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