1 /*
2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_btree.h"
34 #include "xfs_ialloc.h"
35 #include "xfs_alloc.h"
36 #include "xfs_rtalloc.h"
37 #include "xfs_error.h"
38 #include "xfs_bmap.h"
39 
40 
41 /*
42  * Allocation group level functions.
43  */
44 static inline int
xfs_ialloc_cluster_alignment(xfs_alloc_arg_t * args)45 xfs_ialloc_cluster_alignment(
46 	xfs_alloc_arg_t	*args)
47 {
48 	if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
49 	    args->mp->m_sb.sb_inoalignmt >=
50 	     XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp)))
51 		return args->mp->m_sb.sb_inoalignmt;
52 	return 1;
53 }
54 
55 /*
56  * Lookup a record by ino in the btree given by cur.
57  */
58 int					/* error */
xfs_inobt_lookup(struct xfs_btree_cur * cur,xfs_agino_t ino,xfs_lookup_t dir,int * stat)59 xfs_inobt_lookup(
60 	struct xfs_btree_cur	*cur,	/* btree cursor */
61 	xfs_agino_t		ino,	/* starting inode of chunk */
62 	xfs_lookup_t		dir,	/* <=, >=, == */
63 	int			*stat)	/* success/failure */
64 {
65 	cur->bc_rec.i.ir_startino = ino;
66 	cur->bc_rec.i.ir_freecount = 0;
67 	cur->bc_rec.i.ir_free = 0;
68 	return xfs_btree_lookup(cur, dir, stat);
69 }
70 
71 /*
72  * Update the record referred to by cur to the value given.
73  * This either works (return 0) or gets an EFSCORRUPTED error.
74  */
75 STATIC int				/* error */
xfs_inobt_update(struct xfs_btree_cur * cur,xfs_inobt_rec_incore_t * irec)76 xfs_inobt_update(
77 	struct xfs_btree_cur	*cur,	/* btree cursor */
78 	xfs_inobt_rec_incore_t	*irec)	/* btree record */
79 {
80 	union xfs_btree_rec	rec;
81 
82 	rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
83 	rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
84 	rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
85 	return xfs_btree_update(cur, &rec);
86 }
87 
88 /*
89  * Get the data from the pointed-to record.
90  */
91 int					/* error */
xfs_inobt_get_rec(struct xfs_btree_cur * cur,xfs_inobt_rec_incore_t * irec,int * stat)92 xfs_inobt_get_rec(
93 	struct xfs_btree_cur	*cur,	/* btree cursor */
94 	xfs_inobt_rec_incore_t	*irec,	/* btree record */
95 	int			*stat)	/* output: success/failure */
96 {
97 	union xfs_btree_rec	*rec;
98 	int			error;
99 
100 	error = xfs_btree_get_rec(cur, &rec, stat);
101 	if (!error && *stat == 1) {
102 		irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
103 		irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
104 		irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
105 	}
106 	return error;
107 }
108 
109 /*
110  * Verify that the number of free inodes in the AGI is correct.
111  */
112 #ifdef DEBUG
113 STATIC int
xfs_check_agi_freecount(struct xfs_btree_cur * cur,struct xfs_agi * agi)114 xfs_check_agi_freecount(
115 	struct xfs_btree_cur	*cur,
116 	struct xfs_agi		*agi)
117 {
118 	if (cur->bc_nlevels == 1) {
119 		xfs_inobt_rec_incore_t rec;
120 		int		freecount = 0;
121 		int		error;
122 		int		i;
123 
124 		error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
125 		if (error)
126 			return error;
127 
128 		do {
129 			error = xfs_inobt_get_rec(cur, &rec, &i);
130 			if (error)
131 				return error;
132 
133 			if (i) {
134 				freecount += rec.ir_freecount;
135 				error = xfs_btree_increment(cur, 0, &i);
136 				if (error)
137 					return error;
138 			}
139 		} while (i == 1);
140 
141 		if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
142 			ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
143 	}
144 	return 0;
145 }
146 #else
147 #define xfs_check_agi_freecount(cur, agi)	0
148 #endif
149 
150 /*
151  * Initialise a new set of inodes.
152  */
153 STATIC int
xfs_ialloc_inode_init(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,xfs_agblock_t agbno,xfs_agblock_t length,unsigned int gen)154 xfs_ialloc_inode_init(
155 	struct xfs_mount	*mp,
156 	struct xfs_trans	*tp,
157 	xfs_agnumber_t		agno,
158 	xfs_agblock_t		agbno,
159 	xfs_agblock_t		length,
160 	unsigned int		gen)
161 {
162 	struct xfs_buf		*fbuf;
163 	struct xfs_dinode	*free;
164 	int			blks_per_cluster, nbufs, ninodes;
165 	int			version;
166 	int			i, j;
167 	xfs_daddr_t		d;
168 
169 	/*
170 	 * Loop over the new block(s), filling in the inodes.
171 	 * For small block sizes, manipulate the inodes in buffers
172 	 * which are multiples of the blocks size.
173 	 */
174 	if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
175 		blks_per_cluster = 1;
176 		nbufs = length;
177 		ninodes = mp->m_sb.sb_inopblock;
178 	} else {
179 		blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
180 				   mp->m_sb.sb_blocksize;
181 		nbufs = length / blks_per_cluster;
182 		ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
183 	}
184 
185 	/*
186 	 * Figure out what version number to use in the inodes we create.
187 	 * If the superblock version has caught up to the one that supports
188 	 * the new inode format, then use the new inode version.  Otherwise
189 	 * use the old version so that old kernels will continue to be
190 	 * able to use the file system.
191 	 */
192 	if (xfs_sb_version_hasnlink(&mp->m_sb))
193 		version = 2;
194 	else
195 		version = 1;
196 
197 	for (j = 0; j < nbufs; j++) {
198 		/*
199 		 * Get the block.
200 		 */
201 		d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
202 		fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
203 					 mp->m_bsize * blks_per_cluster,
204 					 XBF_LOCK);
205 		if (!fbuf)
206 			return ENOMEM;
207 		/*
208 		 * Initialize all inodes in this buffer and then log them.
209 		 *
210 		 * XXX: It would be much better if we had just one transaction
211 		 *	to log a whole cluster of inodes instead of all the
212 		 *	individual transactions causing a lot of log traffic.
213 		 */
214 		xfs_buf_zero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog);
215 		for (i = 0; i < ninodes; i++) {
216 			int	ioffset = i << mp->m_sb.sb_inodelog;
217 			uint	isize = sizeof(struct xfs_dinode);
218 
219 			free = xfs_make_iptr(mp, fbuf, i);
220 			free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
221 			free->di_version = version;
222 			free->di_gen = cpu_to_be32(gen);
223 			free->di_next_unlinked = cpu_to_be32(NULLAGINO);
224 			xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
225 		}
226 		xfs_trans_inode_alloc_buf(tp, fbuf);
227 	}
228 	return 0;
229 }
230 
231 /*
232  * Allocate new inodes in the allocation group specified by agbp.
233  * Return 0 for success, else error code.
234  */
235 STATIC int				/* error code or 0 */
xfs_ialloc_ag_alloc(xfs_trans_t * tp,xfs_buf_t * agbp,int * alloc)236 xfs_ialloc_ag_alloc(
237 	xfs_trans_t	*tp,		/* transaction pointer */
238 	xfs_buf_t	*agbp,		/* alloc group buffer */
239 	int		*alloc)
240 {
241 	xfs_agi_t	*agi;		/* allocation group header */
242 	xfs_alloc_arg_t	args;		/* allocation argument structure */
243 	xfs_btree_cur_t	*cur;		/* inode btree cursor */
244 	xfs_agnumber_t	agno;
245 	int		error;
246 	int		i;
247 	xfs_agino_t	newino;		/* new first inode's number */
248 	xfs_agino_t	newlen;		/* new number of inodes */
249 	xfs_agino_t	thisino;	/* current inode number, for loop */
250 	int		isaligned = 0;	/* inode allocation at stripe unit */
251 					/* boundary */
252 	struct xfs_perag *pag;
253 
254 	args.tp = tp;
255 	args.mp = tp->t_mountp;
256 
257 	/*
258 	 * Locking will ensure that we don't have two callers in here
259 	 * at one time.
260 	 */
261 	newlen = XFS_IALLOC_INODES(args.mp);
262 	if (args.mp->m_maxicount &&
263 	    args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
264 		return XFS_ERROR(ENOSPC);
265 	args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
266 	/*
267 	 * First try to allocate inodes contiguous with the last-allocated
268 	 * chunk of inodes.  If the filesystem is striped, this will fill
269 	 * an entire stripe unit with inodes.
270  	 */
271 	agi = XFS_BUF_TO_AGI(agbp);
272 	newino = be32_to_cpu(agi->agi_newino);
273 	agno = be32_to_cpu(agi->agi_seqno);
274 	args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
275 			XFS_IALLOC_BLOCKS(args.mp);
276 	if (likely(newino != NULLAGINO &&
277 		  (args.agbno < be32_to_cpu(agi->agi_length)))) {
278 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
279 		args.type = XFS_ALLOCTYPE_THIS_BNO;
280 		args.mod = args.total = args.wasdel = args.isfl =
281 			args.userdata = args.minalignslop = 0;
282 		args.prod = 1;
283 
284 		/*
285 		 * We need to take into account alignment here to ensure that
286 		 * we don't modify the free list if we fail to have an exact
287 		 * block. If we don't have an exact match, and every oher
288 		 * attempt allocation attempt fails, we'll end up cancelling
289 		 * a dirty transaction and shutting down.
290 		 *
291 		 * For an exact allocation, alignment must be 1,
292 		 * however we need to take cluster alignment into account when
293 		 * fixing up the freelist. Use the minalignslop field to
294 		 * indicate that extra blocks might be required for alignment,
295 		 * but not to use them in the actual exact allocation.
296 		 */
297 		args.alignment = 1;
298 		args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
299 
300 		/* Allow space for the inode btree to split. */
301 		args.minleft = args.mp->m_in_maxlevels - 1;
302 		if ((error = xfs_alloc_vextent(&args)))
303 			return error;
304 	} else
305 		args.fsbno = NULLFSBLOCK;
306 
307 	if (unlikely(args.fsbno == NULLFSBLOCK)) {
308 		/*
309 		 * Set the alignment for the allocation.
310 		 * If stripe alignment is turned on then align at stripe unit
311 		 * boundary.
312 		 * If the cluster size is smaller than a filesystem block
313 		 * then we're doing I/O for inodes in filesystem block size
314 		 * pieces, so don't need alignment anyway.
315 		 */
316 		isaligned = 0;
317 		if (args.mp->m_sinoalign) {
318 			ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
319 			args.alignment = args.mp->m_dalign;
320 			isaligned = 1;
321 		} else
322 			args.alignment = xfs_ialloc_cluster_alignment(&args);
323 		/*
324 		 * Need to figure out where to allocate the inode blocks.
325 		 * Ideally they should be spaced out through the a.g.
326 		 * For now, just allocate blocks up front.
327 		 */
328 		args.agbno = be32_to_cpu(agi->agi_root);
329 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
330 		/*
331 		 * Allocate a fixed-size extent of inodes.
332 		 */
333 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
334 		args.mod = args.total = args.wasdel = args.isfl =
335 			args.userdata = args.minalignslop = 0;
336 		args.prod = 1;
337 		/*
338 		 * Allow space for the inode btree to split.
339 		 */
340 		args.minleft = args.mp->m_in_maxlevels - 1;
341 		if ((error = xfs_alloc_vextent(&args)))
342 			return error;
343 	}
344 
345 	/*
346 	 * If stripe alignment is turned on, then try again with cluster
347 	 * alignment.
348 	 */
349 	if (isaligned && args.fsbno == NULLFSBLOCK) {
350 		args.type = XFS_ALLOCTYPE_NEAR_BNO;
351 		args.agbno = be32_to_cpu(agi->agi_root);
352 		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
353 		args.alignment = xfs_ialloc_cluster_alignment(&args);
354 		if ((error = xfs_alloc_vextent(&args)))
355 			return error;
356 	}
357 
358 	if (args.fsbno == NULLFSBLOCK) {
359 		*alloc = 0;
360 		return 0;
361 	}
362 	ASSERT(args.len == args.minlen);
363 
364 	/*
365 	 * Stamp and write the inode buffers.
366 	 *
367 	 * Seed the new inode cluster with a random generation number. This
368 	 * prevents short-term reuse of generation numbers if a chunk is
369 	 * freed and then immediately reallocated. We use random numbers
370 	 * rather than a linear progression to prevent the next generation
371 	 * number from being easily guessable.
372 	 */
373 	error = xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno,
374 			args.len, random32());
375 
376 	if (error)
377 		return error;
378 	/*
379 	 * Convert the results.
380 	 */
381 	newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
382 	be32_add_cpu(&agi->agi_count, newlen);
383 	be32_add_cpu(&agi->agi_freecount, newlen);
384 	pag = xfs_perag_get(args.mp, agno);
385 	pag->pagi_freecount += newlen;
386 	xfs_perag_put(pag);
387 	agi->agi_newino = cpu_to_be32(newino);
388 
389 	/*
390 	 * Insert records describing the new inode chunk into the btree.
391 	 */
392 	cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
393 	for (thisino = newino;
394 	     thisino < newino + newlen;
395 	     thisino += XFS_INODES_PER_CHUNK) {
396 		cur->bc_rec.i.ir_startino = thisino;
397 		cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
398 		cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
399 		error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
400 		if (error) {
401 			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
402 			return error;
403 		}
404 		ASSERT(i == 0);
405 		error = xfs_btree_insert(cur, &i);
406 		if (error) {
407 			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
408 			return error;
409 		}
410 		ASSERT(i == 1);
411 	}
412 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
413 	/*
414 	 * Log allocation group header fields
415 	 */
416 	xfs_ialloc_log_agi(tp, agbp,
417 		XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
418 	/*
419 	 * Modify/log superblock values for inode count and inode free count.
420 	 */
421 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
422 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
423 	*alloc = 1;
424 	return 0;
425 }
426 
427 STATIC xfs_agnumber_t
xfs_ialloc_next_ag(xfs_mount_t * mp)428 xfs_ialloc_next_ag(
429 	xfs_mount_t	*mp)
430 {
431 	xfs_agnumber_t	agno;
432 
433 	spin_lock(&mp->m_agirotor_lock);
434 	agno = mp->m_agirotor;
435 	if (++mp->m_agirotor == mp->m_maxagi)
436 		mp->m_agirotor = 0;
437 	spin_unlock(&mp->m_agirotor_lock);
438 
439 	return agno;
440 }
441 
442 /*
443  * Select an allocation group to look for a free inode in, based on the parent
444  * inode and then mode.  Return the allocation group buffer.
445  */
446 STATIC xfs_buf_t *			/* allocation group buffer */
xfs_ialloc_ag_select(xfs_trans_t * tp,xfs_ino_t parent,umode_t mode,int okalloc)447 xfs_ialloc_ag_select(
448 	xfs_trans_t	*tp,		/* transaction pointer */
449 	xfs_ino_t	parent,		/* parent directory inode number */
450 	umode_t		mode,		/* bits set to indicate file type */
451 	int		okalloc)	/* ok to allocate more space */
452 {
453 	xfs_buf_t	*agbp;		/* allocation group header buffer */
454 	xfs_agnumber_t	agcount;	/* number of ag's in the filesystem */
455 	xfs_agnumber_t	agno;		/* current ag number */
456 	int		flags;		/* alloc buffer locking flags */
457 	xfs_extlen_t	ineed;		/* blocks needed for inode allocation */
458 	xfs_extlen_t	longest = 0;	/* longest extent available */
459 	xfs_mount_t	*mp;		/* mount point structure */
460 	int		needspace;	/* file mode implies space allocated */
461 	xfs_perag_t	*pag;		/* per allocation group data */
462 	xfs_agnumber_t	pagno;		/* parent (starting) ag number */
463 
464 	/*
465 	 * Files of these types need at least one block if length > 0
466 	 * (and they won't fit in the inode, but that's hard to figure out).
467 	 */
468 	needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
469 	mp = tp->t_mountp;
470 	agcount = mp->m_maxagi;
471 	if (S_ISDIR(mode))
472 		pagno = xfs_ialloc_next_ag(mp);
473 	else {
474 		pagno = XFS_INO_TO_AGNO(mp, parent);
475 		if (pagno >= agcount)
476 			pagno = 0;
477 	}
478 	ASSERT(pagno < agcount);
479 	/*
480 	 * Loop through allocation groups, looking for one with a little
481 	 * free space in it.  Note we don't look for free inodes, exactly.
482 	 * Instead, we include whether there is a need to allocate inodes
483 	 * to mean that blocks must be allocated for them,
484 	 * if none are currently free.
485 	 */
486 	agno = pagno;
487 	flags = XFS_ALLOC_FLAG_TRYLOCK;
488 	for (;;) {
489 		pag = xfs_perag_get(mp, agno);
490 		if (!pag->pagi_init) {
491 			if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
492 				agbp = NULL;
493 				goto nextag;
494 			}
495 		} else
496 			agbp = NULL;
497 
498 		if (!pag->pagi_inodeok) {
499 			xfs_ialloc_next_ag(mp);
500 			goto unlock_nextag;
501 		}
502 
503 		/*
504 		 * Is there enough free space for the file plus a block
505 		 * of inodes (if we need to allocate some)?
506 		 */
507 		ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
508 		if (ineed && !pag->pagf_init) {
509 			if (agbp == NULL &&
510 			    xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
511 				agbp = NULL;
512 				goto nextag;
513 			}
514 			(void)xfs_alloc_pagf_init(mp, tp, agno, flags);
515 		}
516 		if (!ineed || pag->pagf_init) {
517 			if (ineed && !(longest = pag->pagf_longest))
518 				longest = pag->pagf_flcount > 0;
519 			if (!ineed ||
520 			    (pag->pagf_freeblks >= needspace + ineed &&
521 			     longest >= ineed &&
522 			     okalloc)) {
523 				if (agbp == NULL &&
524 				    xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
525 					agbp = NULL;
526 					goto nextag;
527 				}
528 				xfs_perag_put(pag);
529 				return agbp;
530 			}
531 		}
532 unlock_nextag:
533 		if (agbp)
534 			xfs_trans_brelse(tp, agbp);
535 nextag:
536 		xfs_perag_put(pag);
537 		/*
538 		 * No point in iterating over the rest, if we're shutting
539 		 * down.
540 		 */
541 		if (XFS_FORCED_SHUTDOWN(mp))
542 			return NULL;
543 		agno++;
544 		if (agno >= agcount)
545 			agno = 0;
546 		if (agno == pagno) {
547 			if (flags == 0)
548 				return NULL;
549 			flags = 0;
550 		}
551 	}
552 }
553 
554 /*
555  * Try to retrieve the next record to the left/right from the current one.
556  */
557 STATIC int
xfs_ialloc_next_rec(struct xfs_btree_cur * cur,xfs_inobt_rec_incore_t * rec,int * done,int left)558 xfs_ialloc_next_rec(
559 	struct xfs_btree_cur	*cur,
560 	xfs_inobt_rec_incore_t	*rec,
561 	int			*done,
562 	int			left)
563 {
564 	int                     error;
565 	int			i;
566 
567 	if (left)
568 		error = xfs_btree_decrement(cur, 0, &i);
569 	else
570 		error = xfs_btree_increment(cur, 0, &i);
571 
572 	if (error)
573 		return error;
574 	*done = !i;
575 	if (i) {
576 		error = xfs_inobt_get_rec(cur, rec, &i);
577 		if (error)
578 			return error;
579 		XFS_WANT_CORRUPTED_RETURN(i == 1);
580 	}
581 
582 	return 0;
583 }
584 
585 STATIC int
xfs_ialloc_get_rec(struct xfs_btree_cur * cur,xfs_agino_t agino,xfs_inobt_rec_incore_t * rec,int * done,int left)586 xfs_ialloc_get_rec(
587 	struct xfs_btree_cur	*cur,
588 	xfs_agino_t		agino,
589 	xfs_inobt_rec_incore_t	*rec,
590 	int			*done,
591 	int			left)
592 {
593 	int                     error;
594 	int			i;
595 
596 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
597 	if (error)
598 		return error;
599 	*done = !i;
600 	if (i) {
601 		error = xfs_inobt_get_rec(cur, rec, &i);
602 		if (error)
603 			return error;
604 		XFS_WANT_CORRUPTED_RETURN(i == 1);
605 	}
606 
607 	return 0;
608 }
609 
610 /*
611  * Visible inode allocation functions.
612  */
613 
614 /*
615  * Allocate an inode on disk.
616  * Mode is used to tell whether the new inode will need space, and whether
617  * it is a directory.
618  *
619  * The arguments IO_agbp and alloc_done are defined to work within
620  * the constraint of one allocation per transaction.
621  * xfs_dialloc() is designed to be called twice if it has to do an
622  * allocation to make more free inodes.  On the first call,
623  * IO_agbp should be set to NULL. If an inode is available,
624  * i.e., xfs_dialloc() did not need to do an allocation, an inode
625  * number is returned.  In this case, IO_agbp would be set to the
626  * current ag_buf and alloc_done set to false.
627  * If an allocation needed to be done, xfs_dialloc would return
628  * the current ag_buf in IO_agbp and set alloc_done to true.
629  * The caller should then commit the current transaction, allocate a new
630  * transaction, and call xfs_dialloc() again, passing in the previous
631  * value of IO_agbp.  IO_agbp should be held across the transactions.
632  * Since the agbp is locked across the two calls, the second call is
633  * guaranteed to have a free inode available.
634  *
635  * Once we successfully pick an inode its number is returned and the
636  * on-disk data structures are updated.  The inode itself is not read
637  * in, since doing so would break ordering constraints with xfs_reclaim.
638  */
639 int
xfs_dialloc(xfs_trans_t * tp,xfs_ino_t parent,umode_t mode,int okalloc,xfs_buf_t ** IO_agbp,boolean_t * alloc_done,xfs_ino_t * inop)640 xfs_dialloc(
641 	xfs_trans_t	*tp,		/* transaction pointer */
642 	xfs_ino_t	parent,		/* parent inode (directory) */
643 	umode_t		mode,		/* mode bits for new inode */
644 	int		okalloc,	/* ok to allocate more space */
645 	xfs_buf_t	**IO_agbp,	/* in/out ag header's buffer */
646 	boolean_t	*alloc_done,	/* true if we needed to replenish
647 					   inode freelist */
648 	xfs_ino_t	*inop)		/* inode number allocated */
649 {
650 	xfs_agnumber_t	agcount;	/* number of allocation groups */
651 	xfs_buf_t	*agbp;		/* allocation group header's buffer */
652 	xfs_agnumber_t	agno;		/* allocation group number */
653 	xfs_agi_t	*agi;		/* allocation group header structure */
654 	xfs_btree_cur_t	*cur;		/* inode allocation btree cursor */
655 	int		error;		/* error return value */
656 	int		i;		/* result code */
657 	int		ialloced;	/* inode allocation status */
658 	int		noroom = 0;	/* no space for inode blk allocation */
659 	xfs_ino_t	ino;		/* fs-relative inode to be returned */
660 	/* REFERENCED */
661 	int		j;		/* result code */
662 	xfs_mount_t	*mp;		/* file system mount structure */
663 	int		offset;		/* index of inode in chunk */
664 	xfs_agino_t	pagino;		/* parent's AG relative inode # */
665 	xfs_agnumber_t	pagno;		/* parent's AG number */
666 	xfs_inobt_rec_incore_t rec;	/* inode allocation record */
667 	xfs_agnumber_t	tagno;		/* testing allocation group number */
668 	xfs_btree_cur_t	*tcur;		/* temp cursor */
669 	xfs_inobt_rec_incore_t trec;	/* temp inode allocation record */
670 	struct xfs_perag *pag;
671 
672 
673 	if (*IO_agbp == NULL) {
674 		/*
675 		 * We do not have an agbp, so select an initial allocation
676 		 * group for inode allocation.
677 		 */
678 		agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
679 		/*
680 		 * Couldn't find an allocation group satisfying the
681 		 * criteria, give up.
682 		 */
683 		if (!agbp) {
684 			*inop = NULLFSINO;
685 			return 0;
686 		}
687 		agi = XFS_BUF_TO_AGI(agbp);
688 		ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
689 	} else {
690 		/*
691 		 * Continue where we left off before.  In this case, we
692 		 * know that the allocation group has free inodes.
693 		 */
694 		agbp = *IO_agbp;
695 		agi = XFS_BUF_TO_AGI(agbp);
696 		ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
697 		ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
698 	}
699 	mp = tp->t_mountp;
700 	agcount = mp->m_sb.sb_agcount;
701 	agno = be32_to_cpu(agi->agi_seqno);
702 	tagno = agno;
703 	pagno = XFS_INO_TO_AGNO(mp, parent);
704 	pagino = XFS_INO_TO_AGINO(mp, parent);
705 
706 	/*
707 	 * If we have already hit the ceiling of inode blocks then clear
708 	 * okalloc so we scan all available agi structures for a free
709 	 * inode.
710 	 */
711 
712 	if (mp->m_maxicount &&
713 	    mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
714 		noroom = 1;
715 		okalloc = 0;
716 	}
717 
718 	/*
719 	 * Loop until we find an allocation group that either has free inodes
720 	 * or in which we can allocate some inodes.  Iterate through the
721 	 * allocation groups upward, wrapping at the end.
722 	 */
723 	*alloc_done = B_FALSE;
724 	while (!agi->agi_freecount) {
725 		/*
726 		 * Don't do anything if we're not supposed to allocate
727 		 * any blocks, just go on to the next ag.
728 		 */
729 		if (okalloc) {
730 			/*
731 			 * Try to allocate some new inodes in the allocation
732 			 * group.
733 			 */
734 			if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
735 				xfs_trans_brelse(tp, agbp);
736 				if (error == ENOSPC) {
737 					*inop = NULLFSINO;
738 					return 0;
739 				} else
740 					return error;
741 			}
742 			if (ialloced) {
743 				/*
744 				 * We successfully allocated some inodes, return
745 				 * the current context to the caller so that it
746 				 * can commit the current transaction and call
747 				 * us again where we left off.
748 				 */
749 				ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
750 				*alloc_done = B_TRUE;
751 				*IO_agbp = agbp;
752 				*inop = NULLFSINO;
753 				return 0;
754 			}
755 		}
756 		/*
757 		 * If it failed, give up on this ag.
758 		 */
759 		xfs_trans_brelse(tp, agbp);
760 		/*
761 		 * Go on to the next ag: get its ag header.
762 		 */
763 nextag:
764 		if (++tagno == agcount)
765 			tagno = 0;
766 		if (tagno == agno) {
767 			*inop = NULLFSINO;
768 			return noroom ? ENOSPC : 0;
769 		}
770 		pag = xfs_perag_get(mp, tagno);
771 		if (pag->pagi_inodeok == 0) {
772 			xfs_perag_put(pag);
773 			goto nextag;
774 		}
775 		error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
776 		xfs_perag_put(pag);
777 		if (error)
778 			goto nextag;
779 		agi = XFS_BUF_TO_AGI(agbp);
780 		ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
781 	}
782 	/*
783 	 * Here with an allocation group that has a free inode.
784 	 * Reset agno since we may have chosen a new ag in the
785 	 * loop above.
786 	 */
787 	agno = tagno;
788 	*IO_agbp = NULL;
789 	pag = xfs_perag_get(mp, agno);
790 
791  restart_pagno:
792 	cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
793 	/*
794 	 * If pagino is 0 (this is the root inode allocation) use newino.
795 	 * This must work because we've just allocated some.
796 	 */
797 	if (!pagino)
798 		pagino = be32_to_cpu(agi->agi_newino);
799 
800 	error = xfs_check_agi_freecount(cur, agi);
801 	if (error)
802 		goto error0;
803 
804 	/*
805 	 * If in the same AG as the parent, try to get near the parent.
806 	 */
807 	if (pagno == agno) {
808 		int		doneleft;	/* done, to the left */
809 		int		doneright;	/* done, to the right */
810 		int		searchdistance = 10;
811 
812 		error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
813 		if (error)
814 			goto error0;
815 		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
816 
817 		error = xfs_inobt_get_rec(cur, &rec, &j);
818 		if (error)
819 			goto error0;
820 		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
821 
822 		if (rec.ir_freecount > 0) {
823 			/*
824 			 * Found a free inode in the same chunk
825 			 * as the parent, done.
826 			 */
827 			goto alloc_inode;
828 		}
829 
830 
831 		/*
832 		 * In the same AG as parent, but parent's chunk is full.
833 		 */
834 
835 		/* duplicate the cursor, search left & right simultaneously */
836 		error = xfs_btree_dup_cursor(cur, &tcur);
837 		if (error)
838 			goto error0;
839 
840 		/*
841 		 * Skip to last blocks looked up if same parent inode.
842 		 */
843 		if (pagino != NULLAGINO &&
844 		    pag->pagl_pagino == pagino &&
845 		    pag->pagl_leftrec != NULLAGINO &&
846 		    pag->pagl_rightrec != NULLAGINO) {
847 			error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
848 						   &trec, &doneleft, 1);
849 			if (error)
850 				goto error1;
851 
852 			error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
853 						   &rec, &doneright, 0);
854 			if (error)
855 				goto error1;
856 		} else {
857 			/* search left with tcur, back up 1 record */
858 			error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
859 			if (error)
860 				goto error1;
861 
862 			/* search right with cur, go forward 1 record. */
863 			error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
864 			if (error)
865 				goto error1;
866 		}
867 
868 		/*
869 		 * Loop until we find an inode chunk with a free inode.
870 		 */
871 		while (!doneleft || !doneright) {
872 			int	useleft;  /* using left inode chunk this time */
873 
874 			if (!--searchdistance) {
875 				/*
876 				 * Not in range - save last search
877 				 * location and allocate a new inode
878 				 */
879 				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
880 				pag->pagl_leftrec = trec.ir_startino;
881 				pag->pagl_rightrec = rec.ir_startino;
882 				pag->pagl_pagino = pagino;
883 				goto newino;
884 			}
885 
886 			/* figure out the closer block if both are valid. */
887 			if (!doneleft && !doneright) {
888 				useleft = pagino -
889 				 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
890 				  rec.ir_startino - pagino;
891 			} else {
892 				useleft = !doneleft;
893 			}
894 
895 			/* free inodes to the left? */
896 			if (useleft && trec.ir_freecount) {
897 				rec = trec;
898 				xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
899 				cur = tcur;
900 
901 				pag->pagl_leftrec = trec.ir_startino;
902 				pag->pagl_rightrec = rec.ir_startino;
903 				pag->pagl_pagino = pagino;
904 				goto alloc_inode;
905 			}
906 
907 			/* free inodes to the right? */
908 			if (!useleft && rec.ir_freecount) {
909 				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
910 
911 				pag->pagl_leftrec = trec.ir_startino;
912 				pag->pagl_rightrec = rec.ir_startino;
913 				pag->pagl_pagino = pagino;
914 				goto alloc_inode;
915 			}
916 
917 			/* get next record to check */
918 			if (useleft) {
919 				error = xfs_ialloc_next_rec(tcur, &trec,
920 								 &doneleft, 1);
921 			} else {
922 				error = xfs_ialloc_next_rec(cur, &rec,
923 								 &doneright, 0);
924 			}
925 			if (error)
926 				goto error1;
927 		}
928 
929 		/*
930 		 * We've reached the end of the btree. because
931 		 * we are only searching a small chunk of the
932 		 * btree each search, there is obviously free
933 		 * inodes closer to the parent inode than we
934 		 * are now. restart the search again.
935 		 */
936 		pag->pagl_pagino = NULLAGINO;
937 		pag->pagl_leftrec = NULLAGINO;
938 		pag->pagl_rightrec = NULLAGINO;
939 		xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
940 		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
941 		goto restart_pagno;
942 	}
943 
944 	/*
945 	 * In a different AG from the parent.
946 	 * See if the most recently allocated block has any free.
947 	 */
948 newino:
949 	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
950 		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
951 					 XFS_LOOKUP_EQ, &i);
952 		if (error)
953 			goto error0;
954 
955 		if (i == 1) {
956 			error = xfs_inobt_get_rec(cur, &rec, &j);
957 			if (error)
958 				goto error0;
959 
960 			if (j == 1 && rec.ir_freecount > 0) {
961 				/*
962 				 * The last chunk allocated in the group
963 				 * still has a free inode.
964 				 */
965 				goto alloc_inode;
966 			}
967 		}
968 	}
969 
970 	/*
971 	 * None left in the last group, search the whole AG
972 	 */
973 	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
974 	if (error)
975 		goto error0;
976 	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
977 
978 	for (;;) {
979 		error = xfs_inobt_get_rec(cur, &rec, &i);
980 		if (error)
981 			goto error0;
982 		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
983 		if (rec.ir_freecount > 0)
984 			break;
985 		error = xfs_btree_increment(cur, 0, &i);
986 		if (error)
987 			goto error0;
988 		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
989 	}
990 
991 alloc_inode:
992 	offset = xfs_ialloc_find_free(&rec.ir_free);
993 	ASSERT(offset >= 0);
994 	ASSERT(offset < XFS_INODES_PER_CHUNK);
995 	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
996 				   XFS_INODES_PER_CHUNK) == 0);
997 	ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
998 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
999 	rec.ir_freecount--;
1000 	error = xfs_inobt_update(cur, &rec);
1001 	if (error)
1002 		goto error0;
1003 	be32_add_cpu(&agi->agi_freecount, -1);
1004 	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1005 	pag->pagi_freecount--;
1006 
1007 	error = xfs_check_agi_freecount(cur, agi);
1008 	if (error)
1009 		goto error0;
1010 
1011 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1012 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1013 	xfs_perag_put(pag);
1014 	*inop = ino;
1015 	return 0;
1016 error1:
1017 	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1018 error0:
1019 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1020 	xfs_perag_put(pag);
1021 	return error;
1022 }
1023 
1024 /*
1025  * Free disk inode.  Carefully avoids touching the incore inode, all
1026  * manipulations incore are the caller's responsibility.
1027  * The on-disk inode is not changed by this operation, only the
1028  * btree (free inode mask) is changed.
1029  */
1030 int
xfs_difree(xfs_trans_t * tp,xfs_ino_t inode,xfs_bmap_free_t * flist,int * delete,xfs_ino_t * first_ino)1031 xfs_difree(
1032 	xfs_trans_t	*tp,		/* transaction pointer */
1033 	xfs_ino_t	inode,		/* inode to be freed */
1034 	xfs_bmap_free_t	*flist,		/* extents to free */
1035 	int		*delete,	/* set if inode cluster was deleted */
1036 	xfs_ino_t	*first_ino)	/* first inode in deleted cluster */
1037 {
1038 	/* REFERENCED */
1039 	xfs_agblock_t	agbno;	/* block number containing inode */
1040 	xfs_buf_t	*agbp;	/* buffer containing allocation group header */
1041 	xfs_agino_t	agino;	/* inode number relative to allocation group */
1042 	xfs_agnumber_t	agno;	/* allocation group number */
1043 	xfs_agi_t	*agi;	/* allocation group header */
1044 	xfs_btree_cur_t	*cur;	/* inode btree cursor */
1045 	int		error;	/* error return value */
1046 	int		i;	/* result code */
1047 	int		ilen;	/* inodes in an inode cluster */
1048 	xfs_mount_t	*mp;	/* mount structure for filesystem */
1049 	int		off;	/* offset of inode in inode chunk */
1050 	xfs_inobt_rec_incore_t rec;	/* btree record */
1051 	struct xfs_perag *pag;
1052 
1053 	mp = tp->t_mountp;
1054 
1055 	/*
1056 	 * Break up inode number into its components.
1057 	 */
1058 	agno = XFS_INO_TO_AGNO(mp, inode);
1059 	if (agno >= mp->m_sb.sb_agcount)  {
1060 		xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1061 			__func__, agno, mp->m_sb.sb_agcount);
1062 		ASSERT(0);
1063 		return XFS_ERROR(EINVAL);
1064 	}
1065 	agino = XFS_INO_TO_AGINO(mp, inode);
1066 	if (inode != XFS_AGINO_TO_INO(mp, agno, agino))  {
1067 		xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1068 			__func__, (unsigned long long)inode,
1069 			(unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1070 		ASSERT(0);
1071 		return XFS_ERROR(EINVAL);
1072 	}
1073 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1074 	if (agbno >= mp->m_sb.sb_agblocks)  {
1075 		xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1076 			__func__, agbno, mp->m_sb.sb_agblocks);
1077 		ASSERT(0);
1078 		return XFS_ERROR(EINVAL);
1079 	}
1080 	/*
1081 	 * Get the allocation group header.
1082 	 */
1083 	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1084 	if (error) {
1085 		xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1086 			__func__, error);
1087 		return error;
1088 	}
1089 	agi = XFS_BUF_TO_AGI(agbp);
1090 	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1091 	ASSERT(agbno < be32_to_cpu(agi->agi_length));
1092 	/*
1093 	 * Initialize the cursor.
1094 	 */
1095 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1096 
1097 	error = xfs_check_agi_freecount(cur, agi);
1098 	if (error)
1099 		goto error0;
1100 
1101 	/*
1102 	 * Look for the entry describing this inode.
1103 	 */
1104 	if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1105 		xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1106 			__func__, error);
1107 		goto error0;
1108 	}
1109 	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1110 	error = xfs_inobt_get_rec(cur, &rec, &i);
1111 	if (error) {
1112 		xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1113 			__func__, error);
1114 		goto error0;
1115 	}
1116 	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1117 	/*
1118 	 * Get the offset in the inode chunk.
1119 	 */
1120 	off = agino - rec.ir_startino;
1121 	ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1122 	ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1123 	/*
1124 	 * Mark the inode free & increment the count.
1125 	 */
1126 	rec.ir_free |= XFS_INOBT_MASK(off);
1127 	rec.ir_freecount++;
1128 
1129 	/*
1130 	 * When an inode cluster is free, it becomes eligible for removal
1131 	 */
1132 	if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1133 	    (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1134 
1135 		*delete = 1;
1136 		*first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1137 
1138 		/*
1139 		 * Remove the inode cluster from the AGI B+Tree, adjust the
1140 		 * AGI and Superblock inode counts, and mark the disk space
1141 		 * to be freed when the transaction is committed.
1142 		 */
1143 		ilen = XFS_IALLOC_INODES(mp);
1144 		be32_add_cpu(&agi->agi_count, -ilen);
1145 		be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1146 		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1147 		pag = xfs_perag_get(mp, agno);
1148 		pag->pagi_freecount -= ilen - 1;
1149 		xfs_perag_put(pag);
1150 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1151 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1152 
1153 		if ((error = xfs_btree_delete(cur, &i))) {
1154 			xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1155 				__func__, error);
1156 			goto error0;
1157 		}
1158 
1159 		xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1160 				agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1161 				XFS_IALLOC_BLOCKS(mp), flist, mp);
1162 	} else {
1163 		*delete = 0;
1164 
1165 		error = xfs_inobt_update(cur, &rec);
1166 		if (error) {
1167 			xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1168 				__func__, error);
1169 			goto error0;
1170 		}
1171 
1172 		/*
1173 		 * Change the inode free counts and log the ag/sb changes.
1174 		 */
1175 		be32_add_cpu(&agi->agi_freecount, 1);
1176 		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1177 		pag = xfs_perag_get(mp, agno);
1178 		pag->pagi_freecount++;
1179 		xfs_perag_put(pag);
1180 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1181 	}
1182 
1183 	error = xfs_check_agi_freecount(cur, agi);
1184 	if (error)
1185 		goto error0;
1186 
1187 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1188 	return 0;
1189 
1190 error0:
1191 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1192 	return error;
1193 }
1194 
1195 STATIC int
xfs_imap_lookup(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,xfs_agino_t agino,xfs_agblock_t agbno,xfs_agblock_t * chunk_agbno,xfs_agblock_t * offset_agbno,int flags)1196 xfs_imap_lookup(
1197 	struct xfs_mount	*mp,
1198 	struct xfs_trans	*tp,
1199 	xfs_agnumber_t		agno,
1200 	xfs_agino_t		agino,
1201 	xfs_agblock_t		agbno,
1202 	xfs_agblock_t		*chunk_agbno,
1203 	xfs_agblock_t		*offset_agbno,
1204 	int			flags)
1205 {
1206 	struct xfs_inobt_rec_incore rec;
1207 	struct xfs_btree_cur	*cur;
1208 	struct xfs_buf		*agbp;
1209 	int			error;
1210 	int			i;
1211 
1212 	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1213 	if (error) {
1214 		xfs_alert(mp,
1215 			"%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1216 			__func__, error, agno);
1217 		return error;
1218 	}
1219 
1220 	/*
1221 	 * Lookup the inode record for the given agino. If the record cannot be
1222 	 * found, then it's an invalid inode number and we should abort. Once
1223 	 * we have a record, we need to ensure it contains the inode number
1224 	 * we are looking up.
1225 	 */
1226 	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1227 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1228 	if (!error) {
1229 		if (i)
1230 			error = xfs_inobt_get_rec(cur, &rec, &i);
1231 		if (!error && i == 0)
1232 			error = EINVAL;
1233 	}
1234 
1235 	xfs_trans_brelse(tp, agbp);
1236 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1237 	if (error)
1238 		return error;
1239 
1240 	/* check that the returned record contains the required inode */
1241 	if (rec.ir_startino > agino ||
1242 	    rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino)
1243 		return EINVAL;
1244 
1245 	/* for untrusted inodes check it is allocated first */
1246 	if ((flags & XFS_IGET_UNTRUSTED) &&
1247 	    (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1248 		return EINVAL;
1249 
1250 	*chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1251 	*offset_agbno = agbno - *chunk_agbno;
1252 	return 0;
1253 }
1254 
1255 /*
1256  * Return the location of the inode in imap, for mapping it into a buffer.
1257  */
1258 int
xfs_imap(xfs_mount_t * mp,xfs_trans_t * tp,xfs_ino_t ino,struct xfs_imap * imap,uint flags)1259 xfs_imap(
1260 	xfs_mount_t	 *mp,	/* file system mount structure */
1261 	xfs_trans_t	 *tp,	/* transaction pointer */
1262 	xfs_ino_t	ino,	/* inode to locate */
1263 	struct xfs_imap	*imap,	/* location map structure */
1264 	uint		flags)	/* flags for inode btree lookup */
1265 {
1266 	xfs_agblock_t	agbno;	/* block number of inode in the alloc group */
1267 	xfs_agino_t	agino;	/* inode number within alloc group */
1268 	xfs_agnumber_t	agno;	/* allocation group number */
1269 	int		blks_per_cluster; /* num blocks per inode cluster */
1270 	xfs_agblock_t	chunk_agbno;	/* first block in inode chunk */
1271 	xfs_agblock_t	cluster_agbno;	/* first block in inode cluster */
1272 	int		error;	/* error code */
1273 	int		offset;	/* index of inode in its buffer */
1274 	int		offset_agbno;	/* blks from chunk start to inode */
1275 
1276 	ASSERT(ino != NULLFSINO);
1277 
1278 	/*
1279 	 * Split up the inode number into its parts.
1280 	 */
1281 	agno = XFS_INO_TO_AGNO(mp, ino);
1282 	agino = XFS_INO_TO_AGINO(mp, ino);
1283 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1284 	if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1285 	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1286 #ifdef DEBUG
1287 		/*
1288 		 * Don't output diagnostic information for untrusted inodes
1289 		 * as they can be invalid without implying corruption.
1290 		 */
1291 		if (flags & XFS_IGET_UNTRUSTED)
1292 			return XFS_ERROR(EINVAL);
1293 		if (agno >= mp->m_sb.sb_agcount) {
1294 			xfs_alert(mp,
1295 				"%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1296 				__func__, agno, mp->m_sb.sb_agcount);
1297 		}
1298 		if (agbno >= mp->m_sb.sb_agblocks) {
1299 			xfs_alert(mp,
1300 		"%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1301 				__func__, (unsigned long long)agbno,
1302 				(unsigned long)mp->m_sb.sb_agblocks);
1303 		}
1304 		if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1305 			xfs_alert(mp,
1306 		"%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1307 				__func__, ino,
1308 				XFS_AGINO_TO_INO(mp, agno, agino));
1309 		}
1310 		xfs_stack_trace();
1311 #endif /* DEBUG */
1312 		return XFS_ERROR(EINVAL);
1313 	}
1314 
1315 	blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1316 
1317 	/*
1318 	 * For bulkstat and handle lookups, we have an untrusted inode number
1319 	 * that we have to verify is valid. We cannot do this just by reading
1320 	 * the inode buffer as it may have been unlinked and removed leaving
1321 	 * inodes in stale state on disk. Hence we have to do a btree lookup
1322 	 * in all cases where an untrusted inode number is passed.
1323 	 */
1324 	if (flags & XFS_IGET_UNTRUSTED) {
1325 		error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1326 					&chunk_agbno, &offset_agbno, flags);
1327 		if (error)
1328 			return error;
1329 		goto out_map;
1330 	}
1331 
1332 	/*
1333 	 * If the inode cluster size is the same as the blocksize or
1334 	 * smaller we get to the buffer by simple arithmetics.
1335 	 */
1336 	if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
1337 		offset = XFS_INO_TO_OFFSET(mp, ino);
1338 		ASSERT(offset < mp->m_sb.sb_inopblock);
1339 
1340 		imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1341 		imap->im_len = XFS_FSB_TO_BB(mp, 1);
1342 		imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1343 		return 0;
1344 	}
1345 
1346 	/*
1347 	 * If the inode chunks are aligned then use simple maths to
1348 	 * find the location. Otherwise we have to do a btree
1349 	 * lookup to find the location.
1350 	 */
1351 	if (mp->m_inoalign_mask) {
1352 		offset_agbno = agbno & mp->m_inoalign_mask;
1353 		chunk_agbno = agbno - offset_agbno;
1354 	} else {
1355 		error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1356 					&chunk_agbno, &offset_agbno, flags);
1357 		if (error)
1358 			return error;
1359 	}
1360 
1361 out_map:
1362 	ASSERT(agbno >= chunk_agbno);
1363 	cluster_agbno = chunk_agbno +
1364 		((offset_agbno / blks_per_cluster) * blks_per_cluster);
1365 	offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1366 		XFS_INO_TO_OFFSET(mp, ino);
1367 
1368 	imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1369 	imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1370 	imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1371 
1372 	/*
1373 	 * If the inode number maps to a block outside the bounds
1374 	 * of the file system then return NULL rather than calling
1375 	 * read_buf and panicing when we get an error from the
1376 	 * driver.
1377 	 */
1378 	if ((imap->im_blkno + imap->im_len) >
1379 	    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1380 		xfs_alert(mp,
1381 	"%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1382 			__func__, (unsigned long long) imap->im_blkno,
1383 			(unsigned long long) imap->im_len,
1384 			XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1385 		return XFS_ERROR(EINVAL);
1386 	}
1387 	return 0;
1388 }
1389 
1390 /*
1391  * Compute and fill in value of m_in_maxlevels.
1392  */
1393 void
xfs_ialloc_compute_maxlevels(xfs_mount_t * mp)1394 xfs_ialloc_compute_maxlevels(
1395 	xfs_mount_t	*mp)		/* file system mount structure */
1396 {
1397 	int		level;
1398 	uint		maxblocks;
1399 	uint		maxleafents;
1400 	int		minleafrecs;
1401 	int		minnoderecs;
1402 
1403 	maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1404 		XFS_INODES_PER_CHUNK_LOG;
1405 	minleafrecs = mp->m_alloc_mnr[0];
1406 	minnoderecs = mp->m_alloc_mnr[1];
1407 	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1408 	for (level = 1; maxblocks > 1; level++)
1409 		maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1410 	mp->m_in_maxlevels = level;
1411 }
1412 
1413 /*
1414  * Log specified fields for the ag hdr (inode section)
1415  */
1416 void
xfs_ialloc_log_agi(xfs_trans_t * tp,xfs_buf_t * bp,int fields)1417 xfs_ialloc_log_agi(
1418 	xfs_trans_t	*tp,		/* transaction pointer */
1419 	xfs_buf_t	*bp,		/* allocation group header buffer */
1420 	int		fields)		/* bitmask of fields to log */
1421 {
1422 	int			first;		/* first byte number */
1423 	int			last;		/* last byte number */
1424 	static const short	offsets[] = {	/* field starting offsets */
1425 					/* keep in sync with bit definitions */
1426 		offsetof(xfs_agi_t, agi_magicnum),
1427 		offsetof(xfs_agi_t, agi_versionnum),
1428 		offsetof(xfs_agi_t, agi_seqno),
1429 		offsetof(xfs_agi_t, agi_length),
1430 		offsetof(xfs_agi_t, agi_count),
1431 		offsetof(xfs_agi_t, agi_root),
1432 		offsetof(xfs_agi_t, agi_level),
1433 		offsetof(xfs_agi_t, agi_freecount),
1434 		offsetof(xfs_agi_t, agi_newino),
1435 		offsetof(xfs_agi_t, agi_dirino),
1436 		offsetof(xfs_agi_t, agi_unlinked),
1437 		sizeof(xfs_agi_t)
1438 	};
1439 #ifdef DEBUG
1440 	xfs_agi_t		*agi;	/* allocation group header */
1441 
1442 	agi = XFS_BUF_TO_AGI(bp);
1443 	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1444 #endif
1445 	/*
1446 	 * Compute byte offsets for the first and last fields.
1447 	 */
1448 	xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1449 	/*
1450 	 * Log the allocation group inode header buffer.
1451 	 */
1452 	xfs_trans_log_buf(tp, bp, first, last);
1453 }
1454 
1455 #ifdef DEBUG
1456 STATIC void
xfs_check_agi_unlinked(struct xfs_agi * agi)1457 xfs_check_agi_unlinked(
1458 	struct xfs_agi		*agi)
1459 {
1460 	int			i;
1461 
1462 	for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1463 		ASSERT(agi->agi_unlinked[i]);
1464 }
1465 #else
1466 #define xfs_check_agi_unlinked(agi)
1467 #endif
1468 
1469 /*
1470  * Read in the allocation group header (inode allocation section)
1471  */
1472 int
xfs_read_agi(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,struct xfs_buf ** bpp)1473 xfs_read_agi(
1474 	struct xfs_mount	*mp,	/* file system mount structure */
1475 	struct xfs_trans	*tp,	/* transaction pointer */
1476 	xfs_agnumber_t		agno,	/* allocation group number */
1477 	struct xfs_buf		**bpp)	/* allocation group hdr buf */
1478 {
1479 	struct xfs_agi		*agi;	/* allocation group header */
1480 	int			agi_ok;	/* agi is consistent */
1481 	int			error;
1482 
1483 	ASSERT(agno != NULLAGNUMBER);
1484 
1485 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1486 			XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1487 			XFS_FSS_TO_BB(mp, 1), 0, bpp);
1488 	if (error)
1489 		return error;
1490 
1491 	ASSERT(!xfs_buf_geterror(*bpp));
1492 	agi = XFS_BUF_TO_AGI(*bpp);
1493 
1494 	/*
1495 	 * Validate the magic number of the agi block.
1496 	 */
1497 	agi_ok = agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC) &&
1498 		XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)) &&
1499 		be32_to_cpu(agi->agi_seqno) == agno;
1500 	if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1501 			XFS_RANDOM_IALLOC_READ_AGI))) {
1502 		XFS_CORRUPTION_ERROR("xfs_read_agi", XFS_ERRLEVEL_LOW,
1503 				     mp, agi);
1504 		xfs_trans_brelse(tp, *bpp);
1505 		return XFS_ERROR(EFSCORRUPTED);
1506 	}
1507 
1508 	xfs_buf_set_ref(*bpp, XFS_AGI_REF);
1509 
1510 	xfs_check_agi_unlinked(agi);
1511 	return 0;
1512 }
1513 
1514 int
xfs_ialloc_read_agi(struct xfs_mount * mp,struct xfs_trans * tp,xfs_agnumber_t agno,struct xfs_buf ** bpp)1515 xfs_ialloc_read_agi(
1516 	struct xfs_mount	*mp,	/* file system mount structure */
1517 	struct xfs_trans	*tp,	/* transaction pointer */
1518 	xfs_agnumber_t		agno,	/* allocation group number */
1519 	struct xfs_buf		**bpp)	/* allocation group hdr buf */
1520 {
1521 	struct xfs_agi		*agi;	/* allocation group header */
1522 	struct xfs_perag	*pag;	/* per allocation group data */
1523 	int			error;
1524 
1525 	error = xfs_read_agi(mp, tp, agno, bpp);
1526 	if (error)
1527 		return error;
1528 
1529 	agi = XFS_BUF_TO_AGI(*bpp);
1530 	pag = xfs_perag_get(mp, agno);
1531 	if (!pag->pagi_init) {
1532 		pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1533 		pag->pagi_count = be32_to_cpu(agi->agi_count);
1534 		pag->pagi_init = 1;
1535 	}
1536 
1537 	/*
1538 	 * It's possible for these to be out of sync if
1539 	 * we are in the middle of a forced shutdown.
1540 	 */
1541 	ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1542 		XFS_FORCED_SHUTDOWN(mp));
1543 	xfs_perag_put(pag);
1544 	return 0;
1545 }
1546 
1547 /*
1548  * Read in the agi to initialise the per-ag data in the mount structure
1549  */
1550 int
xfs_ialloc_pagi_init(xfs_mount_t * mp,xfs_trans_t * tp,xfs_agnumber_t agno)1551 xfs_ialloc_pagi_init(
1552 	xfs_mount_t	*mp,		/* file system mount structure */
1553 	xfs_trans_t	*tp,		/* transaction pointer */
1554 	xfs_agnumber_t	agno)		/* allocation group number */
1555 {
1556 	xfs_buf_t	*bp = NULL;
1557 	int		error;
1558 
1559 	error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1560 	if (error)
1561 		return error;
1562 	if (bp)
1563 		xfs_trans_brelse(tp, bp);
1564 	return 0;
1565 }
1566