);
6701 		break;
6702 	default:
6703 		ASSERT(0);
6704 		error = -EFSCORRUPTED;
6705 	}
6706 
6707 	return error;
6708 }
6709