1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* Copyright (c) 2000-2006 Silicon Graphics, Inc.
4	* All Rights Reserved.
5	*/
6	#include "xfs.h"
7	#include "xfs_fs.h"
8	#include "xfs_shared.h"
9	#include "xfs_format.h"
10	#include "xfs_log_format.h"
11	#include "xfs_trans_resv.h"
12	#include "xfs_bit.h"
13	#include "xfs_sb.h"
14	#include "xfs_mount.h"
15	#include "xfs_defer.h"
16	#include "xfs_dir2.h"
17	#include "xfs_inode.h"
18	#include "xfs_btree.h"
19	#include "xfs_trans.h"
20	#include "xfs_alloc.h"
21	#include "xfs_bmap.h"
22	#include "xfs_bmap_util.h"
23	#include "xfs_bmap_btree.h"
24	#include "xfs_rtbitmap.h"
25	#include "xfs_errortag.h"
26	#include "xfs_error.h"
27	#include "xfs_quota.h"
28	#include "xfs_trans_space.h"
29	#include "xfs_buf_item.h"
30	#include "xfs_trace.h"
31	#include "xfs_attr_leaf.h"
32	#include "xfs_filestream.h"
33	#include "xfs_rmap.h"
34	#include "xfs_ag.h"
35	#include "xfs_ag_resv.h"
36	#include "xfs_refcount.h"
37	#include "xfs_icache.h"
38	#include "xfs_iomap.h"
39	#include "xfs_health.h"
40	#include "xfs_bmap_item.h"
41	#include "xfs_symlink_remote.h"
42
43	struct kmem_cache *xfs_bmap_intent_cache;
44
45	/*
46	* Miscellaneous helper functions
47	*/
48
49	/*
50	* Compute and fill in the value of the maximum depth of a bmap btree
51	* in this filesystem. Done once, during mount.
52	*/
53	void
54	xfs_bmap_compute_maxlevels(
55	xfs_mount_t *mp, /* file system mount structure */
56	int whichfork) /* data or attr fork */
57	{
58	uint64_t maxblocks; /* max blocks at this level */
59	xfs_extnum_t maxleafents; /* max leaf entries possible */
60	int level; /* btree level */
61	int maxrootrecs; /* max records in root block */
62	int minleafrecs; /* min records in leaf block */
63	int minnoderecs; /* min records in node block */
64	int sz; /* root block size */
65
66	/*
67	* The maximum number of extents in a fork, hence the maximum number of
68	* leaf entries, is controlled by the size of the on-disk extent count.
69	*
70	* Note that we can no longer assume that if we are in ATTR1 that the
71	* fork offset of all the inodes will be
72	* (xfs_default_attroffset(ip) >> 3) because we could have mounted with
73	* ATTR2 and then mounted back with ATTR1, keeping the i_forkoff's fixed
74	* but probably at various positions. Therefore, for both ATTR1 and
75	* ATTR2 we have to assume the worst case scenario of a minimum size
76	* available.
77	*/
78	maxleafents = xfs_iext_max_nextents(xfs_has_large_extent_counts(mp),
79	whichfork);
80	if (whichfork == XFS_DATA_FORK)
81	sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
82	else
83	sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
84
85	maxrootrecs = xfs_bmdr_maxrecs(blocklen: sz, leaf: 0);
86	minleafrecs = mp->m_bmap_dmnr[0];
87	minnoderecs = mp->m_bmap_dmnr[1];
88	maxblocks = howmany_64(maxleafents, minleafrecs);
89	for (level = 1; maxblocks > 1; level++) {
90	if (maxblocks <= maxrootrecs)
91	maxblocks = 1;
92	else
93	maxblocks = howmany_64(maxblocks, minnoderecs);
94	}
95	mp->m_bm_maxlevels[whichfork] = level;
96	ASSERT(mp->m_bm_maxlevels[whichfork] <= xfs_bmbt_maxlevels_ondisk());
97	}
98
99	unsigned int
100	xfs_bmap_compute_attr_offset(
101	struct xfs_mount *mp)
102	{
103	if (mp->m_sb.sb_inodesize == 256)
104	return XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(MINABTPTRS);
105	return XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
106	}
107
108	STATIC int /* error */
109	xfs_bmbt_lookup_eq(
110	struct xfs_btree_cur *cur,
111	struct xfs_bmbt_irec *irec,
112	int *stat) /* success/failure */
113	{
114	cur->bc_rec.b = *irec;
115	return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
116	}
117
118	STATIC int /* error */
119	xfs_bmbt_lookup_first(
120	struct xfs_btree_cur *cur,
121	int *stat) /* success/failure */
122	{
123	cur->bc_rec.b.br_startoff = 0;
124	cur->bc_rec.b.br_startblock = 0;
125	cur->bc_rec.b.br_blockcount = 0;
126	return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
127	}
128
129	/*
130	* Check if the inode needs to be converted to btree format.
131	*/
132	static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
133	{
134	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
135
136	return whichfork != XFS_COW_FORK &&
137	ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
138	ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork);
139	}
140
141	/*
142	* Check if the inode should be converted to extent format.
143	*/
144	static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
145	{
146	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
147
148	return whichfork != XFS_COW_FORK &&
149	ifp->if_format == XFS_DINODE_FMT_BTREE &&
150	ifp->if_nextents <= XFS_IFORK_MAXEXT(ip, whichfork);
151	}
152
153	/*
154	* Update the record referred to by cur to the value given by irec
155	* This either works (return 0) or gets an EFSCORRUPTED error.
156	*/
157	STATIC int
158	xfs_bmbt_update(
159	struct xfs_btree_cur *cur,
160	struct xfs_bmbt_irec *irec)
161	{
162	union xfs_btree_rec rec;
163
164	xfs_bmbt_disk_set_all(r: &rec.bmbt, s: irec);
165	return xfs_btree_update(cur, &rec);
166	}
167
168	/*
169	* Compute the worst-case number of indirect blocks that will be used
170	* for ip's delayed extent of length "len".
171	*/
172	STATIC xfs_filblks_t
173	xfs_bmap_worst_indlen(
174	xfs_inode_t *ip, /* incore inode pointer */
175	xfs_filblks_t len) /* delayed extent length */
176	{
177	int level; /* btree level number */
178	int maxrecs; /* maximum record count at this level */
179	xfs_mount_t *mp; /* mount structure */
180	xfs_filblks_t rval; /* return value */
181
182	mp = ip->i_mount;
183	maxrecs = mp->m_bmap_dmxr[0];
184	for (level = 0, rval = 0;
185	level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
186	level++) {
187	len += maxrecs - 1;
188	do_div(len, maxrecs);
189	rval += len;
190	if (len == 1)
191	return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
192	level - 1;
193	if (level == 0)
194	maxrecs = mp->m_bmap_dmxr[1];
195	}
196	return rval;
197	}
198
199	/*
200	* Calculate the default attribute fork offset for newly created inodes.
201	*/
202	uint
203	xfs_default_attroffset(
204	struct xfs_inode *ip)
205	{
206	if (ip->i_df.if_format == XFS_DINODE_FMT_DEV)
207	return roundup(sizeof(xfs_dev_t), 8);
208	return M_IGEO(ip->i_mount)->attr_fork_offset;
209	}
210
211	/*
212	* Helper routine to reset inode i_forkoff field when switching attribute fork
213	* from local to extent format - we reset it where possible to make space
214	* available for inline data fork extents.
215	*/
216	STATIC void
217	xfs_bmap_forkoff_reset(
218	xfs_inode_t *ip,
219	int whichfork)
220	{
221	if (whichfork == XFS_ATTR_FORK &&
222	ip->i_df.if_format != XFS_DINODE_FMT_DEV &&
223	ip->i_df.if_format != XFS_DINODE_FMT_BTREE) {
224	uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
225
226	if (dfl_forkoff > ip->i_forkoff)
227	ip->i_forkoff = dfl_forkoff;
228	}
229	}
230
231	static int
232	xfs_bmap_read_buf(
233	struct xfs_mount *mp, /* file system mount point */
234	struct xfs_trans *tp, /* transaction pointer */
235	xfs_fsblock_t fsbno, /* file system block number */
236	struct xfs_buf **bpp) /* buffer for fsbno */
237	{
238	struct xfs_buf *bp; /* return value */
239	int error;
240
241	if (!xfs_verify_fsbno(mp, fsbno))
242	return -EFSCORRUPTED;
243	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
244	XFS_FSB_TO_DADDR(mp, fsbno), mp->m_bsize, 0, &bp,
245	&xfs_bmbt_buf_ops);
246	if (!error) {
247	xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
248	*bpp = bp;
249	}
250	return error;
251	}
252
253	#ifdef DEBUG
254	STATIC struct xfs_buf *
255	xfs_bmap_get_bp(
256	struct xfs_btree_cur *cur,
257	xfs_fsblock_t bno)
258	{
259	struct xfs_log_item *lip;
260	int i;
261
262	if (!cur)
263	return NULL;
264
265	for (i = 0; i < cur->bc_maxlevels; i++) {
266	if (!cur->bc_levels[i].bp)
267	break;
268	if (xfs_buf_daddr(cur->bc_levels[i].bp) == bno)
269	return cur->bc_levels[i].bp;
270	}
271
272	/* Chase down all the log items to see if the bp is there */
273	list_for_each_entry(lip, &cur->bc_tp->t_items, li_trans) {
274	struct xfs_buf_log_item *bip = (struct xfs_buf_log_item *)lip;
275
276	if (bip->bli_item.li_type == XFS_LI_BUF &&
277	xfs_buf_daddr(bip->bli_buf) == bno)
278	return bip->bli_buf;
279	}
280
281	return NULL;
282	}
283
284	STATIC void
285	xfs_check_block(
286	struct xfs_btree_block *block,
287	xfs_mount_t *mp,
288	int root,
289	short sz)
290	{
291	int i, j, dmxr;
292	__be64 *pp, *thispa; /* pointer to block address */
293	xfs_bmbt_key_t *prevp, *keyp;
294
295	ASSERT(be16_to_cpu(block->bb_level) > 0);
296
297	prevp = NULL;
298	for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
299	dmxr = mp->m_bmap_dmxr[0];
300	keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
301
302	if (prevp) {
303	ASSERT(be64_to_cpu(prevp->br_startoff) <
304	be64_to_cpu(keyp->br_startoff));
305	}
306	prevp = keyp;
307
308	/*
309	* Compare the block numbers to see if there are dups.
310	*/
311	if (root)
312	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
313	else
314	pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
315
316	for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
317	if (root)
318	thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
319	else
320	thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
321	if (*thispa == *pp) {
322	xfs_warn(mp, "%s: thispa(%d) == pp(%d) %lld",
323	__func__, j, i,
324	(unsigned long long)be64_to_cpu(*thispa));
325	xfs_err(mp, "%s: ptrs are equal in node\n",
326	__func__);
327	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
328	}
329	}
330	}
331	}
332
333	/*
334	* Check that the extents for the inode ip are in the right order in all
335	* btree leaves. THis becomes prohibitively expensive for large extent count
336	* files, so don't bother with inodes that have more than 10,000 extents in
337	* them. The btree record ordering checks will still be done, so for such large
338	* bmapbt constructs that is going to catch most corruptions.
339	*/
340	STATIC void
341	xfs_bmap_check_leaf_extents(
342	struct xfs_btree_cur *cur, /* btree cursor or null */
343	xfs_inode_t *ip, /* incore inode pointer */
344	int whichfork) /* data or attr fork */
345	{
346	struct xfs_mount *mp = ip->i_mount;
347	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
348	struct xfs_btree_block *block; /* current btree block */
349	xfs_fsblock_t bno; /* block # of "block" */
350	struct xfs_buf *bp; /* buffer for "block" */
351	int error; /* error return value */
352	xfs_extnum_t i=0, j; /* index into the extents list */
353	int level; /* btree level, for checking */
354	__be64 *pp; /* pointer to block address */
355	xfs_bmbt_rec_t *ep; /* pointer to current extent */
356	xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
357	xfs_bmbt_rec_t *nextp; /* pointer to next extent */
358	int bp_release = 0;
359
360	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
361	return;
362
363	/* skip large extent count inodes */
364	if (ip->i_df.if_nextents > 10000)
365	return;
366
367	bno = NULLFSBLOCK;
368	block = ifp->if_broot;
369	/*
370	* Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
371	*/
372	level = be16_to_cpu(block->bb_level);
373	ASSERT(level > 0);
374	xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
375	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
376	bno = be64_to_cpu(*pp);
377
378	ASSERT(bno != NULLFSBLOCK);
379	ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
380	ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
381
382	/*
383	* Go down the tree until leaf level is reached, following the first
384	* pointer (leftmost) at each level.
385	*/
386	while (level-- > 0) {
387	/* See if buf is in cur first */
388	bp_release = 0;
389	bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
390	if (!bp) {
391	bp_release = 1;
392	error = xfs_bmap_read_buf(mp, NULL, bno, &bp);
393	if (xfs_metadata_is_sick(error))
394	xfs_btree_mark_sick(cur);
395	if (error)
396	goto error_norelse;
397	}
398	block = XFS_BUF_TO_BLOCK(bp);
399	if (level == 0)
400	break;
401
402	/*
403	* Check this block for basic sanity (increasing keys and
404	* no duplicate blocks).
405	*/
406
407	xfs_check_block(block, mp, 0, 0);
408	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
409	bno = be64_to_cpu(*pp);
410	if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, bno))) {
411	xfs_btree_mark_sick(cur);
412	error = -EFSCORRUPTED;
413	goto error0;
414	}
415	if (bp_release) {
416	bp_release = 0;
417	xfs_trans_brelse(NULL, bp);
418	}
419	}
420
421	/*
422	* Here with bp and block set to the leftmost leaf node in the tree.
423	*/
424	i = 0;
425
426	/*
427	* Loop over all leaf nodes checking that all extents are in the right order.
428	*/
429	for (;;) {
430	xfs_fsblock_t nextbno;
431	xfs_extnum_t num_recs;
432
433
434	num_recs = xfs_btree_get_numrecs(block);
435
436	/*
437	* Read-ahead the next leaf block, if any.
438	*/
439
440	nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
441
442	/*
443	* Check all the extents to make sure they are OK.
444	* If we had a previous block, the last entry should
445	* conform with the first entry in this one.
446	*/
447
448	ep = XFS_BMBT_REC_ADDR(mp, block, 1);
449	if (i) {
450	ASSERT(xfs_bmbt_disk_get_startoff(&last) +
451	xfs_bmbt_disk_get_blockcount(&last) <=
452	xfs_bmbt_disk_get_startoff(ep));
453	}
454	for (j = 1; j < num_recs; j++) {
455	nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
456	ASSERT(xfs_bmbt_disk_get_startoff(ep) +
457	xfs_bmbt_disk_get_blockcount(ep) <=
458	xfs_bmbt_disk_get_startoff(nextp));
459	ep = nextp;
460	}
461
462	last = *ep;
463	i += num_recs;
464	if (bp_release) {
465	bp_release = 0;
466	xfs_trans_brelse(NULL, bp);
467	}
468	bno = nextbno;
469	/*
470	* If we've reached the end, stop.
471	*/
472	if (bno == NULLFSBLOCK)
473	break;
474
475	bp_release = 0;
476	bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
477	if (!bp) {
478	bp_release = 1;
479	error = xfs_bmap_read_buf(mp, NULL, bno, &bp);
480	if (xfs_metadata_is_sick(error))
481	xfs_btree_mark_sick(cur);
482	if (error)
483	goto error_norelse;
484	}
485	block = XFS_BUF_TO_BLOCK(bp);
486	}
487
488	return;
489
490	error0:
491	xfs_warn(mp, "%s: at error0", __func__);
492	if (bp_release)
493	xfs_trans_brelse(NULL, bp);
494	error_norelse:
495	xfs_warn(mp, "%s: BAD after btree leaves for %llu extents",
496	__func__, i);
497	xfs_err(mp, "%s: CORRUPTED BTREE OR SOMETHING", __func__);
498	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
499	return;
500	}
501
502	/*
503	* Validate that the bmbt_irecs being returned from bmapi are valid
504	* given the caller's original parameters. Specifically check the
505	* ranges of the returned irecs to ensure that they only extend beyond
506	* the given parameters if the XFS_BMAPI_ENTIRE flag was set.
507	*/
508	STATIC void
509	xfs_bmap_validate_ret(
510	xfs_fileoff_t bno,
511	xfs_filblks_t len,
512	uint32_t flags,
513	xfs_bmbt_irec_t *mval,
514	int nmap,
515	int ret_nmap)
516	{
517	int i; /* index to map values */
518
519	ASSERT(ret_nmap <= nmap);
520
521	for (i = 0; i < ret_nmap; i++) {
522	ASSERT(mval[i].br_blockcount > 0);
523	if (!(flags & XFS_BMAPI_ENTIRE)) {
524	ASSERT(mval[i].br_startoff >= bno);
525	ASSERT(mval[i].br_blockcount <= len);
526	ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
527	bno + len);
528	} else {
529	ASSERT(mval[i].br_startoff < bno + len);
530	ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
531	bno);
532	}
533	ASSERT(i == 0 ||
534	mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
535	mval[i].br_startoff);
536	ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
537	mval[i].br_startblock != HOLESTARTBLOCK);
538	ASSERT(mval[i].br_state == XFS_EXT_NORM ||
539	mval[i].br_state == XFS_EXT_UNWRITTEN);
540	}
541	}
542
543	#else
544	#define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
545	#define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap) do { } while (0)
546	#endif /* DEBUG */
547
548	/*
549	* Inode fork format manipulation functions
550	*/
551
552	/*
553	* Convert the inode format to extent format if it currently is in btree format,
554	* but the extent list is small enough that it fits into the extent format.
555	*
556	* Since the extents are already in-core, all we have to do is give up the space
557	* for the btree root and pitch the leaf block.
558	*/
559	STATIC int /* error */
560	xfs_bmap_btree_to_extents(
561	struct xfs_trans *tp, /* transaction pointer */
562	struct xfs_inode *ip, /* incore inode pointer */
563	struct xfs_btree_cur *cur, /* btree cursor */
564	int *logflagsp, /* inode logging flags */
565	int whichfork) /* data or attr fork */
566	{
567	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
568	struct xfs_mount *mp = ip->i_mount;
569	struct xfs_btree_block *rblock = ifp->if_broot;
570	struct xfs_btree_block *cblock;/* child btree block */
571	xfs_fsblock_t cbno; /* child block number */
572	struct xfs_buf *cbp; /* child block's buffer */
573	int error; /* error return value */
574	__be64 *pp; /* ptr to block address */
575	struct xfs_owner_info oinfo;
576
577	/* check if we actually need the extent format first: */
578	if (!xfs_bmap_wants_extents(ip, whichfork))
579	return 0;
580
581	ASSERT(cur);
582	ASSERT(whichfork != XFS_COW_FORK);
583	ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
584	ASSERT(be16_to_cpu(rblock->bb_level) == 1);
585	ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
586	ASSERT(xfs_bmbt_maxrecs(mp, blocklen: ifp->if_broot_bytes, leaf: 0) == 1);
587
588	pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
589	cbno = be64_to_cpu(*pp);
590	#ifdef DEBUG
591	if (XFS_IS_CORRUPT(cur->bc_mp, !xfs_verify_fsbno(mp, cbno))) {
592	xfs_btree_mark_sick(cur);
593	return -EFSCORRUPTED;
594	}
595	#endif
596	error = xfs_bmap_read_buf(mp, tp, cbno, &cbp);
597	if (xfs_metadata_is_sick(error))
598	xfs_btree_mark_sick(cur);
599	if (error)
600	return error;
601	cblock = XFS_BUF_TO_BLOCK(cbp);
602	if ((error = xfs_btree_check_block(cur, block: cblock, level: 0, bp: cbp)))
603	return error;
604
605	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
606	error = xfs_free_extent_later(cur->bc_tp, cbno, 1, &oinfo,
607	XFS_AG_RESV_NONE, false);
608	if (error)
609	return error;
610
611	ip->i_nblocks--;
612	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
613	xfs_trans_binval(tp, cbp);
614	if (cur->bc_levels[0].bp == cbp)
615	cur->bc_levels[0].bp = NULL;
616	xfs_iroot_realloc(ip, -1, whichfork);
617	ASSERT(ifp->if_broot == NULL);
618	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
619	*logflagsp |= XFS_ILOG_CORE | xfs_ilog_fext(w: whichfork);
620	return 0;
621	}
622
623	/*
624	* Convert an extents-format file into a btree-format file.
625	* The new file will have a root block (in the inode) and a single child block.
626	*/
627	STATIC int /* error */
628	xfs_bmap_extents_to_btree(
629	struct xfs_trans *tp, /* transaction pointer */
630	struct xfs_inode *ip, /* incore inode pointer */
631	struct xfs_btree_cur **curp, /* cursor returned to caller */
632	int wasdel, /* converting a delayed alloc */
633	int *logflagsp, /* inode logging flags */
634	int whichfork) /* data or attr fork */
635	{
636	struct xfs_btree_block *ablock; /* allocated (child) bt block */
637	struct xfs_buf *abp; /* buffer for ablock */
638	struct xfs_alloc_arg args; /* allocation arguments */
639	struct xfs_bmbt_rec *arp; /* child record pointer */
640	struct xfs_btree_block *block; /* btree root block */
641	struct xfs_btree_cur *cur; /* bmap btree cursor */
642	int error; /* error return value */
643	struct xfs_ifork *ifp; /* inode fork pointer */
644	struct xfs_bmbt_key *kp; /* root block key pointer */
645	struct xfs_mount *mp; /* mount structure */
646	xfs_bmbt_ptr_t *pp; /* root block address pointer */
647	struct xfs_iext_cursor icur;
648	struct xfs_bmbt_irec rec;
649	xfs_extnum_t cnt = 0;
650
651	mp = ip->i_mount;
652	ASSERT(whichfork != XFS_COW_FORK);
653	ifp = xfs_ifork_ptr(ip, whichfork);
654	ASSERT(ifp->if_format == XFS_DINODE_FMT_EXTENTS);
655
656	/*
657	* Make space in the inode incore. This needs to be undone if we fail
658	* to expand the root.
659	*/
660	xfs_iroot_realloc(ip, 1, whichfork);
661
662	/*
663	* Fill in the root.
664	*/
665	block = ifp->if_broot;
666	xfs_bmbt_init_block(ip, block, NULL, 1, 1);
667	/*
668	* Need a cursor. Can't allocate until bb_level is filled in.
669	*/
670	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
671	if (wasdel)
672	cur->bc_flags |= XFS_BTREE_BMBT_WASDEL;
673	/*
674	* Convert to a btree with two levels, one record in root.
675	*/
676	ifp->if_format = XFS_DINODE_FMT_BTREE;
677	memset(&args, 0, sizeof(args));
678	args.tp = tp;
679	args.mp = mp;
680	xfs_rmap_ino_bmbt_owner(&args.oinfo, ip->i_ino, whichfork);
681
682	args.minlen = args.maxlen = args.prod = 1;
683	args.wasdel = wasdel;
684	*logflagsp = 0;
685	error = xfs_alloc_vextent_start_ag(&args,
686	XFS_INO_TO_FSB(mp, ip->i_ino));
687	if (error)
688	goto out_root_realloc;
689
690	/*
691	* Allocation can't fail, the space was reserved.
692	*/
693	if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
694	error = -ENOSPC;
695	goto out_root_realloc;
696	}
697
698	cur->bc_bmap.allocated++;
699	ip->i_nblocks++;
700	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
701	error = xfs_trans_get_buf(tp, mp->m_ddev_targp,
702	XFS_FSB_TO_DADDR(mp, args.fsbno),
703	mp->m_bsize, 0, &abp);
704	if (error)
705	goto out_unreserve_dquot;
706
707	/*
708	* Fill in the child block.
709	*/
710	ablock = XFS_BUF_TO_BLOCK(abp);
711	xfs_bmbt_init_block(ip, ablock, abp, 0, 0);
712
713	for_each_xfs_iext(ifp, &icur, &rec) {
714	if (isnullstartblock(rec.br_startblock))
715	continue;
716	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1 + cnt);
717	xfs_bmbt_disk_set_all(r: arp, s: &rec);
718	cnt++;
719	}
720	ASSERT(cnt == ifp->if_nextents);
721	xfs_btree_set_numrecs(ablock, cnt);
722
723	/*
724	* Fill in the root key and pointer.
725	*/
726	kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
727	arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
728	kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
729	pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
730	be16_to_cpu(block->bb_level)));
731	*pp = cpu_to_be64(args.fsbno);
732
733	/*
734	* Do all this logging at the end so that
735	* the root is at the right level.
736	*/
737	xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
738	xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
739	ASSERT(*curp == NULL);
740	*curp = cur;
741	*logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(w: whichfork);
742	return 0;
743
744	out_unreserve_dquot:
745	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
746	out_root_realloc:
747	xfs_iroot_realloc(ip, -1, whichfork);
748	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
749	ASSERT(ifp->if_broot == NULL);
750	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
751
752	return error;
753	}
754
755	/*
756	* Convert a local file to an extents file.
757	* This code is out of bounds for data forks of regular files,
758	* since the file data needs to get logged so things will stay consistent.
759	* (The bmap-level manipulations are ok, though).
760	*/
761	void
762	xfs_bmap_local_to_extents_empty(
763	struct xfs_trans *tp,
764	struct xfs_inode *ip,
765	int whichfork)
766	{
767	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
768
769	ASSERT(whichfork != XFS_COW_FORK);
770	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
771	ASSERT(ifp->if_bytes == 0);
772	ASSERT(ifp->if_nextents == 0);
773
774	xfs_bmap_forkoff_reset(ip, whichfork);
775	ifp->if_data = NULL;
776	ifp->if_height = 0;
777	ifp->if_format = XFS_DINODE_FMT_EXTENTS;
778	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
779	}
780
781
782	STATIC int /* error */
783	xfs_bmap_local_to_extents(
784	xfs_trans_t *tp, /* transaction pointer */
785	xfs_inode_t *ip, /* incore inode pointer */
786	xfs_extlen_t total, /* total blocks needed by transaction */
787	int *logflagsp, /* inode logging flags */
788	int whichfork,
789	void (*init_fn)(struct xfs_trans *tp,
790	struct xfs_buf *bp,
791	struct xfs_inode *ip,
792	struct xfs_ifork *ifp))
793	{
794	int error = 0;
795	int flags; /* logging flags returned */
796	struct xfs_ifork *ifp; /* inode fork pointer */
797	xfs_alloc_arg_t args; /* allocation arguments */
798	struct xfs_buf *bp; /* buffer for extent block */
799	struct xfs_bmbt_irec rec;
800	struct xfs_iext_cursor icur;
801
802	/*
803	* We don't want to deal with the case of keeping inode data inline yet.
804	* So sending the data fork of a regular inode is invalid.
805	*/
806	ASSERT(!(S_ISREG(VFS_I(ip)->i_mode) && whichfork == XFS_DATA_FORK));
807	ifp = xfs_ifork_ptr(ip, whichfork);
808	ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
809
810	if (!ifp->if_bytes) {
811	xfs_bmap_local_to_extents_empty(tp: tp, ip: ip, whichfork);
812	flags = XFS_ILOG_CORE;
813	goto done;
814	}
815
816	flags = 0;
817	error = 0;
818	memset(&args, 0, sizeof(args));
819	args.tp = tp;
820	args.mp = ip->i_mount;
821	args.total = total;
822	args.minlen = args.maxlen = args.prod = 1;
823	xfs_rmap_ino_owner(&args.oinfo, ip->i_ino, whichfork, 0);
824
825	/*
826	* Allocate a block. We know we need only one, since the
827	* file currently fits in an inode.
828	*/
829	args.total = total;
830	args.minlen = args.maxlen = args.prod = 1;
831	error = xfs_alloc_vextent_start_ag(&args,
832	XFS_INO_TO_FSB(args.mp, ip->i_ino));
833	if (error)
834	goto done;
835
836	/* Can't fail, the space was reserved. */
837	ASSERT(args.fsbno != NULLFSBLOCK);
838	ASSERT(args.len == 1);
839	error = xfs_trans_get_buf(tp, args.mp->m_ddev_targp,
840	XFS_FSB_TO_DADDR(args.mp, args.fsbno),
841	args.mp->m_bsize, 0, &bp);
842	if (error)
843	goto done;
844
845	/*
846	* Initialize the block, copy the data and log the remote buffer.
847	*
848	* The callout is responsible for logging because the remote format
849	* might differ from the local format and thus we don't know how much to
850	* log here. Note that init_fn must also set the buffer log item type
851	* correctly.
852	*/
853	init_fn(tp, bp, ip, ifp);
854
855	/* account for the change in fork size */
856	xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
857	xfs_bmap_local_to_extents_empty(tp: tp, ip: ip, whichfork);
858	flags |= XFS_ILOG_CORE;
859
860	ifp->if_data = NULL;
861	ifp->if_height = 0;
862
863	rec.br_startoff = 0;
864	rec.br_startblock = args.fsbno;
865	rec.br_blockcount = 1;
866	rec.br_state = XFS_EXT_NORM;
867	xfs_iext_first(ifp, &icur);
868	xfs_iext_insert(ip, &icur, &rec, 0);
869
870	ifp->if_nextents = 1;
871	ip->i_nblocks = 1;
872	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
873	flags |= xfs_ilog_fext(w: whichfork);
874
875	done:
876	*logflagsp = flags;
877	return error;
878	}
879
880	/*
881	* Called from xfs_bmap_add_attrfork to handle btree format files.
882	*/
883	STATIC int /* error */
884	xfs_bmap_add_attrfork_btree(
885	xfs_trans_t *tp, /* transaction pointer */
886	xfs_inode_t *ip, /* incore inode pointer */
887	int *flags) /* inode logging flags */
888	{
889	struct xfs_btree_block *block = ip->i_df.if_broot;
890	struct xfs_btree_cur *cur; /* btree cursor */
891	int error; /* error return value */
892	xfs_mount_t *mp; /* file system mount struct */
893	int stat; /* newroot status */
894
895	mp = ip->i_mount;
896
897	if (XFS_BMAP_BMDR_SPACE(block) <= xfs_inode_data_fork_size(ip))
898	*flags |= XFS_ILOG_DBROOT;
899	else {
900	cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
901	error = xfs_bmbt_lookup_first(cur, &stat);
902	if (error)
903	goto error0;
904	/* must be at least one entry */
905	if (XFS_IS_CORRUPT(mp, stat != 1)) {
906	xfs_btree_mark_sick(cur);
907	error = -EFSCORRUPTED;
908	goto error0;
909	}
910	if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
911	goto error0;
912	if (stat == 0) {
913	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
914	return -ENOSPC;
915	}
916	cur->bc_bmap.allocated = 0;
917	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
918	}
919	return 0;
920	error0:
921	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
922	return error;
923	}
924
925	/*
926	* Called from xfs_bmap_add_attrfork to handle extents format files.
927	*/
928	STATIC int /* error */
929	xfs_bmap_add_attrfork_extents(
930	struct xfs_trans *tp, /* transaction pointer */
931	struct xfs_inode *ip, /* incore inode pointer */
932	int *flags) /* inode logging flags */
933	{
934	struct xfs_btree_cur *cur; /* bmap btree cursor */
935	int error; /* error return value */
936
937	if (ip->i_df.if_nextents * sizeof(struct xfs_bmbt_rec) <=
938	xfs_inode_data_fork_size(ip))
939	return 0;
940	cur = NULL;
941	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0, flags,
942	XFS_DATA_FORK);
943	if (cur) {
944	cur->bc_bmap.allocated = 0;
945	xfs_btree_del_cursor(cur, error);
946	}
947	return error;
948	}
949
950	/*
951	* Called from xfs_bmap_add_attrfork to handle local format files. Each
952	* different data fork content type needs a different callout to do the
953	* conversion. Some are basic and only require special block initialisation
954	* callouts for the data formating, others (directories) are so specialised they
955	* handle everything themselves.
956	*
957	* XXX (dgc): investigate whether directory conversion can use the generic
958	* formatting callout. It should be possible - it's just a very complex
959	* formatter.
960	*/
961	STATIC int /* error */
962	xfs_bmap_add_attrfork_local(
963	struct xfs_trans *tp, /* transaction pointer */
964	struct xfs_inode *ip, /* incore inode pointer */
965	int *flags) /* inode logging flags */
966	{
967	struct xfs_da_args dargs; /* args for dir/attr code */
968
969	if (ip->i_df.if_bytes <= xfs_inode_data_fork_size(ip))
970	return 0;
971
972	if (S_ISDIR(VFS_I(ip)->i_mode)) {
973	memset(&dargs, 0, sizeof(dargs));
974	dargs.geo = ip->i_mount->m_dir_geo;
975	dargs.dp = ip;
976	dargs.total = dargs.geo->fsbcount;
977	dargs.whichfork = XFS_DATA_FORK;
978	dargs.trans = tp;
979	return xfs_dir2_sf_to_block(args: &dargs);
980	}
981
982	if (S_ISLNK(VFS_I(ip)->i_mode))
983	return xfs_bmap_local_to_extents(tp, ip, 1, flags,
984	XFS_DATA_FORK,
985	xfs_symlink_local_to_remote);
986
987	/* should only be called for types that support local format data */
988	ASSERT(0);
989	xfs_bmap_mark_sick(ip, XFS_ATTR_FORK);
990	return -EFSCORRUPTED;
991	}
992
993	/*
994	* Set an inode attr fork offset based on the format of the data fork.
995	*/
996	static int
997	xfs_bmap_set_attrforkoff(
998	struct xfs_inode *ip,
999	int size,
1000	int *version)
1001	{
1002	int default_size = xfs_default_attroffset(ip) >> 3;
1003
1004	switch (ip->i_df.if_format) {
1005	case XFS_DINODE_FMT_DEV:
1006	ip->i_forkoff = default_size;
1007	break;
1008	case XFS_DINODE_FMT_LOCAL:
1009	case XFS_DINODE_FMT_EXTENTS:
1010	case XFS_DINODE_FMT_BTREE:
1011	ip->i_forkoff = xfs_attr_shortform_bytesfit(dp: ip, bytes: size);
1012	if (!ip->i_forkoff)
1013	ip->i_forkoff = default_size;
1014	else if (xfs_has_attr2(ip->i_mount) && version)
1015	*version = 2;
1016	break;
1017	default:
1018	ASSERT(0);
1019	return -EINVAL;
1020	}
1021
1022	return 0;
1023	}
1024
1025	/*
1026	* Convert inode from non-attributed to attributed.
1027	* Must not be in a transaction, ip must not be locked.
1028	*/
1029	int /* error code */
1030	xfs_bmap_add_attrfork(
1031	xfs_inode_t *ip, /* incore inode pointer */
1032	int size, /* space new attribute needs */
1033	int rsvd) /* xact may use reserved blks */
1034	{
1035	xfs_mount_t *mp; /* mount structure */
1036	xfs_trans_t *tp; /* transaction pointer */
1037	int blks; /* space reservation */
1038	int version = 1; /* superblock attr version */
1039	int logflags; /* logging flags */
1040	int error; /* error return value */
1041
1042	ASSERT(xfs_inode_has_attr_fork(ip) == 0);
1043
1044	mp = ip->i_mount;
1045	ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1046
1047	blks = XFS_ADDAFORK_SPACE_RES(mp);
1048
1049	error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_addafork, blks, 0,
1050	rsvd, &tp);
1051	if (error)
1052	return error;
1053	if (xfs_inode_has_attr_fork(ip))
1054	goto trans_cancel;
1055
1056	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1057	error = xfs_bmap_set_attrforkoff(ip: ip, size, version: &version);
1058	if (error)
1059	goto trans_cancel;
1060
1061	xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
1062	logflags = 0;
1063	switch (ip->i_df.if_format) {
1064	case XFS_DINODE_FMT_LOCAL:
1065	error = xfs_bmap_add_attrfork_local(tp, ip, &logflags);
1066	break;
1067	case XFS_DINODE_FMT_EXTENTS:
1068	error = xfs_bmap_add_attrfork_extents(tp, ip, &logflags);
1069	break;
1070	case XFS_DINODE_FMT_BTREE:
1071	error = xfs_bmap_add_attrfork_btree(tp, ip, &logflags);
1072	break;
1073	default:
1074	error = 0;
1075	break;
1076	}
1077	if (logflags)
1078	xfs_trans_log_inode(tp, ip, logflags);
1079	if (error)
1080	goto trans_cancel;
1081	if (!xfs_has_attr(mp) ||
1082	(!xfs_has_attr2(mp) && version == 2)) {
1083	bool log_sb = false;
1084
1085	spin_lock(&mp->m_sb_lock);
1086	if (!xfs_has_attr(mp)) {
1087	xfs_add_attr(mp);
1088	log_sb = true;
1089	}
1090	if (!xfs_has_attr2(mp) && version == 2) {
1091	xfs_add_attr2(mp);
1092	log_sb = true;
1093	}
1094	spin_unlock(&mp->m_sb_lock);
1095	if (log_sb)
1096	xfs_log_sb(tp);
1097	}
1098
1099	error = xfs_trans_commit(tp);
1100	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1101	return error;
1102
1103	trans_cancel:
1104	xfs_trans_cancel(tp);
1105	xfs_iunlock(ip, XFS_ILOCK_EXCL);
1106	return error;
1107	}
1108
1109	/*
1110	* Internal and external extent tree search functions.
1111	*/
1112
1113	struct xfs_iread_state {
1114	struct xfs_iext_cursor icur;
1115	xfs_extnum_t loaded;
1116	};
1117
1118	int
1119	xfs_bmap_complain_bad_rec(
1120	struct xfs_inode *ip,
1121	int whichfork,
1122	xfs_failaddr_t fa,
1123	const struct xfs_bmbt_irec *irec)
1124	{
1125	struct xfs_mount *mp = ip->i_mount;
1126	const char *forkname;
1127
1128	switch (whichfork) {
1129	case XFS_DATA_FORK: forkname = "data"; break;
1130	case XFS_ATTR_FORK: forkname = "attr"; break;
1131	case XFS_COW_FORK: forkname = "CoW"; break;
1132	default: forkname = "???"; break;
1133	}
1134
1135	xfs_warn(mp,
1136	"Bmap BTree record corruption in inode 0x%llx %s fork detected at %pS!",
1137	ip->i_ino, forkname, fa);
1138	xfs_warn(mp,
1139	"Offset 0x%llx, start block 0x%llx, block count 0x%llx state 0x%x",
1140	irec->br_startoff, irec->br_startblock, irec->br_blockcount,
1141	irec->br_state);
1142
1143	return -EFSCORRUPTED;
1144	}
1145
1146	/* Stuff every bmbt record from this block into the incore extent map. */
1147	static int
1148	xfs_iread_bmbt_block(
1149	struct xfs_btree_cur *cur,
1150	int level,
1151	void *priv)
1152	{
1153	struct xfs_iread_state *ir = priv;
1154	struct xfs_mount *mp = cur->bc_mp;
1155	struct xfs_inode *ip = cur->bc_ino.ip;
1156	struct xfs_btree_block *block;
1157	struct xfs_buf *bp;
1158	struct xfs_bmbt_rec *frp;
1159	xfs_extnum_t num_recs;
1160	xfs_extnum_t j;
1161	int whichfork = cur->bc_ino.whichfork;
1162	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1163
1164	block = xfs_btree_get_block(cur, level, bpp: &bp);
1165
1166	/* Abort if we find more records than nextents. */
1167	num_recs = xfs_btree_get_numrecs(block);
1168	if (unlikely(ir->loaded + num_recs > ifp->if_nextents)) {
1169	xfs_warn(ip->i_mount, "corrupt dinode %llu, (btree extents).",
1170	(unsigned long long)ip->i_ino);
1171	xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, block,
1172	sizeof(*block), __this_address);
1173	xfs_bmap_mark_sick(ip, whichfork);
1174	return -EFSCORRUPTED;
1175	}
1176
1177	/* Copy records into the incore cache. */
1178	frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1179	for (j = 0; j < num_recs; j++, frp++, ir->loaded++) {
1180	struct xfs_bmbt_irec new;
1181	xfs_failaddr_t fa;
1182
1183	xfs_bmbt_disk_get_all(frp, &new);
1184	fa = xfs_bmap_validate_extent(ip, whichfork, &new);
1185	if (fa) {
1186	xfs_inode_verifier_error(ip, -EFSCORRUPTED,
1187	"xfs_iread_extents(2)", frp,
1188	sizeof(*frp), fa);
1189	xfs_bmap_mark_sick(ip, whichfork);
1190	return xfs_bmap_complain_bad_rec(ip, whichfork, fa,
1191	&new);
1192	}
1193	xfs_iext_insert(ip, &ir->icur, &new,
1194	xfs_bmap_fork_to_state(whichfork));
1195	trace_xfs_read_extent(ip, &ir->icur,
1196	xfs_bmap_fork_to_state(whichfork), _THIS_IP_);
1197	xfs_iext_next(ifp, &ir->icur);
1198	}
1199
1200	return 0;
1201	}
1202
1203	/*
1204	* Read in extents from a btree-format inode.
1205	*/
1206	int
1207	xfs_iread_extents(
1208	struct xfs_trans *tp,
1209	struct xfs_inode *ip,
1210	int whichfork)
1211	{
1212	struct xfs_iread_state ir;
1213	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1214	struct xfs_mount *mp = ip->i_mount;
1215	struct xfs_btree_cur *cur;
1216	int error;
1217
1218	if (!xfs_need_iread_extents(ifp))
1219	return 0;
1220
1221	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1222
1223	ir.loaded = 0;
1224	xfs_iext_first(ifp, &ir.icur);
1225	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
1226	error = xfs_btree_visit_blocks(cur, fn: xfs_iread_bmbt_block,
1227	XFS_BTREE_VISIT_RECORDS, data: &ir);
1228	xfs_btree_del_cursor(cur, error);
1229	if (error)
1230	goto out;
1231
1232	if (XFS_IS_CORRUPT(mp, ir.loaded != ifp->if_nextents)) {
1233	xfs_bmap_mark_sick(ip, whichfork);
1234	error = -EFSCORRUPTED;
1235	goto out;
1236	}
1237	ASSERT(ir.loaded == xfs_iext_count(ifp));
1238	/*
1239	* Use release semantics so that we can use acquire semantics in
1240	* xfs_need_iread_extents and be guaranteed to see a valid mapping tree
1241	* after that load.
1242	*/
1243	smp_store_release(&ifp->if_needextents, 0);
1244	return 0;
1245	out:
1246	if (xfs_metadata_is_sick(error))
1247	xfs_bmap_mark_sick(ip, whichfork);
1248	xfs_iext_destroy(ifp);
1249	return error;
1250	}
1251
1252	/*
1253	* Returns the relative block number of the first unused block(s) in the given
1254	* fork with at least "len" logically contiguous blocks free. This is the
1255	* lowest-address hole if the fork has holes, else the first block past the end
1256	* of fork. Return 0 if the fork is currently local (in-inode).
1257	*/
1258	int /* error */
1259	xfs_bmap_first_unused(
1260	struct xfs_trans *tp, /* transaction pointer */
1261	struct xfs_inode *ip, /* incore inode */
1262	xfs_extlen_t len, /* size of hole to find */
1263	xfs_fileoff_t *first_unused, /* unused block */
1264	int whichfork) /* data or attr fork */
1265	{
1266	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1267	struct xfs_bmbt_irec got;
1268	struct xfs_iext_cursor icur;
1269	xfs_fileoff_t lastaddr = 0;
1270	xfs_fileoff_t lowest, max;
1271	int error;
1272
1273	if (ifp->if_format == XFS_DINODE_FMT_LOCAL) {
1274	*first_unused = 0;
1275	return 0;
1276	}
1277
1278	ASSERT(xfs_ifork_has_extents(ifp));
1279
1280	error = xfs_iread_extents(tp, ip, whichfork);
1281	if (error)
1282	return error;
1283
1284	lowest = max = *first_unused;
1285	for_each_xfs_iext(ifp, &icur, &got) {
1286	/*
1287	* See if the hole before this extent will work.
1288	*/
1289	if (got.br_startoff >= lowest + len &&
1290	got.br_startoff - max >= len)
1291	break;
1292	lastaddr = got.br_startoff + got.br_blockcount;
1293	max = XFS_FILEOFF_MAX(lastaddr, lowest);
1294	}
1295
1296	*first_unused = max;
1297	return 0;
1298	}
1299
1300	/*
1301	* Returns the file-relative block number of the last block - 1 before
1302	* last_block (input value) in the file.
1303	* This is not based on i_size, it is based on the extent records.
1304	* Returns 0 for local files, as they do not have extent records.
1305	*/
1306	int /* error */
1307	xfs_bmap_last_before(
1308	struct xfs_trans *tp, /* transaction pointer */
1309	struct xfs_inode *ip, /* incore inode */
1310	xfs_fileoff_t *last_block, /* last block */
1311	int whichfork) /* data or attr fork */
1312	{
1313	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1314	struct xfs_bmbt_irec got;
1315	struct xfs_iext_cursor icur;
1316	int error;
1317
1318	switch (ifp->if_format) {
1319	case XFS_DINODE_FMT_LOCAL:
1320	*last_block = 0;
1321	return 0;
1322	case XFS_DINODE_FMT_BTREE:
1323	case XFS_DINODE_FMT_EXTENTS:
1324	break;
1325	default:
1326	ASSERT(0);
1327	xfs_bmap_mark_sick(ip, whichfork);
1328	return -EFSCORRUPTED;
1329	}
1330
1331	error = xfs_iread_extents(tp, ip, whichfork);
1332	if (error)
1333	return error;
1334
1335	if (!xfs_iext_lookup_extent_before(ip, ifp, last_block, &icur, &got))
1336	*last_block = 0;
1337	return 0;
1338	}
1339
1340	int
1341	xfs_bmap_last_extent(
1342	struct xfs_trans *tp,
1343	struct xfs_inode *ip,
1344	int whichfork,
1345	struct xfs_bmbt_irec *rec,
1346	int *is_empty)
1347	{
1348	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1349	struct xfs_iext_cursor icur;
1350	int error;
1351
1352	error = xfs_iread_extents(tp, ip, whichfork);
1353	if (error)
1354	return error;
1355
1356	xfs_iext_last(ifp, &icur);
1357	if (!xfs_iext_get_extent(ifp, &icur, rec))
1358	*is_empty = 1;
1359	else
1360	*is_empty = 0;
1361	return 0;
1362	}
1363
1364	/*
1365	* Check the last inode extent to determine whether this allocation will result
1366	* in blocks being allocated at the end of the file. When we allocate new data
1367	* blocks at the end of the file which do not start at the previous data block,
1368	* we will try to align the new blocks at stripe unit boundaries.
1369	*
1370	* Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1371	* at, or past the EOF.
1372	*/
1373	STATIC int
1374	xfs_bmap_isaeof(
1375	struct xfs_bmalloca *bma,
1376	int whichfork)
1377	{
1378	struct xfs_bmbt_irec rec;
1379	int is_empty;
1380	int error;
1381
1382	bma->aeof = false;
1383	error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1384	&is_empty);
1385	if (error)
1386	return error;
1387
1388	if (is_empty) {
1389	bma->aeof = true;
1390	return 0;
1391	}
1392
1393	/*
1394	* Check if we are allocation or past the last extent, or at least into
1395	* the last delayed allocated extent.
1396	*/
1397	bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1398	(bma->offset >= rec.br_startoff &&
1399	isnullstartblock(rec.br_startblock));
1400	return 0;
1401	}
1402
1403	/*
1404	* Returns the file-relative block number of the first block past eof in
1405	* the file. This is not based on i_size, it is based on the extent records.
1406	* Returns 0 for local files, as they do not have extent records.
1407	*/
1408	int
1409	xfs_bmap_last_offset(
1410	struct xfs_inode *ip,
1411	xfs_fileoff_t *last_block,
1412	int whichfork)
1413	{
1414	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
1415	struct xfs_bmbt_irec rec;
1416	int is_empty;
1417	int error;
1418
1419	*last_block = 0;
1420
1421	if (ifp->if_format == XFS_DINODE_FMT_LOCAL)
1422	return 0;
1423
1424	if (XFS_IS_CORRUPT(ip->i_mount, !xfs_ifork_has_extents(ifp))) {
1425	xfs_bmap_mark_sick(ip, whichfork);
1426	return -EFSCORRUPTED;
1427	}
1428
1429	error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1430	if (error || is_empty)
1431	return error;
1432
1433	*last_block = rec.br_startoff + rec.br_blockcount;
1434	return 0;
1435	}
1436
1437	/*
1438	* Extent tree manipulation functions used during allocation.
1439	*/
1440
1441	/*
1442	* Convert a delayed allocation to a real allocation.
1443	*/
1444	STATIC int /* error */
1445	xfs_bmap_add_extent_delay_real(
1446	struct xfs_bmalloca *bma,
1447	int whichfork)
1448	{
1449	struct xfs_mount *mp = bma->ip->i_mount;
1450	struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
1451	struct xfs_bmbt_irec *new = &bma->got;
1452	int error; /* error return value */
1453	int i; /* temp state */
1454	xfs_fileoff_t new_endoff; /* end offset of new entry */
1455	xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1456	/* left is 0, right is 1, prev is 2 */
1457	int rval=0; /* return value (logging flags) */
1458	uint32_t state = xfs_bmap_fork_to_state(whichfork);
1459	xfs_filblks_t da_new; /* new count del alloc blocks used */
1460	xfs_filblks_t da_old; /* old count del alloc blocks used */
1461	xfs_filblks_t temp=0; /* value for da_new calculations */
1462	int tmp_rval; /* partial logging flags */
1463	struct xfs_bmbt_irec old;
1464
1465	ASSERT(whichfork != XFS_ATTR_FORK);
1466	ASSERT(!isnullstartblock(new->br_startblock));
1467	ASSERT(!bma->cur || (bma->cur->bc_flags & XFS_BTREE_BMBT_WASDEL));
1468
1469	XFS_STATS_INC(mp, xs_add_exlist);
1470
1471	#define LEFT r[0]
1472	#define RIGHT r[1]
1473	#define PREV r[2]
1474
1475	/*
1476	* Set up a bunch of variables to make the tests simpler.
1477	*/
1478	xfs_iext_get_extent(ifp, &bma->icur, &PREV);
1479	new_endoff = new->br_startoff + new->br_blockcount;
1480	ASSERT(isnullstartblock(PREV.br_startblock));
1481	ASSERT(PREV.br_startoff <= new->br_startoff);
1482	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1483
1484	da_old = startblockval(PREV.br_startblock);
1485	da_new = 0;
1486
1487	/*
1488	* Set flags determining what part of the previous delayed allocation
1489	* extent is being replaced by a real allocation.
1490	*/
1491	if (PREV.br_startoff == new->br_startoff)
1492	state |= BMAP_LEFT_FILLING;
1493	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1494	state |= BMAP_RIGHT_FILLING;
1495
1496	/*
1497	* Check and set flags if this segment has a left neighbor.
1498	* Don't set contiguous if the combined extent would be too large.
1499	*/
1500	if (xfs_iext_peek_prev_extent(ifp, &bma->icur, &LEFT)) {
1501	state |= BMAP_LEFT_VALID;
1502	if (isnullstartblock(LEFT.br_startblock))
1503	state |= BMAP_LEFT_DELAY;
1504	}
1505
1506	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1507	LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1508	LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1509	LEFT.br_state == new->br_state &&
1510	LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
1511	state |= BMAP_LEFT_CONTIG;
1512
1513	/*
1514	* Check and set flags if this segment has a right neighbor.
1515	* Don't set contiguous if the combined extent would be too large.
1516	* Also check for all-three-contiguous being too large.
1517	*/
1518	if (xfs_iext_peek_next_extent(ifp, &bma->icur, &RIGHT)) {
1519	state |= BMAP_RIGHT_VALID;
1520	if (isnullstartblock(RIGHT.br_startblock))
1521	state |= BMAP_RIGHT_DELAY;
1522	}
1523
1524	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1525	new_endoff == RIGHT.br_startoff &&
1526	new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1527	new->br_state == RIGHT.br_state &&
1528	new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
1529	((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1530	BMAP_RIGHT_FILLING)) !=
1531	(BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1532	BMAP_RIGHT_FILLING) ||
1533	LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1534	<= XFS_MAX_BMBT_EXTLEN))
1535	state |= BMAP_RIGHT_CONTIG;
1536
1537	error = 0;
1538	/*
1539	* Switch out based on the FILLING and CONTIG state bits.
1540	*/
1541	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1542	BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1543	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1544	BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1545	/*
1546	* Filling in all of a previously delayed allocation extent.
1547	* The left and right neighbors are both contiguous with new.
1548	*/
1549	LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
1550
1551	xfs_iext_remove(bma->ip, &bma->icur, state);
1552	xfs_iext_remove(bma->ip, &bma->icur, state);
1553	xfs_iext_prev(ifp, &bma->icur);
1554	xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1555	ifp->if_nextents--;
1556
1557	if (bma->cur == NULL)
1558	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1559	else {
1560	rval = XFS_ILOG_CORE;
1561	error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1562	if (error)
1563	goto done;
1564	if (XFS_IS_CORRUPT(mp, i != 1)) {
1565	xfs_btree_mark_sick(cur: bma->cur);
1566	error = -EFSCORRUPTED;
1567	goto done;
1568	}
1569	error = xfs_btree_delete(bma->cur, &i);
1570	if (error)
1571	goto done;
1572	if (XFS_IS_CORRUPT(mp, i != 1)) {
1573	xfs_btree_mark_sick(cur: bma->cur);
1574	error = -EFSCORRUPTED;
1575	goto done;
1576	}
1577	error = xfs_btree_decrement(bma->cur, 0, &i);
1578	if (error)
1579	goto done;
1580	if (XFS_IS_CORRUPT(mp, i != 1)) {
1581	xfs_btree_mark_sick(cur: bma->cur);
1582	error = -EFSCORRUPTED;
1583	goto done;
1584	}
1585	error = xfs_bmbt_update(bma->cur, &LEFT);
1586	if (error)
1587	goto done;
1588	}
1589	break;
1590
1591	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1592	/*
1593	* Filling in all of a previously delayed allocation extent.
1594	* The left neighbor is contiguous, the right is not.
1595	*/
1596	old = LEFT;
1597	LEFT.br_blockcount += PREV.br_blockcount;
1598
1599	xfs_iext_remove(bma->ip, &bma->icur, state);
1600	xfs_iext_prev(ifp, &bma->icur);
1601	xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1602
1603	if (bma->cur == NULL)
1604	rval = XFS_ILOG_DEXT;
1605	else {
1606	rval = 0;
1607	error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1608	if (error)
1609	goto done;
1610	if (XFS_IS_CORRUPT(mp, i != 1)) {
1611	xfs_btree_mark_sick(cur: bma->cur);
1612	error = -EFSCORRUPTED;
1613	goto done;
1614	}
1615	error = xfs_bmbt_update(bma->cur, &LEFT);
1616	if (error)
1617	goto done;
1618	}
1619	break;
1620
1621	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1622	/*
1623	* Filling in all of a previously delayed allocation extent.
1624	* The right neighbor is contiguous, the left is not. Take care
1625	* with delay -> unwritten extent allocation here because the
1626	* delalloc record we are overwriting is always written.
1627	*/
1628	PREV.br_startblock = new->br_startblock;
1629	PREV.br_blockcount += RIGHT.br_blockcount;
1630	PREV.br_state = new->br_state;
1631
1632	xfs_iext_next(ifp, &bma->icur);
1633	xfs_iext_remove(bma->ip, &bma->icur, state);
1634	xfs_iext_prev(ifp, &bma->icur);
1635	xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1636
1637	if (bma->cur == NULL)
1638	rval = XFS_ILOG_DEXT;
1639	else {
1640	rval = 0;
1641	error = xfs_bmbt_lookup_eq(bma->cur, &RIGHT, &i);
1642	if (error)
1643	goto done;
1644	if (XFS_IS_CORRUPT(mp, i != 1)) {
1645	xfs_btree_mark_sick(cur: bma->cur);
1646	error = -EFSCORRUPTED;
1647	goto done;
1648	}
1649	error = xfs_bmbt_update(bma->cur, &PREV);
1650	if (error)
1651	goto done;
1652	}
1653	break;
1654
1655	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1656	/*
1657	* Filling in all of a previously delayed allocation extent.
1658	* Neither the left nor right neighbors are contiguous with
1659	* the new one.
1660	*/
1661	PREV.br_startblock = new->br_startblock;
1662	PREV.br_state = new->br_state;
1663	xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1664	ifp->if_nextents++;
1665
1666	if (bma->cur == NULL)
1667	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1668	else {
1669	rval = XFS_ILOG_CORE;
1670	error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1671	if (error)
1672	goto done;
1673	if (XFS_IS_CORRUPT(mp, i != 0)) {
1674	xfs_btree_mark_sick(cur: bma->cur);
1675	error = -EFSCORRUPTED;
1676	goto done;
1677	}
1678	error = xfs_btree_insert(bma->cur, &i);
1679	if (error)
1680	goto done;
1681	if (XFS_IS_CORRUPT(mp, i != 1)) {
1682	xfs_btree_mark_sick(cur: bma->cur);
1683	error = -EFSCORRUPTED;
1684	goto done;
1685	}
1686	}
1687	break;
1688
1689	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1690	/*
1691	* Filling in the first part of a previous delayed allocation.
1692	* The left neighbor is contiguous.
1693	*/
1694	old = LEFT;
1695	temp = PREV.br_blockcount - new->br_blockcount;
1696	da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1697	startblockval(PREV.br_startblock));
1698
1699	LEFT.br_blockcount += new->br_blockcount;
1700
1701	PREV.br_blockcount = temp;
1702	PREV.br_startoff += new->br_blockcount;
1703	PREV.br_startblock = nullstartblock(da_new);
1704
1705	xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1706	xfs_iext_prev(ifp, &bma->icur);
1707	xfs_iext_update_extent(bma->ip, state, &bma->icur, &LEFT);
1708
1709	if (bma->cur == NULL)
1710	rval = XFS_ILOG_DEXT;
1711	else {
1712	rval = 0;
1713	error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1714	if (error)
1715	goto done;
1716	if (XFS_IS_CORRUPT(mp, i != 1)) {
1717	xfs_btree_mark_sick(cur: bma->cur);
1718	error = -EFSCORRUPTED;
1719	goto done;
1720	}
1721	error = xfs_bmbt_update(bma->cur, &LEFT);
1722	if (error)
1723	goto done;
1724	}
1725	break;
1726
1727	case BMAP_LEFT_FILLING:
1728	/*
1729	* Filling in the first part of a previous delayed allocation.
1730	* The left neighbor is not contiguous.
1731	*/
1732	xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1733	ifp->if_nextents++;
1734
1735	if (bma->cur == NULL)
1736	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1737	else {
1738	rval = XFS_ILOG_CORE;
1739	error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1740	if (error)
1741	goto done;
1742	if (XFS_IS_CORRUPT(mp, i != 0)) {
1743	xfs_btree_mark_sick(cur: bma->cur);
1744	error = -EFSCORRUPTED;
1745	goto done;
1746	}
1747	error = xfs_btree_insert(bma->cur, &i);
1748	if (error)
1749	goto done;
1750	if (XFS_IS_CORRUPT(mp, i != 1)) {
1751	xfs_btree_mark_sick(cur: bma->cur);
1752	error = -EFSCORRUPTED;
1753	goto done;
1754	}
1755	}
1756
1757	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1758	error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1759	&bma->cur, 1, &tmp_rval, whichfork);
1760	rval |= tmp_rval;
1761	if (error)
1762	goto done;
1763	}
1764
1765	temp = PREV.br_blockcount - new->br_blockcount;
1766	da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1767	startblockval(PREV.br_startblock) -
1768	(bma->cur ? bma->cur->bc_bmap.allocated : 0));
1769
1770	PREV.br_startoff = new_endoff;
1771	PREV.br_blockcount = temp;
1772	PREV.br_startblock = nullstartblock(da_new);
1773	xfs_iext_next(ifp, &bma->icur);
1774	xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1775	xfs_iext_prev(ifp, &bma->icur);
1776	break;
1777
1778	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1779	/*
1780	* Filling in the last part of a previous delayed allocation.
1781	* The right neighbor is contiguous with the new allocation.
1782	*/
1783	old = RIGHT;
1784	RIGHT.br_startoff = new->br_startoff;
1785	RIGHT.br_startblock = new->br_startblock;
1786	RIGHT.br_blockcount += new->br_blockcount;
1787
1788	if (bma->cur == NULL)
1789	rval = XFS_ILOG_DEXT;
1790	else {
1791	rval = 0;
1792	error = xfs_bmbt_lookup_eq(bma->cur, &old, &i);
1793	if (error)
1794	goto done;
1795	if (XFS_IS_CORRUPT(mp, i != 1)) {
1796	xfs_btree_mark_sick(cur: bma->cur);
1797	error = -EFSCORRUPTED;
1798	goto done;
1799	}
1800	error = xfs_bmbt_update(bma->cur, &RIGHT);
1801	if (error)
1802	goto done;
1803	}
1804
1805	temp = PREV.br_blockcount - new->br_blockcount;
1806	da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1807	startblockval(PREV.br_startblock));
1808
1809	PREV.br_blockcount = temp;
1810	PREV.br_startblock = nullstartblock(da_new);
1811
1812	xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1813	xfs_iext_next(ifp, &bma->icur);
1814	xfs_iext_update_extent(bma->ip, state, &bma->icur, &RIGHT);
1815	break;
1816
1817	case BMAP_RIGHT_FILLING:
1818	/*
1819	* Filling in the last part of a previous delayed allocation.
1820	* The right neighbor is not contiguous.
1821	*/
1822	xfs_iext_update_extent(bma->ip, state, &bma->icur, new);
1823	ifp->if_nextents++;
1824
1825	if (bma->cur == NULL)
1826	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1827	else {
1828	rval = XFS_ILOG_CORE;
1829	error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1830	if (error)
1831	goto done;
1832	if (XFS_IS_CORRUPT(mp, i != 0)) {
1833	xfs_btree_mark_sick(cur: bma->cur);
1834	error = -EFSCORRUPTED;
1835	goto done;
1836	}
1837	error = xfs_btree_insert(bma->cur, &i);
1838	if (error)
1839	goto done;
1840	if (XFS_IS_CORRUPT(mp, i != 1)) {
1841	xfs_btree_mark_sick(cur: bma->cur);
1842	error = -EFSCORRUPTED;
1843	goto done;
1844	}
1845	}
1846
1847	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1848	error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1849	&bma->cur, 1, &tmp_rval, whichfork);
1850	rval |= tmp_rval;
1851	if (error)
1852	goto done;
1853	}
1854
1855	temp = PREV.br_blockcount - new->br_blockcount;
1856	da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
1857	startblockval(PREV.br_startblock) -
1858	(bma->cur ? bma->cur->bc_bmap.allocated : 0));
1859
1860	PREV.br_startblock = nullstartblock(da_new);
1861	PREV.br_blockcount = temp;
1862	xfs_iext_insert(bma->ip, &bma->icur, &PREV, state);
1863	xfs_iext_next(ifp, &bma->icur);
1864	break;
1865
1866	case 0:
1867	/*
1868	* Filling in the middle part of a previous delayed allocation.
1869	* Contiguity is impossible here.
1870	* This case is avoided almost all the time.
1871	*
1872	* We start with a delayed allocation:
1873	*
1874	* +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1875	* PREV @ idx
1876	*
1877	* and we are allocating:
1878	* +rrrrrrrrrrrrrrrrr+
1879	* new
1880	*
1881	* and we set it up for insertion as:
1882	* +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1883	* new
1884	* PREV @ idx LEFT RIGHT
1885	* inserted at idx + 1
1886	*/
1887	old = PREV;
1888
1889	/* LEFT is the new middle */
1890	LEFT = *new;
1891
1892	/* RIGHT is the new right */
1893	RIGHT.br_state = PREV.br_state;
1894	RIGHT.br_startoff = new_endoff;
1895	RIGHT.br_blockcount =
1896	PREV.br_startoff + PREV.br_blockcount - new_endoff;
1897	RIGHT.br_startblock =
1898	nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1899	RIGHT.br_blockcount));
1900
1901	/* truncate PREV */
1902	PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
1903	PREV.br_startblock =
1904	nullstartblock(xfs_bmap_worst_indlen(bma->ip,
1905	PREV.br_blockcount));
1906	xfs_iext_update_extent(bma->ip, state, &bma->icur, &PREV);
1907
1908	xfs_iext_next(ifp, &bma->icur);
1909	xfs_iext_insert(bma->ip, &bma->icur, &RIGHT, state);
1910	xfs_iext_insert(bma->ip, &bma->icur, &LEFT, state);
1911	ifp->if_nextents++;
1912
1913	if (bma->cur == NULL)
1914	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1915	else {
1916	rval = XFS_ILOG_CORE;
1917	error = xfs_bmbt_lookup_eq(bma->cur, new, &i);
1918	if (error)
1919	goto done;
1920	if (XFS_IS_CORRUPT(mp, i != 0)) {
1921	xfs_btree_mark_sick(cur: bma->cur);
1922	error = -EFSCORRUPTED;
1923	goto done;
1924	}
1925	error = xfs_btree_insert(bma->cur, &i);
1926	if (error)
1927	goto done;
1928	if (XFS_IS_CORRUPT(mp, i != 1)) {
1929	xfs_btree_mark_sick(cur: bma->cur);
1930	error = -EFSCORRUPTED;
1931	goto done;
1932	}
1933	}
1934
1935	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1936	error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1937	&bma->cur, 1, &tmp_rval, whichfork);
1938	rval |= tmp_rval;
1939	if (error)
1940	goto done;
1941	}
1942
1943	da_new = startblockval(PREV.br_startblock) +
1944	startblockval(RIGHT.br_startblock);
1945	break;
1946
1947	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1948	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1949	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1950	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1951	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1952	case BMAP_LEFT_CONTIG:
1953	case BMAP_RIGHT_CONTIG:
1954	/*
1955	* These cases are all impossible.
1956	*/
1957	ASSERT(0);
1958	}
1959
1960	/* add reverse mapping unless caller opted out */
1961	if (!(bma->flags & XFS_BMAPI_NORMAP))
1962	xfs_rmap_map_extent(tp: bma->tp, ip: bma->ip, whichfork, imap: new);
1963
1964	/* convert to a btree if necessary */
1965	if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1966	int tmp_logflags; /* partial log flag return val */
1967
1968	ASSERT(bma->cur == NULL);
1969	error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1970	&bma->cur, da_old > 0, &tmp_logflags,
1971	whichfork);
1972	bma->logflags |= tmp_logflags;
1973	if (error)
1974	goto done;
1975	}
1976
1977	if (da_new != da_old)
1978	xfs_mod_delalloc(mp, (int64_t)da_new - da_old);
1979
1980	if (bma->cur) {
1981	da_new += bma->cur->bc_bmap.allocated;
1982	bma->cur->bc_bmap.allocated = 0;
1983	}
1984
1985	/* adjust for changes in reserved delayed indirect blocks */
1986	if (da_new != da_old) {
1987	ASSERT(state == 0 || da_new < da_old);
1988	error = xfs_mod_fdblocks(mp, (int64_t)(da_old - da_new),
1989	false);
1990	}
1991
1992	xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1993	done:
1994	if (whichfork != XFS_COW_FORK)
1995	bma->logflags |= rval;
1996	return error;
1997	#undef LEFT
1998	#undef RIGHT
1999	#undef PREV
2000	}
2001
2002	/*
2003	* Convert an unwritten allocation to a real allocation or vice versa.
2004	*/
2005	int /* error */
2006	xfs_bmap_add_extent_unwritten_real(
2007	struct xfs_trans *tp,
2008	xfs_inode_t *ip, /* incore inode pointer */
2009	int whichfork,
2010	struct xfs_iext_cursor *icur,
2011	struct xfs_btree_cur **curp, /* if *curp is null, not a btree */
2012	xfs_bmbt_irec_t *new, /* new data to add to file extents */
2013	int *logflagsp) /* inode logging flags */
2014	{
2015	struct xfs_btree_cur *cur; /* btree cursor */
2016	int error; /* error return value */
2017	int i; /* temp state */
2018	struct xfs_ifork *ifp; /* inode fork pointer */
2019	xfs_fileoff_t new_endoff; /* end offset of new entry */
2020	xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2021	/* left is 0, right is 1, prev is 2 */
2022	int rval=0; /* return value (logging flags) */
2023	uint32_t state = xfs_bmap_fork_to_state(whichfork);
2024	struct xfs_mount *mp = ip->i_mount;
2025	struct xfs_bmbt_irec old;
2026
2027	*logflagsp = 0;
2028
2029	cur = *curp;
2030	ifp = xfs_ifork_ptr(ip, whichfork);
2031
2032	ASSERT(!isnullstartblock(new->br_startblock));
2033
2034	XFS_STATS_INC(mp, xs_add_exlist);
2035
2036	#define LEFT r[0]
2037	#define RIGHT r[1]
2038	#define PREV r[2]
2039
2040	/*
2041	* Set up a bunch of variables to make the tests simpler.
2042	*/
2043	error = 0;
2044	xfs_iext_get_extent(ifp, icur, &PREV);
2045	ASSERT(new->br_state != PREV.br_state);
2046	new_endoff = new->br_startoff + new->br_blockcount;
2047	ASSERT(PREV.br_startoff <= new->br_startoff);
2048	ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2049
2050	/*
2051	* Set flags determining what part of the previous oldext allocation
2052	* extent is being replaced by a newext allocation.
2053	*/
2054	if (PREV.br_startoff == new->br_startoff)
2055	state |= BMAP_LEFT_FILLING;
2056	if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2057	state |= BMAP_RIGHT_FILLING;
2058
2059	/*
2060	* Check and set flags if this segment has a left neighbor.
2061	* Don't set contiguous if the combined extent would be too large.
2062	*/
2063	if (xfs_iext_peek_prev_extent(ifp, icur, &LEFT)) {
2064	state |= BMAP_LEFT_VALID;
2065	if (isnullstartblock(LEFT.br_startblock))
2066	state |= BMAP_LEFT_DELAY;
2067	}
2068
2069	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2070	LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2071	LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2072	LEFT.br_state == new->br_state &&
2073	LEFT.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2074	state |= BMAP_LEFT_CONTIG;
2075
2076	/*
2077	* Check and set flags if this segment has a right neighbor.
2078	* Don't set contiguous if the combined extent would be too large.
2079	* Also check for all-three-contiguous being too large.
2080	*/
2081	if (xfs_iext_peek_next_extent(ifp, icur, &RIGHT)) {
2082	state |= BMAP_RIGHT_VALID;
2083	if (isnullstartblock(RIGHT.br_startblock))
2084	state |= BMAP_RIGHT_DELAY;
2085	}
2086
2087	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2088	new_endoff == RIGHT.br_startoff &&
2089	new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2090	new->br_state == RIGHT.br_state &&
2091	new->br_blockcount + RIGHT.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2092	((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2093	BMAP_RIGHT_FILLING)) !=
2094	(BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2095	BMAP_RIGHT_FILLING) ||
2096	LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2097	<= XFS_MAX_BMBT_EXTLEN))
2098	state |= BMAP_RIGHT_CONTIG;
2099
2100	/*
2101	* Switch out based on the FILLING and CONTIG state bits.
2102	*/
2103	switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2104	BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2105	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2106	BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2107	/*
2108	* Setting all of a previous oldext extent to newext.
2109	* The left and right neighbors are both contiguous with new.
2110	*/
2111	LEFT.br_blockcount += PREV.br_blockcount + RIGHT.br_blockcount;
2112
2113	xfs_iext_remove(ip, icur, state);
2114	xfs_iext_remove(ip, icur, state);
2115	xfs_iext_prev(ifp, icur);
2116	xfs_iext_update_extent(ip, state, icur, &LEFT);
2117	ifp->if_nextents -= 2;
2118	if (cur == NULL)
2119	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2120	else {
2121	rval = XFS_ILOG_CORE;
2122	error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2123	if (error)
2124	goto done;
2125	if (XFS_IS_CORRUPT(mp, i != 1)) {
2126	xfs_btree_mark_sick(cur);
2127	error = -EFSCORRUPTED;
2128	goto done;
2129	}
2130	if ((error = xfs_btree_delete(cur, &i)))
2131	goto done;
2132	if (XFS_IS_CORRUPT(mp, i != 1)) {
2133	xfs_btree_mark_sick(cur);
2134	error = -EFSCORRUPTED;
2135	goto done;
2136	}
2137	if ((error = xfs_btree_decrement(cur, 0, &i)))
2138	goto done;
2139	if (XFS_IS_CORRUPT(mp, i != 1)) {
2140	xfs_btree_mark_sick(cur);
2141	error = -EFSCORRUPTED;
2142	goto done;
2143	}
2144	if ((error = xfs_btree_delete(cur, &i)))
2145	goto done;
2146	if (XFS_IS_CORRUPT(mp, i != 1)) {
2147	xfs_btree_mark_sick(cur);
2148	error = -EFSCORRUPTED;
2149	goto done;
2150	}
2151	if ((error = xfs_btree_decrement(cur, 0, &i)))
2152	goto done;
2153	if (XFS_IS_CORRUPT(mp, i != 1)) {
2154	xfs_btree_mark_sick(cur);
2155	error = -EFSCORRUPTED;
2156	goto done;
2157	}
2158	error = xfs_bmbt_update(cur, &LEFT);
2159	if (error)
2160	goto done;
2161	}
2162	break;
2163
2164	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2165	/*
2166	* Setting all of a previous oldext extent to newext.
2167	* The left neighbor is contiguous, the right is not.
2168	*/
2169	LEFT.br_blockcount += PREV.br_blockcount;
2170
2171	xfs_iext_remove(ip, icur, state);
2172	xfs_iext_prev(ifp, icur);
2173	xfs_iext_update_extent(ip, state, icur, &LEFT);
2174	ifp->if_nextents--;
2175	if (cur == NULL)
2176	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2177	else {
2178	rval = XFS_ILOG_CORE;
2179	error = xfs_bmbt_lookup_eq(cur, &PREV, &i);
2180	if (error)
2181	goto done;
2182	if (XFS_IS_CORRUPT(mp, i != 1)) {
2183	xfs_btree_mark_sick(cur);
2184	error = -EFSCORRUPTED;
2185	goto done;
2186	}
2187	if ((error = xfs_btree_delete(cur, &i)))
2188	goto done;
2189	if (XFS_IS_CORRUPT(mp, i != 1)) {
2190	xfs_btree_mark_sick(cur);
2191	error = -EFSCORRUPTED;
2192	goto done;
2193	}
2194	if ((error = xfs_btree_decrement(cur, 0, &i)))
2195	goto done;
2196	if (XFS_IS_CORRUPT(mp, i != 1)) {
2197	xfs_btree_mark_sick(cur);
2198	error = -EFSCORRUPTED;
2199	goto done;
2200	}
2201	error = xfs_bmbt_update(cur, &LEFT);
2202	if (error)
2203	goto done;
2204	}
2205	break;
2206
2207	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2208	/*
2209	* Setting all of a previous oldext extent to newext.
2210	* The right neighbor is contiguous, the left is not.
2211	*/
2212	PREV.br_blockcount += RIGHT.br_blockcount;
2213	PREV.br_state = new->br_state;
2214
2215	xfs_iext_next(ifp, icur);
2216	xfs_iext_remove(ip, icur, state);
2217	xfs_iext_prev(ifp, icur);
2218	xfs_iext_update_extent(ip, state, icur, &PREV);
2219	ifp->if_nextents--;
2220
2221	if (cur == NULL)
2222	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2223	else {
2224	rval = XFS_ILOG_CORE;
2225	error = xfs_bmbt_lookup_eq(cur, &RIGHT, &i);
2226	if (error)
2227	goto done;
2228	if (XFS_IS_CORRUPT(mp, i != 1)) {
2229	xfs_btree_mark_sick(cur);
2230	error = -EFSCORRUPTED;
2231	goto done;
2232	}
2233	if ((error = xfs_btree_delete(cur, &i)))
2234	goto done;
2235	if (XFS_IS_CORRUPT(mp, i != 1)) {
2236	xfs_btree_mark_sick(cur);
2237	error = -EFSCORRUPTED;
2238	goto done;
2239	}
2240	if ((error = xfs_btree_decrement(cur, 0, &i)))
2241	goto done;
2242	if (XFS_IS_CORRUPT(mp, i != 1)) {
2243	xfs_btree_mark_sick(cur);
2244	error = -EFSCORRUPTED;
2245	goto done;
2246	}
2247	error = xfs_bmbt_update(cur, &PREV);
2248	if (error)
2249	goto done;
2250	}
2251	break;
2252
2253	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2254	/*
2255	* Setting all of a previous oldext extent to newext.
2256	* Neither the left nor right neighbors are contiguous with
2257	* the new one.
2258	*/
2259	PREV.br_state = new->br_state;
2260	xfs_iext_update_extent(ip, state, icur, &PREV);
2261
2262	if (cur == NULL)
2263	rval = XFS_ILOG_DEXT;
2264	else {
2265	rval = 0;
2266	error = xfs_bmbt_lookup_eq(cur, new, &i);
2267	if (error)
2268	goto done;
2269	if (XFS_IS_CORRUPT(mp, i != 1)) {
2270	xfs_btree_mark_sick(cur);
2271	error = -EFSCORRUPTED;
2272	goto done;
2273	}
2274	error = xfs_bmbt_update(cur, &PREV);
2275	if (error)
2276	goto done;
2277	}
2278	break;
2279
2280	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2281	/*
2282	* Setting the first part of a previous oldext extent to newext.
2283	* The left neighbor is contiguous.
2284	*/
2285	LEFT.br_blockcount += new->br_blockcount;
2286
2287	old = PREV;
2288	PREV.br_startoff += new->br_blockcount;
2289	PREV.br_startblock += new->br_blockcount;
2290	PREV.br_blockcount -= new->br_blockcount;
2291
2292	xfs_iext_update_extent(ip, state, icur, &PREV);
2293	xfs_iext_prev(ifp, icur);
2294	xfs_iext_update_extent(ip, state, icur, &LEFT);
2295
2296	if (cur == NULL)
2297	rval = XFS_ILOG_DEXT;
2298	else {
2299	rval = 0;
2300	error = xfs_bmbt_lookup_eq(cur, &old, &i);
2301	if (error)
2302	goto done;
2303	if (XFS_IS_CORRUPT(mp, i != 1)) {
2304	xfs_btree_mark_sick(cur);
2305	error = -EFSCORRUPTED;
2306	goto done;
2307	}
2308	error = xfs_bmbt_update(cur, &PREV);
2309	if (error)
2310	goto done;
2311	error = xfs_btree_decrement(cur, 0, &i);
2312	if (error)
2313	goto done;
2314	error = xfs_bmbt_update(cur, &LEFT);
2315	if (error)
2316	goto done;
2317	}
2318	break;
2319
2320	case BMAP_LEFT_FILLING:
2321	/*
2322	* Setting the first part of a previous oldext extent to newext.
2323	* The left neighbor is not contiguous.
2324	*/
2325	old = PREV;
2326	PREV.br_startoff += new->br_blockcount;
2327	PREV.br_startblock += new->br_blockcount;
2328	PREV.br_blockcount -= new->br_blockcount;
2329
2330	xfs_iext_update_extent(ip, state, icur, &PREV);
2331	xfs_iext_insert(ip, icur, new, state);
2332	ifp->if_nextents++;
2333
2334	if (cur == NULL)
2335	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2336	else {
2337	rval = XFS_ILOG_CORE;
2338	error = xfs_bmbt_lookup_eq(cur, &old, &i);
2339	if (error)
2340	goto done;
2341	if (XFS_IS_CORRUPT(mp, i != 1)) {
2342	xfs_btree_mark_sick(cur);
2343	error = -EFSCORRUPTED;
2344	goto done;
2345	}
2346	error = xfs_bmbt_update(cur, &PREV);
2347	if (error)
2348	goto done;
2349	cur->bc_rec.b = *new;
2350	if ((error = xfs_btree_insert(cur, &i)))
2351	goto done;
2352	if (XFS_IS_CORRUPT(mp, i != 1)) {
2353	xfs_btree_mark_sick(cur);
2354	error = -EFSCORRUPTED;
2355	goto done;
2356	}
2357	}
2358	break;
2359
2360	case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2361	/*
2362	* Setting the last part of a previous oldext extent to newext.
2363	* The right neighbor is contiguous with the new allocation.
2364	*/
2365	old = PREV;
2366	PREV.br_blockcount -= new->br_blockcount;
2367
2368	RIGHT.br_startoff = new->br_startoff;
2369	RIGHT.br_startblock = new->br_startblock;
2370	RIGHT.br_blockcount += new->br_blockcount;
2371
2372	xfs_iext_update_extent(ip, state, icur, &PREV);
2373	xfs_iext_next(ifp, icur);
2374	xfs_iext_update_extent(ip, state, icur, &RIGHT);
2375
2376	if (cur == NULL)
2377	rval = XFS_ILOG_DEXT;
2378	else {
2379	rval = 0;
2380	error = xfs_bmbt_lookup_eq(cur, &old, &i);
2381	if (error)
2382	goto done;
2383	if (XFS_IS_CORRUPT(mp, i != 1)) {
2384	xfs_btree_mark_sick(cur);
2385	error = -EFSCORRUPTED;
2386	goto done;
2387	}
2388	error = xfs_bmbt_update(cur, &PREV);
2389	if (error)
2390	goto done;
2391	error = xfs_btree_increment(cur, 0, &i);
2392	if (error)
2393	goto done;
2394	error = xfs_bmbt_update(cur, &RIGHT);
2395	if (error)
2396	goto done;
2397	}
2398	break;
2399
2400	case BMAP_RIGHT_FILLING:
2401	/*
2402	* Setting the last part of a previous oldext extent to newext.
2403	* The right neighbor is not contiguous.
2404	*/
2405	old = PREV;
2406	PREV.br_blockcount -= new->br_blockcount;
2407
2408	xfs_iext_update_extent(ip, state, icur, &PREV);
2409	xfs_iext_next(ifp, icur);
2410	xfs_iext_insert(ip, icur, new, state);
2411	ifp->if_nextents++;
2412
2413	if (cur == NULL)
2414	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2415	else {
2416	rval = XFS_ILOG_CORE;
2417	error = xfs_bmbt_lookup_eq(cur, &old, &i);
2418	if (error)
2419	goto done;
2420	if (XFS_IS_CORRUPT(mp, i != 1)) {
2421	xfs_btree_mark_sick(cur);
2422	error = -EFSCORRUPTED;
2423	goto done;
2424	}
2425	error = xfs_bmbt_update(cur, &PREV);
2426	if (error)
2427	goto done;
2428	error = xfs_bmbt_lookup_eq(cur, new, &i);
2429	if (error)
2430	goto done;
2431	if (XFS_IS_CORRUPT(mp, i != 0)) {
2432	xfs_btree_mark_sick(cur);
2433	error = -EFSCORRUPTED;
2434	goto done;
2435	}
2436	if ((error = xfs_btree_insert(cur, &i)))
2437	goto done;
2438	if (XFS_IS_CORRUPT(mp, i != 1)) {
2439	xfs_btree_mark_sick(cur);
2440	error = -EFSCORRUPTED;
2441	goto done;
2442	}
2443	}
2444	break;
2445
2446	case 0:
2447	/*
2448	* Setting the middle part of a previous oldext extent to
2449	* newext. Contiguity is impossible here.
2450	* One extent becomes three extents.
2451	*/
2452	old = PREV;
2453	PREV.br_blockcount = new->br_startoff - PREV.br_startoff;
2454
2455	r[0] = *new;
2456	r[1].br_startoff = new_endoff;
2457	r[1].br_blockcount =
2458	old.br_startoff + old.br_blockcount - new_endoff;
2459	r[1].br_startblock = new->br_startblock + new->br_blockcount;
2460	r[1].br_state = PREV.br_state;
2461
2462	xfs_iext_update_extent(ip, state, icur, &PREV);
2463	xfs_iext_next(ifp, icur);
2464	xfs_iext_insert(ip, icur, &r[1], state);
2465	xfs_iext_insert(ip, icur, &r[0], state);
2466	ifp->if_nextents += 2;
2467
2468	if (cur == NULL)
2469	rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2470	else {
2471	rval = XFS_ILOG_CORE;
2472	error = xfs_bmbt_lookup_eq(cur, &old, &i);
2473	if (error)
2474	goto done;
2475	if (XFS_IS_CORRUPT(mp, i != 1)) {
2476	xfs_btree_mark_sick(cur);
2477	error = -EFSCORRUPTED;
2478	goto done;
2479	}
2480	/* new right extent - oldext */
2481	error = xfs_bmbt_update(cur, &r[1]);
2482	if (error)
2483	goto done;
2484	/* new left extent - oldext */
2485	cur->bc_rec.b = PREV;
2486	if ((error = xfs_btree_insert(cur, &i)))
2487	goto done;
2488	if (XFS_IS_CORRUPT(mp, i != 1)) {
2489	xfs_btree_mark_sick(cur);
2490	error = -EFSCORRUPTED;
2491	goto done;
2492	}
2493	/*
2494	* Reset the cursor to the position of the new extent
2495	* we are about to insert as we can't trust it after
2496	* the previous insert.
2497	*/
2498	error = xfs_bmbt_lookup_eq(cur, new, &i);
2499	if (error)
2500	goto done;
2501	if (XFS_IS_CORRUPT(mp, i != 0)) {
2502	xfs_btree_mark_sick(cur);
2503	error = -EFSCORRUPTED;
2504	goto done;
2505	}
2506	/* new middle extent - newext */
2507	if ((error = xfs_btree_insert(cur, &i)))
2508	goto done;
2509	if (XFS_IS_CORRUPT(mp, i != 1)) {
2510	xfs_btree_mark_sick(cur);
2511	error = -EFSCORRUPTED;
2512	goto done;
2513	}
2514	}
2515	break;
2516
2517	case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2518	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2519	case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2520	case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2521	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2522	case BMAP_LEFT_CONTIG:
2523	case BMAP_RIGHT_CONTIG:
2524	/*
2525	* These cases are all impossible.
2526	*/
2527	ASSERT(0);
2528	}
2529
2530	/* update reverse mappings */
2531	xfs_rmap_convert_extent(mp, tp, ip: ip, whichfork, imap: new);
2532
2533	/* convert to a btree if necessary */
2534	if (xfs_bmap_needs_btree(ip, whichfork)) {
2535	int tmp_logflags; /* partial log flag return val */
2536
2537	ASSERT(cur == NULL);
2538	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
2539	&tmp_logflags, whichfork);
2540	*logflagsp |= tmp_logflags;
2541	if (error)
2542	goto done;
2543	}
2544
2545	/* clear out the allocated field, done with it now in any case. */
2546	if (cur) {
2547	cur->bc_bmap.allocated = 0;
2548	*curp = cur;
2549	}
2550
2551	xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
2552	done:
2553	*logflagsp |= rval;
2554	return error;
2555	#undef LEFT
2556	#undef RIGHT
2557	#undef PREV
2558	}
2559
2560	/*
2561	* Convert a hole to a delayed allocation.
2562	*/
2563	STATIC void
2564	xfs_bmap_add_extent_hole_delay(
2565	xfs_inode_t *ip, /* incore inode pointer */
2566	int whichfork,
2567	struct xfs_iext_cursor *icur,
2568	xfs_bmbt_irec_t *new) /* new data to add to file extents */
2569	{
2570	struct xfs_ifork *ifp; /* inode fork pointer */
2571	xfs_bmbt_irec_t left; /* left neighbor extent entry */
2572	xfs_filblks_t newlen=0; /* new indirect size */
2573	xfs_filblks_t oldlen=0; /* old indirect size */
2574	xfs_bmbt_irec_t right; /* right neighbor extent entry */
2575	uint32_t state = xfs_bmap_fork_to_state(whichfork);
2576	xfs_filblks_t temp; /* temp for indirect calculations */
2577
2578	ifp = xfs_ifork_ptr(ip, whichfork);
2579	ASSERT(isnullstartblock(new->br_startblock));
2580
2581	/*
2582	* Check and set flags if this segment has a left neighbor
2583	*/
2584	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2585	state |= BMAP_LEFT_VALID;
2586	if (isnullstartblock(left.br_startblock))
2587	state |= BMAP_LEFT_DELAY;
2588	}
2589
2590	/*
2591	* Check and set flags if the current (right) segment exists.
2592	* If it doesn't exist, we're converting the hole at end-of-file.
2593	*/
2594	if (xfs_iext_get_extent(ifp, icur, &right)) {
2595	state |= BMAP_RIGHT_VALID;
2596	if (isnullstartblock(right.br_startblock))
2597	state |= BMAP_RIGHT_DELAY;
2598	}
2599
2600	/*
2601	* Set contiguity flags on the left and right neighbors.
2602	* Don't let extents get too large, even if the pieces are contiguous.
2603	*/
2604	if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2605	left.br_startoff + left.br_blockcount == new->br_startoff &&
2606	left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2607	state |= BMAP_LEFT_CONTIG;
2608
2609	if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2610	new->br_startoff + new->br_blockcount == right.br_startoff &&
2611	new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2612	(!(state & BMAP_LEFT_CONTIG) ||
2613	(left.br_blockcount + new->br_blockcount +
2614	right.br_blockcount <= XFS_MAX_BMBT_EXTLEN)))
2615	state |= BMAP_RIGHT_CONTIG;
2616
2617	/*
2618	* Switch out based on the contiguity flags.
2619	*/
2620	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2621	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2622	/*
2623	* New allocation is contiguous with delayed allocations
2624	* on the left and on the right.
2625	* Merge all three into a single extent record.
2626	*/
2627	temp = left.br_blockcount + new->br_blockcount +
2628	right.br_blockcount;
2629
2630	oldlen = startblockval(left.br_startblock) +
2631	startblockval(new->br_startblock) +
2632	startblockval(right.br_startblock);
2633	newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2634	oldlen);
2635	left.br_startblock = nullstartblock(newlen);
2636	left.br_blockcount = temp;
2637
2638	xfs_iext_remove(ip, icur, state);
2639	xfs_iext_prev(ifp, icur);
2640	xfs_iext_update_extent(ip, state, icur, &left);
2641	break;
2642
2643	case BMAP_LEFT_CONTIG:
2644	/*
2645	* New allocation is contiguous with a delayed allocation
2646	* on the left.
2647	* Merge the new allocation with the left neighbor.
2648	*/
2649	temp = left.br_blockcount + new->br_blockcount;
2650
2651	oldlen = startblockval(left.br_startblock) +
2652	startblockval(new->br_startblock);
2653	newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2654	oldlen);
2655	left.br_blockcount = temp;
2656	left.br_startblock = nullstartblock(newlen);
2657
2658	xfs_iext_prev(ifp, icur);
2659	xfs_iext_update_extent(ip, state, icur, &left);
2660	break;
2661
2662	case BMAP_RIGHT_CONTIG:
2663	/*
2664	* New allocation is contiguous with a delayed allocation
2665	* on the right.
2666	* Merge the new allocation with the right neighbor.
2667	*/
2668	temp = new->br_blockcount + right.br_blockcount;
2669	oldlen = startblockval(new->br_startblock) +
2670	startblockval(right.br_startblock);
2671	newlen = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2672	oldlen);
2673	right.br_startoff = new->br_startoff;
2674	right.br_startblock = nullstartblock(newlen);
2675	right.br_blockcount = temp;
2676	xfs_iext_update_extent(ip, state, icur, &right);
2677	break;
2678
2679	case 0:
2680	/*
2681	* New allocation is not contiguous with another
2682	* delayed allocation.
2683	* Insert a new entry.
2684	*/
2685	oldlen = newlen = 0;
2686	xfs_iext_insert(ip, icur, new, state);
2687	break;
2688	}
2689	if (oldlen != newlen) {
2690	ASSERT(oldlen > newlen);
2691	xfs_mod_fdblocks(ip->i_mount, (int64_t)(oldlen - newlen),
2692	false);
2693	/*
2694	* Nothing to do for disk quota accounting here.
2695	*/
2696	xfs_mod_delalloc(ip->i_mount, (int64_t)newlen - oldlen);
2697	}
2698	}
2699
2700	/*
2701	* Convert a hole to a real allocation.
2702	*/
2703	STATIC int /* error */
2704	xfs_bmap_add_extent_hole_real(
2705	struct xfs_trans *tp,
2706	struct xfs_inode *ip,
2707	int whichfork,
2708	struct xfs_iext_cursor *icur,
2709	struct xfs_btree_cur **curp,
2710	struct xfs_bmbt_irec *new,
2711	int *logflagsp,
2712	uint32_t flags)
2713	{
2714	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
2715	struct xfs_mount *mp = ip->i_mount;
2716	struct xfs_btree_cur *cur = *curp;
2717	int error; /* error return value */
2718	int i; /* temp state */
2719	xfs_bmbt_irec_t left; /* left neighbor extent entry */
2720	xfs_bmbt_irec_t right; /* right neighbor extent entry */
2721	int rval=0; /* return value (logging flags) */
2722	uint32_t state = xfs_bmap_fork_to_state(whichfork);
2723	struct xfs_bmbt_irec old;
2724
2725	ASSERT(!isnullstartblock(new->br_startblock));
2726	ASSERT(!cur || !(cur->bc_flags & XFS_BTREE_BMBT_WASDEL));
2727
2728	XFS_STATS_INC(mp, xs_add_exlist);
2729
2730	/*
2731	* Check and set flags if this segment has a left neighbor.
2732	*/
2733	if (xfs_iext_peek_prev_extent(ifp, icur, &left)) {
2734	state |= BMAP_LEFT_VALID;
2735	if (isnullstartblock(left.br_startblock))
2736	state |= BMAP_LEFT_DELAY;
2737	}
2738
2739	/*
2740	* Check and set flags if this segment has a current value.
2741	* Not true if we're inserting into the "hole" at eof.
2742	*/
2743	if (xfs_iext_get_extent(ifp, icur, &right)) {
2744	state |= BMAP_RIGHT_VALID;
2745	if (isnullstartblock(right.br_startblock))
2746	state |= BMAP_RIGHT_DELAY;
2747	}
2748
2749	/*
2750	* We're inserting a real allocation between "left" and "right".
2751	* Set the contiguity flags. Don't let extents get too large.
2752	*/
2753	if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2754	left.br_startoff + left.br_blockcount == new->br_startoff &&
2755	left.br_startblock + left.br_blockcount == new->br_startblock &&
2756	left.br_state == new->br_state &&
2757	left.br_blockcount + new->br_blockcount <= XFS_MAX_BMBT_EXTLEN)
2758	state |= BMAP_LEFT_CONTIG;
2759
2760	if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2761	new->br_startoff + new->br_blockcount == right.br_startoff &&
2762	new->br_startblock + new->br_blockcount == right.br_startblock &&
2763	new->br_state == right.br_state &&
2764	new->br_blockcount + right.br_blockcount <= XFS_MAX_BMBT_EXTLEN &&
2765	(!(state & BMAP_LEFT_CONTIG) ||
2766	left.br_blockcount + new->br_blockcount +
2767	right.br_blockcount <= XFS_MAX_BMBT_EXTLEN))
2768	state |= BMAP_RIGHT_CONTIG;
2769
2770	error = 0;
2771	/*
2772	* Select which case we're in here, and implement it.
2773	*/
2774	switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2775	case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2776	/*
2777	* New allocation is contiguous with real allocations on the
2778	* left and on the right.
2779	* Merge all three into a single extent record.
2780	*/
2781	left.br_blockcount += new->br_blockcount + right.br_blockcount;
2782
2783	xfs_iext_remove(ip, icur, state);
2784	xfs_iext_prev(ifp, icur);
2785	xfs_iext_update_extent(ip, state, icur, &left);
2786	ifp->if_nextents--;
2787
2788	if (cur == NULL) {
2789	rval = XFS_ILOG_CORE | xfs_ilog_fext(w: whichfork);
2790	} else {
2791	rval = XFS_ILOG_CORE;
2792	error = xfs_bmbt_lookup_eq(cur, &right, &i);
2793	if (error)
2794	goto done;
2795	if (XFS_IS_CORRUPT(mp, i != 1)) {
2796	xfs_btree_mark_sick(cur);
2797	error = -EFSCORRUPTED;
2798	goto done;
2799	}
2800	error = xfs_btree_delete(cur, &i);
2801	if (error)
2802	goto done;
2803	if (XFS_IS_CORRUPT(mp, i != 1)) {
2804	xfs_btree_mark_sick(cur);
2805	error = -EFSCORRUPTED;
2806	goto done;
2807	}
2808	error = xfs_btree_decrement(cur, 0, &i);
2809	if (error)
2810	goto done;
2811	if (XFS_IS_CORRUPT(mp, i != 1)) {
2812	xfs_btree_mark_sick(cur);
2813	error = -EFSCORRUPTED;
2814	goto done;
2815	}
2816	error = xfs_bmbt_update(cur, &left);
2817	if (error)
2818	goto done;
2819	}
2820	break;
2821
2822	case BMAP_LEFT_CONTIG:
2823	/*
2824	* New allocation is contiguous with a real allocation
2825	* on the left.
2826	* Merge the new allocation with the left neighbor.
2827	*/
2828	old = left;
2829	left.br_blockcount += new->br_blockcount;
2830
2831	xfs_iext_prev(ifp, icur);
2832	xfs_iext_update_extent(ip, state, icur, &left);
2833
2834	if (cur == NULL) {
2835	rval = xfs_ilog_fext(w: whichfork);
2836	} else {
2837	rval = 0;
2838	error = xfs_bmbt_lookup_eq(cur, &old, &i);
2839	if (error)
2840	goto done;
2841	if (XFS_IS_CORRUPT(mp, i != 1)) {
2842	xfs_btree_mark_sick(cur);
2843	error = -EFSCORRUPTED;
2844	goto done;
2845	}
2846	error = xfs_bmbt_update(cur, &left);
2847	if (error)
2848	goto done;
2849	}
2850	break;
2851
2852	case BMAP_RIGHT_CONTIG:
2853	/*
2854	* New allocation is contiguous with a real allocation
2855	* on the right.
2856	* Merge the new allocation with the right neighbor.
2857	*/
2858	old = right;
2859
2860	right.br_startoff = new->br_startoff;
2861	right.br_startblock = new->br_startblock;
2862	right.br_blockcount += new->br_blockcount;
2863	xfs_iext_update_extent(ip, state, icur, &right);
2864
2865	if (cur == NULL) {
2866	rval = xfs_ilog_fext(w: whichfork);
2867	} else {
2868	rval = 0;
2869	error = xfs_bmbt_lookup_eq(cur, &old, &i);
2870	if (error)
2871	goto done;
2872	if (XFS_IS_CORRUPT(mp, i != 1)) {
2873	xfs_btree_mark_sick(cur);
2874	error = -EFSCORRUPTED;
2875	goto done;
2876	}
2877	error = xfs_bmbt_update(cur, &right);
2878	if (error)
2879	goto done;
2880	}
2881	break;
2882
2883	case 0:
2884	/*
2885	* New allocation is not contiguous with another
2886	* real allocation.
2887	* Insert a new entry.
2888	*/
2889	xfs_iext_insert(ip, icur, new, state);
2890	ifp->if_nextents++;
2891
2892	if (cur == NULL) {
2893	rval = XFS_ILOG_CORE | xfs_ilog_fext(w: whichfork);
2894	} else {
2895	rval = XFS_ILOG_CORE;
2896	error = xfs_bmbt_lookup_eq(cur, new, &i);
2897	if (error)
2898	goto done;
2899	if (XFS_IS_CORRUPT(mp, i != 0)) {
2900	xfs_btree_mark_sick(cur);
2901	error = -EFSCORRUPTED;
2902	goto done;
2903	}
2904	error = xfs_btree_insert(cur, &i);
2905	if (error)
2906	goto done;
2907	if (XFS_IS_CORRUPT(mp, i != 1)) {
2908	xfs_btree_mark_sick(cur);
2909	error = -EFSCORRUPTED;
2910	goto done;
2911	}
2912	}
2913	break;
2914	}
2915
2916	/* add reverse mapping unless caller opted out */
2917	if (!(flags & XFS_BMAPI_NORMAP))
2918	xfs_rmap_map_extent(tp, ip, whichfork, imap: new);
2919
2920	/* convert to a btree if necessary */
2921	if (xfs_bmap_needs_btree(ip, whichfork)) {
2922	int tmp_logflags; /* partial log flag return val */
2923
2924	ASSERT(cur == NULL);
2925	error = xfs_bmap_extents_to_btree(tp, ip, curp, 0,
2926	&tmp_logflags, whichfork);
2927	*logflagsp |= tmp_logflags;
2928	cur = *curp;
2929	if (error)
2930	goto done;
2931	}
2932
2933	/* clear out the allocated field, done with it now in any case. */
2934	if (cur)
2935	cur->bc_bmap.allocated = 0;
2936
2937	xfs_bmap_check_leaf_extents(cur, ip, whichfork);
2938	done:
2939	*logflagsp |= rval;
2940	return error;
2941	}
2942
2943	/*
2944	* Functions used in the extent read, allocate and remove paths
2945	*/
2946
2947	/*
2948	* Adjust the size of the new extent based on i_extsize and rt extsize.
2949	*/
2950	int
2951	xfs_bmap_extsize_align(
2952	xfs_mount_t *mp,
2953	xfs_bmbt_irec_t *gotp, /* next extent pointer */
2954	xfs_bmbt_irec_t *prevp, /* previous extent pointer */
2955	xfs_extlen_t extsz, /* align to this extent size */
2956	int rt, /* is this a realtime inode? */
2957	int eof, /* is extent at end-of-file? */
2958	int delay, /* creating delalloc extent? */
2959	int convert, /* overwriting unwritten extent? */
2960	xfs_fileoff_t *offp, /* in/out: aligned offset */
2961	xfs_extlen_t *lenp) /* in/out: aligned length */
2962	{
2963	xfs_fileoff_t orig_off; /* original offset */
2964	xfs_extlen_t orig_alen; /* original length */
2965	xfs_fileoff_t orig_end; /* original off+len */
2966	xfs_fileoff_t nexto; /* next file offset */
2967	xfs_fileoff_t prevo; /* previous file offset */
2968	xfs_fileoff_t align_off; /* temp for offset */
2969	xfs_extlen_t align_alen; /* temp for length */
2970	xfs_extlen_t temp; /* temp for calculations */
2971
2972	if (convert)
2973	return 0;
2974
2975	orig_off = align_off = *offp;
2976	orig_alen = align_alen = *lenp;
2977	orig_end = orig_off + orig_alen;
2978
2979	/*
2980	* If this request overlaps an existing extent, then don't
2981	* attempt to perform any additional alignment.
2982	*/
2983	if (!delay && !eof &&
2984	(orig_off >= gotp->br_startoff) &&
2985	(orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2986	return 0;
2987	}
2988
2989	/*
2990	* If the file offset is unaligned vs. the extent size
2991	* we need to align it. This will be possible unless
2992	* the file was previously written with a kernel that didn't
2993	* perform this alignment, or if a truncate shot us in the
2994	* foot.
2995	*/
2996	div_u64_rem(orig_off, extsz, &temp);
2997	if (temp) {
2998	align_alen += temp;
2999	align_off -= temp;
3000	}
3001
3002	/* Same adjustment for the end of the requested area. */
3003	temp = (align_alen % extsz);
3004	if (temp)
3005	align_alen += extsz - temp;
3006
3007	/*
3008	* For large extent hint sizes, the aligned extent might be larger than
3009	* XFS_BMBT_MAX_EXTLEN. In that case, reduce the size by an extsz so
3010	* that it pulls the length back under XFS_BMBT_MAX_EXTLEN. The outer
3011	* allocation loops handle short allocation just fine, so it is safe to
3012	* do this. We only want to do it when we are forced to, though, because
3013	* it means more allocation operations are required.
3014	*/
3015	while (align_alen > XFS_MAX_BMBT_EXTLEN)
3016	align_alen -= extsz;
3017	ASSERT(align_alen <= XFS_MAX_BMBT_EXTLEN);
3018
3019	/*
3020	* If the previous block overlaps with this proposed allocation
3021	* then move the start forward without adjusting the length.
3022	*/
3023	if (prevp->br_startoff != NULLFILEOFF) {
3024	if (prevp->br_startblock == HOLESTARTBLOCK)
3025	prevo = prevp->br_startoff;
3026	else
3027	prevo = prevp->br_startoff + prevp->br_blockcount;
3028	} else
3029	prevo = 0;
3030	if (align_off != orig_off && align_off < prevo)
3031	align_off = prevo;
3032	/*
3033	* If the next block overlaps with this proposed allocation
3034	* then move the start back without adjusting the length,
3035	* but not before offset 0.
3036	* This may of course make the start overlap previous block,
3037	* and if we hit the offset 0 limit then the next block
3038	* can still overlap too.
3039	*/
3040	if (!eof && gotp->br_startoff != NULLFILEOFF) {
3041	if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3042	(!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3043	nexto = gotp->br_startoff + gotp->br_blockcount;
3044	else
3045	nexto = gotp->br_startoff;
3046	} else
3047	nexto = NULLFILEOFF;
3048	if (!eof &&
3049	align_off + align_alen != orig_end &&
3050	align_off + align_alen > nexto)
3051	align_off = nexto > align_alen ? nexto - align_alen : 0;
3052	/*
3053	* If we're now overlapping the next or previous extent that
3054	* means we can't fit an extsz piece in this hole. Just move
3055	* the start forward to the first valid spot and set
3056	* the length so we hit the end.
3057	*/
3058	if (align_off != orig_off && align_off < prevo)
3059	align_off = prevo;
3060	if (align_off + align_alen != orig_end &&
3061	align_off + align_alen > nexto &&
3062	nexto != NULLFILEOFF) {
3063	ASSERT(nexto > prevo);
3064	align_alen = nexto - align_off;
3065	}
3066
3067	/*
3068	* If realtime, and the result isn't a multiple of the realtime
3069	* extent size we need to remove blocks until it is.
3070	*/
3071	if (rt && (temp = xfs_extlen_to_rtxmod(mp, align_alen))) {
3072	/*
3073	* We're not covering the original request, or
3074	* we won't be able to once we fix the length.
3075	*/
3076	if (orig_off < align_off ||
3077	orig_end > align_off + align_alen ||
3078	align_alen - temp < orig_alen)
3079	return -EINVAL;
3080	/*
3081	* Try to fix it by moving the start up.
3082	*/
3083	if (align_off + temp <= orig_off) {
3084	align_alen -= temp;
3085	align_off += temp;
3086	}
3087	/*
3088	* Try to fix it by moving the end in.
3089	*/
3090	else if (align_off + align_alen - temp >= orig_end)
3091	align_alen -= temp;
3092	/*
3093	* Set the start to the minimum then trim the length.
3094	*/
3095	else {
3096	align_alen -= orig_off - align_off;
3097	align_off = orig_off;
3098	align_alen -= xfs_extlen_to_rtxmod(mp, align_alen);
3099	}
3100	/*
3101	* Result doesn't cover the request, fail it.
3102	*/
3103	if (orig_off < align_off || orig_end > align_off + align_alen)
3104	return -EINVAL;
3105	} else {
3106	ASSERT(orig_off >= align_off);
3107	/* see XFS_BMBT_MAX_EXTLEN handling above */
3108	ASSERT(orig_end <= align_off + align_alen ||
3109	align_alen + extsz > XFS_MAX_BMBT_EXTLEN);
3110	}
3111
3112	#ifdef DEBUG
3113	if (!eof && gotp->br_startoff != NULLFILEOFF)
3114	ASSERT(align_off + align_alen <= gotp->br_startoff);
3115	if (prevp->br_startoff != NULLFILEOFF)
3116	ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3117	#endif
3118
3119	*lenp = align_alen;
3120	*offp = align_off;
3121	return 0;
3122	}
3123
3124	#define XFS_ALLOC_GAP_UNITS 4
3125
3126	/* returns true if ap->blkno was modified */
3127	bool
3128	xfs_bmap_adjacent(
3129	struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3130	{
3131	xfs_fsblock_t adjust; /* adjustment to block numbers */
3132	xfs_mount_t *mp; /* mount point structure */
3133	int rt; /* true if inode is realtime */
3134
3135	#define ISVALID(x,y) \
3136	(rt ? \
3137	(x) < mp->m_sb.sb_rblocks : \
3138	XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3139	XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3140	XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3141
3142	mp = ap->ip->i_mount;
3143	rt = XFS_IS_REALTIME_INODE(ap->ip) &&
3144	(ap->datatype & XFS_ALLOC_USERDATA);
3145	/*
3146	* If allocating at eof, and there's a previous real block,
3147	* try to use its last block as our starting point.
3148	*/
3149	if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3150	!isnullstartblock(ap->prev.br_startblock) &&
3151	ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3152	ap->prev.br_startblock)) {
3153	ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3154	/*
3155	* Adjust for the gap between prevp and us.
3156	*/
3157	adjust = ap->offset -
3158	(ap->prev.br_startoff + ap->prev.br_blockcount);
3159	if (adjust &&
3160	ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3161	ap->blkno += adjust;
3162	return true;
3163	}
3164	/*
3165	* If not at eof, then compare the two neighbor blocks.
3166	* Figure out whether either one gives us a good starting point,
3167	* and pick the better one.
3168	*/
3169	if (!ap->eof) {
3170	xfs_fsblock_t gotbno; /* right side block number */
3171	xfs_fsblock_t gotdiff=0; /* right side difference */
3172	xfs_fsblock_t prevbno; /* left side block number */
3173	xfs_fsblock_t prevdiff=0; /* left side difference */
3174
3175	/*
3176	* If there's a previous (left) block, select a requested
3177	* start block based on it.
3178	*/
3179	if (ap->prev.br_startoff != NULLFILEOFF &&
3180	!isnullstartblock(ap->prev.br_startblock) &&
3181	(prevbno = ap->prev.br_startblock +
3182	ap->prev.br_blockcount) &&
3183	ISVALID(prevbno, ap->prev.br_startblock)) {
3184	/*
3185	* Calculate gap to end of previous block.
3186	*/
3187	adjust = prevdiff = ap->offset -
3188	(ap->prev.br_startoff +
3189	ap->prev.br_blockcount);
3190	/*
3191	* Figure the startblock based on the previous block's
3192	* end and the gap size.
3193	* Heuristic!
3194	* If the gap is large relative to the piece we're
3195	* allocating, or using it gives us an invalid block
3196	* number, then just use the end of the previous block.
3197	*/
3198	if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3199	ISVALID(prevbno + prevdiff,
3200	ap->prev.br_startblock))
3201	prevbno += adjust;
3202	else
3203	prevdiff += adjust;
3204	}
3205	/*
3206	* No previous block or can't follow it, just default.
3207	*/
3208	else
3209	prevbno = NULLFSBLOCK;
3210	/*
3211	* If there's a following (right) block, select a requested
3212	* start block based on it.
3213	*/
3214	if (!isnullstartblock(ap->got.br_startblock)) {
3215	/*
3216	* Calculate gap to start of next block.
3217	*/
3218	adjust = gotdiff = ap->got.br_startoff - ap->offset;
3219	/*
3220	* Figure the startblock based on the next block's
3221	* start and the gap size.
3222	*/
3223	gotbno = ap->got.br_startblock;
3224	/*
3225	* Heuristic!
3226	* If the gap is large relative to the piece we're
3227	* allocating, or using it gives us an invalid block
3228	* number, then just use the start of the next block
3229	* offset by our length.
3230	*/
3231	if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3232	ISVALID(gotbno - gotdiff, gotbno))
3233	gotbno -= adjust;
3234	else if (ISVALID(gotbno - ap->length, gotbno)) {
3235	gotbno -= ap->length;
3236	gotdiff += adjust - ap->length;
3237	} else
3238	gotdiff += adjust;
3239	}
3240	/*
3241	* No next block, just default.
3242	*/
3243	else
3244	gotbno = NULLFSBLOCK;
3245	/*
3246	* If both valid, pick the better one, else the only good
3247	* one, else ap->blkno is already set (to 0 or the inode block).
3248	*/
3249	if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK) {
3250	ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3251	return true;
3252	}
3253	if (prevbno != NULLFSBLOCK) {
3254	ap->blkno = prevbno;
3255	return true;
3256	}
3257	if (gotbno != NULLFSBLOCK) {
3258	ap->blkno = gotbno;
3259	return true;
3260	}
3261	}
3262	#undef ISVALID
3263	return false;
3264	}
3265
3266	int
3267	xfs_bmap_longest_free_extent(
3268	struct xfs_perag *pag,
3269	struct xfs_trans *tp,
3270	xfs_extlen_t *blen)
3271	{
3272	xfs_extlen_t longest;
3273	int error = 0;
3274
3275	if (!xfs_perag_initialised_agf(pag)) {
3276	error = xfs_alloc_read_agf(pag, tp, XFS_ALLOC_FLAG_TRYLOCK,
3277	NULL);
3278	if (error)
3279	return error;
3280	}
3281
3282	longest = xfs_alloc_longest_free_extent(pag,
3283	xfs_alloc_min_freelist(pag->pag_mount, pag),
3284	xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE));
3285	if (*blen < longest)
3286	*blen = longest;
3287
3288	return 0;
3289	}
3290
3291	static xfs_extlen_t
3292	xfs_bmap_select_minlen(
3293	struct xfs_bmalloca *ap,
3294	struct xfs_alloc_arg *args,
3295	xfs_extlen_t blen)
3296	{
3297
3298	/*
3299	* Since we used XFS_ALLOC_FLAG_TRYLOCK in _longest_free_extent(), it is
3300	* possible that there is enough contiguous free space for this request.
3301	*/
3302	if (blen < ap->minlen)
3303	return ap->minlen;
3304
3305	/*
3306	* If the best seen length is less than the request length,
3307	* use the best as the minimum, otherwise we've got the maxlen we
3308	* were asked for.
3309	*/
3310	if (blen < args->maxlen)
3311	return blen;
3312	return args->maxlen;
3313	}
3314
3315	static int
3316	xfs_bmap_btalloc_select_lengths(
3317	struct xfs_bmalloca *ap,
3318	struct xfs_alloc_arg *args,
3319	xfs_extlen_t *blen)
3320	{
3321	struct xfs_mount *mp = args->mp;
3322	struct xfs_perag *pag;
3323	xfs_agnumber_t agno, startag;
3324	int error = 0;
3325
3326	if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3327	args->total = ap->minlen;
3328	args->minlen = ap->minlen;
3329	return 0;
3330	}
3331
3332	args->total = ap->total;
3333	startag = XFS_FSB_TO_AGNO(mp, ap->blkno);
3334	if (startag == NULLAGNUMBER)
3335	startag = 0;
3336
3337	*blen = 0;
3338	for_each_perag_wrap(mp, startag, agno, pag) {
3339	error = xfs_bmap_longest_free_extent(pag, args->tp, blen);
3340	if (error && error != -EAGAIN)
3341	break;
3342	error = 0;
3343	if (*blen >= args->maxlen)
3344	break;
3345	}
3346	if (pag)
3347	xfs_perag_rele(pag);
3348
3349	args->minlen = xfs_bmap_select_minlen(ap, args, *blen);
3350	return error;
3351	}
3352
3353	/* Update all inode and quota accounting for the allocation we just did. */
3354	void
3355	xfs_bmap_alloc_account(
3356	struct xfs_bmalloca *ap)
3357	{
3358	bool isrt = XFS_IS_REALTIME_INODE(ap->ip) &&
3359	!(ap->flags & XFS_BMAPI_ATTRFORK);
3360	uint fld;
3361
3362	if (ap->flags & XFS_BMAPI_COWFORK) {
3363	/*
3364	* COW fork blocks are in-core only and thus are treated as
3365	* in-core quota reservation (like delalloc blocks) even when
3366	* converted to real blocks. The quota reservation is not
3367	* accounted to disk until blocks are remapped to the data
3368	* fork. So if these blocks were previously delalloc, we
3369	* already have quota reservation and there's nothing to do
3370	* yet.
3371	*/
3372	if (ap->wasdel) {
3373	xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)ap->length);
3374	return;
3375	}
3376
3377	/*
3378	* Otherwise, we've allocated blocks in a hole. The transaction
3379	* has acquired in-core quota reservation for this extent.
3380	* Rather than account these as real blocks, however, we reduce
3381	* the transaction quota reservation based on the allocation.
3382	* This essentially transfers the transaction quota reservation
3383	* to that of a delalloc extent.
3384	*/
3385	ap->ip->i_delayed_blks += ap->length;
3386	xfs_trans_mod_dquot_byino(ap->tp, ap->ip, isrt ?
3387	XFS_TRANS_DQ_RES_RTBLKS : XFS_TRANS_DQ_RES_BLKS,
3388	-(long)ap->length);
3389	return;
3390	}
3391
3392	/* data/attr fork only */
3393	ap->ip->i_nblocks += ap->length;
3394	xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3395	if (ap->wasdel) {
3396	ap->ip->i_delayed_blks -= ap->length;
3397	xfs_mod_delalloc(ap->ip->i_mount, -(int64_t)ap->length);
3398	fld = isrt ? XFS_TRANS_DQ_DELRTBCOUNT : XFS_TRANS_DQ_DELBCOUNT;
3399	} else {
3400	fld = isrt ? XFS_TRANS_DQ_RTBCOUNT : XFS_TRANS_DQ_BCOUNT;
3401	}
3402
3403	xfs_trans_mod_dquot_byino(ap->tp, ap->ip, fld, ap->length);
3404	}
3405
3406	static int
3407	xfs_bmap_compute_alignments(
3408	struct xfs_bmalloca *ap,
3409	struct xfs_alloc_arg *args)
3410	{
3411	struct xfs_mount *mp = args->mp;
3412	xfs_extlen_t align = 0; /* minimum allocation alignment */
3413	int stripe_align = 0;
3414
3415	/* stripe alignment for allocation is determined by mount parameters */
3416	if (mp->m_swidth && xfs_has_swalloc(mp))
3417	stripe_align = mp->m_swidth;
3418	else if (mp->m_dalign)
3419	stripe_align = mp->m_dalign;
3420
3421	if (ap->flags & XFS_BMAPI_COWFORK)
3422	align = xfs_get_cowextsz_hint(ap->ip);
3423	else if (ap->datatype & XFS_ALLOC_USERDATA)
3424	align = xfs_get_extsz_hint(ap->ip);
3425	if (align) {
3426	if (xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, align, 0,
3427	ap->eof, 0, ap->conv, &ap->offset,
3428	&ap->length))
3429	ASSERT(0);
3430	ASSERT(ap->length);
3431	}
3432
3433	/* apply extent size hints if obtained earlier */
3434	if (align) {
3435	args->prod = align;
3436	div_u64_rem(ap->offset, args->prod, &args->mod);
3437	if (args->mod)
3438	args->mod = args->prod - args->mod;
3439	} else if (mp->m_sb.sb_blocksize >= PAGE_SIZE) {
3440	args->prod = 1;
3441	args->mod = 0;
3442	} else {
3443	args->prod = PAGE_SIZE >> mp->m_sb.sb_blocklog;
3444	div_u64_rem(ap->offset, args->prod, &args->mod);
3445	if (args->mod)
3446	args->mod = args->prod - args->mod;
3447	}
3448
3449	return stripe_align;
3450	}
3451
3452	static void
3453	xfs_bmap_process_allocated_extent(
3454	struct xfs_bmalloca *ap,
3455	struct xfs_alloc_arg *args,
3456	xfs_fileoff_t orig_offset,
3457	xfs_extlen_t orig_length)
3458	{
3459	ap->blkno = args->fsbno;
3460	ap->length = args->len;
3461	/*
3462	* If the extent size hint is active, we tried to round the
3463	* caller's allocation request offset down to extsz and the
3464	* length up to another extsz boundary. If we found a free
3465	* extent we mapped it in starting at this new offset. If the
3466	* newly mapped space isn't long enough to cover any of the
3467	* range of offsets that was originally requested, move the
3468	* mapping up so that we can fill as much of the caller's
3469	* original request as possible. Free space is apparently
3470	* very fragmented so we're unlikely to be able to satisfy the
3471	* hints anyway.
3472	*/
3473	if (ap->length <= orig_length)
3474	ap->offset = orig_offset;
3475	else if (ap->offset + ap->length < orig_offset + orig_length)
3476	ap->offset = orig_offset + orig_length - ap->length;
3477	xfs_bmap_alloc_account(ap);
3478	}
3479
3480	#ifdef DEBUG
3481	static int
3482	xfs_bmap_exact_minlen_extent_alloc(
3483	struct xfs_bmalloca *ap)
3484	{
3485	struct xfs_mount *mp = ap->ip->i_mount;
3486	struct xfs_alloc_arg args = { .tp = ap->tp, .mp = mp };
3487	xfs_fileoff_t orig_offset;
3488	xfs_extlen_t orig_length;
3489	int error;
3490
3491	ASSERT(ap->length);
3492
3493	if (ap->minlen != 1) {
3494	ap->blkno = NULLFSBLOCK;
3495	ap->length = 0;
3496	return 0;
3497	}
3498
3499	orig_offset = ap->offset;
3500	orig_length = ap->length;
3501
3502	args.alloc_minlen_only = 1;
3503
3504	xfs_bmap_compute_alignments(ap, &args);
3505
3506	/*
3507	* Unlike the longest extent available in an AG, we don't track
3508	* the length of an AG's shortest extent.
3509	* XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT is a debug only knob and
3510	* hence we can afford to start traversing from the 0th AG since
3511	* we need not be concerned about a drop in performance in
3512	* "debug only" code paths.
3513	*/
3514	ap->blkno = XFS_AGB_TO_FSB(mp, 0, 0);
3515
3516	args.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
3517	args.minlen = args.maxlen = ap->minlen;
3518	args.total = ap->total;
3519
3520	args.alignment = 1;
3521	args.minalignslop = 0;
3522
3523	args.minleft = ap->minleft;
3524	args.wasdel = ap->wasdel;
3525	args.resv = XFS_AG_RESV_NONE;
3526	args.datatype = ap->datatype;
3527
3528	error = xfs_alloc_vextent_first_ag(&args, ap->blkno);
3529	if (error)
3530	return error;
3531
3532	if (args.fsbno != NULLFSBLOCK) {
3533	xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3534	orig_length);
3535	} else {
3536	ap->blkno = NULLFSBLOCK;
3537	ap->length = 0;
3538	}
3539
3540	return 0;
3541	}
3542	#else
3543
3544	#define xfs_bmap_exact_minlen_extent_alloc(bma) (-EFSCORRUPTED)
3545
3546	#endif
3547
3548	/*
3549	* If we are not low on available data blocks and we are allocating at
3550	* EOF, optimise allocation for contiguous file extension and/or stripe
3551	* alignment of the new extent.
3552	*
3553	* NOTE: ap->aeof is only set if the allocation length is >= the
3554	* stripe unit and the allocation offset is at the end of file.
3555	*/
3556	static int
3557	xfs_bmap_btalloc_at_eof(
3558	struct xfs_bmalloca *ap,
3559	struct xfs_alloc_arg *args,
3560	xfs_extlen_t blen,
3561	int stripe_align,
3562	bool ag_only)
3563	{
3564	struct xfs_mount *mp = args->mp;
3565	struct xfs_perag *caller_pag = args->pag;
3566	int error;
3567
3568	/*
3569	* If there are already extents in the file, try an exact EOF block
3570	* allocation to extend the file as a contiguous extent. If that fails,
3571	* or it's the first allocation in a file, just try for a stripe aligned
3572	* allocation.
3573	*/
3574	if (ap->offset) {
3575	xfs_extlen_t nextminlen = 0;
3576
3577	/*
3578	* Compute the minlen+alignment for the next case. Set slop so
3579	* that the value of minlen+alignment+slop doesn't go up between
3580	* the calls.
3581	*/
3582	args->alignment = 1;
3583	if (blen > stripe_align && blen <= args->maxlen)
3584	nextminlen = blen - stripe_align;
3585	else
3586	nextminlen = args->minlen;
3587	if (nextminlen + stripe_align > args->minlen + 1)
3588	args->minalignslop = nextminlen + stripe_align -
3589	args->minlen - 1;
3590	else
3591	args->minalignslop = 0;
3592
3593	if (!caller_pag)
3594	args->pag = xfs_perag_get(mp, XFS_FSB_TO_AGNO(mp, ap->blkno));
3595	error = xfs_alloc_vextent_exact_bno(args, ap->blkno);
3596	if (!caller_pag) {
3597	xfs_perag_put(pag: args->pag);
3598	args->pag = NULL;
3599	}
3600	if (error)
3601	return error;
3602
3603	if (args->fsbno != NULLFSBLOCK)
3604	return 0;
3605	/*
3606	* Exact allocation failed. Reset to try an aligned allocation
3607	* according to the original allocation specification.
3608	*/
3609	args->alignment = stripe_align;
3610	args->minlen = nextminlen;
3611	args->minalignslop = 0;
3612	} else {
3613	/*
3614	* Adjust minlen to try and preserve alignment if we
3615	* can't guarantee an aligned maxlen extent.
3616	*/
3617	args->alignment = stripe_align;
3618	if (blen > args->alignment &&
3619	blen <= args->maxlen + args->alignment)
3620	args->minlen = blen - args->alignment;
3621	args->minalignslop = 0;
3622	}
3623
3624	if (ag_only) {
3625	error = xfs_alloc_vextent_near_bno(args, ap->blkno);
3626	} else {
3627	args->pag = NULL;
3628	error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3629	ASSERT(args->pag == NULL);
3630	args->pag = caller_pag;
3631	}
3632	if (error)
3633	return error;
3634
3635	if (args->fsbno != NULLFSBLOCK)
3636	return 0;
3637
3638	/*
3639	* Allocation failed, so turn return the allocation args to their
3640	* original non-aligned state so the caller can proceed on allocation
3641	* failure as if this function was never called.
3642	*/
3643	args->alignment = 1;
3644	return 0;
3645	}
3646
3647	/*
3648	* We have failed multiple allocation attempts so now are in a low space
3649	* allocation situation. Try a locality first full filesystem minimum length
3650	* allocation whilst still maintaining necessary total block reservation
3651	* requirements.
3652	*
3653	* If that fails, we are now critically low on space, so perform a last resort
3654	* allocation attempt: no reserve, no locality, blocking, minimum length, full
3655	* filesystem free space scan. We also indicate to future allocations in this
3656	* transaction that we are critically low on space so they don't waste time on
3657	* allocation modes that are unlikely to succeed.
3658	*/
3659	int
3660	xfs_bmap_btalloc_low_space(
3661	struct xfs_bmalloca *ap,
3662	struct xfs_alloc_arg *args)
3663	{
3664	int error;
3665
3666	if (args->minlen > ap->minlen) {
3667	args->minlen = ap->minlen;
3668	error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3669	if (error || args->fsbno != NULLFSBLOCK)
3670	return error;
3671	}
3672
3673	/* Last ditch attempt before failure is declared. */
3674	args->total = ap->minlen;
3675	error = xfs_alloc_vextent_first_ag(args, 0);
3676	if (error)
3677	return error;
3678	ap->tp->t_flags |= XFS_TRANS_LOWMODE;
3679	return 0;
3680	}
3681
3682	static int
3683	xfs_bmap_btalloc_filestreams(
3684	struct xfs_bmalloca *ap,
3685	struct xfs_alloc_arg *args,
3686	int stripe_align)
3687	{
3688	xfs_extlen_t blen = 0;
3689	int error = 0;
3690
3691
3692	error = xfs_filestream_select_ag(ap, args, &blen);
3693	if (error)
3694	return error;
3695	ASSERT(args->pag);
3696
3697	/*
3698	* If we are in low space mode, then optimal allocation will fail so
3699	* prepare for minimal allocation and jump to the low space algorithm
3700	* immediately.
3701	*/
3702	if (ap->tp->t_flags & XFS_TRANS_LOWMODE) {
3703	args->minlen = ap->minlen;
3704	ASSERT(args->fsbno == NULLFSBLOCK);
3705	goto out_low_space;
3706	}
3707
3708	args->minlen = xfs_bmap_select_minlen(ap, args, blen);
3709	if (ap->aeof)
3710	error = xfs_bmap_btalloc_at_eof(ap, args, blen, stripe_align,
3711	true);
3712
3713	if (!error && args->fsbno == NULLFSBLOCK)
3714	error = xfs_alloc_vextent_near_bno(args, ap->blkno);
3715
3716	out_low_space:
3717	/*
3718	* We are now done with the perag reference for the filestreams
3719	* association provided by xfs_filestream_select_ag(). Release it now as
3720	* we've either succeeded, had a fatal error or we are out of space and
3721	* need to do a full filesystem scan for free space which will take it's
3722	* own references.
3723	*/
3724	xfs_perag_rele(pag: args->pag);
3725	args->pag = NULL;
3726	if (error || args->fsbno != NULLFSBLOCK)
3727	return error;
3728
3729	return xfs_bmap_btalloc_low_space(ap, args);
3730	}
3731
3732	static int
3733	xfs_bmap_btalloc_best_length(
3734	struct xfs_bmalloca *ap,
3735	struct xfs_alloc_arg *args,
3736	int stripe_align)
3737	{
3738	xfs_extlen_t blen = 0;
3739	int error;
3740
3741	ap->blkno = XFS_INO_TO_FSB(args->mp, ap->ip->i_ino);
3742	xfs_bmap_adjacent(ap);
3743
3744	/*
3745	* Search for an allocation group with a single extent large enough for
3746	* the request. If one isn't found, then adjust the minimum allocation
3747	* size to the largest space found.
3748	*/
3749	error = xfs_bmap_btalloc_select_lengths(ap, args, &blen);
3750	if (error)
3751	return error;
3752
3753	/*
3754	* Don't attempt optimal EOF allocation if previous allocations barely
3755	* succeeded due to being near ENOSPC. It is highly unlikely we'll get
3756	* optimal or even aligned allocations in this case, so don't waste time
3757	* trying.
3758	*/
3759	if (ap->aeof && !(ap->tp->t_flags & XFS_TRANS_LOWMODE)) {
3760	error = xfs_bmap_btalloc_at_eof(ap, args, blen, stripe_align,
3761	false);
3762	if (error || args->fsbno != NULLFSBLOCK)
3763	return error;
3764	}
3765
3766	error = xfs_alloc_vextent_start_ag(args, ap->blkno);
3767	if (error || args->fsbno != NULLFSBLOCK)
3768	return error;
3769
3770	return xfs_bmap_btalloc_low_space(ap, args);
3771	}
3772
3773	static int
3774	xfs_bmap_btalloc(
3775	struct xfs_bmalloca *ap)
3776	{
3777	struct xfs_mount *mp = ap->ip->i_mount;
3778	struct xfs_alloc_arg args = {
3779	.tp = ap->tp,
3780	.mp = mp,
3781	.fsbno = NULLFSBLOCK,
3782	.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE,
3783	.minleft = ap->minleft,
3784	.wasdel = ap->wasdel,
3785	.resv = XFS_AG_RESV_NONE,
3786	.datatype = ap->datatype,
3787	.alignment = 1,
3788	.minalignslop = 0,
3789	};
3790	xfs_fileoff_t orig_offset;
3791	xfs_extlen_t orig_length;
3792	int error;
3793	int stripe_align;
3794
3795	ASSERT(ap->length);
3796	orig_offset = ap->offset;
3797	orig_length = ap->length;
3798
3799	stripe_align = xfs_bmap_compute_alignments(ap, args: &args);
3800
3801	/* Trim the allocation back to the maximum an AG can fit. */
3802	args.maxlen = min(ap->length, mp->m_ag_max_usable);
3803
3804	if ((ap->datatype & XFS_ALLOC_USERDATA) &&
3805	xfs_inode_is_filestream(ap->ip))
3806	error = xfs_bmap_btalloc_filestreams(ap, args: &args, stripe_align);
3807	else
3808	error = xfs_bmap_btalloc_best_length(ap, args: &args, stripe_align);
3809	if (error)
3810	return error;
3811
3812	if (args.fsbno != NULLFSBLOCK) {
3813	xfs_bmap_process_allocated_extent(ap, &args, orig_offset,
3814	orig_length);
3815	} else {
3816	ap->blkno = NULLFSBLOCK;
3817	ap->length = 0;
3818	}
3819	return 0;
3820	}
3821
3822	/* Trim extent to fit a logical block range. */
3823	void
3824	xfs_trim_extent(
3825	struct xfs_bmbt_irec *irec,
3826	xfs_fileoff_t bno,
3827	xfs_filblks_t len)
3828	{
3829	xfs_fileoff_t distance;
3830	xfs_fileoff_t end = bno + len;
3831
3832	if (irec->br_startoff + irec->br_blockcount <= bno ||
3833	irec->br_startoff >= end) {
3834	irec->br_blockcount = 0;
3835	return;
3836	}
3837
3838	if (irec->br_startoff < bno) {
3839	distance = bno - irec->br_startoff;
3840	if (isnullstartblock(irec->br_startblock))
3841	irec->br_startblock = DELAYSTARTBLOCK;
3842	if (irec->br_startblock != DELAYSTARTBLOCK &&
3843	irec->br_startblock != HOLESTARTBLOCK)
3844	irec->br_startblock += distance;
3845	irec->br_startoff += distance;
3846	irec->br_blockcount -= distance;
3847	}
3848
3849	if (end < irec->br_startoff + irec->br_blockcount) {
3850	distance = irec->br_startoff + irec->br_blockcount - end;
3851	irec->br_blockcount -= distance;
3852	}
3853	}
3854
3855	/*
3856	* Trim the returned map to the required bounds
3857	*/
3858	STATIC void
3859	xfs_bmapi_trim_map(
3860	struct xfs_bmbt_irec *mval,
3861	struct xfs_bmbt_irec *got,
3862	xfs_fileoff_t *bno,
3863	xfs_filblks_t len,
3864	xfs_fileoff_t obno,
3865	xfs_fileoff_t end,
3866	int n,
3867	uint32_t flags)
3868	{
3869	if ((flags & XFS_BMAPI_ENTIRE) ||
3870	got->br_startoff + got->br_blockcount <= obno) {
3871	*mval = *got;
3872	if (isnullstartblock(got->br_startblock))
3873	mval->br_startblock = DELAYSTARTBLOCK;
3874	return;
3875	}
3876
3877	if (obno > *bno)
3878	*bno = obno;
3879	ASSERT((*bno >= obno) || (n == 0));
3880	ASSERT(*bno < end);
3881	mval->br_startoff = *bno;
3882	if (isnullstartblock(got->br_startblock))
3883	mval->br_startblock = DELAYSTARTBLOCK;
3884	else
3885	mval->br_startblock = got->br_startblock +
3886	(*bno - got->br_startoff);
3887	/*
3888	* Return the minimum of what we got and what we asked for for
3889	* the length. We can use the len variable here because it is
3890	* modified below and we could have been there before coming
3891	* here if the first part of the allocation didn't overlap what
3892	* was asked for.
3893	*/
3894	mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3895	got->br_blockcount - (*bno - got->br_startoff));
3896	mval->br_state = got->br_state;
3897	ASSERT(mval->br_blockcount <= len);
3898	return;
3899	}
3900
3901	/*
3902	* Update and validate the extent map to return
3903	*/
3904	STATIC void
3905	xfs_bmapi_update_map(
3906	struct xfs_bmbt_irec **map,
3907	xfs_fileoff_t *bno,
3908	xfs_filblks_t *len,
3909	xfs_fileoff_t obno,
3910	xfs_fileoff_t end,
3911	int *n,
3912	uint32_t flags)
3913	{
3914	xfs_bmbt_irec_t *mval = *map;
3915
3916	ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3917	((mval->br_startoff + mval->br_blockcount) <= end));
3918	ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3919	(mval->br_startoff < obno));
3920
3921	*bno = mval->br_startoff + mval->br_blockcount;
3922	*len = end - *bno;
3923	if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3924	/* update previous map with new information */
3925	ASSERT(mval->br_startblock == mval[-1].br_startblock);
3926	ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3927	ASSERT(mval->br_state == mval[-1].br_state);
3928	mval[-1].br_blockcount = mval->br_blockcount;
3929	mval[-1].br_state = mval->br_state;
3930	} else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3931	mval[-1].br_startblock != DELAYSTARTBLOCK &&
3932	mval[-1].br_startblock != HOLESTARTBLOCK &&
3933	mval->br_startblock == mval[-1].br_startblock +
3934	mval[-1].br_blockcount &&
3935	mval[-1].br_state == mval->br_state) {
3936	ASSERT(mval->br_startoff ==
3937	mval[-1].br_startoff + mval[-1].br_blockcount);
3938	mval[-1].br_blockcount += mval->br_blockcount;
3939	} else if (*n > 0 &&
3940	mval->br_startblock == DELAYSTARTBLOCK &&
3941	mval[-1].br_startblock == DELAYSTARTBLOCK &&
3942	mval->br_startoff ==
3943	mval[-1].br_startoff + mval[-1].br_blockcount) {
3944	mval[-1].br_blockcount += mval->br_blockcount;
3945	mval[-1].br_state = mval->br_state;
3946	} else if (!((*n == 0) &&
3947	((mval->br_startoff + mval->br_blockcount) <=
3948	obno))) {
3949	mval++;
3950	(*n)++;
3951	}
3952	*map = mval;
3953	}
3954
3955	/*
3956	* Map file blocks to filesystem blocks without allocation.
3957	*/
3958	int
3959	xfs_bmapi_read(
3960	struct xfs_inode *ip,
3961	xfs_fileoff_t bno,
3962	xfs_filblks_t len,
3963	struct xfs_bmbt_irec *mval,
3964	int *nmap,
3965	uint32_t flags)
3966	{
3967	struct xfs_mount *mp = ip->i_mount;
3968	int whichfork = xfs_bmapi_whichfork(flags);
3969	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
3970	struct xfs_bmbt_irec got;
3971	xfs_fileoff_t obno;
3972	xfs_fileoff_t end;
3973	struct xfs_iext_cursor icur;
3974	int error;
3975	bool eof = false;
3976	int n = 0;
3977
3978	ASSERT(*nmap >= 1);
3979	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_ENTIRE)));
3980	xfs_assert_ilocked(ip, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL);
3981
3982	if (WARN_ON_ONCE(!ifp)) {
3983	xfs_bmap_mark_sick(ip, whichfork);
3984	return -EFSCORRUPTED;
3985	}
3986
3987	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
3988	XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
3989	xfs_bmap_mark_sick(ip, whichfork);
3990	return -EFSCORRUPTED;
3991	}
3992
3993	if (xfs_is_shutdown(mp))
3994	return -EIO;
3995
3996	XFS_STATS_INC(mp, xs_blk_mapr);
3997
3998	error = xfs_iread_extents(NULL, ip, whichfork);
3999	if (error)
4000	return error;
4001
4002	if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got))
4003	eof = true;
4004	end = bno + len;
4005	obno = bno;
4006
4007	while (bno < end && n < *nmap) {
4008	/* Reading past eof, act as though there's a hole up to end. */
4009	if (eof)
4010	got.br_startoff = end;
4011	if (got.br_startoff > bno) {
4012	/* Reading in a hole. */
4013	mval->br_startoff = bno;
4014	mval->br_startblock = HOLESTARTBLOCK;
4015	mval->br_blockcount =
4016	XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4017	mval->br_state = XFS_EXT_NORM;
4018	bno += mval->br_blockcount;
4019	len -= mval->br_blockcount;
4020	mval++;
4021	n++;
4022	continue;
4023	}
4024
4025	/* set up the extent map to return. */
4026	xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4027	xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4028
4029	/* If we're done, stop now. */
4030	if (bno >= end || n >= *nmap)
4031	break;
4032
4033	/* Else go on to the next record. */
4034	if (!xfs_iext_next_extent(ifp, &icur, &got))
4035	eof = true;
4036	}
4037	*nmap = n;
4038	return 0;
4039	}
4040
4041	/*
4042	* Add a delayed allocation extent to an inode. Blocks are reserved from the
4043	* global pool and the extent inserted into the inode in-core extent tree.
4044	*
4045	* On entry, got refers to the first extent beyond the offset of the extent to
4046	* allocate or eof is specified if no such extent exists. On return, got refers
4047	* to the extent record that was inserted to the inode fork.
4048	*
4049	* Note that the allocated extent may have been merged with contiguous extents
4050	* during insertion into the inode fork. Thus, got does not reflect the current
4051	* state of the inode fork on return. If necessary, the caller can use lastx to
4052	* look up the updated record in the inode fork.
4053	*/
4054	int
4055	xfs_bmapi_reserve_delalloc(
4056	struct xfs_inode *ip,
4057	int whichfork,
4058	xfs_fileoff_t off,
4059	xfs_filblks_t len,
4060	xfs_filblks_t prealloc,
4061	struct xfs_bmbt_irec *got,
4062	struct xfs_iext_cursor *icur,
4063	int eof)
4064	{
4065	struct xfs_mount *mp = ip->i_mount;
4066	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4067	xfs_extlen_t alen;
4068	xfs_extlen_t indlen;
4069	int error;
4070	xfs_fileoff_t aoff = off;
4071
4072	/*
4073	* Cap the alloc length. Keep track of prealloc so we know whether to
4074	* tag the inode before we return.
4075	*/
4076	alen = XFS_FILBLKS_MIN(len + prealloc, XFS_MAX_BMBT_EXTLEN);
4077	if (!eof)
4078	alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4079	if (prealloc && alen >= len)
4080	prealloc = alen - len;
4081
4082	/* Figure out the extent size, adjust alen */
4083	if (whichfork == XFS_COW_FORK) {
4084	struct xfs_bmbt_irec prev;
4085	xfs_extlen_t extsz = xfs_get_cowextsz_hint(ip);
4086
4087	if (!xfs_iext_peek_prev_extent(ifp, icur, &prev))
4088	prev.br_startoff = NULLFILEOFF;
4089
4090	error = xfs_bmap_extsize_align(mp, got, &prev, extsz, 0, eof,
4091	1, 0, &aoff, &alen);
4092	ASSERT(!error);
4093	}
4094
4095	/*
4096	* Make a transaction-less quota reservation for delayed allocation
4097	* blocks. This number gets adjusted later. We return if we haven't
4098	* allocated blocks already inside this loop.
4099	*/
4100	error = xfs_quota_reserve_blkres(ip, alen);
4101	if (error)
4102	return error;
4103
4104	/*
4105	* Split changing sb for alen and indlen since they could be coming
4106	* from different places.
4107	*/
4108	indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4109	ASSERT(indlen > 0);
4110
4111	error = xfs_mod_fdblocks(mp, -((int64_t)alen), false);
4112	if (error)
4113	goto out_unreserve_quota;
4114
4115	error = xfs_mod_fdblocks(mp, -((int64_t)indlen), false);
4116	if (error)
4117	goto out_unreserve_blocks;
4118
4119
4120	ip->i_delayed_blks += alen;
4121	xfs_mod_delalloc(ip->i_mount, alen + indlen);
4122
4123	got->br_startoff = aoff;
4124	got->br_startblock = nullstartblock(indlen);
4125	got->br_blockcount = alen;
4126	got->br_state = XFS_EXT_NORM;
4127
4128	xfs_bmap_add_extent_hole_delay(ip, whichfork, icur, got);
4129
4130	/*
4131	* Tag the inode if blocks were preallocated. Note that COW fork
4132	* preallocation can occur at the start or end of the extent, even when
4133	* prealloc == 0, so we must also check the aligned offset and length.
4134	*/
4135	if (whichfork == XFS_DATA_FORK && prealloc)
4136	xfs_inode_set_eofblocks_tag(ip);
4137	if (whichfork == XFS_COW_FORK && (prealloc || aoff < off || alen > len))
4138	xfs_inode_set_cowblocks_tag(ip);
4139
4140	return 0;
4141
4142	out_unreserve_blocks:
4143	xfs_mod_fdblocks(mp, alen, false);
4144	out_unreserve_quota:
4145	if (XFS_IS_QUOTA_ON(mp))
4146	xfs_quota_unreserve_blkres(ip, alen);
4147	return error;
4148	}
4149
4150	static int
4151	xfs_bmap_alloc_userdata(
4152	struct xfs_bmalloca *bma)
4153	{
4154	struct xfs_mount *mp = bma->ip->i_mount;
4155	int whichfork = xfs_bmapi_whichfork(bma->flags);
4156	int error;
4157
4158	/*
4159	* Set the data type being allocated. For the data fork, the first data
4160	* in the file is treated differently to all other allocations. For the
4161	* attribute fork, we only need to ensure the allocated range is not on
4162	* the busy list.
4163	*/
4164	bma->datatype = XFS_ALLOC_NOBUSY;
4165	if (whichfork == XFS_DATA_FORK || whichfork == XFS_COW_FORK) {
4166	bma->datatype |= XFS_ALLOC_USERDATA;
4167	if (bma->offset == 0)
4168	bma->datatype |= XFS_ALLOC_INITIAL_USER_DATA;
4169
4170	if (mp->m_dalign && bma->length >= mp->m_dalign) {
4171	error = xfs_bmap_isaeof(bma, whichfork);
4172	if (error)
4173	return error;
4174	}
4175
4176	if (XFS_IS_REALTIME_INODE(bma->ip))
4177	return xfs_bmap_rtalloc(bma);
4178	}
4179
4180	if (unlikely(XFS_TEST_ERROR(false, mp,
4181	XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4182	return xfs_bmap_exact_minlen_extent_alloc(bma);
4183
4184	return xfs_bmap_btalloc(ap: bma);
4185	}
4186
4187	static int
4188	xfs_bmapi_allocate(
4189	struct xfs_bmalloca *bma)
4190	{
4191	struct xfs_mount *mp = bma->ip->i_mount;
4192	int whichfork = xfs_bmapi_whichfork(bma->flags);
4193	struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
4194	int tmp_logflags = 0;
4195	int error;
4196
4197	ASSERT(bma->length > 0);
4198
4199	/*
4200	* For the wasdelay case, we could also just allocate the stuff asked
4201	* for in this bmap call but that wouldn't be as good.
4202	*/
4203	if (bma->wasdel) {
4204	bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4205	bma->offset = bma->got.br_startoff;
4206	if (!xfs_iext_peek_prev_extent(ifp, &bma->icur, &bma->prev))
4207	bma->prev.br_startoff = NULLFILEOFF;
4208	} else {
4209	bma->length = XFS_FILBLKS_MIN(bma->length, XFS_MAX_BMBT_EXTLEN);
4210	if (!bma->eof)
4211	bma->length = XFS_FILBLKS_MIN(bma->length,
4212	bma->got.br_startoff - bma->offset);
4213	}
4214
4215	if (bma->flags & XFS_BMAPI_CONTIG)
4216	bma->minlen = bma->length;
4217	else
4218	bma->minlen = 1;
4219
4220	if (bma->flags & XFS_BMAPI_METADATA) {
4221	if (unlikely(XFS_TEST_ERROR(false, mp,
4222	XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT)))
4223	error = xfs_bmap_exact_minlen_extent_alloc(bma);
4224	else
4225	error = xfs_bmap_btalloc(ap: bma);
4226	} else {
4227	error = xfs_bmap_alloc_userdata(bma);
4228	}
4229	if (error || bma->blkno == NULLFSBLOCK)
4230	return error;
4231
4232	if (bma->flags & XFS_BMAPI_ZERO) {
4233	error = xfs_zero_extent(bma->ip, bma->blkno, bma->length);
4234	if (error)
4235	return error;
4236	}
4237
4238	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur)
4239	bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4240	/*
4241	* Bump the number of extents we've allocated
4242	* in this call.
4243	*/
4244	bma->nallocs++;
4245
4246	if (bma->cur && bma->wasdel)
4247	bma->cur->bc_flags |= XFS_BTREE_BMBT_WASDEL;
4248
4249	bma->got.br_startoff = bma->offset;
4250	bma->got.br_startblock = bma->blkno;
4251	bma->got.br_blockcount = bma->length;
4252	bma->got.br_state = XFS_EXT_NORM;
4253
4254	if (bma->flags & XFS_BMAPI_PREALLOC)
4255	bma->got.br_state = XFS_EXT_UNWRITTEN;
4256
4257	if (bma->wasdel)
4258	error = xfs_bmap_add_extent_delay_real(bma, whichfork);
4259	else
4260	error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip,
4261	whichfork, &bma->icur, &bma->cur, &bma->got,
4262	&bma->logflags, bma->flags);
4263
4264	bma->logflags |= tmp_logflags;
4265	if (error)
4266	return error;
4267
4268	/*
4269	* Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4270	* or xfs_bmap_add_extent_hole_real might have merged it into one of
4271	* the neighbouring ones.
4272	*/
4273	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4274
4275	ASSERT(bma->got.br_startoff <= bma->offset);
4276	ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4277	bma->offset + bma->length);
4278	ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4279	bma->got.br_state == XFS_EXT_UNWRITTEN);
4280	return 0;
4281	}
4282
4283	STATIC int
4284	xfs_bmapi_convert_unwritten(
4285	struct xfs_bmalloca *bma,
4286	struct xfs_bmbt_irec *mval,
4287	xfs_filblks_t len,
4288	uint32_t flags)
4289	{
4290	int whichfork = xfs_bmapi_whichfork(flags);
4291	struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
4292	int tmp_logflags = 0;
4293	int error;
4294
4295	/* check if we need to do unwritten->real conversion */
4296	if (mval->br_state == XFS_EXT_UNWRITTEN &&
4297	(flags & XFS_BMAPI_PREALLOC))
4298	return 0;
4299
4300	/* check if we need to do real->unwritten conversion */
4301	if (mval->br_state == XFS_EXT_NORM &&
4302	(flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4303	(XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4304	return 0;
4305
4306	/*
4307	* Modify (by adding) the state flag, if writing.
4308	*/
4309	ASSERT(mval->br_blockcount <= len);
4310	if (ifp->if_format == XFS_DINODE_FMT_BTREE && !bma->cur) {
4311	bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4312	bma->ip, whichfork);
4313	}
4314	mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4315	? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4316
4317	/*
4318	* Before insertion into the bmbt, zero the range being converted
4319	* if required.
4320	*/
4321	if (flags & XFS_BMAPI_ZERO) {
4322	error = xfs_zero_extent(bma->ip, mval->br_startblock,
4323	mval->br_blockcount);
4324	if (error)
4325	return error;
4326	}
4327
4328	error = xfs_bmap_add_extent_unwritten_real(tp: bma->tp, ip: bma->ip, whichfork,
4329	icur: &bma->icur, curp: &bma->cur, new: mval, logflagsp: &tmp_logflags);
4330	/*
4331	* Log the inode core unconditionally in the unwritten extent conversion
4332	* path because the conversion might not have done so (e.g., if the
4333	* extent count hasn't changed). We need to make sure the inode is dirty
4334	* in the transaction for the sake of fsync(), even if nothing has
4335	* changed, because fsync() will not force the log for this transaction
4336	* unless it sees the inode pinned.
4337	*
4338	* Note: If we're only converting cow fork extents, there aren't
4339	* any on-disk updates to make, so we don't need to log anything.
4340	*/
4341	if (whichfork != XFS_COW_FORK)
4342	bma->logflags |= tmp_logflags | XFS_ILOG_CORE;
4343	if (error)
4344	return error;
4345
4346	/*
4347	* Update our extent pointer, given that
4348	* xfs_bmap_add_extent_unwritten_real might have merged it into one
4349	* of the neighbouring ones.
4350	*/
4351	xfs_iext_get_extent(ifp, &bma->icur, &bma->got);
4352
4353	/*
4354	* We may have combined previously unwritten space with written space,
4355	* so generate another request.
4356	*/
4357	if (mval->br_blockcount < len)
4358	return -EAGAIN;
4359	return 0;
4360	}
4361
4362	xfs_extlen_t
4363	xfs_bmapi_minleft(
4364	struct xfs_trans *tp,
4365	struct xfs_inode *ip,
4366	int fork)
4367	{
4368	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, fork);
4369
4370	if (tp && tp->t_highest_agno != NULLAGNUMBER)
4371	return 0;
4372	if (ifp->if_format != XFS_DINODE_FMT_BTREE)
4373	return 1;
4374	return be16_to_cpu(ifp->if_broot->bb_level) + 1;
4375	}
4376
4377	/*
4378	* Log whatever the flags say, even if error. Otherwise we might miss detecting
4379	* a case where the data is changed, there's an error, and it's not logged so we
4380	* don't shutdown when we should. Don't bother logging extents/btree changes if
4381	* we converted to the other format.
4382	*/
4383	static void
4384	xfs_bmapi_finish(
4385	struct xfs_bmalloca *bma,
4386	int whichfork,
4387	int error)
4388	{
4389	struct xfs_ifork *ifp = xfs_ifork_ptr(bma->ip, whichfork);
4390
4391	if ((bma->logflags & xfs_ilog_fext(w: whichfork)) &&
4392	ifp->if_format != XFS_DINODE_FMT_EXTENTS)
4393	bma->logflags &= ~xfs_ilog_fext(w: whichfork);
4394	else if ((bma->logflags & xfs_ilog_fbroot(w: whichfork)) &&
4395	ifp->if_format != XFS_DINODE_FMT_BTREE)
4396	bma->logflags &= ~xfs_ilog_fbroot(w: whichfork);
4397
4398	if (bma->logflags)
4399	xfs_trans_log_inode(bma->tp, bma->ip, bma->logflags);
4400	if (bma->cur)
4401	xfs_btree_del_cursor(cur: bma->cur, error);
4402	}
4403
4404	/*
4405	* Map file blocks to filesystem blocks, and allocate blocks or convert the
4406	* extent state if necessary. Details behaviour is controlled by the flags
4407	* parameter. Only allocates blocks from a single allocation group, to avoid
4408	* locking problems.
4409	*/
4410	int
4411	xfs_bmapi_write(
4412	struct xfs_trans *tp, /* transaction pointer */
4413	struct xfs_inode *ip, /* incore inode */
4414	xfs_fileoff_t bno, /* starting file offs. mapped */
4415	xfs_filblks_t len, /* length to map in file */
4416	uint32_t flags, /* XFS_BMAPI_... */
4417	xfs_extlen_t total, /* total blocks needed */
4418	struct xfs_bmbt_irec *mval, /* output: map values */
4419	int *nmap) /* i/o: mval size/count */
4420	{
4421	struct xfs_bmalloca bma = {
4422	.tp = tp,
4423	.ip = ip,
4424	.total = total,
4425	};
4426	struct xfs_mount *mp = ip->i_mount;
4427	int whichfork = xfs_bmapi_whichfork(flags);
4428	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4429	xfs_fileoff_t end; /* end of mapped file region */
4430	bool eof = false; /* after the end of extents */
4431	int error; /* error return */
4432	int n; /* current extent index */
4433	xfs_fileoff_t obno; /* old block number (offset) */
4434
4435	#ifdef DEBUG
4436	xfs_fileoff_t orig_bno; /* original block number value */
4437	int orig_flags; /* original flags arg value */
4438	xfs_filblks_t orig_len; /* original value of len arg */
4439	struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4440	int orig_nmap; /* original value of *nmap */
4441
4442	orig_bno = bno;
4443	orig_len = len;
4444	orig_flags = flags;
4445	orig_mval = mval;
4446	orig_nmap = *nmap;
4447	#endif
4448
4449	ASSERT(*nmap >= 1);
4450	ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4451	ASSERT(tp != NULL);
4452	ASSERT(len > 0);
4453	ASSERT(ifp->if_format != XFS_DINODE_FMT_LOCAL);
4454	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
4455	ASSERT(!(flags & XFS_BMAPI_REMAP));
4456
4457	/* zeroing is for currently only for data extents, not metadata */
4458	ASSERT((flags & (XFS_BMAPI_METADATA | XFS_BMAPI_ZERO)) !=
4459	(XFS_BMAPI_METADATA | XFS_BMAPI_ZERO));
4460	/*
4461	* we can allocate unwritten extents or pre-zero allocated blocks,
4462	* but it makes no sense to do both at once. This would result in
4463	* zeroing the unwritten extent twice, but it still being an
4464	* unwritten extent....
4465	*/
4466	ASSERT((flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO)) !=
4467	(XFS_BMAPI_PREALLOC | XFS_BMAPI_ZERO));
4468
4469	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4470	XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4471	xfs_bmap_mark_sick(ip, whichfork);
4472	return -EFSCORRUPTED;
4473	}
4474
4475	if (xfs_is_shutdown(mp))
4476	return -EIO;
4477
4478	XFS_STATS_INC(mp, xs_blk_mapw);
4479
4480	error = xfs_iread_extents(tp, ip, whichfork);
4481	if (error)
4482	goto error0;
4483
4484	if (!xfs_iext_lookup_extent(ip, ifp, bno, &bma.icur, &bma.got))
4485	eof = true;
4486	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4487	bma.prev.br_startoff = NULLFILEOFF;
4488	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4489
4490	n = 0;
4491	end = bno + len;
4492	obno = bno;
4493	while (bno < end && n < *nmap) {
4494	bool need_alloc = false, wasdelay = false;
4495
4496	/* in hole or beyond EOF? */
4497	if (eof || bma.got.br_startoff > bno) {
4498	/*
4499	* CoW fork conversions should /never/ hit EOF or
4500	* holes. There should always be something for us
4501	* to work on.
4502	*/
4503	ASSERT(!((flags & XFS_BMAPI_CONVERT) &&
4504	(flags & XFS_BMAPI_COWFORK)));
4505
4506	need_alloc = true;
4507	} else if (isnullstartblock(bma.got.br_startblock)) {
4508	wasdelay = true;
4509	}
4510
4511	/*
4512	* First, deal with the hole before the allocated space
4513	* that we found, if any.
4514	*/
4515	if (need_alloc || wasdelay) {
4516	bma.eof = eof;
4517	bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4518	bma.wasdel = wasdelay;
4519	bma.offset = bno;
4520	bma.flags = flags;
4521
4522	/*
4523	* There's a 32/64 bit type mismatch between the
4524	* allocation length request (which can be 64 bits in
4525	* length) and the bma length request, which is
4526	* xfs_extlen_t and therefore 32 bits. Hence we have to
4527	* check for 32-bit overflows and handle them here.
4528	*/
4529	if (len > (xfs_filblks_t)XFS_MAX_BMBT_EXTLEN)
4530	bma.length = XFS_MAX_BMBT_EXTLEN;
4531	else
4532	bma.length = len;
4533
4534	ASSERT(len > 0);
4535	ASSERT(bma.length > 0);
4536	error = xfs_bmapi_allocate(bma: &bma);
4537	if (error)
4538	goto error0;
4539	if (bma.blkno == NULLFSBLOCK)
4540	break;
4541
4542	/*
4543	* If this is a CoW allocation, record the data in
4544	* the refcount btree for orphan recovery.
4545	*/
4546	if (whichfork == XFS_COW_FORK)
4547	xfs_refcount_alloc_cow_extent(tp, bma.blkno,
4548	bma.length);
4549	}
4550
4551	/* Deal with the allocated space we found. */
4552	xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4553	end, n, flags);
4554
4555	/* Execute unwritten extent conversion if necessary */
4556	error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4557	if (error == -EAGAIN)
4558	continue;
4559	if (error)
4560	goto error0;
4561
4562	/* update the extent map to return */
4563	xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4564
4565	/*
4566	* If we're done, stop now. Stop when we've allocated
4567	* XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4568	* the transaction may get too big.
4569	*/
4570	if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4571	break;
4572
4573	/* Else go on to the next record. */
4574	bma.prev = bma.got;
4575	if (!xfs_iext_next_extent(ifp, &bma.icur, &bma.got))
4576	eof = true;
4577	}
4578	*nmap = n;
4579
4580	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4581	whichfork);
4582	if (error)
4583	goto error0;
4584
4585	ASSERT(ifp->if_format != XFS_DINODE_FMT_BTREE ||
4586	ifp->if_nextents > XFS_IFORK_MAXEXT(ip, whichfork));
4587	xfs_bmapi_finish(bma: &bma, whichfork, error: 0);
4588	xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4589	orig_nmap, *nmap);
4590	return 0;
4591	error0:
4592	xfs_bmapi_finish(bma: &bma, whichfork, error);
4593	return error;
4594	}
4595
4596	/*
4597	* Convert an existing delalloc extent to real blocks based on file offset. This
4598	* attempts to allocate the entire delalloc extent and may require multiple
4599	* invocations to allocate the target offset if a large enough physical extent
4600	* is not available.
4601	*/
4602	int
4603	xfs_bmapi_convert_delalloc(
4604	struct xfs_inode *ip,
4605	int whichfork,
4606	xfs_off_t offset,
4607	struct iomap *iomap,
4608	unsigned int *seq)
4609	{
4610	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4611	struct xfs_mount *mp = ip->i_mount;
4612	xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
4613	struct xfs_bmalloca bma = { NULL };
4614	uint16_t flags = 0;
4615	struct xfs_trans *tp;
4616	int error;
4617
4618	if (whichfork == XFS_COW_FORK)
4619	flags |= IOMAP_F_SHARED;
4620
4621	/*
4622	* Space for the extent and indirect blocks was reserved when the
4623	* delalloc extent was created so there's no need to do so here.
4624	*/
4625	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0,
4626	XFS_TRANS_RESERVE, &tp);
4627	if (error)
4628	return error;
4629
4630	xfs_ilock(ip, XFS_ILOCK_EXCL);
4631	xfs_trans_ijoin(tp, ip, 0);
4632
4633	error = xfs_iext_count_may_overflow(ip, whichfork,
4634	XFS_IEXT_ADD_NOSPLIT_CNT);
4635	if (error == -EFBIG)
4636	error = xfs_iext_count_upgrade(tp, ip,
4637	XFS_IEXT_ADD_NOSPLIT_CNT);
4638	if (error)
4639	goto out_trans_cancel;
4640
4641	if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb, &bma.icur, &bma.got) ||
4642	bma.got.br_startoff > offset_fsb) {
4643	/*
4644	* No extent found in the range we are trying to convert. This
4645	* should only happen for the COW fork, where another thread
4646	* might have moved the extent to the data fork in the meantime.
4647	*/
4648	WARN_ON_ONCE(whichfork != XFS_COW_FORK);
4649	error = -EAGAIN;
4650	goto out_trans_cancel;
4651	}
4652
4653	/*
4654	* If we find a real extent here we raced with another thread converting
4655	* the extent. Just return the real extent at this offset.
4656	*/
4657	if (!isnullstartblock(bma.got.br_startblock)) {
4658	xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags,
4659	xfs_iomap_inode_sequence(ip, flags));
4660	*seq = READ_ONCE(ifp->if_seq);
4661	goto out_trans_cancel;
4662	}
4663
4664	bma.tp = tp;
4665	bma.ip = ip;
4666	bma.wasdel = true;
4667	bma.offset = bma.got.br_startoff;
4668	bma.length = max_t(xfs_filblks_t, bma.got.br_blockcount,
4669	XFS_MAX_BMBT_EXTLEN);
4670	bma.minleft = xfs_bmapi_minleft(tp, ip, whichfork);
4671
4672	/*
4673	* When we're converting the delalloc reservations backing dirty pages
4674	* in the page cache, we must be careful about how we create the new
4675	* extents:
4676	*
4677	* New CoW fork extents are created unwritten, turned into real extents
4678	* when we're about to write the data to disk, and mapped into the data
4679	* fork after the write finishes. End of story.
4680	*
4681	* New data fork extents must be mapped in as unwritten and converted
4682	* to real extents after the write succeeds to avoid exposing stale
4683	* disk contents if we crash.
4684	*/
4685	bma.flags = XFS_BMAPI_PREALLOC;
4686	if (whichfork == XFS_COW_FORK)
4687	bma.flags |= XFS_BMAPI_COWFORK;
4688
4689	if (!xfs_iext_peek_prev_extent(ifp, &bma.icur, &bma.prev))
4690	bma.prev.br_startoff = NULLFILEOFF;
4691
4692	error = xfs_bmapi_allocate(bma: &bma);
4693	if (error)
4694	goto out_finish;
4695
4696	error = -ENOSPC;
4697	if (WARN_ON_ONCE(bma.blkno == NULLFSBLOCK))
4698	goto out_finish;
4699	if (WARN_ON_ONCE(!xfs_valid_startblock(ip, bma.got.br_startblock))) {
4700	xfs_bmap_mark_sick(ip, whichfork);
4701	error = -EFSCORRUPTED;
4702	goto out_finish;
4703	}
4704
4705	XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, bma.length));
4706	XFS_STATS_INC(mp, xs_xstrat_quick);
4707
4708	ASSERT(!isnullstartblock(bma.got.br_startblock));
4709	xfs_bmbt_to_iomap(ip, iomap, &bma.got, 0, flags,
4710	xfs_iomap_inode_sequence(ip, flags));
4711	*seq = READ_ONCE(ifp->if_seq);
4712
4713	if (whichfork == XFS_COW_FORK)
4714	xfs_refcount_alloc_cow_extent(tp, bma.blkno, bma.length);
4715
4716	error = xfs_bmap_btree_to_extents(tp, ip, bma.cur, &bma.logflags,
4717	whichfork);
4718	if (error)
4719	goto out_finish;
4720
4721	xfs_bmapi_finish(bma: &bma, whichfork, error: 0);
4722	error = xfs_trans_commit(tp);
4723	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4724	return error;
4725
4726	out_finish:
4727	xfs_bmapi_finish(bma: &bma, whichfork, error);
4728	out_trans_cancel:
4729	xfs_trans_cancel(tp);
4730	xfs_iunlock(ip, XFS_ILOCK_EXCL);
4731	return error;
4732	}
4733
4734	int
4735	xfs_bmapi_remap(
4736	struct xfs_trans *tp,
4737	struct xfs_inode *ip,
4738	xfs_fileoff_t bno,
4739	xfs_filblks_t len,
4740	xfs_fsblock_t startblock,
4741	uint32_t flags)
4742	{
4743	struct xfs_mount *mp = ip->i_mount;
4744	struct xfs_ifork *ifp;
4745	struct xfs_btree_cur *cur = NULL;
4746	struct xfs_bmbt_irec got;
4747	struct xfs_iext_cursor icur;
4748	int whichfork = xfs_bmapi_whichfork(flags);
4749	int logflags = 0, error;
4750
4751	ifp = xfs_ifork_ptr(ip, whichfork);
4752	ASSERT(len > 0);
4753	ASSERT(len <= (xfs_filblks_t)XFS_MAX_BMBT_EXTLEN);
4754	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
4755	ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC |
4756	XFS_BMAPI_NORMAP)));
4757	ASSERT((flags & (XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC)) !=
4758	(XFS_BMAPI_ATTRFORK | XFS_BMAPI_PREALLOC));
4759
4760	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
4761	XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
4762	xfs_bmap_mark_sick(ip, whichfork);
4763	return -EFSCORRUPTED;
4764	}
4765
4766	if (xfs_is_shutdown(mp))
4767	return -EIO;
4768
4769	error = xfs_iread_extents(tp, ip, whichfork);
4770	if (error)
4771	return error;
4772
4773	if (xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
4774	/* make sure we only reflink into a hole. */
4775	ASSERT(got.br_startoff > bno);
4776	ASSERT(got.br_startoff - bno >= len);
4777	}
4778
4779	ip->i_nblocks += len;
4780	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4781
4782	if (ifp->if_format == XFS_DINODE_FMT_BTREE)
4783	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
4784
4785	got.br_startoff = bno;
4786	got.br_startblock = startblock;
4787	got.br_blockcount = len;
4788	if (flags & XFS_BMAPI_PREALLOC)
4789	got.br_state = XFS_EXT_UNWRITTEN;
4790	else
4791	got.br_state = XFS_EXT_NORM;
4792
4793	error = xfs_bmap_add_extent_hole_real(tp, ip, whichfork, &icur,
4794	&cur, &got, &logflags, flags);
4795	if (error)
4796	goto error0;
4797
4798	error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags, whichfork);
4799
4800	error0:
4801	if (ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS)
4802	logflags &= ~XFS_ILOG_DEXT;
4803	else if (ip->i_df.if_format != XFS_DINODE_FMT_BTREE)
4804	logflags &= ~XFS_ILOG_DBROOT;
4805
4806	if (logflags)
4807	xfs_trans_log_inode(tp, ip, logflags);
4808	if (cur)
4809	xfs_btree_del_cursor(cur, error);
4810	return error;
4811	}
4812
4813	/*
4814	* When a delalloc extent is split (e.g., due to a hole punch), the original
4815	* indlen reservation must be shared across the two new extents that are left
4816	* behind.
4817	*
4818	* Given the original reservation and the worst case indlen for the two new
4819	* extents (as calculated by xfs_bmap_worst_indlen()), split the original
4820	* reservation fairly across the two new extents. If necessary, steal available
4821	* blocks from a deleted extent to make up a reservation deficiency (e.g., if
4822	* ores == 1). The number of stolen blocks is returned. The availability and
4823	* subsequent accounting of stolen blocks is the responsibility of the caller.
4824	*/
4825	static xfs_filblks_t
4826	xfs_bmap_split_indlen(
4827	xfs_filblks_t ores, /* original res. */
4828	xfs_filblks_t *indlen1, /* ext1 worst indlen */
4829	xfs_filblks_t *indlen2, /* ext2 worst indlen */
4830	xfs_filblks_t avail) /* stealable blocks */
4831	{
4832	xfs_filblks_t len1 = *indlen1;
4833	xfs_filblks_t len2 = *indlen2;
4834	xfs_filblks_t nres = len1 + len2; /* new total res. */
4835	xfs_filblks_t stolen = 0;
4836	xfs_filblks_t resfactor;
4837
4838	/*
4839	* Steal as many blocks as we can to try and satisfy the worst case
4840	* indlen for both new extents.
4841	*/
4842	if (ores < nres && avail)
4843	stolen = XFS_FILBLKS_MIN(nres - ores, avail);
4844	ores += stolen;
4845
4846	/* nothing else to do if we've satisfied the new reservation */
4847	if (ores >= nres)
4848	return stolen;
4849
4850	/*
4851	* We can't meet the total required reservation for the two extents.
4852	* Calculate the percent of the overall shortage between both extents
4853	* and apply this percentage to each of the requested indlen values.
4854	* This distributes the shortage fairly and reduces the chances that one
4855	* of the two extents is left with nothing when extents are repeatedly
4856	* split.
4857	*/
4858	resfactor = (ores * 100);
4859	do_div(resfactor, nres);
4860	len1 *= resfactor;
4861	do_div(len1, 100);
4862	len2 *= resfactor;
4863	do_div(len2, 100);
4864	ASSERT(len1 + len2 <= ores);
4865	ASSERT(len1 < *indlen1 && len2 < *indlen2);
4866
4867	/*
4868	* Hand out the remainder to each extent. If one of the two reservations
4869	* is zero, we want to make sure that one gets a block first. The loop
4870	* below starts with len1, so hand len2 a block right off the bat if it
4871	* is zero.
4872	*/
4873	ores -= (len1 + len2);
4874	ASSERT((*indlen1 - len1) + (*indlen2 - len2) >= ores);
4875	if (ores && !len2 && *indlen2) {
4876	len2++;
4877	ores--;
4878	}
4879	while (ores) {
4880	if (len1 < *indlen1) {
4881	len1++;
4882	ores--;
4883	}
4884	if (!ores)
4885	break;
4886	if (len2 < *indlen2) {
4887	len2++;
4888	ores--;
4889	}
4890	}
4891
4892	*indlen1 = len1;
4893	*indlen2 = len2;
4894
4895	return stolen;
4896	}
4897
4898	int
4899	xfs_bmap_del_extent_delay(
4900	struct xfs_inode *ip,
4901	int whichfork,
4902	struct xfs_iext_cursor *icur,
4903	struct xfs_bmbt_irec *got,
4904	struct xfs_bmbt_irec *del)
4905	{
4906	struct xfs_mount *mp = ip->i_mount;
4907	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
4908	struct xfs_bmbt_irec new;
4909	int64_t da_old, da_new, da_diff = 0;
4910	xfs_fileoff_t del_endoff, got_endoff;
4911	xfs_filblks_t got_indlen, new_indlen, stolen;
4912	uint32_t state = xfs_bmap_fork_to_state(whichfork);
4913	int error = 0;
4914	bool isrt;
4915
4916	XFS_STATS_INC(mp, xs_del_exlist);
4917
4918	isrt = xfs_ifork_is_realtime(ip, whichfork);
4919	del_endoff = del->br_startoff + del->br_blockcount;
4920	got_endoff = got->br_startoff + got->br_blockcount;
4921	da_old = startblockval(got->br_startblock);
4922	da_new = 0;
4923
4924	ASSERT(del->br_blockcount > 0);
4925	ASSERT(got->br_startoff <= del->br_startoff);
4926	ASSERT(got_endoff >= del_endoff);
4927
4928	if (isrt)
4929	xfs_mod_frextents(mp, xfs_rtb_to_rtx(mp, del->br_blockcount));
4930
4931	/*
4932	* Update the inode delalloc counter now and wait to update the
4933	* sb counters as we might have to borrow some blocks for the
4934	* indirect block accounting.
4935	*/
4936	ASSERT(!isrt);
4937	error = xfs_quota_unreserve_blkres(ip, del->br_blockcount);
4938	if (error)
4939	return error;
4940	ip->i_delayed_blks -= del->br_blockcount;
4941
4942	if (got->br_startoff == del->br_startoff)
4943	state |= BMAP_LEFT_FILLING;
4944	if (got_endoff == del_endoff)
4945	state |= BMAP_RIGHT_FILLING;
4946
4947	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
4948	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
4949	/*
4950	* Matches the whole extent. Delete the entry.
4951	*/
4952	xfs_iext_remove(ip, icur, state);
4953	xfs_iext_prev(ifp, icur);
4954	break;
4955	case BMAP_LEFT_FILLING:
4956	/*
4957	* Deleting the first part of the extent.
4958	*/
4959	got->br_startoff = del_endoff;
4960	got->br_blockcount -= del->br_blockcount;
4961	da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4962	got->br_blockcount), da_old);
4963	got->br_startblock = nullstartblock((int)da_new);
4964	xfs_iext_update_extent(ip, state, icur, got);
4965	break;
4966	case BMAP_RIGHT_FILLING:
4967	/*
4968	* Deleting the last part of the extent.
4969	*/
4970	got->br_blockcount = got->br_blockcount - del->br_blockcount;
4971	da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip,
4972	got->br_blockcount), da_old);
4973	got->br_startblock = nullstartblock((int)da_new);
4974	xfs_iext_update_extent(ip, state, icur, got);
4975	break;
4976	case 0:
4977	/*
4978	* Deleting the middle of the extent.
4979	*
4980	* Distribute the original indlen reservation across the two new
4981	* extents. Steal blocks from the deleted extent if necessary.
4982	* Stealing blocks simply fudges the fdblocks accounting below.
4983	* Warn if either of the new indlen reservations is zero as this
4984	* can lead to delalloc problems.
4985	*/
4986	got->br_blockcount = del->br_startoff - got->br_startoff;
4987	got_indlen = xfs_bmap_worst_indlen(ip, got->br_blockcount);
4988
4989	new.br_blockcount = got_endoff - del_endoff;
4990	new_indlen = xfs_bmap_worst_indlen(ip, new.br_blockcount);
4991
4992	WARN_ON_ONCE(!got_indlen || !new_indlen);
4993	stolen = xfs_bmap_split_indlen(da_old, &got_indlen, &new_indlen,
4994	del->br_blockcount);
4995
4996	got->br_startblock = nullstartblock((int)got_indlen);
4997
4998	new.br_startoff = del_endoff;
4999	new.br_state = got->br_state;
5000	new.br_startblock = nullstartblock((int)new_indlen);
5001
5002	xfs_iext_update_extent(ip, state, icur, got);
5003	xfs_iext_next(ifp, icur);
5004	xfs_iext_insert(ip, icur, &new, state);
5005
5006	da_new = got_indlen + new_indlen - stolen;
5007	del->br_blockcount -= stolen;
5008	break;
5009	}
5010
5011	ASSERT(da_old >= da_new);
5012	da_diff = da_old - da_new;
5013	if (!isrt)
5014	da_diff += del->br_blockcount;
5015	if (da_diff) {
5016	xfs_mod_fdblocks(mp, da_diff, false);
5017	xfs_mod_delalloc(mp, -da_diff);
5018	}
5019	return error;
5020	}
5021
5022	void
5023	xfs_bmap_del_extent_cow(
5024	struct xfs_inode *ip,
5025	struct xfs_iext_cursor *icur,
5026	struct xfs_bmbt_irec *got,
5027	struct xfs_bmbt_irec *del)
5028	{
5029	struct xfs_mount *mp = ip->i_mount;
5030	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
5031	struct xfs_bmbt_irec new;
5032	xfs_fileoff_t del_endoff, got_endoff;
5033	uint32_t state = BMAP_COWFORK;
5034
5035	XFS_STATS_INC(mp, xs_del_exlist);
5036
5037	del_endoff = del->br_startoff + del->br_blockcount;
5038	got_endoff = got->br_startoff + got->br_blockcount;
5039
5040	ASSERT(del->br_blockcount > 0);
5041	ASSERT(got->br_startoff <= del->br_startoff);
5042	ASSERT(got_endoff >= del_endoff);
5043	ASSERT(!isnullstartblock(got->br_startblock));
5044
5045	if (got->br_startoff == del->br_startoff)
5046	state |= BMAP_LEFT_FILLING;
5047	if (got_endoff == del_endoff)
5048	state |= BMAP_RIGHT_FILLING;
5049
5050	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5051	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5052	/*
5053	* Matches the whole extent. Delete the entry.
5054	*/
5055	xfs_iext_remove(ip, icur, state);
5056	xfs_iext_prev(ifp, icur);
5057	break;
5058	case BMAP_LEFT_FILLING:
5059	/*
5060	* Deleting the first part of the extent.
5061	*/
5062	got->br_startoff = del_endoff;
5063	got->br_blockcount -= del->br_blockcount;
5064	got->br_startblock = del->br_startblock + del->br_blockcount;
5065	xfs_iext_update_extent(ip, state, icur, got);
5066	break;
5067	case BMAP_RIGHT_FILLING:
5068	/*
5069	* Deleting the last part of the extent.
5070	*/
5071	got->br_blockcount -= del->br_blockcount;
5072	xfs_iext_update_extent(ip, state, icur, got);
5073	break;
5074	case 0:
5075	/*
5076	* Deleting the middle of the extent.
5077	*/
5078	got->br_blockcount = del->br_startoff - got->br_startoff;
5079
5080	new.br_startoff = del_endoff;
5081	new.br_blockcount = got_endoff - del_endoff;
5082	new.br_state = got->br_state;
5083	new.br_startblock = del->br_startblock + del->br_blockcount;
5084
5085	xfs_iext_update_extent(ip, state, icur, got);
5086	xfs_iext_next(ifp, icur);
5087	xfs_iext_insert(ip, icur, &new, state);
5088	break;
5089	}
5090	ip->i_delayed_blks -= del->br_blockcount;
5091	}
5092
5093	/*
5094	* Called by xfs_bmapi to update file extent records and the btree
5095	* after removing space.
5096	*/
5097	STATIC int /* error */
5098	xfs_bmap_del_extent_real(
5099	xfs_inode_t *ip, /* incore inode pointer */
5100	xfs_trans_t *tp, /* current transaction pointer */
5101	struct xfs_iext_cursor *icur,
5102	struct xfs_btree_cur *cur, /* if null, not a btree */
5103	xfs_bmbt_irec_t *del, /* data to remove from extents */
5104	int *logflagsp, /* inode logging flags */
5105	int whichfork, /* data or attr fork */
5106	uint32_t bflags) /* bmapi flags */
5107	{
5108	xfs_fsblock_t del_endblock=0; /* first block past del */
5109	xfs_fileoff_t del_endoff; /* first offset past del */
5110	int do_fx; /* free extent at end of routine */
5111	int error; /* error return value */
5112	struct xfs_bmbt_irec got; /* current extent entry */
5113	xfs_fileoff_t got_endoff; /* first offset past got */
5114	int i; /* temp state */
5115	struct xfs_ifork *ifp; /* inode fork pointer */
5116	xfs_mount_t *mp; /* mount structure */
5117	xfs_filblks_t nblks; /* quota/sb block count */
5118	xfs_bmbt_irec_t new; /* new record to be inserted */
5119	/* REFERENCED */
5120	uint qfield; /* quota field to update */
5121	uint32_t state = xfs_bmap_fork_to_state(whichfork);
5122	struct xfs_bmbt_irec old;
5123
5124	*logflagsp = 0;
5125
5126	mp = ip->i_mount;
5127	XFS_STATS_INC(mp, xs_del_exlist);
5128
5129	ifp = xfs_ifork_ptr(ip, whichfork);
5130	ASSERT(del->br_blockcount > 0);
5131	xfs_iext_get_extent(ifp, icur, &got);
5132	ASSERT(got.br_startoff <= del->br_startoff);
5133	del_endoff = del->br_startoff + del->br_blockcount;
5134	got_endoff = got.br_startoff + got.br_blockcount;
5135	ASSERT(got_endoff >= del_endoff);
5136	ASSERT(!isnullstartblock(got.br_startblock));
5137	qfield = 0;
5138
5139	/*
5140	* If it's the case where the directory code is running with no block
5141	* reservation, and the deleted block is in the middle of its extent,
5142	* and the resulting insert of an extent would cause transformation to
5143	* btree format, then reject it. The calling code will then swap blocks
5144	* around instead. We have to do this now, rather than waiting for the
5145	* conversion to btree format, since the transaction will be dirty then.
5146	*/
5147	if (tp->t_blk_res == 0 &&
5148	ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
5149	ifp->if_nextents >= XFS_IFORK_MAXEXT(ip, whichfork) &&
5150	del->br_startoff > got.br_startoff && del_endoff < got_endoff)
5151	return -ENOSPC;
5152
5153	*logflagsp = XFS_ILOG_CORE;
5154	if (xfs_ifork_is_realtime(ip, whichfork)) {
5155	if (!(bflags & XFS_BMAPI_REMAP)) {
5156	error = xfs_rtfree_blocks(tp, del->br_startblock,
5157	del->br_blockcount);
5158	if (error)
5159	return error;
5160	}
5161
5162	do_fx = 0;
5163	qfield = XFS_TRANS_DQ_RTBCOUNT;
5164	} else {
5165	do_fx = 1;
5166	qfield = XFS_TRANS_DQ_BCOUNT;
5167	}
5168	nblks = del->br_blockcount;
5169
5170	del_endblock = del->br_startblock + del->br_blockcount;
5171	if (cur) {
5172	error = xfs_bmbt_lookup_eq(cur, &got, &i);
5173	if (error)
5174	return error;
5175	if (XFS_IS_CORRUPT(mp, i != 1)) {
5176	xfs_btree_mark_sick(cur);
5177	return -EFSCORRUPTED;
5178	}
5179	}
5180
5181	if (got.br_startoff == del->br_startoff)
5182	state |= BMAP_LEFT_FILLING;
5183	if (got_endoff == del_endoff)
5184	state |= BMAP_RIGHT_FILLING;
5185
5186	switch (state & (BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING)) {
5187	case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
5188	/*
5189	* Matches the whole extent. Delete the entry.
5190	*/
5191	xfs_iext_remove(ip, icur, state);
5192	xfs_iext_prev(ifp, icur);
5193	ifp->if_nextents--;
5194
5195	*logflagsp |= XFS_ILOG_CORE;
5196	if (!cur) {
5197	*logflagsp |= xfs_ilog_fext(w: whichfork);
5198	break;
5199	}
5200	if ((error = xfs_btree_delete(cur, &i)))
5201	return error;
5202	if (XFS_IS_CORRUPT(mp, i != 1)) {
5203	xfs_btree_mark_sick(cur);
5204	return -EFSCORRUPTED;
5205	}
5206	break;
5207	case BMAP_LEFT_FILLING:
5208	/*
5209	* Deleting the first part of the extent.
5210	*/
5211	got.br_startoff = del_endoff;
5212	got.br_startblock = del_endblock;
5213	got.br_blockcount -= del->br_blockcount;
5214	xfs_iext_update_extent(ip, state, icur, &got);
5215	if (!cur) {
5216	*logflagsp |= xfs_ilog_fext(w: whichfork);
5217	break;
5218	}
5219	error = xfs_bmbt_update(cur, &got);
5220	if (error)
5221	return error;
5222	break;
5223	case BMAP_RIGHT_FILLING:
5224	/*
5225	* Deleting the last part of the extent.
5226	*/
5227	got.br_blockcount -= del->br_blockcount;
5228	xfs_iext_update_extent(ip, state, icur, &got);
5229	if (!cur) {
5230	*logflagsp |= xfs_ilog_fext(w: whichfork);
5231	break;
5232	}
5233	error = xfs_bmbt_update(cur, &got);
5234	if (error)
5235	return error;
5236	break;
5237	case 0:
5238	/*
5239	* Deleting the middle of the extent.
5240	*/
5241
5242	old = got;
5243
5244	got.br_blockcount = del->br_startoff - got.br_startoff;
5245	xfs_iext_update_extent(ip, state, icur, &got);
5246
5247	new.br_startoff = del_endoff;
5248	new.br_blockcount = got_endoff - del_endoff;
5249	new.br_state = got.br_state;
5250	new.br_startblock = del_endblock;
5251
5252	*logflagsp |= XFS_ILOG_CORE;
5253	if (cur) {
5254	error = xfs_bmbt_update(cur, &got);
5255	if (error)
5256	return error;
5257	error = xfs_btree_increment(cur, 0, &i);
5258	if (error)
5259	return error;
5260	cur->bc_rec.b = new;
5261	error = xfs_btree_insert(cur, &i);
5262	if (error && error != -ENOSPC)
5263	return error;
5264	/*
5265	* If get no-space back from btree insert, it tried a
5266	* split, and we have a zero block reservation. Fix up
5267	* our state and return the error.
5268	*/
5269	if (error == -ENOSPC) {
5270	/*
5271	* Reset the cursor, don't trust it after any
5272	* insert operation.
5273	*/
5274	error = xfs_bmbt_lookup_eq(cur, &got, &i);
5275	if (error)
5276	return error;
5277	if (XFS_IS_CORRUPT(mp, i != 1)) {
5278	xfs_btree_mark_sick(cur);
5279	return -EFSCORRUPTED;
5280	}
5281	/*
5282	* Update the btree record back
5283	* to the original value.
5284	*/
5285	error = xfs_bmbt_update(cur, &old);
5286	if (error)
5287	return error;
5288	/*
5289	* Reset the extent record back
5290	* to the original value.
5291	*/
5292	xfs_iext_update_extent(ip, state, icur, &old);
5293	*logflagsp = 0;
5294	return -ENOSPC;
5295	}
5296	if (XFS_IS_CORRUPT(mp, i != 1)) {
5297	xfs_btree_mark_sick(cur);
5298	return -EFSCORRUPTED;
5299	}
5300	} else
5301	*logflagsp |= xfs_ilog_fext(w: whichfork);
5302
5303	ifp->if_nextents++;
5304	xfs_iext_next(ifp, icur);
5305	xfs_iext_insert(ip, icur, &new, state);
5306	break;
5307	}
5308
5309	/* remove reverse mapping */
5310	xfs_rmap_unmap_extent(tp: tp, ip: ip, whichfork, imap: del);
5311
5312	/*
5313	* If we need to, add to list of extents to delete.
5314	*/
5315	if (do_fx && !(bflags & XFS_BMAPI_REMAP)) {
5316	if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
5317	xfs_refcount_decrease_extent(tp: tp, irec: del);
5318	} else {
5319	error = xfs_free_extent_later(tp, del->br_startblock,
5320	del->br_blockcount, NULL,
5321	XFS_AG_RESV_NONE,
5322	((bflags & XFS_BMAPI_NODISCARD) ||
5323	del->br_state == XFS_EXT_UNWRITTEN));
5324	if (error)
5325	return error;
5326	}
5327	}
5328
5329	/*
5330	* Adjust inode # blocks in the file.
5331	*/
5332	if (nblks)
5333	ip->i_nblocks -= nblks;
5334	/*
5335	* Adjust quota data.
5336	*/
5337	if (qfield && !(bflags & XFS_BMAPI_REMAP))
5338	xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5339
5340	return 0;
5341	}
5342
5343	/*
5344	* Unmap (remove) blocks from a file.
5345	* If nexts is nonzero then the number of extents to remove is limited to
5346	* that value. If not all extents in the block range can be removed then
5347	* *done is set.
5348	*/
5349	static int
5350	__xfs_bunmapi(
5351	struct xfs_trans *tp, /* transaction pointer */
5352	struct xfs_inode *ip, /* incore inode */
5353	xfs_fileoff_t start, /* first file offset deleted */
5354	xfs_filblks_t *rlen, /* i/o: amount remaining */
5355	uint32_t flags, /* misc flags */
5356	xfs_extnum_t nexts) /* number of extents max */
5357	{
5358	struct xfs_btree_cur *cur; /* bmap btree cursor */
5359	struct xfs_bmbt_irec del; /* extent being deleted */
5360	int error; /* error return value */
5361	xfs_extnum_t extno; /* extent number in list */
5362	struct xfs_bmbt_irec got; /* current extent record */
5363	struct xfs_ifork *ifp; /* inode fork pointer */
5364	int isrt; /* freeing in rt area */
5365	int logflags; /* transaction logging flags */
5366	xfs_extlen_t mod; /* rt extent offset */
5367	struct xfs_mount *mp = ip->i_mount;
5368	int tmp_logflags; /* partial logging flags */
5369	int wasdel; /* was a delayed alloc extent */
5370	int whichfork; /* data or attribute fork */
5371	xfs_filblks_t len = *rlen; /* length to unmap in file */
5372	xfs_fileoff_t end;
5373	struct xfs_iext_cursor icur;
5374	bool done = false;
5375
5376	trace_xfs_bunmap(ip, start, len, flags, _RET_IP_);
5377
5378	whichfork = xfs_bmapi_whichfork(flags);
5379	ASSERT(whichfork != XFS_COW_FORK);
5380	ifp = xfs_ifork_ptr(ip, whichfork);
5381	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp))) {
5382	xfs_bmap_mark_sick(ip, whichfork);
5383	return -EFSCORRUPTED;
5384	}
5385	if (xfs_is_shutdown(mp))
5386	return -EIO;
5387
5388	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
5389	ASSERT(len > 0);
5390	ASSERT(nexts >= 0);
5391
5392	error = xfs_iread_extents(tp, ip, whichfork);
5393	if (error)
5394	return error;
5395
5396	if (xfs_iext_count(ifp) == 0) {
5397	*rlen = 0;
5398	return 0;
5399	}
5400	XFS_STATS_INC(mp, xs_blk_unmap);
5401	isrt = xfs_ifork_is_realtime(ip, whichfork);
5402	end = start + len;
5403
5404	if (!xfs_iext_lookup_extent_before(ip, ifp, &end, &icur, &got)) {
5405	*rlen = 0;
5406	return 0;
5407	}
5408	end--;
5409
5410	logflags = 0;
5411	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
5412	ASSERT(ifp->if_format == XFS_DINODE_FMT_BTREE);
5413	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5414	} else
5415	cur = NULL;
5416
5417	if (isrt) {
5418	/*
5419	* Synchronize by locking the bitmap inode.
5420	*/
5421	xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
5422	xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5423	xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
5424	xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
5425	}
5426
5427	extno = 0;
5428	while (end != (xfs_fileoff_t)-1 && end >= start &&
5429	(nexts == 0 || extno < nexts)) {
5430	/*
5431	* Is the found extent after a hole in which end lives?
5432	* Just back up to the previous extent, if so.
5433	*/
5434	if (got.br_startoff > end &&
5435	!xfs_iext_prev_extent(ifp, &icur, &got)) {
5436	done = true;
5437	break;
5438	}
5439	/*
5440	* Is the last block of this extent before the range
5441	* we're supposed to delete? If so, we're done.
5442	*/
5443	end = XFS_FILEOFF_MIN(end,
5444	got.br_startoff + got.br_blockcount - 1);
5445	if (end < start)
5446	break;
5447	/*
5448	* Then deal with the (possibly delayed) allocated space
5449	* we found.
5450	*/
5451	del = got;
5452	wasdel = isnullstartblock(del.br_startblock);
5453
5454	if (got.br_startoff < start) {
5455	del.br_startoff = start;
5456	del.br_blockcount -= start - got.br_startoff;
5457	if (!wasdel)
5458	del.br_startblock += start - got.br_startoff;
5459	}
5460	if (del.br_startoff + del.br_blockcount > end + 1)
5461	del.br_blockcount = end + 1 - del.br_startoff;
5462
5463	if (!isrt || (flags & XFS_BMAPI_REMAP))
5464	goto delete;
5465
5466	mod = xfs_rtb_to_rtxoff(mp,
5467	del.br_startblock + del.br_blockcount);
5468	if (mod) {
5469	/*
5470	* Realtime extent not lined up at the end.
5471	* The extent could have been split into written
5472	* and unwritten pieces, or we could just be
5473	* unmapping part of it. But we can't really
5474	* get rid of part of a realtime extent.
5475	*/
5476	if (del.br_state == XFS_EXT_UNWRITTEN) {
5477	/*
5478	* This piece is unwritten, or we're not
5479	* using unwritten extents. Skip over it.
5480	*/
5481	ASSERT((flags & XFS_BMAPI_REMAP) || end >= mod);
5482	end -= mod > del.br_blockcount ?
5483	del.br_blockcount : mod;
5484	if (end < got.br_startoff &&
5485	!xfs_iext_prev_extent(ifp, &icur, &got)) {
5486	done = true;
5487	break;
5488	}
5489	continue;
5490	}
5491	/*
5492	* It's written, turn it unwritten.
5493	* This is better than zeroing it.
5494	*/
5495	ASSERT(del.br_state == XFS_EXT_NORM);
5496	ASSERT(tp->t_blk_res > 0);
5497	/*
5498	* If this spans a realtime extent boundary,
5499	* chop it back to the start of the one we end at.
5500	*/
5501	if (del.br_blockcount > mod) {
5502	del.br_startoff += del.br_blockcount - mod;
5503	del.br_startblock += del.br_blockcount - mod;
5504	del.br_blockcount = mod;
5505	}
5506	del.br_state = XFS_EXT_UNWRITTEN;
5507	error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5508	whichfork, icur: &icur, curp: &cur, new: &del,
5509	logflagsp: &logflags);
5510	if (error)
5511	goto error0;
5512	goto nodelete;
5513	}
5514
5515	mod = xfs_rtb_to_rtxoff(mp, del.br_startblock);
5516	if (mod) {
5517	xfs_extlen_t off = mp->m_sb.sb_rextsize - mod;
5518
5519	/*
5520	* Realtime extent is lined up at the end but not
5521	* at the front. We'll get rid of full extents if
5522	* we can.
5523	*/
5524	if (del.br_blockcount > off) {
5525	del.br_blockcount -= off;
5526	del.br_startoff += off;
5527	del.br_startblock += off;
5528	} else if (del.br_startoff == start &&
5529	(del.br_state == XFS_EXT_UNWRITTEN ||
5530	tp->t_blk_res == 0)) {
5531	/*
5532	* Can't make it unwritten. There isn't
5533	* a full extent here so just skip it.
5534	*/
5535	ASSERT(end >= del.br_blockcount);
5536	end -= del.br_blockcount;
5537	if (got.br_startoff > end &&
5538	!xfs_iext_prev_extent(ifp, &icur, &got)) {
5539	done = true;
5540	break;
5541	}
5542	continue;
5543	} else if (del.br_state == XFS_EXT_UNWRITTEN) {
5544	struct xfs_bmbt_irec prev;
5545	xfs_fileoff_t unwrite_start;
5546
5547	/*
5548	* This one is already unwritten.
5549	* It must have a written left neighbor.
5550	* Unwrite the killed part of that one and
5551	* try again.
5552	*/
5553	if (!xfs_iext_prev_extent(ifp, &icur, &prev))
5554	ASSERT(0);
5555	ASSERT(prev.br_state == XFS_EXT_NORM);
5556	ASSERT(!isnullstartblock(prev.br_startblock));
5557	ASSERT(del.br_startblock ==
5558	prev.br_startblock + prev.br_blockcount);
5559	unwrite_start = max3(start,
5560	del.br_startoff - mod,
5561	prev.br_startoff);
5562	mod = unwrite_start - prev.br_startoff;
5563	prev.br_startoff = unwrite_start;
5564	prev.br_startblock += mod;
5565	prev.br_blockcount -= mod;
5566	prev.br_state = XFS_EXT_UNWRITTEN;
5567	error = xfs_bmap_add_extent_unwritten_real(tp,
5568	ip, whichfork, icur: &icur, curp: &cur,
5569	new: &prev, logflagsp: &logflags);
5570	if (error)
5571	goto error0;
5572	goto nodelete;
5573	} else {
5574	ASSERT(del.br_state == XFS_EXT_NORM);
5575	del.br_state = XFS_EXT_UNWRITTEN;
5576	error = xfs_bmap_add_extent_unwritten_real(tp,
5577	ip, whichfork, icur: &icur, curp: &cur,
5578	new: &del, logflagsp: &logflags);
5579	if (error)
5580	goto error0;
5581	goto nodelete;
5582	}
5583	}
5584
5585	delete:
5586	if (wasdel) {
5587	error = xfs_bmap_del_extent_delay(ip, whichfork, icur: &icur,
5588	got: &got, del: &del);
5589	} else {
5590	error = xfs_bmap_del_extent_real(ip, tp, &icur, cur,
5591	&del, &tmp_logflags, whichfork,
5592	flags);
5593	logflags |= tmp_logflags;
5594	}
5595
5596	if (error)
5597	goto error0;
5598
5599	end = del.br_startoff - 1;
5600	nodelete:
5601	/*
5602	* If not done go on to the next (previous) record.
5603	*/
5604	if (end != (xfs_fileoff_t)-1 && end >= start) {
5605	if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5606	(got.br_startoff > end &&
5607	!xfs_iext_prev_extent(ifp, &icur, &got))) {
5608	done = true;
5609	break;
5610	}
5611	extno++;
5612	}
5613	}
5614	if (done || end == (xfs_fileoff_t)-1 || end < start)
5615	*rlen = 0;
5616	else
5617	*rlen = end - start + 1;
5618
5619	/*
5620	* Convert to a btree if necessary.
5621	*/
5622	if (xfs_bmap_needs_btree(ip, whichfork)) {
5623	ASSERT(cur == NULL);
5624	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
5625	&tmp_logflags, whichfork);
5626	logflags |= tmp_logflags;
5627	} else {
5628	error = xfs_bmap_btree_to_extents(tp, ip, cur, &logflags,
5629	whichfork);
5630	}
5631
5632	error0:
5633	/*
5634	* Log everything. Do this after conversion, there's no point in
5635	* logging the extent records if we've converted to btree format.
5636	*/
5637	if ((logflags & xfs_ilog_fext(w: whichfork)) &&
5638	ifp->if_format != XFS_DINODE_FMT_EXTENTS)
5639	logflags &= ~xfs_ilog_fext(w: whichfork);
5640	else if ((logflags & xfs_ilog_fbroot(w: whichfork)) &&
5641	ifp->if_format != XFS_DINODE_FMT_BTREE)
5642	logflags &= ~xfs_ilog_fbroot(w: whichfork);
5643	/*
5644	* Log inode even in the error case, if the transaction
5645	* is dirty we'll need to shut down the filesystem.
5646	*/
5647	if (logflags)
5648	xfs_trans_log_inode(tp, ip, logflags);
5649	if (cur) {
5650	if (!error)
5651	cur->bc_bmap.allocated = 0;
5652	xfs_btree_del_cursor(cur, error);
5653	}
5654	return error;
5655	}
5656
5657	/* Unmap a range of a file. */
5658	int
5659	xfs_bunmapi(
5660	xfs_trans_t *tp,
5661	struct xfs_inode *ip,
5662	xfs_fileoff_t bno,
5663	xfs_filblks_t len,
5664	uint32_t flags,
5665	xfs_extnum_t nexts,
5666	int *done)
5667	{
5668	int error;
5669
5670	error = __xfs_bunmapi(tp, ip, bno, &len, flags, nexts);
5671	*done = (len == 0);
5672	return error;
5673	}
5674
5675	/*
5676	* Determine whether an extent shift can be accomplished by a merge with the
5677	* extent that precedes the target hole of the shift.
5678	*/
5679	STATIC bool
5680	xfs_bmse_can_merge(
5681	struct xfs_bmbt_irec *left, /* preceding extent */
5682	struct xfs_bmbt_irec *got, /* current extent to shift */
5683	xfs_fileoff_t shift) /* shift fsb */
5684	{
5685	xfs_fileoff_t startoff;
5686
5687	startoff = got->br_startoff - shift;
5688
5689	/*
5690	* The extent, once shifted, must be adjacent in-file and on-disk with
5691	* the preceding extent.
5692	*/
5693	if ((left->br_startoff + left->br_blockcount != startoff) ||
5694	(left->br_startblock + left->br_blockcount != got->br_startblock) ||
5695	(left->br_state != got->br_state) ||
5696	(left->br_blockcount + got->br_blockcount > XFS_MAX_BMBT_EXTLEN))
5697	return false;
5698
5699	return true;
5700	}
5701
5702	/*
5703	* A bmap extent shift adjusts the file offset of an extent to fill a preceding
5704	* hole in the file. If an extent shift would result in the extent being fully
5705	* adjacent to the extent that currently precedes the hole, we can merge with
5706	* the preceding extent rather than do the shift.
5707	*
5708	* This function assumes the caller has verified a shift-by-merge is possible
5709	* with the provided extents via xfs_bmse_can_merge().
5710	*/
5711	STATIC int
5712	xfs_bmse_merge(
5713	struct xfs_trans *tp,
5714	struct xfs_inode *ip,
5715	int whichfork,
5716	xfs_fileoff_t shift, /* shift fsb */
5717	struct xfs_iext_cursor *icur,
5718	struct xfs_bmbt_irec *got, /* extent to shift */
5719	struct xfs_bmbt_irec *left, /* preceding extent */
5720	struct xfs_btree_cur *cur,
5721	int *logflags) /* output */
5722	{
5723	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
5724	struct xfs_bmbt_irec new;
5725	xfs_filblks_t blockcount;
5726	int error, i;
5727	struct xfs_mount *mp = ip->i_mount;
5728
5729	blockcount = left->br_blockcount + got->br_blockcount;
5730
5731	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
5732	ASSERT(xfs_bmse_can_merge(left, got, shift));
5733
5734	new = *left;
5735	new.br_blockcount = blockcount;
5736
5737	/*
5738	* Update the on-disk extent count, the btree if necessary and log the
5739	* inode.
5740	*/
5741	ifp->if_nextents--;
5742	*logflags |= XFS_ILOG_CORE;
5743	if (!cur) {
5744	*logflags |= XFS_ILOG_DEXT;
5745	goto done;
5746	}
5747
5748	/* lookup and remove the extent to merge */
5749	error = xfs_bmbt_lookup_eq(cur, got, &i);
5750	if (error)
5751	return error;
5752	if (XFS_IS_CORRUPT(mp, i != 1)) {
5753	xfs_btree_mark_sick(cur);
5754	return -EFSCORRUPTED;
5755	}
5756
5757	error = xfs_btree_delete(cur, &i);
5758	if (error)
5759	return error;
5760	if (XFS_IS_CORRUPT(mp, i != 1)) {
5761	xfs_btree_mark_sick(cur);
5762	return -EFSCORRUPTED;
5763	}
5764
5765	/* lookup and update size of the previous extent */
5766	error = xfs_bmbt_lookup_eq(cur, left, &i);
5767	if (error)
5768	return error;
5769	if (XFS_IS_CORRUPT(mp, i != 1)) {
5770	xfs_btree_mark_sick(cur);
5771	return -EFSCORRUPTED;
5772	}
5773
5774	error = xfs_bmbt_update(cur, &new);
5775	if (error)
5776	return error;
5777
5778	/* change to extent format if required after extent removal */
5779	error = xfs_bmap_btree_to_extents(tp, ip, cur, logflags, whichfork);
5780	if (error)
5781	return error;
5782
5783	done:
5784	xfs_iext_remove(ip, icur, 0);
5785	xfs_iext_prev(ifp, icur);
5786	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5787	&new);
5788
5789	/* update reverse mapping. rmap functions merge the rmaps for us */
5790	xfs_rmap_unmap_extent(tp, ip, whichfork, imap: got);
5791	memcpy(&new, got, sizeof(new));
5792	new.br_startoff = left->br_startoff + left->br_blockcount;
5793	xfs_rmap_map_extent(tp, ip, whichfork, imap: &new);
5794	return 0;
5795	}
5796
5797	static int
5798	xfs_bmap_shift_update_extent(
5799	struct xfs_trans *tp,
5800	struct xfs_inode *ip,
5801	int whichfork,
5802	struct xfs_iext_cursor *icur,
5803	struct xfs_bmbt_irec *got,
5804	struct xfs_btree_cur *cur,
5805	int *logflags,
5806	xfs_fileoff_t startoff)
5807	{
5808	struct xfs_mount *mp = ip->i_mount;
5809	struct xfs_bmbt_irec prev = *got;
5810	int error, i;
5811
5812	*logflags |= XFS_ILOG_CORE;
5813
5814	got->br_startoff = startoff;
5815
5816	if (cur) {
5817	error = xfs_bmbt_lookup_eq(cur, &prev, &i);
5818	if (error)
5819	return error;
5820	if (XFS_IS_CORRUPT(mp, i != 1)) {
5821	xfs_btree_mark_sick(cur);
5822	return -EFSCORRUPTED;
5823	}
5824
5825	error = xfs_bmbt_update(cur, got);
5826	if (error)
5827	return error;
5828	} else {
5829	*logflags |= XFS_ILOG_DEXT;
5830	}
5831
5832	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), icur,
5833	got);
5834
5835	/* update reverse mapping */
5836	xfs_rmap_unmap_extent(tp, ip, whichfork, imap: &prev);
5837	xfs_rmap_map_extent(tp, ip, whichfork, imap: got);
5838	return 0;
5839	}
5840
5841	int
5842	xfs_bmap_collapse_extents(
5843	struct xfs_trans *tp,
5844	struct xfs_inode *ip,
5845	xfs_fileoff_t *next_fsb,
5846	xfs_fileoff_t offset_shift_fsb,
5847	bool *done)
5848	{
5849	int whichfork = XFS_DATA_FORK;
5850	struct xfs_mount *mp = ip->i_mount;
5851	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
5852	struct xfs_btree_cur *cur = NULL;
5853	struct xfs_bmbt_irec got, prev;
5854	struct xfs_iext_cursor icur;
5855	xfs_fileoff_t new_startoff;
5856	int error = 0;
5857	int logflags = 0;
5858
5859	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5860	XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5861	xfs_bmap_mark_sick(ip, whichfork);
5862	return -EFSCORRUPTED;
5863	}
5864
5865	if (xfs_is_shutdown(mp))
5866	return -EIO;
5867
5868	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
5869
5870	error = xfs_iread_extents(tp, ip, whichfork);
5871	if (error)
5872	return error;
5873
5874	if (ifp->if_format == XFS_DINODE_FMT_BTREE)
5875	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5876
5877	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
5878	*done = true;
5879	goto del_cursor;
5880	}
5881	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
5882	xfs_bmap_mark_sick(ip, whichfork);
5883	error = -EFSCORRUPTED;
5884	goto del_cursor;
5885	}
5886
5887	new_startoff = got.br_startoff - offset_shift_fsb;
5888	if (xfs_iext_peek_prev_extent(ifp, &icur, &prev)) {
5889	if (new_startoff < prev.br_startoff + prev.br_blockcount) {
5890	error = -EINVAL;
5891	goto del_cursor;
5892	}
5893
5894	if (xfs_bmse_can_merge(&prev, &got, offset_shift_fsb)) {
5895	error = xfs_bmse_merge(tp, ip, whichfork,
5896	offset_shift_fsb, &icur, &got, &prev,
5897	cur, &logflags);
5898	if (error)
5899	goto del_cursor;
5900	goto done;
5901	}
5902	} else {
5903	if (got.br_startoff < offset_shift_fsb) {
5904	error = -EINVAL;
5905	goto del_cursor;
5906	}
5907	}
5908
5909	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
5910	cur, &logflags, new_startoff);
5911	if (error)
5912	goto del_cursor;
5913
5914	done:
5915	if (!xfs_iext_next_extent(ifp, &icur, &got)) {
5916	*done = true;
5917	goto del_cursor;
5918	}
5919
5920	*next_fsb = got.br_startoff;
5921	del_cursor:
5922	if (cur)
5923	xfs_btree_del_cursor(cur, error);
5924	if (logflags)
5925	xfs_trans_log_inode(tp, ip, logflags);
5926	return error;
5927	}
5928
5929	/* Make sure we won't be right-shifting an extent past the maximum bound. */
5930	int
5931	xfs_bmap_can_insert_extents(
5932	struct xfs_inode *ip,
5933	xfs_fileoff_t off,
5934	xfs_fileoff_t shift)
5935	{
5936	struct xfs_bmbt_irec got;
5937	int is_empty;
5938	int error = 0;
5939
5940	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);
5941
5942	if (xfs_is_shutdown(ip->i_mount))
5943	return -EIO;
5944
5945	xfs_ilock(ip, XFS_ILOCK_EXCL);
5946	error = xfs_bmap_last_extent(NULL, ip, XFS_DATA_FORK, &got, &is_empty);
5947	if (!error && !is_empty && got.br_startoff >= off &&
5948	((got.br_startoff + shift) & BMBT_STARTOFF_MASK) < got.br_startoff)
5949	error = -EINVAL;
5950	xfs_iunlock(ip, XFS_ILOCK_EXCL);
5951
5952	return error;
5953	}
5954
5955	int
5956	xfs_bmap_insert_extents(
5957	struct xfs_trans *tp,
5958	struct xfs_inode *ip,
5959	xfs_fileoff_t *next_fsb,
5960	xfs_fileoff_t offset_shift_fsb,
5961	bool *done,
5962	xfs_fileoff_t stop_fsb)
5963	{
5964	int whichfork = XFS_DATA_FORK;
5965	struct xfs_mount *mp = ip->i_mount;
5966	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
5967	struct xfs_btree_cur *cur = NULL;
5968	struct xfs_bmbt_irec got, next;
5969	struct xfs_iext_cursor icur;
5970	xfs_fileoff_t new_startoff;
5971	int error = 0;
5972	int logflags = 0;
5973
5974	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
5975	XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
5976	xfs_bmap_mark_sick(ip, whichfork);
5977	return -EFSCORRUPTED;
5978	}
5979
5980	if (xfs_is_shutdown(mp))
5981	return -EIO;
5982
5983	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
5984
5985	error = xfs_iread_extents(tp, ip, whichfork);
5986	if (error)
5987	return error;
5988
5989	if (ifp->if_format == XFS_DINODE_FMT_BTREE)
5990	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5991
5992	if (*next_fsb == NULLFSBLOCK) {
5993	xfs_iext_last(ifp, &icur);
5994	if (!xfs_iext_get_extent(ifp, &icur, &got) ||
5995	stop_fsb > got.br_startoff) {
5996	*done = true;
5997	goto del_cursor;
5998	}
5999	} else {
6000	if (!xfs_iext_lookup_extent(ip, ifp, *next_fsb, &icur, &got)) {
6001	*done = true;
6002	goto del_cursor;
6003	}
6004	}
6005	if (XFS_IS_CORRUPT(mp, isnullstartblock(got.br_startblock))) {
6006	xfs_bmap_mark_sick(ip, whichfork);
6007	error = -EFSCORRUPTED;
6008	goto del_cursor;
6009	}
6010
6011	if (XFS_IS_CORRUPT(mp, stop_fsb > got.br_startoff)) {
6012	xfs_bmap_mark_sick(ip, whichfork);
6013	error = -EFSCORRUPTED;
6014	goto del_cursor;
6015	}
6016
6017	new_startoff = got.br_startoff + offset_shift_fsb;
6018	if (xfs_iext_peek_next_extent(ifp, &icur, &next)) {
6019	if (new_startoff + got.br_blockcount > next.br_startoff) {
6020	error = -EINVAL;
6021	goto del_cursor;
6022	}
6023
6024	/*
6025	* Unlike a left shift (which involves a hole punch), a right
6026	* shift does not modify extent neighbors in any way. We should
6027	* never find mergeable extents in this scenario. Check anyways
6028	* and warn if we encounter two extents that could be one.
6029	*/
6030	if (xfs_bmse_can_merge(&got, &next, offset_shift_fsb))
6031	WARN_ON_ONCE(1);
6032	}
6033
6034	error = xfs_bmap_shift_update_extent(tp, ip, whichfork, &icur, &got,
6035	cur, &logflags, new_startoff);
6036	if (error)
6037	goto del_cursor;
6038
6039	if (!xfs_iext_prev_extent(ifp, &icur, &got) ||
6040	stop_fsb >= got.br_startoff + got.br_blockcount) {
6041	*done = true;
6042	goto del_cursor;
6043	}
6044
6045	*next_fsb = got.br_startoff;
6046	del_cursor:
6047	if (cur)
6048	xfs_btree_del_cursor(cur, error);
6049	if (logflags)
6050	xfs_trans_log_inode(tp, ip, logflags);
6051	return error;
6052	}
6053
6054	/*
6055	* Splits an extent into two extents at split_fsb block such that it is the
6056	* first block of the current_ext. @ext is a target extent to be split.
6057	* @split_fsb is a block where the extents is split. If split_fsb lies in a
6058	* hole or the first block of extents, just return 0.
6059	*/
6060	int
6061	xfs_bmap_split_extent(
6062	struct xfs_trans *tp,
6063	struct xfs_inode *ip,
6064	xfs_fileoff_t split_fsb)
6065	{
6066	int whichfork = XFS_DATA_FORK;
6067	struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
6068	struct xfs_btree_cur *cur = NULL;
6069	struct xfs_bmbt_irec got;
6070	struct xfs_bmbt_irec new; /* split extent */
6071	struct xfs_mount *mp = ip->i_mount;
6072	xfs_fsblock_t gotblkcnt; /* new block count for got */
6073	struct xfs_iext_cursor icur;
6074	int error = 0;
6075	int logflags = 0;
6076	int i = 0;
6077
6078	if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(ifp)) ||
6079	XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
6080	xfs_bmap_mark_sick(ip, whichfork);
6081	return -EFSCORRUPTED;
6082	}
6083
6084	if (xfs_is_shutdown(mp))
6085	return -EIO;
6086
6087	/* Read in all the extents */
6088	error = xfs_iread_extents(tp, ip, whichfork);
6089	if (error)
6090	return error;
6091
6092	/*
6093	* If there are not extents, or split_fsb lies in a hole we are done.
6094	*/
6095	if (!xfs_iext_lookup_extent(ip, ifp, split_fsb, &icur, &got) ||
6096	got.br_startoff >= split_fsb)
6097	return 0;
6098
6099	gotblkcnt = split_fsb - got.br_startoff;
6100	new.br_startoff = split_fsb;
6101	new.br_startblock = got.br_startblock + gotblkcnt;
6102	new.br_blockcount = got.br_blockcount - gotblkcnt;
6103	new.br_state = got.br_state;
6104
6105	if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
6106	cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
6107	error = xfs_bmbt_lookup_eq(cur, &got, &i);
6108	if (error)
6109	goto del_cursor;
6110	if (XFS_IS_CORRUPT(mp, i != 1)) {
6111	xfs_btree_mark_sick(cur);
6112	error = -EFSCORRUPTED;
6113	goto del_cursor;
6114	}
6115	}
6116
6117	got.br_blockcount = gotblkcnt;
6118	xfs_iext_update_extent(ip, xfs_bmap_fork_to_state(whichfork), &icur,
6119	&got);
6120
6121	logflags = XFS_ILOG_CORE;
6122	if (cur) {
6123	error = xfs_bmbt_update(cur, &got);
6124	if (error)
6125	goto del_cursor;
6126	} else
6127	logflags |= XFS_ILOG_DEXT;
6128
6129	/* Add new extent */
6130	xfs_iext_next(ifp, &icur);
6131	xfs_iext_insert(ip, &icur, &new, 0);
6132	ifp->if_nextents++;
6133
6134	if (cur) {
6135	error = xfs_bmbt_lookup_eq(cur, &new, &i);
6136	if (error)
6137	goto del_cursor;
6138	if (XFS_IS_CORRUPT(mp, i != 0)) {
6139	xfs_btree_mark_sick(cur);
6140	error = -EFSCORRUPTED;
6141	goto del_cursor;
6142	}
6143	error = xfs_btree_insert(cur, &i);
6144	if (error)
6145	goto del_cursor;
6146	if (XFS_IS_CORRUPT(mp, i != 1)) {
6147	xfs_btree_mark_sick(cur);
6148	error = -EFSCORRUPTED;
6149	goto del_cursor;
6150	}
6151	}
6152
6153	/*
6154	* Convert to a btree if necessary.
6155	*/
6156	if (xfs_bmap_needs_btree(ip, whichfork)) {
6157	int tmp_logflags; /* partial log flag return val */
6158
6159	ASSERT(cur == NULL);
6160	error = xfs_bmap_extents_to_btree(tp, ip, &cur, 0,
6161	&tmp_logflags, whichfork);
6162	logflags |= tmp_logflags;
6163	}
6164
6165	del_cursor:
6166	if (cur) {
6167	cur->bc_bmap.allocated = 0;
6168	xfs_btree_del_cursor(cur, error);
6169	}
6170
6171	if (logflags)
6172	xfs_trans_log_inode(tp, ip, logflags);
6173	return error;
6174	}
6175
6176	/* Record a bmap intent. */
6177	static inline void
6178	__xfs_bmap_add(
6179	struct xfs_trans *tp,
6180	enum xfs_bmap_intent_type type,
6181	struct xfs_inode *ip,
6182	int whichfork,
6183	struct xfs_bmbt_irec *bmap)
6184	{
6185	struct xfs_bmap_intent *bi;
6186
6187	if ((whichfork != XFS_DATA_FORK && whichfork != XFS_ATTR_FORK) ||
6188	bmap->br_startblock == HOLESTARTBLOCK ||
6189	bmap->br_startblock == DELAYSTARTBLOCK)
6190	return;
6191
6192	bi = kmem_cache_alloc(xfs_bmap_intent_cache, GFP_KERNEL | __GFP_NOFAIL);
6193	INIT_LIST_HEAD(&bi->bi_list);
6194	bi->bi_type = type;
6195	bi->bi_owner = ip;
6196	bi->bi_whichfork = whichfork;
6197	bi->bi_bmap = *bmap;
6198
6199	xfs_bmap_defer_add(tp, bi);
6200	}
6201
6202	/* Map an extent into a file. */
6203	void
6204	xfs_bmap_map_extent(
6205	struct xfs_trans *tp,
6206	struct xfs_inode *ip,
6207	int whichfork,
6208	struct xfs_bmbt_irec *PREV)
6209	{
6210	__xfs_bmap_add(tp, type: XFS_BMAP_MAP, ip, whichfork, bmap: PREV);
6211	}
6212
6213	/* Unmap an extent out of a file. */
6214	void
6215	xfs_bmap_unmap_extent(
6216	struct xfs_trans *tp,
6217	struct xfs_inode *ip,
6218	int whichfork,
6219	struct xfs_bmbt_irec *PREV)
6220	{
6221	__xfs_bmap_add(tp, type: XFS_BMAP_UNMAP, ip, whichfork, bmap: PREV);
6222	}
6223
6224	/*
6225	* Process one of the deferred bmap operations. We pass back the
6226	* btree cursor to maintain our lock on the bmapbt between calls.
6227	*/
6228	int
6229	xfs_bmap_finish_one(
6230	struct xfs_trans *tp,
6231	struct xfs_bmap_intent *bi)
6232	{
6233	struct xfs_bmbt_irec *bmap = &bi->bi_bmap;
6234	int error = 0;
6235	int flags = 0;
6236
6237	if (bi->bi_whichfork == XFS_ATTR_FORK)
6238	flags |= XFS_BMAPI_ATTRFORK;
6239
6240	ASSERT(tp->t_highest_agno == NULLAGNUMBER);
6241
6242	trace_xfs_bmap_deferred(bi);
6243
6244	if (XFS_TEST_ERROR(false, tp->t_mountp, XFS_ERRTAG_BMAP_FINISH_ONE))
6245	return -EIO;
6246
6247	switch (bi->bi_type) {
6248	case XFS_BMAP_MAP:
6249	if (bi->bi_bmap.br_state == XFS_EXT_UNWRITTEN)
6250	flags |= XFS_BMAPI_PREALLOC;
6251	error = xfs_bmapi_remap(tp, bi->bi_owner, bmap->br_startoff,
6252	bmap->br_blockcount, bmap->br_startblock,
6253	flags);
6254	bmap->br_blockcount = 0;
6255	break;
6256	case XFS_BMAP_UNMAP:
6257	error = __xfs_bunmapi(tp, bi->bi_owner, bmap->br_startoff,
6258	&bmap->br_blockcount, flags | XFS_BMAPI_REMAP,
6259	1);
6260	break;
6261	default:
6262	ASSERT(0);
6263	xfs_bmap_mark_sick(ip: bi->bi_owner, whichfork: bi->bi_whichfork);
6264	error = -EFSCORRUPTED;
6265	}
6266
6267	return error;
6268	}
6269
6270	/* Check that an extent does not have invalid flags or bad ranges. */
6271	xfs_failaddr_t
6272	xfs_bmap_validate_extent_raw(
6273	struct xfs_mount *mp,
6274	bool rtfile,
6275	int whichfork,
6276	struct xfs_bmbt_irec *irec)
6277	{
6278	if (!xfs_verify_fileext(mp, irec->br_startoff, irec->br_blockcount))
6279	return __this_address;
6280
6281	if (rtfile && whichfork == XFS_DATA_FORK) {
6282	if (!xfs_verify_rtbext(mp, irec->br_startblock,
6283	irec->br_blockcount))
6284	return __this_address;
6285	} else {
6286	if (!xfs_verify_fsbext(mp, irec->br_startblock,
6287	irec->br_blockcount))
6288	return __this_address;
6289	}
6290	if (irec->br_state != XFS_EXT_NORM && whichfork != XFS_DATA_FORK)
6291	return __this_address;
6292	return NULL;
6293	}
6294
6295	int __init
6296	xfs_bmap_intent_init_cache(void)
6297	{
6298	xfs_bmap_intent_cache = kmem_cache_create("xfs_bmap_intent",
6299	sizeof(struct xfs_bmap_intent),
6300	0, 0, NULL);
6301
6302	return xfs_bmap_intent_cache != NULL ? 0 : -ENOMEM;
6303	}
6304
6305	void
6306	xfs_bmap_intent_destroy_cache(void)
6307	{
6308	kmem_cache_destroy(xfs_bmap_intent_cache);
6309	xfs_bmap_intent_cache = NULL;
6310	}
6311
6312	/* Check that an inode's extent does not have invalid flags or bad ranges. */
6313	xfs_failaddr_t
6314	xfs_bmap_validate_extent(
6315	struct xfs_inode *ip,
6316	int whichfork,
6317	struct xfs_bmbt_irec *irec)
6318	{
6319	return xfs_bmap_validate_extent_raw(ip->i_mount,
6320	XFS_IS_REALTIME_INODE(ip), whichfork, irec);
6321	}
6322
6323	/*
6324	* Used in xfs_itruncate_extents(). This is the maximum number of extents
6325	* freed from a file in a single transaction.
6326	*/
6327	#define XFS_ITRUNC_MAX_EXTENTS 2
6328
6329	/*
6330	* Unmap every extent in part of an inode's fork. We don't do any higher level
6331	* invalidation work at all.
6332	*/
6333	int
6334	xfs_bunmapi_range(
6335	struct xfs_trans **tpp,
6336	struct xfs_inode *ip,
6337	uint32_t flags,
6338	xfs_fileoff_t startoff,
6339	xfs_fileoff_t endoff)
6340	{
6341	xfs_filblks_t unmap_len = endoff - startoff + 1;
6342	int error = 0;
6343
6344	xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
6345
6346	while (unmap_len > 0) {
6347	ASSERT((*tpp)->t_highest_agno == NULLAGNUMBER);
6348	error = __xfs_bunmapi(*tpp, ip, startoff, &unmap_len, flags,
6349	XFS_ITRUNC_MAX_EXTENTS);
6350	if (error)
6351	goto out;
6352
6353	/* free the just unmapped extents */
6354	error = xfs_defer_finish(tp: tpp);
6355	if (error)
6356	goto out;
6357	}
6358	out:
6359	return error;
6360	}
6361
6362	struct xfs_bmap_query_range {
6363	xfs_bmap_query_range_fn fn;
6364	void *priv;
6365	};
6366
6367	/* Format btree record and pass to our callback. */
6368	STATIC int
6369	xfs_bmap_query_range_helper(
6370	struct xfs_btree_cur *cur,
6371	const union xfs_btree_rec *rec,
6372	void *priv)
6373	{
6374	struct xfs_bmap_query_range *query = priv;
6375	struct xfs_bmbt_irec irec;
6376	xfs_failaddr_t fa;
6377
6378	xfs_bmbt_disk_get_all(r: &rec->bmbt, s: &irec);
6379	fa = xfs_bmap_validate_extent(cur->bc_ino.ip, cur->bc_ino.whichfork,
6380	&irec);
6381	if (fa) {
6382	xfs_btree_mark_sick(cur);
6383	return xfs_bmap_complain_bad_rec(cur->bc_ino.ip,
6384	cur->bc_ino.whichfork, fa, &irec);
6385	}
6386
6387	return query->fn(cur, &irec, query->priv);
6388	}
6389
6390	/* Find all bmaps. */
6391	int
6392	xfs_bmap_query_all(
6393	struct xfs_btree_cur *cur,
6394	xfs_bmap_query_range_fn fn,
6395	void *priv)
6396	{
6397	struct xfs_bmap_query_range query = {
6398	.priv = priv,
6399	.fn = fn,
6400	};
6401
6402	return xfs_btree_query_all(cur, fn: xfs_bmap_query_range_helper, priv: &query);
6403	}
6404