blob: 54deb2d12ac6bdfeb55ca2e05b56c81ed169e1a5 [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 Chinner70a98832013-10-23 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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_sb.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110028#include "xfs_btree.h"
29#include "xfs_ialloc.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110030#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_rtalloc.h"
33#include "xfs_error.h"
34#include "xfs_bmap.h"
Dave Chinner983d09f2013-04-03 16:11:15 +110035#include "xfs_cksum.h"
Dave Chinner239880e2013-10-23 10:50:10 +110036#include "xfs_trans.h"
Dave Chinner983d09f2013-04-03 16:11:15 +110037#include "xfs_buf_item.h"
Dave Chinnerddf6ad02013-06-27 16:04:56 +100038#include "xfs_icreate_item.h"
Dave Chinner7bb85ef2013-08-12 20:49:34 +100039#include "xfs_icache.h"
Dave Chinnerd1230312013-11-01 15:27:19 +110040#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43/*
44 * Allocation group level functions.
45 */
David Chinner75de2a92008-03-27 18:00:38 +110046static inline int
47xfs_ialloc_cluster_alignment(
Dave Chinner7a1df152014-12-04 09:42:21 +110048 struct xfs_mount *mp)
David Chinner75de2a92008-03-27 18:00:38 +110049{
Dave Chinner7a1df152014-12-04 09:42:21 +110050 if (xfs_sb_version_hasalign(&mp->m_sb) &&
51 mp->m_sb.sb_inoalignmt >=
52 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
53 return mp->m_sb.sb_inoalignmt;
David Chinner75de2a92008-03-27 18:00:38 +110054 return 1;
55}
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57/*
Christoph Hellwig21875502009-08-31 20:58:21 -030058 * Lookup a record by ino in the btree given by cur.
Christoph Hellwigfe033cc2008-10-30 16:56:09 +110059 */
Christoph Hellwig81e25172009-09-01 19:56:55 -040060int /* error */
Christoph Hellwig21875502009-08-31 20:58:21 -030061xfs_inobt_lookup(
Christoph Hellwigfe033cc2008-10-30 16:56:09 +110062 struct xfs_btree_cur *cur, /* btree cursor */
63 xfs_agino_t ino, /* starting inode of chunk */
Christoph Hellwig21875502009-08-31 20:58:21 -030064 xfs_lookup_t dir, /* <=, >=, == */
Christoph Hellwigfe033cc2008-10-30 16:56:09 +110065 int *stat) /* success/failure */
66{
67 cur->bc_rec.i.ir_startino = ino;
Brian Foster54190402015-05-29 09:03:04 +100068 cur->bc_rec.i.ir_holemask = 0;
69 cur->bc_rec.i.ir_count = 0;
Christoph Hellwig21875502009-08-31 20:58:21 -030070 cur->bc_rec.i.ir_freecount = 0;
71 cur->bc_rec.i.ir_free = 0;
72 return xfs_btree_lookup(cur, dir, stat);
Christoph Hellwigfe033cc2008-10-30 16:56:09 +110073}
74
75/*
Christoph Hellwigafabc242009-08-31 20:57:03 -030076 * Update the record referred to by cur to the value given.
Christoph Hellwig278d0ca2008-10-30 16:56:32 +110077 * This either works (return 0) or gets an EFSCORRUPTED error.
78 */
79STATIC int /* error */
80xfs_inobt_update(
81 struct xfs_btree_cur *cur, /* btree cursor */
Christoph Hellwigafabc242009-08-31 20:57:03 -030082 xfs_inobt_rec_incore_t *irec) /* btree record */
Christoph Hellwig278d0ca2008-10-30 16:56:32 +110083{
84 union xfs_btree_rec rec;
85
Christoph Hellwigafabc242009-08-31 20:57:03 -030086 rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
Brian Foster54190402015-05-29 09:03:04 +100087 if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
88 rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
89 rec.inobt.ir_u.sp.ir_count = irec->ir_count;
90 rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
91 } else {
92 /* ir_holemask/ir_count not supported on-disk */
93 rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
94 }
Christoph Hellwigafabc242009-08-31 20:57:03 -030095 rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
Christoph Hellwig278d0ca2008-10-30 16:56:32 +110096 return xfs_btree_update(cur, &rec);
97}
98
99/*
Christoph Hellwig8cc938f2008-10-30 16:58:11 +1100100 * Get the data from the pointed-to record.
101 */
102int /* error */
103xfs_inobt_get_rec(
104 struct xfs_btree_cur *cur, /* btree cursor */
Christoph Hellwig2e287a72009-08-31 20:56:58 -0300105 xfs_inobt_rec_incore_t *irec, /* btree record */
Christoph Hellwig8cc938f2008-10-30 16:58:11 +1100106 int *stat) /* output: success/failure */
107{
108 union xfs_btree_rec *rec;
109 int error;
110
111 error = xfs_btree_get_rec(cur, &rec, stat);
Brian Foster54190402015-05-29 09:03:04 +1000112 if (error || *stat == 0)
113 return error;
114
115 irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
116 if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
117 irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
118 irec->ir_count = rec->inobt.ir_u.sp.ir_count;
119 irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
120 } else {
121 /*
122 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
123 * values for full inode chunks.
124 */
125 irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
126 irec->ir_count = XFS_INODES_PER_CHUNK;
127 irec->ir_freecount =
128 be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
Christoph Hellwig8cc938f2008-10-30 16:58:11 +1100129 }
Brian Foster54190402015-05-29 09:03:04 +1000130 irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
131
132 return 0;
Christoph Hellwig8cc938f2008-10-30 16:58:11 +1100133}
134
135/*
Brian Foster0aa0a752014-04-24 16:00:53 +1000136 * Insert a single inobt record. Cursor must already point to desired location.
137 */
138STATIC int
139xfs_inobt_insert_rec(
140 struct xfs_btree_cur *cur,
Brian Foster54190402015-05-29 09:03:04 +1000141 __uint16_t holemask,
142 __uint8_t count,
Brian Foster0aa0a752014-04-24 16:00:53 +1000143 __int32_t freecount,
144 xfs_inofree_t free,
145 int *stat)
146{
Brian Foster54190402015-05-29 09:03:04 +1000147 cur->bc_rec.i.ir_holemask = holemask;
148 cur->bc_rec.i.ir_count = count;
Brian Foster0aa0a752014-04-24 16:00:53 +1000149 cur->bc_rec.i.ir_freecount = freecount;
150 cur->bc_rec.i.ir_free = free;
151 return xfs_btree_insert(cur, stat);
152}
153
154/*
155 * Insert records describing a newly allocated inode chunk into the inobt.
156 */
157STATIC int
158xfs_inobt_insert(
159 struct xfs_mount *mp,
160 struct xfs_trans *tp,
161 struct xfs_buf *agbp,
162 xfs_agino_t newino,
163 xfs_agino_t newlen,
164 xfs_btnum_t btnum)
165{
166 struct xfs_btree_cur *cur;
167 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
168 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
169 xfs_agino_t thisino;
170 int i;
171 int error;
172
173 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
174
175 for (thisino = newino;
176 thisino < newino + newlen;
177 thisino += XFS_INODES_PER_CHUNK) {
178 error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
179 if (error) {
180 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
181 return error;
182 }
183 ASSERT(i == 0);
184
Brian Foster54190402015-05-29 09:03:04 +1000185 error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
186 XFS_INODES_PER_CHUNK,
187 XFS_INODES_PER_CHUNK,
Brian Foster0aa0a752014-04-24 16:00:53 +1000188 XFS_INOBT_ALL_FREE, &i);
189 if (error) {
190 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
191 return error;
192 }
193 ASSERT(i == 1);
194 }
195
196 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
197
198 return 0;
199}
200
201/*
Dave Chinner0b48db82009-08-31 20:57:09 -0300202 * Verify that the number of free inodes in the AGI is correct.
203 */
204#ifdef DEBUG
205STATIC int
206xfs_check_agi_freecount(
207 struct xfs_btree_cur *cur,
208 struct xfs_agi *agi)
209{
210 if (cur->bc_nlevels == 1) {
211 xfs_inobt_rec_incore_t rec;
212 int freecount = 0;
213 int error;
214 int i;
215
Christoph Hellwig21875502009-08-31 20:58:21 -0300216 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
Dave Chinner0b48db82009-08-31 20:57:09 -0300217 if (error)
218 return error;
219
220 do {
221 error = xfs_inobt_get_rec(cur, &rec, &i);
222 if (error)
223 return error;
224
225 if (i) {
226 freecount += rec.ir_freecount;
227 error = xfs_btree_increment(cur, 0, &i);
228 if (error)
229 return error;
230 }
231 } while (i == 1);
232
233 if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
234 ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
235 }
236 return 0;
237}
238#else
239#define xfs_check_agi_freecount(cur, agi) 0
240#endif
241
242/*
Dave Chinner28c8e412013-06-27 16:04:55 +1000243 * Initialise a new set of inodes. When called without a transaction context
244 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
245 * than logging them (which in a transaction context puts them into the AIL
246 * for writeback rather than the xfsbufd queue).
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300247 */
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000248int
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300249xfs_ialloc_inode_init(
250 struct xfs_mount *mp,
251 struct xfs_trans *tp,
Dave Chinner28c8e412013-06-27 16:04:55 +1000252 struct list_head *buffer_list,
Brian Foster463958a2015-05-29 09:05:49 +1000253 int icount,
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300254 xfs_agnumber_t agno,
255 xfs_agblock_t agbno,
256 xfs_agblock_t length,
257 unsigned int gen)
258{
259 struct xfs_buf *fbuf;
260 struct xfs_dinode *free;
Jie Liu6e0c7b82013-12-13 15:51:49 +1100261 int nbufs, blks_per_cluster, inodes_per_cluster;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300262 int version;
263 int i, j;
264 xfs_daddr_t d;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100265 xfs_ino_t ino = 0;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300266
267 /*
Jie Liu6e0c7b82013-12-13 15:51:49 +1100268 * Loop over the new block(s), filling in the inodes. For small block
269 * sizes, manipulate the inodes in buffers which are multiples of the
270 * blocks size.
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300271 */
Jie Liu6e0c7b82013-12-13 15:51:49 +1100272 blks_per_cluster = xfs_icluster_size_fsb(mp);
273 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
274 nbufs = length / blks_per_cluster;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300275
276 /*
Christoph Hellwig93848a92013-04-03 16:11:17 +1100277 * Figure out what version number to use in the inodes we create. If
278 * the superblock version has caught up to the one that supports the new
279 * inode format, then use the new inode version. Otherwise use the old
280 * version so that old kernels will continue to be able to use the file
281 * system.
282 *
283 * For v3 inodes, we also need to write the inode number into the inode,
284 * so calculate the first inode number of the chunk here as
285 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
286 * across multiple filesystem blocks (such as a cluster) and so cannot
287 * be used in the cluster buffer loop below.
288 *
289 * Further, because we are writing the inode directly into the buffer
290 * and calculating a CRC on the entire inode, we have ot log the entire
291 * inode so that the entire range the CRC covers is present in the log.
292 * That means for v3 inode we log the entire buffer rather than just the
293 * inode cores.
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300294 */
Christoph Hellwig93848a92013-04-03 16:11:17 +1100295 if (xfs_sb_version_hascrc(&mp->m_sb)) {
296 version = 3;
297 ino = XFS_AGINO_TO_INO(mp, agno,
298 XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000299
300 /*
301 * log the initialisation that is about to take place as an
302 * logical operation. This means the transaction does not
303 * need to log the physical changes to the inode buffers as log
304 * recovery will know what initialisation is actually needed.
305 * Hence we only need to log the buffers as "ordered" buffers so
306 * they track in the AIL as if they were physically logged.
307 */
308 if (tp)
Brian Foster463958a2015-05-29 09:05:49 +1000309 xfs_icreate_log(tp, agno, agbno, icount,
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000310 mp->m_sb.sb_inodesize, length, gen);
Dave Chinner263997a2014-05-20 07:46:40 +1000311 } else
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300312 version = 2;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300313
314 for (j = 0; j < nbufs; j++) {
315 /*
316 * Get the block.
317 */
318 d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
319 fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
Dave Chinner7c4cebe2012-11-23 14:24:23 +1100320 mp->m_bsize * blks_per_cluster,
321 XBF_UNMAPPED);
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000322 if (!fbuf)
Dave Chinner24513372014-06-25 14:58:08 +1000323 return -ENOMEM;
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000324
325 /* Initialize the inode buffers and log them appropriately. */
Dave Chinner1813dd62012-11-14 17:54:40 +1100326 fbuf->b_ops = &xfs_inode_buf_ops;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100327 xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
Jie Liu6e0c7b82013-12-13 15:51:49 +1100328 for (i = 0; i < inodes_per_cluster; i++) {
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300329 int ioffset = i << mp->m_sb.sb_inodelog;
Christoph Hellwig93848a92013-04-03 16:11:17 +1100330 uint isize = xfs_dinode_size(version);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300331
332 free = xfs_make_iptr(mp, fbuf, i);
333 free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
334 free->di_version = version;
335 free->di_gen = cpu_to_be32(gen);
336 free->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig93848a92013-04-03 16:11:17 +1100337
338 if (version == 3) {
339 free->di_ino = cpu_to_be64(ino);
340 ino++;
Eric Sandeence748ea2015-07-29 11:53:31 +1000341 uuid_copy(&free->di_uuid,
342 &mp->m_sb.sb_meta_uuid);
Christoph Hellwig93848a92013-04-03 16:11:17 +1100343 xfs_dinode_calc_crc(mp, free);
Dave Chinner28c8e412013-06-27 16:04:55 +1000344 } else if (tp) {
Christoph Hellwig93848a92013-04-03 16:11:17 +1100345 /* just log the inode core */
346 xfs_trans_log_buf(tp, fbuf, ioffset,
347 ioffset + isize - 1);
348 }
349 }
Dave Chinner28c8e412013-06-27 16:04:55 +1000350
351 if (tp) {
352 /*
353 * Mark the buffer as an inode allocation buffer so it
354 * sticks in AIL at the point of this allocation
355 * transaction. This ensures the they are on disk before
356 * the tail of the log can be moved past this
357 * transaction (i.e. by preventing relogging from moving
358 * it forward in the log).
359 */
360 xfs_trans_inode_alloc_buf(tp, fbuf);
361 if (version == 3) {
Dave Chinnerddf6ad02013-06-27 16:04:56 +1000362 /*
363 * Mark the buffer as ordered so that they are
364 * not physically logged in the transaction but
365 * still tracked in the AIL as part of the
366 * transaction and pin the log appropriately.
367 */
368 xfs_trans_ordered_buf(tp, fbuf);
Dave Chinner28c8e412013-06-27 16:04:55 +1000369 xfs_trans_log_buf(tp, fbuf, 0,
370 BBTOB(fbuf->b_length) - 1);
371 }
372 } else {
373 fbuf->b_flags |= XBF_DONE;
374 xfs_buf_delwri_queue(fbuf, buffer_list);
375 xfs_buf_relse(fbuf);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300376 }
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300377 }
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000378 return 0;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300379}
380
381/*
Brian Foster56d11152015-05-29 09:18:32 +1000382 * Align startino and allocmask for a recently allocated sparse chunk such that
383 * they are fit for insertion (or merge) into the on-disk inode btrees.
384 *
385 * Background:
386 *
387 * When enabled, sparse inode support increases the inode alignment from cluster
388 * size to inode chunk size. This means that the minimum range between two
389 * non-adjacent inode records in the inobt is large enough for a full inode
390 * record. This allows for cluster sized, cluster aligned block allocation
391 * without need to worry about whether the resulting inode record overlaps with
392 * another record in the tree. Without this basic rule, we would have to deal
393 * with the consequences of overlap by potentially undoing recent allocations in
394 * the inode allocation codepath.
395 *
396 * Because of this alignment rule (which is enforced on mount), there are two
397 * inobt possibilities for newly allocated sparse chunks. One is that the
398 * aligned inode record for the chunk covers a range of inodes not already
399 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
400 * other is that a record already exists at the aligned startino that considers
401 * the newly allocated range as sparse. In the latter case, record content is
402 * merged in hope that sparse inode chunks fill to full chunks over time.
403 */
404STATIC void
405xfs_align_sparse_ino(
406 struct xfs_mount *mp,
407 xfs_agino_t *startino,
408 uint16_t *allocmask)
409{
410 xfs_agblock_t agbno;
411 xfs_agblock_t mod;
412 int offset;
413
414 agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
415 mod = agbno % mp->m_sb.sb_inoalignmt;
416 if (!mod)
417 return;
418
419 /* calculate the inode offset and align startino */
420 offset = mod << mp->m_sb.sb_inopblog;
421 *startino -= offset;
422
423 /*
424 * Since startino has been aligned down, left shift allocmask such that
425 * it continues to represent the same physical inodes relative to the
426 * new startino.
427 */
428 *allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
429}
430
431/*
432 * Determine whether the source inode record can merge into the target. Both
433 * records must be sparse, the inode ranges must match and there must be no
434 * allocation overlap between the records.
435 */
436STATIC bool
437__xfs_inobt_can_merge(
438 struct xfs_inobt_rec_incore *trec, /* tgt record */
439 struct xfs_inobt_rec_incore *srec) /* src record */
440{
441 uint64_t talloc;
442 uint64_t salloc;
443
444 /* records must cover the same inode range */
445 if (trec->ir_startino != srec->ir_startino)
446 return false;
447
448 /* both records must be sparse */
449 if (!xfs_inobt_issparse(trec->ir_holemask) ||
450 !xfs_inobt_issparse(srec->ir_holemask))
451 return false;
452
453 /* both records must track some inodes */
454 if (!trec->ir_count || !srec->ir_count)
455 return false;
456
457 /* can't exceed capacity of a full record */
458 if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
459 return false;
460
461 /* verify there is no allocation overlap */
462 talloc = xfs_inobt_irec_to_allocmask(trec);
463 salloc = xfs_inobt_irec_to_allocmask(srec);
464 if (talloc & salloc)
465 return false;
466
467 return true;
468}
469
470/*
471 * Merge the source inode record into the target. The caller must call
472 * __xfs_inobt_can_merge() to ensure the merge is valid.
473 */
474STATIC void
475__xfs_inobt_rec_merge(
476 struct xfs_inobt_rec_incore *trec, /* target */
477 struct xfs_inobt_rec_incore *srec) /* src */
478{
479 ASSERT(trec->ir_startino == srec->ir_startino);
480
481 /* combine the counts */
482 trec->ir_count += srec->ir_count;
483 trec->ir_freecount += srec->ir_freecount;
484
485 /*
486 * Merge the holemask and free mask. For both fields, 0 bits refer to
487 * allocated inodes. We combine the allocated ranges with bitwise AND.
488 */
489 trec->ir_holemask &= srec->ir_holemask;
490 trec->ir_free &= srec->ir_free;
491}
492
493/*
494 * Insert a new sparse inode chunk into the associated inode btree. The inode
495 * record for the sparse chunk is pre-aligned to a startino that should match
496 * any pre-existing sparse inode record in the tree. This allows sparse chunks
497 * to fill over time.
498 *
499 * This function supports two modes of handling preexisting records depending on
500 * the merge flag. If merge is true, the provided record is merged with the
501 * existing record and updated in place. The merged record is returned in nrec.
502 * If merge is false, an existing record is replaced with the provided record.
503 * If no preexisting record exists, the provided record is always inserted.
504 *
505 * It is considered corruption if a merge is requested and not possible. Given
506 * the sparse inode alignment constraints, this should never happen.
507 */
508STATIC int
509xfs_inobt_insert_sprec(
510 struct xfs_mount *mp,
511 struct xfs_trans *tp,
512 struct xfs_buf *agbp,
513 int btnum,
514 struct xfs_inobt_rec_incore *nrec, /* in/out: new/merged rec. */
515 bool merge) /* merge or replace */
516{
517 struct xfs_btree_cur *cur;
518 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
519 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
520 int error;
521 int i;
522 struct xfs_inobt_rec_incore rec;
523
524 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, btnum);
525
526 /* the new record is pre-aligned so we know where to look */
527 error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
528 if (error)
529 goto error;
530 /* if nothing there, insert a new record and return */
531 if (i == 0) {
532 error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
533 nrec->ir_count, nrec->ir_freecount,
534 nrec->ir_free, &i);
535 if (error)
536 goto error;
537 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
538
539 goto out;
540 }
541
542 /*
543 * A record exists at this startino. Merge or replace the record
544 * depending on what we've been asked to do.
545 */
546 if (merge) {
547 error = xfs_inobt_get_rec(cur, &rec, &i);
548 if (error)
549 goto error;
550 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
551 XFS_WANT_CORRUPTED_GOTO(mp,
552 rec.ir_startino == nrec->ir_startino,
553 error);
554
555 /*
556 * This should never fail. If we have coexisting records that
557 * cannot merge, something is seriously wrong.
558 */
559 XFS_WANT_CORRUPTED_GOTO(mp, __xfs_inobt_can_merge(nrec, &rec),
560 error);
561
562 trace_xfs_irec_merge_pre(mp, agno, rec.ir_startino,
563 rec.ir_holemask, nrec->ir_startino,
564 nrec->ir_holemask);
565
566 /* merge to nrec to output the updated record */
567 __xfs_inobt_rec_merge(nrec, &rec);
568
569 trace_xfs_irec_merge_post(mp, agno, nrec->ir_startino,
570 nrec->ir_holemask);
571
572 error = xfs_inobt_rec_check_count(mp, nrec);
573 if (error)
574 goto error;
575 }
576
577 error = xfs_inobt_update(cur, nrec);
578 if (error)
579 goto error;
580
581out:
582 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
583 return 0;
584error:
585 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
586 return error;
587}
588
589/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 * Allocate new inodes in the allocation group specified by agbp.
591 * Return 0 for success, else error code.
592 */
593STATIC int /* error code or 0 */
594xfs_ialloc_ag_alloc(
595 xfs_trans_t *tp, /* transaction pointer */
596 xfs_buf_t *agbp, /* alloc group buffer */
597 int *alloc)
598{
599 xfs_agi_t *agi; /* allocation group header */
600 xfs_alloc_arg_t args; /* allocation argument structure */
David Chinner92821e22007-05-24 15:26:31 +1000601 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 xfs_agino_t newino; /* new first inode's number */
604 xfs_agino_t newlen; /* new number of inodes */
Glen Overby3ccb8b52006-03-29 09:52:28 +1000605 int isaligned = 0; /* inode allocation at stripe unit */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 /* boundary */
Brian Foster56d11152015-05-29 09:18:32 +1000607 uint16_t allocmask = (uint16_t) -1; /* init. to full chunk */
608 struct xfs_inobt_rec_incore rec;
Dave Chinner44b56e02010-01-11 11:47:43 +0000609 struct xfs_perag *pag;
Brian Foster1cdadee2015-05-29 09:19:29 +1000610 int do_sparse = 0;
611
Mark Tinguelya0041682012-09-20 13:16:45 -0500612 memset(&args, 0, sizeof(args));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 args.tp = tp;
614 args.mp = tp->t_mountp;
Brian Foster1cdadee2015-05-29 09:19:29 +1000615 args.fsbno = NULLFSBLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Brian Foster46fc58d2015-06-04 13:03:34 +1000617#ifdef DEBUG
618 /* randomly do sparse inode allocations */
619 if (xfs_sb_version_hassparseinodes(&tp->t_mountp->m_sb) &&
620 args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks)
621 do_sparse = prandom_u32() & 1;
622#endif
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 /*
625 * Locking will ensure that we don't have two callers in here
626 * at one time.
627 */
Jie Liu71783432013-12-13 15:51:46 +1100628 newlen = args.mp->m_ialloc_inos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 if (args.mp->m_maxicount &&
George Wang74f9ce12015-05-29 07:39:34 +1000630 percpu_counter_read_positive(&args.mp->m_icount) + newlen >
Dave Chinner501ab322015-02-23 21:19:28 +1100631 args.mp->m_maxicount)
Dave Chinner24513372014-06-25 14:58:08 +1000632 return -ENOSPC;
Jie Liu126cd102013-12-13 15:51:48 +1100633 args.minlen = args.maxlen = args.mp->m_ialloc_blks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 /*
Glen Overby3ccb8b52006-03-29 09:52:28 +1000635 * First try to allocate inodes contiguous with the last-allocated
636 * chunk of inodes. If the filesystem is striped, this will fill
637 * an entire stripe unit with inodes.
Dave Chinner28c8e412013-06-27 16:04:55 +1000638 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 agi = XFS_BUF_TO_AGI(agbp);
Glen Overby3ccb8b52006-03-29 09:52:28 +1000640 newino = be32_to_cpu(agi->agi_newino);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300641 agno = be32_to_cpu(agi->agi_seqno);
Nathan Scott019ff2d2006-04-11 15:45:05 +1000642 args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
Jie Liu126cd102013-12-13 15:51:48 +1100643 args.mp->m_ialloc_blks;
Brian Foster1cdadee2015-05-29 09:19:29 +1000644 if (do_sparse)
645 goto sparse_alloc;
Nathan Scott019ff2d2006-04-11 15:45:05 +1000646 if (likely(newino != NULLAGINO &&
647 (args.agbno < be32_to_cpu(agi->agi_length)))) {
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300648 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
Glen Overby3ccb8b52006-03-29 09:52:28 +1000649 args.type = XFS_ALLOCTYPE_THIS_BNO;
Glen Overby3ccb8b52006-03-29 09:52:28 +1000650 args.prod = 1;
David Chinner75de2a92008-03-27 18:00:38 +1100651
Glen Overby3ccb8b52006-03-29 09:52:28 +1000652 /*
David Chinner75de2a92008-03-27 18:00:38 +1100653 * We need to take into account alignment here to ensure that
654 * we don't modify the free list if we fail to have an exact
655 * block. If we don't have an exact match, and every oher
656 * attempt allocation attempt fails, we'll end up cancelling
657 * a dirty transaction and shutting down.
658 *
659 * For an exact allocation, alignment must be 1,
660 * however we need to take cluster alignment into account when
661 * fixing up the freelist. Use the minalignslop field to
662 * indicate that extra blocks might be required for alignment,
663 * but not to use them in the actual exact allocation.
Glen Overby3ccb8b52006-03-29 09:52:28 +1000664 */
David Chinner75de2a92008-03-27 18:00:38 +1100665 args.alignment = 1;
Dave Chinner7a1df152014-12-04 09:42:21 +1100666 args.minalignslop = xfs_ialloc_cluster_alignment(args.mp) - 1;
David Chinner75de2a92008-03-27 18:00:38 +1100667
668 /* Allow space for the inode btree to split. */
Christoph Hellwig0d87e652009-02-09 08:37:14 +0100669 args.minleft = args.mp->m_in_maxlevels - 1;
Glen Overby3ccb8b52006-03-29 09:52:28 +1000670 if ((error = xfs_alloc_vextent(&args)))
671 return error;
Brian Fostere480a722014-03-07 16:19:14 +1100672
673 /*
674 * This request might have dirtied the transaction if the AG can
675 * satisfy the request, but the exact block was not available.
676 * If the allocation did fail, subsequent requests will relax
677 * the exact agbno requirement and increase the alignment
678 * instead. It is critical that the total size of the request
679 * (len + alignment + slop) does not increase from this point
680 * on, so reset minalignslop to ensure it is not included in
681 * subsequent requests.
682 */
683 args.minalignslop = 0;
Brian Foster1cdadee2015-05-29 09:19:29 +1000684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Glen Overby3ccb8b52006-03-29 09:52:28 +1000686 if (unlikely(args.fsbno == NULLFSBLOCK)) {
687 /*
688 * Set the alignment for the allocation.
689 * If stripe alignment is turned on then align at stripe unit
690 * boundary.
Nathan Scott019ff2d2006-04-11 15:45:05 +1000691 * If the cluster size is smaller than a filesystem block
692 * then we're doing I/O for inodes in filesystem block size
Glen Overby3ccb8b52006-03-29 09:52:28 +1000693 * pieces, so don't need alignment anyway.
694 */
695 isaligned = 0;
696 if (args.mp->m_sinoalign) {
697 ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
698 args.alignment = args.mp->m_dalign;
699 isaligned = 1;
David Chinner75de2a92008-03-27 18:00:38 +1100700 } else
Dave Chinner7a1df152014-12-04 09:42:21 +1100701 args.alignment = xfs_ialloc_cluster_alignment(args.mp);
Glen Overby3ccb8b52006-03-29 09:52:28 +1000702 /*
703 * Need to figure out where to allocate the inode blocks.
704 * Ideally they should be spaced out through the a.g.
705 * For now, just allocate blocks up front.
706 */
707 args.agbno = be32_to_cpu(agi->agi_root);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300708 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
Glen Overby3ccb8b52006-03-29 09:52:28 +1000709 /*
710 * Allocate a fixed-size extent of inodes.
711 */
712 args.type = XFS_ALLOCTYPE_NEAR_BNO;
Glen Overby3ccb8b52006-03-29 09:52:28 +1000713 args.prod = 1;
714 /*
715 * Allow space for the inode btree to split.
716 */
Christoph Hellwig0d87e652009-02-09 08:37:14 +0100717 args.minleft = args.mp->m_in_maxlevels - 1;
Glen Overby3ccb8b52006-03-29 09:52:28 +1000718 if ((error = xfs_alloc_vextent(&args)))
719 return error;
720 }
Nathan Scott019ff2d2006-04-11 15:45:05 +1000721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 /*
723 * If stripe alignment is turned on, then try again with cluster
724 * alignment.
725 */
726 if (isaligned && args.fsbno == NULLFSBLOCK) {
727 args.type = XFS_ALLOCTYPE_NEAR_BNO;
Christoph Hellwig16259e72005-11-02 15:11:25 +1100728 args.agbno = be32_to_cpu(agi->agi_root);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300729 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
Dave Chinner7a1df152014-12-04 09:42:21 +1100730 args.alignment = xfs_ialloc_cluster_alignment(args.mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 if ((error = xfs_alloc_vextent(&args)))
732 return error;
733 }
734
Brian Foster56d11152015-05-29 09:18:32 +1000735 /*
736 * Finally, try a sparse allocation if the filesystem supports it and
737 * the sparse allocation length is smaller than a full chunk.
738 */
739 if (xfs_sb_version_hassparseinodes(&args.mp->m_sb) &&
740 args.mp->m_ialloc_min_blks < args.mp->m_ialloc_blks &&
741 args.fsbno == NULLFSBLOCK) {
Brian Foster1cdadee2015-05-29 09:19:29 +1000742sparse_alloc:
Brian Foster56d11152015-05-29 09:18:32 +1000743 args.type = XFS_ALLOCTYPE_NEAR_BNO;
744 args.agbno = be32_to_cpu(agi->agi_root);
745 args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
746 args.alignment = args.mp->m_sb.sb_spino_align;
747 args.prod = 1;
748
749 args.minlen = args.mp->m_ialloc_min_blks;
750 args.maxlen = args.minlen;
751
752 /*
753 * The inode record will be aligned to full chunk size. We must
754 * prevent sparse allocation from AG boundaries that result in
755 * invalid inode records, such as records that start at agbno 0
756 * or extend beyond the AG.
757 *
758 * Set min agbno to the first aligned, non-zero agbno and max to
759 * the last aligned agbno that is at least one full chunk from
760 * the end of the AG.
761 */
762 args.min_agbno = args.mp->m_sb.sb_inoalignmt;
763 args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
764 args.mp->m_sb.sb_inoalignmt) -
765 args.mp->m_ialloc_blks;
766
767 error = xfs_alloc_vextent(&args);
768 if (error)
769 return error;
770
771 newlen = args.len << args.mp->m_sb.sb_inopblog;
Brian Foster46fc58d2015-06-04 13:03:34 +1000772 ASSERT(newlen <= XFS_INODES_PER_CHUNK);
Brian Foster56d11152015-05-29 09:18:32 +1000773 allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
774 }
775
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 if (args.fsbno == NULLFSBLOCK) {
777 *alloc = 0;
778 return 0;
779 }
780 ASSERT(args.len == args.minlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
David Chinner359346a2008-04-29 12:53:32 +1000782 /*
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300783 * Stamp and write the inode buffers.
784 *
David Chinner359346a2008-04-29 12:53:32 +1000785 * Seed the new inode cluster with a random generation number. This
786 * prevents short-term reuse of generation numbers if a chunk is
787 * freed and then immediately reallocated. We use random numbers
788 * rather than a linear progression to prevent the next generation
789 * number from being easily guessable.
790 */
Brian Foster463958a2015-05-29 09:05:49 +1000791 error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, agno,
792 args.agbno, args.len, prandom_u32());
Christoph Hellwigd42f08f2008-11-28 14:23:38 +1100793
Chandra Seetharaman2a30f36d2011-09-20 13:56:55 +0000794 if (error)
795 return error;
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300796 /*
797 * Convert the results.
798 */
799 newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
Brian Foster56d11152015-05-29 09:18:32 +1000800
801 if (xfs_inobt_issparse(~allocmask)) {
802 /*
803 * We've allocated a sparse chunk. Align the startino and mask.
804 */
805 xfs_align_sparse_ino(args.mp, &newino, &allocmask);
806
807 rec.ir_startino = newino;
808 rec.ir_holemask = ~allocmask;
809 rec.ir_count = newlen;
810 rec.ir_freecount = newlen;
811 rec.ir_free = XFS_INOBT_ALL_FREE;
812
813 /*
814 * Insert the sparse record into the inobt and allow for a merge
815 * if necessary. If a merge does occur, rec is updated to the
816 * merged record.
817 */
818 error = xfs_inobt_insert_sprec(args.mp, tp, agbp, XFS_BTNUM_INO,
819 &rec, true);
820 if (error == -EFSCORRUPTED) {
821 xfs_alert(args.mp,
822 "invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
823 XFS_AGINO_TO_INO(args.mp, agno,
824 rec.ir_startino),
825 rec.ir_holemask, rec.ir_count);
826 xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
827 }
828 if (error)
829 return error;
830
831 /*
832 * We can't merge the part we've just allocated as for the inobt
833 * due to finobt semantics. The original record may or may not
834 * exist independent of whether physical inodes exist in this
835 * sparse chunk.
836 *
837 * We must update the finobt record based on the inobt record.
838 * rec contains the fully merged and up to date inobt record
839 * from the previous call. Set merge false to replace any
840 * existing record with this one.
841 */
842 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
843 error = xfs_inobt_insert_sprec(args.mp, tp, agbp,
844 XFS_BTNUM_FINO, &rec,
845 false);
846 if (error)
847 return error;
848 }
849 } else {
850 /* full chunk - insert new records to both btrees */
851 error = xfs_inobt_insert(args.mp, tp, agbp, newino, newlen,
852 XFS_BTNUM_INO);
853 if (error)
854 return error;
855
856 if (xfs_sb_version_hasfinobt(&args.mp->m_sb)) {
857 error = xfs_inobt_insert(args.mp, tp, agbp, newino,
858 newlen, XFS_BTNUM_FINO);
859 if (error)
860 return error;
861 }
862 }
863
864 /*
865 * Update AGI counts and newino.
866 */
Marcin Slusarz413d57c2008-02-13 15:03:29 -0800867 be32_add_cpu(&agi->agi_count, newlen);
868 be32_add_cpu(&agi->agi_freecount, newlen);
Dave Chinner44b56e02010-01-11 11:47:43 +0000869 pag = xfs_perag_get(args.mp, agno);
870 pag->pagi_freecount += newlen;
871 xfs_perag_put(pag);
Christoph Hellwig16259e72005-11-02 15:11:25 +1100872 agi->agi_newino = cpu_to_be32(newino);
Dave Chinner85c0b2a2009-08-31 20:56:51 -0300873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 * Log allocation group header fields
876 */
877 xfs_ialloc_log_agi(tp, agbp,
878 XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
879 /*
880 * Modify/log superblock values for inode count and inode free count.
881 */
882 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
883 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
884 *alloc = 1;
885 return 0;
886}
887
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000888STATIC xfs_agnumber_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889xfs_ialloc_next_ag(
890 xfs_mount_t *mp)
891{
892 xfs_agnumber_t agno;
893
894 spin_lock(&mp->m_agirotor_lock);
895 agno = mp->m_agirotor;
Carlos Maiolino8aea3ff2012-09-20 10:32:36 -0300896 if (++mp->m_agirotor >= mp->m_maxagi)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 mp->m_agirotor = 0;
898 spin_unlock(&mp->m_agirotor_lock);
899
900 return agno;
901}
902
903/*
904 * Select an allocation group to look for a free inode in, based on the parent
Zhi Yong Wu2f21ff12013-08-12 03:14:58 +0000905 * inode and the mode. Return the allocation group buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 */
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400907STATIC xfs_agnumber_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908xfs_ialloc_ag_select(
909 xfs_trans_t *tp, /* transaction pointer */
910 xfs_ino_t parent, /* parent directory inode number */
Al Viro576b1d62011-07-26 02:50:15 -0400911 umode_t mode, /* bits set to indicate file type */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 int okalloc) /* ok to allocate more space */
913{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 xfs_agnumber_t agcount; /* number of ag's in the filesystem */
915 xfs_agnumber_t agno; /* current ag number */
916 int flags; /* alloc buffer locking flags */
917 xfs_extlen_t ineed; /* blocks needed for inode allocation */
918 xfs_extlen_t longest = 0; /* longest extent available */
919 xfs_mount_t *mp; /* mount point structure */
920 int needspace; /* file mode implies space allocated */
921 xfs_perag_t *pag; /* per allocation group data */
922 xfs_agnumber_t pagno; /* parent (starting) ag number */
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400923 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925 /*
926 * Files of these types need at least one block if length > 0
927 * (and they won't fit in the inode, but that's hard to figure out).
928 */
929 needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
930 mp = tp->t_mountp;
931 agcount = mp->m_maxagi;
932 if (S_ISDIR(mode))
933 pagno = xfs_ialloc_next_ag(mp);
934 else {
935 pagno = XFS_INO_TO_AGNO(mp, parent);
936 if (pagno >= agcount)
937 pagno = 0;
938 }
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 ASSERT(pagno < agcount);
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400941
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 /*
943 * Loop through allocation groups, looking for one with a little
944 * free space in it. Note we don't look for free inodes, exactly.
945 * Instead, we include whether there is a need to allocate inodes
946 * to mean that blocks must be allocated for them,
947 * if none are currently free.
948 */
949 agno = pagno;
950 flags = XFS_ALLOC_FLAG_TRYLOCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 for (;;) {
Dave Chinner44b56e02010-01-11 11:47:43 +0000952 pag = xfs_perag_get(mp, agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 if (!pag->pagi_inodeok) {
954 xfs_ialloc_next_ag(mp);
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400955 goto nextag;
956 }
957
958 if (!pag->pagi_init) {
959 error = xfs_ialloc_pagi_init(mp, tp, agno);
960 if (error)
961 goto nextag;
962 }
963
964 if (pag->pagi_freecount) {
965 xfs_perag_put(pag);
966 return agno;
967 }
968
969 if (!okalloc)
970 goto nextag;
971
972 if (!pag->pagf_init) {
973 error = xfs_alloc_pagf_init(mp, tp, agno, flags);
974 if (error)
975 goto nextag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 }
977
978 /*
Dave Chinner7a1df152014-12-04 09:42:21 +1100979 * Check that there is enough free space for the file plus a
980 * chunk of inodes if we need to allocate some. If this is the
981 * first pass across the AGs, take into account the potential
982 * space needed for alignment of inode chunks when checking the
983 * longest contiguous free space in the AG - this prevents us
984 * from getting ENOSPC because we have free space larger than
985 * m_ialloc_blks but alignment constraints prevent us from using
986 * it.
987 *
988 * If we can't find an AG with space for full alignment slack to
989 * be taken into account, we must be near ENOSPC in all AGs.
990 * Hence we don't include alignment for the second pass and so
991 * if we fail allocation due to alignment issues then it is most
992 * likely a real ENOSPC condition.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 */
Brian Foster066a1882015-05-29 08:55:20 +1000994 ineed = mp->m_ialloc_min_blks;
Dave Chinner7a1df152014-12-04 09:42:21 +1100995 if (flags && ineed > 1)
996 ineed += xfs_ialloc_cluster_alignment(mp);
Christoph Hellwig55d6af62012-07-04 10:54:49 -0400997 longest = pag->pagf_longest;
998 if (!longest)
999 longest = pag->pagf_flcount > 0;
1000
1001 if (pag->pagf_freeblks >= needspace + ineed &&
1002 longest >= ineed) {
1003 xfs_perag_put(pag);
1004 return agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006nextag:
Dave Chinner44b56e02010-01-11 11:47:43 +00001007 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 /*
1009 * No point in iterating over the rest, if we're shutting
1010 * down.
1011 */
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001012 if (XFS_FORCED_SHUTDOWN(mp))
Christoph Hellwig55d6af62012-07-04 10:54:49 -04001013 return NULLAGNUMBER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 agno++;
1015 if (agno >= agcount)
1016 agno = 0;
1017 if (agno == pagno) {
Dave Chinner1c1c6eb2010-01-11 11:47:44 +00001018 if (flags == 0)
Christoph Hellwig55d6af62012-07-04 10:54:49 -04001019 return NULLAGNUMBER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 flags = 0;
1021 }
1022 }
1023}
1024
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001025/*
1026 * Try to retrieve the next record to the left/right from the current one.
1027 */
1028STATIC int
1029xfs_ialloc_next_rec(
1030 struct xfs_btree_cur *cur,
1031 xfs_inobt_rec_incore_t *rec,
1032 int *done,
1033 int left)
1034{
1035 int error;
1036 int i;
1037
1038 if (left)
1039 error = xfs_btree_decrement(cur, 0, &i);
1040 else
1041 error = xfs_btree_increment(cur, 0, &i);
1042
1043 if (error)
1044 return error;
1045 *done = !i;
1046 if (i) {
1047 error = xfs_inobt_get_rec(cur, rec, &i);
1048 if (error)
1049 return error;
Eric Sandeen5fb5aee2015-02-23 22:39:13 +11001050 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001051 }
1052
1053 return 0;
1054}
1055
Dave Chinnerbd169562009-08-31 20:58:28 -03001056STATIC int
1057xfs_ialloc_get_rec(
1058 struct xfs_btree_cur *cur,
1059 xfs_agino_t agino,
1060 xfs_inobt_rec_incore_t *rec,
Jie Liu43df2ee2013-06-24 13:37:23 +08001061 int *done)
Dave Chinnerbd169562009-08-31 20:58:28 -03001062{
1063 int error;
1064 int i;
1065
1066 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
1067 if (error)
1068 return error;
1069 *done = !i;
1070 if (i) {
1071 error = xfs_inobt_get_rec(cur, rec, &i);
1072 if (error)
1073 return error;
Eric Sandeen5fb5aee2015-02-23 22:39:13 +11001074 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
Dave Chinnerbd169562009-08-31 20:58:28 -03001075 }
1076
1077 return 0;
1078}
Dave Chinner0b48db82009-08-31 20:57:09 -03001079
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080/*
Brian Foster26dd5212015-05-29 09:20:10 +10001081 * Return the offset of the first free inode in the record. If the inode chunk
1082 * is sparsely allocated, we convert the record holemask to inode granularity
1083 * and mask off the unallocated regions from the inode free mask.
Brian Fosterd4cc5402015-05-29 08:50:21 +10001084 */
1085STATIC int
1086xfs_inobt_first_free_inode(
1087 struct xfs_inobt_rec_incore *rec)
1088{
Brian Foster26dd5212015-05-29 09:20:10 +10001089 xfs_inofree_t realfree;
1090
1091 /* if there are no holes, return the first available offset */
1092 if (!xfs_inobt_issparse(rec->ir_holemask))
1093 return xfs_lowbit64(rec->ir_free);
1094
1095 realfree = xfs_inobt_irec_to_allocmask(rec);
1096 realfree &= rec->ir_free;
1097
1098 return xfs_lowbit64(realfree);
Brian Fosterd4cc5402015-05-29 08:50:21 +10001099}
1100
1101/*
Brian Foster6dd86382014-04-24 16:00:53 +10001102 * Allocate an inode using the inobt-only algorithm.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 */
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001104STATIC int
Brian Foster6dd86382014-04-24 16:00:53 +10001105xfs_dialloc_ag_inobt(
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001106 struct xfs_trans *tp,
1107 struct xfs_buf *agbp,
1108 xfs_ino_t parent,
1109 xfs_ino_t *inop)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110{
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001111 struct xfs_mount *mp = tp->t_mountp;
1112 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1113 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1114 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1115 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1116 struct xfs_perag *pag;
1117 struct xfs_btree_cur *cur, *tcur;
1118 struct xfs_inobt_rec_incore rec, trec;
1119 xfs_ino_t ino;
1120 int error;
1121 int offset;
1122 int i, j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Dave Chinner44b56e02010-01-11 11:47:43 +00001124 pag = xfs_perag_get(mp, agno);
Dave Chinnerbd169562009-08-31 20:58:28 -03001125
Christoph Hellwig4bb61062012-07-04 10:54:48 -04001126 ASSERT(pag->pagi_init);
1127 ASSERT(pag->pagi_inodeok);
1128 ASSERT(pag->pagi_freecount > 0);
1129
Dave Chinnerbd169562009-08-31 20:58:28 -03001130 restart_pagno:
Brian Foster57bd3db2014-04-24 16:00:50 +10001131 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 /*
1133 * If pagino is 0 (this is the root inode allocation) use newino.
1134 * This must work because we've just allocated some.
1135 */
1136 if (!pagino)
Christoph Hellwig16259e72005-11-02 15:11:25 +11001137 pagino = be32_to_cpu(agi->agi_newino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Dave Chinner0b48db82009-08-31 20:57:09 -03001139 error = xfs_check_agi_freecount(cur, agi);
1140 if (error)
1141 goto error0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 /*
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001144 * If in the same AG as the parent, try to get near the parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 */
1146 if (pagno == agno) {
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001147 int doneleft; /* done, to the left */
1148 int doneright; /* done, to the right */
Dave Chinnerbd169562009-08-31 20:58:28 -03001149 int searchdistance = 10;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001150
Christoph Hellwig21875502009-08-31 20:58:21 -03001151 error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001152 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 goto error0;
Eric Sandeenc29aad42015-02-23 22:39:08 +11001154 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001155
1156 error = xfs_inobt_get_rec(cur, &rec, &j);
1157 if (error)
1158 goto error0;
Eric Sandeenc29aad42015-02-23 22:39:08 +11001159 XFS_WANT_CORRUPTED_GOTO(mp, j == 1, error0);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001160
1161 if (rec.ir_freecount > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 /*
1163 * Found a free inode in the same chunk
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001164 * as the parent, done.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 */
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001166 goto alloc_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001168
1169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 /*
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001171 * In the same AG as parent, but parent's chunk is full.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001174 /* duplicate the cursor, search left & right simultaneously */
1175 error = xfs_btree_dup_cursor(cur, &tcur);
1176 if (error)
1177 goto error0;
1178
Dave Chinnerbd169562009-08-31 20:58:28 -03001179 /*
1180 * Skip to last blocks looked up if same parent inode.
1181 */
1182 if (pagino != NULLAGINO &&
1183 pag->pagl_pagino == pagino &&
1184 pag->pagl_leftrec != NULLAGINO &&
1185 pag->pagl_rightrec != NULLAGINO) {
1186 error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
Jie Liu43df2ee2013-06-24 13:37:23 +08001187 &trec, &doneleft);
Dave Chinnerbd169562009-08-31 20:58:28 -03001188 if (error)
1189 goto error1;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001190
Dave Chinnerbd169562009-08-31 20:58:28 -03001191 error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
Jie Liu43df2ee2013-06-24 13:37:23 +08001192 &rec, &doneright);
Dave Chinnerbd169562009-08-31 20:58:28 -03001193 if (error)
1194 goto error1;
1195 } else {
1196 /* search left with tcur, back up 1 record */
1197 error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1198 if (error)
1199 goto error1;
1200
1201 /* search right with cur, go forward 1 record. */
1202 error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1203 if (error)
1204 goto error1;
1205 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001206
1207 /*
1208 * Loop until we find an inode chunk with a free inode.
1209 */
1210 while (!doneleft || !doneright) {
1211 int useleft; /* using left inode chunk this time */
1212
Dave Chinnerbd169562009-08-31 20:58:28 -03001213 if (!--searchdistance) {
1214 /*
1215 * Not in range - save last search
1216 * location and allocate a new inode
1217 */
Eric Sandeen3b826382009-10-30 09:27:07 +01001218 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
Dave Chinnerbd169562009-08-31 20:58:28 -03001219 pag->pagl_leftrec = trec.ir_startino;
1220 pag->pagl_rightrec = rec.ir_startino;
1221 pag->pagl_pagino = pagino;
1222 goto newino;
1223 }
1224
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001225 /* figure out the closer block if both are valid. */
1226 if (!doneleft && !doneright) {
1227 useleft = pagino -
1228 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1229 rec.ir_startino - pagino;
1230 } else {
1231 useleft = !doneleft;
1232 }
1233
1234 /* free inodes to the left? */
1235 if (useleft && trec.ir_freecount) {
1236 rec = trec;
1237 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1238 cur = tcur;
Dave Chinnerbd169562009-08-31 20:58:28 -03001239
1240 pag->pagl_leftrec = trec.ir_startino;
1241 pag->pagl_rightrec = rec.ir_startino;
1242 pag->pagl_pagino = pagino;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001243 goto alloc_inode;
1244 }
1245
1246 /* free inodes to the right? */
1247 if (!useleft && rec.ir_freecount) {
1248 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
Dave Chinnerbd169562009-08-31 20:58:28 -03001249
1250 pag->pagl_leftrec = trec.ir_startino;
1251 pag->pagl_rightrec = rec.ir_startino;
1252 pag->pagl_pagino = pagino;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001253 goto alloc_inode;
1254 }
1255
1256 /* get next record to check */
1257 if (useleft) {
1258 error = xfs_ialloc_next_rec(tcur, &trec,
1259 &doneleft, 1);
1260 } else {
1261 error = xfs_ialloc_next_rec(cur, &rec,
1262 &doneright, 0);
1263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 goto error1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 }
Dave Chinnerbd169562009-08-31 20:58:28 -03001267
1268 /*
1269 * We've reached the end of the btree. because
1270 * we are only searching a small chunk of the
1271 * btree each search, there is obviously free
1272 * inodes closer to the parent inode than we
1273 * are now. restart the search again.
1274 */
1275 pag->pagl_pagino = NULLAGINO;
1276 pag->pagl_leftrec = NULLAGINO;
1277 pag->pagl_rightrec = NULLAGINO;
1278 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1279 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1280 goto restart_pagno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 /*
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001284 * In a different AG from the parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 * See if the most recently allocated block has any free.
1286 */
Dave Chinnerbd169562009-08-31 20:58:28 -03001287newino:
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001288 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
Christoph Hellwig21875502009-08-31 20:58:21 -03001289 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1290 XFS_LOOKUP_EQ, &i);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001291 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 goto error0;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001293
1294 if (i == 1) {
1295 error = xfs_inobt_get_rec(cur, &rec, &j);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 if (error)
1297 goto error0;
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001298
1299 if (j == 1 && rec.ir_freecount > 0) {
1300 /*
1301 * The last chunk allocated in the group
1302 * still has a free inode.
1303 */
1304 goto alloc_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 }
1306 }
Dave Chinnerbd169562009-08-31 20:58:28 -03001307 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001308
Dave Chinnerbd169562009-08-31 20:58:28 -03001309 /*
1310 * None left in the last group, search the whole AG
1311 */
1312 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1313 if (error)
1314 goto error0;
Eric Sandeenc29aad42015-02-23 22:39:08 +11001315 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
Dave Chinnerbd169562009-08-31 20:58:28 -03001316
1317 for (;;) {
1318 error = xfs_inobt_get_rec(cur, &rec, &i);
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001319 if (error)
1320 goto error0;
Eric Sandeenc29aad42015-02-23 22:39:08 +11001321 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
Dave Chinnerbd169562009-08-31 20:58:28 -03001322 if (rec.ir_freecount > 0)
1323 break;
1324 error = xfs_btree_increment(cur, 0, &i);
1325 if (error)
1326 goto error0;
Eric Sandeenc29aad42015-02-23 22:39:08 +11001327 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 }
Christoph Hellwig4254b0b2009-08-31 20:57:14 -03001329
1330alloc_inode:
Brian Fosterd4cc5402015-05-29 08:50:21 +10001331 offset = xfs_inobt_first_free_inode(&rec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 ASSERT(offset >= 0);
1333 ASSERT(offset < XFS_INODES_PER_CHUNK);
1334 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1335 XFS_INODES_PER_CHUNK) == 0);
1336 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
Christoph Hellwig0d87e652009-02-09 08:37:14 +01001337 rec.ir_free &= ~XFS_INOBT_MASK(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 rec.ir_freecount--;
Christoph Hellwigafabc242009-08-31 20:57:03 -03001339 error = xfs_inobt_update(cur, &rec);
1340 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 goto error0;
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001342 be32_add_cpu(&agi->agi_freecount, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
Dave Chinner44b56e02010-01-11 11:47:43 +00001344 pag->pagi_freecount--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Dave Chinner0b48db82009-08-31 20:57:09 -03001346 error = xfs_check_agi_freecount(cur, agi);
1347 if (error)
1348 goto error0;
1349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1351 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
Dave Chinner44b56e02010-01-11 11:47:43 +00001352 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 *inop = ino;
1354 return 0;
1355error1:
1356 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1357error0:
1358 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
Dave Chinner44b56e02010-01-11 11:47:43 +00001359 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 return error;
1361}
1362
1363/*
Brian Foster6dd86382014-04-24 16:00:53 +10001364 * Use the free inode btree to allocate an inode based on distance from the
1365 * parent. Note that the provided cursor may be deleted and replaced.
1366 */
1367STATIC int
1368xfs_dialloc_ag_finobt_near(
1369 xfs_agino_t pagino,
1370 struct xfs_btree_cur **ocur,
1371 struct xfs_inobt_rec_incore *rec)
1372{
1373 struct xfs_btree_cur *lcur = *ocur; /* left search cursor */
1374 struct xfs_btree_cur *rcur; /* right search cursor */
1375 struct xfs_inobt_rec_incore rrec;
1376 int error;
1377 int i, j;
1378
1379 error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1380 if (error)
1381 return error;
1382
1383 if (i == 1) {
1384 error = xfs_inobt_get_rec(lcur, rec, &i);
1385 if (error)
1386 return error;
Eric Sandeen5fb5aee2015-02-23 22:39:13 +11001387 XFS_WANT_CORRUPTED_RETURN(lcur->bc_mp, i == 1);
Brian Foster6dd86382014-04-24 16:00:53 +10001388
1389 /*
1390 * See if we've landed in the parent inode record. The finobt
1391 * only tracks chunks with at least one free inode, so record
1392 * existence is enough.
1393 */
1394 if (pagino >= rec->ir_startino &&
1395 pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1396 return 0;
1397 }
1398
1399 error = xfs_btree_dup_cursor(lcur, &rcur);
1400 if (error)
1401 return error;
1402
1403 error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1404 if (error)
1405 goto error_rcur;
1406 if (j == 1) {
1407 error = xfs_inobt_get_rec(rcur, &rrec, &j);
1408 if (error)
1409 goto error_rcur;
Eric Sandeenc29aad42015-02-23 22:39:08 +11001410 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, j == 1, error_rcur);
Brian Foster6dd86382014-04-24 16:00:53 +10001411 }
1412
Eric Sandeenc29aad42015-02-23 22:39:08 +11001413 XFS_WANT_CORRUPTED_GOTO(lcur->bc_mp, i == 1 || j == 1, error_rcur);
Brian Foster6dd86382014-04-24 16:00:53 +10001414 if (i == 1 && j == 1) {
1415 /*
1416 * Both the left and right records are valid. Choose the closer
1417 * inode chunk to the target.
1418 */
1419 if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1420 (rrec.ir_startino - pagino)) {
1421 *rec = rrec;
1422 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1423 *ocur = rcur;
1424 } else {
1425 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1426 }
1427 } else if (j == 1) {
1428 /* only the right record is valid */
1429 *rec = rrec;
1430 xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1431 *ocur = rcur;
1432 } else if (i == 1) {
1433 /* only the left record is valid */
1434 xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1435 }
1436
1437 return 0;
1438
1439error_rcur:
1440 xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1441 return error;
1442}
1443
1444/*
1445 * Use the free inode btree to find a free inode based on a newino hint. If
1446 * the hint is NULL, find the first free inode in the AG.
1447 */
1448STATIC int
1449xfs_dialloc_ag_finobt_newino(
1450 struct xfs_agi *agi,
1451 struct xfs_btree_cur *cur,
1452 struct xfs_inobt_rec_incore *rec)
1453{
1454 int error;
1455 int i;
1456
1457 if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
Dave Chinnere68ed772014-09-29 10:43:15 +10001458 error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1459 XFS_LOOKUP_EQ, &i);
Brian Foster6dd86382014-04-24 16:00:53 +10001460 if (error)
1461 return error;
1462 if (i == 1) {
1463 error = xfs_inobt_get_rec(cur, rec, &i);
1464 if (error)
1465 return error;
Eric Sandeen5fb5aee2015-02-23 22:39:13 +11001466 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
Brian Foster6dd86382014-04-24 16:00:53 +10001467 return 0;
1468 }
1469 }
1470
1471 /*
1472 * Find the first inode available in the AG.
1473 */
1474 error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1475 if (error)
1476 return error;
Eric Sandeen5fb5aee2015-02-23 22:39:13 +11001477 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
Brian Foster6dd86382014-04-24 16:00:53 +10001478
1479 error = xfs_inobt_get_rec(cur, rec, &i);
1480 if (error)
1481 return error;
Eric Sandeen5fb5aee2015-02-23 22:39:13 +11001482 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
Brian Foster6dd86382014-04-24 16:00:53 +10001483
1484 return 0;
1485}
1486
1487/*
1488 * Update the inobt based on a modification made to the finobt. Also ensure that
1489 * the records from both trees are equivalent post-modification.
1490 */
1491STATIC int
1492xfs_dialloc_ag_update_inobt(
1493 struct xfs_btree_cur *cur, /* inobt cursor */
1494 struct xfs_inobt_rec_incore *frec, /* finobt record */
1495 int offset) /* inode offset */
1496{
1497 struct xfs_inobt_rec_incore rec;
1498 int error;
1499 int i;
1500
1501 error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1502 if (error)
1503 return error;
Eric Sandeen5fb5aee2015-02-23 22:39:13 +11001504 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
Brian Foster6dd86382014-04-24 16:00:53 +10001505
1506 error = xfs_inobt_get_rec(cur, &rec, &i);
1507 if (error)
1508 return error;
Eric Sandeen5fb5aee2015-02-23 22:39:13 +11001509 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
Brian Foster6dd86382014-04-24 16:00:53 +10001510 ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1511 XFS_INODES_PER_CHUNK) == 0);
1512
1513 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1514 rec.ir_freecount--;
1515
Eric Sandeen5fb5aee2015-02-23 22:39:13 +11001516 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, (rec.ir_free == frec->ir_free) &&
Brian Foster6dd86382014-04-24 16:00:53 +10001517 (rec.ir_freecount == frec->ir_freecount));
1518
kbuild test robotb72091f2014-12-01 08:24:58 +11001519 return xfs_inobt_update(cur, &rec);
Brian Foster6dd86382014-04-24 16:00:53 +10001520}
1521
1522/*
1523 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1524 * back to the inobt search algorithm.
1525 *
1526 * The caller selected an AG for us, and made sure that free inodes are
1527 * available.
1528 */
1529STATIC int
1530xfs_dialloc_ag(
1531 struct xfs_trans *tp,
1532 struct xfs_buf *agbp,
1533 xfs_ino_t parent,
1534 xfs_ino_t *inop)
1535{
1536 struct xfs_mount *mp = tp->t_mountp;
1537 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1538 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1539 xfs_agnumber_t pagno = XFS_INO_TO_AGNO(mp, parent);
1540 xfs_agino_t pagino = XFS_INO_TO_AGINO(mp, parent);
1541 struct xfs_perag *pag;
1542 struct xfs_btree_cur *cur; /* finobt cursor */
1543 struct xfs_btree_cur *icur; /* inobt cursor */
1544 struct xfs_inobt_rec_incore rec;
1545 xfs_ino_t ino;
1546 int error;
1547 int offset;
1548 int i;
1549
1550 if (!xfs_sb_version_hasfinobt(&mp->m_sb))
1551 return xfs_dialloc_ag_inobt(tp, agbp, parent, inop);
1552
1553 pag = xfs_perag_get(mp, agno);
1554
1555 /*
1556 * If pagino is 0 (this is the root inode allocation) use newino.
1557 * This must work because we've just allocated some.
1558 */
1559 if (!pagino)
1560 pagino = be32_to_cpu(agi->agi_newino);
1561
1562 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
1563
1564 error = xfs_check_agi_freecount(cur, agi);
1565 if (error)
1566 goto error_cur;
1567
1568 /*
1569 * The search algorithm depends on whether we're in the same AG as the
1570 * parent. If so, find the closest available inode to the parent. If
1571 * not, consider the agi hint or find the first free inode in the AG.
1572 */
1573 if (agno == pagno)
1574 error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1575 else
1576 error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1577 if (error)
1578 goto error_cur;
1579
Brian Fosterd4cc5402015-05-29 08:50:21 +10001580 offset = xfs_inobt_first_free_inode(&rec);
Brian Foster6dd86382014-04-24 16:00:53 +10001581 ASSERT(offset >= 0);
1582 ASSERT(offset < XFS_INODES_PER_CHUNK);
1583 ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1584 XFS_INODES_PER_CHUNK) == 0);
1585 ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
1586
1587 /*
1588 * Modify or remove the finobt record.
1589 */
1590 rec.ir_free &= ~XFS_INOBT_MASK(offset);
1591 rec.ir_freecount--;
1592 if (rec.ir_freecount)
1593 error = xfs_inobt_update(cur, &rec);
1594 else
1595 error = xfs_btree_delete(cur, &i);
1596 if (error)
1597 goto error_cur;
1598
1599 /*
1600 * The finobt has now been updated appropriately. We haven't updated the
1601 * agi and superblock yet, so we can create an inobt cursor and validate
1602 * the original freecount. If all is well, make the equivalent update to
1603 * the inobt using the finobt record and offset information.
1604 */
1605 icur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
1606
1607 error = xfs_check_agi_freecount(icur, agi);
1608 if (error)
1609 goto error_icur;
1610
1611 error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1612 if (error)
1613 goto error_icur;
1614
1615 /*
1616 * Both trees have now been updated. We must update the perag and
1617 * superblock before we can check the freecount for each btree.
1618 */
1619 be32_add_cpu(&agi->agi_freecount, -1);
1620 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1621 pag->pagi_freecount--;
1622
1623 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1624
1625 error = xfs_check_agi_freecount(icur, agi);
1626 if (error)
1627 goto error_icur;
1628 error = xfs_check_agi_freecount(cur, agi);
1629 if (error)
1630 goto error_icur;
1631
1632 xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1633 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1634 xfs_perag_put(pag);
1635 *inop = ino;
1636 return 0;
1637
1638error_icur:
1639 xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1640error_cur:
1641 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1642 xfs_perag_put(pag);
1643 return error;
1644}
1645
1646/*
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001647 * Allocate an inode on disk.
1648 *
1649 * Mode is used to tell whether the new inode will need space, and whether it
1650 * is a directory.
1651 *
1652 * This function is designed to be called twice if it has to do an allocation
1653 * to make more free inodes. On the first call, *IO_agbp should be set to NULL.
1654 * If an inode is available without having to performn an allocation, an inode
Carlos Maiolinocd856db2012-10-20 11:08:19 -03001655 * number is returned. In this case, *IO_agbp is set to NULL. If an allocation
1656 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
1657 * The caller should then commit the current transaction, allocate a
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001658 * new transaction, and call xfs_dialloc() again, passing in the previous value
1659 * of *IO_agbp. IO_agbp should be held across the transactions. Since the AGI
1660 * buffer is locked across the two calls, the second call is guaranteed to have
1661 * a free inode available.
1662 *
1663 * Once we successfully pick an inode its number is returned and the on-disk
1664 * data structures are updated. The inode itself is not read in, since doing so
1665 * would break ordering constraints with xfs_reclaim.
1666 */
1667int
1668xfs_dialloc(
1669 struct xfs_trans *tp,
1670 xfs_ino_t parent,
1671 umode_t mode,
1672 int okalloc,
1673 struct xfs_buf **IO_agbp,
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001674 xfs_ino_t *inop)
1675{
Christoph Hellwig55d6af62012-07-04 10:54:49 -04001676 struct xfs_mount *mp = tp->t_mountp;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001677 struct xfs_buf *agbp;
1678 xfs_agnumber_t agno;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001679 int error;
1680 int ialloced;
1681 int noroom = 0;
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001682 xfs_agnumber_t start_agno;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001683 struct xfs_perag *pag;
1684
Christoph Hellwig4bb61062012-07-04 10:54:48 -04001685 if (*IO_agbp) {
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001686 /*
Christoph Hellwig4bb61062012-07-04 10:54:48 -04001687 * If the caller passes in a pointer to the AGI buffer,
1688 * continue where we left off before. In this case, we
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001689 * know that the allocation group has free inodes.
1690 */
1691 agbp = *IO_agbp;
Christoph Hellwig4bb61062012-07-04 10:54:48 -04001692 goto out_alloc;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001693 }
Christoph Hellwig4bb61062012-07-04 10:54:48 -04001694
1695 /*
1696 * We do not have an agbp, so select an initial allocation
1697 * group for inode allocation.
1698 */
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001699 start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
1700 if (start_agno == NULLAGNUMBER) {
Christoph Hellwig4bb61062012-07-04 10:54:48 -04001701 *inop = NULLFSINO;
1702 return 0;
1703 }
Christoph Hellwig55d6af62012-07-04 10:54:49 -04001704
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001705 /*
1706 * If we have already hit the ceiling of inode blocks then clear
1707 * okalloc so we scan all available agi structures for a free
1708 * inode.
George Wang74f9ce12015-05-29 07:39:34 +10001709 *
1710 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1711 * which will sacrifice the preciseness but improve the performance.
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001712 */
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001713 if (mp->m_maxicount &&
George Wang74f9ce12015-05-29 07:39:34 +10001714 percpu_counter_read_positive(&mp->m_icount) + mp->m_ialloc_inos
1715 > mp->m_maxicount) {
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001716 noroom = 1;
1717 okalloc = 0;
1718 }
1719
1720 /*
1721 * Loop until we find an allocation group that either has free inodes
1722 * or in which we can allocate some inodes. Iterate through the
1723 * allocation groups upward, wrapping at the end.
1724 */
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001725 agno = start_agno;
1726 for (;;) {
1727 pag = xfs_perag_get(mp, agno);
1728 if (!pag->pagi_inodeok) {
1729 xfs_ialloc_next_ag(mp);
1730 goto nextag;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001731 }
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001732
1733 if (!pag->pagi_init) {
1734 error = xfs_ialloc_pagi_init(mp, tp, agno);
1735 if (error)
1736 goto out_error;
1737 }
1738
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001739 /*
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001740 * Do a first racy fast path check if this AG is usable.
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001741 */
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001742 if (!pag->pagi_freecount && !okalloc)
1743 goto nextag;
1744
Christoph Hellwigc4982112012-08-07 02:02:02 -04001745 /*
1746 * Then read in the AGI buffer and recheck with the AGI buffer
1747 * lock held.
1748 */
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001749 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1750 if (error)
1751 goto out_error;
1752
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001753 if (pag->pagi_freecount) {
1754 xfs_perag_put(pag);
1755 goto out_alloc;
1756 }
1757
Christoph Hellwigc4982112012-08-07 02:02:02 -04001758 if (!okalloc)
1759 goto nextag_relse_buffer;
1760
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001761
1762 error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1763 if (error) {
1764 xfs_trans_brelse(tp, agbp);
1765
Dave Chinner24513372014-06-25 14:58:08 +10001766 if (error != -ENOSPC)
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001767 goto out_error;
1768
1769 xfs_perag_put(pag);
1770 *inop = NULLFSINO;
1771 return 0;
1772 }
1773
1774 if (ialloced) {
1775 /*
1776 * We successfully allocated some inodes, return
1777 * the current context to the caller so that it
1778 * can commit the current transaction and call
1779 * us again where we left off.
1780 */
1781 ASSERT(pag->pagi_freecount > 0);
1782 xfs_perag_put(pag);
1783
1784 *IO_agbp = agbp;
1785 *inop = NULLFSINO;
1786 return 0;
1787 }
1788
Christoph Hellwigc4982112012-08-07 02:02:02 -04001789nextag_relse_buffer:
1790 xfs_trans_brelse(tp, agbp);
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001791nextag:
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001792 xfs_perag_put(pag);
1793 if (++agno == mp->m_sb.sb_agcount)
1794 agno = 0;
1795 if (agno == start_agno) {
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001796 *inop = NULLFSINO;
Dave Chinner24513372014-06-25 14:58:08 +10001797 return noroom ? -ENOSPC : 0;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001798 }
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001799 }
1800
Christoph Hellwig4bb61062012-07-04 10:54:48 -04001801out_alloc:
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001802 *IO_agbp = NULL;
1803 return xfs_dialloc_ag(tp, agbp, parent, inop);
Christoph Hellwigbe60fe52012-07-04 10:54:50 -04001804out_error:
1805 xfs_perag_put(pag);
Eric Sandeenb474c7a2014-06-22 15:04:54 +10001806 return error;
Christoph Hellwigf2ecc5e2012-07-04 10:54:46 -04001807}
1808
Brian Foster10ae3dc72015-05-29 09:22:52 +10001809/*
1810 * Free the blocks of an inode chunk. We must consider that the inode chunk
1811 * might be sparse and only free the regions that are allocated as part of the
1812 * chunk.
1813 */
1814STATIC void
1815xfs_difree_inode_chunk(
1816 struct xfs_mount *mp,
1817 xfs_agnumber_t agno,
1818 struct xfs_inobt_rec_incore *rec,
1819 struct xfs_bmap_free *flist)
1820{
1821 xfs_agblock_t sagbno = XFS_AGINO_TO_AGBNO(mp, rec->ir_startino);
1822 int startidx, endidx;
1823 int nextbit;
1824 xfs_agblock_t agbno;
1825 int contigblk;
1826 DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1827
1828 if (!xfs_inobt_issparse(rec->ir_holemask)) {
1829 /* not sparse, calculate extent info directly */
1830 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno,
1831 XFS_AGINO_TO_AGBNO(mp, rec->ir_startino)),
1832 mp->m_ialloc_blks, flist, mp);
1833 return;
1834 }
1835
1836 /* holemask is only 16-bits (fits in an unsigned long) */
1837 ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1838 holemask[0] = rec->ir_holemask;
1839
1840 /*
1841 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1842 * holemask and convert the start/end index of each range to an extent.
1843 * We start with the start and end index both pointing at the first 0 in
1844 * the mask.
1845 */
1846 startidx = endidx = find_first_zero_bit(holemask,
1847 XFS_INOBT_HOLEMASK_BITS);
1848 nextbit = startidx + 1;
1849 while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1850 nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1851 nextbit);
1852 /*
1853 * If the next zero bit is contiguous, update the end index of
1854 * the current range and continue.
1855 */
1856 if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1857 nextbit == endidx + 1) {
1858 endidx = nextbit;
1859 goto next;
1860 }
1861
1862 /*
1863 * nextbit is not contiguous with the current end index. Convert
1864 * the current start/end to an extent and add it to the free
1865 * list.
1866 */
1867 agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1868 mp->m_sb.sb_inopblock;
1869 contigblk = ((endidx - startidx + 1) *
1870 XFS_INODES_PER_HOLEMASK_BIT) /
1871 mp->m_sb.sb_inopblock;
1872
1873 ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1874 ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1875 xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno, agbno), contigblk,
1876 flist, mp);
1877
1878 /* reset range to current bit and carry on... */
1879 startidx = endidx = nextbit;
1880
1881next:
1882 nextbit++;
1883 }
1884}
1885
Brian Foster2b64ee52014-04-24 16:00:53 +10001886STATIC int
1887xfs_difree_inobt(
1888 struct xfs_mount *mp,
1889 struct xfs_trans *tp,
1890 struct xfs_buf *agbp,
1891 xfs_agino_t agino,
1892 struct xfs_bmap_free *flist,
Brian Foster09b56602015-05-29 09:26:03 +10001893 struct xfs_icluster *xic,
Brian Foster2b64ee52014-04-24 16:00:53 +10001894 struct xfs_inobt_rec_incore *orec)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895{
Brian Foster2b64ee52014-04-24 16:00:53 +10001896 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
1897 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
1898 struct xfs_perag *pag;
1899 struct xfs_btree_cur *cur;
1900 struct xfs_inobt_rec_incore rec;
1901 int ilen;
1902 int error;
1903 int i;
1904 int off;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001906 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
Brian Foster2b64ee52014-04-24 16:00:53 +10001907 ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1908
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 /*
1910 * Initialize the cursor.
1911 */
Brian Foster57bd3db2014-04-24 16:00:50 +10001912 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
Dave Chinner0b48db82009-08-31 20:57:09 -03001914 error = xfs_check_agi_freecount(cur, agi);
1915 if (error)
1916 goto error0;
1917
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 /*
1919 * Look for the entry describing this inode.
1920 */
Christoph Hellwig21875502009-08-31 20:58:21 -03001921 if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001922 xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1923 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 goto error0;
1925 }
Eric Sandeenc29aad42015-02-23 22:39:08 +11001926 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
Christoph Hellwig2e287a72009-08-31 20:56:58 -03001927 error = xfs_inobt_get_rec(cur, &rec, &i);
1928 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001929 xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1930 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 goto error0;
1932 }
Eric Sandeenc29aad42015-02-23 22:39:08 +11001933 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 /*
1935 * Get the offset in the inode chunk.
1936 */
1937 off = agino - rec.ir_startino;
1938 ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
Christoph Hellwig0d87e652009-02-09 08:37:14 +01001939 ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 /*
1941 * Mark the inode free & increment the count.
1942 */
Christoph Hellwig0d87e652009-02-09 08:37:14 +01001943 rec.ir_free |= XFS_INOBT_MASK(off);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 rec.ir_freecount++;
1945
1946 /*
Brian Foster999633d2015-05-29 08:51:37 +10001947 * When an inode chunk is free, it becomes eligible for removal. Don't
1948 * remove the chunk if the block size is large enough for multiple inode
1949 * chunks (that might not be free).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 */
Josef Jeff Sipek1bd960e2008-02-29 13:58:40 +11001951 if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
Brian Foster999633d2015-05-29 08:51:37 +10001952 rec.ir_free == XFS_INOBT_ALL_FREE &&
1953 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
Brian Foster09b56602015-05-29 09:26:03 +10001954 xic->deleted = 1;
1955 xic->first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1956 xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
1958 /*
1959 * Remove the inode cluster from the AGI B+Tree, adjust the
1960 * AGI and Superblock inode counts, and mark the disk space
1961 * to be freed when the transaction is committed.
1962 */
Brian Foster999633d2015-05-29 08:51:37 +10001963 ilen = rec.ir_freecount;
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001964 be32_add_cpu(&agi->agi_count, -ilen);
1965 be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
Dave Chinner44b56e02010-01-11 11:47:43 +00001967 pag = xfs_perag_get(mp, agno);
1968 pag->pagi_freecount -= ilen - 1;
1969 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1971 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1972
Christoph Hellwig91cca5df2008-10-30 16:58:01 +11001973 if ((error = xfs_btree_delete(cur, &i))) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001974 xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1975 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 goto error0;
1977 }
1978
Brian Foster10ae3dc72015-05-29 09:22:52 +10001979 xfs_difree_inode_chunk(mp, agno, &rec, flist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 } else {
Brian Foster09b56602015-05-29 09:26:03 +10001981 xic->deleted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
Christoph Hellwigafabc242009-08-31 20:57:03 -03001983 error = xfs_inobt_update(cur, &rec);
1984 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001985 xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1986 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 goto error0;
1988 }
Christoph Hellwigafabc242009-08-31 20:57:03 -03001989
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 /*
1991 * Change the inode free counts and log the ag/sb changes.
1992 */
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001993 be32_add_cpu(&agi->agi_freecount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
Dave Chinner44b56e02010-01-11 11:47:43 +00001995 pag = xfs_perag_get(mp, agno);
1996 pag->pagi_freecount++;
1997 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1999 }
2000
Dave Chinner0b48db82009-08-31 20:57:09 -03002001 error = xfs_check_agi_freecount(cur, agi);
2002 if (error)
2003 goto error0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
Brian Foster2b64ee52014-04-24 16:00:53 +10002005 *orec = rec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2007 return 0;
2008
2009error0:
2010 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2011 return error;
2012}
2013
Brian Foster2b64ee52014-04-24 16:00:53 +10002014/*
Brian Foster3efa4ff2014-04-24 16:00:53 +10002015 * Free an inode in the free inode btree.
2016 */
2017STATIC int
2018xfs_difree_finobt(
2019 struct xfs_mount *mp,
2020 struct xfs_trans *tp,
2021 struct xfs_buf *agbp,
2022 xfs_agino_t agino,
2023 struct xfs_inobt_rec_incore *ibtrec) /* inobt record */
2024{
2025 struct xfs_agi *agi = XFS_BUF_TO_AGI(agbp);
2026 xfs_agnumber_t agno = be32_to_cpu(agi->agi_seqno);
2027 struct xfs_btree_cur *cur;
2028 struct xfs_inobt_rec_incore rec;
2029 int offset = agino - ibtrec->ir_startino;
2030 int error;
2031 int i;
2032
2033 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_FINO);
2034
2035 error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2036 if (error)
2037 goto error;
2038 if (i == 0) {
2039 /*
2040 * If the record does not exist in the finobt, we must have just
2041 * freed an inode in a previously fully allocated chunk. If not,
2042 * something is out of sync.
2043 */
Eric Sandeenc29aad42015-02-23 22:39:08 +11002044 XFS_WANT_CORRUPTED_GOTO(mp, ibtrec->ir_freecount == 1, error);
Brian Foster3efa4ff2014-04-24 16:00:53 +10002045
Brian Foster54190402015-05-29 09:03:04 +10002046 error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2047 ibtrec->ir_count,
2048 ibtrec->ir_freecount,
Brian Foster3efa4ff2014-04-24 16:00:53 +10002049 ibtrec->ir_free, &i);
2050 if (error)
2051 goto error;
2052 ASSERT(i == 1);
2053
2054 goto out;
2055 }
2056
2057 /*
2058 * Read and update the existing record. We could just copy the ibtrec
2059 * across here, but that would defeat the purpose of having redundant
2060 * metadata. By making the modifications independently, we can catch
2061 * corruptions that we wouldn't see if we just copied from one record
2062 * to another.
2063 */
2064 error = xfs_inobt_get_rec(cur, &rec, &i);
2065 if (error)
2066 goto error;
Eric Sandeenc29aad42015-02-23 22:39:08 +11002067 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, error);
Brian Foster3efa4ff2014-04-24 16:00:53 +10002068
2069 rec.ir_free |= XFS_INOBT_MASK(offset);
2070 rec.ir_freecount++;
2071
Eric Sandeenc29aad42015-02-23 22:39:08 +11002072 XFS_WANT_CORRUPTED_GOTO(mp, (rec.ir_free == ibtrec->ir_free) &&
Brian Foster3efa4ff2014-04-24 16:00:53 +10002073 (rec.ir_freecount == ibtrec->ir_freecount),
2074 error);
2075
2076 /*
2077 * The content of inobt records should always match between the inobt
2078 * and finobt. The lifecycle of records in the finobt is different from
2079 * the inobt in that the finobt only tracks records with at least one
2080 * free inode. Hence, if all of the inodes are free and we aren't
2081 * keeping inode chunks permanently on disk, remove the record.
2082 * Otherwise, update the record with the new information.
Brian Foster999633d2015-05-29 08:51:37 +10002083 *
2084 * Note that we currently can't free chunks when the block size is large
2085 * enough for multiple chunks. Leave the finobt record to remain in sync
2086 * with the inobt.
Brian Foster3efa4ff2014-04-24 16:00:53 +10002087 */
Brian Foster999633d2015-05-29 08:51:37 +10002088 if (rec.ir_free == XFS_INOBT_ALL_FREE &&
2089 mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK &&
Brian Foster3efa4ff2014-04-24 16:00:53 +10002090 !(mp->m_flags & XFS_MOUNT_IKEEP)) {
2091 error = xfs_btree_delete(cur, &i);
2092 if (error)
2093 goto error;
2094 ASSERT(i == 1);
2095 } else {
2096 error = xfs_inobt_update(cur, &rec);
2097 if (error)
2098 goto error;
2099 }
2100
2101out:
2102 error = xfs_check_agi_freecount(cur, agi);
2103 if (error)
2104 goto error;
2105
2106 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2107 return 0;
2108
2109error:
2110 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2111 return error;
2112}
2113
2114/*
Brian Foster2b64ee52014-04-24 16:00:53 +10002115 * Free disk inode. Carefully avoids touching the incore inode, all
2116 * manipulations incore are the caller's responsibility.
2117 * The on-disk inode is not changed by this operation, only the
2118 * btree (free inode mask) is changed.
2119 */
2120int
2121xfs_difree(
2122 struct xfs_trans *tp, /* transaction pointer */
2123 xfs_ino_t inode, /* inode to be freed */
2124 struct xfs_bmap_free *flist, /* extents to free */
Brian Foster09b56602015-05-29 09:26:03 +10002125 struct xfs_icluster *xic) /* cluster info if deleted */
Brian Foster2b64ee52014-04-24 16:00:53 +10002126{
2127 /* REFERENCED */
2128 xfs_agblock_t agbno; /* block number containing inode */
2129 struct xfs_buf *agbp; /* buffer for allocation group header */
2130 xfs_agino_t agino; /* allocation group inode number */
2131 xfs_agnumber_t agno; /* allocation group number */
2132 int error; /* error return value */
2133 struct xfs_mount *mp; /* mount structure for filesystem */
2134 struct xfs_inobt_rec_incore rec;/* btree record */
2135
2136 mp = tp->t_mountp;
2137
2138 /*
2139 * Break up inode number into its components.
2140 */
2141 agno = XFS_INO_TO_AGNO(mp, inode);
2142 if (agno >= mp->m_sb.sb_agcount) {
2143 xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
2144 __func__, agno, mp->m_sb.sb_agcount);
2145 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10002146 return -EINVAL;
Brian Foster2b64ee52014-04-24 16:00:53 +10002147 }
2148 agino = XFS_INO_TO_AGINO(mp, inode);
2149 if (inode != XFS_AGINO_TO_INO(mp, agno, agino)) {
2150 xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2151 __func__, (unsigned long long)inode,
2152 (unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
2153 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10002154 return -EINVAL;
Brian Foster2b64ee52014-04-24 16:00:53 +10002155 }
2156 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2157 if (agbno >= mp->m_sb.sb_agblocks) {
2158 xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2159 __func__, agbno, mp->m_sb.sb_agblocks);
2160 ASSERT(0);
Dave Chinner24513372014-06-25 14:58:08 +10002161 return -EINVAL;
Brian Foster2b64ee52014-04-24 16:00:53 +10002162 }
2163 /*
2164 * Get the allocation group header.
2165 */
2166 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2167 if (error) {
2168 xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2169 __func__, error);
2170 return error;
2171 }
2172
2173 /*
2174 * Fix up the inode allocation btree.
2175 */
Brian Foster09b56602015-05-29 09:26:03 +10002176 error = xfs_difree_inobt(mp, tp, agbp, agino, flist, xic, &rec);
Brian Foster2b64ee52014-04-24 16:00:53 +10002177 if (error)
2178 goto error0;
2179
Brian Foster3efa4ff2014-04-24 16:00:53 +10002180 /*
2181 * Fix up the free inode btree.
2182 */
2183 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
2184 error = xfs_difree_finobt(mp, tp, agbp, agino, &rec);
2185 if (error)
2186 goto error0;
2187 }
2188
Brian Foster2b64ee52014-04-24 16:00:53 +10002189 return 0;
2190
2191error0:
2192 return error;
2193}
2194
Dave Chinner7124fe02010-06-24 11:15:33 +10002195STATIC int
2196xfs_imap_lookup(
2197 struct xfs_mount *mp,
2198 struct xfs_trans *tp,
2199 xfs_agnumber_t agno,
2200 xfs_agino_t agino,
2201 xfs_agblock_t agbno,
2202 xfs_agblock_t *chunk_agbno,
2203 xfs_agblock_t *offset_agbno,
2204 int flags)
2205{
2206 struct xfs_inobt_rec_incore rec;
2207 struct xfs_btree_cur *cur;
2208 struct xfs_buf *agbp;
Dave Chinner7124fe02010-06-24 11:15:33 +10002209 int error;
2210 int i;
2211
2212 error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
2213 if (error) {
Dave Chinner53487782011-03-07 10:05:35 +11002214 xfs_alert(mp,
2215 "%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2216 __func__, error, agno);
Dave Chinner7124fe02010-06-24 11:15:33 +10002217 return error;
2218 }
2219
2220 /*
Dave Chinner4536f2a2010-08-24 11:42:30 +10002221 * Lookup the inode record for the given agino. If the record cannot be
2222 * found, then it's an invalid inode number and we should abort. Once
2223 * we have a record, we need to ensure it contains the inode number
2224 * we are looking up.
Dave Chinner7124fe02010-06-24 11:15:33 +10002225 */
Brian Foster57bd3db2014-04-24 16:00:50 +10002226 cur = xfs_inobt_init_cursor(mp, tp, agbp, agno, XFS_BTNUM_INO);
Dave Chinner4536f2a2010-08-24 11:42:30 +10002227 error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
Dave Chinner7124fe02010-06-24 11:15:33 +10002228 if (!error) {
2229 if (i)
2230 error = xfs_inobt_get_rec(cur, &rec, &i);
2231 if (!error && i == 0)
Dave Chinner24513372014-06-25 14:58:08 +10002232 error = -EINVAL;
Dave Chinner7124fe02010-06-24 11:15:33 +10002233 }
2234
2235 xfs_trans_brelse(tp, agbp);
Brian Fosterf3070802015-08-19 10:00:53 +10002236 xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
Dave Chinner7124fe02010-06-24 11:15:33 +10002237 if (error)
2238 return error;
2239
Dave Chinner4536f2a2010-08-24 11:42:30 +10002240 /* check that the returned record contains the required inode */
2241 if (rec.ir_startino > agino ||
Jie Liu71783432013-12-13 15:51:46 +11002242 rec.ir_startino + mp->m_ialloc_inos <= agino)
Dave Chinner24513372014-06-25 14:58:08 +10002243 return -EINVAL;
Dave Chinner4536f2a2010-08-24 11:42:30 +10002244
Dave Chinner7124fe02010-06-24 11:15:33 +10002245 /* for untrusted inodes check it is allocated first */
Dave Chinner19207792010-06-24 11:15:47 +10002246 if ((flags & XFS_IGET_UNTRUSTED) &&
Dave Chinner7124fe02010-06-24 11:15:33 +10002247 (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
Dave Chinner24513372014-06-25 14:58:08 +10002248 return -EINVAL;
Dave Chinner7124fe02010-06-24 11:15:33 +10002249
2250 *chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2251 *offset_agbno = agbno - *chunk_agbno;
2252 return 0;
2253}
2254
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255/*
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002256 * Return the location of the inode in imap, for mapping it into a buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258int
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002259xfs_imap(
2260 xfs_mount_t *mp, /* file system mount structure */
2261 xfs_trans_t *tp, /* transaction pointer */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 xfs_ino_t ino, /* inode to locate */
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002263 struct xfs_imap *imap, /* location map structure */
2264 uint flags) /* flags for inode btree lookup */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265{
2266 xfs_agblock_t agbno; /* block number of inode in the alloc group */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 xfs_agino_t agino; /* inode number within alloc group */
2268 xfs_agnumber_t agno; /* allocation group number */
2269 int blks_per_cluster; /* num blocks per inode cluster */
2270 xfs_agblock_t chunk_agbno; /* first block in inode chunk */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 xfs_agblock_t cluster_agbno; /* first block in inode cluster */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 int error; /* error code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 int offset; /* index of inode in its buffer */
Dave Chinner836a94a2013-08-12 20:49:44 +10002274 xfs_agblock_t offset_agbno; /* blks from chunk start to inode */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
2276 ASSERT(ino != NULLFSINO);
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002277
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 /*
2279 * Split up the inode number into its parts.
2280 */
2281 agno = XFS_INO_TO_AGNO(mp, ino);
2282 agino = XFS_INO_TO_AGINO(mp, ino);
2283 agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2284 if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
2285 ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
2286#ifdef DEBUG
Dave Chinner19207792010-06-24 11:15:47 +10002287 /*
2288 * Don't output diagnostic information for untrusted inodes
2289 * as they can be invalid without implying corruption.
2290 */
2291 if (flags & XFS_IGET_UNTRUSTED)
Dave Chinner24513372014-06-25 14:58:08 +10002292 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 if (agno >= mp->m_sb.sb_agcount) {
Dave Chinner53487782011-03-07 10:05:35 +11002294 xfs_alert(mp,
2295 "%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
2296 __func__, agno, mp->m_sb.sb_agcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 }
2298 if (agbno >= mp->m_sb.sb_agblocks) {
Dave Chinner53487782011-03-07 10:05:35 +11002299 xfs_alert(mp,
2300 "%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2301 __func__, (unsigned long long)agbno,
2302 (unsigned long)mp->m_sb.sb_agblocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 }
2304 if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
Dave Chinner53487782011-03-07 10:05:35 +11002305 xfs_alert(mp,
2306 "%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2307 __func__, ino,
2308 XFS_AGINO_TO_INO(mp, agno, agino));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 }
Nathan Scott745b1f472006-09-28 11:02:23 +10002310 xfs_stack_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311#endif /* DEBUG */
Dave Chinner24513372014-06-25 14:58:08 +10002312 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 }
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002314
Jie Liuf9e5abc2013-12-13 15:51:49 +11002315 blks_per_cluster = xfs_icluster_size_fsb(mp);
Dave Chinner7124fe02010-06-24 11:15:33 +10002316
2317 /*
2318 * For bulkstat and handle lookups, we have an untrusted inode number
2319 * that we have to verify is valid. We cannot do this just by reading
2320 * the inode buffer as it may have been unlinked and removed leaving
2321 * inodes in stale state on disk. Hence we have to do a btree lookup
2322 * in all cases where an untrusted inode number is passed.
2323 */
Dave Chinner19207792010-06-24 11:15:47 +10002324 if (flags & XFS_IGET_UNTRUSTED) {
Dave Chinner7124fe02010-06-24 11:15:33 +10002325 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2326 &chunk_agbno, &offset_agbno, flags);
2327 if (error)
2328 return error;
2329 goto out_map;
2330 }
2331
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002332 /*
2333 * If the inode cluster size is the same as the blocksize or
2334 * smaller we get to the buffer by simple arithmetics.
2335 */
Jie Liuf9e5abc2013-12-13 15:51:49 +11002336 if (blks_per_cluster == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 offset = XFS_INO_TO_OFFSET(mp, ino);
2338 ASSERT(offset < mp->m_sb.sb_inopblock);
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002339
2340 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
2341 imap->im_len = XFS_FSB_TO_BB(mp, 1);
2342 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 return 0;
2344 }
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002345
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002346 /*
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002347 * If the inode chunks are aligned then use simple maths to
2348 * find the location. Otherwise we have to do a btree
2349 * lookup to find the location.
2350 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 if (mp->m_inoalign_mask) {
2352 offset_agbno = agbno & mp->m_inoalign_mask;
2353 chunk_agbno = agbno - offset_agbno;
2354 } else {
Dave Chinner7124fe02010-06-24 11:15:33 +10002355 error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
2356 &chunk_agbno, &offset_agbno, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 if (error)
2358 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 }
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002360
Dave Chinner7124fe02010-06-24 11:15:33 +10002361out_map:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 ASSERT(agbno >= chunk_agbno);
2363 cluster_agbno = chunk_agbno +
2364 ((offset_agbno / blks_per_cluster) * blks_per_cluster);
2365 offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2366 XFS_INO_TO_OFFSET(mp, ino);
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002367
2368 imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
2369 imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
2370 imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
2371
2372 /*
2373 * If the inode number maps to a block outside the bounds
2374 * of the file system then return NULL rather than calling
2375 * read_buf and panicing when we get an error from the
2376 * driver.
2377 */
2378 if ((imap->im_blkno + imap->im_len) >
2379 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
Dave Chinner53487782011-03-07 10:05:35 +11002380 xfs_alert(mp,
2381 "%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2382 __func__, (unsigned long long) imap->im_blkno,
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002383 (unsigned long long) imap->im_len,
2384 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
Dave Chinner24513372014-06-25 14:58:08 +10002385 return -EINVAL;
Christoph Hellwig94e1b692008-11-28 14:23:41 +11002386 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388}
2389
2390/*
2391 * Compute and fill in value of m_in_maxlevels.
2392 */
2393void
2394xfs_ialloc_compute_maxlevels(
2395 xfs_mount_t *mp) /* file system mount structure */
2396{
2397 int level;
2398 uint maxblocks;
2399 uint maxleafents;
2400 int minleafrecs;
2401 int minnoderecs;
2402
2403 maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
2404 XFS_INODES_PER_CHUNK_LOG;
2405 minleafrecs = mp->m_alloc_mnr[0];
2406 minnoderecs = mp->m_alloc_mnr[1];
2407 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
2408 for (level = 1; maxblocks > 1; level++)
2409 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
2410 mp->m_in_maxlevels = level;
2411}
2412
2413/*
Brian Fosteraafc3c22014-04-24 16:00:52 +10002414 * Log specified fields for the ag hdr (inode section). The growth of the agi
2415 * structure over time requires that we interpret the buffer as two logical
2416 * regions delineated by the end of the unlinked list. This is due to the size
2417 * of the hash table and its location in the middle of the agi.
2418 *
2419 * For example, a request to log a field before agi_unlinked and a field after
2420 * agi_unlinked could cause us to log the entire hash table and use an excessive
2421 * amount of log space. To avoid this behavior, log the region up through
2422 * agi_unlinked in one call and the region after agi_unlinked through the end of
2423 * the structure in another.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 */
2425void
2426xfs_ialloc_log_agi(
2427 xfs_trans_t *tp, /* transaction pointer */
2428 xfs_buf_t *bp, /* allocation group header buffer */
2429 int fields) /* bitmask of fields to log */
2430{
2431 int first; /* first byte number */
2432 int last; /* last byte number */
2433 static const short offsets[] = { /* field starting offsets */
2434 /* keep in sync with bit definitions */
2435 offsetof(xfs_agi_t, agi_magicnum),
2436 offsetof(xfs_agi_t, agi_versionnum),
2437 offsetof(xfs_agi_t, agi_seqno),
2438 offsetof(xfs_agi_t, agi_length),
2439 offsetof(xfs_agi_t, agi_count),
2440 offsetof(xfs_agi_t, agi_root),
2441 offsetof(xfs_agi_t, agi_level),
2442 offsetof(xfs_agi_t, agi_freecount),
2443 offsetof(xfs_agi_t, agi_newino),
2444 offsetof(xfs_agi_t, agi_dirino),
2445 offsetof(xfs_agi_t, agi_unlinked),
Brian Fosteraafc3c22014-04-24 16:00:52 +10002446 offsetof(xfs_agi_t, agi_free_root),
2447 offsetof(xfs_agi_t, agi_free_level),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 sizeof(xfs_agi_t)
2449 };
2450#ifdef DEBUG
2451 xfs_agi_t *agi; /* allocation group header */
2452
2453 agi = XFS_BUF_TO_AGI(bp);
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002454 ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455#endif
Brian Fosteraafc3c22014-04-24 16:00:52 +10002456
Dave Chinner61fe1352013-04-03 16:11:30 +11002457 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF);
Brian Fosteraafc3c22014-04-24 16:00:52 +10002458
2459 /*
2460 * Compute byte offsets for the first and last fields in the first
2461 * region and log the agi buffer. This only logs up through
2462 * agi_unlinked.
2463 */
2464 if (fields & XFS_AGI_ALL_BITS_R1) {
2465 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2466 &first, &last);
2467 xfs_trans_log_buf(tp, bp, first, last);
2468 }
2469
2470 /*
2471 * Mask off the bits in the first region and calculate the first and
2472 * last field offsets for any bits in the second region.
2473 */
2474 fields &= ~XFS_AGI_ALL_BITS_R1;
2475 if (fields) {
2476 xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2477 &first, &last);
2478 xfs_trans_log_buf(tp, bp, first, last);
2479 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480}
2481
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002482#ifdef DEBUG
2483STATIC void
2484xfs_check_agi_unlinked(
2485 struct xfs_agi *agi)
2486{
2487 int i;
2488
2489 for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
2490 ASSERT(agi->agi_unlinked[i]);
2491}
2492#else
2493#define xfs_check_agi_unlinked(agi)
2494#endif
2495
Dave Chinner983d09f2013-04-03 16:11:15 +11002496static bool
Dave Chinner612cfbf2012-11-14 17:52:32 +11002497xfs_agi_verify(
Dave Chinner3702ce62012-11-12 22:54:05 +11002498 struct xfs_buf *bp)
2499{
2500 struct xfs_mount *mp = bp->b_target->bt_mount;
2501 struct xfs_agi *agi = XFS_BUF_TO_AGI(bp);
Dave Chinner3702ce62012-11-12 22:54:05 +11002502
Dave Chinner983d09f2013-04-03 16:11:15 +11002503 if (xfs_sb_version_hascrc(&mp->m_sb) &&
Eric Sandeence748ea2015-07-29 11:53:31 +10002504 !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
Dave Chinner983d09f2013-04-03 16:11:15 +11002505 return false;
Dave Chinner3702ce62012-11-12 22:54:05 +11002506 /*
2507 * Validate the magic number of the agi block.
2508 */
Dave Chinner983d09f2013-04-03 16:11:15 +11002509 if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
2510 return false;
2511 if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2512 return false;
Dave Chinner3702ce62012-11-12 22:54:05 +11002513
Eric Sandeene1b05722014-09-09 11:47:24 +10002514 if (be32_to_cpu(agi->agi_level) > XFS_BTREE_MAXLEVELS)
2515 return false;
Dave Chinner3702ce62012-11-12 22:54:05 +11002516 /*
2517 * during growfs operations, the perag is not fully initialised,
2518 * so we can't use it for any useful checking. growfs ensures we can't
2519 * use it by using uncached buffers that don't have the perag attached
2520 * so we can detect and avoid this problem.
2521 */
Dave Chinner983d09f2013-04-03 16:11:15 +11002522 if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
2523 return false;
Dave Chinner3702ce62012-11-12 22:54:05 +11002524
Dave Chinner3702ce62012-11-12 22:54:05 +11002525 xfs_check_agi_unlinked(agi);
Dave Chinner983d09f2013-04-03 16:11:15 +11002526 return true;
Dave Chinner612cfbf2012-11-14 17:52:32 +11002527}
2528
Dave Chinner1813dd62012-11-14 17:54:40 +11002529static void
2530xfs_agi_read_verify(
Dave Chinner612cfbf2012-11-14 17:52:32 +11002531 struct xfs_buf *bp)
2532{
Dave Chinner983d09f2013-04-03 16:11:15 +11002533 struct xfs_mount *mp = bp->b_target->bt_mount;
Dave Chinner983d09f2013-04-03 16:11:15 +11002534
Eric Sandeence5028c2014-02-27 15:23:10 +11002535 if (xfs_sb_version_hascrc(&mp->m_sb) &&
2536 !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
Dave Chinner24513372014-06-25 14:58:08 +10002537 xfs_buf_ioerror(bp, -EFSBADCRC);
Eric Sandeence5028c2014-02-27 15:23:10 +11002538 else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp,
2539 XFS_ERRTAG_IALLOC_READ_AGI,
2540 XFS_RANDOM_IALLOC_READ_AGI))
Dave Chinner24513372014-06-25 14:58:08 +10002541 xfs_buf_ioerror(bp, -EFSCORRUPTED);
Eric Sandeence5028c2014-02-27 15:23:10 +11002542
2543 if (bp->b_error)
2544 xfs_verifier_error(bp);
Dave Chinner612cfbf2012-11-14 17:52:32 +11002545}
2546
Dave Chinnerb0f539d2012-11-14 17:53:49 +11002547static void
Dave Chinner1813dd62012-11-14 17:54:40 +11002548xfs_agi_write_verify(
Dave Chinner612cfbf2012-11-14 17:52:32 +11002549 struct xfs_buf *bp)
2550{
Dave Chinner983d09f2013-04-03 16:11:15 +11002551 struct xfs_mount *mp = bp->b_target->bt_mount;
2552 struct xfs_buf_log_item *bip = bp->b_fspriv;
2553
2554 if (!xfs_agi_verify(bp)) {
Dave Chinner24513372014-06-25 14:58:08 +10002555 xfs_buf_ioerror(bp, -EFSCORRUPTED);
Eric Sandeence5028c2014-02-27 15:23:10 +11002556 xfs_verifier_error(bp);
Dave Chinner983d09f2013-04-03 16:11:15 +11002557 return;
2558 }
2559
2560 if (!xfs_sb_version_hascrc(&mp->m_sb))
2561 return;
2562
2563 if (bip)
2564 XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
Eric Sandeenf1dbcd72014-02-27 15:18:23 +11002565 xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
Dave Chinner3702ce62012-11-12 22:54:05 +11002566}
2567
Dave Chinner1813dd62012-11-14 17:54:40 +11002568const struct xfs_buf_ops xfs_agi_buf_ops = {
2569 .verify_read = xfs_agi_read_verify,
2570 .verify_write = xfs_agi_write_verify,
2571};
2572
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573/*
2574 * Read in the allocation group header (inode allocation section)
2575 */
2576int
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002577xfs_read_agi(
2578 struct xfs_mount *mp, /* file system mount structure */
2579 struct xfs_trans *tp, /* transaction pointer */
2580 xfs_agnumber_t agno, /* allocation group number */
2581 struct xfs_buf **bpp) /* allocation group hdr buf */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582{
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002583 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584
Dave Chinnerd1230312013-11-01 15:27:19 +11002585 trace_xfs_read_agi(mp, agno);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002586
Dave Chinnerd1230312013-11-01 15:27:19 +11002587 ASSERT(agno != NULLAGNUMBER);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002588 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
Dave Chinner1813dd62012-11-14 17:54:40 +11002590 XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 if (error)
2592 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002593
Christoph Hellwig38f23232011-10-10 16:52:45 +00002594 xfs_buf_set_ref(*bpp, XFS_AGI_REF);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002595 return 0;
2596}
2597
2598int
2599xfs_ialloc_read_agi(
2600 struct xfs_mount *mp, /* file system mount structure */
2601 struct xfs_trans *tp, /* transaction pointer */
2602 xfs_agnumber_t agno, /* allocation group number */
2603 struct xfs_buf **bpp) /* allocation group hdr buf */
2604{
2605 struct xfs_agi *agi; /* allocation group header */
2606 struct xfs_perag *pag; /* per allocation group data */
2607 int error;
2608
Dave Chinnerd1230312013-11-01 15:27:19 +11002609 trace_xfs_ialloc_read_agi(mp, agno);
2610
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002611 error = xfs_read_agi(mp, tp, agno, bpp);
2612 if (error)
2613 return error;
2614
2615 agi = XFS_BUF_TO_AGI(*bpp);
Dave Chinner44b56e02010-01-11 11:47:43 +00002616 pag = xfs_perag_get(mp, agno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 if (!pag->pagi_init) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11002618 pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
David Chinner92821e22007-05-24 15:26:31 +10002619 pag->pagi_count = be32_to_cpu(agi->agi_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 pag->pagi_init = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 }
2622
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11002623 /*
2624 * It's possible for these to be out of sync if
2625 * we are in the middle of a forced shutdown.
2626 */
2627 ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2628 XFS_FORCED_SHUTDOWN(mp));
Dave Chinner44b56e02010-01-11 11:47:43 +00002629 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 return 0;
2631}
David Chinner92821e22007-05-24 15:26:31 +10002632
2633/*
2634 * Read in the agi to initialise the per-ag data in the mount structure
2635 */
2636int
2637xfs_ialloc_pagi_init(
2638 xfs_mount_t *mp, /* file system mount structure */
2639 xfs_trans_t *tp, /* transaction pointer */
2640 xfs_agnumber_t agno) /* allocation group number */
2641{
2642 xfs_buf_t *bp = NULL;
2643 int error;
2644
2645 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
2646 if (error)
2647 return error;
2648 if (bp)
2649 xfs_trans_brelse(tp, bp);
2650 return 0;
2651}