xref: /linux/fs/xfs/libxfs/xfs_ialloc.c (revision 9141c5d389a9ec80121de3125cde7c713726ceea)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2002,2005 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_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
16 #include "xfs_ialloc.h"
17 #include "xfs_ialloc_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_errortag.h"
20 #include "xfs_error.h"
21 #include "xfs_bmap.h"
22 #include "xfs_trans.h"
23 #include "xfs_buf_item.h"
24 #include "xfs_icreate_item.h"
25 #include "xfs_icache.h"
26 #include "xfs_trace.h"
27 #include "xfs_log.h"
28 #include "xfs_rmap.h"
29 #include "xfs_ag.h"
30 #include "xfs_health.h"
31 
32 /*
33  * Lookup a record by ino in the btree given by cur.
34  */
35 int					/* error */
xfs_inobt_lookup(struct xfs_btree_cur * cur,xfs_agino_t ino,xfs_lookup_t dir,int * stat)36 xfs_inobt_lookup(
37 	struct xfs_btree_cur	*cur,	/* btree cursor */
38 	xfs_agino_t		ino,	/* starting inode of chunk */
39 	xfs_lookup_t		dir,	/* <=, >=, == */
40 	int			*stat)	/* success/failure */
41 {
42 	cur->bc_rec.i.ir_startino = ino;
43 	cur->bc_rec.i.ir_holemask = 0;
44 	cur->bc_rec.i.ir_count = 0;
45 	cur->bc_rec.i.ir_freecount = 0;
46 	cur->bc_rec.i.ir_free = 0;
47 	return xfs_btree_lookup(cur, dir, stat);
48 }
49 
50 /*
51  * Update the record referred to by cur to the value given.
52  * This either works (return 0) or gets an EFSCORRUPTED error.
53  */
54 STATIC int				/* error */
xfs_inobt_update(struct xfs_btree_cur * cur,xfs_inobt_rec_incore_t * irec)55 xfs_inobt_update(
56 	struct xfs_btree_cur	*cur,	/* btree cursor */
57 	xfs_inobt_rec_incore_t	*irec)	/* btree record */
58 {
59 	union xfs_btree_rec	rec;
60 
61 	rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
62 	if (xfs_has_sparseinodes(cur->bc_mp)) {
63 		rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
64 		rec.inobt.ir_u.sp.ir_count = irec->ir_count;
65 		rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
66 	} else {
67 		/* ir_holemask/ir_count not supported on-disk */
68 		rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
69 	}
70 	rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
71 	return xfs_btree_update(cur, &rec);
72 }
73 
74 /* Convert on-disk btree record to incore inobt record. */
75 void
xfs_inobt_btrec_to_irec(struct xfs_mount * mp,const union xfs_btree_rec * rec,struct xfs_inobt_rec_incore * irec)76 xfs_inobt_btrec_to_irec(
77 	struct xfs_mount		*mp,
78 	const union xfs_btree_rec	*rec,
79 	struct xfs_inobt_rec_incore	*irec)
80 {
81 	irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
82 	if (xfs_has_sparseinodes(mp)) {
83 		irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
84 		irec->ir_count = rec->inobt.ir_u.sp.ir_count;
85 		irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
86 	} else {
87 		/*
88 		 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
89 		 * values for full inode chunks.
90 		 */
91 		irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
92 		irec->ir_count = XFS_INODES_PER_CHUNK;
93 		irec->ir_freecount =
94 				be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
95 	}
96 	irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
97 }
98 
99 /* Compute the freecount of an incore inode record. */
100 uint8_t
xfs_inobt_rec_freecount(const struct xfs_inobt_rec_incore * irec)101 xfs_inobt_rec_freecount(
102 	const struct xfs_inobt_rec_incore	*irec)
103 {
104 	uint64_t				realfree = irec->ir_free;
105 
106 	if (xfs_inobt_issparse(irec->ir_holemask))
107 		realfree &= xfs_inobt_irec_to_allocmask(irec);
108 	return hweight64(realfree);
109 }
110 
111 /* Simple checks for inode records. */
112 xfs_failaddr_t
xfs_inobt_check_irec(struct xfs_perag * pag,const struct xfs_inobt_rec_incore * irec)113 xfs_inobt_check_irec(
114 	struct xfs_perag			*pag,
115 	const struct xfs_inobt_rec_incore	*irec)
116 {
117 	/* Record has to be properly aligned within the AG. */
118 	if (!xfs_verify_agino(pag, irec->ir_startino))
119 		return __this_address;
120 	if (!xfs_verify_agino(pag,
121 				irec->ir_startino + XFS_INODES_PER_CHUNK - 1))
122 		return __this_address;
123 	if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT ||
124 	    irec->ir_count > XFS_INODES_PER_CHUNK)
125 		return __this_address;
126 	if (irec->ir_freecount > XFS_INODES_PER_CHUNK)
127 		return __this_address;
128 
129 	if (xfs_inobt_rec_freecount(irec) != irec->ir_freecount)
130 		return __this_address;
131 
132 	return NULL;
133 }
134 
135 static inline int
xfs_inobt_complain_bad_rec(struct xfs_btree_cur * cur,xfs_failaddr_t fa,const struct xfs_inobt_rec_incore * irec)136 xfs_inobt_complain_bad_rec(
137 	struct xfs_btree_cur		*cur,
138 	xfs_failaddr_t			fa,
139 	const struct xfs_inobt_rec_incore *irec)
140 {
141 	struct xfs_mount		*mp = cur->bc_mp;
142 
143 	xfs_warn(mp,
144 		"%sbt record corruption in AG %d detected at %pS!",
145 		cur->bc_ops->name, cur->bc_group->xg_gno, fa);
146 	xfs_warn(mp,
147 "start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x",
148 		irec->ir_startino, irec->ir_count, irec->ir_freecount,
149 		irec->ir_free, irec->ir_holemask);
150 	xfs_btree_mark_sick(cur);
151 	return -EFSCORRUPTED;
152 }
153 
154 /*
155  * Get the data from the pointed-to record.
156  */
157 int
xfs_inobt_get_rec(struct xfs_btree_cur * cur,struct xfs_inobt_rec_incore * irec,int * stat)158 xfs_inobt_get_rec(
159 	struct xfs_btree_cur		*cur,
160 	struct xfs_inobt_rec_incore	*irec,
161 	int				*stat)
162 {
163 	struct xfs_mount		*mp = cur->bc_mp;
164 	union xfs_btree_rec		*rec;
165 	xfs_failaddr_t			fa;
166 	int				error;
167 
168 	error = xfs_btree_get_rec(cur, &rec, stat);
169 	if (error || *stat == 0)
170 		return error;
171 
172 	xfs_inobt_btrec_to_irec(mp, rec, irec);
173 	fa = xfs_inobt_check_irec(to_perag(cur->bc_group), irec);
174 	if (fa)
175 		return xfs_inobt_complain_bad_rec(cur, fa, irec);
176 
177 	return 0;
178 }
179 
180 /*
181  * Insert a single inobt record. Cursor must already point to desired location.
182  */
183 int
xfs_inobt_insert_rec(struct xfs_btree_cur * cur,uint16_t holemask,uint8_t count,int32_t freecount,xfs_inofree_t free,int * stat)184 xfs_inobt_insert_rec(
185 	struct xfs_btree_cur	*cur,
186 	uint16_t		holemask,
187 	uint8_t			count,
188 	int32_t			freecount,
189 	xfs_inofree_t		free,
190 	int			*stat)
191 {
192 	cur->bc_rec.i.ir_holemask = holemask;
193 	cur->bc_rec.i.ir_count = count;
194 	cur->bc_rec.i.ir_freecount = freecount;
195 	cur->bc_rec.i.ir_free = free;
196 	return xfs_btree_insert(cur, stat);
197 }
198 
199 /*
200  * Insert records describing a newly allocated inode chunk into the inobt.
201  */
202 STATIC int
xfs_inobt_insert(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp,xfs_agino_t newino,xfs_agino_t newlen,bool is_finobt)203 xfs_inobt_insert(
204 	struct xfs_perag	*pag,
205 	struct xfs_trans	*tp,
206 	struct xfs_buf		*agbp,
207 	xfs_agino_t		newino,
208 	xfs_agino_t		newlen,
209 	bool			is_finobt)
210 {
211 	struct xfs_btree_cur	*cur;
212 	xfs_agino_t		thisino;
213 	int			i;
214 	int			error;
215 
216 	if (is_finobt)
217 		cur = xfs_finobt_init_cursor(pag, tp, agbp);
218 	else
219 		cur = xfs_inobt_init_cursor(pag, tp, agbp);
220 
221 	for (thisino = newino;
222 	     thisino < newino + newlen;
223 	     thisino += XFS_INODES_PER_CHUNK) {
224 		error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
225 		if (error) {
226 			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
227 			return error;
228 		}
229 		ASSERT(i == 0);
230 
231 		error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
232 					     XFS_INODES_PER_CHUNK,
233 					     XFS_INODES_PER_CHUNK,
234 					     XFS_INOBT_ALL_FREE, &i);
235 		if (error) {
236 			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
237 			return error;
238 		}
239 		ASSERT(i == 1);
240 	}
241 
242 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
243 
244 	return 0;
245 }
246 
247 /*
248  * Verify that the number of free inodes in the AGI is correct.
249  */
250 #ifdef DEBUG
251 static int
xfs_check_agi_freecount(struct xfs_btree_cur * cur)252 xfs_check_agi_freecount(
253 	struct xfs_btree_cur	*cur)
254 {
255 	if (cur->bc_nlevels == 1) {
256 		xfs_inobt_rec_incore_t rec;
257 		int		freecount = 0;
258 		int		error;
259 		int		i;
260 
261 		error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
262 		if (error)
263 			return error;
264 
265 		do {
266 			error = xfs_inobt_get_rec(cur, &rec, &i);
267 			if (error)
268 				return error;
269 
270 			if (i) {
271 				freecount += rec.ir_freecount;
272 				error = xfs_btree_increment(cur, 0, &i);
273 				if (error)
274 					return error;
275 			}
276 		} while (i == 1);
277 
278 		if (!xfs_is_shutdown(cur->bc_mp)) {
279 			ASSERT(freecount ==
280 				to_perag(cur->bc_group)->pagi_freecount);
281 		}
282 	}
283 	return 0;
284 }
285 #else
286 #define xfs_check_agi_freecount(cur)	0
287 #endif
288 
289 /*
290  * Initialise a new set of inodes. When called without a transaction context
291  * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
292  * than logging them (which in a transaction context puts them into the AIL
293  * for writeback rather than the xfsbufd queue).
294  */
295 int
xfs_ialloc_inode_init(struct xfs_mount * mp,struct xfs_trans * tp,struct list_head * buffer_list,int icount,xfs_agnumber_t agno,xfs_agblock_t agbno,xfs_agblock_t length,unsigned int gen)296 xfs_ialloc_inode_init(
297 	struct xfs_mount	*mp,
298 	struct xfs_trans	*tp,
299 	struct list_head	*buffer_list,
300 	int			icount,
301 	xfs_agnumber_t		agno,
302 	xfs_agblock_t		agbno,
303 	xfs_agblock_t		length,
304 	unsigned int		gen)
305 {
306 	struct xfs_buf		*fbuf;
307 	struct xfs_dinode	*free;
308 	int			nbufs;
309 	int			version;
310 	int			i, j;
311 	xfs_daddr_t		d;
312 	xfs_ino_t		ino = 0;
313 	int			error;
314 
315 	/*
316 	 * Loop over the new block(s), filling in the inodes.  For small block
317 	 * sizes, manipulate the inodes in buffers  which are multiples of the
318 	 * blocks size.
319 	 */
320 	nbufs = length / M_IGEO(mp)->blocks_per_cluster;
321 
322 	/*
323 	 * Figure out what version number to use in the inodes we create.  If
324 	 * the superblock version has caught up to the one that supports the new
325 	 * inode format, then use the new inode version.  Otherwise use the old
326 	 * version so that old kernels will continue to be able to use the file
327 	 * system.
328 	 *
329 	 * For v3 inodes, we also need to write the inode number into the inode,
330 	 * so calculate the first inode number of the chunk here as
331 	 * XFS_AGB_TO_AGINO() only works within a filesystem block, not
332 	 * across multiple filesystem blocks (such as a cluster) and so cannot
333 	 * be used in the cluster buffer loop below.
334 	 *
335 	 * Further, because we are writing the inode directly into the buffer
336 	 * and calculating a CRC on the entire inode, we have ot log the entire
337 	 * inode so that the entire range the CRC covers is present in the log.
338 	 * That means for v3 inode we log the entire buffer rather than just the
339 	 * inode cores.
340 	 */
341 	if (xfs_has_v3inodes(mp)) {
342 		version = 3;
343 		ino = XFS_AGINO_TO_INO(mp, agno, XFS_AGB_TO_AGINO(mp, agbno));
344 
345 		/*
346 		 * log the initialisation that is about to take place as an
347 		 * logical operation. This means the transaction does not
348 		 * need to log the physical changes to the inode buffers as log
349 		 * recovery will know what initialisation is actually needed.
350 		 * Hence we only need to log the buffers as "ordered" buffers so
351 		 * they track in the AIL as if they were physically logged.
352 		 */
353 		if (tp)
354 			xfs_icreate_log(tp, agno, agbno, icount,
355 					mp->m_sb.sb_inodesize, length, gen);
356 	} else
357 		version = 2;
358 
359 	for (j = 0; j < nbufs; j++) {
360 		/*
361 		 * Get the block.
362 		 */
363 		d = XFS_AGB_TO_DADDR(mp, agno, agbno +
364 				(j * M_IGEO(mp)->blocks_per_cluster));
365 		error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
366 				mp->m_bsize * M_IGEO(mp)->blocks_per_cluster,
367 				XBF_UNMAPPED, &fbuf);
368 		if (error)
369 			return error;
370 
371 		/* Initialize the inode buffers and log them appropriately. */
372 		fbuf->b_ops = &xfs_inode_buf_ops;
373 		xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
374 		for (i = 0; i < M_IGEO(mp)->inodes_per_cluster; i++) {
375 			int	ioffset = i << mp->m_sb.sb_inodelog;
376 
377 			free = xfs_make_iptr(mp, fbuf, i);
378 			free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
379 			free->di_version = version;
380 			free->di_gen = cpu_to_be32(gen);
381 			free->di_next_unlinked = cpu_to_be32(NULLAGINO);
382 
383 			if (version == 3) {
384 				free->di_ino = cpu_to_be64(ino);
385 				ino++;
386 				uuid_copy(&free->di_uuid,
387 					  &mp->m_sb.sb_meta_uuid);
388 				xfs_dinode_calc_crc(mp, free);
389 			} else if (tp) {
390 				/* just log the inode core */
391 				xfs_trans_log_buf(tp, fbuf, ioffset,
392 					  ioffset + XFS_DINODE_SIZE(mp) - 1);
393 			}
394 		}
395 
396 		if (tp) {
397 			/*
398 			 * Mark the buffer as an inode allocation buffer so it
399 			 * sticks in AIL at the point of this allocation
400 			 * transaction. This ensures the they are on disk before
401 			 * the tail of the log can be moved past this
402 			 * transaction (i.e. by preventing relogging from moving
403 			 * it forward in the log).
404 			 */
405 			xfs_trans_inode_alloc_buf(tp, fbuf);
406 			if (version == 3) {
407 				/*
408 				 * Mark the buffer as ordered so that they are
409 				 * not physically logged in the transaction but
410 				 * still tracked in the AIL as part of the
411 				 * transaction and pin the log appropriately.
412 				 */
413 				xfs_trans_ordered_buf(tp, fbuf);
414 			}
415 		} else {
416 			fbuf->b_flags |= XBF_DONE;
417 			xfs_buf_delwri_queue(fbuf, buffer_list);
418 			xfs_buf_relse(fbuf);
419 		}
420 	}
421 	return 0;
422 }
423 
424 /*
425  * Align startino and allocmask for a recently allocated sparse chunk such that
426  * they are fit for insertion (or merge) into the on-disk inode btrees.
427  *
428  * Background:
429  *
430  * When enabled, sparse inode support increases the inode alignment from cluster
431  * size to inode chunk size. This means that the minimum range between two
432  * non-adjacent inode records in the inobt is large enough for a full inode
433  * record. This allows for cluster sized, cluster aligned block allocation
434  * without need to worry about whether the resulting inode record overlaps with
435  * another record in the tree. Without this basic rule, we would have to deal
436  * with the consequences of overlap by potentially undoing recent allocations in
437  * the inode allocation codepath.
438  *
439  * Because of this alignment rule (which is enforced on mount), there are two
440  * inobt possibilities for newly allocated sparse chunks. One is that the
441  * aligned inode record for the chunk covers a range of inodes not already
442  * covered in the inobt (i.e., it is safe to insert a new sparse record). The
443  * other is that a record already exists at the aligned startino that considers
444  * the newly allocated range as sparse. In the latter case, record content is
445  * merged in hope that sparse inode chunks fill to full chunks over time.
446  */
447 STATIC void
xfs_align_sparse_ino(struct xfs_mount * mp,xfs_agino_t * startino,uint16_t * allocmask)448 xfs_align_sparse_ino(
449 	struct xfs_mount		*mp,
450 	xfs_agino_t			*startino,
451 	uint16_t			*allocmask)
452 {
453 	xfs_agblock_t			agbno;
454 	xfs_agblock_t			mod;
455 	int				offset;
456 
457 	agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
458 	mod = agbno % mp->m_sb.sb_inoalignmt;
459 	if (!mod)
460 		return;
461 
462 	/* calculate the inode offset and align startino */
463 	offset = XFS_AGB_TO_AGINO(mp, mod);
464 	*startino -= offset;
465 
466 	/*
467 	 * Since startino has been aligned down, left shift allocmask such that
468 	 * it continues to represent the same physical inodes relative to the
469 	 * new startino.
470 	 */
471 	*allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
472 }
473 
474 /*
475  * Determine whether the source inode record can merge into the target. Both
476  * records must be sparse, the inode ranges must match and there must be no
477  * allocation overlap between the records.
478  */
479 STATIC bool
__xfs_inobt_can_merge(struct xfs_inobt_rec_incore * trec,struct xfs_inobt_rec_incore * srec)480 __xfs_inobt_can_merge(
481 	struct xfs_inobt_rec_incore	*trec,	/* tgt record */
482 	struct xfs_inobt_rec_incore	*srec)	/* src record */
483 {
484 	uint64_t			talloc;
485 	uint64_t			salloc;
486 
487 	/* records must cover the same inode range */
488 	if (trec->ir_startino != srec->ir_startino)
489 		return false;
490 
491 	/* both records must be sparse */
492 	if (!xfs_inobt_issparse(trec->ir_holemask) ||
493 	    !xfs_inobt_issparse(srec->ir_holemask))
494 		return false;
495 
496 	/* both records must track some inodes */
497 	if (!trec->ir_count || !srec->ir_count)
498 		return false;
499 
500 	/* can't exceed capacity of a full record */
501 	if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
502 		return false;
503 
504 	/* verify there is no allocation overlap */
505 	talloc = xfs_inobt_irec_to_allocmask(trec);
506 	salloc = xfs_inobt_irec_to_allocmask(srec);
507 	if (talloc & salloc)
508 		return false;
509 
510 	return true;
511 }
512 
513 /*
514  * Merge the source inode record into the target. The caller must call
515  * __xfs_inobt_can_merge() to ensure the merge is valid.
516  */
517 STATIC void
__xfs_inobt_rec_merge(struct xfs_inobt_rec_incore * trec,struct xfs_inobt_rec_incore * srec)518 __xfs_inobt_rec_merge(
519 	struct xfs_inobt_rec_incore	*trec,	/* target */
520 	struct xfs_inobt_rec_incore	*srec)	/* src */
521 {
522 	ASSERT(trec->ir_startino == srec->ir_startino);
523 
524 	/* combine the counts */
525 	trec->ir_count += srec->ir_count;
526 	trec->ir_freecount += srec->ir_freecount;
527 
528 	/*
529 	 * Merge the holemask and free mask. For both fields, 0 bits refer to
530 	 * allocated inodes. We combine the allocated ranges with bitwise AND.
531 	 */
532 	trec->ir_holemask &= srec->ir_holemask;
533 	trec->ir_free &= srec->ir_free;
534 }
535 
536 /*
537  * Insert a new sparse inode chunk into the associated inode allocation btree.
538  * The inode record for the sparse chunk is pre-aligned to a startino that
539  * should match any pre-existing sparse inode record in the tree. This allows
540  * sparse chunks to fill over time.
541  *
542  * If no preexisting record exists, the provided record is inserted.
543  * If there is a preexisting record, the provided record is merged with the
544  * existing record and updated in place. The merged record is returned in nrec.
545  *
546  * It is considered corruption if a merge is requested and not possible. Given
547  * the sparse inode alignment constraints, this should never happen.
548  */
549 STATIC int
xfs_inobt_insert_sprec(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp,struct xfs_inobt_rec_incore * nrec)550 xfs_inobt_insert_sprec(
551 	struct xfs_perag		*pag,
552 	struct xfs_trans		*tp,
553 	struct xfs_buf			*agbp,
554 	struct xfs_inobt_rec_incore	*nrec)	/* in/out: new/merged rec. */
555 {
556 	struct xfs_mount		*mp = pag_mount(pag);
557 	struct xfs_btree_cur		*cur;
558 	int				error;
559 	int				i;
560 	struct xfs_inobt_rec_incore	rec;
561 
562 	cur = xfs_inobt_init_cursor(pag, tp, agbp);
563 
564 	/* the new record is pre-aligned so we know where to look */
565 	error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
566 	if (error)
567 		goto error;
568 	/* if nothing there, insert a new record and return */
569 	if (i == 0) {
570 		error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
571 					     nrec->ir_count, nrec->ir_freecount,
572 					     nrec->ir_free, &i);
573 		if (error)
574 			goto error;
575 		if (XFS_IS_CORRUPT(mp, i != 1)) {
576 			xfs_btree_mark_sick(cur);
577 			error = -EFSCORRUPTED;
578 			goto error;
579 		}
580 
581 		goto out;
582 	}
583 
584 	/*
585 	 * A record exists at this startino.  Merge the records.
586 	 */
587 	error = xfs_inobt_get_rec(cur, &rec, &i);
588 	if (error)
589 		goto error;
590 	if (XFS_IS_CORRUPT(mp, i != 1)) {
591 		xfs_btree_mark_sick(cur);
592 		error = -EFSCORRUPTED;
593 		goto error;
594 	}
595 	if (XFS_IS_CORRUPT(mp, rec.ir_startino != nrec->ir_startino)) {
596 		xfs_btree_mark_sick(cur);
597 		error = -EFSCORRUPTED;
598 		goto error;
599 	}
600 
601 	/*
602 	 * This should never fail. If we have coexisting records that
603 	 * cannot merge, something is seriously wrong.
604 	 */
605 	if (XFS_IS_CORRUPT(mp, !__xfs_inobt_can_merge(nrec, &rec))) {
606 		xfs_btree_mark_sick(cur);
607 		error = -EFSCORRUPTED;
608 		goto error;
609 	}
610 
611 	trace_xfs_irec_merge_pre(pag, &rec, nrec);
612 
613 	/* merge to nrec to output the updated record */
614 	__xfs_inobt_rec_merge(nrec, &rec);
615 
616 	trace_xfs_irec_merge_post(pag, nrec);
617 
618 	error = xfs_inobt_rec_check_count(mp, nrec);
619 	if (error)
620 		goto error;
621 
622 	error = xfs_inobt_update(cur, nrec);
623 	if (error)
624 		goto error;
625 
626 out:
627 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
628 	return 0;
629 error:
630 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
631 	return error;
632 }
633 
634 /*
635  * Insert a new sparse inode chunk into the free inode btree. The inode
636  * record for the sparse chunk is pre-aligned to a startino that should match
637  * any pre-existing sparse inode record in the tree. This allows sparse chunks
638  * to fill over time.
639  *
640  * The new record is always inserted, overwriting a pre-existing record if
641  * there is one.
642  */
643 STATIC int
xfs_finobt_insert_sprec(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp,struct xfs_inobt_rec_incore * nrec)644 xfs_finobt_insert_sprec(
645 	struct xfs_perag		*pag,
646 	struct xfs_trans		*tp,
647 	struct xfs_buf			*agbp,
648 	struct xfs_inobt_rec_incore	*nrec)	/* in/out: new rec. */
649 {
650 	struct xfs_mount		*mp = pag_mount(pag);
651 	struct xfs_btree_cur		*cur;
652 	int				error;
653 	int				i;
654 
655 	cur = xfs_finobt_init_cursor(pag, tp, agbp);
656 
657 	/* the new record is pre-aligned so we know where to look */
658 	error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
659 	if (error)
660 		goto error;
661 	/* if nothing there, insert a new record and return */
662 	if (i == 0) {
663 		error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
664 					     nrec->ir_count, nrec->ir_freecount,
665 					     nrec->ir_free, &i);
666 		if (error)
667 			goto error;
668 		if (XFS_IS_CORRUPT(mp, i != 1)) {
669 			xfs_btree_mark_sick(cur);
670 			error = -EFSCORRUPTED;
671 			goto error;
672 		}
673 	} else {
674 		error = xfs_inobt_update(cur, nrec);
675 		if (error)
676 			goto error;
677 	}
678 
679 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
680 	return 0;
681 error:
682 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
683 	return error;
684 }
685 
686 
687 /*
688  * Allocate new inodes in the allocation group specified by agbp.  Returns 0 if
689  * inodes were allocated in this AG; -EAGAIN if there was no space in this AG so
690  * the caller knows it can try another AG, a hard -ENOSPC when over the maximum
691  * inode count threshold, or the usual negative error code for other errors.
692  */
693 STATIC int
xfs_ialloc_ag_alloc(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp)694 xfs_ialloc_ag_alloc(
695 	struct xfs_perag	*pag,
696 	struct xfs_trans	*tp,
697 	struct xfs_buf		*agbp)
698 {
699 	struct xfs_agi		*agi;
700 	struct xfs_alloc_arg	args;
701 	int			error;
702 	xfs_agino_t		newino;		/* new first inode's number */
703 	xfs_agino_t		newlen;		/* new number of inodes */
704 	int			isaligned = 0;	/* inode allocation at stripe */
705 						/* unit boundary */
706 	/* init. to full chunk */
707 	struct xfs_inobt_rec_incore rec;
708 	struct xfs_ino_geometry	*igeo = M_IGEO(tp->t_mountp);
709 	uint16_t		allocmask = (uint16_t) -1;
710 	int			do_sparse = 0;
711 
712 	memset(&args, 0, sizeof(args));
713 	args.tp = tp;
714 	args.mp = tp->t_mountp;
715 	args.fsbno = NULLFSBLOCK;
716 	args.oinfo = XFS_RMAP_OINFO_INODES;
717 	args.pag = pag;
718 
719 #ifdef DEBUG
720 	/* randomly do sparse inode allocations */
721 	if (xfs_has_sparseinodes(tp->t_mountp) &&
722 	    igeo->ialloc_min_blks < igeo->ialloc_blks)
723 		do_sparse = get_random_u32_below(2);
724 #endif
725 
726 	/*
727 	 * Locking will ensure that we don't have two callers in here
728 	 * at one time.
729 	 */
730 	newlen = igeo->ialloc_inos;
731 	if (igeo->maxicount &&
732 	    percpu_counter_read_positive(&args.mp->m_icount) + newlen >
733 							igeo->maxicount)
734 		return -ENOSPC;
735 	args.minlen = args.maxlen = igeo->ialloc_blks;
736 	/*
737 	 * First try to allocate inodes contiguous with the last-allocated
738 	 * chunk of inodes.  If the filesystem is striped, this will fill
739 	 * an entire stripe unit with inodes.
740 	 */
741 	agi = agbp->b_addr;
742 	newino = be32_to_cpu(agi->agi_newino);
743 	args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
744 		     igeo->ialloc_blks;
745 	if (do_sparse)
746 		goto sparse_alloc;
747 	if (likely(newino != NULLAGINO &&
748 		  (args.agbno < be32_to_cpu(agi->agi_length)))) {
749 		args.prod = 1;
750 
751 		/*
752 		 * We need to take into account alignment here to ensure that
753 		 * we don't modify the free list if we fail to have an exact
754 		 * block. If we don't have an exact match, and every oher
755 		 * attempt allocation attempt fails, we'll end up cancelling
756 		 * a dirty transaction and shutting down.
757 		 *
758 		 * For an exact allocation, alignment must be 1,
759 		 * however we need to take cluster alignment into account when
760 		 * fixing up the freelist. Use the minalignslop field to
761 		 * indicate that extra blocks might be required for alignment,
762 		 * but not to use them in the actual exact allocation.
763 		 */
764 		args.alignment = 1;
765 		args.minalignslop = igeo->cluster_align - 1;
766 
767 		/* Allow space for the inode btree to split. */
768 		args.minleft = igeo->inobt_maxlevels;
769 		error = xfs_alloc_vextent_exact_bno(&args,
770 				xfs_agbno_to_fsb(pag, args.agbno));
771 		if (error)
772 			return error;
773 
774 		/*
775 		 * This request might have dirtied the transaction if the AG can
776 		 * satisfy the request, but the exact block was not available.
777 		 * If the allocation did fail, subsequent requests will relax
778 		 * the exact agbno requirement and increase the alignment
779 		 * instead. It is critical that the total size of the request
780 		 * (len + alignment + slop) does not increase from this point
781 		 * on, so reset minalignslop to ensure it is not included in
782 		 * subsequent requests.
783 		 */
784 		args.minalignslop = 0;
785 	}
786 
787 	if (unlikely(args.fsbno == NULLFSBLOCK)) {
788 		/*
789 		 * Set the alignment for the allocation.
790 		 * If stripe alignment is turned on then align at stripe unit
791 		 * boundary.
792 		 * If the cluster size is smaller than a filesystem block
793 		 * then we're doing I/O for inodes in filesystem block size
794 		 * pieces, so don't need alignment anyway.
795 		 */
796 		isaligned = 0;
797 		if (igeo->ialloc_align) {
798 			ASSERT(!xfs_has_noalign(args.mp));
799 			args.alignment = args.mp->m_dalign;
800 			isaligned = 1;
801 		} else
802 			args.alignment = igeo->cluster_align;
803 		/*
804 		 * Allocate a fixed-size extent of inodes.
805 		 */
806 		args.prod = 1;
807 		/*
808 		 * Allow space for the inode btree to split.
809 		 */
810 		args.minleft = igeo->inobt_maxlevels;
811 		error = xfs_alloc_vextent_near_bno(&args,
812 				xfs_agbno_to_fsb(pag,
813 					be32_to_cpu(agi->agi_root)));
814 		if (error)
815 			return error;
816 	}
817 
818 	/*
819 	 * If stripe alignment is turned on, then try again with cluster
820 	 * alignment.
821 	 */
822 	if (isaligned && args.fsbno == NULLFSBLOCK) {
823 		args.alignment = igeo->cluster_align;
824 		error = xfs_alloc_vextent_near_bno(&args,
825 				xfs_agbno_to_fsb(pag,
826 					be32_to_cpu(agi->agi_root)));
827 		if (error)
828 			return error;
829 	}
830 
831 	/*
832 	 * Finally, try a sparse allocation if the filesystem supports it and
833 	 * the sparse allocation length is smaller than a full chunk.
834 	 */
835 	if (xfs_has_sparseinodes(args.mp) &&
836 	    igeo->ialloc_min_blks < igeo->ialloc_blks &&
837 	    args.fsbno == NULLFSBLOCK) {
838 sparse_alloc:
839 		args.alignment = args.mp->m_sb.sb_spino_align;
840 		args.prod = 1;
841 
842 		args.minlen = igeo->ialloc_min_blks;
843 		args.maxlen = args.minlen;
844 
845 		/*
846 		 * The inode record will be aligned to full chunk size. We must
847 		 * prevent sparse allocation from AG boundaries that result in
848 		 * invalid inode records, such as records that start at agbno 0
849 		 * or extend beyond the AG.
850 		 *
851 		 * Set min agbno to the first aligned, non-zero agbno and max to
852 		 * the last aligned agbno that is at least one full chunk from
853 		 * the end of the AG.
854 		 */
855 		args.min_agbno = args.mp->m_sb.sb_inoalignmt;
856 		args.max_agbno = round_down(xfs_ag_block_count(args.mp,
857 							pag_agno(pag)),
858 					    args.mp->m_sb.sb_inoalignmt) -
859 				 igeo->ialloc_blks;
860 
861 		error = xfs_alloc_vextent_near_bno(&args,
862 				xfs_agbno_to_fsb(pag,
863 					be32_to_cpu(agi->agi_root)));
864 		if (error)
865 			return error;
866 
867 		newlen = XFS_AGB_TO_AGINO(args.mp, args.len);
868 		ASSERT(newlen <= XFS_INODES_PER_CHUNK);
869 		allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
870 	}
871 
872 	if (args.fsbno == NULLFSBLOCK)
873 		return -EAGAIN;
874 
875 	ASSERT(args.len == args.minlen);
876 
877 	/*
878 	 * Stamp and write the inode buffers.
879 	 *
880 	 * Seed the new inode cluster with a random generation number. This
881 	 * prevents short-term reuse of generation numbers if a chunk is
882 	 * freed and then immediately reallocated. We use random numbers
883 	 * rather than a linear progression to prevent the next generation
884 	 * number from being easily guessable.
885 	 */
886 	error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, pag_agno(pag),
887 			args.agbno, args.len, get_random_u32());
888 
889 	if (error)
890 		return error;
891 	/*
892 	 * Convert the results.
893 	 */
894 	newino = XFS_AGB_TO_AGINO(args.mp, args.agbno);
895 
896 	if (xfs_inobt_issparse(~allocmask)) {
897 		/*
898 		 * We've allocated a sparse chunk. Align the startino and mask.
899 		 */
900 		xfs_align_sparse_ino(args.mp, &newino, &allocmask);
901 
902 		rec.ir_startino = newino;
903 		rec.ir_holemask = ~allocmask;
904 		rec.ir_count = newlen;
905 		rec.ir_freecount = newlen;
906 		rec.ir_free = XFS_INOBT_ALL_FREE;
907 
908 		/*
909 		 * Insert the sparse record into the inobt and allow for a merge
910 		 * if necessary. If a merge does occur, rec is updated to the
911 		 * merged record.
912 		 */
913 		error = xfs_inobt_insert_sprec(pag, tp, agbp, &rec);
914 		if (error == -EFSCORRUPTED) {
915 			xfs_alert(args.mp,
916 	"invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
917 				  xfs_agino_to_ino(pag, rec.ir_startino),
918 				  rec.ir_holemask, rec.ir_count);
919 			xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
920 		}
921 		if (error)
922 			return error;
923 
924 		/*
925 		 * We can't merge the part we've just allocated as for the inobt
926 		 * due to finobt semantics. The original record may or may not
927 		 * exist independent of whether physical inodes exist in this
928 		 * sparse chunk.
929 		 *
930 		 * We must update the finobt record based on the inobt record.
931 		 * rec contains the fully merged and up to date inobt record
932 		 * from the previous call. Set merge false to replace any
933 		 * existing record with this one.
934 		 */
935 		if (xfs_has_finobt(args.mp)) {
936 			error = xfs_finobt_insert_sprec(pag, tp, agbp, &rec);
937 			if (error)
938 				return error;
939 		}
940 	} else {
941 		/* full chunk - insert new records to both btrees */
942 		error = xfs_inobt_insert(pag, tp, agbp, newino, newlen, false);
943 		if (error)
944 			return error;
945 
946 		if (xfs_has_finobt(args.mp)) {
947 			error = xfs_inobt_insert(pag, tp, agbp, newino,
948 						 newlen, true);
949 			if (error)
950 				return error;
951 		}
952 	}
953 
954 	/*
955 	 * Update AGI counts and newino.
956 	 */
957 	be32_add_cpu(&agi->agi_count, newlen);
958 	be32_add_cpu(&agi->agi_freecount, newlen);
959 	pag->pagi_freecount += newlen;
960 	pag->pagi_count += newlen;
961 	agi->agi_newino = cpu_to_be32(newino);
962 
963 	/*
964 	 * Log allocation group header fields
965 	 */
966 	xfs_ialloc_log_agi(tp, agbp,
967 		XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
968 	/*
969 	 * Modify/log superblock values for inode count and inode free count.
970 	 */
971 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
972 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
973 	return 0;
974 }
975 
976 /*
977  * Try to retrieve the next record to the left/right from the current one.
978  */
979 STATIC int
xfs_ialloc_next_rec(struct xfs_btree_cur * cur,xfs_inobt_rec_incore_t * rec,int * done,int left)980 xfs_ialloc_next_rec(
981 	struct xfs_btree_cur	*cur,
982 	xfs_inobt_rec_incore_t	*rec,
983 	int			*done,
984 	int			left)
985 {
986 	int                     error;
987 	int			i;
988 
989 	if (left)
990 		error = xfs_btree_decrement(cur, 0, &i);
991 	else
992 		error = xfs_btree_increment(cur, 0, &i);
993 
994 	if (error)
995 		return error;
996 	*done = !i;
997 	if (i) {
998 		error = xfs_inobt_get_rec(cur, rec, &i);
999 		if (error)
1000 			return error;
1001 		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1002 			xfs_btree_mark_sick(cur);
1003 			return -EFSCORRUPTED;
1004 		}
1005 	}
1006 
1007 	return 0;
1008 }
1009 
1010 STATIC int
xfs_ialloc_get_rec(struct xfs_btree_cur * cur,xfs_agino_t agino,xfs_inobt_rec_incore_t * rec,int * done)1011 xfs_ialloc_get_rec(
1012 	struct xfs_btree_cur	*cur,
1013 	xfs_agino_t		agino,
1014 	xfs_inobt_rec_incore_t	*rec,
1015 	int			*done)
1016 {
1017 	int                     error;
1018 	int			i;
1019 
1020 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
1021 	if (error)
1022 		return error;
1023 	*done = !i;
1024 	if (i) {
1025 		error = xfs_inobt_get_rec(cur, rec, &i);
1026 		if (error)
1027 			return error;
1028 		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1029 			xfs_btree_mark_sick(cur);
1030 			return -EFSCORRUPTED;
1031 		}
1032 	}
1033 
1034 	return 0;
1035 }
1036 
1037 /*
1038  * Return the offset of the first free inode in the record. If the inode chunk
1039  * is sparsely allocated, we convert the record holemask to inode granularity
1040  * and mask off the unallocated regions from the inode free mask.
1041  */
1042 STATIC int
xfs_inobt_first_free_inode(struct xfs_inobt_rec_incore * rec)1043 xfs_inobt_first_free_inode(
1044 	struct xfs_inobt_rec_incore	*rec)
1045 {
1046 	xfs_inofree_t			realfree;
1047 
1048 	/* if there are no holes, return the first available offset */
1049 	if (!xfs_inobt_issparse(rec->ir_holemask))
1050 		return xfs_lowbit64(rec->ir_free);
1051 
1052 	realfree = xfs_inobt_irec_to_allocmask(rec);
1053 	realfree &= rec->ir_free;
1054 
1055 	return xfs_lowbit64(realfree);
1056 }
1057 
1058 /*
1059  * If this AG has corrupt inodes, check if allocating this inode would fail
1060  * with corruption errors.  Returns 0 if we're clear, or EAGAIN to try again
1061  * somewhere else.
1062  */
1063 static int
xfs_dialloc_check_ino(struct xfs_perag * pag,struct xfs_trans * tp,xfs_ino_t ino)1064 xfs_dialloc_check_ino(
1065 	struct xfs_perag	*pag,
1066 	struct xfs_trans	*tp,
1067 	xfs_ino_t		ino)
1068 {
1069 	struct xfs_imap		imap;
1070 	struct xfs_buf		*bp;
1071 	int			error;
1072 
1073 	error = xfs_imap(pag, tp, ino, &imap, 0);
1074 	if (error)
1075 		return -EAGAIN;
1076 
1077 	error = xfs_imap_to_bp(pag_mount(pag), tp, &imap, &bp);
1078 	if (error)
1079 		return -EAGAIN;
1080 
1081 	xfs_trans_brelse(tp, bp);
1082 	return 0;
1083 }
1084 
1085 /*
1086  * Allocate an inode using the inobt-only algorithm.
1087  */
1088 STATIC int
xfs_dialloc_ag_inobt(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp,xfs_ino_t parent,xfs_ino_t * inop)1089 xfs_dialloc_ag_inobt(
1090 	struct xfs_perag	*pag,
1091 	struct xfs_trans	*tp,
1092 	struct xfs_buf		*agbp,
1093 	xfs_ino_t		parent,
1094 	xfs_ino_t		*inop)
1095 {
1096 	struct xfs_mount	*mp = tp->t_mountp;
1097 	struct xfs_agi		*agi = agbp->b_addr;
1098 	xfs_agnumber_t		pagno = XFS_INO_TO_AGNO(mp, parent);
1099 	xfs_agino_t		pagino = XFS_INO_TO_AGINO(mp, parent);
1100 	struct xfs_btree_cur	*cur, *tcur;
1101 	struct xfs_inobt_rec_incore rec, trec;
1102 	xfs_ino_t		ino;
1103 	int			error;
1104 	int			offset;
1105 	int			i, j;
1106 	int			searchdistance = 10;
1107 
1108 	ASSERT(xfs_perag_initialised_agi(pag));
1109 	ASSERT(xfs_perag_allows_inodes(pag));
1110 	ASSERT(pag->pagi_freecount > 0);
1111 
1112  restart_pagno:
1113 	cur = xfs_inobt_init_cursor(pag, tp, agbp);
1114 	/*
1115 	 * If pagino is 0 (this is the root inode allocation) use newino.
1116 	 * This must work because we've just allocated some.
1117 	 */
1118 	if (!pagino)
1119 		pagino = be32_to_cpu(agi->agi_newino);
1120 
1121 	error = xfs_check_agi_freecount(cur);
1122 	if (error)
1123 		goto error0;
1124 
1125 	/*
1126 	 * If in the same AG as the parent, try to get near the parent.
1127 	 */
1128 	if (pagno == pag_agno(pag)) {
1129 		int		doneleft;	/* done, to the left */
1130 		int		doneright;	/* done, to the right */
1131 
1132 		error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1133 		if (error)
1134 			goto error0;
1135 		if (XFS_IS_CORRUPT(mp, i != 1)) {
1136 			xfs_btree_mark_sick(cur);
1137 			error = -EFSCORRUPTED;
1138 			goto error0;
1139 		}
1140 
1141 		error = xfs_inobt_get_rec(cur, &rec, &j);
1142 		if (error)
1143 			goto error0;
1144 		if (XFS_IS_CORRUPT(mp, j != 1)) {
1145 			xfs_btree_mark_sick(cur);
1146 			error = -EFSCORRUPTED;
1147 			goto error0;
1148 		}
1149 
1150 		if (rec.ir_freecount > 0) {
1151 			/*
1152 			 * Found a free inode in the same chunk
1153 			 * as the parent, done.
1154 			 */
1155 			goto alloc_inode;
1156 		}
1157 
1158 
1159 		/*
1160 		 * In the same AG as parent, but parent's chunk is full.
1161 		 */
1162 
1163 		/* duplicate the cursor, search left & right simultaneously */
1164 		error = xfs_btree_dup_cursor(cur, &tcur);
1165 		if (error)
1166 			goto error0;
1167 
1168 		/*
1169 		 * Skip to last blocks looked up if same parent inode.
1170 		 */
1171 		if (pagino != NULLAGINO &&
1172 		    pag->pagl_pagino == pagino &&
1173 		    pag->pagl_leftrec != NULLAGINO &&
1174 		    pag->pagl_rightrec != NULLAGINO) {
1175 			error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
1176 						   &trec, &doneleft);
1177 			if (error)
1178 				goto error1;
1179 
1180 			error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
1181 						   &rec, &doneright);
1182 			if (error)
1183 				goto error1;
1184 		} else {
1185 			/* search left with tcur, back up 1 record */
1186 			error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1187 			if (error)
1188 				goto error1;
1189 
1190 			/* search right with cur, go forward 1 record. */
1191 			error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1192 			if (error)
1193 				goto error1;
1194 		}
1195 
1196 		/*
1197 		 * Loop until we find an inode chunk with a free inode.
1198 		 */
1199 		while (--searchdistance > 0 && (!doneleft || !doneright)) {
1200 			int	useleft;  /* using left inode chunk this time */
1201 
1202 			/* figure out the closer block if both are valid. */
1203 			if (!doneleft && !doneright) {
1204 				useleft = pagino -
1205 				 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1206 				  rec.ir_startino - pagino;
1207 			} else {
1208 				useleft = !doneleft;
1209 			}
1210 
1211 			/* free inodes to the left? */
1212 			if (useleft && trec.ir_freecount) {
1213 				xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1214 				cur = tcur;
1215 
1216 				pag->pagl_leftrec = trec.ir_startino;
1217 				pag->pagl_rightrec = rec.ir_startino;
1218 				pag->pagl_pagino = pagino;
1219 				rec = trec;
1220 				goto alloc_inode;
1221 			}
1222 
1223 			/* free inodes to the right? */
1224 			if (!useleft && rec.ir_freecount) {
1225 				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1226 
1227 				pag->pagl_leftrec = trec.ir_startino;
1228 				pag->pagl_rightrec = rec.ir_startino;
1229 				pag->pagl_pagino = pagino;
1230 				goto alloc_inode;
1231 			}
1232 
1233 			/* get next record to check */
1234 			if (useleft) {
1235 				error = xfs_ialloc_next_rec(tcur, &trec,
1236 								 &doneleft, 1);
1237 			} else {
1238 				error = xfs_ialloc_next_rec(cur, &rec,
1239 								 &doneright, 0);
1240 			}
1241 			if (error)
1242 				goto error1;
1243 		}
1244 
1245 		if (searchdistance <= 0) {
1246 			/*
1247 			 * Not in range - save last search
1248 			 * location and allocate a new inode
1249 			 */
1250 			xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1251 			pag->pagl_leftrec = trec.ir_startino;
1252 			pag->pagl_rightrec = rec.ir_startino;
1253 			pag->pagl_pagino = pagino;
1254 
1255 		} else {
1256 			/*
1257 			 * We've reached the end of the btree. because
1258 			 * we are only searching a small chunk of the
1259 			 * btree each search, there is obviously free
1260 			 * inodes closer to the parent inode than we
1261 			 * are now. restart the search again.
1262 			 */
1263 			pag->pagl_pagino = NULLAGINO;
1264 			pag->pagl_leftrec = NULLAGINO;
1265 			pag->pagl_rightrec = NULLAGINO;
1266 			xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1267 			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1268 			goto restart_pagno;
1269 		}
1270 	}
1271 
1272 	/*
1273 	 * In a different AG from the parent.
1274 	 * See if the most recently allocated block has any free.
1275 	 */
1276 	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1277 		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1278 					 XFS_LOOKUP_EQ, &i);
1279 		if (error)
1280 			goto error0;
1281 
1282 		if (i == 1) {
1283 			error = xfs_inobt_get_rec(cur, &rec, &j);
1284 			if (error)
1285 				goto error0;
1286 
1287 			if (j == 1 && rec.ir_freecount > 0) {
1288 				/*
1289 				 * The last chunk allocated in the group
1290 				 * still has a free inode.
1291 				 */
1292 				goto alloc_inode;
1293 			}
1294 		}
1295 	}
1296 
1297 	/*
1298 	 * None left in the last group, search the whole AG
1299 	 */
1300 	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1301 	if (error)
1302 		goto error0;
1303 	if (XFS_IS_CORRUPT(mp, i != 1)) {
1304 		xfs_btree_mark_sick(cur);
1305 		error = -EFSCORRUPTED;
1306 		goto error0;
1307 	}
1308 
1309 	for (;;) {
1310 		error = xfs_inobt_get_rec(cur, &rec, &i);
1311 		if (error)
1312 			goto error0;
1313 		if (XFS_IS_CORRUPT(mp, i != 1)) {
1314 			xfs_btree_mark_sick(cur);
1315 			error = -EFSCORRUPTED;
1316 			goto error0;
1317 		}
1318 		if (rec.ir_freecount > 0)
1319 			break;
1320 		error = xfs_btree_increment(cur, 0, &i);
1321 		if (error)
1322 			goto error0;
1323 		if (XFS_IS_CORRUPT(mp, i != 1)) {
1324 			xfs_btree_mark_sick(cur);
1325 			error = -EFSCORRUPTED;
1326 			goto error0;
1327 		}
1328 	}
1329 
1330 alloc_inode:
1331 	offset = xfs_inobt_first_free_inode(&rec);
1332 	ASSERT(offset >= 0);
1333 	ASSERT(offset < XFS_INODES_PER_CHUNK);
1334 	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1335 				   XFS_INODES_PER_CHUNK) == 0);
1336 	ino = xfs_agino_to_ino(pag, rec.ir_startino + offset);
1337 
1338 	if (xfs_ag_has_sickness(pag, XFS_SICK_AG_INODES)) {
1339 		error = xfs_dialloc_check_ino(pag, tp, ino);
1340 		if (error)
1341 			goto error0;
1342 	}
1343 
1344 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1345 	rec.ir_freecount--;
1346 	error = xfs_inobt_update(cur, &rec);
1347 	if (error)
1348 		goto error0;
1349 	be32_add_cpu(&agi->agi_freecount, -1);
1350 	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1351 	pag->pagi_freecount--;
1352 
1353 	error = xfs_check_agi_freecount(cur);
1354 	if (error)
1355 		goto error0;
1356 
1357 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1358 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1359 	*inop = ino;
1360 	return 0;
1361 error1:
1362 	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1363 error0:
1364 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1365 	return error;
1366 }
1367 
1368 /*
1369  * Use the free inode btree to allocate an inode based on distance from the
1370  * parent. Note that the provided cursor may be deleted and replaced.
1371  */
1372 STATIC int
xfs_dialloc_ag_finobt_near(xfs_agino_t pagino,struct xfs_btree_cur ** ocur,struct xfs_inobt_rec_incore * rec)1373 xfs_dialloc_ag_finobt_near(
1374 	xfs_agino_t			pagino,
1375 	struct xfs_btree_cur		**ocur,
1376 	struct xfs_inobt_rec_incore	*rec)
1377 {
1378 	struct xfs_btree_cur		*lcur = *ocur;	/* left search cursor */
1379 	struct xfs_btree_cur		*rcur;	/* right search cursor */
1380 	struct xfs_inobt_rec_incore	rrec;
1381 	int				error;
1382 	int				i, j;
1383 
1384 	error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1385 	if (error)
1386 		return error;
1387 
1388 	if (i == 1) {
1389 		error = xfs_inobt_get_rec(lcur, rec, &i);
1390 		if (error)
1391 			return error;
1392 		if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1)) {
1393 			xfs_btree_mark_sick(lcur);
1394 			return -EFSCORRUPTED;
1395 		}
1396 
1397 		/*
1398 		 * See if we've landed in the parent inode record. The finobt
1399 		 * only tracks chunks with at least one free inode, so record
1400 		 * existence is enough.
1401 		 */
1402 		if (pagino >= rec->ir_startino &&
1403 		    pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1404 			return 0;
1405 	}
1406 
1407 	error = xfs_btree_dup_cursor(lcur, &rcur);
1408 	if (error)
1409 		return error;
1410 
1411 	error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1412 	if (error)
1413 		goto error_rcur;
1414 	if (j == 1) {
1415 		error = xfs_inobt_get_rec(rcur, &rrec, &j);
1416 		if (error)
1417 			goto error_rcur;
1418 		if (XFS_IS_CORRUPT(lcur->bc_mp, j != 1)) {
1419 			xfs_btree_mark_sick(lcur);
1420 			error = -EFSCORRUPTED;
1421 			goto error_rcur;
1422 		}
1423 	}
1424 
1425 	if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1 && j != 1)) {
1426 		xfs_btree_mark_sick(lcur);
1427 		error = -EFSCORRUPTED;
1428 		goto error_rcur;
1429 	}
1430 	if (i == 1 && j == 1) {
1431 		/*
1432 		 * Both the left and right records are valid. Choose the closer
1433 		 * inode chunk to the target.
1434 		 */
1435 		if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1436 		    (rrec.ir_startino - pagino)) {
1437 			*rec = rrec;
1438 			xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1439 			*ocur = rcur;
1440 		} else {
1441 			xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1442 		}
1443 	} else if (j == 1) {
1444 		/* only the right record is valid */
1445 		*rec = rrec;
1446 		xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1447 		*ocur = rcur;
1448 	} else if (i == 1) {
1449 		/* only the left record is valid */
1450 		xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1451 	}
1452 
1453 	return 0;
1454 
1455 error_rcur:
1456 	xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1457 	return error;
1458 }
1459 
1460 /*
1461  * Use the free inode btree to find a free inode based on a newino hint. If
1462  * the hint is NULL, find the first free inode in the AG.
1463  */
1464 STATIC int
xfs_dialloc_ag_finobt_newino(struct xfs_agi * agi,struct xfs_btree_cur * cur,struct xfs_inobt_rec_incore * rec)1465 xfs_dialloc_ag_finobt_newino(
1466 	struct xfs_agi			*agi,
1467 	struct xfs_btree_cur		*cur,
1468 	struct xfs_inobt_rec_incore	*rec)
1469 {
1470 	int error;
1471 	int i;
1472 
1473 	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1474 		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1475 					 XFS_LOOKUP_EQ, &i);
1476 		if (error)
1477 			return error;
1478 		if (i == 1) {
1479 			error = xfs_inobt_get_rec(cur, rec, &i);
1480 			if (error)
1481 				return error;
1482 			if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1483 				xfs_btree_mark_sick(cur);
1484 				return -EFSCORRUPTED;
1485 			}
1486 			return 0;
1487 		}
1488 	}
1489 
1490 	/*
1491 	 * Find the first inode available in the AG.
1492 	 */
1493 	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1494 	if (error)
1495 		return error;
1496 	if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1497 		xfs_btree_mark_sick(cur);
1498 		return -EFSCORRUPTED;
1499 	}
1500 
1501 	error = xfs_inobt_get_rec(cur, rec, &i);
1502 	if (error)
1503 		return error;
1504 	if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1505 		xfs_btree_mark_sick(cur);
1506 		return -EFSCORRUPTED;
1507 	}
1508 
1509 	return 0;
1510 }
1511 
1512 /*
1513  * Update the inobt based on a modification made to the finobt. Also ensure that
1514  * the records from both trees are equivalent post-modification.
1515  */
1516 STATIC int
xfs_dialloc_ag_update_inobt(struct xfs_btree_cur * cur,struct xfs_inobt_rec_incore * frec,int offset)1517 xfs_dialloc_ag_update_inobt(
1518 	struct xfs_btree_cur		*cur,	/* inobt cursor */
1519 	struct xfs_inobt_rec_incore	*frec,	/* finobt record */
1520 	int				offset) /* inode offset */
1521 {
1522 	struct xfs_inobt_rec_incore	rec;
1523 	int				error;
1524 	int				i;
1525 
1526 	error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1527 	if (error)
1528 		return error;
1529 	if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1530 		xfs_btree_mark_sick(cur);
1531 		return -EFSCORRUPTED;
1532 	}
1533 
1534 	error = xfs_inobt_get_rec(cur, &rec, &i);
1535 	if (error)
1536 		return error;
1537 	if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
1538 		xfs_btree_mark_sick(cur);
1539 		return -EFSCORRUPTED;
1540 	}
1541 	ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1542 				   XFS_INODES_PER_CHUNK) == 0);
1543 
1544 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1545 	rec.ir_freecount--;
1546 
1547 	if (XFS_IS_CORRUPT(cur->bc_mp,
1548 			   rec.ir_free != frec->ir_free ||
1549 			   rec.ir_freecount != frec->ir_freecount)) {
1550 		xfs_btree_mark_sick(cur);
1551 		return -EFSCORRUPTED;
1552 	}
1553 
1554 	return xfs_inobt_update(cur, &rec);
1555 }
1556 
1557 /*
1558  * Allocate an inode using the free inode btree, if available. Otherwise, fall
1559  * back to the inobt search algorithm.
1560  *
1561  * The caller selected an AG for us, and made sure that free inodes are
1562  * available.
1563  */
1564 static int
xfs_dialloc_ag(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp,xfs_ino_t parent,xfs_ino_t * inop)1565 xfs_dialloc_ag(
1566 	struct xfs_perag	*pag,
1567 	struct xfs_trans	*tp,
1568 	struct xfs_buf		*agbp,
1569 	xfs_ino_t		parent,
1570 	xfs_ino_t		*inop)
1571 {
1572 	struct xfs_mount		*mp = tp->t_mountp;
1573 	struct xfs_agi			*agi = agbp->b_addr;
1574 	xfs_agnumber_t			pagno = XFS_INO_TO_AGNO(mp, parent);
1575 	xfs_agino_t			pagino = XFS_INO_TO_AGINO(mp, parent);
1576 	struct xfs_btree_cur		*cur;	/* finobt cursor */
1577 	struct xfs_btree_cur		*icur;	/* inobt cursor */
1578 	struct xfs_inobt_rec_incore	rec;
1579 	xfs_ino_t			ino;
1580 	int				error;
1581 	int				offset;
1582 	int				i;
1583 
1584 	if (!xfs_has_finobt(mp))
1585 		return xfs_dialloc_ag_inobt(pag, tp, agbp, parent, inop);
1586 
1587 	/*
1588 	 * If pagino is 0 (this is the root inode allocation) use newino.
1589 	 * This must work because we've just allocated some.
1590 	 */
1591 	if (!pagino)
1592 		pagino = be32_to_cpu(agi->agi_newino);
1593 
1594 	cur = xfs_finobt_init_cursor(pag, tp, agbp);
1595 
1596 	error = xfs_check_agi_freecount(cur);
1597 	if (error)
1598 		goto error_cur;
1599 
1600 	/*
1601 	 * The search algorithm depends on whether we're in the same AG as the
1602 	 * parent. If so, find the closest available inode to the parent. If
1603 	 * not, consider the agi hint or find the first free inode in the AG.
1604 	 */
1605 	if (pag_agno(pag) == pagno)
1606 		error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1607 	else
1608 		error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1609 	if (error)
1610 		goto error_cur;
1611 
1612 	offset = xfs_inobt_first_free_inode(&rec);
1613 	ASSERT(offset >= 0);
1614 	ASSERT(offset < XFS_INODES_PER_CHUNK);
1615 	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1616 				   XFS_INODES_PER_CHUNK) == 0);
1617 	ino = xfs_agino_to_ino(pag, rec.ir_startino + offset);
1618 
1619 	if (xfs_ag_has_sickness(pag, XFS_SICK_AG_INODES)) {
1620 		error = xfs_dialloc_check_ino(pag, tp, ino);
1621 		if (error)
1622 			goto error_cur;
1623 	}
1624 
1625 	/*
1626 	 * Modify or remove the finobt record.
1627 	 */
1628 	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1629 	rec.ir_freecount--;
1630 	if (rec.ir_freecount)
1631 		error = xfs_inobt_update(cur, &rec);
1632 	else
1633 		error = xfs_btree_delete(cur, &i);
1634 	if (error)
1635 		goto error_cur;
1636 
1637 	/*
1638 	 * The finobt has now been updated appropriately. We haven't updated the
1639 	 * agi and superblock yet, so we can create an inobt cursor and validate
1640 	 * the original freecount. If all is well, make the equivalent update to
1641 	 * the inobt using the finobt record and offset information.
1642 	 */
1643 	icur = xfs_inobt_init_cursor(pag, tp, agbp);
1644 
1645 	error = xfs_check_agi_freecount(icur);
1646 	if (error)
1647 		goto error_icur;
1648 
1649 	error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1650 	if (error)
1651 		goto error_icur;
1652 
1653 	/*
1654 	 * Both trees have now been updated. We must update the perag and
1655 	 * superblock before we can check the freecount for each btree.
1656 	 */
1657 	be32_add_cpu(&agi->agi_freecount, -1);
1658 	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1659 	pag->pagi_freecount--;
1660 
1661 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1662 
1663 	error = xfs_check_agi_freecount(icur);
1664 	if (error)
1665 		goto error_icur;
1666 	error = xfs_check_agi_freecount(cur);
1667 	if (error)
1668 		goto error_icur;
1669 
1670 	xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1671 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1672 	*inop = ino;
1673 	return 0;
1674 
1675 error_icur:
1676 	xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1677 error_cur:
1678 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1679 	return error;
1680 }
1681 
1682 static int
xfs_dialloc_roll(struct xfs_trans ** tpp,struct xfs_buf * agibp)1683 xfs_dialloc_roll(
1684 	struct xfs_trans	**tpp,
1685 	struct xfs_buf		*agibp)
1686 {
1687 	struct xfs_trans	*tp = *tpp;
1688 	struct xfs_dquot_acct	*dqinfo;
1689 	int			error;
1690 
1691 	/*
1692 	 * Hold to on to the agibp across the commit so no other allocation can
1693 	 * come in and take the free inodes we just allocated for our caller.
1694 	 */
1695 	xfs_trans_bhold(tp, agibp);
1696 
1697 	/*
1698 	 * We want the quota changes to be associated with the next transaction,
1699 	 * NOT this one. So, detach the dqinfo from this and attach it to the
1700 	 * next transaction.
1701 	 */
1702 	dqinfo = tp->t_dqinfo;
1703 	tp->t_dqinfo = NULL;
1704 
1705 	error = xfs_trans_roll(&tp);
1706 
1707 	/* Re-attach the quota info that we detached from prev trx. */
1708 	tp->t_dqinfo = dqinfo;
1709 
1710 	/*
1711 	 * Join the buffer even on commit error so that the buffer is released
1712 	 * when the caller cancels the transaction and doesn't have to handle
1713 	 * this error case specially.
1714 	 */
1715 	xfs_trans_bjoin(tp, agibp);
1716 	*tpp = tp;
1717 	return error;
1718 }
1719 
1720 static bool
xfs_dialloc_good_ag(struct xfs_perag * pag,struct xfs_trans * tp,umode_t mode,int flags,bool ok_alloc)1721 xfs_dialloc_good_ag(
1722 	struct xfs_perag	*pag,
1723 	struct xfs_trans	*tp,
1724 	umode_t			mode,
1725 	int			flags,
1726 	bool			ok_alloc)
1727 {
1728 	struct xfs_mount	*mp = tp->t_mountp;
1729 	xfs_extlen_t		ineed;
1730 	xfs_extlen_t		longest = 0;
1731 	int			needspace;
1732 	int			error;
1733 
1734 	if (!pag)
1735 		return false;
1736 	if (!xfs_perag_allows_inodes(pag))
1737 		return false;
1738 
1739 	if (!xfs_perag_initialised_agi(pag)) {
1740 		error = xfs_ialloc_read_agi(pag, tp, 0, NULL);
1741 		if (error)
1742 			return false;
1743 	}
1744 
1745 	if (pag->pagi_freecount)
1746 		return true;
1747 	if (!ok_alloc)
1748 		return false;
1749 
1750 	if (!xfs_perag_initialised_agf(pag)) {
1751 		error = xfs_alloc_read_agf(pag, tp, flags, NULL);
1752 		if (error)
1753 			return false;
1754 	}
1755 
1756 	/*
1757 	 * Check that there is enough free space for the file plus a chunk of
1758 	 * inodes if we need to allocate some. If this is the first pass across
1759 	 * the AGs, take into account the potential space needed for alignment
1760 	 * of inode chunks when checking the longest contiguous free space in
1761 	 * the AG - this prevents us from getting ENOSPC because we have free
1762 	 * space larger than ialloc_blks but alignment constraints prevent us
1763 	 * from using it.
1764 	 *
1765 	 * If we can't find an AG with space for full alignment slack to be
1766 	 * taken into account, we must be near ENOSPC in all AGs.  Hence we
1767 	 * don't include alignment for the second pass and so if we fail
1768 	 * allocation due to alignment issues then it is most likely a real
1769 	 * ENOSPC condition.
1770 	 *
1771 	 * XXX(dgc): this calculation is now bogus thanks to the per-ag
1772 	 * reservations that xfs_alloc_fix_freelist() now does via
1773 	 * xfs_alloc_space_available(). When the AG fills up, pagf_freeblks will
1774 	 * be more than large enough for the check below to succeed, but
1775 	 * xfs_alloc_space_available() will fail because of the non-zero
1776 	 * metadata reservation and hence we won't actually be able to allocate
1777 	 * more inodes in this AG. We do soooo much unnecessary work near ENOSPC
1778 	 * because of this.
1779 	 */
1780 	ineed = M_IGEO(mp)->ialloc_min_blks;
1781 	if (flags && ineed > 1)
1782 		ineed += M_IGEO(mp)->cluster_align;
1783 	longest = pag->pagf_longest;
1784 	if (!longest)
1785 		longest = pag->pagf_flcount > 0;
1786 	needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
1787 
1788 	if (pag->pagf_freeblks < needspace + ineed || longest < ineed)
1789 		return false;
1790 	return true;
1791 }
1792 
1793 static int
xfs_dialloc_try_ag(struct xfs_perag * pag,struct xfs_trans ** tpp,xfs_ino_t parent,xfs_ino_t * new_ino,bool ok_alloc)1794 xfs_dialloc_try_ag(
1795 	struct xfs_perag	*pag,
1796 	struct xfs_trans	**tpp,
1797 	xfs_ino_t		parent,
1798 	xfs_ino_t		*new_ino,
1799 	bool			ok_alloc)
1800 {
1801 	struct xfs_buf		*agbp;
1802 	xfs_ino_t		ino;
1803 	int			error;
1804 
1805 	/*
1806 	 * Then read in the AGI buffer and recheck with the AGI buffer
1807 	 * lock held.
1808 	 */
1809 	error = xfs_ialloc_read_agi(pag, *tpp, 0, &agbp);
1810 	if (error)
1811 		return error;
1812 
1813 	if (!pag->pagi_freecount) {
1814 		if (!ok_alloc) {
1815 			error = -EAGAIN;
1816 			goto out_release;
1817 		}
1818 
1819 		error = xfs_ialloc_ag_alloc(pag, *tpp, agbp);
1820 		if (error < 0)
1821 			goto out_release;
1822 
1823 		/*
1824 		 * We successfully allocated space for an inode cluster in this
1825 		 * AG.  Roll the transaction so that we can allocate one of the
1826 		 * new inodes.
1827 		 */
1828 		ASSERT(pag->pagi_freecount > 0);
1829 		error = xfs_dialloc_roll(tpp, agbp);
1830 		if (error)
1831 			goto out_release;
1832 	}
1833 
1834 	/* Allocate an inode in the found AG */
1835 	error = xfs_dialloc_ag(pag, *tpp, agbp, parent, &ino);
1836 	if (!error)
1837 		*new_ino = ino;
1838 	return error;
1839 
1840 out_release:
1841 	xfs_trans_brelse(*tpp, agbp);
1842 	return error;
1843 }
1844 
1845 /*
1846  * Pick an AG for the new inode.
1847  *
1848  * Directories, symlinks, and regular files frequently allocate at least one
1849  * block, so factor that potential expansion when we examine whether an AG has
1850  * enough space for file creation.  Try to keep metadata files all in the same
1851  * AG.
1852  */
1853 static inline xfs_agnumber_t
xfs_dialloc_pick_ag(struct xfs_mount * mp,struct xfs_inode * dp,umode_t mode)1854 xfs_dialloc_pick_ag(
1855 	struct xfs_mount	*mp,
1856 	struct xfs_inode	*dp,
1857 	umode_t			mode)
1858 {
1859 	xfs_agnumber_t		start_agno;
1860 
1861 	if (!dp)
1862 		return 0;
1863 	if (xfs_is_metadir_inode(dp)) {
1864 		if (mp->m_sb.sb_logstart)
1865 			return XFS_FSB_TO_AGNO(mp, mp->m_sb.sb_logstart);
1866 		return 0;
1867 	}
1868 
1869 	if (S_ISDIR(mode))
1870 		return (atomic_inc_return(&mp->m_agirotor) - 1) % mp->m_maxagi;
1871 
1872 	start_agno = XFS_INO_TO_AGNO(mp, dp->i_ino);
1873 	if (start_agno >= mp->m_maxagi)
1874 		start_agno = 0;
1875 
1876 	return start_agno;
1877 }
1878 
1879 /*
1880  * Allocate an on-disk inode.
1881  *
1882  * Mode is used to tell whether the new inode is a directory and hence where to
1883  * locate it. The on-disk inode that is allocated will be returned in @new_ino
1884  * on success, otherwise an error will be set to indicate the failure (e.g.
1885  * -ENOSPC).
1886  */
1887 int
xfs_dialloc(struct xfs_trans ** tpp,const struct xfs_icreate_args * args,xfs_ino_t * new_ino)1888 xfs_dialloc(
1889 	struct xfs_trans	**tpp,
1890 	const struct xfs_icreate_args *args,
1891 	xfs_ino_t		*new_ino)
1892 {
1893 	struct xfs_mount	*mp = (*tpp)->t_mountp;
1894 	struct xfs_perag	*pag;
1895 	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
1896 	xfs_ino_t		ino = NULLFSINO;
1897 	xfs_ino_t		parent = args->pip ? args->pip->i_ino : 0;
1898 	xfs_agnumber_t		agno;
1899 	xfs_agnumber_t		start_agno;
1900 	umode_t			mode = args->mode & S_IFMT;
1901 	bool			ok_alloc = true;
1902 	bool			low_space = false;
1903 	int			flags;
1904 	int			error = 0;
1905 
1906 	start_agno = xfs_dialloc_pick_ag(mp, args->pip, mode);
1907 
1908 	/*
1909 	 * If we have already hit the ceiling of inode blocks then clear
1910 	 * ok_alloc so we scan all available agi structures for a free
1911 	 * inode.
1912 	 *
1913 	 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1914 	 * which will sacrifice the preciseness but improve the performance.
1915 	 */
1916 	if (igeo->maxicount &&
1917 	    percpu_counter_read_positive(&mp->m_icount) + igeo->ialloc_inos
1918 							> igeo->maxicount) {
1919 		ok_alloc = false;
1920 	}
1921 
1922 	/*
1923 	 * If we are near to ENOSPC, we want to prefer allocation from AGs that
1924 	 * have free inodes in them rather than use up free space allocating new
1925 	 * inode chunks. Hence we turn off allocation for the first non-blocking
1926 	 * pass through the AGs if we are near ENOSPC to consume free inodes
1927 	 * that we can immediately allocate, but then we allow allocation on the
1928 	 * second pass if we fail to find an AG with free inodes in it.
1929 	 */
1930 	if (percpu_counter_read_positive(&mp->m_fdblocks) <
1931 			mp->m_low_space[XFS_LOWSP_1_PCNT]) {
1932 		ok_alloc = false;
1933 		low_space = true;
1934 	}
1935 
1936 	/*
1937 	 * Loop until we find an allocation group that either has free inodes
1938 	 * or in which we can allocate some inodes.  Iterate through the
1939 	 * allocation groups upward, wrapping at the end.
1940 	 */
1941 	flags = XFS_ALLOC_FLAG_TRYLOCK;
1942 retry:
1943 	for_each_perag_wrap_at(mp, start_agno, mp->m_maxagi, agno, pag) {
1944 		if (xfs_dialloc_good_ag(pag, *tpp, mode, flags, ok_alloc)) {
1945 			error = xfs_dialloc_try_ag(pag, tpp, parent,
1946 					&ino, ok_alloc);
1947 			if (error != -EAGAIN)
1948 				break;
1949 			error = 0;
1950 		}
1951 
1952 		if (xfs_is_shutdown(mp)) {
1953 			error = -EFSCORRUPTED;
1954 			break;
1955 		}
1956 	}
1957 	if (pag)
1958 		xfs_perag_rele(pag);
1959 	if (error)
1960 		return error;
1961 	if (ino == NULLFSINO) {
1962 		if (flags) {
1963 			flags = 0;
1964 			if (low_space)
1965 				ok_alloc = true;
1966 			goto retry;
1967 		}
1968 		return -ENOSPC;
1969 	}
1970 
1971 	/*
1972 	 * Protect against obviously corrupt allocation btree records. Later
1973 	 * xfs_iget checks will catch re-allocation of other active in-memory
1974 	 * and on-disk inodes. If we don't catch reallocating the parent inode
1975 	 * here we will deadlock in xfs_iget() so we have to do these checks
1976 	 * first.
1977 	 */
1978 	if (ino == parent || !xfs_verify_dir_ino(mp, ino)) {
1979 		xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
1980 		xfs_agno_mark_sick(mp, XFS_INO_TO_AGNO(mp, ino),
1981 				XFS_SICK_AG_INOBT);
1982 		return -EFSCORRUPTED;
1983 	}
1984 
1985 	*new_ino = ino;
1986 	return 0;
1987 }
1988 
1989 /*
1990  * Free the blocks of an inode chunk. We must consider that the inode chunk
1991  * might be sparse and only free the regions that are allocated as part of the
1992  * chunk.
1993  */
1994 static int
xfs_difree_inode_chunk(struct xfs_trans * tp,struct xfs_perag * pag,struct xfs_inobt_rec_incore * rec)1995 xfs_difree_inode_chunk(
1996 	struct xfs_trans		*tp,
1997 	struct xfs_perag		*pag,
1998 	struct xfs_inobt_rec_incore	*rec)
1999 {
2000 	struct xfs_mount		*mp = tp->t_mountp;
2001 	xfs_agblock_t			sagbno = XFS_AGINO_TO_AGBNO(mp,
2002 							rec->ir_startino);
2003 	int				startidx, endidx;
2004 	int				nextbit;
2005 	xfs_agblock_t			agbno;
2006 	int				contigblk;
2007 	DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
2008 
2009 	if (!xfs_inobt_issparse(rec->ir_holemask)) {
2010 		/* not sparse, calculate extent info directly */
2011 		return xfs_free_extent_later(tp, xfs_agbno_to_fsb(pag, sagbno),
2012 				M_IGEO(mp)->ialloc_blks, &XFS_RMAP_OINFO_INODES,
2013 				XFS_AG_RESV_NONE, 0);
2014 	}
2015 
2016 	/* holemask is only 16-bits (fits in an unsigned long) */
2017 	ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
2018 	holemask[0] = rec->ir_holemask;
2019 
2020 	/*
2021 	 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
2022 	 * holemask and convert the start/end index of each range to an extent.
2023 	 * We start with the start and end index both pointing at the first 0 in
2024 	 * the mask.
2025 	 */
2026 	startidx = endidx = find_first_zero_bit(holemask,
2027 						XFS_INOBT_HOLEMASK_BITS);
2028 	nextbit = startidx + 1;
2029 	while (startidx < XFS_INOBT_HOLEMASK_BITS) {
2030 		int error;
2031 
2032 		nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
2033 					     nextbit);
2034 		/*
2035 		 * If the next zero bit is contiguous, update the end index of
2036 		 * the current range and continue.
2037 		 */
2038 		if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
2039 		    nextbit == endidx + 1) {
2040 			endidx = nextbit;
2041 			goto next;
2042 		}
2043 
2044 		/*
2045 		 * nextbit is not contiguous with the current end index. Convert
2046 		 * the current start/end to an extent and add it to the free
2047 		 * list.
2048 		 */
2049 		agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
2050 				  mp->m_sb.sb_inopblock;
2051 		contigblk = ((endidx - startidx + 1) *
2052 			     XFS_INODES_PER_HOLEMASK_BIT) /
2053 			    mp->m_sb.sb_inopblock;
2054 
2055 		ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
2056 		ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
2057 		error = xfs_free_extent_later(tp, xfs_agbno_to_fsb(pag, agbno),
2058 				contigblk, &XFS_RMAP_OINFO_INODES,
2059 				XFS_AG_RESV_NONE, 0);
2060 		if (error)
2061 			return error;
2062 
2063 		/* reset range to current bit and carry on... */
2064 		startidx = endidx = nextbit;
2065 
2066 next:
2067 		nextbit++;
2068 	}
2069 	return 0;
2070 }
2071 
2072 STATIC int
xfs_difree_inobt(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp,xfs_agino_t agino,struct xfs_icluster * xic,struct xfs_inobt_rec_incore * orec)2073 xfs_difree_inobt(
2074 	struct xfs_perag		*pag,
2075 	struct xfs_trans		*tp,
2076 	struct xfs_buf			*agbp,
2077 	xfs_agino_t			agino,
2078 	struct xfs_icluster		*xic,
2079 	struct xfs_inobt_rec_incore	*orec)
2080 {
2081 	struct xfs_mount		*mp = pag_mount(pag);
2082 	struct xfs_agi			*agi = agbp->b_addr;
2083 	struct xfs_btree_cur		*cur;
2084 	struct xfs_inobt_rec_incore	rec;
2085 	int				ilen;
2086 	int				error;
2087 	int				i;
2088 	int				off;
2089 
2090 	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2091 	ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
2092 
2093 	/*
2094 	 * Initialize the cursor.
2095 	 */
2096 	cur = xfs_inobt_init_cursor(pag, tp, agbp);
2097 
2098 	error = xfs_check_agi_freecount(cur);
2099 	if (error)
2100 		goto error0;
2101 
2102 	/*
2103 	 * Look for the entry describing this inode.
2104 	 */
2105 	if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
2106 		xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
2107 			__func__, error);
2108 		goto error0;
2109 	}
2110 	if (XFS_IS_CORRUPT(mp, i != 1)) {
2111 		xfs_btree_mark_sick(cur);
2112 		error = -EFSCORRUPTED;
2113 		goto error0;
2114 	}
2115 	error = xfs_inobt_get_rec(cur, &rec, &i);
2116 	if (error) {
2117 		xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
2118 			__func__, error);
2119 		goto error0;
2120 	}
2121 	if (XFS_IS_CORRUPT(mp, i != 1)) {
2122 		xfs_btree_mark_sick(cur);
2123 		error = -EFSCORRUPTED;
2124 		goto error0;
2125 	}
2126 	/*
2127 	 * Get the offset in the inode chunk.
2128 	 */
2129 	off = agino - rec.ir_startino;
2130 	ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
2131 	ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
2132 	/*
2133 	 * Mark the inode free & increment the count.
2134 	 */
2135 	rec.ir_free |= XFS_INOBT_MASK(off);
2136 	rec.ir_freecount++;
2137 
2138 	/*
2139 	 * When an inode chunk is free, it becomes eligible for removal. Don't
2140 	 * remove the chunk if the block size is large enough for multiple inode
2141 	 * chunks (that might not be free).
2142 	 */
2143 	if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE &&
2144 	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
2145 		xic->deleted = true;
2146 		xic->first_ino = xfs_agino_to_ino(pag, rec.ir_startino);
2147 		xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
2148 
2149 		/*
2150 		 * Remove the inode cluster from the AGI B+Tree, adjust the
2151 		 * AGI and Superblock inode counts, and mark the disk space
2152 		 * to be freed when the transaction is committed.
2153 		 */
2154 		ilen = rec.ir_freecount;
2155 		be32_add_cpu(&agi->agi_count, -ilen);
2156 		be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
2157 		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
2158 		pag->pagi_freecount -= ilen - 1;
2159 		pag->pagi_count -= ilen;
2160 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
2161 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
2162 
2163 		if ((error = xfs_btree_delete(cur, &i))) {
2164 			xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
2165 				__func__, error);
2166 			goto error0;
2167 		}
2168 
2169 		error = xfs_difree_inode_chunk(tp, pag, &rec);
2170 		if (error)
2171 			goto error0;
2172 	} else {
2173 		xic->deleted = false;
2174 
2175 		error = xfs_inobt_update(cur, &rec);
2176 		if (error) {
2177 			xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
2178 				__func__, error);
2179 			goto error0;
2180 		}
2181 
2182 		/*
2183 		 * Change the inode free counts and log the ag/sb changes.
2184 		 */
2185 		be32_add_cpu(&agi->agi_freecount, 1);
2186 		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
2187 		pag->pagi_freecount++;
2188 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2189 	}
2190 
2191 	error = xfs_check_agi_freecount(cur);
2192 	if (error)
2193 		goto error0;
2194 
2195 	*orec = rec;
2196 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2197 	return 0;
2198 
2199 error0:
2200 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2201 	return error;
2202 }
2203 
2204 /*
2205  * Free an inode in the free inode btree.
2206  */
2207 STATIC int
xfs_difree_finobt(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agbp,xfs_agino_t agino,struct xfs_inobt_rec_incore * ibtrec)2208 xfs_difree_finobt(
2209 	struct xfs_perag		*pag,
2210 	struct xfs_trans		*tp,
2211 	struct xfs_buf			*agbp,
2212 	xfs_agino_t			agino,
2213 	struct xfs_inobt_rec_incore	*ibtrec) /* inobt record */
2214 {
2215 	struct xfs_mount		*mp = pag_mount(pag);
2216 	struct xfs_btree_cur		*cur;
2217 	struct xfs_inobt_rec_incore	rec;
2218 	int				offset = agino - ibtrec->ir_startino;
2219 	int				error;
2220 	int				i;
2221 
2222 	cur = xfs_finobt_init_cursor(pag, tp, agbp);
2223 
2224 	error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2225 	if (error)
2226 		goto error;
2227 	if (i == 0) {
2228 		/*
2229 		 * If the record does not exist in the finobt, we must have just
2230 		 * freed an inode in a previously fully allocated chunk. If not,
2231 		 * something is out of sync.
2232 		 */
2233 		if (XFS_IS_CORRUPT(mp, ibtrec->ir_freecount != 1)) {
2234 			xfs_btree_mark_sick(cur);
2235 			error = -EFSCORRUPTED;
2236 			goto error;
2237 		}
2238 
2239 		error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2240 					     ibtrec->ir_count,
2241 					     ibtrec->ir_freecount,
2242 					     ibtrec->ir_free, &i);
2243 		if (error)
2244 			goto error;
2245 		ASSERT(i == 1);
2246 
2247 		goto out;
2248 	}
2249 
2250 	/*
2251 	 * Read and update the existing record. We could just copy the ibtrec
2252 	 * across here, but that would defeat the purpose of having redundant
2253 	 * metadata. By making the modifications independently, we can catch
2254 	 * corruptions that we wouldn't see if we just copied from one record
2255 	 * to another.
2256 	 */
2257 	error = xfs_inobt_get_rec(cur, &rec, &i);
2258 	if (error)
2259 		goto error;
2260 	if (XFS_IS_CORRUPT(mp, i != 1)) {
2261 		xfs_btree_mark_sick(cur);
2262 		error = -EFSCORRUPTED;
2263 		goto error;
2264 	}
2265 
2266 	rec.ir_free |= XFS_INOBT_MASK(offset);
2267 	rec.ir_freecount++;
2268 
2269 	if (XFS_IS_CORRUPT(mp,
2270 			   rec.ir_free != ibtrec->ir_free ||
2271 			   rec.ir_freecount != ibtrec->ir_freecount)) {
2272 		xfs_btree_mark_sick(cur);
2273 		error = -EFSCORRUPTED;
2274 		goto error;
2275 	}
2276 
2277 	/*
2278 	 * The content of inobt records should always match between the inobt
2279 	 * and finobt. The lifecycle of records in the finobt is different from
2280 	 * the inobt in that the finobt only tracks records with at least one
2281 	 * free inode. Hence, if all of the inodes are free and we aren't
2282 	 * keeping inode chunks permanently on disk, remove the record.
2283 	 * Otherwise, update the record with the new information.
2284 	 *
2285 	 * Note that we currently can't free chunks when the block size is large
2286 	 * enough for multiple chunks. Leave the finobt record to remain in sync
2287 	 * with the inobt.
2288 	 */
2289 	if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE &&
2290 	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
2291 		error = xfs_btree_delete(cur, &i);
2292 		if (error)
2293 			goto error;
2294 		ASSERT(i == 1);
2295 	} else {
2296 		error = xfs_inobt_update(cur, &rec);
2297 		if (error)
2298 			goto error;
2299 	}
2300 
2301 out:
2302 	error = xfs_check_agi_freecount(cur);
2303 	if (error)
2304 		goto error;
2305 
2306 	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2307 	return 0;
2308 
2309 error:
2310 	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2311 	return error;
2312 }
2313 
2314 /*
2315  * Free disk inode.  Carefully avoids touching the incore inode, all
2316  * manipulations incore are the caller's responsibility.
2317  * The on-disk inode is not changed by this operation, only the
2318  * btree (free inode mask) is changed.
2319  */
2320 int
xfs_difree(struct xfs_trans * tp,struct xfs_perag * pag,xfs_ino_t inode,struct xfs_icluster * xic)2321 xfs_difree(
2322 	struct xfs_trans	*tp,
2323 	struct xfs_perag	*pag,
2324 	xfs_ino_t		inode,
2325 	struct xfs_icluster	*xic)
2326 {
2327 	/* REFERENCED */
2328 	xfs_agblock_t		agbno;	/* block number containing inode */
2329 	struct xfs_buf		*agbp;	/* buffer for allocation group header */
2330 	xfs_agino_t		agino;	/* allocation group inode number */
2331 	int			error;	/* error return value */
2332 	struct xfs_mount	*mp = tp->t_mountp;
2333 	struct xfs_inobt_rec_incore rec;/* btree record */
2334 
2335 	/*
2336 	 * Break up inode number into its components.
2337 	 */
2338 	if (pag_agno(pag) != XFS_INO_TO_AGNO(mp, inode)) {
2339 		xfs_warn(mp, "%s: agno != pag_agno(pag) (%d != %d).",
2340 			__func__, XFS_INO_TO_AGNO(mp, inode), pag_agno(pag));
2341 		ASSERT(0);
2342 		return -EINVAL;
2343 	}
2344 	agino = XFS_INO_TO_AGINO(mp, inode);
2345 	if (inode != xfs_agino_to_ino(pag, agino))  {
2346 		xfs_warn(mp, "%s: inode != xfs_agino_to_ino() (%llu != %llu).",
2347 			__func__, (unsigned long long)inode,
2348 			(unsigned long long)xfs_agino_to_ino(pag, agino));
2349 		ASSERT(0);
2350 		return -EINVAL;
2351 	}
2352 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2353 	if (agbno >= xfs_ag_block_count(mp, pag_agno(pag))) {
2354 		xfs_warn(mp, "%s: agbno >= xfs_ag_block_count (%d >= %d).",
2355 			__func__, agbno, xfs_ag_block_count(mp, pag_agno(pag)));
2356 		ASSERT(0);
2357 		return -EINVAL;
2358 	}
2359 	/*
2360 	 * Get the allocation group header.
2361 	 */
2362 	error = xfs_ialloc_read_agi(pag, tp, 0, &agbp);
2363 	if (error) {
2364 		xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2365 			__func__, error);
2366 		return error;
2367 	}
2368 
2369 	/*
2370 	 * Fix up the inode allocation btree.
2371 	 */
2372 	error = xfs_difree_inobt(pag, tp, agbp, agino, xic, &rec);
2373 	if (error)
2374 		goto error0;
2375 
2376 	/*
2377 	 * Fix up the free inode btree.
2378 	 */
2379 	if (xfs_has_finobt(mp)) {
2380 		error = xfs_difree_finobt(pag, tp, agbp, agino, &rec);
2381 		if (error)
2382 			goto error0;
2383 	}
2384 
2385 	return 0;
2386 
2387 error0:
2388 	return error;
2389 }
2390 
2391 STATIC int
xfs_imap_lookup(struct xfs_perag * pag,struct xfs_trans * tp,xfs_agino_t agino,xfs_agblock_t agbno,xfs_agblock_t * chunk_agbno,xfs_agblock_t * offset_agbno,int flags)2392 xfs_imap_lookup(
2393 	struct xfs_perag	*pag,
2394 	struct xfs_trans	*tp,
2395 	xfs_agino_t		agino,
2396 	xfs_agblock_t		agbno,
2397 	xfs_agblock_t		*chunk_agbno,
2398 	xfs_agblock_t		*offset_agbno,
2399 	int			flags)
2400 {
2401 	struct xfs_mount	*mp = pag_mount(pag);
2402 	struct xfs_inobt_rec_incore rec;
2403 	struct xfs_btree_cur	*cur;
2404 	struct xfs_buf		*agbp;
2405 	int			error;
2406 	int			i;
2407 
2408 	error = xfs_ialloc_read_agi(pag, tp, 0, &agbp);
2409 	if (error) {
2410 		xfs_alert(mp,
2411 			"%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2412 			__func__, error, pag_agno(pag));
2413 		return error;
2414 	}
2415 
2416 	/*
2417 	 * Lookup the inode record for the given agino. If the record cannot be
2418 	 * found, then it's an invalid inode number and we should abort. Once
2419 	 * we have a record, we need to ensure it contains the inode number
2420 	 * we are looking up.
2421 	 */
2422 	cur = xfs_inobt_init_cursor(pag, tp, agbp);
2423 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2424 	if (!error) {
2425 		if (i)
2426 			error = xfs_inobt_get_rec(cur, &rec, &i);
2427 		if (!error && i == 0)
2428 			error = -EINVAL;
2429 	}
2430 
2431 	xfs_trans_brelse(tp, agbp);
2432 	xfs_btree_del_cursor(cur, error);
2433 	if (error)
2434 		return error;
2435 
2436 	/* check that the returned record contains the required inode */
2437 	if (rec.ir_startino > agino ||
2438 	    rec.ir_startino + M_IGEO(mp)->ialloc_inos <= agino)
2439 		return -EINVAL;
2440 
2441 	/* for untrusted inodes check it is allocated first */
2442 	if ((flags & XFS_IGET_UNTRUSTED) &&
2443 	    (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2444 		return -EINVAL;
2445 
2446 	*chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2447 	*offset_agbno = agbno - *chunk_agbno;
2448 	return 0;
2449 }
2450 
2451 /*
2452  * Return the location of the inode in imap, for mapping it into a buffer.
2453  */
2454 int
xfs_imap(struct xfs_perag * pag,struct xfs_trans * tp,xfs_ino_t ino,struct xfs_imap * imap,uint flags)2455 xfs_imap(
2456 	struct xfs_perag	*pag,
2457 	struct xfs_trans	*tp,
2458 	xfs_ino_t		ino,	/* inode to locate */
2459 	struct xfs_imap		*imap,	/* location map structure */
2460 	uint			flags)	/* flags for inode btree lookup */
2461 {
2462 	struct xfs_mount	*mp = pag_mount(pag);
2463 	xfs_agblock_t		agbno;	/* block number of inode in the alloc group */
2464 	xfs_agino_t		agino;	/* inode number within alloc group */
2465 	xfs_agblock_t		chunk_agbno;	/* first block in inode chunk */
2466 	xfs_agblock_t		cluster_agbno;	/* first block in inode cluster */
2467 	int			error;	/* error code */
2468 	int			offset;	/* index of inode in its buffer */
2469 	xfs_agblock_t		offset_agbno;	/* blks from chunk start to inode */
2470 
2471 	ASSERT(ino != NULLFSINO);
2472 
2473 	/*
2474 	 * Split up the inode number into its parts.
2475 	 */
2476 	agino = XFS_INO_TO_AGINO(mp, ino);
2477 	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2478 	if (agbno >= xfs_ag_block_count(mp, pag_agno(pag)) ||
2479 	    ino != xfs_agino_to_ino(pag, agino)) {
2480 		error = -EINVAL;
2481 #ifdef DEBUG
2482 		/*
2483 		 * Don't output diagnostic information for untrusted inodes
2484 		 * as they can be invalid without implying corruption.
2485 		 */
2486 		if (flags & XFS_IGET_UNTRUSTED)
2487 			return error;
2488 		if (agbno >= xfs_ag_block_count(mp, pag_agno(pag))) {
2489 			xfs_alert(mp,
2490 		"%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2491 				__func__, (unsigned long long)agbno,
2492 				(unsigned long)xfs_ag_block_count(mp,
2493 							pag_agno(pag)));
2494 		}
2495 		if (ino != xfs_agino_to_ino(pag, agino)) {
2496 			xfs_alert(mp,
2497 		"%s: ino (0x%llx) != xfs_agino_to_ino() (0x%llx)",
2498 				__func__, ino,
2499 				xfs_agino_to_ino(pag, agino));
2500 		}
2501 		xfs_stack_trace();
2502 #endif /* DEBUG */
2503 		return error;
2504 	}
2505 
2506 	/*
2507 	 * For bulkstat and handle lookups, we have an untrusted inode number
2508 	 * that we have to verify is valid. We cannot do this just by reading
2509 	 * the inode buffer as it may have been unlinked and removed leaving
2510 	 * inodes in stale state on disk. Hence we have to do a btree lookup
2511 	 * in all cases where an untrusted inode number is passed.
2512 	 */
2513 	if (flags & XFS_IGET_UNTRUSTED) {
2514 		error = xfs_imap_lookup(pag, tp, agino, agbno,
2515 					&chunk_agbno, &offset_agbno, flags);
2516 		if (error)
2517 			return error;
2518 		goto out_map;
2519 	}
2520 
2521 	/*
2522 	 * If the inode cluster size is the same as the blocksize or
2523 	 * smaller we get to the buffer by simple arithmetics.
2524 	 */
2525 	if (M_IGEO(mp)->blocks_per_cluster == 1) {
2526 		offset = XFS_INO_TO_OFFSET(mp, ino);
2527 		ASSERT(offset < mp->m_sb.sb_inopblock);
2528 
2529 		imap->im_blkno = xfs_agbno_to_daddr(pag, agbno);
2530 		imap->im_len = XFS_FSB_TO_BB(mp, 1);
2531 		imap->im_boffset = (unsigned short)(offset <<
2532 							mp->m_sb.sb_inodelog);
2533 		return 0;
2534 	}
2535 
2536 	/*
2537 	 * If the inode chunks are aligned then use simple maths to
2538 	 * find the location. Otherwise we have to do a btree
2539 	 * lookup to find the location.
2540 	 */
2541 	if (M_IGEO(mp)->inoalign_mask) {
2542 		offset_agbno = agbno & M_IGEO(mp)->inoalign_mask;
2543 		chunk_agbno = agbno - offset_agbno;
2544 	} else {
2545 		error = xfs_imap_lookup(pag, tp, agino, agbno,
2546 					&chunk_agbno, &offset_agbno, flags);
2547 		if (error)
2548 			return error;
2549 	}
2550 
2551 out_map:
2552 	ASSERT(agbno >= chunk_agbno);
2553 	cluster_agbno = chunk_agbno +
2554 		((offset_agbno / M_IGEO(mp)->blocks_per_cluster) *
2555 		 M_IGEO(mp)->blocks_per_cluster);
2556 	offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2557 		XFS_INO_TO_OFFSET(mp, ino);
2558 
2559 	imap->im_blkno = xfs_agbno_to_daddr(pag, cluster_agbno);
2560 	imap->im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
2561 	imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2562 
2563 	/*
2564 	 * If the inode number maps to a block outside the bounds
2565 	 * of the file system then return NULL rather than calling
2566 	 * read_buf and panicing when we get an error from the
2567 	 * driver.
2568 	 */
2569 	if ((imap->im_blkno + imap->im_len) >
2570 	    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2571 		xfs_alert(mp,
2572 	"%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2573 			__func__, (unsigned long long) imap->im_blkno,
2574 			(unsigned long long) imap->im_len,
2575 			XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2576 		return -EINVAL;
2577 	}
2578 	return 0;
2579 }
2580 
2581 /*
2582  * Log specified fields for the ag hdr (inode section). The growth of the agi
2583  * structure over time requires that we interpret the buffer as two logical
2584  * regions delineated by the end of the unlinked list. This is due to the size
2585  * of the hash table and its location in the middle of the agi.
2586  *
2587  * For example, a request to log a field before agi_unlinked and a field after
2588  * agi_unlinked could cause us to log the entire hash table and use an excessive
2589  * amount of log space. To avoid this behavior, log the region up through
2590  * agi_unlinked in one call and the region after agi_unlinked through the end of
2591  * the structure in another.
2592  */
2593 void
xfs_ialloc_log_agi(struct xfs_trans * tp,struct xfs_buf * bp,uint32_t fields)2594 xfs_ialloc_log_agi(
2595 	struct xfs_trans	*tp,
2596 	struct xfs_buf		*bp,
2597 	uint32_t		fields)
2598 {
2599 	int			first;		/* first byte number */
2600 	int			last;		/* last byte number */
2601 	static const short	offsets[] = {	/* field starting offsets */
2602 					/* keep in sync with bit definitions */
2603 		offsetof(xfs_agi_t, agi_magicnum),
2604 		offsetof(xfs_agi_t, agi_versionnum),
2605 		offsetof(xfs_agi_t, agi_seqno),
2606 		offsetof(xfs_agi_t, agi_length),
2607 		offsetof(xfs_agi_t, agi_count),
2608 		offsetof(xfs_agi_t, agi_root),
2609 		offsetof(xfs_agi_t, agi_level),
2610 		offsetof(xfs_agi_t, agi_freecount),
2611 		offsetof(xfs_agi_t, agi_newino),
2612 		offsetof(xfs_agi_t, agi_dirino),
2613 		offsetof(xfs_agi_t, agi_unlinked),
2614 		offsetof(xfs_agi_t, agi_free_root),
2615 		offsetof(xfs_agi_t, agi_free_level),
2616 		offsetof(xfs_agi_t, agi_iblocks),
2617 		sizeof(xfs_agi_t)
2618 	};
2619 #ifdef DEBUG
2620 	struct xfs_agi		*agi = bp->b_addr;
2621 
2622 	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2623 #endif
2624 
2625 	/*
2626 	 * Compute byte offsets for the first and last fields in the first
2627 	 * region and log the agi buffer. This only logs up through
2628 	 * agi_unlinked.
2629 	 */
2630 	if (fields & XFS_AGI_ALL_BITS_R1) {
2631 		xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2632 				  &first, &last);
2633 		xfs_trans_log_buf(tp, bp, first, last);
2634 	}
2635 
2636 	/*
2637 	 * Mask off the bits in the first region and calculate the first and
2638 	 * last field offsets for any bits in the second region.
2639 	 */
2640 	fields &= ~XFS_AGI_ALL_BITS_R1;
2641 	if (fields) {
2642 		xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2643 				  &first, &last);
2644 		xfs_trans_log_buf(tp, bp, first, last);
2645 	}
2646 }
2647 
2648 static xfs_failaddr_t
xfs_agi_verify(struct xfs_buf * bp)2649 xfs_agi_verify(
2650 	struct xfs_buf		*bp)
2651 {
2652 	struct xfs_mount	*mp = bp->b_mount;
2653 	struct xfs_agi		*agi = bp->b_addr;
2654 	xfs_failaddr_t		fa;
2655 	uint32_t		agi_seqno = be32_to_cpu(agi->agi_seqno);
2656 	uint32_t		agi_length = be32_to_cpu(agi->agi_length);
2657 	int			i;
2658 
2659 	if (xfs_has_crc(mp)) {
2660 		if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2661 			return __this_address;
2662 		if (!xfs_log_check_lsn(mp, be64_to_cpu(agi->agi_lsn)))
2663 			return __this_address;
2664 	}
2665 
2666 	/*
2667 	 * Validate the magic number of the agi block.
2668 	 */
2669 	if (!xfs_verify_magic(bp, agi->agi_magicnum))
2670 		return __this_address;
2671 	if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2672 		return __this_address;
2673 
2674 	fa = xfs_validate_ag_length(bp, agi_seqno, agi_length);
2675 	if (fa)
2676 		return fa;
2677 
2678 	if (be32_to_cpu(agi->agi_level) < 1 ||
2679 	    be32_to_cpu(agi->agi_level) > M_IGEO(mp)->inobt_maxlevels)
2680 		return __this_address;
2681 
2682 	if (xfs_has_finobt(mp) &&
2683 	    (be32_to_cpu(agi->agi_free_level) < 1 ||
2684 	     be32_to_cpu(agi->agi_free_level) > M_IGEO(mp)->inobt_maxlevels))
2685 		return __this_address;
2686 
2687 	for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
2688 		if (agi->agi_unlinked[i] == cpu_to_be32(NULLAGINO))
2689 			continue;
2690 		if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i])))
2691 			return __this_address;
2692 	}
2693 
2694 	return NULL;
2695 }
2696 
2697 static void
xfs_agi_read_verify(struct xfs_buf * bp)2698 xfs_agi_read_verify(
2699 	struct xfs_buf	*bp)
2700 {
2701 	struct xfs_mount *mp = bp->b_mount;
2702 	xfs_failaddr_t	fa;
2703 
2704 	if (xfs_has_crc(mp) &&
2705 	    !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2706 		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2707 	else {
2708 		fa = xfs_agi_verify(bp);
2709 		if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI))
2710 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2711 	}
2712 }
2713 
2714 static void
xfs_agi_write_verify(struct xfs_buf * bp)2715 xfs_agi_write_verify(
2716 	struct xfs_buf	*bp)
2717 {
2718 	struct xfs_mount	*mp = bp->b_mount;
2719 	struct xfs_buf_log_item	*bip = bp->b_log_item;
2720 	struct xfs_agi		*agi = bp->b_addr;
2721 	xfs_failaddr_t		fa;
2722 
2723 	fa = xfs_agi_verify(bp);
2724 	if (fa) {
2725 		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2726 		return;
2727 	}
2728 
2729 	if (!xfs_has_crc(mp))
2730 		return;
2731 
2732 	if (bip)
2733 		agi->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2734 	xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2735 }
2736 
2737 const struct xfs_buf_ops xfs_agi_buf_ops = {
2738 	.name = "xfs_agi",
2739 	.magic = { cpu_to_be32(XFS_AGI_MAGIC), cpu_to_be32(XFS_AGI_MAGIC) },
2740 	.verify_read = xfs_agi_read_verify,
2741 	.verify_write = xfs_agi_write_verify,
2742 	.verify_struct = xfs_agi_verify,
2743 };
2744 
2745 /*
2746  * Read in the allocation group header (inode allocation section)
2747  */
2748 int
xfs_read_agi(struct xfs_perag * pag,struct xfs_trans * tp,xfs_buf_flags_t flags,struct xfs_buf ** agibpp)2749 xfs_read_agi(
2750 	struct xfs_perag	*pag,
2751 	struct xfs_trans	*tp,
2752 	xfs_buf_flags_t		flags,
2753 	struct xfs_buf		**agibpp)
2754 {
2755 	struct xfs_mount	*mp = pag_mount(pag);
2756 	int			error;
2757 
2758 	trace_xfs_read_agi(pag);
2759 
2760 	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2761 			XFS_AG_DADDR(mp, pag_agno(pag), XFS_AGI_DADDR(mp)),
2762 			XFS_FSS_TO_BB(mp, 1), flags, agibpp, &xfs_agi_buf_ops);
2763 	if (xfs_metadata_is_sick(error))
2764 		xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
2765 	if (error)
2766 		return error;
2767 	if (tp)
2768 		xfs_trans_buf_set_type(tp, *agibpp, XFS_BLFT_AGI_BUF);
2769 
2770 	xfs_buf_set_ref(*agibpp, XFS_AGI_REF);
2771 	return 0;
2772 }
2773 
2774 /*
2775  * Read in the agi and initialise the per-ag data. If the caller supplies a
2776  * @agibpp, return the locked AGI buffer to them, otherwise release it.
2777  */
2778 int
xfs_ialloc_read_agi(struct xfs_perag * pag,struct xfs_trans * tp,int flags,struct xfs_buf ** agibpp)2779 xfs_ialloc_read_agi(
2780 	struct xfs_perag	*pag,
2781 	struct xfs_trans	*tp,
2782 	int			flags,
2783 	struct xfs_buf		**agibpp)
2784 {
2785 	struct xfs_buf		*agibp;
2786 	struct xfs_agi		*agi;
2787 	int			error;
2788 
2789 	trace_xfs_ialloc_read_agi(pag);
2790 
2791 	error = xfs_read_agi(pag, tp,
2792 			(flags & XFS_IALLOC_FLAG_TRYLOCK) ? XBF_TRYLOCK : 0,
2793 			&agibp);
2794 	if (error)
2795 		return error;
2796 
2797 	agi = agibp->b_addr;
2798 	if (!xfs_perag_initialised_agi(pag)) {
2799 		pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2800 		pag->pagi_count = be32_to_cpu(agi->agi_count);
2801 		set_bit(XFS_AGSTATE_AGI_INIT, &pag->pag_opstate);
2802 	}
2803 
2804 	/*
2805 	 * It's possible for these to be out of sync if
2806 	 * we are in the middle of a forced shutdown.
2807 	 */
2808 	ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2809 		xfs_is_shutdown(pag_mount(pag)));
2810 	if (agibpp)
2811 		*agibpp = agibp;
2812 	else
2813 		xfs_trans_brelse(tp, agibp);
2814 	return 0;
2815 }
2816 
2817 /* How many inodes are backed by inode clusters ondisk? */
2818 STATIC int
xfs_ialloc_count_ondisk(struct xfs_btree_cur * cur,xfs_agino_t low,xfs_agino_t high,unsigned int * allocated)2819 xfs_ialloc_count_ondisk(
2820 	struct xfs_btree_cur		*cur,
2821 	xfs_agino_t			low,
2822 	xfs_agino_t			high,
2823 	unsigned int			*allocated)
2824 {
2825 	struct xfs_inobt_rec_incore	irec;
2826 	unsigned int			ret = 0;
2827 	int				has_record;
2828 	int				error;
2829 
2830 	error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record);
2831 	if (error)
2832 		return error;
2833 
2834 	while (has_record) {
2835 		unsigned int		i, hole_idx;
2836 
2837 		error = xfs_inobt_get_rec(cur, &irec, &has_record);
2838 		if (error)
2839 			return error;
2840 		if (irec.ir_startino > high)
2841 			break;
2842 
2843 		for (i = 0; i < XFS_INODES_PER_CHUNK; i++) {
2844 			if (irec.ir_startino + i < low)
2845 				continue;
2846 			if (irec.ir_startino + i > high)
2847 				break;
2848 
2849 			hole_idx = i / XFS_INODES_PER_HOLEMASK_BIT;
2850 			if (!(irec.ir_holemask & (1U << hole_idx)))
2851 				ret++;
2852 		}
2853 
2854 		error = xfs_btree_increment(cur, 0, &has_record);
2855 		if (error)
2856 			return error;
2857 	}
2858 
2859 	*allocated = ret;
2860 	return 0;
2861 }
2862 
2863 /* Is there an inode record covering a given extent? */
2864 int
xfs_ialloc_has_inodes_at_extent(struct xfs_btree_cur * cur,xfs_agblock_t bno,xfs_extlen_t len,enum xbtree_recpacking * outcome)2865 xfs_ialloc_has_inodes_at_extent(
2866 	struct xfs_btree_cur	*cur,
2867 	xfs_agblock_t		bno,
2868 	xfs_extlen_t		len,
2869 	enum xbtree_recpacking	*outcome)
2870 {
2871 	xfs_agino_t		agino;
2872 	xfs_agino_t		last_agino;
2873 	unsigned int		allocated;
2874 	int			error;
2875 
2876 	agino = XFS_AGB_TO_AGINO(cur->bc_mp, bno);
2877 	last_agino = XFS_AGB_TO_AGINO(cur->bc_mp, bno + len) - 1;
2878 
2879 	error = xfs_ialloc_count_ondisk(cur, agino, last_agino, &allocated);
2880 	if (error)
2881 		return error;
2882 
2883 	if (allocated == 0)
2884 		*outcome = XBTREE_RECPACKING_EMPTY;
2885 	else if (allocated == last_agino - agino + 1)
2886 		*outcome = XBTREE_RECPACKING_FULL;
2887 	else
2888 		*outcome = XBTREE_RECPACKING_SPARSE;
2889 	return 0;
2890 }
2891 
2892 struct xfs_ialloc_count_inodes {
2893 	xfs_agino_t			count;
2894 	xfs_agino_t			freecount;
2895 };
2896 
2897 /* Record inode counts across all inobt records. */
2898 STATIC int
xfs_ialloc_count_inodes_rec(struct xfs_btree_cur * cur,const union xfs_btree_rec * rec,void * priv)2899 xfs_ialloc_count_inodes_rec(
2900 	struct xfs_btree_cur		*cur,
2901 	const union xfs_btree_rec	*rec,
2902 	void				*priv)
2903 {
2904 	struct xfs_inobt_rec_incore	irec;
2905 	struct xfs_ialloc_count_inodes	*ci = priv;
2906 	xfs_failaddr_t			fa;
2907 
2908 	xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec);
2909 	fa = xfs_inobt_check_irec(to_perag(cur->bc_group), &irec);
2910 	if (fa)
2911 		return xfs_inobt_complain_bad_rec(cur, fa, &irec);
2912 
2913 	ci->count += irec.ir_count;
2914 	ci->freecount += irec.ir_freecount;
2915 
2916 	return 0;
2917 }
2918 
2919 /* Count allocated and free inodes under an inobt. */
2920 int
xfs_ialloc_count_inodes(struct xfs_btree_cur * cur,xfs_agino_t * count,xfs_agino_t * freecount)2921 xfs_ialloc_count_inodes(
2922 	struct xfs_btree_cur		*cur,
2923 	xfs_agino_t			*count,
2924 	xfs_agino_t			*freecount)
2925 {
2926 	struct xfs_ialloc_count_inodes	ci = {0};
2927 	int				error;
2928 
2929 	ASSERT(xfs_btree_is_ino(cur->bc_ops));
2930 	error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci);
2931 	if (error)
2932 		return error;
2933 
2934 	*count = ci.count;
2935 	*freecount = ci.freecount;
2936 	return 0;
2937 }
2938 
2939 /*
2940  * Initialize inode-related geometry information.
2941  *
2942  * Compute the inode btree min and max levels and set maxicount.
2943  *
2944  * Set the inode cluster size.  This may still be overridden by the file
2945  * system block size if it is larger than the chosen cluster size.
2946  *
2947  * For v5 filesystems, scale the cluster size with the inode size to keep a
2948  * constant ratio of inode per cluster buffer, but only if mkfs has set the
2949  * inode alignment value appropriately for larger cluster sizes.
2950  *
2951  * Then compute the inode cluster alignment information.
2952  */
2953 void
xfs_ialloc_setup_geometry(struct xfs_mount * mp)2954 xfs_ialloc_setup_geometry(
2955 	struct xfs_mount	*mp)
2956 {
2957 	struct xfs_sb		*sbp = &mp->m_sb;
2958 	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
2959 	uint64_t		icount;
2960 	uint			inodes;
2961 
2962 	igeo->new_diflags2 = 0;
2963 	if (xfs_has_bigtime(mp))
2964 		igeo->new_diflags2 |= XFS_DIFLAG2_BIGTIME;
2965 	if (xfs_has_large_extent_counts(mp))
2966 		igeo->new_diflags2 |= XFS_DIFLAG2_NREXT64;
2967 
2968 	/* Compute inode btree geometry. */
2969 	igeo->agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
2970 	igeo->inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, true);
2971 	igeo->inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, false);
2972 	igeo->inobt_mnr[0] = igeo->inobt_mxr[0] / 2;
2973 	igeo->inobt_mnr[1] = igeo->inobt_mxr[1] / 2;
2974 
2975 	igeo->ialloc_inos = max_t(uint16_t, XFS_INODES_PER_CHUNK,
2976 			sbp->sb_inopblock);
2977 	igeo->ialloc_blks = igeo->ialloc_inos >> sbp->sb_inopblog;
2978 
2979 	if (sbp->sb_spino_align)
2980 		igeo->ialloc_min_blks = sbp->sb_spino_align;
2981 	else
2982 		igeo->ialloc_min_blks = igeo->ialloc_blks;
2983 
2984 	/* Compute and fill in value of m_ino_geo.inobt_maxlevels. */
2985 	inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2986 	igeo->inobt_maxlevels = xfs_btree_compute_maxlevels(igeo->inobt_mnr,
2987 			inodes);
2988 	ASSERT(igeo->inobt_maxlevels <= xfs_iallocbt_maxlevels_ondisk());
2989 
2990 	/*
2991 	 * Set the maximum inode count for this filesystem, being careful not
2992 	 * to use obviously garbage sb_inopblog/sb_inopblock values.  Regular
2993 	 * users should never get here due to failing sb verification, but
2994 	 * certain users (xfs_db) need to be usable even with corrupt metadata.
2995 	 */
2996 	if (sbp->sb_imax_pct && igeo->ialloc_blks) {
2997 		/*
2998 		 * Make sure the maximum inode count is a multiple
2999 		 * of the units we allocate inodes in.
3000 		 */
3001 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
3002 		do_div(icount, 100);
3003 		do_div(icount, igeo->ialloc_blks);
3004 		igeo->maxicount = XFS_FSB_TO_INO(mp,
3005 				icount * igeo->ialloc_blks);
3006 	} else {
3007 		igeo->maxicount = 0;
3008 	}
3009 
3010 	/*
3011 	 * Compute the desired size of an inode cluster buffer size, which
3012 	 * starts at 8K and (on v5 filesystems) scales up with larger inode
3013 	 * sizes.
3014 	 *
3015 	 * Preserve the desired inode cluster size because the sparse inodes
3016 	 * feature uses that desired size (not the actual size) to compute the
3017 	 * sparse inode alignment.  The mount code validates this value, so we
3018 	 * cannot change the behavior.
3019 	 */
3020 	igeo->inode_cluster_size_raw = XFS_INODE_BIG_CLUSTER_SIZE;
3021 	if (xfs_has_v3inodes(mp)) {
3022 		int	new_size = igeo->inode_cluster_size_raw;
3023 
3024 		new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
3025 		if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
3026 			igeo->inode_cluster_size_raw = new_size;
3027 	}
3028 
3029 	/* Calculate inode cluster ratios. */
3030 	if (igeo->inode_cluster_size_raw > mp->m_sb.sb_blocksize)
3031 		igeo->blocks_per_cluster = XFS_B_TO_FSBT(mp,
3032 				igeo->inode_cluster_size_raw);
3033 	else
3034 		igeo->blocks_per_cluster = 1;
3035 	igeo->inode_cluster_size = XFS_FSB_TO_B(mp, igeo->blocks_per_cluster);
3036 	igeo->inodes_per_cluster = XFS_FSB_TO_INO(mp, igeo->blocks_per_cluster);
3037 
3038 	/* Calculate inode cluster alignment. */
3039 	if (xfs_has_align(mp) &&
3040 	    mp->m_sb.sb_inoalignmt >= igeo->blocks_per_cluster)
3041 		igeo->cluster_align = mp->m_sb.sb_inoalignmt;
3042 	else
3043 		igeo->cluster_align = 1;
3044 	igeo->inoalign_mask = igeo->cluster_align - 1;
3045 	igeo->cluster_align_inodes = XFS_FSB_TO_INO(mp, igeo->cluster_align);
3046 
3047 	/*
3048 	 * If we are using stripe alignment, check whether
3049 	 * the stripe unit is a multiple of the inode alignment
3050 	 */
3051 	if (mp->m_dalign && igeo->inoalign_mask &&
3052 	    !(mp->m_dalign & igeo->inoalign_mask))
3053 		igeo->ialloc_align = mp->m_dalign;
3054 	else
3055 		igeo->ialloc_align = 0;
3056 
3057 	if (mp->m_sb.sb_blocksize > PAGE_SIZE)
3058 		igeo->min_folio_order = mp->m_sb.sb_blocklog - PAGE_SHIFT;
3059 	else
3060 		igeo->min_folio_order = 0;
3061 }
3062 
3063 /* Compute the location of the root directory inode that is laid out by mkfs. */
3064 xfs_ino_t
xfs_ialloc_calc_rootino(struct xfs_mount * mp,int sunit)3065 xfs_ialloc_calc_rootino(
3066 	struct xfs_mount	*mp,
3067 	int			sunit)
3068 {
3069 	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
3070 	xfs_agblock_t		first_bno;
3071 
3072 	/*
3073 	 * Pre-calculate the geometry of AG 0.  We know what it looks like
3074 	 * because libxfs knows how to create allocation groups now.
3075 	 *
3076 	 * first_bno is the first block in which mkfs could possibly have
3077 	 * allocated the root directory inode, once we factor in the metadata
3078 	 * that mkfs formats before it.  Namely, the four AG headers...
3079 	 */
3080 	first_bno = howmany(4 * mp->m_sb.sb_sectsize, mp->m_sb.sb_blocksize);
3081 
3082 	/* ...the two free space btree roots... */
3083 	first_bno += 2;
3084 
3085 	/* ...the inode btree root... */
3086 	first_bno += 1;
3087 
3088 	/* ...the initial AGFL... */
3089 	first_bno += xfs_alloc_min_freelist(mp, NULL);
3090 
3091 	/* ...the free inode btree root... */
3092 	if (xfs_has_finobt(mp))
3093 		first_bno++;
3094 
3095 	/* ...the reverse mapping btree root... */
3096 	if (xfs_has_rmapbt(mp))
3097 		first_bno++;
3098 
3099 	/* ...the reference count btree... */
3100 	if (xfs_has_reflink(mp))
3101 		first_bno++;
3102 
3103 	/*
3104 	 * ...and the log, if it is allocated in the first allocation group.
3105 	 *
3106 	 * This can happen with filesystems that only have a single
3107 	 * allocation group, or very odd geometries created by old mkfs
3108 	 * versions on very small filesystems.
3109 	 */
3110 	if (xfs_ag_contains_log(mp, 0))
3111 		 first_bno += mp->m_sb.sb_logblocks;
3112 
3113 	/*
3114 	 * Now round first_bno up to whatever allocation alignment is given
3115 	 * by the filesystem or was passed in.
3116 	 */
3117 	if (xfs_has_dalign(mp) && igeo->ialloc_align > 0)
3118 		first_bno = roundup(first_bno, sunit);
3119 	else if (xfs_has_align(mp) &&
3120 			mp->m_sb.sb_inoalignmt > 1)
3121 		first_bno = roundup(first_bno, mp->m_sb.sb_inoalignmt);
3122 
3123 	return XFS_AGINO_TO_INO(mp, 0, XFS_AGB_TO_AGINO(mp, first_bno));
3124 }
3125 
3126 /*
3127  * Ensure there are not sparse inode clusters that cross the new EOAG.
3128  *
3129  * This is a no-op for non-spinode filesystems since clusters are always fully
3130  * allocated and checking the bnobt suffices.  However, a spinode filesystem
3131  * could have a record where the upper inodes are free blocks.  If those blocks
3132  * were removed from the filesystem, the inode record would extend beyond EOAG,
3133  * which will be flagged as corruption.
3134  */
3135 int
xfs_ialloc_check_shrink(struct xfs_perag * pag,struct xfs_trans * tp,struct xfs_buf * agibp,xfs_agblock_t new_length)3136 xfs_ialloc_check_shrink(
3137 	struct xfs_perag	*pag,
3138 	struct xfs_trans	*tp,
3139 	struct xfs_buf		*agibp,
3140 	xfs_agblock_t		new_length)
3141 {
3142 	struct xfs_inobt_rec_incore rec;
3143 	struct xfs_btree_cur	*cur;
3144 	xfs_agino_t		agino;
3145 	int			has;
3146 	int			error;
3147 
3148 	if (!xfs_has_sparseinodes(pag_mount(pag)))
3149 		return 0;
3150 
3151 	cur = xfs_inobt_init_cursor(pag, tp, agibp);
3152 
3153 	/* Look up the inobt record that would correspond to the new EOFS. */
3154 	agino = XFS_AGB_TO_AGINO(pag_mount(pag), new_length);
3155 	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has);
3156 	if (error || !has)
3157 		goto out;
3158 
3159 	error = xfs_inobt_get_rec(cur, &rec, &has);
3160 	if (error)
3161 		goto out;
3162 
3163 	if (!has) {
3164 		xfs_ag_mark_sick(pag, XFS_SICK_AG_INOBT);
3165 		error = -EFSCORRUPTED;
3166 		goto out;
3167 	}
3168 
3169 	/* If the record covers inodes that would be beyond EOFS, bail out. */
3170 	if (rec.ir_startino + XFS_INODES_PER_CHUNK > agino) {
3171 		error = -ENOSPC;
3172 		goto out;
3173 	}
3174 out:
3175 	xfs_btree_del_cursor(cur, error);
3176 	return error;
3177 }
3178