blob: 8f3fae1aa98a5bd8df833ccd7efc78e5828408b1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
27#include "xfs_dir.h"
28#include "xfs_dir2.h"
29#include "xfs_dmapi.h"
30#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_dir_sf.h"
35#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110036#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_dinode.h"
38#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110039#include "xfs_btree.h"
40#include "xfs_ialloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_rtalloc.h"
43#include "xfs_error.h"
44#include "xfs_bmap.h"
45
46/*
47 * Log specified fields for the inode given by bp and off.
48 */
49STATIC void
50xfs_ialloc_log_di(
51 xfs_trans_t *tp, /* transaction pointer */
52 xfs_buf_t *bp, /* inode buffer */
53 int off, /* index of inode in buffer */
54 int fields) /* bitmask of fields to log */
55{
56 int first; /* first byte number */
57 int ioffset; /* off in bytes */
58 int last; /* last byte number */
59 xfs_mount_t *mp; /* mount point structure */
60 static const short offsets[] = { /* field offsets */
61 /* keep in sync with bits */
62 offsetof(xfs_dinode_core_t, di_magic),
63 offsetof(xfs_dinode_core_t, di_mode),
64 offsetof(xfs_dinode_core_t, di_version),
65 offsetof(xfs_dinode_core_t, di_format),
66 offsetof(xfs_dinode_core_t, di_onlink),
67 offsetof(xfs_dinode_core_t, di_uid),
68 offsetof(xfs_dinode_core_t, di_gid),
69 offsetof(xfs_dinode_core_t, di_nlink),
70 offsetof(xfs_dinode_core_t, di_projid),
71 offsetof(xfs_dinode_core_t, di_pad),
72 offsetof(xfs_dinode_core_t, di_atime),
73 offsetof(xfs_dinode_core_t, di_mtime),
74 offsetof(xfs_dinode_core_t, di_ctime),
75 offsetof(xfs_dinode_core_t, di_size),
76 offsetof(xfs_dinode_core_t, di_nblocks),
77 offsetof(xfs_dinode_core_t, di_extsize),
78 offsetof(xfs_dinode_core_t, di_nextents),
79 offsetof(xfs_dinode_core_t, di_anextents),
80 offsetof(xfs_dinode_core_t, di_forkoff),
81 offsetof(xfs_dinode_core_t, di_aformat),
82 offsetof(xfs_dinode_core_t, di_dmevmask),
83 offsetof(xfs_dinode_core_t, di_dmstate),
84 offsetof(xfs_dinode_core_t, di_flags),
85 offsetof(xfs_dinode_core_t, di_gen),
86 offsetof(xfs_dinode_t, di_next_unlinked),
87 offsetof(xfs_dinode_t, di_u),
88 offsetof(xfs_dinode_t, di_a),
89 sizeof(xfs_dinode_t)
90 };
91
92
93 ASSERT(offsetof(xfs_dinode_t, di_core) == 0);
94 ASSERT((fields & (XFS_DI_U|XFS_DI_A)) == 0);
95 mp = tp->t_mountp;
96 /*
97 * Get the inode-relative first and last bytes for these fields
98 */
99 xfs_btree_offsets(fields, offsets, XFS_DI_NUM_BITS, &first, &last);
100 /*
101 * Convert to buffer offsets and log it.
102 */
103 ioffset = off << mp->m_sb.sb_inodelog;
104 first += ioffset;
105 last += ioffset;
106 xfs_trans_log_buf(tp, bp, first, last);
107}
108
109/*
110 * Allocation group level functions.
111 */
112
113/*
114 * Allocate new inodes in the allocation group specified by agbp.
115 * Return 0 for success, else error code.
116 */
117STATIC int /* error code or 0 */
118xfs_ialloc_ag_alloc(
119 xfs_trans_t *tp, /* transaction pointer */
120 xfs_buf_t *agbp, /* alloc group buffer */
121 int *alloc)
122{
123 xfs_agi_t *agi; /* allocation group header */
124 xfs_alloc_arg_t args; /* allocation argument structure */
125 int blks_per_cluster; /* fs blocks per inode cluster */
126 xfs_btree_cur_t *cur; /* inode btree cursor */
127 xfs_daddr_t d; /* disk addr of buffer */
128 int error;
129 xfs_buf_t *fbuf; /* new free inodes' buffer */
130 xfs_dinode_t *free; /* new free inode structure */
131 int i; /* inode counter */
132 int j; /* block counter */
133 int nbufs; /* num bufs of new inodes */
134 xfs_agino_t newino; /* new first inode's number */
135 xfs_agino_t newlen; /* new number of inodes */
136 int ninodes; /* num inodes per buf */
137 xfs_agino_t thisino; /* current inode number, for loop */
138 int version; /* inode version number to use */
139 int isaligned; /* inode allocation at stripe unit */
140 /* boundary */
141 xfs_dinode_core_t dic; /* a dinode_core to copy to new */
142 /* inodes */
143
144 args.tp = tp;
145 args.mp = tp->t_mountp;
146
147 /*
148 * Locking will ensure that we don't have two callers in here
149 * at one time.
150 */
151 newlen = XFS_IALLOC_INODES(args.mp);
152 if (args.mp->m_maxicount &&
153 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
154 return XFS_ERROR(ENOSPC);
155 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
156 /*
157 * Set the alignment for the allocation.
158 * If stripe alignment is turned on then align at stripe unit
159 * boundary.
160 * If the cluster size is smaller than a filesystem block
161 * then we're doing I/O for inodes in filesystem block size pieces,
162 * so don't need alignment anyway.
163 */
164 isaligned = 0;
165 if (args.mp->m_sinoalign) {
166 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
167 args.alignment = args.mp->m_dalign;
168 isaligned = 1;
169 } else if (XFS_SB_VERSION_HASALIGN(&args.mp->m_sb) &&
170 args.mp->m_sb.sb_inoalignmt >=
171 XFS_B_TO_FSBT(args.mp, XFS_INODE_CLUSTER_SIZE(args.mp)))
172 args.alignment = args.mp->m_sb.sb_inoalignmt;
173 else
174 args.alignment = 1;
175 agi = XFS_BUF_TO_AGI(agbp);
176 /*
177 * Need to figure out where to allocate the inode blocks.
178 * Ideally they should be spaced out through the a.g.
179 * For now, just allocate blocks up front.
180 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100181 args.agbno = be32_to_cpu(agi->agi_root);
182 args.fsbno = XFS_AGB_TO_FSB(args.mp, be32_to_cpu(agi->agi_seqno),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 args.agbno);
184 /*
185 * Allocate a fixed-size extent of inodes.
186 */
187 args.type = XFS_ALLOCTYPE_NEAR_BNO;
188 args.mod = args.total = args.wasdel = args.isfl = args.userdata =
189 args.minalignslop = 0;
190 args.prod = 1;
191 /*
192 * Allow space for the inode btree to split.
193 */
194 args.minleft = XFS_IN_MAXLEVELS(args.mp) - 1;
195 if ((error = xfs_alloc_vextent(&args)))
196 return error;
197
198 /*
199 * If stripe alignment is turned on, then try again with cluster
200 * alignment.
201 */
202 if (isaligned && args.fsbno == NULLFSBLOCK) {
203 args.type = XFS_ALLOCTYPE_NEAR_BNO;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100204 args.agbno = be32_to_cpu(agi->agi_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 args.fsbno = XFS_AGB_TO_FSB(args.mp,
Christoph Hellwig16259e72005-11-02 15:11:25 +1100206 be32_to_cpu(agi->agi_seqno), args.agbno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (XFS_SB_VERSION_HASALIGN(&args.mp->m_sb) &&
208 args.mp->m_sb.sb_inoalignmt >=
209 XFS_B_TO_FSBT(args.mp, XFS_INODE_CLUSTER_SIZE(args.mp)))
210 args.alignment = args.mp->m_sb.sb_inoalignmt;
211 else
212 args.alignment = 1;
213 if ((error = xfs_alloc_vextent(&args)))
214 return error;
215 }
216
217 if (args.fsbno == NULLFSBLOCK) {
218 *alloc = 0;
219 return 0;
220 }
221 ASSERT(args.len == args.minlen);
222 /*
223 * Convert the results.
224 */
225 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
226 /*
227 * Loop over the new block(s), filling in the inodes.
228 * For small block sizes, manipulate the inodes in buffers
229 * which are multiples of the blocks size.
230 */
231 if (args.mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(args.mp)) {
232 blks_per_cluster = 1;
233 nbufs = (int)args.len;
234 ninodes = args.mp->m_sb.sb_inopblock;
235 } else {
236 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(args.mp) /
237 args.mp->m_sb.sb_blocksize;
238 nbufs = (int)args.len / blks_per_cluster;
239 ninodes = blks_per_cluster * args.mp->m_sb.sb_inopblock;
240 }
241 /*
242 * Figure out what version number to use in the inodes we create.
243 * If the superblock version has caught up to the one that supports
244 * the new inode format, then use the new inode version. Otherwise
245 * use the old version so that old kernels will continue to be
246 * able to use the file system.
247 */
248 if (XFS_SB_VERSION_HASNLINK(&args.mp->m_sb))
249 version = XFS_DINODE_VERSION_2;
250 else
251 version = XFS_DINODE_VERSION_1;
252
253 memset(&dic, 0, sizeof(xfs_dinode_core_t));
254 INT_SET(dic.di_magic, ARCH_CONVERT, XFS_DINODE_MAGIC);
255 INT_SET(dic.di_version, ARCH_CONVERT, version);
256
257 for (j = 0; j < nbufs; j++) {
258 /*
259 * Get the block.
260 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100261 d = XFS_AGB_TO_DADDR(args.mp, be32_to_cpu(agi->agi_seqno),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 args.agbno + (j * blks_per_cluster));
263 fbuf = xfs_trans_get_buf(tp, args.mp->m_ddev_targp, d,
264 args.mp->m_bsize * blks_per_cluster,
265 XFS_BUF_LOCK);
266 ASSERT(fbuf);
267 ASSERT(!XFS_BUF_GETERROR(fbuf));
268 /*
269 * Loop over the inodes in this buffer.
270 */
271
272 for (i = 0; i < ninodes; i++) {
273 free = XFS_MAKE_IPTR(args.mp, fbuf, i);
274 memcpy(&(free->di_core), &dic, sizeof(xfs_dinode_core_t));
275 INT_SET(free->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
276 xfs_ialloc_log_di(tp, fbuf, i,
277 XFS_DI_CORE_BITS | XFS_DI_NEXT_UNLINKED);
278 }
279 xfs_trans_inode_alloc_buf(tp, fbuf);
280 }
Christoph Hellwig16259e72005-11-02 15:11:25 +1100281 be32_add(&agi->agi_count, newlen);
282 be32_add(&agi->agi_freecount, newlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 down_read(&args.mp->m_peraglock);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100284 args.mp->m_perag[be32_to_cpu(agi->agi_seqno)].pagi_freecount += newlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 up_read(&args.mp->m_peraglock);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100286 agi->agi_newino = cpu_to_be32(newino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 /*
288 * Insert records describing the new inode chunk into the btree.
289 */
290 cur = xfs_btree_init_cursor(args.mp, tp, agbp,
Christoph Hellwig16259e72005-11-02 15:11:25 +1100291 be32_to_cpu(agi->agi_seqno),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
293 for (thisino = newino;
294 thisino < newino + newlen;
295 thisino += XFS_INODES_PER_CHUNK) {
296 if ((error = xfs_inobt_lookup_eq(cur, thisino,
297 XFS_INODES_PER_CHUNK, XFS_INOBT_ALL_FREE, &i))) {
298 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
299 return error;
300 }
301 ASSERT(i == 0);
302 if ((error = xfs_inobt_insert(cur, &i))) {
303 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
304 return error;
305 }
306 ASSERT(i == 1);
307 }
308 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
309 /*
310 * Log allocation group header fields
311 */
312 xfs_ialloc_log_agi(tp, agbp,
313 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
314 /*
315 * Modify/log superblock values for inode count and inode free count.
316 */
317 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
318 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
319 *alloc = 1;
320 return 0;
321}
322
323STATIC __inline xfs_agnumber_t
324xfs_ialloc_next_ag(
325 xfs_mount_t *mp)
326{
327 xfs_agnumber_t agno;
328
329 spin_lock(&mp->m_agirotor_lock);
330 agno = mp->m_agirotor;
331 if (++mp->m_agirotor == mp->m_maxagi)
332 mp->m_agirotor = 0;
333 spin_unlock(&mp->m_agirotor_lock);
334
335 return agno;
336}
337
338/*
339 * Select an allocation group to look for a free inode in, based on the parent
340 * inode and then mode. Return the allocation group buffer.
341 */
342STATIC xfs_buf_t * /* allocation group buffer */
343xfs_ialloc_ag_select(
344 xfs_trans_t *tp, /* transaction pointer */
345 xfs_ino_t parent, /* parent directory inode number */
346 mode_t mode, /* bits set to indicate file type */
347 int okalloc) /* ok to allocate more space */
348{
349 xfs_buf_t *agbp; /* allocation group header buffer */
350 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
351 xfs_agnumber_t agno; /* current ag number */
352 int flags; /* alloc buffer locking flags */
353 xfs_extlen_t ineed; /* blocks needed for inode allocation */
354 xfs_extlen_t longest = 0; /* longest extent available */
355 xfs_mount_t *mp; /* mount point structure */
356 int needspace; /* file mode implies space allocated */
357 xfs_perag_t *pag; /* per allocation group data */
358 xfs_agnumber_t pagno; /* parent (starting) ag number */
359
360 /*
361 * Files of these types need at least one block if length > 0
362 * (and they won't fit in the inode, but that's hard to figure out).
363 */
364 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
365 mp = tp->t_mountp;
366 agcount = mp->m_maxagi;
367 if (S_ISDIR(mode))
368 pagno = xfs_ialloc_next_ag(mp);
369 else {
370 pagno = XFS_INO_TO_AGNO(mp, parent);
371 if (pagno >= agcount)
372 pagno = 0;
373 }
374 ASSERT(pagno < agcount);
375 /*
376 * Loop through allocation groups, looking for one with a little
377 * free space in it. Note we don't look for free inodes, exactly.
378 * Instead, we include whether there is a need to allocate inodes
379 * to mean that blocks must be allocated for them,
380 * if none are currently free.
381 */
382 agno = pagno;
383 flags = XFS_ALLOC_FLAG_TRYLOCK;
384 down_read(&mp->m_peraglock);
385 for (;;) {
386 pag = &mp->m_perag[agno];
387 if (!pag->pagi_init) {
388 if (xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
389 agbp = NULL;
390 goto nextag;
391 }
392 } else
393 agbp = NULL;
394
395 if (!pag->pagi_inodeok) {
396 xfs_ialloc_next_ag(mp);
397 goto unlock_nextag;
398 }
399
400 /*
401 * Is there enough free space for the file plus a block
402 * of inodes (if we need to allocate some)?
403 */
404 ineed = pag->pagi_freecount ? 0 : XFS_IALLOC_BLOCKS(mp);
405 if (ineed && !pag->pagf_init) {
406 if (agbp == NULL &&
407 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
408 agbp = NULL;
409 goto nextag;
410 }
411 (void)xfs_alloc_pagf_init(mp, tp, agno, flags);
412 }
413 if (!ineed || pag->pagf_init) {
414 if (ineed && !(longest = pag->pagf_longest))
415 longest = pag->pagf_flcount > 0;
416 if (!ineed ||
417 (pag->pagf_freeblks >= needspace + ineed &&
418 longest >= ineed &&
419 okalloc)) {
420 if (agbp == NULL &&
421 xfs_ialloc_read_agi(mp, tp, agno, &agbp)) {
422 agbp = NULL;
423 goto nextag;
424 }
425 up_read(&mp->m_peraglock);
426 return agbp;
427 }
428 }
429unlock_nextag:
430 if (agbp)
431 xfs_trans_brelse(tp, agbp);
432nextag:
433 /*
434 * No point in iterating over the rest, if we're shutting
435 * down.
436 */
437 if (XFS_FORCED_SHUTDOWN(mp)) {
438 up_read(&mp->m_peraglock);
439 return (xfs_buf_t *)0;
440 }
441 agno++;
442 if (agno >= agcount)
443 agno = 0;
444 if (agno == pagno) {
445 if (flags == 0) {
446 up_read(&mp->m_peraglock);
447 return (xfs_buf_t *)0;
448 }
449 flags = 0;
450 }
451 }
452}
453
454/*
455 * Visible inode allocation functions.
456 */
457
458/*
459 * Allocate an inode on disk.
460 * Mode is used to tell whether the new inode will need space, and whether
461 * it is a directory.
462 *
463 * The arguments IO_agbp and alloc_done are defined to work within
464 * the constraint of one allocation per transaction.
465 * xfs_dialloc() is designed to be called twice if it has to do an
466 * allocation to make more free inodes. On the first call,
467 * IO_agbp should be set to NULL. If an inode is available,
468 * i.e., xfs_dialloc() did not need to do an allocation, an inode
469 * number is returned. In this case, IO_agbp would be set to the
470 * current ag_buf and alloc_done set to false.
471 * If an allocation needed to be done, xfs_dialloc would return
472 * the current ag_buf in IO_agbp and set alloc_done to true.
473 * The caller should then commit the current transaction, allocate a new
474 * transaction, and call xfs_dialloc() again, passing in the previous
475 * value of IO_agbp. IO_agbp should be held across the transactions.
476 * Since the agbp is locked across the two calls, the second call is
477 * guaranteed to have a free inode available.
478 *
479 * Once we successfully pick an inode its number is returned and the
480 * on-disk data structures are updated. The inode itself is not read
481 * in, since doing so would break ordering constraints with xfs_reclaim.
482 */
483int
484xfs_dialloc(
485 xfs_trans_t *tp, /* transaction pointer */
486 xfs_ino_t parent, /* parent inode (directory) */
487 mode_t mode, /* mode bits for new inode */
488 int okalloc, /* ok to allocate more space */
489 xfs_buf_t **IO_agbp, /* in/out ag header's buffer */
490 boolean_t *alloc_done, /* true if we needed to replenish
491 inode freelist */
492 xfs_ino_t *inop) /* inode number allocated */
493{
494 xfs_agnumber_t agcount; /* number of allocation groups */
495 xfs_buf_t *agbp; /* allocation group header's buffer */
496 xfs_agnumber_t agno; /* allocation group number */
497 xfs_agi_t *agi; /* allocation group header structure */
498 xfs_btree_cur_t *cur; /* inode allocation btree cursor */
499 int error; /* error return value */
500 int i; /* result code */
501 int ialloced; /* inode allocation status */
502 int noroom = 0; /* no space for inode blk allocation */
503 xfs_ino_t ino; /* fs-relative inode to be returned */
504 /* REFERENCED */
505 int j; /* result code */
506 xfs_mount_t *mp; /* file system mount structure */
507 int offset; /* index of inode in chunk */
508 xfs_agino_t pagino; /* parent's a.g. relative inode # */
509 xfs_agnumber_t pagno; /* parent's allocation group number */
510 xfs_inobt_rec_t rec; /* inode allocation record */
511 xfs_agnumber_t tagno; /* testing allocation group number */
512 xfs_btree_cur_t *tcur; /* temp cursor */
513 xfs_inobt_rec_t trec; /* temp inode allocation record */
514
515
516 if (*IO_agbp == NULL) {
517 /*
518 * We do not have an agbp, so select an initial allocation
519 * group for inode allocation.
520 */
521 agbp = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
522 /*
523 * Couldn't find an allocation group satisfying the
524 * criteria, give up.
525 */
526 if (!agbp) {
527 *inop = NULLFSINO;
528 return 0;
529 }
530 agi = XFS_BUF_TO_AGI(agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100531 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 } else {
533 /*
534 * Continue where we left off before. In this case, we
535 * know that the allocation group has free inodes.
536 */
537 agbp = *IO_agbp;
538 agi = XFS_BUF_TO_AGI(agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100539 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
540 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 }
542 mp = tp->t_mountp;
543 agcount = mp->m_sb.sb_agcount;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100544 agno = be32_to_cpu(agi->agi_seqno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 tagno = agno;
546 pagno = XFS_INO_TO_AGNO(mp, parent);
547 pagino = XFS_INO_TO_AGINO(mp, parent);
548
549 /*
550 * If we have already hit the ceiling of inode blocks then clear
551 * okalloc so we scan all available agi structures for a free
552 * inode.
553 */
554
555 if (mp->m_maxicount &&
556 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
557 noroom = 1;
558 okalloc = 0;
559 }
560
561 /*
562 * Loop until we find an allocation group that either has free inodes
563 * or in which we can allocate some inodes. Iterate through the
564 * allocation groups upward, wrapping at the end.
565 */
566 *alloc_done = B_FALSE;
567 while (!agi->agi_freecount) {
568 /*
569 * Don't do anything if we're not supposed to allocate
570 * any blocks, just go on to the next ag.
571 */
572 if (okalloc) {
573 /*
574 * Try to allocate some new inodes in the allocation
575 * group.
576 */
577 if ((error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced))) {
578 xfs_trans_brelse(tp, agbp);
579 if (error == ENOSPC) {
580 *inop = NULLFSINO;
581 return 0;
582 } else
583 return error;
584 }
585 if (ialloced) {
586 /*
587 * We successfully allocated some inodes, return
588 * the current context to the caller so that it
589 * can commit the current transaction and call
590 * us again where we left off.
591 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100592 ASSERT(be32_to_cpu(agi->agi_freecount) > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 *alloc_done = B_TRUE;
594 *IO_agbp = agbp;
595 *inop = NULLFSINO;
596 return 0;
597 }
598 }
599 /*
600 * If it failed, give up on this ag.
601 */
602 xfs_trans_brelse(tp, agbp);
603 /*
604 * Go on to the next ag: get its ag header.
605 */
606nextag:
607 if (++tagno == agcount)
608 tagno = 0;
609 if (tagno == agno) {
610 *inop = NULLFSINO;
611 return noroom ? ENOSPC : 0;
612 }
613 down_read(&mp->m_peraglock);
614 if (mp->m_perag[tagno].pagi_inodeok == 0) {
615 up_read(&mp->m_peraglock);
616 goto nextag;
617 }
618 error = xfs_ialloc_read_agi(mp, tp, tagno, &agbp);
619 up_read(&mp->m_peraglock);
620 if (error)
621 goto nextag;
622 agi = XFS_BUF_TO_AGI(agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100623 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 }
625 /*
626 * Here with an allocation group that has a free inode.
627 * Reset agno since we may have chosen a new ag in the
628 * loop above.
629 */
630 agno = tagno;
631 *IO_agbp = NULL;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100632 cur = xfs_btree_init_cursor(mp, tp, agbp, be32_to_cpu(agi->agi_seqno),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
634 /*
635 * If pagino is 0 (this is the root inode allocation) use newino.
636 * This must work because we've just allocated some.
637 */
638 if (!pagino)
Christoph Hellwig16259e72005-11-02 15:11:25 +1100639 pagino = be32_to_cpu(agi->agi_newino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640#ifdef DEBUG
641 if (cur->bc_nlevels == 1) {
642 int freecount = 0;
643
644 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
645 goto error0;
646 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
647 do {
648 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino,
649 &rec.ir_freecount, &rec.ir_free, &i)))
650 goto error0;
651 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
652 freecount += rec.ir_freecount;
653 if ((error = xfs_inobt_increment(cur, 0, &i)))
654 goto error0;
655 } while (i == 1);
656
Christoph Hellwig16259e72005-11-02 15:11:25 +1100657 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 XFS_FORCED_SHUTDOWN(mp));
659 }
660#endif
661 /*
662 * If in the same a.g. as the parent, try to get near the parent.
663 */
664 if (pagno == agno) {
665 if ((error = xfs_inobt_lookup_le(cur, pagino, 0, 0, &i)))
666 goto error0;
667 if (i != 0 &&
668 (error = xfs_inobt_get_rec(cur, &rec.ir_startino,
669 &rec.ir_freecount, &rec.ir_free, &j)) == 0 &&
670 j == 1 &&
671 rec.ir_freecount > 0) {
672 /*
673 * Found a free inode in the same chunk
674 * as parent, done.
675 */
676 }
677 /*
678 * In the same a.g. as parent, but parent's chunk is full.
679 */
680 else {
681 int doneleft; /* done, to the left */
682 int doneright; /* done, to the right */
683
684 if (error)
685 goto error0;
686 ASSERT(i == 1);
687 ASSERT(j == 1);
688 /*
689 * Duplicate the cursor, search left & right
690 * simultaneously.
691 */
692 if ((error = xfs_btree_dup_cursor(cur, &tcur)))
693 goto error0;
694 /*
695 * Search left with tcur, back up 1 record.
696 */
697 if ((error = xfs_inobt_decrement(tcur, 0, &i)))
698 goto error1;
699 doneleft = !i;
700 if (!doneleft) {
701 if ((error = xfs_inobt_get_rec(tcur,
702 &trec.ir_startino,
703 &trec.ir_freecount,
704 &trec.ir_free, &i)))
705 goto error1;
706 XFS_WANT_CORRUPTED_GOTO(i == 1, error1);
707 }
708 /*
709 * Search right with cur, go forward 1 record.
710 */
711 if ((error = xfs_inobt_increment(cur, 0, &i)))
712 goto error1;
713 doneright = !i;
714 if (!doneright) {
715 if ((error = xfs_inobt_get_rec(cur,
716 &rec.ir_startino,
717 &rec.ir_freecount,
718 &rec.ir_free, &i)))
719 goto error1;
720 XFS_WANT_CORRUPTED_GOTO(i == 1, error1);
721 }
722 /*
723 * Loop until we find the closest inode chunk
724 * with a free one.
725 */
726 while (!doneleft || !doneright) {
727 int useleft; /* using left inode
728 chunk this time */
729
730 /*
731 * Figure out which block is closer,
732 * if both are valid.
733 */
734 if (!doneleft && !doneright)
735 useleft =
736 pagino -
737 (trec.ir_startino +
738 XFS_INODES_PER_CHUNK - 1) <
739 rec.ir_startino - pagino;
740 else
741 useleft = !doneleft;
742 /*
743 * If checking the left, does it have
744 * free inodes?
745 */
746 if (useleft && trec.ir_freecount) {
747 /*
748 * Yes, set it up as the chunk to use.
749 */
750 rec = trec;
751 xfs_btree_del_cursor(cur,
752 XFS_BTREE_NOERROR);
753 cur = tcur;
754 break;
755 }
756 /*
757 * If checking the right, does it have
758 * free inodes?
759 */
760 if (!useleft && rec.ir_freecount) {
761 /*
762 * Yes, it's already set up.
763 */
764 xfs_btree_del_cursor(tcur,
765 XFS_BTREE_NOERROR);
766 break;
767 }
768 /*
769 * If used the left, get another one
770 * further left.
771 */
772 if (useleft) {
773 if ((error = xfs_inobt_decrement(tcur, 0,
774 &i)))
775 goto error1;
776 doneleft = !i;
777 if (!doneleft) {
778 if ((error = xfs_inobt_get_rec(
779 tcur,
780 &trec.ir_startino,
781 &trec.ir_freecount,
782 &trec.ir_free, &i)))
783 goto error1;
784 XFS_WANT_CORRUPTED_GOTO(i == 1,
785 error1);
786 }
787 }
788 /*
789 * If used the right, get another one
790 * further right.
791 */
792 else {
793 if ((error = xfs_inobt_increment(cur, 0,
794 &i)))
795 goto error1;
796 doneright = !i;
797 if (!doneright) {
798 if ((error = xfs_inobt_get_rec(
799 cur,
800 &rec.ir_startino,
801 &rec.ir_freecount,
802 &rec.ir_free, &i)))
803 goto error1;
804 XFS_WANT_CORRUPTED_GOTO(i == 1,
805 error1);
806 }
807 }
808 }
809 ASSERT(!doneleft || !doneright);
810 }
811 }
812 /*
813 * In a different a.g. from the parent.
814 * See if the most recently allocated block has any free.
815 */
Christoph Hellwig16259e72005-11-02 15:11:25 +1100816 else if (be32_to_cpu(agi->agi_newino) != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 if ((error = xfs_inobt_lookup_eq(cur,
Christoph Hellwig16259e72005-11-02 15:11:25 +1100818 be32_to_cpu(agi->agi_newino), 0, 0, &i)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 goto error0;
820 if (i == 1 &&
821 (error = xfs_inobt_get_rec(cur, &rec.ir_startino,
822 &rec.ir_freecount, &rec.ir_free, &j)) == 0 &&
823 j == 1 &&
824 rec.ir_freecount > 0) {
825 /*
826 * The last chunk allocated in the group still has
827 * a free inode.
828 */
829 }
830 /*
831 * None left in the last group, search the whole a.g.
832 */
833 else {
834 if (error)
835 goto error0;
836 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
837 goto error0;
838 ASSERT(i == 1);
839 for (;;) {
840 if ((error = xfs_inobt_get_rec(cur,
841 &rec.ir_startino,
842 &rec.ir_freecount, &rec.ir_free,
843 &i)))
844 goto error0;
845 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
846 if (rec.ir_freecount > 0)
847 break;
848 if ((error = xfs_inobt_increment(cur, 0, &i)))
849 goto error0;
850 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
851 }
852 }
853 }
854 offset = XFS_IALLOC_FIND_FREE(&rec.ir_free);
855 ASSERT(offset >= 0);
856 ASSERT(offset < XFS_INODES_PER_CHUNK);
857 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
858 XFS_INODES_PER_CHUNK) == 0);
859 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
860 XFS_INOBT_CLR_FREE(&rec, offset);
861 rec.ir_freecount--;
862 if ((error = xfs_inobt_update(cur, rec.ir_startino, rec.ir_freecount,
863 rec.ir_free)))
864 goto error0;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100865 be32_add(&agi->agi_freecount, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
867 down_read(&mp->m_peraglock);
868 mp->m_perag[tagno].pagi_freecount--;
869 up_read(&mp->m_peraglock);
870#ifdef DEBUG
871 if (cur->bc_nlevels == 1) {
872 int freecount = 0;
873
874 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
875 goto error0;
876 do {
877 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino,
878 &rec.ir_freecount, &rec.ir_free, &i)))
879 goto error0;
880 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
881 freecount += rec.ir_freecount;
882 if ((error = xfs_inobt_increment(cur, 0, &i)))
883 goto error0;
884 } while (i == 1);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100885 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 XFS_FORCED_SHUTDOWN(mp));
887 }
888#endif
889 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
890 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
891 *inop = ino;
892 return 0;
893error1:
894 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
895error0:
896 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
897 return error;
898}
899
900/*
901 * Free disk inode. Carefully avoids touching the incore inode, all
902 * manipulations incore are the caller's responsibility.
903 * The on-disk inode is not changed by this operation, only the
904 * btree (free inode mask) is changed.
905 */
906int
907xfs_difree(
908 xfs_trans_t *tp, /* transaction pointer */
909 xfs_ino_t inode, /* inode to be freed */
910 xfs_bmap_free_t *flist, /* extents to free */
911 int *delete, /* set if inode cluster was deleted */
912 xfs_ino_t *first_ino) /* first inode in deleted cluster */
913{
914 /* REFERENCED */
915 xfs_agblock_t agbno; /* block number containing inode */
916 xfs_buf_t *agbp; /* buffer containing allocation group header */
917 xfs_agino_t agino; /* inode number relative to allocation group */
918 xfs_agnumber_t agno; /* allocation group number */
919 xfs_agi_t *agi; /* allocation group header */
920 xfs_btree_cur_t *cur; /* inode btree cursor */
921 int error; /* error return value */
922 int i; /* result code */
923 int ilen; /* inodes in an inode cluster */
924 xfs_mount_t *mp; /* mount structure for filesystem */
925 int off; /* offset of inode in inode chunk */
926 xfs_inobt_rec_t rec; /* btree record */
927
928 mp = tp->t_mountp;
929
930 /*
931 * Break up inode number into its components.
932 */
933 agno = XFS_INO_TO_AGNO(mp, inode);
934 if (agno >= mp->m_sb.sb_agcount) {
935 cmn_err(CE_WARN,
936 "xfs_difree: agno >= mp->m_sb.sb_agcount (%d >= %d) on %s. Returning EINVAL.",
937 agno, mp->m_sb.sb_agcount, mp->m_fsname);
938 ASSERT(0);
939 return XFS_ERROR(EINVAL);
940 }
941 agino = XFS_INO_TO_AGINO(mp, inode);
942 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
943 cmn_err(CE_WARN,
Christoph Hellwigda1650a2005-11-02 10:21:35 +1100944 "xfs_difree: inode != XFS_AGINO_TO_INO() "
945 "(%llu != %llu) on %s. Returning EINVAL.",
946 (unsigned long long)inode,
947 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino),
948 mp->m_fsname);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 ASSERT(0);
950 return XFS_ERROR(EINVAL);
951 }
952 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
953 if (agbno >= mp->m_sb.sb_agblocks) {
954 cmn_err(CE_WARN,
955 "xfs_difree: agbno >= mp->m_sb.sb_agblocks (%d >= %d) on %s. Returning EINVAL.",
956 agbno, mp->m_sb.sb_agblocks, mp->m_fsname);
957 ASSERT(0);
958 return XFS_ERROR(EINVAL);
959 }
960 /*
961 * Get the allocation group header.
962 */
963 down_read(&mp->m_peraglock);
964 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
965 up_read(&mp->m_peraglock);
966 if (error) {
967 cmn_err(CE_WARN,
968 "xfs_difree: xfs_ialloc_read_agi() returned an error %d on %s. Returning error.",
969 error, mp->m_fsname);
970 return error;
971 }
972 agi = XFS_BUF_TO_AGI(agbp);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100973 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
974 ASSERT(agbno < be32_to_cpu(agi->agi_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 /*
976 * Initialize the cursor.
977 */
978 cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO,
979 (xfs_inode_t *)0, 0);
980#ifdef DEBUG
981 if (cur->bc_nlevels == 1) {
982 int freecount = 0;
983
984 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
985 goto error0;
986 do {
987 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino,
988 &rec.ir_freecount, &rec.ir_free, &i)))
989 goto error0;
990 if (i) {
991 freecount += rec.ir_freecount;
992 if ((error = xfs_inobt_increment(cur, 0, &i)))
993 goto error0;
994 }
995 } while (i == 1);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100996 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 XFS_FORCED_SHUTDOWN(mp));
998 }
999#endif
1000 /*
1001 * Look for the entry describing this inode.
1002 */
1003 if ((error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i))) {
1004 cmn_err(CE_WARN,
1005 "xfs_difree: xfs_inobt_lookup_le returned() an error %d on %s. Returning error.",
1006 error, mp->m_fsname);
1007 goto error0;
1008 }
1009 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1010 if ((error = xfs_inobt_get_rec(cur, &rec.ir_startino, &rec.ir_freecount,
1011 &rec.ir_free, &i))) {
1012 cmn_err(CE_WARN,
1013 "xfs_difree: xfs_inobt_get_rec() returned an error %d on %s. Returning error.",
1014 error, mp->m_fsname);
1015 goto error0;
1016 }
1017 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1018 /*
1019 * Get the offset in the inode chunk.
1020 */
1021 off = agino - rec.ir_startino;
1022 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1023 ASSERT(!XFS_INOBT_IS_FREE(&rec, off));
1024 /*
1025 * Mark the inode free & increment the count.
1026 */
1027 XFS_INOBT_SET_FREE(&rec, off);
1028 rec.ir_freecount++;
1029
1030 /*
1031 * When an inode cluster is free, it becomes elgible for removal
1032 */
1033 if ((mp->m_flags & XFS_MOUNT_IDELETE) &&
1034 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1035
1036 *delete = 1;
1037 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1038
1039 /*
1040 * Remove the inode cluster from the AGI B+Tree, adjust the
1041 * AGI and Superblock inode counts, and mark the disk space
1042 * to be freed when the transaction is committed.
1043 */
1044 ilen = XFS_IALLOC_INODES(mp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001045 be32_add(&agi->agi_count, -ilen);
1046 be32_add(&agi->agi_freecount, -(ilen - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1048 down_read(&mp->m_peraglock);
1049 mp->m_perag[agno].pagi_freecount -= ilen - 1;
1050 up_read(&mp->m_peraglock);
1051 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1052 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1053
1054 if ((error = xfs_inobt_delete(cur, &i))) {
1055 cmn_err(CE_WARN, "xfs_difree: xfs_inobt_delete returned an error %d on %s.\n",
1056 error, mp->m_fsname);
1057 goto error0;
1058 }
1059
1060 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
1061 agno, XFS_INO_TO_AGBNO(mp,rec.ir_startino)),
1062 XFS_IALLOC_BLOCKS(mp), flist, mp);
1063 } else {
1064 *delete = 0;
1065
1066 if ((error = xfs_inobt_update(cur, rec.ir_startino, rec.ir_freecount, rec.ir_free))) {
1067 cmn_err(CE_WARN,
1068 "xfs_difree: xfs_inobt_update() returned an error %d on %s. Returning error.",
1069 error, mp->m_fsname);
1070 goto error0;
1071 }
1072 /*
1073 * Change the inode free counts and log the ag/sb changes.
1074 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001075 be32_add(&agi->agi_freecount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1077 down_read(&mp->m_peraglock);
1078 mp->m_perag[agno].pagi_freecount++;
1079 up_read(&mp->m_peraglock);
1080 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1081 }
1082
1083#ifdef DEBUG
1084 if (cur->bc_nlevels == 1) {
1085 int freecount = 0;
1086
1087 if ((error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &i)))
1088 goto error0;
1089 do {
1090 if ((error = xfs_inobt_get_rec(cur,
1091 &rec.ir_startino,
1092 &rec.ir_freecount,
1093 &rec.ir_free, &i)))
1094 goto error0;
1095 if (i) {
1096 freecount += rec.ir_freecount;
1097 if ((error = xfs_inobt_increment(cur, 0, &i)))
1098 goto error0;
1099 }
1100 } while (i == 1);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001101 ASSERT(freecount == be32_to_cpu(agi->agi_freecount) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 XFS_FORCED_SHUTDOWN(mp));
1103 }
1104#endif
1105 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1106 return 0;
1107
1108error0:
1109 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1110 return error;
1111}
1112
1113/*
1114 * Return the location of the inode in bno/off, for mapping it into a buffer.
1115 */
1116/*ARGSUSED*/
1117int
1118xfs_dilocate(
1119 xfs_mount_t *mp, /* file system mount structure */
1120 xfs_trans_t *tp, /* transaction pointer */
1121 xfs_ino_t ino, /* inode to locate */
1122 xfs_fsblock_t *bno, /* output: block containing inode */
1123 int *len, /* output: num blocks in inode cluster */
1124 int *off, /* output: index in block of inode */
1125 uint flags) /* flags concerning inode lookup */
1126{
1127 xfs_agblock_t agbno; /* block number of inode in the alloc group */
1128 xfs_buf_t *agbp; /* agi buffer */
1129 xfs_agino_t agino; /* inode number within alloc group */
1130 xfs_agnumber_t agno; /* allocation group number */
1131 int blks_per_cluster; /* num blocks per inode cluster */
1132 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
1133 xfs_agino_t chunk_agino; /* first agino in inode chunk */
1134 __int32_t chunk_cnt; /* count of free inodes in chunk */
1135 xfs_inofree_t chunk_free; /* mask of free inodes in chunk */
1136 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
1137 xfs_btree_cur_t *cur; /* inode btree cursor */
1138 int error; /* error code */
1139 int i; /* temp state */
1140 int offset; /* index of inode in its buffer */
1141 int offset_agbno; /* blks from chunk start to inode */
1142
1143 ASSERT(ino != NULLFSINO);
1144 /*
1145 * Split up the inode number into its parts.
1146 */
1147 agno = XFS_INO_TO_AGNO(mp, ino);
1148 agino = XFS_INO_TO_AGINO(mp, ino);
1149 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1150 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1151 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1152#ifdef DEBUG
1153 if (agno >= mp->m_sb.sb_agcount) {
1154 xfs_fs_cmn_err(CE_ALERT, mp,
1155 "xfs_dilocate: agno (%d) >= "
1156 "mp->m_sb.sb_agcount (%d)",
1157 agno, mp->m_sb.sb_agcount);
1158 }
1159 if (agbno >= mp->m_sb.sb_agblocks) {
1160 xfs_fs_cmn_err(CE_ALERT, mp,
1161 "xfs_dilocate: agbno (0x%llx) >= "
1162 "mp->m_sb.sb_agblocks (0x%lx)",
1163 (unsigned long long) agbno,
1164 (unsigned long) mp->m_sb.sb_agblocks);
1165 }
1166 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1167 xfs_fs_cmn_err(CE_ALERT, mp,
1168 "xfs_dilocate: ino (0x%llx) != "
1169 "XFS_AGINO_TO_INO(mp, agno, agino) "
1170 "(0x%llx)",
1171 ino, XFS_AGINO_TO_INO(mp, agno, agino));
1172 }
1173#endif /* DEBUG */
1174 return XFS_ERROR(EINVAL);
1175 }
1176 if ((mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) ||
1177 !(flags & XFS_IMAP_LOOKUP)) {
1178 offset = XFS_INO_TO_OFFSET(mp, ino);
1179 ASSERT(offset < mp->m_sb.sb_inopblock);
1180 *bno = XFS_AGB_TO_FSB(mp, agno, agbno);
1181 *off = offset;
1182 *len = 1;
1183 return 0;
1184 }
1185 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1186 if (*bno != NULLFSBLOCK) {
1187 offset = XFS_INO_TO_OFFSET(mp, ino);
1188 ASSERT(offset < mp->m_sb.sb_inopblock);
1189 cluster_agbno = XFS_FSB_TO_AGBNO(mp, *bno);
1190 *off = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1191 offset;
1192 *len = blks_per_cluster;
1193 return 0;
1194 }
1195 if (mp->m_inoalign_mask) {
1196 offset_agbno = agbno & mp->m_inoalign_mask;
1197 chunk_agbno = agbno - offset_agbno;
1198 } else {
1199 down_read(&mp->m_peraglock);
1200 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1201 up_read(&mp->m_peraglock);
1202 if (error) {
1203#ifdef DEBUG
1204 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_dilocate: "
1205 "xfs_ialloc_read_agi() returned "
1206 "error %d, agno %d",
1207 error, agno);
1208#endif /* DEBUG */
1209 return error;
1210 }
1211 cur = xfs_btree_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO,
1212 (xfs_inode_t *)0, 0);
1213 if ((error = xfs_inobt_lookup_le(cur, agino, 0, 0, &i))) {
1214#ifdef DEBUG
1215 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_dilocate: "
1216 "xfs_inobt_lookup_le() failed");
1217#endif /* DEBUG */
1218 goto error0;
1219 }
1220 if ((error = xfs_inobt_get_rec(cur, &chunk_agino, &chunk_cnt,
1221 &chunk_free, &i))) {
1222#ifdef DEBUG
1223 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_dilocate: "
1224 "xfs_inobt_get_rec() failed");
1225#endif /* DEBUG */
1226 goto error0;
1227 }
1228 if (i == 0) {
1229#ifdef DEBUG
1230 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_dilocate: "
1231 "xfs_inobt_get_rec() failed");
1232#endif /* DEBUG */
1233 error = XFS_ERROR(EINVAL);
1234 }
1235 xfs_trans_brelse(tp, agbp);
1236 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1237 if (error)
1238 return error;
1239 chunk_agbno = XFS_AGINO_TO_AGBNO(mp, chunk_agino);
1240 offset_agbno = agbno - chunk_agbno;
1241 }
1242 ASSERT(agbno >= chunk_agbno);
1243 cluster_agbno = chunk_agbno +
1244 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1245 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1246 XFS_INO_TO_OFFSET(mp, ino);
1247 *bno = XFS_AGB_TO_FSB(mp, agno, cluster_agbno);
1248 *off = offset;
1249 *len = blks_per_cluster;
1250 return 0;
1251error0:
1252 xfs_trans_brelse(tp, agbp);
1253 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1254 return error;
1255}
1256
1257/*
1258 * Compute and fill in value of m_in_maxlevels.
1259 */
1260void
1261xfs_ialloc_compute_maxlevels(
1262 xfs_mount_t *mp) /* file system mount structure */
1263{
1264 int level;
1265 uint maxblocks;
1266 uint maxleafents;
1267 int minleafrecs;
1268 int minnoderecs;
1269
1270 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1271 XFS_INODES_PER_CHUNK_LOG;
1272 minleafrecs = mp->m_alloc_mnr[0];
1273 minnoderecs = mp->m_alloc_mnr[1];
1274 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1275 for (level = 1; maxblocks > 1; level++)
1276 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1277 mp->m_in_maxlevels = level;
1278}
1279
1280/*
1281 * Log specified fields for the ag hdr (inode section)
1282 */
1283void
1284xfs_ialloc_log_agi(
1285 xfs_trans_t *tp, /* transaction pointer */
1286 xfs_buf_t *bp, /* allocation group header buffer */
1287 int fields) /* bitmask of fields to log */
1288{
1289 int first; /* first byte number */
1290 int last; /* last byte number */
1291 static const short offsets[] = { /* field starting offsets */
1292 /* keep in sync with bit definitions */
1293 offsetof(xfs_agi_t, agi_magicnum),
1294 offsetof(xfs_agi_t, agi_versionnum),
1295 offsetof(xfs_agi_t, agi_seqno),
1296 offsetof(xfs_agi_t, agi_length),
1297 offsetof(xfs_agi_t, agi_count),
1298 offsetof(xfs_agi_t, agi_root),
1299 offsetof(xfs_agi_t, agi_level),
1300 offsetof(xfs_agi_t, agi_freecount),
1301 offsetof(xfs_agi_t, agi_newino),
1302 offsetof(xfs_agi_t, agi_dirino),
1303 offsetof(xfs_agi_t, agi_unlinked),
1304 sizeof(xfs_agi_t)
1305 };
1306#ifdef DEBUG
1307 xfs_agi_t *agi; /* allocation group header */
1308
1309 agi = XFS_BUF_TO_AGI(bp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001310 ASSERT(be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311#endif
1312 /*
1313 * Compute byte offsets for the first and last fields.
1314 */
1315 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1316 /*
1317 * Log the allocation group inode header buffer.
1318 */
1319 xfs_trans_log_buf(tp, bp, first, last);
1320}
1321
1322/*
1323 * Read in the allocation group header (inode allocation section)
1324 */
1325int
1326xfs_ialloc_read_agi(
1327 xfs_mount_t *mp, /* file system mount structure */
1328 xfs_trans_t *tp, /* transaction pointer */
1329 xfs_agnumber_t agno, /* allocation group number */
1330 xfs_buf_t **bpp) /* allocation group hdr buf */
1331{
1332 xfs_agi_t *agi; /* allocation group header */
1333 int agi_ok; /* agi is consistent */
1334 xfs_buf_t *bp; /* allocation group hdr buf */
1335 xfs_perag_t *pag; /* per allocation group data */
1336 int error;
1337
1338 ASSERT(agno != NULLAGNUMBER);
1339 error = xfs_trans_read_buf(
1340 mp, tp, mp->m_ddev_targp,
1341 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1342 XFS_FSS_TO_BB(mp, 1), 0, &bp);
1343 if (error)
1344 return error;
1345 ASSERT(bp && !XFS_BUF_GETERROR(bp));
1346
1347 /*
1348 * Validate the magic number of the agi block.
1349 */
1350 agi = XFS_BUF_TO_AGI(bp);
1351 agi_ok =
Christoph Hellwig16259e72005-11-02 15:11:25 +11001352 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1353 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1355 XFS_RANDOM_IALLOC_READ_AGI))) {
1356 XFS_CORRUPTION_ERROR("xfs_ialloc_read_agi", XFS_ERRLEVEL_LOW,
1357 mp, agi);
1358 xfs_trans_brelse(tp, bp);
1359 return XFS_ERROR(EFSCORRUPTED);
1360 }
1361 pag = &mp->m_perag[agno];
1362 if (!pag->pagi_init) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11001363 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 pag->pagi_init = 1;
1365 } else {
1366 /*
1367 * It's possible for these to be out of sync if
1368 * we are in the middle of a forced shutdown.
1369 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001370 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1371 XFS_FORCED_SHUTDOWN(mp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 }
1373
1374#ifdef DEBUG
1375 {
1376 int i;
1377
1378 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1379 ASSERT(agi->agi_unlinked[i]);
1380 }
1381#endif
1382
1383 XFS_BUF_SET_VTYPE_REF(bp, B_FS_AGI, XFS_AGI_REF);
1384 *bpp = bp;
1385 return 0;
1386}