aboutsummaryrefslogtreecommitdiffstats
path: root/fs/xfs/xfs_ialloc.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/xfs/xfs_ialloc.c')
-rw-r--r--fs/xfs/xfs_ialloc.c806
1 files changed, 407 insertions, 399 deletions
diff --git a/fs/xfs/xfs_ialloc.c b/fs/xfs/xfs_ialloc.c
index 3120a3a5e20f..0785797db828 100644
--- a/fs/xfs/xfs_ialloc.c
+++ b/fs/xfs/xfs_ialloc.c
@@ -57,75 +57,35 @@ xfs_ialloc_cluster_alignment(
57} 57}
58 58
59/* 59/*
60 * Lookup the record equal to ino in the btree given by cur. 60 * Lookup a record by ino in the btree given by cur.
61 */
62STATIC int /* error */
63xfs_inobt_lookup_eq(
64 struct xfs_btree_cur *cur, /* btree cursor */
65 xfs_agino_t ino, /* starting inode of chunk */
66 __int32_t fcnt, /* free inode count */
67 xfs_inofree_t free, /* free inode mask */
68 int *stat) /* success/failure */
69{
70 cur->bc_rec.i.ir_startino = ino;
71 cur->bc_rec.i.ir_freecount = fcnt;
72 cur->bc_rec.i.ir_free = free;
73 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
74}
75
76/*
77 * Lookup the first record greater than or equal to ino
78 * in the btree given by cur.
79 */ 61 */
80int /* error */ 62int /* error */
81xfs_inobt_lookup_ge( 63xfs_inobt_lookup(
82 struct xfs_btree_cur *cur, /* btree cursor */ 64 struct xfs_btree_cur *cur, /* btree cursor */
83 xfs_agino_t ino, /* starting inode of chunk */ 65 xfs_agino_t ino, /* starting inode of chunk */
84 __int32_t fcnt, /* free inode count */ 66 xfs_lookup_t dir, /* <=, >=, == */
85 xfs_inofree_t free, /* free inode mask */
86 int *stat) /* success/failure */ 67 int *stat) /* success/failure */
87{ 68{
88 cur->bc_rec.i.ir_startino = ino; 69 cur->bc_rec.i.ir_startino = ino;
89 cur->bc_rec.i.ir_freecount = fcnt; 70 cur->bc_rec.i.ir_freecount = 0;
90 cur->bc_rec.i.ir_free = free; 71 cur->bc_rec.i.ir_free = 0;
91 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat); 72 return xfs_btree_lookup(cur, dir, stat);
92} 73}
93 74
94/* 75/*
95 * Lookup the first record less than or equal to ino 76 * Update the record referred to by cur to the value given.
96 * in the btree given by cur.
97 */
98int /* error */
99xfs_inobt_lookup_le(
100 struct xfs_btree_cur *cur, /* btree cursor */
101 xfs_agino_t ino, /* starting inode of chunk */
102 __int32_t fcnt, /* free inode count */
103 xfs_inofree_t free, /* free inode mask */
104 int *stat) /* success/failure */
105{
106 cur->bc_rec.i.ir_startino = ino;
107 cur->bc_rec.i.ir_freecount = fcnt;
108 cur->bc_rec.i.ir_free = free;
109 return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
110}
111
112/*
113 * Update the record referred to by cur to the value given
114 * by [ino, fcnt, free].
115 * This either works (return 0) or gets an EFSCORRUPTED error. 77 * This either works (return 0) or gets an EFSCORRUPTED error.
116 */ 78 */
117STATIC int /* error */ 79STATIC int /* error */
118xfs_inobt_update( 80xfs_inobt_update(
119 struct xfs_btree_cur *cur, /* btree cursor */ 81 struct xfs_btree_cur *cur, /* btree cursor */
120 xfs_agino_t ino, /* starting inode of chunk */ 82 xfs_inobt_rec_incore_t *irec) /* btree record */
121 __int32_t fcnt, /* free inode count */
122 xfs_inofree_t free) /* free inode mask */
123{ 83{
124 union xfs_btree_rec rec; 84 union xfs_btree_rec rec;
125 85
126 rec.inobt.ir_startino = cpu_to_be32(ino); 86 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
127 rec.inobt.ir_freecount = cpu_to_be32(fcnt); 87 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
128 rec.inobt.ir_free = cpu_to_be64(free); 88 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
129 return xfs_btree_update(cur, &rec); 89 return xfs_btree_update(cur, &rec);
130} 90}
131 91
@@ -135,9 +95,7 @@ xfs_inobt_update(
135int /* error */ 95int /* error */
136xfs_inobt_get_rec( 96xfs_inobt_get_rec(
137 struct xfs_btree_cur *cur, /* btree cursor */ 97 struct xfs_btree_cur *cur, /* btree cursor */
138 xfs_agino_t *ino, /* output: starting inode of chunk */ 98 xfs_inobt_rec_incore_t *irec, /* btree record */
139 __int32_t *fcnt, /* output: number of free inodes */
140 xfs_inofree_t *free, /* output: free inode mask */
141 int *stat) /* output: success/failure */ 99 int *stat) /* output: success/failure */
142{ 100{
143 union xfs_btree_rec *rec; 101 union xfs_btree_rec *rec;
@@ -145,14 +103,136 @@ xfs_inobt_get_rec(
145 103
146 error = xfs_btree_get_rec(cur, &rec, stat); 104 error = xfs_btree_get_rec(cur, &rec, stat);
147 if (!error && *stat == 1) { 105 if (!error && *stat == 1) {
148 *ino = be32_to_cpu(rec->inobt.ir_startino); 106 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
149 *fcnt = be32_to_cpu(rec->inobt.ir_freecount); 107 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
150 *free = be64_to_cpu(rec->inobt.ir_free); 108 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
151 } 109 }
152 return error; 110 return error;
153} 111}
154 112
155/* 113/*
114 * Verify that the number of free inodes in the AGI is correct.
115 */
116#ifdef DEBUG
117STATIC int
118xfs_check_agi_freecount(
119 struct xfs_btree_cur *cur,
120 struct xfs_agi *agi)
121{
122 if (cur->bc_nlevels == 1) {
123 xfs_inobt_rec_incore_t rec;
124 int freecount = 0;
125 int error;
126 int i;
127
128 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
129 if (error)
130 return error;
131
132 do {
133 error = xfs_inobt_get_rec(cur, &rec, &i);
134 if (error)
135 return error;
136
137 if (i) {
138 freecount += rec.ir_freecount;
139 error = xfs_btree_increment(cur, 0, &i);
140 if (error)
141 return error;
142 }
143 } while (i == 1);
144
145 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
146 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
147 }
148 return 0;
149}
150#else
151#define xfs_check_agi_freecount(cur, agi) 0
152#endif
153
154/*
155 * Initialise a new set of inodes.
156 */
157STATIC void
158xfs_ialloc_inode_init(
159 struct xfs_mount *mp,
160 struct xfs_trans *tp,
161 xfs_agnumber_t agno,
162 xfs_agblock_t agbno,
163 xfs_agblock_t length,
164 unsigned int gen)
165{
166 struct xfs_buf *fbuf;
167 struct xfs_dinode *free;
168 int blks_per_cluster, nbufs, ninodes;
169 int version;
170 int i, j;
171 xfs_daddr_t d;
172
173 /*
174 * Loop over the new block(s), filling in the inodes.
175 * For small block sizes, manipulate the inodes in buffers
176 * which are multiples of the blocks size.
177 */
178 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
179 blks_per_cluster = 1;
180 nbufs = length;
181 ninodes = mp->m_sb.sb_inopblock;
182 } else {
183 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
184 mp->m_sb.sb_blocksize;
185 nbufs = length / blks_per_cluster;
186 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
187 }
188
189 /*
190 * Figure out what version number to use in the inodes we create.
191 * If the superblock version has caught up to the one that supports
192 * the new inode format, then use the new inode version. Otherwise
193 * use the old version so that old kernels will continue to be
194 * able to use the file system.
195 */
196 if (xfs_sb_version_hasnlink(&mp->m_sb))
197 version = 2;
198 else
199 version = 1;
200
201 for (j = 0; j < nbufs; j++) {
202 /*
203 * Get the block.
204 */
205 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
206 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
207 mp->m_bsize * blks_per_cluster,
208 XFS_BUF_LOCK);
209 ASSERT(fbuf);
210 ASSERT(!XFS_BUF_GETERROR(fbuf));
211
212 /*
213 * Initialize all inodes in this buffer and then log them.
214 *
215 * XXX: It would be much better if we had just one transaction
216 * to log a whole cluster of inodes instead of all the
217 * individual transactions causing a lot of log traffic.
218 */
219 xfs_biozero(fbuf, 0, ninodes << mp->m_sb.sb_inodelog);
220 for (i = 0; i < ninodes; i++) {
221 int ioffset = i << mp->m_sb.sb_inodelog;
222 uint isize = sizeof(struct xfs_dinode);
223
224 free = xfs_make_iptr(mp, fbuf, i);
225 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
226 free->di_version = version;
227 free->di_gen = cpu_to_be32(gen);
228 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
229 xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
230 }
231 xfs_trans_inode_alloc_buf(tp, fbuf);
232 }
233}
234
235/*
156 * Allocate new inodes in the allocation group specified by agbp. 236 * Allocate new inodes in the allocation group specified by agbp.
157 * Return 0 for success, else error code. 237 * Return 0 for success, else error code.
158 */ 238 */
@@ -164,24 +244,15 @@ xfs_ialloc_ag_alloc(
164{ 244{
165 xfs_agi_t *agi; /* allocation group header */ 245 xfs_agi_t *agi; /* allocation group header */
166 xfs_alloc_arg_t args; /* allocation argument structure */ 246 xfs_alloc_arg_t args; /* allocation argument structure */
167 int blks_per_cluster; /* fs blocks per inode cluster */
168 xfs_btree_cur_t *cur; /* inode btree cursor */ 247 xfs_btree_cur_t *cur; /* inode btree cursor */
169 xfs_daddr_t d; /* disk addr of buffer */
170 xfs_agnumber_t agno; 248 xfs_agnumber_t agno;
171 int error; 249 int error;
172 xfs_buf_t *fbuf; /* new free inodes' buffer */ 250 int i;
173 xfs_dinode_t *free; /* new free inode structure */
174 int i; /* inode counter */
175 int j; /* block counter */
176 int nbufs; /* num bufs of new inodes */
177 xfs_agino_t newino; /* new first inode's number */ 251 xfs_agino_t newino; /* new first inode's number */
178 xfs_agino_t newlen; /* new number of inodes */ 252 xfs_agino_t newlen; /* new number of inodes */
179 int ninodes; /* num inodes per buf */
180 xfs_agino_t thisino; /* current inode number, for loop */ 253 xfs_agino_t thisino; /* current inode number, for loop */
181 int version; /* inode version number to use */
182 int isaligned = 0; /* inode allocation at stripe unit */ 254 int isaligned = 0; /* inode allocation at stripe unit */
183 /* boundary */ 255 /* boundary */
184 unsigned int gen;
185 256
186 args.tp = tp; 257 args.tp = tp;
187 args.mp = tp->t_mountp; 258 args.mp = tp->t_mountp;
@@ -202,12 +273,12 @@ xfs_ialloc_ag_alloc(
202 */ 273 */
203 agi = XFS_BUF_TO_AGI(agbp); 274 agi = XFS_BUF_TO_AGI(agbp);
204 newino = be32_to_cpu(agi->agi_newino); 275 newino = be32_to_cpu(agi->agi_newino);
276 agno = be32_to_cpu(agi->agi_seqno);
205 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) + 277 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
206 XFS_IALLOC_BLOCKS(args.mp); 278 XFS_IALLOC_BLOCKS(args.mp);
207 if (likely(newino != NULLAGINO && 279 if (likely(newino != NULLAGINO &&
208 (args.agbno < be32_to_cpu(agi->agi_length)))) { 280 (args.agbno < be32_to_cpu(agi->agi_length)))) {
209 args.fsbno = XFS_AGB_TO_FSB(args.mp, 281 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
210 be32_to_cpu(agi->agi_seqno), args.agbno);
211 args.type = XFS_ALLOCTYPE_THIS_BNO; 282 args.type = XFS_ALLOCTYPE_THIS_BNO;
212 args.mod = args.total = args.wasdel = args.isfl = 283 args.mod = args.total = args.wasdel = args.isfl =
213 args.userdata = args.minalignslop = 0; 284 args.userdata = args.minalignslop = 0;
@@ -258,8 +329,7 @@ xfs_ialloc_ag_alloc(
258 * For now, just allocate blocks up front. 329 * For now, just allocate blocks up front.
259 */ 330 */
260 args.agbno = be32_to_cpu(agi->agi_root); 331 args.agbno = be32_to_cpu(agi->agi_root);
261 args.fsbno = XFS_AGB_TO_FSB(args.mp, 332 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
262 be32_to_cpu(agi->agi_seqno), args.agbno);
263 /* 333 /*
264 * Allocate a fixed-size extent of inodes. 334 * Allocate a fixed-size extent of inodes.
265 */ 335 */
@@ -282,8 +352,7 @@ xfs_ialloc_ag_alloc(
282 if (isaligned && args.fsbno == NULLFSBLOCK) { 352 if (isaligned && args.fsbno == NULLFSBLOCK) {
283 args.type = XFS_ALLOCTYPE_NEAR_BNO; 353 args.type = XFS_ALLOCTYPE_NEAR_BNO;
284 args.agbno = be32_to_cpu(agi->agi_root); 354 args.agbno = be32_to_cpu(agi->agi_root);
285 args.fsbno = XFS_AGB_TO_FSB(args.mp, 355 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
286 be32_to_cpu(agi->agi_seqno), args.agbno);
287 args.alignment = xfs_ialloc_cluster_alignment(&args); 356 args.alignment = xfs_ialloc_cluster_alignment(&args);
288 if ((error = xfs_alloc_vextent(&args))) 357 if ((error = xfs_alloc_vextent(&args)))
289 return error; 358 return error;
@@ -294,85 +363,30 @@ xfs_ialloc_ag_alloc(
294 return 0; 363 return 0;
295 } 364 }
296 ASSERT(args.len == args.minlen); 365 ASSERT(args.len == args.minlen);
297 /*
298 * Convert the results.
299 */
300 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
301 /*
302 * Loop over the new block(s), filling in the inodes.
303 * For small block sizes, manipulate the inodes in buffers
304 * which are multiples of the blocks size.
305 */
306 if (args.mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(args.mp)) {
307 blks_per_cluster = 1;
308 nbufs = (int)args.len;
309 ninodes = args.mp->m_sb.sb_inopblock;
310 } else {
311 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(args.mp) /
312 args.mp->m_sb.sb_blocksize;
313 nbufs = (int)args.len / blks_per_cluster;
314 ninodes = blks_per_cluster * args.mp->m_sb.sb_inopblock;
315 }
316 /*
317 * Figure out what version number to use in the inodes we create.
318 * If the superblock version has caught up to the one that supports
319 * the new inode format, then use the new inode version. Otherwise
320 * use the old version so that old kernels will continue to be
321 * able to use the file system.
322 */
323 if (xfs_sb_version_hasnlink(&args.mp->m_sb))
324 version = 2;
325 else
326 version = 1;
327 366
328 /* 367 /*
368 * Stamp and write the inode buffers.
369 *
329 * Seed the new inode cluster with a random generation number. This 370 * Seed the new inode cluster with a random generation number. This
330 * prevents short-term reuse of generation numbers if a chunk is 371 * prevents short-term reuse of generation numbers if a chunk is
331 * freed and then immediately reallocated. We use random numbers 372 * freed and then immediately reallocated. We use random numbers
332 * rather than a linear progression to prevent the next generation 373 * rather than a linear progression to prevent the next generation
333 * number from being easily guessable. 374 * number from being easily guessable.
334 */ 375 */
335 gen = random32(); 376 xfs_ialloc_inode_init(args.mp, tp, agno, args.agbno, args.len,
336 for (j = 0; j < nbufs; j++) { 377 random32());
337 /*
338 * Get the block.
339 */
340 d = XFS_AGB_TO_DADDR(args.mp, be32_to_cpu(agi->agi_seqno),
341 args.agbno + (j * blks_per_cluster));
342 fbuf = xfs_trans_get_buf(tp, args.mp->m_ddev_targp, d,
343 args.mp->m_bsize * blks_per_cluster,
344 XFS_BUF_LOCK);
345 ASSERT(fbuf);
346 ASSERT(!XFS_BUF_GETERROR(fbuf));
347 378
348 /* 379 /*
349 * Initialize all inodes in this buffer and then log them. 380 * Convert the results.
350 * 381 */
351 * XXX: It would be much better if we had just one transaction to 382 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
352 * log a whole cluster of inodes instead of all the individual
353 * transactions causing a lot of log traffic.
354 */
355 xfs_biozero(fbuf, 0, ninodes << args.mp->m_sb.sb_inodelog);
356 for (i = 0; i < ninodes; i++) {
357 int ioffset = i << args.mp->m_sb.sb_inodelog;
358 uint isize = sizeof(struct xfs_dinode);
359
360 free = xfs_make_iptr(args.mp, fbuf, i);
361 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
362 free->di_version = version;
363 free->di_gen = cpu_to_be32(gen);
364 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
365 xfs_trans_log_buf(tp, fbuf, ioffset, ioffset + isize - 1);
366 }
367 xfs_trans_inode_alloc_buf(tp, fbuf);
368 }
369 be32_add_cpu(&agi->agi_count, newlen); 383 be32_add_cpu(&agi->agi_count, newlen);
370 be32_add_cpu(&agi->agi_freecount, newlen); 384 be32_add_cpu(&agi->agi_freecount, newlen);
371 agno = be32_to_cpu(agi->agi_seqno);
372 down_read(&args.mp->m_peraglock); 385 down_read(&args.mp->m_peraglock);
373 args.mp->m_perag[agno].pagi_freecount += newlen; 386 args.mp->m_perag[agno].pagi_freecount += newlen;
374 up_read(&args.mp->m_peraglock); 387 up_read(&args.mp->m_peraglock);
375 agi->agi_newino = cpu_to_be32(newino); 388 agi->agi_newino = cpu_to_be32(newino);
389
376 /* 390 /*
377 * Insert records describing the new inode chunk into the btree. 391 * Insert records describing the new inode chunk into the btree.
378 */ 392 */
@@ -380,13 +394,17 @@ xfs_ialloc_ag_alloc(
380 for (thisino = newino; 394 for (thisino = newino;
381 thisino < newino + newlen; 395 thisino < newino + newlen;
382 thisino += XFS_INODES_PER_CHUNK) { 396 thisino += XFS_INODES_PER_CHUNK) {
383 if ((error = xfs_inobt_lookup_eq(cur, thisino, 397 cur->bc_rec.i.ir_startino = thisino;
384 XFS_INODES_PER_CHUNK, XFS_INOBT_ALL_FREE, &i))) { 398 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
399 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
400 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
401 if (error) {
385 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); 402 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
386 return error; 403 return error;
387 } 404 }
388 ASSERT(i == 0); 405 ASSERT(i == 0);
389 if ((error = xfs_btree_insert(cur, &i))) { 406 error = xfs_btree_insert(cur, &i);
407 if (error) {
390 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR); 408 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
391 return error; 409 return error;
392 } 410 }
@@ -539,6 +557,62 @@ nextag:
539} 557}
540 558
541/* 559/*
560 * Try to retrieve the next record to the left/right from the current one.
561 */
562STATIC int
563xfs_ialloc_next_rec(
564 struct xfs_btree_cur *cur,
565 xfs_inobt_rec_incore_t *rec,
566 int *done,
567 int left)
568{
569 int error;
570 int i;
571
572 if (left)
573 error = xfs_btree_decrement(cur, 0, &i);
574 else
575 error = xfs_btree_increment(cur, 0, &i);
576
577 if (error)
578 return error;
579 *done = !i;
580 if (i) {
581 error = xfs_inobt_get_rec(cur, rec, &i);
582 if (error)
583 return error;
584 XFS_WANT_CORRUPTED_RETURN(i == 1);
585 }
586
587 return 0;
588}
589
590STATIC int
591xfs_ialloc_get_rec(
592 struct xfs_btree_cur *cur,
593 xfs_agino_t agino,
594 xfs_inobt_rec_incore_t *rec,
595 int *done,
596 int left)
597{
598 int error;
599 int i;
600
601 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
602 if (error)
603 return error;
604 *done = !i;
605 if (i) {
606 error = xfs_inobt_get_rec(cur, rec, &i);
607 if (error)
608 return error;
609 XFS_WANT_CORRUPTED_RETURN(i == 1);
610 }
611
612 return 0;
613}
614
615/*
542 * Visible inode allocation functions. 616 * Visible inode allocation functions.
543 */ 617 */
544 618
@@ -592,8 +666,8 @@ xfs_dialloc(
592 int j; /* result code */ 666 int j; /* result code */
593 xfs_mount_t *mp; /* file system mount structure */ 667 xfs_mount_t *mp; /* file system mount structure */
594 int offset; /* index of inode in chunk */ 668 int offset; /* index of inode in chunk */
595 xfs_agino_t pagino; /* parent's a.g. relative inode # */ 669 xfs_agino_t pagino; /* parent's AG relative inode # */
596 xfs_agnumber_t pagno; /* parent's allocation group number */ 670 xfs_agnumber_t pagno; /* parent's AG number */
597 xfs_inobt_rec_incore_t rec; /* inode allocation record */ 671 xfs_inobt_rec_incore_t rec; /* inode allocation record */
598 xfs_agnumber_t tagno; /* testing allocation group number */ 672 xfs_agnumber_t tagno; /* testing allocation group number */
599 xfs_btree_cur_t *tcur; /* temp cursor */ 673 xfs_btree_cur_t *tcur; /* temp cursor */
@@ -716,6 +790,8 @@ nextag:
716 */ 790 */
717 agno = tagno; 791 agno = tagno;
718 *IO_agbp = NULL; 792 *IO_agbp = NULL;
793
794 restart_pagno:
719 cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno)); 795 cur = xfs_inobt_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno));
720 /* 796 /*
721 * If pagino is 0 (this is the root inode allocation) use newino. 797 * If pagino is 0 (this is the root inode allocation) use newino.
@@ -723,220 +799,200 @@ nextag:
723 */ 799 */
724 if (!pagino) 800 if (!pagino)
725 pagino = be32_to_cpu(agi->agi_newino); 801 pagino = be32_to_cpu(agi->agi_newino);
726#ifdef DEBUG
727 if (cur->bc_nlevels == 1) {
728 int freecount = 0;
729 802
730 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i))) 803 error = xfs_check_agi_freecount(cur, agi);
731 goto error0; 804 if (error)
732 XFS_WANT_CORRUPTED_GOTO(i == 1, error0); 805 goto error0;
733 do {
734 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino,
735 &rec.ir_freecount, &rec.ir_free, &i)))
736 goto error0;
737 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
738 freecount += rec.ir_freecount;
739 if ((error = xfs_btree_increment(cur, 0, &i)))
740 goto error0;
741 } while (i == 1);
742 806
743 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
744 XFS_FORCED_SHUTDOWN(mp));
745 }
746#endif
747 /* 807 /*
748 * If in the same a.g. as the parent, try to get near the parent. 808 * If in the same AG as the parent, try to get near the parent.
749 */ 809 */
750 if (pagno == agno) { 810 if (pagno == agno) {
751 if ((error = xfs_inobt_lookup_le(cur, pagino, 0, 0, &i))) 811 xfs_perag_t *pag = &mp->m_perag[agno];
812 int doneleft; /* done, to the left */
813 int doneright; /* done, to the right */
814 int searchdistance = 10;
815
816 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
817 if (error)
818 goto error0;
819 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
820
821 error = xfs_inobt_get_rec(cur, &rec, &j);
822 if (error)
752 goto error0; 823 goto error0;
753 if (i != 0 && 824 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
754 (error = xfs_inobt_get_rec(cur, &rec.ir_startino, 825
755 &rec.ir_freecount, &rec.ir_free, &j)) == 0 && 826 if (rec.ir_freecount > 0) {
756 j == 1 &&
757 rec.ir_freecount > 0) {
758 /* 827 /*
759 * Found a free inode in the same chunk 828 * Found a free inode in the same chunk
760 * as parent, done. 829 * as the parent, done.
761 */ 830 */
831 goto alloc_inode;
762 } 832 }
833
834
835 /*
836 * In the same AG as parent, but parent's chunk is full.
837 */
838
839 /* duplicate the cursor, search left & right simultaneously */
840 error = xfs_btree_dup_cursor(cur, &tcur);
841 if (error)
842 goto error0;
843
763 /* 844 /*
764 * In the same a.g. as parent, but parent's chunk is full. 845 * Skip to last blocks looked up if same parent inode.
765 */ 846 */
766 else { 847 if (pagino != NULLAGINO &&
767 int doneleft; /* done, to the left */ 848 pag->pagl_pagino == pagino &&
768 int doneright; /* done, to the right */ 849 pag->pagl_leftrec != NULLAGINO &&
850 pag->pagl_rightrec != NULLAGINO) {
851 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
852 &trec, &doneleft, 1);
853 if (error)
854 goto error1;
769 855
856 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
857 &rec, &doneright, 0);
770 if (error) 858 if (error)
771 goto error0;
772 ASSERT(i == 1);
773 ASSERT(j == 1);
774 /*
775 * Duplicate the cursor, search left & right
776 * simultaneously.
777 */
778 if ((error = xfs_btree_dup_cursor(cur, &tcur)))
779 goto error0;
780 /*
781 * Search left with tcur, back up 1 record.
782 */
783 if ((error = xfs_btree_decrement(tcur, 0, &i)))
784 goto error1; 859 goto error1;
785 doneleft = !i; 860 } else {
786 if (!doneleft) { 861 /* search left with tcur, back up 1 record */
787 if ((error = xfs_inobt_get_rec(tcur, 862 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
788 &trec.ir_startino, 863 if (error)
789 &trec.ir_freecount,
790 &trec.ir_free, &i)))
791 goto error1;
792 XFS_WANT_CORRUPTED_GOTO(i == 1, error1);
793 }
794 /*
795 * Search right with cur, go forward 1 record.
796 */
797 if ((error = xfs_btree_increment(cur, 0, &i)))
798 goto error1; 864 goto error1;
799 doneright = !i;
800 if (!doneright) {
801 if ((error = xfs_inobt_get_rec(cur,
802 &rec.ir_startino,
803 &rec.ir_freecount,
804 &rec.ir_free, &i)))
805 goto error1;
806 XFS_WANT_CORRUPTED_GOTO(i == 1, error1);
807 }
808 /*
809 * Loop until we find the closest inode chunk
810 * with a free one.
811 */
812 while (!doneleft || !doneright) {
813 int useleft; /* using left inode
814 chunk this time */
815 865
866 /* search right with cur, go forward 1 record. */
867 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
868 if (error)
869 goto error1;
870 }
871
872 /*
873 * Loop until we find an inode chunk with a free inode.
874 */
875 while (!doneleft || !doneright) {
876 int useleft; /* using left inode chunk this time */
877
878 if (!--searchdistance) {
816 /* 879 /*
817 * Figure out which block is closer, 880 * Not in range - save last search
818 * if both are valid. 881 * location and allocate a new inode
819 */
820 if (!doneleft && !doneright)
821 useleft =
822 pagino -
823 (trec.ir_startino +
824 XFS_INODES_PER_CHUNK - 1) <
825 rec.ir_startino - pagino;
826 else
827 useleft = !doneleft;
828 /*
829 * If checking the left, does it have
830 * free inodes?
831 */
832 if (useleft && trec.ir_freecount) {
833 /*
834 * Yes, set it up as the chunk to use.
835 */
836 rec = trec;
837 xfs_btree_del_cursor(cur,
838 XFS_BTREE_NOERROR);
839 cur = tcur;
840 break;
841 }
842 /*
843 * If checking the right, does it have
844 * free inodes?
845 */
846 if (!useleft && rec.ir_freecount) {
847 /*
848 * Yes, it's already set up.
849 */
850 xfs_btree_del_cursor(tcur,
851 XFS_BTREE_NOERROR);
852 break;
853 }
854 /*
855 * If used the left, get another one
856 * further left.
857 */
858 if (useleft) {
859 if ((error = xfs_btree_decrement(tcur, 0,
860 &i)))
861 goto error1;
862 doneleft = !i;
863 if (!doneleft) {
864 if ((error = xfs_inobt_get_rec(
865 tcur,
866 &trec.ir_startino,
867 &trec.ir_freecount,
868 &trec.ir_free, &i)))
869 goto error1;
870 XFS_WANT_CORRUPTED_GOTO(i == 1,
871 error1);
872 }
873 }
874 /*
875 * If used the right, get another one
876 * further right.
877 */ 882 */
878 else { 883 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
879 if ((error = xfs_btree_increment(cur, 0, 884 pag->pagl_leftrec = trec.ir_startino;
880 &i))) 885 pag->pagl_rightrec = rec.ir_startino;
881 goto error1; 886 pag->pagl_pagino = pagino;
882 doneright = !i; 887 goto newino;
883 if (!doneright) { 888 }
884 if ((error = xfs_inobt_get_rec( 889
885 cur, 890 /* figure out the closer block if both are valid. */
886 &rec.ir_startino, 891 if (!doneleft && !doneright) {
887 &rec.ir_freecount, 892 useleft = pagino -
888 &rec.ir_free, &i))) 893 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
889 goto error1; 894 rec.ir_startino - pagino;
890 XFS_WANT_CORRUPTED_GOTO(i == 1, 895 } else {
891 error1); 896 useleft = !doneleft;
892 }
893 }
894 } 897 }
895 ASSERT(!doneleft || !doneright); 898
899 /* free inodes to the left? */
900 if (useleft && trec.ir_freecount) {
901 rec = trec;
902 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
903 cur = tcur;
904
905 pag->pagl_leftrec = trec.ir_startino;
906 pag->pagl_rightrec = rec.ir_startino;
907 pag->pagl_pagino = pagino;
908 goto alloc_inode;
909 }
910
911 /* free inodes to the right? */
912 if (!useleft && rec.ir_freecount) {
913 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
914
915 pag->pagl_leftrec = trec.ir_startino;
916 pag->pagl_rightrec = rec.ir_startino;
917 pag->pagl_pagino = pagino;
918 goto alloc_inode;
919 }
920
921 /* get next record to check */
922 if (useleft) {
923 error = xfs_ialloc_next_rec(tcur, &trec,
924 &doneleft, 1);
925 } else {
926 error = xfs_ialloc_next_rec(cur, &rec,
927 &doneright, 0);
928 }
929 if (error)
930 goto error1;
896 } 931 }
932
933 /*
934 * We've reached the end of the btree. because
935 * we are only searching a small chunk of the
936 * btree each search, there is obviously free
937 * inodes closer to the parent inode than we
938 * are now. restart the search again.
939 */
940 pag->pagl_pagino = NULLAGINO;
941 pag->pagl_leftrec = NULLAGINO;
942 pag->pagl_rightrec = NULLAGINO;
943 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
944 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
945 goto restart_pagno;
897 } 946 }
947
898 /* 948 /*
899 * In a different a.g. from the parent. 949 * In a different AG from the parent.
900 * See if the most recently allocated block has any free. 950 * See if the most recently allocated block has any free.
901 */ 951 */
902 else if (be32_to_cpu(agi->agi_newino) != NULLAGINO) { 952newino:
903 if ((error = xfs_inobt_lookup_eq(cur, 953 if (be32_to_cpu(agi->agi_newino) != NULLAGINO) {
904 be32_to_cpu(agi->agi_newino), 0, 0, &i))) 954 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
955 XFS_LOOKUP_EQ, &i);
956 if (error)
905 goto error0; 957 goto error0;
906 if (i == 1 && 958
907 (error = xfs_inobt_get_rec(cur, &rec.ir_startino, 959 if (i == 1) {
908 &rec.ir_freecount, &rec.ir_free, &j)) == 0 && 960 error = xfs_inobt_get_rec(cur, &rec, &j);
909 j == 1 &&
910 rec.ir_freecount > 0) {
911 /*
912 * The last chunk allocated in the group still has
913 * a free inode.
914 */
915 }
916 /*
917 * None left in the last group, search the whole a.g.
918 */
919 else {
920 if (error) 961 if (error)
921 goto error0; 962 goto error0;
922 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i))) 963
923 goto error0; 964 if (j == 1 && rec.ir_freecount > 0) {
924 ASSERT(i == 1); 965 /*
925 for (;;) { 966 * The last chunk allocated in the group
926 if ((error = xfs_inobt_get_rec(cur, 967 * still has a free inode.
927 &rec.ir_startino, 968 */
928 &rec.ir_freecount, &rec.ir_free, 969 goto alloc_inode;
929 &i)))
930 goto error0;
931 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
932 if (rec.ir_freecount > 0)
933 break;
934 if ((error = xfs_btree_increment(cur, 0, &i)))
935 goto error0;
936 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
937 } 970 }
938 } 971 }
939 } 972 }
973
974 /*
975 * None left in the last group, search the whole AG
976 */
977 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
978 if (error)
979 goto error0;
980 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
981
982 for (;;) {
983 error = xfs_inobt_get_rec(cur, &rec, &i);
984 if (error)
985 goto error0;
986 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
987 if (rec.ir_freecount > 0)
988 break;
989 error = xfs_btree_increment(cur, 0, &i);
990 if (error)
991 goto error0;
992 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
993 }
994
995alloc_inode:
940 offset = xfs_ialloc_find_free(&rec.ir_free); 996 offset = xfs_ialloc_find_free(&rec.ir_free);
941 ASSERT(offset >= 0); 997 ASSERT(offset >= 0);
942 ASSERT(offset < XFS_INODES_PER_CHUNK); 998 ASSERT(offset < XFS_INODES_PER_CHUNK);
@@ -945,33 +1001,19 @@ nextag:
945 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset); 1001 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
946 rec.ir_free &= ~XFS_INOBT_MASK(offset); 1002 rec.ir_free &= ~XFS_INOBT_MASK(offset);
947 rec.ir_freecount--; 1003 rec.ir_freecount--;
948 if ((error = xfs_inobt_update(cur, rec.ir_startino, rec.ir_freecount, 1004 error = xfs_inobt_update(cur, &rec);
949 rec.ir_free))) 1005 if (error)
950 goto error0; 1006 goto error0;
951 be32_add_cpu(&agi->agi_freecount, -1); 1007 be32_add_cpu(&agi->agi_freecount, -1);
952 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT); 1008 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
953 down_read(&mp->m_peraglock); 1009 down_read(&mp->m_peraglock);
954 mp->m_perag[tagno].pagi_freecount--; 1010 mp->m_perag[tagno].pagi_freecount--;
955 up_read(&mp->m_peraglock); 1011 up_read(&mp->m_peraglock);
956#ifdef DEBUG
957 if (cur->bc_nlevels == 1) {
958 int freecount = 0;
959 1012
960 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i))) 1013 error = xfs_check_agi_freecount(cur, agi);
961 goto error0; 1014 if (error)
962 do { 1015 goto error0;
963 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino, 1016
964 &rec.ir_freecount, &rec.ir_free, &i)))
965 goto error0;
966 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
967 freecount += rec.ir_freecount;
968 if ((error = xfs_btree_increment(cur, 0, &i)))
969 goto error0;
970 } while (i == 1);
971 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
972 XFS_FORCED_SHUTDOWN(mp));
973 }
974#endif
975 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); 1017 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
976 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1); 1018 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
977 *inop = ino; 1019 *inop = ino;
@@ -1062,38 +1104,23 @@ xfs_difree(
1062 * Initialize the cursor. 1104 * Initialize the cursor.
1063 */ 1105 */
1064 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno); 1106 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1065#ifdef DEBUG
1066 if (cur->bc_nlevels == 1) {
1067 int freecount = 0;
1068 1107
1069 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i))) 1108 error = xfs_check_agi_freecount(cur, agi);
1070 goto error0; 1109 if (error)
1071 do { 1110 goto error0;
1072 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino, 1111
1073 &rec.ir_freecount, &rec.ir_free, &i)))
1074 goto error0;
1075 if (i) {
1076 freecount += rec.ir_freecount;
1077 if ((error = xfs_btree_increment(cur, 0, &i)))
1078 goto error0;
1079 }
1080 } while (i == 1);
1081 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
1082 XFS_FORCED_SHUTDOWN(mp));
1083 }
1084#endif
1085 /* 1112 /*
1086 * Look for the entry describing this inode. 1113 * Look for the entry describing this inode.
1087 */ 1114 */
1088 if ((error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i))) { 1115 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1089 cmn_err(CE_WARN, 1116 cmn_err(CE_WARN,
1090 "xfs_difree: xfs_inobt_lookup_le returned() an error %d on %s. Returning error.", 1117 "xfs_difree: xfs_inobt_lookup returned() an error %d on %s. Returning error.",
1091 error, mp->m_fsname); 1118 error, mp->m_fsname);
1092 goto error0; 1119 goto error0;
1093 } 1120 }
1094 XFS_WANT_CORRUPTED_GOTO(i == 1, error0); 1121 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1095 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino, &rec.ir_freecount, 1122 error = xfs_inobt_get_rec(cur, &rec, &i);
1096 &rec.ir_free, &i))) { 1123 if (error) {
1097 cmn_err(CE_WARN, 1124 cmn_err(CE_WARN,
1098 "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.", 1125 "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.",
1099 error, mp->m_fsname); 1126 error, mp->m_fsname);
@@ -1148,12 +1175,14 @@ xfs_difree(
1148 } else { 1175 } else {
1149 *delete = 0; 1176 *delete = 0;
1150 1177
1151 if ((error = xfs_inobt_update(cur, rec.ir_startino, rec.ir_freecount, rec.ir_free))) { 1178 error = xfs_inobt_update(cur, &rec);
1179 if (error) {
1152 cmn_err(CE_WARN, 1180 cmn_err(CE_WARN,
1153 "xfs_difree: xfs_inobt_update() returned an error %d on %s. Returning error.", 1181 "xfs_difree: xfs_inobt_update returned an error %d on %s.",
1154 error, mp->m_fsname); 1182 error, mp->m_fsname);
1155 goto error0; 1183 goto error0;
1156 } 1184 }
1185
1157 /* 1186 /*
1158 * Change the inode free counts and log the ag/sb changes. 1187 * Change the inode free counts and log the ag/sb changes.
1159 */ 1188 */
@@ -1165,28 +1194,10 @@ xfs_difree(
1165 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1); 1194 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1166 } 1195 }
1167 1196
1168#ifdef DEBUG 1197 error = xfs_check_agi_freecount(cur, agi);
1169 if (cur->bc_nlevels == 1) { 1198 if (error)
1170 int freecount = 0; 1199 goto error0;
1171 1200
1172 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
1173 goto error0;
1174 do {
1175 if ((error = xfs_inobt_get_rec(cur,
1176 &rec.ir_startino,
1177 &rec.ir_freecount,
1178 &rec.ir_free, &i)))
1179 goto error0;
1180 if (i) {
1181 freecount += rec.ir_freecount;
1182 if ((error = xfs_btree_increment(cur, 0, &i)))
1183 goto error0;
1184 }
1185 } while (i == 1);
1186 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
1187 XFS_FORCED_SHUTDOWN(mp));
1188 }
1189#endif
1190 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); 1201 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1191 return 0; 1202 return 0;
1192 1203
@@ -1297,9 +1308,7 @@ xfs_imap(
1297 chunk_agbno = agbno - offset_agbno; 1308 chunk_agbno = agbno - offset_agbno;
1298 } else { 1309 } else {
1299 xfs_btree_cur_t *cur; /* inode btree cursor */ 1310 xfs_btree_cur_t *cur; /* inode btree cursor */
1300 xfs_agino_t chunk_agino; /* first agino in inode chunk */ 1311 xfs_inobt_rec_incore_t chunk_rec;
1301 __int32_t chunk_cnt; /* count of free inodes in chunk */
1302 xfs_inofree_t chunk_free; /* mask of free inodes in chunk */
1303 xfs_buf_t *agbp; /* agi buffer */ 1312 xfs_buf_t *agbp; /* agi buffer */
1304 int i; /* temp state */ 1313 int i; /* temp state */
1305 1314
@@ -1315,15 +1324,14 @@ xfs_imap(
1315 } 1324 }
1316 1325
1317 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno); 1326 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1318 error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i); 1327 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1319 if (error) { 1328 if (error) {
1320 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: " 1329 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1321 "xfs_inobt_lookup_le() failed"); 1330 "xfs_inobt_lookup() failed");
1322 goto error0; 1331 goto error0;
1323 } 1332 }
1324 1333
1325 error = xfs_inobt_get_rec(cur, &chunk_agino, &chunk_cnt, 1334 error = xfs_inobt_get_rec(cur, &chunk_rec, &i);
1326 &chunk_free, &i);
1327 if (error) { 1335 if (error) {
1328 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: " 1336 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
1329 "xfs_inobt_get_rec() failed"); 1337 "xfs_inobt_get_rec() failed");
@@ -1341,7 +1349,7 @@ xfs_imap(
1341 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR); 1349 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1342 if (error) 1350 if (error)
1343 return error; 1351 return error;
1344 chunk_agbno = XFS_AGINO_TO_AGBNO(mp, chunk_agino); 1352 chunk_agbno = XFS_AGINO_TO_AGBNO(mp, chunk_rec.ir_startino);
1345 offset_agbno = agbno - chunk_agbno; 1353 offset_agbno = agbno - chunk_agbno;
1346 } 1354 }
1347 1355