blob: 7a728f9fc0be97d98aeb3a4908716a03d27dbe57 [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"
Dave Chinner70a9883c2013-10-23 10:36:05 +110020#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_bit.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_sb.h"
27#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_btree.h"
31#include "xfs_ialloc.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110032#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_rtalloc.h"
35#include "xfs_error.h"
36#include "xfs_bmap.h"
Dave Chinner983d09f2013-04-03 16:11:15 +110037#include "xfs_cksum.h"
Dave Chinner239880e2013-10-23 10:50:10 +110038#include "xfs_trans.h"
Dave Chinner983d09f2013-04-03 16:11:15 +110039#include "xfs_buf_item.h"
Dave Chinnerddf6ad02013-06-27 16:04:56 +100040#include "xfs_icreate_item.h"
Dave Chinner7bb85ef2013-08-12 20:49:34 +100041#include "xfs_icache.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110042#include "xfs_dinode.h"
Dave Chinnerd1230312013-11-01 15:27:19 +110043#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46/*
47 * Allocation group level functions.
48 */
David Chinner75de2a92008-03-27 18:00:38 +110049static inline int
50xfs_ialloc_cluster_alignment(
51 xfs_alloc_arg_t *args)
52{
53 if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
54 args->mp->m_sb.sb_inoalignmt >=
55 XFS_B_TO_FSBT(args->mp, XFS_INODE_CLUSTER_SIZE(args->mp)))
56 return args->mp->m_sb.sb_inoalignmt;
57 return 1;
58}
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60/*
Christoph Hellwig21875502009-08-31 20:58:21 -030061 * Lookup a record by ino in the btree given by cur.
Christoph Hellwigfe033cc2008-10-30 16:56:09 +110062 */
Christoph Hellwig81e25172009-09-01 19:56:55 -040063int /* error */
Christoph Hellwig21875502009-08-31 20:58:21 -030064xfs_inobt_lookup(
Christoph Hellwigfe033cc2008-10-30 16:56:09 +110065 struct xfs_btree_cur *cur, /* btree cursor */
66 xfs_agino_t ino, /* starting inode of chunk */
Christoph Hellwig21875502009-08-31 20:58:21 -030067 xfs_lookup_t dir, /* <=, >=, == */
Christoph Hellwigfe033cc2008-10-30 16:56:09 +110068 int *stat) /* success/failure */
69{
70 cur->bc_rec.i.ir_startino = ino;
Christoph Hellwig21875502009-08-31 20:58:21 -030071 cur->bc_rec.i.ir_freecount = 0;
72 cur->bc_rec.i.ir_free = 0;
73 return xfs_btree_lookup(cur, dir, stat);
Christoph Hellwigfe033cc2008-10-30 16:56:09 +110074}
75
76/*
Christoph Hellwigafabc242009-08-31 20:57:03 -030077 * Update the record referred to by cur to the value given.
Christoph Hellwig278d0ca2008-10-30 16:56:32 +110078 * This either works (return 0) or gets an EFSCORRUPTED error.
79 */
80STATIC int /* error */
81xfs_inobt_update(
82 struct xfs_btree_cur *cur, /* btree cursor */
Christoph Hellwigafabc242009-08-31 20:57:03 -030083 xfs_inobt_rec_incore_t *irec) /* btree record */
Christoph Hellwig278d0ca2008-10-30 16:56:32 +110084{
85 union xfs_btree_rec rec;
86
Christoph Hellwigafabc242009-08-31 20:57:03 -030087 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
88 rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
89 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
Christoph Hellwig278d0ca2008-10-30 16:56:32 +110090 return xfs_btree_update(cur, &rec);
91}
92
93/*
Christoph Hellwig8cc938f2008-10-30 16:58:11 +110094 * Get the data from the pointed-to record.
95 */
96int /* error */
97xfs_inobt_get_rec(
98 struct xfs_btree_cur *cur, /* btree cursor */
Christoph Hellwig2e287a72009-08-31 20:56:58 -030099 xfs_inobt_rec_incore_t *irec, /* btree record */
Christoph Hellwig8cc938f2008-10-30 16:58:11 +1100100 int *stat) /* output: success/failure */
101{
102 union xfs_btree_rec *rec;
103 int error;
104
105 error = xfs_btree_get_rec(cur, &rec, stat);
106 if (!error && *stat == 1) {
Christoph Hellwig2e287a72009-08-31 20:56:58 -0300107 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
108 irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
109 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
Christoph Hellwig8cc938f2008-10-30 16:58:11 +1100110 }
111 return error;
112}
113
114/*
Dave Chinner0b48db82009-08-31 20:57:09 -0300115 * Verify that the number of free inodes in the AGI is correct.
116 */
117#ifdef DEBUG
118STATIC int
119xfs_check_agi_freecount(
120 struct xfs_btree_cur *cur,
121 struct xfs_agi *agi)
122{
123 if (cur->bc_nlevels == 1) {
124 xfs_inobt_rec_incore_t rec;
125 int freecount = 0;
126 int error;
127 int i;
128
Christoph Hellwig21875502009-08-31 20:58:21 -0300129 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
Dave Chinner0b48db82009-08-31 20:57:09 -0300130 if (error)
131 return error;
132
133 do {
134 error = xfs_inobt_get_rec(cur, &rec, &i);
135 if (error)
136 return error;
137
138 if (i) {
139 freecount += rec.ir_freecount;
140 error = xfs_btree_increment(cur, 0, &i);
141 if (error)
142 return error;
143 }
144 } while (i == 1);
145
146 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
147 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
148 }
149 return 0;
150}
151#else
152#define xfs_check_agi_freecount(cur, agi) 0
153#endif
154
155/*
Dave Chinner28c8e412013-06-27 16:04:55 +1000156 * Initialise a new set of inodes. When called without a transaction context
157 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
158 * than logging them (which in a transaction context puts them into the AIL
159 * for writeback rather than the xfsbufd queue).
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300160 */
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000161int
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300162xfs_ialloc_inode_init(
163 struct xfs_mount *mp,
164 struct xfs_trans *tp,
Dave Chinner28c8e412013-06-27 16:04:55 +1000165 struct list_head *buffer_list,
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300166 xfs_agnumber_t agno,
167 xfs_agblock_t agbno,
168 xfs_agblock_t length,
169 unsigned int gen)
170{
171 struct xfs_buf *fbuf;
172 struct xfs_dinode *free;
173 int blks_per_cluster, nbufs, ninodes;
174 int version;
175 int i, j;
176 xfs_daddr_t d;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100177 xfs_ino_t ino = 0;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300178
179 /*
180 * Loop over the new block(s), filling in the inodes.
181 * For small block sizes, manipulate the inodes in buffers
182 * which are multiples of the blocks size.
183 */
184 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
185 blks_per_cluster = 1;
186 nbufs = length;
187 ninodes = mp->m_sb.sb_inopblock;
188 } else {
189 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
190 mp->m_sb.sb_blocksize;
191 nbufs = length / blks_per_cluster;
192 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
193 }
194
195 /*
Christoph Hellwig93848a92013-04-03 16:11:17 +1100196 * Figure out what version number to use in the inodes we create. If
197 * the superblock version has caught up to the one that supports the new
198 * inode format, then use the new inode version. Otherwise use the old
199 * version so that old kernels will continue to be able to use the file
200 * system.
201 *
202 * For v3 inodes, we also need to write the inode number into the inode,
203 * so calculate the first inode number of the chunk here as
204 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
205 * across multiple filesystem blocks (such as a cluster) and so cannot
206 * be used in the cluster buffer loop below.
207 *
208 * Further, because we are writing the inode directly into the buffer
209 * and calculating a CRC on the entire inode, we have ot log the entire
210 * inode so that the entire range the CRC covers is present in the log.
211 * That means for v3 inode we log the entire buffer rather than just the
212 * inode cores.
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300213 */
Christoph Hellwig93848a92013-04-03 16:11:17 +1100214 if (xfs_sb_version_hascrc(&mp->m_sb)) {
215 version = 3;
216 ino = XFS_AGINO_TO_INO(mp, agno,
217 XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000218
219 /*
220 * log the initialisation that is about to take place as an
221 * logical operation. This means the transaction does not
222 * need to log the physical changes to the inode buffers as log
223 * recovery will know what initialisation is actually needed.
224 * Hence we only need to log the buffers as "ordered" buffers so
225 * they track in the AIL as if they were physically logged.
226 */
227 if (tp)
228 xfs_icreate_log(tp, agno, agbno, XFS_IALLOC_INODES(mp),
229 mp->m_sb.sb_inodesize, length, gen);
Christoph Hellwig93848a92013-04-03 16:11:17 +1100230 } else if (xfs_sb_version_hasnlink(&mp->m_sb))
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300231 version = 2;
232 else
233 version = 1;
234
235 for (j = 0; j < nbufs; j++) {
236 /*
237 * Get the block.
238 */
239 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
240 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
Dave Chinner7c4cebe2012-11-23 14:24:23 +1100241 mp->m_bsize * blks_per_cluster,
242 XBF_UNMAPPED);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000243 if (!fbuf)
244 return ENOMEM;
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000245
246 /* Initialize the inode buffers and log them appropriately. */
Dave Chinner1813dd62012-11-14 17:54:40 +1100247 fbuf->b_ops = &xfs_inode_buf_ops;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100248 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300249 for (i = 0; i < ninodes; i++) {
250 int ioffset = i << mp->m_sb.sb_inodelog;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100251 uint isize = xfs_dinode_size(version);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300252
253 free = xfs_make_iptr(mp, fbuf, i);
254 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
255 free->di_version = version;
256 free->di_gen = cpu_to_be32(gen);
257 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig93848a92013-04-03 16:11:17 +1100258
259 if (version == 3) {
260 free->di_ino = cpu_to_be64(ino);
261 ino++;
262 uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid);
263 xfs_dinode_calc_crc(mp, free);
Dave Chinner28c8e412013-06-27 16:04:55 +1000264 } else if (tp) {
Christoph Hellwig93848a92013-04-03 16:11:17 +1100265 /* just log the inode core */
266 xfs_trans_log_buf(tp, fbuf, ioffset,
267 ioffset + isize - 1);
268 }
269 }
Dave Chinner28c8e412013-06-27 16:04:55 +1000270
271 if (tp) {
272 /*
273 * Mark the buffer as an inode allocation buffer so it
274 * sticks in AIL at the point of this allocation
275 * transaction. This ensures the they are on disk before
276 * the tail of the log can be moved past this
277 * transaction (i.e. by preventing relogging from moving
278 * it forward in the log).
279 */
280 xfs_trans_inode_alloc_buf(tp, fbuf);
281 if (version == 3) {
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000282 /*
283 * Mark the buffer as ordered so that they are
284 * not physically logged in the transaction but
285 * still tracked in the AIL as part of the
286 * transaction and pin the log appropriately.
287 */
288 xfs_trans_ordered_buf(tp, fbuf);
Dave Chinner28c8e412013-06-27 16:04:55 +1000289 xfs_trans_log_buf(tp, fbuf, 0,
290 BBTOB(fbuf->b_length) - 1);
291 }
292 } else {
293 fbuf->b_flags |= XBF_DONE;
294 xfs_buf_delwri_queue(fbuf, buffer_list);
295 xfs_buf_relse(fbuf);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300296 }
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300297 }
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000298 return 0;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300299}
300
301/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 * Allocate new inodes in the allocation group specified by agbp.
303 * Return 0 for success, else error code.
304 */
305STATIC int /* error code or 0 */
306xfs_ialloc_ag_alloc(
307 xfs_trans_t *tp, /* transaction pointer */
308 xfs_buf_t *agbp, /* alloc group buffer */
309 int *alloc)
310{
311 xfs_agi_t *agi; /* allocation group header */
312 xfs_alloc_arg_t args; /* allocation argument structure */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 xfs_btree_cur_t *cur; /* inode btree cursor */
David Chinner92821e22007-05-24 15:26:31 +1000314 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 int error;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300316 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 xfs_agino_t newino; /* new first inode's number */
318 xfs_agino_t newlen; /* new number of inodes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 xfs_agino_t thisino; /* current inode number, for loop */
Glen Overby3ccb8b52006-03-29 09:52:28 +1000320 int isaligned = 0; /* inode allocation at stripe unit */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 /* boundary */
Dave Chinner44b56e02010-01-11 11:47:43 +0000322 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Mark Tinguelya0041682012-09-20 13:16:45 -0500324 memset(&args, 0, sizeof(args));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 args.tp = tp;
326 args.mp = tp->t_mountp;
327
328 /*
329 * Locking will ensure that we don't have two callers in here
330 * at one time.
331 */
332 newlen = XFS_IALLOC_INODES(args.mp);
333 if (args.mp->m_maxicount &&
334 args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
335 return XFS_ERROR(ENOSPC);
336 args.minlen = args.maxlen = XFS_IALLOC_BLOCKS(args.mp);
337 /*
Glen Overby3ccb8b52006-03-29 09:52:28 +1000338 * First try to allocate inodes contiguous with the last-allocated
339 * chunk of inodes. If the filesystem is striped, this will fill
340 * an entire stripe unit with inodes.
Dave Chinner28c8e412013-06-27 16:04:55 +1000341 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 agi = XFS_BUF_TO_AGI(agbp);
Glen Overby3ccb8b52006-03-29 09:52:28 +1000343 newino = be32_to_cpu(agi->agi_newino);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300344 agno = be32_to_cpu(agi->agi_seqno);
Nathan Scott019ff2d2006-04-11 15:45:05 +1000345 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
346 XFS_IALLOC_BLOCKS(args.mp);
347 if (likely(newino != NULLAGINO &&
348 (args.agbno < be32_to_cpu(agi->agi_length)))) {
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300349 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
Glen Overby3ccb8b52006-03-29 09:52:28 +1000350 args.type = XFS_ALLOCTYPE_THIS_BNO;
Glen Overby3ccb8b52006-03-29 09:52:28 +1000351 args.prod = 1;
David Chinner75de2a92008-03-27 18:00:38 +1100352
Glen Overby3ccb8b52006-03-29 09:52:28 +1000353 /*
David Chinner75de2a92008-03-27 18:00:38 +1100354 * We need to take into account alignment here to ensure that
355 * we don't modify the free list if we fail to have an exact
356 * block. If we don't have an exact match, and every oher
357 * attempt allocation attempt fails, we'll end up cancelling
358 * a dirty transaction and shutting down.
359 *
360 * For an exact allocation, alignment must be 1,
361 * however we need to take cluster alignment into account when
362 * fixing up the freelist. Use the minalignslop field to
363 * indicate that extra blocks might be required for alignment,
364 * but not to use them in the actual exact allocation.
Glen Overby3ccb8b52006-03-29 09:52:28 +1000365 */
David Chinner75de2a92008-03-27 18:00:38 +1100366 args.alignment = 1;
367 args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
368
369 /* Allow space for the inode btree to split. */
Christoph Hellwig0d87e652009-02-09 08:37:14 +0100370 args.minleft = args.mp->m_in_maxlevels - 1;
Glen Overby3ccb8b52006-03-29 09:52:28 +1000371 if ((error = xfs_alloc_vextent(&args)))
372 return error;
373 } else
374 args.fsbno = NULLFSBLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Glen Overby3ccb8b52006-03-29 09:52:28 +1000376 if (unlikely(args.fsbno == NULLFSBLOCK)) {
377 /*
378 * Set the alignment for the allocation.
379 * If stripe alignment is turned on then align at stripe unit
380 * boundary.
Nathan Scott019ff2d2006-04-11 15:45:05 +1000381 * If the cluster size is smaller than a filesystem block
382 * then we're doing I/O for inodes in filesystem block size
Glen Overby3ccb8b52006-03-29 09:52:28 +1000383 * pieces, so don't need alignment anyway.
384 */
385 isaligned = 0;
386 if (args.mp->m_sinoalign) {
387 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
388 args.alignment = args.mp->m_dalign;
389 isaligned = 1;
David Chinner75de2a92008-03-27 18:00:38 +1100390 } else
391 args.alignment = xfs_ialloc_cluster_alignment(&args);
Glen Overby3ccb8b52006-03-29 09:52:28 +1000392 /*
393 * Need to figure out where to allocate the inode blocks.
394 * Ideally they should be spaced out through the a.g.
395 * For now, just allocate blocks up front.
396 */
397 args.agbno = be32_to_cpu(agi->agi_root);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300398 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
Glen Overby3ccb8b52006-03-29 09:52:28 +1000399 /*
400 * Allocate a fixed-size extent of inodes.
401 */
402 args.type = XFS_ALLOCTYPE_NEAR_BNO;
Glen Overby3ccb8b52006-03-29 09:52:28 +1000403 args.prod = 1;
404 /*
405 * Allow space for the inode btree to split.
406 */
Christoph Hellwig0d87e652009-02-09 08:37:14 +0100407 args.minleft = args.mp->m_in_maxlevels - 1;
Glen Overby3ccb8b52006-03-29 09:52:28 +1000408 if ((error = xfs_alloc_vextent(&args)))
409 return error;
410 }
Nathan Scott019ff2d2006-04-11 15:45:05 +1000411
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 /*
413 * If stripe alignment is turned on, then try again with cluster
414 * alignment.
415 */
416 if (isaligned && args.fsbno == NULLFSBLOCK) {
417 args.type = XFS_ALLOCTYPE_NEAR_BNO;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100418 args.agbno = be32_to_cpu(agi->agi_root);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300419 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
David Chinner75de2a92008-03-27 18:00:38 +1100420 args.alignment = xfs_ialloc_cluster_alignment(&args);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 if ((error = xfs_alloc_vextent(&args)))
422 return error;
423 }
424
425 if (args.fsbno == NULLFSBLOCK) {
426 *alloc = 0;
427 return 0;
428 }
429 ASSERT(args.len == args.minlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
David Chinner359346a2008-04-29 12:53:32 +1000431 /*
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300432 * Stamp and write the inode buffers.
433 *
David Chinner359346a2008-04-29 12:53:32 +1000434 * Seed the new inode cluster with a random generation number. This
435 * prevents short-term reuse of generation numbers if a chunk is
436 * freed and then immediately reallocated. We use random numbers
437 * rather than a linear progression to prevent the next generation
438 * number from being easily guessable.
439 */
Dave Chinner28c8e412013-06-27 16:04:55 +1000440 error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno,
Akinobu Mitaecb34032013-03-04 21:58:20 +0900441 args.len, prandom_u32());
Christoph Hellwigd42f08f2008-11-28 14:23:38 +1100442
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000443 if (error)
444 return error;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300445 /*
446 * Convert the results.
447 */
448 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800449 be32_add_cpu(&agi->agi_count, newlen);
450 be32_add_cpu(&agi->agi_freecount, newlen);
Dave Chinner44b56e02010-01-11 11:47:43 +0000451 pag = xfs_perag_get(args.mp, agno);
452 pag->pagi_freecount += newlen;
453 xfs_perag_put(pag);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100454 agi->agi_newino = cpu_to_be32(newino);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300455
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 /*
457 * Insert records describing the new inode chunk into the btree.
458 */
Christoph Hellwig561f7d12008-10-30 16:53:59 +1100459 cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 for (thisino = newino;
461 thisino < newino + newlen;
462 thisino += XFS_INODES_PER_CHUNK) {
Christoph Hellwig21875502009-08-31 20:58:21 -0300463 cur->bc_rec.i.ir_startino = thisino;
464 cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
465 cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
466 error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
467 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
469 return error;
470 }
471 ASSERT(i == 0);
Christoph Hellwig21875502009-08-31 20:58:21 -0300472 error = xfs_btree_insert(cur, &i);
473 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
475 return error;
476 }
477 ASSERT(i == 1);
478 }
479 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
480 /*
481 * Log allocation group header fields
482 */
483 xfs_ialloc_log_agi(tp, agbp,
484 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
485 /*
486 * Modify/log superblock values for inode count and inode free count.
487 */
488 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
489 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
490 *alloc = 1;
491 return 0;
492}
493
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000494STATIC xfs_agnumber_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495xfs_ialloc_next_ag(
496 xfs_mount_t *mp)
497{
498 xfs_agnumber_t agno;
499
500 spin_lock(&mp->m_agirotor_lock);
501 agno = mp->m_agirotor;
Carlos Maiolino8aea3ff2012-09-20 10:32:36 -0300502 if (++mp->m_agirotor >= mp->m_maxagi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 mp->m_agirotor = 0;
504 spin_unlock(&mp->m_agirotor_lock);
505
506 return agno;
507}
508
509/*
510 * Select an allocation group to look for a free inode in, based on the parent
Zhi Yong Wu2f21ff12013-08-12 03:14:58 +0000511 * inode and the mode. Return the allocation group buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 */
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400513STATIC xfs_agnumber_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514xfs_ialloc_ag_select(
515 xfs_trans_t *tp, /* transaction pointer */
516 xfs_ino_t parent, /* parent directory inode number */
Al Viro576b1d62011-07-26 02:50:15 -0400517 umode_t mode, /* bits set to indicate file type */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 int okalloc) /* ok to allocate more space */
519{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
521 xfs_agnumber_t agno; /* current ag number */
522 int flags; /* alloc buffer locking flags */
523 xfs_extlen_t ineed; /* blocks needed for inode allocation */
524 xfs_extlen_t longest = 0; /* longest extent available */
525 xfs_mount_t *mp; /* mount point structure */
526 int needspace; /* file mode implies space allocated */
527 xfs_perag_t *pag; /* per allocation group data */
528 xfs_agnumber_t pagno; /* parent (starting) ag number */
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400529 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
531 /*
532 * Files of these types need at least one block if length > 0
533 * (and they won't fit in the inode, but that's hard to figure out).
534 */
535 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
536 mp = tp->t_mountp;
537 agcount = mp->m_maxagi;
538 if (S_ISDIR(mode))
539 pagno = xfs_ialloc_next_ag(mp);
540 else {
541 pagno = XFS_INO_TO_AGNO(mp, parent);
542 if (pagno >= agcount)
543 pagno = 0;
544 }
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 ASSERT(pagno < agcount);
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 /*
549 * Loop through allocation groups, looking for one with a little
550 * free space in it. Note we don't look for free inodes, exactly.
551 * Instead, we include whether there is a need to allocate inodes
552 * to mean that blocks must be allocated for them,
553 * if none are currently free.
554 */
555 agno = pagno;
556 flags = XFS_ALLOC_FLAG_TRYLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 for (;;) {
Dave Chinner44b56e02010-01-11 11:47:43 +0000558 pag = xfs_perag_get(mp, agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 if (!pag->pagi_inodeok) {
560 xfs_ialloc_next_ag(mp);
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400561 goto nextag;
562 }
563
564 if (!pag->pagi_init) {
565 error = xfs_ialloc_pagi_init(mp, tp, agno);
566 if (error)
567 goto nextag;
568 }
569
570 if (pag->pagi_freecount) {
571 xfs_perag_put(pag);
572 return agno;
573 }
574
575 if (!okalloc)
576 goto nextag;
577
578 if (!pag->pagf_init) {
579 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
580 if (error)
581 goto nextag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 }
583
584 /*
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400585 * Is there enough free space for the file plus a block of
586 * inodes? (if we need to allocate some)?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400588 ineed = XFS_IALLOC_BLOCKS(mp);
589 longest = pag->pagf_longest;
590 if (!longest)
591 longest = pag->pagf_flcount > 0;
592
593 if (pag->pagf_freeblks >= needspace + ineed &&
594 longest >= ineed) {
595 xfs_perag_put(pag);
596 return agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598nextag:
Dave Chinner44b56e02010-01-11 11:47:43 +0000599 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 /*
601 * No point in iterating over the rest, if we're shutting
602 * down.
603 */
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000604 if (XFS_FORCED_SHUTDOWN(mp))
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400605 return NULLAGNUMBER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 agno++;
607 if (agno >= agcount)
608 agno = 0;
609 if (agno == pagno) {
Dave Chinner1c1c6eb2010-01-11 11:47:44 +0000610 if (flags == 0)
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400611 return NULLAGNUMBER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 flags = 0;
613 }
614 }
615}
616
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300617/*
618 * Try to retrieve the next record to the left/right from the current one.
619 */
620STATIC int
621xfs_ialloc_next_rec(
622 struct xfs_btree_cur *cur,
623 xfs_inobt_rec_incore_t *rec,
624 int *done,
625 int left)
626{
627 int error;
628 int i;
629
630 if (left)
631 error = xfs_btree_decrement(cur, 0, &i);
632 else
633 error = xfs_btree_increment(cur, 0, &i);
634
635 if (error)
636 return error;
637 *done = !i;
638 if (i) {
639 error = xfs_inobt_get_rec(cur, rec, &i);
640 if (error)
641 return error;
642 XFS_WANT_CORRUPTED_RETURN(i == 1);
643 }
644
645 return 0;
646}
647
Dave Chinnerbd169562009-08-31 20:58:28 -0300648STATIC int
649xfs_ialloc_get_rec(
650 struct xfs_btree_cur *cur,
651 xfs_agino_t agino,
652 xfs_inobt_rec_incore_t *rec,
Jie Liu43df2ee2013-06-24 13:37:23 +0800653 int *done)
Dave Chinnerbd169562009-08-31 20:58:28 -0300654{
655 int error;
656 int i;
657
658 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
659 if (error)
660 return error;
661 *done = !i;
662 if (i) {
663 error = xfs_inobt_get_rec(cur, rec, &i);
664 if (error)
665 return error;
666 XFS_WANT_CORRUPTED_RETURN(i == 1);
667 }
668
669 return 0;
670}
Dave Chinner0b48db82009-08-31 20:57:09 -0300671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672/*
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400673 * Allocate an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 *
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400675 * The caller selected an AG for us, and made sure that free inodes are
676 * available.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 */
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400678STATIC int
679xfs_dialloc_ag(
680 struct xfs_trans *tp,
681 struct xfs_buf *agbp,
682 xfs_ino_t parent,
683 xfs_ino_t *inop)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400685 struct xfs_mount *mp = tp->t_mountp;
686 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
687 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
688 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
689 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
690 struct xfs_perag *pag;
691 struct xfs_btree_cur *cur, *tcur;
692 struct xfs_inobt_rec_incore rec, trec;
693 xfs_ino_t ino;
694 int error;
695 int offset;
696 int i, j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
Dave Chinner44b56e02010-01-11 11:47:43 +0000698 pag = xfs_perag_get(mp, agno);
Dave Chinnerbd169562009-08-31 20:58:28 -0300699
Christoph Hellwig4bb61062012-07-04 10:54:48 -0400700 ASSERT(pag->pagi_init);
701 ASSERT(pag->pagi_inodeok);
702 ASSERT(pag->pagi_freecount > 0);
703
Dave Chinnerbd169562009-08-31 20:58:28 -0300704 restart_pagno:
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400705 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 /*
707 * If pagino is 0 (this is the root inode allocation) use newino.
708 * This must work because we've just allocated some.
709 */
710 if (!pagino)
Christoph Hellwig16259e72005-11-02 15:11:25 +1100711 pagino = be32_to_cpu(agi->agi_newino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Dave Chinner0b48db82009-08-31 20:57:09 -0300713 error = xfs_check_agi_freecount(cur, agi);
714 if (error)
715 goto error0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 /*
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300718 * If in the same AG as the parent, try to get near the parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 */
720 if (pagno == agno) {
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300721 int doneleft; /* done, to the left */
722 int doneright; /* done, to the right */
Dave Chinnerbd169562009-08-31 20:58:28 -0300723 int searchdistance = 10;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300724
Christoph Hellwig21875502009-08-31 20:58:21 -0300725 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300726 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 goto error0;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300728 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
729
730 error = xfs_inobt_get_rec(cur, &rec, &j);
731 if (error)
732 goto error0;
Brian Fosterb1210992013-08-27 17:15:45 -0400733 XFS_WANT_CORRUPTED_GOTO(j == 1, error0);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300734
735 if (rec.ir_freecount > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 /*
737 * Found a free inode in the same chunk
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300738 * as the parent, done.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 */
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300740 goto alloc_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300742
743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 /*
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300745 * In the same AG as parent, but parent's chunk is full.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300748 /* duplicate the cursor, search left & right simultaneously */
749 error = xfs_btree_dup_cursor(cur, &tcur);
750 if (error)
751 goto error0;
752
Dave Chinnerbd169562009-08-31 20:58:28 -0300753 /*
754 * Skip to last blocks looked up if same parent inode.
755 */
756 if (pagino != NULLAGINO &&
757 pag->pagl_pagino == pagino &&
758 pag->pagl_leftrec != NULLAGINO &&
759 pag->pagl_rightrec != NULLAGINO) {
760 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
Jie Liu43df2ee2013-06-24 13:37:23 +0800761 &trec, &doneleft);
Dave Chinnerbd169562009-08-31 20:58:28 -0300762 if (error)
763 goto error1;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300764
Dave Chinnerbd169562009-08-31 20:58:28 -0300765 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
Jie Liu43df2ee2013-06-24 13:37:23 +0800766 &rec, &doneright);
Dave Chinnerbd169562009-08-31 20:58:28 -0300767 if (error)
768 goto error1;
769 } else {
770 /* search left with tcur, back up 1 record */
771 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
772 if (error)
773 goto error1;
774
775 /* search right with cur, go forward 1 record. */
776 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
777 if (error)
778 goto error1;
779 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300780
781 /*
782 * Loop until we find an inode chunk with a free inode.
783 */
784 while (!doneleft || !doneright) {
785 int useleft; /* using left inode chunk this time */
786
Dave Chinnerbd169562009-08-31 20:58:28 -0300787 if (!--searchdistance) {
788 /*
789 * Not in range - save last search
790 * location and allocate a new inode
791 */
Eric Sandeen3b826382009-10-30 09:27:07 +0100792 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
Dave Chinnerbd169562009-08-31 20:58:28 -0300793 pag->pagl_leftrec = trec.ir_startino;
794 pag->pagl_rightrec = rec.ir_startino;
795 pag->pagl_pagino = pagino;
796 goto newino;
797 }
798
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300799 /* figure out the closer block if both are valid. */
800 if (!doneleft && !doneright) {
801 useleft = pagino -
802 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
803 rec.ir_startino - pagino;
804 } else {
805 useleft = !doneleft;
806 }
807
808 /* free inodes to the left? */
809 if (useleft && trec.ir_freecount) {
810 rec = trec;
811 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
812 cur = tcur;
Dave Chinnerbd169562009-08-31 20:58:28 -0300813
814 pag->pagl_leftrec = trec.ir_startino;
815 pag->pagl_rightrec = rec.ir_startino;
816 pag->pagl_pagino = pagino;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300817 goto alloc_inode;
818 }
819
820 /* free inodes to the right? */
821 if (!useleft && rec.ir_freecount) {
822 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
Dave Chinnerbd169562009-08-31 20:58:28 -0300823
824 pag->pagl_leftrec = trec.ir_startino;
825 pag->pagl_rightrec = rec.ir_startino;
826 pag->pagl_pagino = pagino;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300827 goto alloc_inode;
828 }
829
830 /* get next record to check */
831 if (useleft) {
832 error = xfs_ialloc_next_rec(tcur, &trec,
833 &doneleft, 1);
834 } else {
835 error = xfs_ialloc_next_rec(cur, &rec,
836 &doneright, 0);
837 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 }
Dave Chinnerbd169562009-08-31 20:58:28 -0300841
842 /*
843 * We've reached the end of the btree. because
844 * we are only searching a small chunk of the
845 * btree each search, there is obviously free
846 * inodes closer to the parent inode than we
847 * are now. restart the search again.
848 */
849 pag->pagl_pagino = NULLAGINO;
850 pag->pagl_leftrec = NULLAGINO;
851 pag->pagl_rightrec = NULLAGINO;
852 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
853 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
854 goto restart_pagno;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 /*
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300858 * In a different AG from the parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 * See if the most recently allocated block has any free.
860 */
Dave Chinnerbd169562009-08-31 20:58:28 -0300861newino:
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200862 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
Christoph Hellwig21875502009-08-31 20:58:21 -0300863 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
864 XFS_LOOKUP_EQ, &i);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300865 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 goto error0;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300867
868 if (i == 1) {
869 error = xfs_inobt_get_rec(cur, &rec, &j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 if (error)
871 goto error0;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300872
873 if (j == 1 && rec.ir_freecount > 0) {
874 /*
875 * The last chunk allocated in the group
876 * still has a free inode.
877 */
878 goto alloc_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 }
880 }
Dave Chinnerbd169562009-08-31 20:58:28 -0300881 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300882
Dave Chinnerbd169562009-08-31 20:58:28 -0300883 /*
884 * None left in the last group, search the whole AG
885 */
886 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
887 if (error)
888 goto error0;
889 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
890
891 for (;;) {
892 error = xfs_inobt_get_rec(cur, &rec, &i);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300893 if (error)
894 goto error0;
895 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
Dave Chinnerbd169562009-08-31 20:58:28 -0300896 if (rec.ir_freecount > 0)
897 break;
898 error = xfs_btree_increment(cur, 0, &i);
899 if (error)
900 goto error0;
901 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -0300903
904alloc_inode:
Christoph Hellwig824c3132012-07-04 10:54:45 -0400905 offset = xfs_lowbit64(rec.ir_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 ASSERT(offset >= 0);
907 ASSERT(offset < XFS_INODES_PER_CHUNK);
908 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
909 XFS_INODES_PER_CHUNK) == 0);
910 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
Christoph Hellwig0d87e652009-02-09 08:37:14 +0100911 rec.ir_free &= ~XFS_INOBT_MASK(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 rec.ir_freecount--;
Christoph Hellwigafabc242009-08-31 20:57:03 -0300913 error = xfs_inobt_update(cur, &rec);
914 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 goto error0;
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800916 be32_add_cpu(&agi->agi_freecount, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
Dave Chinner44b56e02010-01-11 11:47:43 +0000918 pag->pagi_freecount--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Dave Chinner0b48db82009-08-31 20:57:09 -0300920 error = xfs_check_agi_freecount(cur, agi);
921 if (error)
922 goto error0;
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
925 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
Dave Chinner44b56e02010-01-11 11:47:43 +0000926 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 *inop = ino;
928 return 0;
929error1:
930 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
931error0:
932 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
Dave Chinner44b56e02010-01-11 11:47:43 +0000933 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 return error;
935}
936
937/*
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400938 * Allocate an inode on disk.
939 *
940 * Mode is used to tell whether the new inode will need space, and whether it
941 * is a directory.
942 *
943 * This function is designed to be called twice if it has to do an allocation
944 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
945 * If an inode is available without having to performn an allocation, an inode
Carlos Maiolinocd856db2012-10-20 11:08:19 -0300946 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
947 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
948 * The caller should then commit the current transaction, allocate a
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400949 * new transaction, and call xfs_dialloc() again, passing in the previous value
950 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
951 * buffer is locked across the two calls, the second call is guaranteed to have
952 * a free inode available.
953 *
954 * Once we successfully pick an inode its number is returned and the on-disk
955 * data structures are updated. The inode itself is not read in, since doing so
956 * would break ordering constraints with xfs_reclaim.
957 */
958int
959xfs_dialloc(
960 struct xfs_trans *tp,
961 xfs_ino_t parent,
962 umode_t mode,
963 int okalloc,
964 struct xfs_buf **IO_agbp,
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400965 xfs_ino_t *inop)
966{
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400967 struct xfs_mount *mp = tp->t_mountp;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400968 struct xfs_buf *agbp;
969 xfs_agnumber_t agno;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400970 int error;
971 int ialloced;
972 int noroom = 0;
Christoph Hellwigbe60fe52012-07-04 10:54:50 -0400973 xfs_agnumber_t start_agno;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400974 struct xfs_perag *pag;
975
Christoph Hellwig4bb61062012-07-04 10:54:48 -0400976 if (*IO_agbp) {
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400977 /*
Christoph Hellwig4bb61062012-07-04 10:54:48 -0400978 * If the caller passes in a pointer to the AGI buffer,
979 * continue where we left off before. In this case, we
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400980 * know that the allocation group has free inodes.
981 */
982 agbp = *IO_agbp;
Christoph Hellwig4bb61062012-07-04 10:54:48 -0400983 goto out_alloc;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400984 }
Christoph Hellwig4bb61062012-07-04 10:54:48 -0400985
986 /*
987 * We do not have an agbp, so select an initial allocation
988 * group for inode allocation.
989 */
Christoph Hellwigbe60fe52012-07-04 10:54:50 -0400990 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
991 if (start_agno == NULLAGNUMBER) {
Christoph Hellwig4bb61062012-07-04 10:54:48 -0400992 *inop = NULLFSINO;
993 return 0;
994 }
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400995
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -0400996 /*
997 * If we have already hit the ceiling of inode blocks then clear
998 * okalloc so we scan all available agi structures for a free
999 * inode.
1000 */
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001001 if (mp->m_maxicount &&
1002 mp->m_sb.sb_icount + XFS_IALLOC_INODES(mp) > mp->m_maxicount) {
1003 noroom = 1;
1004 okalloc = 0;
1005 }
1006
1007 /*
1008 * Loop until we find an allocation group that either has free inodes
1009 * or in which we can allocate some inodes. Iterate through the
1010 * allocation groups upward, wrapping at the end.
1011 */
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001012 agno = start_agno;
1013 for (;;) {
1014 pag = xfs_perag_get(mp, agno);
1015 if (!pag->pagi_inodeok) {
1016 xfs_ialloc_next_ag(mp);
1017 goto nextag;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001018 }
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001019
1020 if (!pag->pagi_init) {
1021 error = xfs_ialloc_pagi_init(mp, tp, agno);
1022 if (error)
1023 goto out_error;
1024 }
1025
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001026 /*
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001027 * Do a first racy fast path check if this AG is usable.
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001028 */
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001029 if (!pag->pagi_freecount && !okalloc)
1030 goto nextag;
1031
Christoph Hellwigc4982112012-08-07 02:02:02 -04001032 /*
1033 * Then read in the AGI buffer and recheck with the AGI buffer
1034 * lock held.
1035 */
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001036 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1037 if (error)
1038 goto out_error;
1039
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001040 if (pag->pagi_freecount) {
1041 xfs_perag_put(pag);
1042 goto out_alloc;
1043 }
1044
Christoph Hellwigc4982112012-08-07 02:02:02 -04001045 if (!okalloc)
1046 goto nextag_relse_buffer;
1047
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001048
1049 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1050 if (error) {
1051 xfs_trans_brelse(tp, agbp);
1052
1053 if (error != ENOSPC)
1054 goto out_error;
1055
1056 xfs_perag_put(pag);
1057 *inop = NULLFSINO;
1058 return 0;
1059 }
1060
1061 if (ialloced) {
1062 /*
1063 * We successfully allocated some inodes, return
1064 * the current context to the caller so that it
1065 * can commit the current transaction and call
1066 * us again where we left off.
1067 */
1068 ASSERT(pag->pagi_freecount > 0);
1069 xfs_perag_put(pag);
1070
1071 *IO_agbp = agbp;
1072 *inop = NULLFSINO;
1073 return 0;
1074 }
1075
Christoph Hellwigc4982112012-08-07 02:02:02 -04001076nextag_relse_buffer:
1077 xfs_trans_brelse(tp, agbp);
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001078nextag:
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001079 xfs_perag_put(pag);
1080 if (++agno == mp->m_sb.sb_agcount)
1081 agno = 0;
1082 if (agno == start_agno) {
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001083 *inop = NULLFSINO;
1084 return noroom ? ENOSPC : 0;
1085 }
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001086 }
1087
Christoph Hellwig4bb61062012-07-04 10:54:48 -04001088out_alloc:
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001089 *IO_agbp = NULL;
1090 return xfs_dialloc_ag(tp, agbp, parent, inop);
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001091out_error:
1092 xfs_perag_put(pag);
1093 return XFS_ERROR(error);
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001094}
1095
1096/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 * Free disk inode. Carefully avoids touching the incore inode, all
1098 * manipulations incore are the caller's responsibility.
1099 * The on-disk inode is not changed by this operation, only the
1100 * btree (free inode mask) is changed.
1101 */
1102int
1103xfs_difree(
1104 xfs_trans_t *tp, /* transaction pointer */
1105 xfs_ino_t inode, /* inode to be freed */
1106 xfs_bmap_free_t *flist, /* extents to free */
1107 int *delete, /* set if inode cluster was deleted */
1108 xfs_ino_t *first_ino) /* first inode in deleted cluster */
1109{
1110 /* REFERENCED */
1111 xfs_agblock_t agbno; /* block number containing inode */
1112 xfs_buf_t *agbp; /* buffer containing allocation group header */
1113 xfs_agino_t agino; /* inode number relative to allocation group */
1114 xfs_agnumber_t agno; /* allocation group number */
1115 xfs_agi_t *agi; /* allocation group header */
1116 xfs_btree_cur_t *cur; /* inode btree cursor */
1117 int error; /* error return value */
1118 int i; /* result code */
1119 int ilen; /* inodes in an inode cluster */
1120 xfs_mount_t *mp; /* mount structure for filesystem */
1121 int off; /* offset of inode in inode chunk */
Christoph Hellwig61a25842006-09-28 10:57:04 +10001122 xfs_inobt_rec_incore_t rec; /* btree record */
Dave Chinner44b56e02010-01-11 11:47:43 +00001123 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
1125 mp = tp->t_mountp;
1126
1127 /*
1128 * Break up inode number into its components.
1129 */
1130 agno = XFS_INO_TO_AGNO(mp, inode);
1131 if (agno >= mp->m_sb.sb_agcount) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001132 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1133 __func__, agno, mp->m_sb.sb_agcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 ASSERT(0);
1135 return XFS_ERROR(EINVAL);
1136 }
1137 agino = XFS_INO_TO_AGINO(mp, inode);
1138 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001139 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1140 __func__, (unsigned long long)inode,
1141 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 ASSERT(0);
1143 return XFS_ERROR(EINVAL);
1144 }
1145 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1146 if (agbno >= mp->m_sb.sb_agblocks) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001147 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1148 __func__, agbno, mp->m_sb.sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 ASSERT(0);
1150 return XFS_ERROR(EINVAL);
1151 }
1152 /*
1153 * Get the allocation group header.
1154 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001157 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1158 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 return error;
1160 }
1161 agi = XFS_BUF_TO_AGI(agbp);
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001162 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
Christoph Hellwig16259e72005-11-02 15:11:25 +11001163 ASSERT(agbno < be32_to_cpu(agi->agi_length));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 /*
1165 * Initialize the cursor.
1166 */
Christoph Hellwig561f7d12008-10-30 16:53:59 +11001167 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
Dave Chinner0b48db82009-08-31 20:57:09 -03001169 error = xfs_check_agi_freecount(cur, agi);
1170 if (error)
1171 goto error0;
1172
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 /*
1174 * Look for the entry describing this inode.
1175 */
Christoph Hellwig21875502009-08-31 20:58:21 -03001176 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001177 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1178 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 goto error0;
1180 }
1181 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
Christoph Hellwig2e287a72009-08-31 20:56:58 -03001182 error = xfs_inobt_get_rec(cur, &rec, &i);
1183 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001184 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1185 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 goto error0;
1187 }
1188 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1189 /*
1190 * Get the offset in the inode chunk.
1191 */
1192 off = agino - rec.ir_startino;
1193 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
Christoph Hellwig0d87e652009-02-09 08:37:14 +01001194 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 /*
1196 * Mark the inode free & increment the count.
1197 */
Christoph Hellwig0d87e652009-02-09 08:37:14 +01001198 rec.ir_free |= XFS_INOBT_MASK(off);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 rec.ir_freecount++;
1200
1201 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001202 * When an inode cluster is free, it becomes eligible for removal
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 */
Josef Jeff Sipek1bd960e2008-02-29 13:58:40 +11001204 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 (rec.ir_freecount == XFS_IALLOC_INODES(mp))) {
1206
1207 *delete = 1;
1208 *first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1209
1210 /*
1211 * Remove the inode cluster from the AGI B+Tree, adjust the
1212 * AGI and Superblock inode counts, and mark the disk space
1213 * to be freed when the transaction is committed.
1214 */
1215 ilen = XFS_IALLOC_INODES(mp);
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001216 be32_add_cpu(&agi->agi_count, -ilen);
1217 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
Dave Chinner44b56e02010-01-11 11:47:43 +00001219 pag = xfs_perag_get(mp, agno);
1220 pag->pagi_freecount -= ilen - 1;
1221 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1223 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1224
Christoph Hellwig91cca5df2008-10-30 16:58:01 +11001225 if ((error = xfs_btree_delete(cur, &i))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001226 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1227 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 goto error0;
1229 }
1230
1231 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp,
Ben Myers8e825e32013-12-10 14:59:31 -06001232 agno, XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 XFS_IALLOC_BLOCKS(mp), flist, mp);
1234 } else {
1235 *delete = 0;
1236
Christoph Hellwigafabc242009-08-31 20:57:03 -03001237 error = xfs_inobt_update(cur, &rec);
1238 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001239 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1240 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 goto error0;
1242 }
Christoph Hellwigafabc242009-08-31 20:57:03 -03001243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 /*
1245 * Change the inode free counts and log the ag/sb changes.
1246 */
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001247 be32_add_cpu(&agi->agi_freecount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
Dave Chinner44b56e02010-01-11 11:47:43 +00001249 pag = xfs_perag_get(mp, agno);
1250 pag->pagi_freecount++;
1251 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1253 }
1254
Dave Chinner0b48db82009-08-31 20:57:09 -03001255 error = xfs_check_agi_freecount(cur, agi);
1256 if (error)
1257 goto error0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1260 return 0;
1261
1262error0:
1263 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1264 return error;
1265}
1266
Dave Chinner7124fe02010-06-24 11:15:33 +10001267STATIC int
1268xfs_imap_lookup(
1269 struct xfs_mount *mp,
1270 struct xfs_trans *tp,
1271 xfs_agnumber_t agno,
1272 xfs_agino_t agino,
1273 xfs_agblock_t agbno,
1274 xfs_agblock_t *chunk_agbno,
1275 xfs_agblock_t *offset_agbno,
1276 int flags)
1277{
1278 struct xfs_inobt_rec_incore rec;
1279 struct xfs_btree_cur *cur;
1280 struct xfs_buf *agbp;
Dave Chinner7124fe02010-06-24 11:15:33 +10001281 int error;
1282 int i;
1283
1284 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1285 if (error) {
Dave Chinner53487782011-03-07 10:05:35 +11001286 xfs_alert(mp,
1287 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1288 __func__, error, agno);
Dave Chinner7124fe02010-06-24 11:15:33 +10001289 return error;
1290 }
1291
1292 /*
Dave Chinner4536f2a2010-08-24 11:42:30 +10001293 * Lookup the inode record for the given agino. If the record cannot be
1294 * found, then it's an invalid inode number and we should abort. Once
1295 * we have a record, we need to ensure it contains the inode number
1296 * we are looking up.
Dave Chinner7124fe02010-06-24 11:15:33 +10001297 */
1298 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
Dave Chinner4536f2a2010-08-24 11:42:30 +10001299 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
Dave Chinner7124fe02010-06-24 11:15:33 +10001300 if (!error) {
1301 if (i)
1302 error = xfs_inobt_get_rec(cur, &rec, &i);
1303 if (!error && i == 0)
1304 error = EINVAL;
1305 }
1306
1307 xfs_trans_brelse(tp, agbp);
1308 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1309 if (error)
1310 return error;
1311
Dave Chinner4536f2a2010-08-24 11:42:30 +10001312 /* check that the returned record contains the required inode */
1313 if (rec.ir_startino > agino ||
1314 rec.ir_startino + XFS_IALLOC_INODES(mp) <= agino)
1315 return EINVAL;
1316
Dave Chinner7124fe02010-06-24 11:15:33 +10001317 /* for untrusted inodes check it is allocated first */
Dave Chinner19207792010-06-24 11:15:47 +10001318 if ((flags & XFS_IGET_UNTRUSTED) &&
Dave Chinner7124fe02010-06-24 11:15:33 +10001319 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1320 return EINVAL;
1321
1322 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1323 *offset_agbno = agbno - *chunk_agbno;
1324 return 0;
1325}
1326
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327/*
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001328 * Return the location of the inode in imap, for mapping it into a buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330int
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001331xfs_imap(
1332 xfs_mount_t *mp, /* file system mount structure */
1333 xfs_trans_t *tp, /* transaction pointer */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 xfs_ino_t ino, /* inode to locate */
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001335 struct xfs_imap *imap, /* location map structure */
1336 uint flags) /* flags for inode btree lookup */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337{
1338 xfs_agblock_t agbno; /* block number of inode in the alloc group */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 xfs_agino_t agino; /* inode number within alloc group */
1340 xfs_agnumber_t agno; /* allocation group number */
1341 int blks_per_cluster; /* num blocks per inode cluster */
1342 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 int error; /* error code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 int offset; /* index of inode in its buffer */
Dave Chinner836a94a2013-08-12 20:49:44 +10001346 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
1348 ASSERT(ino != NULLFSINO);
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 /*
1351 * Split up the inode number into its parts.
1352 */
1353 agno = XFS_INO_TO_AGNO(mp, ino);
1354 agino = XFS_INO_TO_AGINO(mp, ino);
1355 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1356 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1357 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1358#ifdef DEBUG
Dave Chinner19207792010-06-24 11:15:47 +10001359 /*
1360 * Don't output diagnostic information for untrusted inodes
1361 * as they can be invalid without implying corruption.
1362 */
1363 if (flags & XFS_IGET_UNTRUSTED)
Nathan Scott4d1a2ed2006-06-09 17:12:28 +10001364 return XFS_ERROR(EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 if (agno >= mp->m_sb.sb_agcount) {
Dave Chinner53487782011-03-07 10:05:35 +11001366 xfs_alert(mp,
1367 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1368 __func__, agno, mp->m_sb.sb_agcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 }
1370 if (agbno >= mp->m_sb.sb_agblocks) {
Dave Chinner53487782011-03-07 10:05:35 +11001371 xfs_alert(mp,
1372 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1373 __func__, (unsigned long long)agbno,
1374 (unsigned long)mp->m_sb.sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 }
1376 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
Dave Chinner53487782011-03-07 10:05:35 +11001377 xfs_alert(mp,
1378 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1379 __func__, ino,
1380 XFS_AGINO_TO_INO(mp, agno, agino));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 }
Nathan Scott745b1f472006-09-28 11:02:23 +10001382 xfs_stack_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383#endif /* DEBUG */
1384 return XFS_ERROR(EINVAL);
1385 }
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001386
Dave Chinner7124fe02010-06-24 11:15:33 +10001387 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_blocklog;
1388
1389 /*
1390 * For bulkstat and handle lookups, we have an untrusted inode number
1391 * that we have to verify is valid. We cannot do this just by reading
1392 * the inode buffer as it may have been unlinked and removed leaving
1393 * inodes in stale state on disk. Hence we have to do a btree lookup
1394 * in all cases where an untrusted inode number is passed.
1395 */
Dave Chinner19207792010-06-24 11:15:47 +10001396 if (flags & XFS_IGET_UNTRUSTED) {
Dave Chinner7124fe02010-06-24 11:15:33 +10001397 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1398 &chunk_agbno, &offset_agbno, flags);
1399 if (error)
1400 return error;
1401 goto out_map;
1402 }
1403
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001404 /*
1405 * If the inode cluster size is the same as the blocksize or
1406 * smaller we get to the buffer by simple arithmetics.
1407 */
1408 if (XFS_INODE_CLUSTER_SIZE(mp) <= mp->m_sb.sb_blocksize) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 offset = XFS_INO_TO_OFFSET(mp, ino);
1410 ASSERT(offset < mp->m_sb.sb_inopblock);
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001411
1412 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1413 imap->im_len = XFS_FSB_TO_BB(mp, 1);
1414 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 return 0;
1416 }
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001417
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001418 /*
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001419 * If the inode chunks are aligned then use simple maths to
1420 * find the location. Otherwise we have to do a btree
1421 * lookup to find the location.
1422 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (mp->m_inoalign_mask) {
1424 offset_agbno = agbno & mp->m_inoalign_mask;
1425 chunk_agbno = agbno - offset_agbno;
1426 } else {
Dave Chinner7124fe02010-06-24 11:15:33 +10001427 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1428 &chunk_agbno, &offset_agbno, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 if (error)
1430 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 }
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001432
Dave Chinner7124fe02010-06-24 11:15:33 +10001433out_map:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 ASSERT(agbno >= chunk_agbno);
1435 cluster_agbno = chunk_agbno +
1436 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
1437 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1438 XFS_INO_TO_OFFSET(mp, ino);
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001439
1440 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1441 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1442 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1443
1444 /*
1445 * If the inode number maps to a block outside the bounds
1446 * of the file system then return NULL rather than calling
1447 * read_buf and panicing when we get an error from the
1448 * driver.
1449 */
1450 if ((imap->im_blkno + imap->im_len) >
1451 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
Dave Chinner53487782011-03-07 10:05:35 +11001452 xfs_alert(mp,
1453 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1454 __func__, (unsigned long long) imap->im_blkno,
Christoph Hellwig94e1b692008-11-28 14:23:41 +11001455 (unsigned long long) imap->im_len,
1456 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1457 return XFS_ERROR(EINVAL);
1458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460}
1461
1462/*
1463 * Compute and fill in value of m_in_maxlevels.
1464 */
1465void
1466xfs_ialloc_compute_maxlevels(
1467 xfs_mount_t *mp) /* file system mount structure */
1468{
1469 int level;
1470 uint maxblocks;
1471 uint maxleafents;
1472 int minleafrecs;
1473 int minnoderecs;
1474
1475 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1476 XFS_INODES_PER_CHUNK_LOG;
1477 minleafrecs = mp->m_alloc_mnr[0];
1478 minnoderecs = mp->m_alloc_mnr[1];
1479 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1480 for (level = 1; maxblocks > 1; level++)
1481 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1482 mp->m_in_maxlevels = level;
1483}
1484
1485/*
1486 * Log specified fields for the ag hdr (inode section)
1487 */
1488void
1489xfs_ialloc_log_agi(
1490 xfs_trans_t *tp, /* transaction pointer */
1491 xfs_buf_t *bp, /* allocation group header buffer */
1492 int fields) /* bitmask of fields to log */
1493{
1494 int first; /* first byte number */
1495 int last; /* last byte number */
1496 static const short offsets[] = { /* field starting offsets */
1497 /* keep in sync with bit definitions */
1498 offsetof(xfs_agi_t, agi_magicnum),
1499 offsetof(xfs_agi_t, agi_versionnum),
1500 offsetof(xfs_agi_t, agi_seqno),
1501 offsetof(xfs_agi_t, agi_length),
1502 offsetof(xfs_agi_t, agi_count),
1503 offsetof(xfs_agi_t, agi_root),
1504 offsetof(xfs_agi_t, agi_level),
1505 offsetof(xfs_agi_t, agi_freecount),
1506 offsetof(xfs_agi_t, agi_newino),
1507 offsetof(xfs_agi_t, agi_dirino),
1508 offsetof(xfs_agi_t, agi_unlinked),
1509 sizeof(xfs_agi_t)
1510 };
1511#ifdef DEBUG
1512 xfs_agi_t *agi; /* allocation group header */
1513
1514 agi = XFS_BUF_TO_AGI(bp);
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001515 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516#endif
1517 /*
1518 * Compute byte offsets for the first and last fields.
1519 */
1520 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1521 /*
1522 * Log the allocation group inode header buffer.
1523 */
Dave Chinner61fe1352013-04-03 16:11:30 +11001524 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 xfs_trans_log_buf(tp, bp, first, last);
1526}
1527
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001528#ifdef DEBUG
1529STATIC void
1530xfs_check_agi_unlinked(
1531 struct xfs_agi *agi)
1532{
1533 int i;
1534
1535 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1536 ASSERT(agi->agi_unlinked[i]);
1537}
1538#else
1539#define xfs_check_agi_unlinked(agi)
1540#endif
1541
Dave Chinner983d09f2013-04-03 16:11:15 +11001542static bool
Dave Chinner612cfbf2012-11-14 17:52:32 +11001543xfs_agi_verify(
Dave Chinner3702ce62012-11-12 22:54:05 +11001544 struct xfs_buf *bp)
1545{
1546 struct xfs_mount *mp = bp->b_target->bt_mount;
1547 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
Dave Chinner3702ce62012-11-12 22:54:05 +11001548
Dave Chinner983d09f2013-04-03 16:11:15 +11001549 if (xfs_sb_version_hascrc(&mp->m_sb) &&
1550 !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid))
1551 return false;
Dave Chinner3702ce62012-11-12 22:54:05 +11001552 /*
1553 * Validate the magic number of the agi block.
1554 */
Dave Chinner983d09f2013-04-03 16:11:15 +11001555 if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
1556 return false;
1557 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
1558 return false;
Dave Chinner3702ce62012-11-12 22:54:05 +11001559
1560 /*
1561 * during growfs operations, the perag is not fully initialised,
1562 * so we can't use it for any useful checking. growfs ensures we can't
1563 * use it by using uncached buffers that don't have the perag attached
1564 * so we can detect and avoid this problem.
1565 */
Dave Chinner983d09f2013-04-03 16:11:15 +11001566 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
1567 return false;
Dave Chinner3702ce62012-11-12 22:54:05 +11001568
Dave Chinner3702ce62012-11-12 22:54:05 +11001569 xfs_check_agi_unlinked(agi);
Dave Chinner983d09f2013-04-03 16:11:15 +11001570 return true;
Dave Chinner612cfbf2012-11-14 17:52:32 +11001571}
1572
Dave Chinner1813dd62012-11-14 17:54:40 +11001573static void
1574xfs_agi_read_verify(
Dave Chinner612cfbf2012-11-14 17:52:32 +11001575 struct xfs_buf *bp)
1576{
Dave Chinner983d09f2013-04-03 16:11:15 +11001577 struct xfs_mount *mp = bp->b_target->bt_mount;
1578 int agi_ok = 1;
1579
1580 if (xfs_sb_version_hascrc(&mp->m_sb))
1581 agi_ok = xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
1582 offsetof(struct xfs_agi, agi_crc));
1583 agi_ok = agi_ok && xfs_agi_verify(bp);
1584
1585 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IALLOC_READ_AGI,
1586 XFS_RANDOM_IALLOC_READ_AGI))) {
1587 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
1588 xfs_buf_ioerror(bp, EFSCORRUPTED);
1589 }
Dave Chinner612cfbf2012-11-14 17:52:32 +11001590}
1591
Dave Chinnerb0f539d2012-11-14 17:53:49 +11001592static void
Dave Chinner1813dd62012-11-14 17:54:40 +11001593xfs_agi_write_verify(
Dave Chinner612cfbf2012-11-14 17:52:32 +11001594 struct xfs_buf *bp)
1595{
Dave Chinner983d09f2013-04-03 16:11:15 +11001596 struct xfs_mount *mp = bp->b_target->bt_mount;
1597 struct xfs_buf_log_item *bip = bp->b_fspriv;
1598
1599 if (!xfs_agi_verify(bp)) {
1600 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
1601 xfs_buf_ioerror(bp, EFSCORRUPTED);
1602 return;
1603 }
1604
1605 if (!xfs_sb_version_hascrc(&mp->m_sb))
1606 return;
1607
1608 if (bip)
1609 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
1610 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length),
1611 offsetof(struct xfs_agi, agi_crc));
Dave Chinner3702ce62012-11-12 22:54:05 +11001612}
1613
Dave Chinner1813dd62012-11-14 17:54:40 +11001614const struct xfs_buf_ops xfs_agi_buf_ops = {
1615 .verify_read = xfs_agi_read_verify,
1616 .verify_write = xfs_agi_write_verify,
1617};
1618
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619/*
1620 * Read in the allocation group header (inode allocation section)
1621 */
1622int
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001623xfs_read_agi(
1624 struct xfs_mount *mp, /* file system mount structure */
1625 struct xfs_trans *tp, /* transaction pointer */
1626 xfs_agnumber_t agno, /* allocation group number */
1627 struct xfs_buf **bpp) /* allocation group hdr buf */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628{
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001629 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
Dave Chinnerd1230312013-11-01 15:27:19 +11001631 trace_xfs_read_agi(mp, agno);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001632
Dave Chinnerd1230312013-11-01 15:27:19 +11001633 ASSERT(agno != NULLAGNUMBER);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001634 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
Dave Chinner1813dd62012-11-14 17:54:40 +11001636 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if (error)
1638 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001639
Chandra Seetharaman5a52c2a582011-07-22 23:39:51 +00001640 ASSERT(!xfs_buf_geterror(*bpp));
Christoph Hellwig38f23232011-10-10 16:52:45 +00001641 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001642 return 0;
1643}
1644
1645int
1646xfs_ialloc_read_agi(
1647 struct xfs_mount *mp, /* file system mount structure */
1648 struct xfs_trans *tp, /* transaction pointer */
1649 xfs_agnumber_t agno, /* allocation group number */
1650 struct xfs_buf **bpp) /* allocation group hdr buf */
1651{
1652 struct xfs_agi *agi; /* allocation group header */
1653 struct xfs_perag *pag; /* per allocation group data */
1654 int error;
1655
Dave Chinnerd1230312013-11-01 15:27:19 +11001656 trace_xfs_ialloc_read_agi(mp, agno);
1657
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001658 error = xfs_read_agi(mp, tp, agno, bpp);
1659 if (error)
1660 return error;
1661
1662 agi = XFS_BUF_TO_AGI(*bpp);
Dave Chinner44b56e02010-01-11 11:47:43 +00001663 pag = xfs_perag_get(mp, agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 if (!pag->pagi_init) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11001665 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
David Chinner92821e22007-05-24 15:26:31 +10001666 pag->pagi_count = be32_to_cpu(agi->agi_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 pag->pagi_init = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 }
1669
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001670 /*
1671 * It's possible for these to be out of sync if
1672 * we are in the middle of a forced shutdown.
1673 */
1674 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1675 XFS_FORCED_SHUTDOWN(mp));
Dave Chinner44b56e02010-01-11 11:47:43 +00001676 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 return 0;
1678}
David Chinner92821e22007-05-24 15:26:31 +10001679
1680/*
1681 * Read in the agi to initialise the per-ag data in the mount structure
1682 */
1683int
1684xfs_ialloc_pagi_init(
1685 xfs_mount_t *mp, /* file system mount structure */
1686 xfs_trans_t *tp, /* transaction pointer */
1687 xfs_agnumber_t agno) /* allocation group number */
1688{
1689 xfs_buf_t *bp = NULL;
1690 int error;
1691
1692 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1693 if (error)
1694 return error;
1695 if (bp)
1696 xfs_trans_brelse(tp, bp);
1697 return 0;
1698}