blob: e569bf5d6cf0d9509356e93aef781048202c669b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Olaf Weber3e57ecf2006-06-09 14:48:12 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * 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 */
Robert P. J. Day40ebd812007-11-23 16:30:51 +110018#include <linux/log2.h>
19
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_inum.h"
26#include "xfs_imap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_trans.h"
28#include "xfs_trans_priv.h"
29#include "xfs_sb.h"
30#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_dir2.h"
32#include "xfs_dmapi.h"
33#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110042#include "xfs_inode_item.h"
43#include "xfs_btree.h"
44#include "xfs_alloc.h"
45#include "xfs_ialloc.h"
46#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "xfs_rw.h"
48#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include "xfs_utils.h"
50#include "xfs_dir2_trace.h"
51#include "xfs_quota.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include "xfs_acl.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100053#include "xfs_filestream.h"
Christoph Hellwig739bfb22007-08-29 10:58:01 +100054#include "xfs_vnodeops.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056kmem_zone_t *xfs_ifork_zone;
57kmem_zone_t *xfs_inode_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59/*
60 * Used in xfs_itruncate(). This is the maximum number of extents
61 * freed from a file in a single transaction.
62 */
63#define XFS_ITRUNC_MAX_EXTENTS 2
64
65STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
66STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
67STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
68STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#ifdef DEBUG
71/*
72 * Make sure that the extents in the given memory buffer
73 * are valid.
74 */
75STATIC void
76xfs_validate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +110077 xfs_ifork_t *ifp,
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 int nrecs,
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 xfs_exntfmt_t fmt)
80{
81 xfs_bmbt_irec_t irec;
Christoph Hellwiga6f64d42007-08-16 16:23:40 +100082 xfs_bmbt_rec_host_t rec;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 int i;
84
85 for (i = 0; i < nrecs; i++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +100086 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
87 rec.l0 = get_unaligned(&ep->l0);
88 rec.l1 = get_unaligned(&ep->l1);
89 xfs_bmbt_get_all(&rec, &irec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 if (fmt == XFS_EXTFMT_NOSTATE)
91 ASSERT(irec.br_state == XFS_EXT_NORM);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 }
93}
94#else /* DEBUG */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +100095#define xfs_validate_extents(ifp, nrecs, fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096#endif /* DEBUG */
97
98/*
99 * Check that none of the inode's in the buffer have a next
100 * unlinked field of 0.
101 */
102#if defined(DEBUG)
103void
104xfs_inobp_check(
105 xfs_mount_t *mp,
106 xfs_buf_t *bp)
107{
108 int i;
109 int j;
110 xfs_dinode_t *dip;
111
112 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
113
114 for (i = 0; i < j; i++) {
115 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
116 i * mp->m_sb.sb_inodesize);
117 if (!dip->di_next_unlinked) {
118 xfs_fs_cmn_err(CE_ALERT, mp,
119 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
120 bp);
121 ASSERT(dip->di_next_unlinked);
122 }
123 }
124}
125#endif
126
127/*
David Chinner4ae29b42008-03-06 13:43:34 +1100128 * Find the buffer associated with the given inode map
129 * We do basic validation checks on the buffer once it has been
130 * retrieved from disk.
131 */
132STATIC int
133xfs_imap_to_bp(
134 xfs_mount_t *mp,
135 xfs_trans_t *tp,
136 xfs_imap_t *imap,
137 xfs_buf_t **bpp,
138 uint buf_flags,
139 uint imap_flags)
140{
141 int error;
142 int i;
143 int ni;
144 xfs_buf_t *bp;
145
146 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
David Chinnera3f74ff2008-03-06 13:43:42 +1100147 (int)imap->im_len, buf_flags, &bp);
David Chinner4ae29b42008-03-06 13:43:34 +1100148 if (error) {
David Chinnera3f74ff2008-03-06 13:43:42 +1100149 if (error != EAGAIN) {
150 cmn_err(CE_WARN,
151 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
David Chinner4ae29b42008-03-06 13:43:34 +1100152 "an error %d on %s. Returning error.",
153 error, mp->m_fsname);
David Chinnera3f74ff2008-03-06 13:43:42 +1100154 } else {
155 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
156 }
David Chinner4ae29b42008-03-06 13:43:34 +1100157 return error;
158 }
159
160 /*
161 * Validate the magic number and version of every inode in the buffer
162 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
163 */
164#ifdef DEBUG
165 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
166#else /* usual case */
167 ni = 1;
168#endif
169
170 for (i = 0; i < ni; i++) {
171 int di_ok;
172 xfs_dinode_t *dip;
173
174 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
175 (i << mp->m_sb.sb_inodelog));
176 di_ok = be16_to_cpu(dip->di_core.di_magic) == XFS_DINODE_MAGIC &&
177 XFS_DINODE_GOOD_VERSION(dip->di_core.di_version);
178 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
179 XFS_ERRTAG_ITOBP_INOTOBP,
180 XFS_RANDOM_ITOBP_INOTOBP))) {
181 if (imap_flags & XFS_IMAP_BULKSTAT) {
182 xfs_trans_brelse(tp, bp);
183 return XFS_ERROR(EINVAL);
184 }
185 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
186 XFS_ERRLEVEL_HIGH, mp, dip);
187#ifdef DEBUG
188 cmn_err(CE_PANIC,
189 "Device %s - bad inode magic/vsn "
190 "daddr %lld #%d (magic=%x)",
191 XFS_BUFTARG_NAME(mp->m_ddev_targp),
192 (unsigned long long)imap->im_blkno, i,
193 be16_to_cpu(dip->di_core.di_magic));
194#endif
195 xfs_trans_brelse(tp, bp);
196 return XFS_ERROR(EFSCORRUPTED);
197 }
198 }
199
200 xfs_inobp_check(mp, bp);
201
202 /*
203 * Mark the buffer as an inode buffer now that it looks good
204 */
205 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
206
207 *bpp = bp;
208 return 0;
209}
210
211/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * This routine is called to map an inode number within a file
213 * system to the buffer containing the on-disk version of the
214 * inode. It returns a pointer to the buffer containing the
215 * on-disk inode in the bpp parameter, and in the dip parameter
216 * it returns a pointer to the on-disk inode within that buffer.
217 *
218 * If a non-zero error is returned, then the contents of bpp and
219 * dipp are undefined.
220 *
221 * Use xfs_imap() to determine the size and location of the
222 * buffer to read from disk.
223 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000224STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225xfs_inotobp(
226 xfs_mount_t *mp,
227 xfs_trans_t *tp,
228 xfs_ino_t ino,
229 xfs_dinode_t **dipp,
230 xfs_buf_t **bpp,
231 int *offset)
232{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 xfs_imap_t imap;
234 xfs_buf_t *bp;
235 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 imap.im_blkno = 0;
238 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
David Chinner4ae29b42008-03-06 13:43:34 +1100239 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
David Chinner4ae29b42008-03-06 13:43:34 +1100242 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, 0);
243 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
247 *bpp = bp;
248 *offset = imap.im_boffset;
249 return 0;
250}
251
252
253/*
254 * This routine is called to map an inode to the buffer containing
255 * the on-disk version of the inode. It returns a pointer to the
256 * buffer containing the on-disk inode in the bpp parameter, and in
257 * the dip parameter it returns a pointer to the on-disk inode within
258 * that buffer.
259 *
260 * If a non-zero error is returned, then the contents of bpp and
261 * dipp are undefined.
262 *
263 * If the inode is new and has not yet been initialized, use xfs_imap()
264 * to determine the size and location of the buffer to read from disk.
265 * If the inode has already been mapped to its buffer and read in once,
266 * then use the mapping information stored in the inode rather than
267 * calling xfs_imap(). This allows us to avoid the overhead of looking
268 * at the inode btree for small block file systems (see xfs_dilocate()).
269 * We can tell whether the inode has been mapped in before by comparing
270 * its disk block address to 0. Only uninitialized inodes will have
271 * 0 for the disk block address.
272 */
273int
274xfs_itobp(
275 xfs_mount_t *mp,
276 xfs_trans_t *tp,
277 xfs_inode_t *ip,
278 xfs_dinode_t **dipp,
279 xfs_buf_t **bpp,
Nathan Scottb12dd342006-03-17 17:26:04 +1100280 xfs_daddr_t bno,
David Chinnera3f74ff2008-03-06 13:43:42 +1100281 uint imap_flags,
282 uint buf_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283{
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000284 xfs_imap_t imap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 xfs_buf_t *bp;
286 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
288 if (ip->i_blkno == (xfs_daddr_t)0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 imap.im_blkno = bno;
David Chinner4ae29b42008-03-06 13:43:34 +1100290 error = xfs_imap(mp, tp, ip->i_ino, &imap,
291 XFS_IMAP_LOOKUP | imap_flags);
292 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
295 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 * Fill in the fields in the inode that will be used to
297 * map the inode to its buffer from now on.
298 */
299 ip->i_blkno = imap.im_blkno;
300 ip->i_len = imap.im_len;
301 ip->i_boffset = imap.im_boffset;
302 } else {
303 /*
304 * We've already mapped the inode once, so just use the
305 * mapping that we saved the first time.
306 */
307 imap.im_blkno = ip->i_blkno;
308 imap.im_len = ip->i_len;
309 imap.im_boffset = ip->i_boffset;
310 }
311 ASSERT(bno == 0 || bno == imap.im_blkno);
312
David Chinnera3f74ff2008-03-06 13:43:42 +1100313 error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, imap_flags);
David Chinner4ae29b42008-03-06 13:43:34 +1100314 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 return error;
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000316
David Chinnera3f74ff2008-03-06 13:43:42 +1100317 if (!bp) {
318 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
319 ASSERT(tp == NULL);
320 *bpp = NULL;
321 return EAGAIN;
322 }
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
325 *bpp = bp;
326 return 0;
327}
328
329/*
330 * Move inode type and inode format specific information from the
331 * on-disk inode to the in-core inode. For fifos, devs, and sockets
332 * this means set if_rdev to the proper value. For files, directories,
333 * and symlinks this means to bring in the in-line data or extent
334 * pointers. For a file in B-tree format, only the root is immediately
335 * brought in-core. The rest will be in-lined in if_extents when it
336 * is first referenced (see xfs_iread_extents()).
337 */
338STATIC int
339xfs_iformat(
340 xfs_inode_t *ip,
341 xfs_dinode_t *dip)
342{
343 xfs_attr_shortform_t *atp;
344 int size;
345 int error;
346 xfs_fsize_t di_size;
347 ip->i_df.if_ext_max =
348 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
349 error = 0;
350
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000351 if (unlikely(be32_to_cpu(dip->di_core.di_nextents) +
352 be16_to_cpu(dip->di_core.di_anextents) >
353 be64_to_cpu(dip->di_core.di_nblocks))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100354 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
355 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 (unsigned long long)ip->i_ino,
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000357 (int)(be32_to_cpu(dip->di_core.di_nextents) +
358 be16_to_cpu(dip->di_core.di_anextents)),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 (unsigned long long)
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000360 be64_to_cpu(dip->di_core.di_nblocks));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
362 ip->i_mount, dip);
363 return XFS_ERROR(EFSCORRUPTED);
364 }
365
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000366 if (unlikely(dip->di_core.di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100367 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
368 "corrupt dinode %Lu, forkoff = 0x%x.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 (unsigned long long)ip->i_ino,
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000370 dip->di_core.di_forkoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
372 ip->i_mount, dip);
373 return XFS_ERROR(EFSCORRUPTED);
374 }
375
376 switch (ip->i_d.di_mode & S_IFMT) {
377 case S_IFIFO:
378 case S_IFCHR:
379 case S_IFBLK:
380 case S_IFSOCK:
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000381 if (unlikely(dip->di_core.di_format != XFS_DINODE_FMT_DEV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
383 ip->i_mount, dip);
384 return XFS_ERROR(EFSCORRUPTED);
385 }
386 ip->i_d.di_size = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000387 ip->i_size = 0;
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000388 ip->i_df.if_u2.if_rdev = be32_to_cpu(dip->di_u.di_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 break;
390
391 case S_IFREG:
392 case S_IFLNK:
393 case S_IFDIR:
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000394 switch (dip->di_core.di_format) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 case XFS_DINODE_FMT_LOCAL:
396 /*
397 * no local regular files yet
398 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000399 if (unlikely((be16_to_cpu(dip->di_core.di_mode) & S_IFMT) == S_IFREG)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100400 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
401 "corrupt inode %Lu "
402 "(local format for regular file).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 (unsigned long long) ip->i_ino);
404 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
405 XFS_ERRLEVEL_LOW,
406 ip->i_mount, dip);
407 return XFS_ERROR(EFSCORRUPTED);
408 }
409
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000410 di_size = be64_to_cpu(dip->di_core.di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100412 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
413 "corrupt inode %Lu "
414 "(bad size %Ld for local inode).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 (unsigned long long) ip->i_ino,
416 (long long) di_size);
417 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
418 XFS_ERRLEVEL_LOW,
419 ip->i_mount, dip);
420 return XFS_ERROR(EFSCORRUPTED);
421 }
422
423 size = (int)di_size;
424 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
425 break;
426 case XFS_DINODE_FMT_EXTENTS:
427 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
428 break;
429 case XFS_DINODE_FMT_BTREE:
430 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
431 break;
432 default:
433 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
434 ip->i_mount);
435 return XFS_ERROR(EFSCORRUPTED);
436 }
437 break;
438
439 default:
440 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
441 return XFS_ERROR(EFSCORRUPTED);
442 }
443 if (error) {
444 return error;
445 }
446 if (!XFS_DFORK_Q(dip))
447 return 0;
448 ASSERT(ip->i_afp == NULL);
449 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
450 ip->i_afp->if_ext_max =
451 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000452 switch (dip->di_core.di_aformat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 case XFS_DINODE_FMT_LOCAL:
454 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
Nathan Scott3b244aa2006-03-17 17:29:25 +1100455 size = be16_to_cpu(atp->hdr.totsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
457 break;
458 case XFS_DINODE_FMT_EXTENTS:
459 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
460 break;
461 case XFS_DINODE_FMT_BTREE:
462 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
463 break;
464 default:
465 error = XFS_ERROR(EFSCORRUPTED);
466 break;
467 }
468 if (error) {
469 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
470 ip->i_afp = NULL;
471 xfs_idestroy_fork(ip, XFS_DATA_FORK);
472 }
473 return error;
474}
475
476/*
477 * The file is in-lined in the on-disk inode.
478 * If it fits into if_inline_data, then copy
479 * it there, otherwise allocate a buffer for it
480 * and copy the data there. Either way, set
481 * if_data to point at the data.
482 * If we allocate a buffer for the data, make
483 * sure that its size is a multiple of 4 and
484 * record the real size in i_real_bytes.
485 */
486STATIC int
487xfs_iformat_local(
488 xfs_inode_t *ip,
489 xfs_dinode_t *dip,
490 int whichfork,
491 int size)
492{
493 xfs_ifork_t *ifp;
494 int real_size;
495
496 /*
497 * If the size is unreasonable, then something
498 * is wrong and we just bail out rather than crash in
499 * kmem_alloc() or memcpy() below.
500 */
501 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100502 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
503 "corrupt inode %Lu "
504 "(bad size %d for local fork, size = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 (unsigned long long) ip->i_ino, size,
506 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
507 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
508 ip->i_mount, dip);
509 return XFS_ERROR(EFSCORRUPTED);
510 }
511 ifp = XFS_IFORK_PTR(ip, whichfork);
512 real_size = 0;
513 if (size == 0)
514 ifp->if_u1.if_data = NULL;
515 else if (size <= sizeof(ifp->if_u2.if_inline_data))
516 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
517 else {
518 real_size = roundup(size, 4);
519 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
520 }
521 ifp->if_bytes = size;
522 ifp->if_real_bytes = real_size;
523 if (size)
524 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
525 ifp->if_flags &= ~XFS_IFEXTENTS;
526 ifp->if_flags |= XFS_IFINLINE;
527 return 0;
528}
529
530/*
531 * The file consists of a set of extents all
532 * of which fit into the on-disk inode.
533 * If there are few enough extents to fit into
534 * the if_inline_ext, then copy them there.
535 * Otherwise allocate a buffer for them and copy
536 * them into it. Either way, set if_extents
537 * to point at the extents.
538 */
539STATIC int
540xfs_iformat_extents(
541 xfs_inode_t *ip,
542 xfs_dinode_t *dip,
543 int whichfork)
544{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000545 xfs_bmbt_rec_t *dp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 xfs_ifork_t *ifp;
547 int nex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 int size;
549 int i;
550
551 ifp = XFS_IFORK_PTR(ip, whichfork);
552 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
553 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
554
555 /*
556 * If the number of extents is unreasonable, then something
557 * is wrong and we just bail out rather than crash in
558 * kmem_alloc() or memcpy() below.
559 */
560 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100561 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
562 "corrupt inode %Lu ((a)extents = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 (unsigned long long) ip->i_ino, nex);
564 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
565 ip->i_mount, dip);
566 return XFS_ERROR(EFSCORRUPTED);
567 }
568
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100569 ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 if (nex == 0)
571 ifp->if_u1.if_extents = NULL;
572 else if (nex <= XFS_INLINE_EXTS)
573 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100574 else
575 xfs_iext_add(ifp, 0, nex);
576
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 ifp->if_bytes = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 if (size) {
579 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000580 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100581 for (i = 0; i < nex; i++, dp++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000582 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Christoph Hellwigcd8b0a92007-08-16 16:24:15 +1000583 ep->l0 = be64_to_cpu(get_unaligned(&dp->l0));
584 ep->l1 = be64_to_cpu(get_unaligned(&dp->l1));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 }
Eric Sandeen3a59c942007-07-11 11:09:47 +1000586 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 if (whichfork != XFS_DATA_FORK ||
588 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
589 if (unlikely(xfs_check_nostate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100590 ifp, 0, nex))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
592 XFS_ERRLEVEL_LOW,
593 ip->i_mount);
594 return XFS_ERROR(EFSCORRUPTED);
595 }
596 }
597 ifp->if_flags |= XFS_IFEXTENTS;
598 return 0;
599}
600
601/*
602 * The file has too many extents to fit into
603 * the inode, so they are in B-tree format.
604 * Allocate a buffer for the root of the B-tree
605 * and copy the root into it. The i_extents
606 * field will remain NULL until all of the
607 * extents are read in (when they are needed).
608 */
609STATIC int
610xfs_iformat_btree(
611 xfs_inode_t *ip,
612 xfs_dinode_t *dip,
613 int whichfork)
614{
615 xfs_bmdr_block_t *dfp;
616 xfs_ifork_t *ifp;
617 /* REFERENCED */
618 int nrecs;
619 int size;
620
621 ifp = XFS_IFORK_PTR(ip, whichfork);
622 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
623 size = XFS_BMAP_BROOT_SPACE(dfp);
624 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
625
626 /*
627 * blow out if -- fork has less extents than can fit in
628 * fork (fork shouldn't be a btree format), root btree
629 * block has more records than can fit into the fork,
630 * or the number of extents is greater than the number of
631 * blocks.
632 */
633 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
634 || XFS_BMDR_SPACE_CALC(nrecs) >
635 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
636 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100637 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
638 "corrupt inode %Lu (btree).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 (unsigned long long) ip->i_ino);
640 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
641 ip->i_mount);
642 return XFS_ERROR(EFSCORRUPTED);
643 }
644
645 ifp->if_broot_bytes = size;
646 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
647 ASSERT(ifp->if_broot != NULL);
648 /*
649 * Copy and convert from the on-disk structure
650 * to the in-memory structure.
651 */
652 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
653 ifp->if_broot, size);
654 ifp->if_flags &= ~XFS_IFEXTENTS;
655 ifp->if_flags |= XFS_IFBROOT;
656
657 return 0;
658}
659
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660void
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000661xfs_dinode_from_disk(
662 xfs_icdinode_t *to,
663 xfs_dinode_core_t *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000665 to->di_magic = be16_to_cpu(from->di_magic);
666 to->di_mode = be16_to_cpu(from->di_mode);
667 to->di_version = from ->di_version;
668 to->di_format = from->di_format;
669 to->di_onlink = be16_to_cpu(from->di_onlink);
670 to->di_uid = be32_to_cpu(from->di_uid);
671 to->di_gid = be32_to_cpu(from->di_gid);
672 to->di_nlink = be32_to_cpu(from->di_nlink);
673 to->di_projid = be16_to_cpu(from->di_projid);
674 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
675 to->di_flushiter = be16_to_cpu(from->di_flushiter);
676 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
677 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
678 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
679 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
680 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
681 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
682 to->di_size = be64_to_cpu(from->di_size);
683 to->di_nblocks = be64_to_cpu(from->di_nblocks);
684 to->di_extsize = be32_to_cpu(from->di_extsize);
685 to->di_nextents = be32_to_cpu(from->di_nextents);
686 to->di_anextents = be16_to_cpu(from->di_anextents);
687 to->di_forkoff = from->di_forkoff;
688 to->di_aformat = from->di_aformat;
689 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
690 to->di_dmstate = be16_to_cpu(from->di_dmstate);
691 to->di_flags = be16_to_cpu(from->di_flags);
692 to->di_gen = be32_to_cpu(from->di_gen);
693}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000695void
696xfs_dinode_to_disk(
697 xfs_dinode_core_t *to,
698 xfs_icdinode_t *from)
699{
700 to->di_magic = cpu_to_be16(from->di_magic);
701 to->di_mode = cpu_to_be16(from->di_mode);
702 to->di_version = from ->di_version;
703 to->di_format = from->di_format;
704 to->di_onlink = cpu_to_be16(from->di_onlink);
705 to->di_uid = cpu_to_be32(from->di_uid);
706 to->di_gid = cpu_to_be32(from->di_gid);
707 to->di_nlink = cpu_to_be32(from->di_nlink);
708 to->di_projid = cpu_to_be16(from->di_projid);
709 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
710 to->di_flushiter = cpu_to_be16(from->di_flushiter);
711 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
712 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
713 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
714 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
715 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
716 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
717 to->di_size = cpu_to_be64(from->di_size);
718 to->di_nblocks = cpu_to_be64(from->di_nblocks);
719 to->di_extsize = cpu_to_be32(from->di_extsize);
720 to->di_nextents = cpu_to_be32(from->di_nextents);
721 to->di_anextents = cpu_to_be16(from->di_anextents);
722 to->di_forkoff = from->di_forkoff;
723 to->di_aformat = from->di_aformat;
724 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
725 to->di_dmstate = cpu_to_be16(from->di_dmstate);
726 to->di_flags = cpu_to_be16(from->di_flags);
727 to->di_gen = cpu_to_be32(from->di_gen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728}
729
730STATIC uint
731_xfs_dic2xflags(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 __uint16_t di_flags)
733{
734 uint flags = 0;
735
736 if (di_flags & XFS_DIFLAG_ANY) {
737 if (di_flags & XFS_DIFLAG_REALTIME)
738 flags |= XFS_XFLAG_REALTIME;
739 if (di_flags & XFS_DIFLAG_PREALLOC)
740 flags |= XFS_XFLAG_PREALLOC;
741 if (di_flags & XFS_DIFLAG_IMMUTABLE)
742 flags |= XFS_XFLAG_IMMUTABLE;
743 if (di_flags & XFS_DIFLAG_APPEND)
744 flags |= XFS_XFLAG_APPEND;
745 if (di_flags & XFS_DIFLAG_SYNC)
746 flags |= XFS_XFLAG_SYNC;
747 if (di_flags & XFS_DIFLAG_NOATIME)
748 flags |= XFS_XFLAG_NOATIME;
749 if (di_flags & XFS_DIFLAG_NODUMP)
750 flags |= XFS_XFLAG_NODUMP;
751 if (di_flags & XFS_DIFLAG_RTINHERIT)
752 flags |= XFS_XFLAG_RTINHERIT;
753 if (di_flags & XFS_DIFLAG_PROJINHERIT)
754 flags |= XFS_XFLAG_PROJINHERIT;
755 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
756 flags |= XFS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100757 if (di_flags & XFS_DIFLAG_EXTSIZE)
758 flags |= XFS_XFLAG_EXTSIZE;
759 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
760 flags |= XFS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000761 if (di_flags & XFS_DIFLAG_NODEFRAG)
762 flags |= XFS_XFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000763 if (di_flags & XFS_DIFLAG_FILESTREAM)
764 flags |= XFS_XFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 }
766
767 return flags;
768}
769
770uint
771xfs_ip2xflags(
772 xfs_inode_t *ip)
773{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000774 xfs_icdinode_t *dic = &ip->i_d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Nathan Scotta916e2b2006-06-09 17:12:17 +1000776 return _xfs_dic2xflags(dic->di_flags) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100777 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
780uint
781xfs_dic2xflags(
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100782 xfs_dinode_t *dip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100784 xfs_dinode_core_t *dic = &dip->di_core;
785
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000786 return _xfs_dic2xflags(be16_to_cpu(dic->di_flags)) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100787 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
789
790/*
791 * Given a mount structure and an inode number, return a pointer
Nathan Scottc41564b2006-03-29 08:55:14 +1000792 * to a newly allocated in-core inode corresponding to the given
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 * inode number.
794 *
795 * Initialize the inode's attributes and extent pointers if it
796 * already has them (it will not if the inode has no links).
797 */
798int
799xfs_iread(
800 xfs_mount_t *mp,
801 xfs_trans_t *tp,
802 xfs_ino_t ino,
803 xfs_inode_t **ipp,
Nathan Scott745b1f472006-09-28 11:02:23 +1000804 xfs_daddr_t bno,
805 uint imap_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806{
807 xfs_buf_t *bp;
808 xfs_dinode_t *dip;
809 xfs_inode_t *ip;
810 int error;
811
812 ASSERT(xfs_inode_zone != NULL);
813
814 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
815 ip->i_ino = ino;
816 ip->i_mount = mp;
Christoph Hellwigb677c212007-08-29 11:46:28 +1000817 atomic_set(&ip->i_iocount, 0);
David Chinnerf273ab82006-09-28 11:06:03 +1000818 spin_lock_init(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
820 /*
821 * Get pointer's to the on-disk inode and the buffer containing it.
822 * If the inode number refers to a block outside the file system
823 * then xfs_itobp() will return NULL. In this case we should
824 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
825 * know that this is a new incore inode.
826 */
David Chinnera3f74ff2008-03-06 13:43:42 +1100827 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags, XFS_BUF_LOCK);
Nathan Scottb12dd342006-03-17 17:26:04 +1100828 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 kmem_zone_free(xfs_inode_zone, ip);
830 return error;
831 }
832
833 /*
834 * Initialize inode's trace buffers.
835 * Do this before xfs_iformat in case it adds entries.
836 */
Lachlan McIlroycf441ee2008-02-07 16:42:19 +1100837#ifdef XFS_INODE_TRACE
838 ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_SLEEP);
Christoph Hellwig1543d792007-08-29 11:46:47 +1000839#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840#ifdef XFS_BMAP_TRACE
841 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
842#endif
843#ifdef XFS_BMBT_TRACE
844 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
845#endif
846#ifdef XFS_RW_TRACE
847 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
848#endif
849#ifdef XFS_ILOCK_TRACE
850 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
851#endif
852#ifdef XFS_DIR2_TRACE
853 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
854#endif
855
856 /*
857 * If we got something that isn't an inode it means someone
858 * (nfs or dmi) has a stale handle.
859 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000860 if (be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 kmem_zone_free(xfs_inode_zone, ip);
862 xfs_trans_brelse(tp, bp);
863#ifdef DEBUG
864 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
865 "dip->di_core.di_magic (0x%x) != "
866 "XFS_DINODE_MAGIC (0x%x)",
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000867 be16_to_cpu(dip->di_core.di_magic),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 XFS_DINODE_MAGIC);
869#endif /* DEBUG */
870 return XFS_ERROR(EINVAL);
871 }
872
873 /*
874 * If the on-disk inode is already linked to a directory
875 * entry, copy all of the inode into the in-core inode.
876 * xfs_iformat() handles copying in the inode format
877 * specific information.
878 * Otherwise, just get the truly permanent information.
879 */
880 if (dip->di_core.di_mode) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000881 xfs_dinode_from_disk(&ip->i_d, &dip->di_core);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 error = xfs_iformat(ip, dip);
883 if (error) {
884 kmem_zone_free(xfs_inode_zone, ip);
885 xfs_trans_brelse(tp, bp);
886#ifdef DEBUG
887 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
888 "xfs_iformat() returned error %d",
889 error);
890#endif /* DEBUG */
891 return error;
892 }
893 } else {
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000894 ip->i_d.di_magic = be16_to_cpu(dip->di_core.di_magic);
895 ip->i_d.di_version = dip->di_core.di_version;
896 ip->i_d.di_gen = be32_to_cpu(dip->di_core.di_gen);
897 ip->i_d.di_flushiter = be16_to_cpu(dip->di_core.di_flushiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 /*
899 * Make sure to pull in the mode here as well in
900 * case the inode is released without being used.
901 * This ensures that xfs_inactive() will see that
902 * the inode is already free and not try to mess
903 * with the uninitialized part of it.
904 */
905 ip->i_d.di_mode = 0;
906 /*
907 * Initialize the per-fork minima and maxima for a new
908 * inode here. xfs_iformat will do it for old inodes.
909 */
910 ip->i_df.if_ext_max =
911 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
912 }
913
914 INIT_LIST_HEAD(&ip->i_reclaim);
915
916 /*
917 * The inode format changed when we moved the link count and
918 * made it 32 bits long. If this is an old format inode,
919 * convert it in memory to look like a new one. If it gets
920 * flushed to disk we will convert back before flushing or
921 * logging it. We zero out the new projid field and the old link
922 * count field. We'll handle clearing the pad field (the remains
923 * of the old uuid field) when we actually convert the inode to
924 * the new format. We don't change the version number so that we
925 * can distinguish this from a real new format inode.
926 */
927 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
928 ip->i_d.di_nlink = ip->i_d.di_onlink;
929 ip->i_d.di_onlink = 0;
930 ip->i_d.di_projid = 0;
931 }
932
933 ip->i_delayed_blks = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000934 ip->i_size = ip->i_d.di_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
936 /*
937 * Mark the buffer containing the inode as something to keep
938 * around for a while. This helps to keep recently accessed
939 * meta-data in-core longer.
940 */
941 XFS_BUF_SET_REF(bp, XFS_INO_REF);
942
943 /*
944 * Use xfs_trans_brelse() to release the buffer containing the
945 * on-disk inode, because it was acquired with xfs_trans_read_buf()
946 * in xfs_itobp() above. If tp is NULL, this is just a normal
947 * brelse(). If we're within a transaction, then xfs_trans_brelse()
948 * will only release the buffer if it is not dirty within the
949 * transaction. It will be OK to release the buffer in this case,
950 * because inodes on disk are never destroyed and we will be
951 * locking the new in-core inode before putting it in the hash
952 * table where other processes can find it. Thus we don't have
953 * to worry about the inode being changed just because we released
954 * the buffer.
955 */
956 xfs_trans_brelse(tp, bp);
957 *ipp = ip;
958 return 0;
959}
960
961/*
962 * Read in extents from a btree-format inode.
963 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
964 */
965int
966xfs_iread_extents(
967 xfs_trans_t *tp,
968 xfs_inode_t *ip,
969 int whichfork)
970{
971 int error;
972 xfs_ifork_t *ifp;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100973 xfs_extnum_t nextents;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 size_t size;
975
976 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
977 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
978 ip->i_mount);
979 return XFS_ERROR(EFSCORRUPTED);
980 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100981 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
982 size = nextents * sizeof(xfs_bmbt_rec_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 ifp = XFS_IFORK_PTR(ip, whichfork);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 /*
986 * We know that the size is valid (it's checked in iformat_btree)
987 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 ifp->if_lastex = NULLEXTNUM;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100989 ifp->if_bytes = ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 ifp->if_flags |= XFS_IFEXTENTS;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100991 xfs_iext_add(ifp, 0, nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 error = xfs_bmap_read_extents(tp, ip, whichfork);
993 if (error) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100994 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 ifp->if_flags &= ~XFS_IFEXTENTS;
996 return error;
997 }
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000998 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 return 0;
1000}
1001
1002/*
1003 * Allocate an inode on disk and return a copy of its in-core version.
1004 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1005 * appropriately within the inode. The uid and gid for the inode are
1006 * set according to the contents of the given cred structure.
1007 *
1008 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1009 * has a free inode available, call xfs_iget()
1010 * to obtain the in-core version of the allocated inode. Finally,
1011 * fill in the inode and log its initial contents. In this case,
1012 * ialloc_context would be set to NULL and call_again set to false.
1013 *
1014 * If xfs_dialloc() does not have an available inode,
1015 * it will replenish its supply by doing an allocation. Since we can
1016 * only do one allocation within a transaction without deadlocks, we
1017 * must commit the current transaction before returning the inode itself.
1018 * In this case, therefore, we will set call_again to true and return.
1019 * The caller should then commit the current transaction, start a new
1020 * transaction, and call xfs_ialloc() again to actually get the inode.
1021 *
1022 * To ensure that some other process does not grab the inode that
1023 * was allocated during the first call to xfs_ialloc(), this routine
1024 * also returns the [locked] bp pointing to the head of the freelist
1025 * as ialloc_context. The caller should hold this buffer across
1026 * the commit and pass it back into this routine on the second call.
David Chinnerb11f94d2007-07-11 11:09:33 +10001027 *
1028 * If we are allocating quota inodes, we do not have a parent inode
1029 * to attach to or associate with (i.e. pip == NULL) because they
1030 * are not linked into the directory structure - they are attached
1031 * directly to the superblock - and so have no parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 */
1033int
1034xfs_ialloc(
1035 xfs_trans_t *tp,
1036 xfs_inode_t *pip,
1037 mode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -07001038 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 xfs_dev_t rdev,
1040 cred_t *cr,
1041 xfs_prid_t prid,
1042 int okalloc,
1043 xfs_buf_t **ialloc_context,
1044 boolean_t *call_again,
1045 xfs_inode_t **ipp)
1046{
1047 xfs_ino_t ino;
1048 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001049 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 uint flags;
1051 int error;
1052
1053 /*
1054 * Call the space management code to pick
1055 * the on-disk inode to be allocated.
1056 */
David Chinnerb11f94d2007-07-11 11:09:33 +10001057 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 ialloc_context, call_again, &ino);
1059 if (error != 0) {
1060 return error;
1061 }
1062 if (*call_again || ino == NULLFSINO) {
1063 *ipp = NULL;
1064 return 0;
1065 }
1066 ASSERT(*ialloc_context == NULL);
1067
1068 /*
1069 * Get the in-core inode with the lock held exclusively.
1070 * This is because we're setting fields here we need
1071 * to prevent others from looking at until we're done.
1072 */
1073 error = xfs_trans_iget(tp->t_mountp, tp, ino,
Nathan Scott745b1f472006-09-28 11:02:23 +10001074 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 if (error != 0) {
1076 return error;
1077 }
1078 ASSERT(ip != NULL);
1079
1080 vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 ip->i_d.di_mode = (__uint16_t)mode;
1082 ip->i_d.di_onlink = 0;
1083 ip->i_d.di_nlink = nlink;
1084 ASSERT(ip->i_d.di_nlink == nlink);
1085 ip->i_d.di_uid = current_fsuid(cr);
1086 ip->i_d.di_gid = current_fsgid(cr);
1087 ip->i_d.di_projid = prid;
1088 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1089
1090 /*
1091 * If the superblock version is up to where we support new format
1092 * inodes and this is currently an old format inode, then change
1093 * the inode version number now. This way we only do the conversion
1094 * here rather than here and in the flush/logging code.
1095 */
Eric Sandeen62118702008-03-06 13:44:28 +11001096 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1098 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1099 /*
1100 * We've already zeroed the old link count, the projid field,
1101 * and the pad field.
1102 */
1103 }
1104
1105 /*
1106 * Project ids won't be stored on disk if we are using a version 1 inode.
1107 */
David Chinner2a82b8b2007-07-11 11:09:12 +10001108 if ((prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 xfs_bump_ino_vers2(tp, ip);
1110
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001111 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 ip->i_d.di_gid = pip->i_d.di_gid;
1113 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1114 ip->i_d.di_mode |= S_ISGID;
1115 }
1116 }
1117
1118 /*
1119 * If the group ID of the new file does not match the effective group
1120 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1121 * (and only if the irix_sgid_inherit compatibility variable is set).
1122 */
1123 if ((irix_sgid_inherit) &&
1124 (ip->i_d.di_mode & S_ISGID) &&
1125 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1126 ip->i_d.di_mode &= ~S_ISGID;
1127 }
1128
1129 ip->i_d.di_size = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001130 ip->i_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 ip->i_d.di_nextents = 0;
1132 ASSERT(ip->i_d.di_nblocks == 0);
1133 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1134 /*
1135 * di_gen will have been taken care of in xfs_iread.
1136 */
1137 ip->i_d.di_extsize = 0;
1138 ip->i_d.di_dmevmask = 0;
1139 ip->i_d.di_dmstate = 0;
1140 ip->i_d.di_flags = 0;
1141 flags = XFS_ILOG_CORE;
1142 switch (mode & S_IFMT) {
1143 case S_IFIFO:
1144 case S_IFCHR:
1145 case S_IFBLK:
1146 case S_IFSOCK:
1147 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1148 ip->i_df.if_u2.if_rdev = rdev;
1149 ip->i_df.if_flags = 0;
1150 flags |= XFS_ILOG_DEV;
1151 break;
1152 case S_IFREG:
David Chinnerb11f94d2007-07-11 11:09:33 +10001153 if (pip && xfs_inode_is_filestream(pip)) {
David Chinner2a82b8b2007-07-11 11:09:12 +10001154 error = xfs_filestream_associate(pip, ip);
1155 if (error < 0)
1156 return -error;
1157 if (!error)
1158 xfs_iflags_set(ip, XFS_IFILESTREAM);
1159 }
1160 /* fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +10001162 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Nathan Scott365ca832005-06-21 15:39:12 +10001163 uint di_flags = 0;
1164
1165 if ((mode & S_IFMT) == S_IFDIR) {
1166 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1167 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001168 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1169 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1170 ip->i_d.di_extsize = pip->i_d.di_extsize;
1171 }
1172 } else if ((mode & S_IFMT) == S_IFREG) {
Christoph Hellwig613d7042007-10-11 17:44:08 +10001173 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +10001174 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001175 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1176 di_flags |= XFS_DIFLAG_EXTSIZE;
1177 ip->i_d.di_extsize = pip->i_d.di_extsize;
1178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 }
1180 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1181 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +10001182 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1184 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +10001185 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1187 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +10001188 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1190 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +10001191 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1192 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1193 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +10001194 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1195 xfs_inherit_nodefrag)
1196 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +10001197 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1198 di_flags |= XFS_DIFLAG_FILESTREAM;
Nathan Scott365ca832005-06-21 15:39:12 +10001199 ip->i_d.di_flags |= di_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 }
1201 /* FALLTHROUGH */
1202 case S_IFLNK:
1203 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1204 ip->i_df.if_flags = XFS_IFEXTENTS;
1205 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1206 ip->i_df.if_u1.if_extents = NULL;
1207 break;
1208 default:
1209 ASSERT(0);
1210 }
1211 /*
1212 * Attribute fork settings for new inode.
1213 */
1214 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1215 ip->i_d.di_anextents = 0;
1216
1217 /*
1218 * Log the new values stuffed into the inode.
1219 */
1220 xfs_trans_log_inode(tp, ip, flags);
1221
Nathan Scottb83bd132006-06-09 16:48:30 +10001222 /* now that we have an i_mode we can setup inode ops and unlock */
Christoph Hellwig745f6912007-08-30 17:20:39 +10001223 xfs_initialize_vnode(tp->t_mountp, vp, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
1225 *ipp = ip;
1226 return 0;
1227}
1228
1229/*
1230 * Check to make sure that there are no blocks allocated to the
1231 * file beyond the size of the file. We don't check this for
1232 * files with fixed size extents or real time extents, but we
1233 * at least do it for regular files.
1234 */
1235#ifdef DEBUG
1236void
1237xfs_isize_check(
1238 xfs_mount_t *mp,
1239 xfs_inode_t *ip,
1240 xfs_fsize_t isize)
1241{
1242 xfs_fileoff_t map_first;
1243 int nimaps;
1244 xfs_bmbt_irec_t imaps[2];
1245
1246 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1247 return;
1248
Eric Sandeen71ddabb2007-11-23 16:29:42 +11001249 if (XFS_IS_REALTIME_INODE(ip))
1250 return;
1251
1252 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 return;
1254
1255 nimaps = 2;
1256 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1257 /*
1258 * The filesystem could be shutting down, so bmapi may return
1259 * an error.
1260 */
1261 if (xfs_bmapi(NULL, ip, map_first,
1262 (XFS_B_TO_FSB(mp,
1263 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1264 map_first),
1265 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001266 NULL, NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 return;
1268 ASSERT(nimaps == 1);
1269 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1270}
1271#endif /* DEBUG */
1272
1273/*
1274 * Calculate the last possible buffered byte in a file. This must
1275 * include data that was buffered beyond the EOF by the write code.
1276 * This also needs to deal with overflowing the xfs_fsize_t type
1277 * which can happen for sizes near the limit.
1278 *
1279 * We also need to take into account any blocks beyond the EOF. It
1280 * may be the case that they were buffered by a write which failed.
1281 * In that case the pages will still be in memory, but the inode size
1282 * will never have been updated.
1283 */
1284xfs_fsize_t
1285xfs_file_last_byte(
1286 xfs_inode_t *ip)
1287{
1288 xfs_mount_t *mp;
1289 xfs_fsize_t last_byte;
1290 xfs_fileoff_t last_block;
1291 xfs_fileoff_t size_last_block;
1292 int error;
1293
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001294 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296 mp = ip->i_mount;
1297 /*
1298 * Only check for blocks beyond the EOF if the extents have
1299 * been read in. This eliminates the need for the inode lock,
1300 * and it also saves us from looking when it really isn't
1301 * necessary.
1302 */
1303 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1304 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1305 XFS_DATA_FORK);
1306 if (error) {
1307 last_block = 0;
1308 }
1309 } else {
1310 last_block = 0;
1311 }
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001312 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1314
1315 last_byte = XFS_FSB_TO_B(mp, last_block);
1316 if (last_byte < 0) {
1317 return XFS_MAXIOFFSET(mp);
1318 }
1319 last_byte += (1 << mp->m_writeio_log);
1320 if (last_byte < 0) {
1321 return XFS_MAXIOFFSET(mp);
1322 }
1323 return last_byte;
1324}
1325
1326#if defined(XFS_RW_TRACE)
1327STATIC void
1328xfs_itrunc_trace(
1329 int tag,
1330 xfs_inode_t *ip,
1331 int flag,
1332 xfs_fsize_t new_size,
1333 xfs_off_t toss_start,
1334 xfs_off_t toss_finish)
1335{
1336 if (ip->i_rwtrace == NULL) {
1337 return;
1338 }
1339
1340 ktrace_enter(ip->i_rwtrace,
1341 (void*)((long)tag),
1342 (void*)ip,
1343 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1344 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1345 (void*)((long)flag),
1346 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1347 (void*)(unsigned long)(new_size & 0xffffffff),
1348 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1349 (void*)(unsigned long)(toss_start & 0xffffffff),
1350 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1351 (void*)(unsigned long)(toss_finish & 0xffffffff),
1352 (void*)(unsigned long)current_cpu(),
Yingping Luf1fdc842006-03-22 12:44:15 +11001353 (void*)(unsigned long)current_pid(),
1354 (void*)NULL,
1355 (void*)NULL,
1356 (void*)NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357}
1358#else
1359#define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1360#endif
1361
1362/*
1363 * Start the truncation of the file to new_size. The new size
1364 * must be smaller than the current size. This routine will
1365 * clear the buffer and page caches of file data in the removed
1366 * range, and xfs_itruncate_finish() will remove the underlying
1367 * disk blocks.
1368 *
1369 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1370 * must NOT have the inode lock held at all. This is because we're
1371 * calling into the buffer/page cache code and we can't hold the
1372 * inode lock when we do so.
1373 *
David Chinner38e22992006-03-22 12:47:15 +11001374 * We need to wait for any direct I/Os in flight to complete before we
1375 * proceed with the truncate. This is needed to prevent the extents
1376 * being read or written by the direct I/Os from being removed while the
1377 * I/O is in flight as there is no other method of synchronising
1378 * direct I/O with the truncate operation. Also, because we hold
1379 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1380 * started until the truncate completes and drops the lock. Essentially,
1381 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1382 * between direct I/Os and the truncate operation.
1383 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1385 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1386 * in the case that the caller is locking things out of order and
1387 * may not be able to call xfs_itruncate_finish() with the inode lock
1388 * held without dropping the I/O lock. If the caller must drop the
1389 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1390 * must be called again with all the same restrictions as the initial
1391 * call.
1392 */
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001393int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394xfs_itruncate_start(
1395 xfs_inode_t *ip,
1396 uint flags,
1397 xfs_fsize_t new_size)
1398{
1399 xfs_fsize_t last_byte;
1400 xfs_off_t toss_start;
1401 xfs_mount_t *mp;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001402 bhv_vnode_t *vp;
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001403 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001405 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001406 ASSERT((new_size == 0) || (new_size <= ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1408 (flags == XFS_ITRUNC_MAYBE));
1409
1410 mp = ip->i_mount;
1411 vp = XFS_ITOV(ip);
Yingping Lu9fa80462006-03-22 12:44:35 +11001412
Lachlan McIlroyc734c792007-12-18 16:17:41 +11001413 /* wait for the completion of any pending DIOs */
1414 if (new_size < ip->i_size)
1415 vn_iowait(ip);
1416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 /*
Nathan Scott67fcaa72006-06-09 17:00:52 +10001418 * Call toss_pages or flushinval_pages to get rid of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 * overlapping the region being removed. We have to use
Nathan Scott67fcaa72006-06-09 17:00:52 +10001420 * the less efficient flushinval_pages in the case that the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 * caller may not be able to finish the truncate without
1422 * dropping the inode's I/O lock. Make sure
1423 * to catch any pages brought in by buffers overlapping
1424 * the EOF by searching out beyond the isize by our
1425 * block size. We round new_size up to a block boundary
1426 * so that we don't toss things on the same block as
1427 * new_size but before it.
1428 *
Nathan Scott67fcaa72006-06-09 17:00:52 +10001429 * Before calling toss_page or flushinval_pages, make sure to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 * call remapf() over the same region if the file is mapped.
1431 * This frees up mapped file references to the pages in the
Nathan Scott67fcaa72006-06-09 17:00:52 +10001432 * given range and for the flushinval_pages case it ensures
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 * that we get the latest mapped changes flushed out.
1434 */
1435 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1436 toss_start = XFS_FSB_TO_B(mp, toss_start);
1437 if (toss_start < 0) {
1438 /*
1439 * The place to start tossing is beyond our maximum
1440 * file size, so there is no way that the data extended
1441 * out there.
1442 */
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001443 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 }
1445 last_byte = xfs_file_last_byte(ip);
1446 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1447 last_byte);
1448 if (last_byte > toss_start) {
1449 if (flags & XFS_ITRUNC_DEFINITE) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001450 xfs_tosspages(ip, toss_start,
1451 -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 } else {
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001453 error = xfs_flushinval_pages(ip, toss_start,
1454 -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 }
1456 }
1457
1458#ifdef DEBUG
1459 if (new_size == 0) {
1460 ASSERT(VN_CACHED(vp) == 0);
1461 }
1462#endif
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001463 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464}
1465
1466/*
David Chinnerf6485052008-04-17 16:50:04 +10001467 * Shrink the file to the given new_size. The new size must be smaller than
1468 * the current size. This will free up the underlying blocks in the removed
1469 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 *
David Chinnerf6485052008-04-17 16:50:04 +10001471 * The transaction passed to this routine must have made a permanent log
1472 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1473 * given transaction and start new ones, so make sure everything involved in
1474 * the transaction is tidy before calling here. Some transaction will be
1475 * returned to the caller to be committed. The incoming transaction must
1476 * already include the inode, and both inode locks must be held exclusively.
1477 * The inode must also be "held" within the transaction. On return the inode
1478 * will be "held" within the returned transaction. This routine does NOT
1479 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 *
David Chinnerf6485052008-04-17 16:50:04 +10001481 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1482 * indicates the fork which is to be truncated. For the attribute fork we only
1483 * support truncation to size 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 *
David Chinnerf6485052008-04-17 16:50:04 +10001485 * We use the sync parameter to indicate whether or not the first transaction
1486 * we perform might have to be synchronous. For the attr fork, it needs to be
1487 * so if the unlink of the inode is not yet known to be permanent in the log.
1488 * This keeps us from freeing and reusing the blocks of the attribute fork
1489 * before the unlink of the inode becomes permanent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 *
David Chinnerf6485052008-04-17 16:50:04 +10001491 * For the data fork, we normally have to run synchronously if we're being
1492 * called out of the inactive path or we're being called out of the create path
1493 * where we're truncating an existing file. Either way, the truncate needs to
1494 * be sync so blocks don't reappear in the file with altered data in case of a
1495 * crash. wsync filesystems can run the first case async because anything that
1496 * shrinks the inode has to run sync so by the time we're called here from
1497 * inactive, the inode size is permanently set to 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 *
David Chinnerf6485052008-04-17 16:50:04 +10001499 * Calls from the truncate path always need to be sync unless we're in a wsync
1500 * filesystem and the file has already been unlinked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 *
David Chinnerf6485052008-04-17 16:50:04 +10001502 * The caller is responsible for correctly setting the sync parameter. It gets
1503 * too hard for us to guess here which path we're being called out of just
1504 * based on inode state.
1505 *
1506 * If we get an error, we must return with the inode locked and linked into the
1507 * current transaction. This keeps things simple for the higher level code,
1508 * because it always knows that the inode is locked and held in the transaction
1509 * that returns to it whether errors occur or not. We don't mark the inode
1510 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 */
1512int
1513xfs_itruncate_finish(
1514 xfs_trans_t **tp,
1515 xfs_inode_t *ip,
1516 xfs_fsize_t new_size,
1517 int fork,
1518 int sync)
1519{
1520 xfs_fsblock_t first_block;
1521 xfs_fileoff_t first_unmap_block;
1522 xfs_fileoff_t last_block;
1523 xfs_filblks_t unmap_len=0;
1524 xfs_mount_t *mp;
1525 xfs_trans_t *ntp;
1526 int done;
1527 int committed;
1528 xfs_bmap_free_t free_list;
1529 int error;
1530
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001531 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001532 ASSERT((new_size == 0) || (new_size <= ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 ASSERT(*tp != NULL);
1534 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1535 ASSERT(ip->i_transp == *tp);
1536 ASSERT(ip->i_itemp != NULL);
1537 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1538
1539
1540 ntp = *tp;
1541 mp = (ntp)->t_mountp;
1542 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1543
1544 /*
1545 * We only support truncating the entire attribute fork.
1546 */
1547 if (fork == XFS_ATTR_FORK) {
1548 new_size = 0LL;
1549 }
1550 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1551 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1552 /*
1553 * The first thing we do is set the size to new_size permanently
1554 * on disk. This way we don't have to worry about anyone ever
1555 * being able to look at the data being freed even in the face
1556 * of a crash. What we're getting around here is the case where
1557 * we free a block, it is allocated to another file, it is written
1558 * to, and then we crash. If the new data gets written to the
1559 * file but the log buffers containing the free and reallocation
1560 * don't, then we'd end up with garbage in the blocks being freed.
1561 * As long as we make the new_size permanent before actually
1562 * freeing any blocks it doesn't matter if they get writtten to.
1563 *
1564 * The callers must signal into us whether or not the size
1565 * setting here must be synchronous. There are a few cases
1566 * where it doesn't have to be synchronous. Those cases
1567 * occur if the file is unlinked and we know the unlink is
1568 * permanent or if the blocks being truncated are guaranteed
1569 * to be beyond the inode eof (regardless of the link count)
1570 * and the eof value is permanent. Both of these cases occur
1571 * only on wsync-mounted filesystems. In those cases, we're
1572 * guaranteed that no user will ever see the data in the blocks
1573 * that are being truncated so the truncate can run async.
1574 * In the free beyond eof case, the file may wind up with
1575 * more blocks allocated to it than it needs if we crash
1576 * and that won't get fixed until the next time the file
1577 * is re-opened and closed but that's ok as that shouldn't
1578 * be too many blocks.
1579 *
1580 * However, we can't just make all wsync xactions run async
1581 * because there's one call out of the create path that needs
1582 * to run sync where it's truncating an existing file to size
1583 * 0 whose size is > 0.
1584 *
1585 * It's probably possible to come up with a test in this
1586 * routine that would correctly distinguish all the above
1587 * cases from the values of the function parameters and the
1588 * inode state but for sanity's sake, I've decided to let the
1589 * layers above just tell us. It's simpler to correctly figure
1590 * out in the layer above exactly under what conditions we
1591 * can run async and I think it's easier for others read and
1592 * follow the logic in case something has to be changed.
1593 * cscope is your friend -- rcc.
1594 *
1595 * The attribute fork is much simpler.
1596 *
1597 * For the attribute fork we allow the caller to tell us whether
1598 * the unlink of the inode that led to this call is yet permanent
1599 * in the on disk log. If it is not and we will be freeing extents
1600 * in this inode then we make the first transaction synchronous
1601 * to make sure that the unlink is permanent by the time we free
1602 * the blocks.
1603 */
1604 if (fork == XFS_DATA_FORK) {
1605 if (ip->i_d.di_nextents > 0) {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001606 /*
1607 * If we are not changing the file size then do
1608 * not update the on-disk file size - we may be
1609 * called from xfs_inactive_free_eofblocks(). If we
1610 * update the on-disk file size and then the system
1611 * crashes before the contents of the file are
1612 * flushed to disk then the files may be full of
1613 * holes (ie NULL files bug).
1614 */
1615 if (ip->i_size != new_size) {
1616 ip->i_d.di_size = new_size;
1617 ip->i_size = new_size;
1618 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 }
1621 } else if (sync) {
1622 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1623 if (ip->i_d.di_anextents > 0)
1624 xfs_trans_set_sync(ntp);
1625 }
1626 ASSERT(fork == XFS_DATA_FORK ||
1627 (fork == XFS_ATTR_FORK &&
1628 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1629 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1630
1631 /*
1632 * Since it is possible for space to become allocated beyond
1633 * the end of the file (in a crash where the space is allocated
1634 * but the inode size is not yet updated), simply remove any
1635 * blocks which show up between the new EOF and the maximum
1636 * possible file size. If the first block to be removed is
1637 * beyond the maximum file size (ie it is the same as last_block),
1638 * then there is nothing to do.
1639 */
1640 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1641 ASSERT(first_unmap_block <= last_block);
1642 done = 0;
1643 if (last_block == first_unmap_block) {
1644 done = 1;
1645 } else {
1646 unmap_len = last_block - first_unmap_block + 1;
1647 }
1648 while (!done) {
1649 /*
1650 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1651 * will tell us whether it freed the entire range or
1652 * not. If this is a synchronous mount (wsync),
1653 * then we can tell bunmapi to keep all the
1654 * transactions asynchronous since the unlink
1655 * transaction that made this inode inactive has
1656 * already hit the disk. There's no danger of
1657 * the freed blocks being reused, there being a
1658 * crash, and the reused blocks suddenly reappearing
1659 * in this file with garbage in them once recovery
1660 * runs.
1661 */
1662 XFS_BMAP_INIT(&free_list, &first_block);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10001663 error = xfs_bunmapi(ntp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001664 first_unmap_block, unmap_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 XFS_BMAPI_AFLAG(fork) |
1666 (sync ? 0 : XFS_BMAPI_ASYNC),
1667 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001668 &first_block, &free_list,
1669 NULL, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 if (error) {
1671 /*
1672 * If the bunmapi call encounters an error,
1673 * return to the caller where the transaction
1674 * can be properly aborted. We just need to
1675 * make sure we're not holding any resources
1676 * that we were not when we came in.
1677 */
1678 xfs_bmap_cancel(&free_list);
1679 return error;
1680 }
1681
1682 /*
1683 * Duplicate the transaction that has the permanent
1684 * reservation and commit the old transaction.
1685 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001686 error = xfs_bmap_finish(tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 ntp = *tp;
David Chinnerf6485052008-04-17 16:50:04 +10001688 if (committed) {
1689 /* link the inode into the next xact in the chain */
1690 xfs_trans_ijoin(ntp, ip,
1691 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1692 xfs_trans_ihold(ntp, ip);
1693 }
1694
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 if (error) {
1696 /*
David Chinnerf6485052008-04-17 16:50:04 +10001697 * If the bmap finish call encounters an error, return
1698 * to the caller where the transaction can be properly
1699 * aborted. We just need to make sure we're not
1700 * holding any resources that we were not when we came
1701 * in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 *
David Chinnerf6485052008-04-17 16:50:04 +10001703 * Aborting from this point might lose some blocks in
1704 * the file system, but oh well.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 */
1706 xfs_bmap_cancel(&free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 return error;
1708 }
1709
1710 if (committed) {
1711 /*
David Chinnerf6485052008-04-17 16:50:04 +10001712 * Mark the inode dirty so it will be logged and
David Chinnere5720ee2008-04-10 12:21:18 +10001713 * moved forward in the log as part of every commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1716 }
David Chinnerf6485052008-04-17 16:50:04 +10001717
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 ntp = xfs_trans_dup(ntp);
David Chinnere5720ee2008-04-10 12:21:18 +10001719 error = xfs_trans_commit(*tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 *tp = ntp;
David Chinnere5720ee2008-04-10 12:21:18 +10001721
David Chinnerf6485052008-04-17 16:50:04 +10001722 /* link the inode into the next transaction in the chain */
1723 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1724 xfs_trans_ihold(ntp, ip);
1725
1726 if (!error)
1727 error = xfs_trans_reserve(ntp, 0,
1728 XFS_ITRUNCATE_LOG_RES(mp), 0,
1729 XFS_TRANS_PERM_LOG_RES,
1730 XFS_ITRUNCATE_LOG_COUNT);
1731 if (error)
1732 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 }
1734 /*
1735 * Only update the size in the case of the data fork, but
1736 * always re-log the inode so that our permanent transaction
1737 * can keep on rolling it forward in the log.
1738 */
1739 if (fork == XFS_DATA_FORK) {
1740 xfs_isize_check(mp, ip, new_size);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001741 /*
1742 * If we are not changing the file size then do
1743 * not update the on-disk file size - we may be
1744 * called from xfs_inactive_free_eofblocks(). If we
1745 * update the on-disk file size and then the system
1746 * crashes before the contents of the file are
1747 * flushed to disk then the files may be full of
1748 * holes (ie NULL files bug).
1749 */
1750 if (ip->i_size != new_size) {
1751 ip->i_d.di_size = new_size;
1752 ip->i_size = new_size;
1753 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 }
1755 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1756 ASSERT((new_size != 0) ||
1757 (fork == XFS_ATTR_FORK) ||
1758 (ip->i_delayed_blks == 0));
1759 ASSERT((new_size != 0) ||
1760 (fork == XFS_ATTR_FORK) ||
1761 (ip->i_d.di_nextents == 0));
1762 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1763 return 0;
1764}
1765
1766
1767/*
1768 * xfs_igrow_start
1769 *
1770 * Do the first part of growing a file: zero any data in the last
1771 * block that is beyond the old EOF. We need to do this before
1772 * the inode is joined to the transaction to modify the i_size.
1773 * That way we can drop the inode lock and call into the buffer
1774 * cache to get the buffer mapping the EOF.
1775 */
1776int
1777xfs_igrow_start(
1778 xfs_inode_t *ip,
1779 xfs_fsize_t new_size,
1780 cred_t *credp)
1781{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001782 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001783 ASSERT(new_size > ip->i_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 /*
1786 * Zero any pages that may have been created by
1787 * xfs_write_file() beyond the end of the file
1788 * and any blocks between the old and new file sizes.
1789 */
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10001790 return xfs_zero_eof(ip, new_size, ip->i_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791}
1792
1793/*
1794 * xfs_igrow_finish
1795 *
1796 * This routine is called to extend the size of a file.
1797 * The inode must have both the iolock and the ilock locked
1798 * for update and it must be a part of the current transaction.
1799 * The xfs_igrow_start() function must have been called previously.
1800 * If the change_flag is not zero, the inode change timestamp will
1801 * be updated.
1802 */
1803void
1804xfs_igrow_finish(
1805 xfs_trans_t *tp,
1806 xfs_inode_t *ip,
1807 xfs_fsize_t new_size,
1808 int change_flag)
1809{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001810 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 ASSERT(ip->i_transp == tp);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001812 ASSERT(new_size > ip->i_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
1814 /*
1815 * Update the file size. Update the inode change timestamp
1816 * if change_flag set.
1817 */
1818 ip->i_d.di_size = new_size;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001819 ip->i_size = new_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 if (change_flag)
1821 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1822 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1823
1824}
1825
1826
1827/*
1828 * This is called when the inode's link count goes to 0.
1829 * We place the on-disk inode on a list in the AGI. It
1830 * will be pulled from this list when the inode is freed.
1831 */
1832int
1833xfs_iunlink(
1834 xfs_trans_t *tp,
1835 xfs_inode_t *ip)
1836{
1837 xfs_mount_t *mp;
1838 xfs_agi_t *agi;
1839 xfs_dinode_t *dip;
1840 xfs_buf_t *agibp;
1841 xfs_buf_t *ibp;
1842 xfs_agnumber_t agno;
1843 xfs_daddr_t agdaddr;
1844 xfs_agino_t agino;
1845 short bucket_index;
1846 int offset;
1847 int error;
1848 int agi_ok;
1849
1850 ASSERT(ip->i_d.di_nlink == 0);
1851 ASSERT(ip->i_d.di_mode != 0);
1852 ASSERT(ip->i_transp == tp);
1853
1854 mp = tp->t_mountp;
1855
1856 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1857 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1858
1859 /*
1860 * Get the agi buffer first. It ensures lock ordering
1861 * on the list.
1862 */
1863 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1864 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001865 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 return error;
Vlad Apostolov859d7182007-10-11 17:44:18 +10001867
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 /*
1869 * Validate the magic number of the agi block.
1870 */
1871 agi = XFS_BUF_TO_AGI(agibp);
1872 agi_ok =
Christoph Hellwig16259e72005-11-02 15:11:25 +11001873 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1874 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1876 XFS_RANDOM_IUNLINK))) {
1877 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1878 xfs_trans_brelse(tp, agibp);
1879 return XFS_ERROR(EFSCORRUPTED);
1880 }
1881 /*
1882 * Get the index into the agi hash table for the
1883 * list this inode will go on.
1884 */
1885 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1886 ASSERT(agino != 0);
1887 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1888 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001889 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Christoph Hellwig16259e72005-11-02 15:11:25 +11001891 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 /*
1893 * There is already another inode in the bucket we need
1894 * to add ourselves to. Add us at the front of the list.
1895 * Here we put the head pointer into our next pointer,
1896 * and then we fall through to point the head at us.
1897 */
David Chinnera3f74ff2008-03-06 13:43:42 +11001898 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
Vlad Apostolovc319b582007-11-23 16:27:51 +11001899 if (error)
1900 return error;
1901
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001902 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 /* both on-disk, don't endian flip twice */
1904 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1905 offset = ip->i_boffset +
1906 offsetof(xfs_dinode_t, di_next_unlinked);
1907 xfs_trans_inode_buf(tp, ibp);
1908 xfs_trans_log_buf(tp, ibp, offset,
1909 (offset + sizeof(xfs_agino_t) - 1));
1910 xfs_inobp_check(mp, ibp);
1911 }
1912
1913 /*
1914 * Point the bucket head pointer at the inode being inserted.
1915 */
1916 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001917 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 offset = offsetof(xfs_agi_t, agi_unlinked) +
1919 (sizeof(xfs_agino_t) * bucket_index);
1920 xfs_trans_log_buf(tp, agibp, offset,
1921 (offset + sizeof(xfs_agino_t) - 1));
1922 return 0;
1923}
1924
1925/*
1926 * Pull the on-disk inode from the AGI unlinked list.
1927 */
1928STATIC int
1929xfs_iunlink_remove(
1930 xfs_trans_t *tp,
1931 xfs_inode_t *ip)
1932{
1933 xfs_ino_t next_ino;
1934 xfs_mount_t *mp;
1935 xfs_agi_t *agi;
1936 xfs_dinode_t *dip;
1937 xfs_buf_t *agibp;
1938 xfs_buf_t *ibp;
1939 xfs_agnumber_t agno;
1940 xfs_daddr_t agdaddr;
1941 xfs_agino_t agino;
1942 xfs_agino_t next_agino;
1943 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001944 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001946 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 int error;
1948 int agi_ok;
1949
1950 /*
1951 * First pull the on-disk inode from the AGI unlinked list.
1952 */
1953 mp = tp->t_mountp;
1954
1955 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1956 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1957
1958 /*
1959 * Get the agi buffer first. It ensures lock ordering
1960 * on the list.
1961 */
1962 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1963 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1964 if (error) {
1965 cmn_err(CE_WARN,
1966 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1967 error, mp->m_fsname);
1968 return error;
1969 }
1970 /*
1971 * Validate the magic number of the agi block.
1972 */
1973 agi = XFS_BUF_TO_AGI(agibp);
1974 agi_ok =
Christoph Hellwig16259e72005-11-02 15:11:25 +11001975 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1976 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1978 XFS_RANDOM_IUNLINK_REMOVE))) {
1979 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1980 mp, agi);
1981 xfs_trans_brelse(tp, agibp);
1982 cmn_err(CE_WARN,
1983 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1984 mp->m_fsname);
1985 return XFS_ERROR(EFSCORRUPTED);
1986 }
1987 /*
1988 * Get the index into the agi hash table for the
1989 * list this inode will go on.
1990 */
1991 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1992 ASSERT(agino != 0);
1993 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001994 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 ASSERT(agi->agi_unlinked[bucket_index]);
1996
Christoph Hellwig16259e72005-11-02 15:11:25 +11001997 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 /*
1999 * We're at the head of the list. Get the inode's
2000 * on-disk buffer to see if there is anyone after us
2001 * on the list. Only modify our next pointer if it
2002 * is not already NULLAGINO. This saves us the overhead
2003 * of dealing with the buffer when there is no need to
2004 * change it.
2005 */
David Chinnera3f74ff2008-03-06 13:43:42 +11002006 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 if (error) {
2008 cmn_err(CE_WARN,
2009 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2010 error, mp->m_fsname);
2011 return error;
2012 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002013 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 ASSERT(next_agino != 0);
2015 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002016 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 offset = ip->i_boffset +
2018 offsetof(xfs_dinode_t, di_next_unlinked);
2019 xfs_trans_inode_buf(tp, ibp);
2020 xfs_trans_log_buf(tp, ibp, offset,
2021 (offset + sizeof(xfs_agino_t) - 1));
2022 xfs_inobp_check(mp, ibp);
2023 } else {
2024 xfs_trans_brelse(tp, ibp);
2025 }
2026 /*
2027 * Point the bucket head pointer at the next inode.
2028 */
2029 ASSERT(next_agino != 0);
2030 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002031 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 offset = offsetof(xfs_agi_t, agi_unlinked) +
2033 (sizeof(xfs_agino_t) * bucket_index);
2034 xfs_trans_log_buf(tp, agibp, offset,
2035 (offset + sizeof(xfs_agino_t) - 1));
2036 } else {
2037 /*
2038 * We need to search the list for the inode being freed.
2039 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002040 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 last_ibp = NULL;
2042 while (next_agino != agino) {
2043 /*
2044 * If the last inode wasn't the one pointing to
2045 * us, then release its buffer since we're not
2046 * going to do anything with it.
2047 */
2048 if (last_ibp != NULL) {
2049 xfs_trans_brelse(tp, last_ibp);
2050 }
2051 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2052 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2053 &last_ibp, &last_offset);
2054 if (error) {
2055 cmn_err(CE_WARN,
2056 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2057 error, mp->m_fsname);
2058 return error;
2059 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002060 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 ASSERT(next_agino != NULLAGINO);
2062 ASSERT(next_agino != 0);
2063 }
2064 /*
2065 * Now last_ibp points to the buffer previous to us on
2066 * the unlinked list. Pull us from the list.
2067 */
David Chinnera3f74ff2008-03-06 13:43:42 +11002068 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 if (error) {
2070 cmn_err(CE_WARN,
2071 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2072 error, mp->m_fsname);
2073 return error;
2074 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002075 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 ASSERT(next_agino != 0);
2077 ASSERT(next_agino != agino);
2078 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002079 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 offset = ip->i_boffset +
2081 offsetof(xfs_dinode_t, di_next_unlinked);
2082 xfs_trans_inode_buf(tp, ibp);
2083 xfs_trans_log_buf(tp, ibp, offset,
2084 (offset + sizeof(xfs_agino_t) - 1));
2085 xfs_inobp_check(mp, ibp);
2086 } else {
2087 xfs_trans_brelse(tp, ibp);
2088 }
2089 /*
2090 * Point the previous inode on the list to the next inode.
2091 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002092 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 ASSERT(next_agino != 0);
2094 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2095 xfs_trans_inode_buf(tp, last_ibp);
2096 xfs_trans_log_buf(tp, last_ibp, offset,
2097 (offset + sizeof(xfs_agino_t) - 1));
2098 xfs_inobp_check(mp, last_ibp);
2099 }
2100 return 0;
2101}
2102
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002103STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104xfs_ifree_cluster(
2105 xfs_inode_t *free_ip,
2106 xfs_trans_t *tp,
2107 xfs_ino_t inum)
2108{
2109 xfs_mount_t *mp = free_ip->i_mount;
2110 int blks_per_cluster;
2111 int nbufs;
2112 int ninodes;
2113 int i, j, found, pre_flushed;
2114 xfs_daddr_t blkno;
2115 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 xfs_inode_t *ip, **ip_found;
2117 xfs_inode_log_item_t *iip;
2118 xfs_log_item_t *lip;
David Chinnerda353b02007-08-28 14:00:13 +10002119 xfs_perag_t *pag = xfs_get_perag(mp, inum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120
2121 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2122 blks_per_cluster = 1;
2123 ninodes = mp->m_sb.sb_inopblock;
2124 nbufs = XFS_IALLOC_BLOCKS(mp);
2125 } else {
2126 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2127 mp->m_sb.sb_blocksize;
2128 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2129 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2130 }
2131
2132 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2133
2134 for (j = 0; j < nbufs; j++, inum += ninodes) {
2135 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2136 XFS_INO_TO_AGBNO(mp, inum));
2137
2138
2139 /*
2140 * Look for each inode in memory and attempt to lock it,
2141 * we can be racing with flush and tail pushing here.
2142 * any inode we get the locks on, add to an array of
2143 * inode items to process later.
2144 *
2145 * The get the buffer lock, we could beat a flush
2146 * or tail pushing thread to the lock here, in which
2147 * case they will go looking for the inode buffer
2148 * and fail, we need some other form of interlock
2149 * here.
2150 */
2151 found = 0;
2152 for (i = 0; i < ninodes; i++) {
David Chinnerda353b02007-08-28 14:00:13 +10002153 read_lock(&pag->pag_ici_lock);
2154 ip = radix_tree_lookup(&pag->pag_ici_root,
2155 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156
2157 /* Inode not in memory or we found it already,
2158 * nothing to do
2159 */
David Chinner7a18c382006-11-11 18:04:54 +11002160 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
David Chinnerda353b02007-08-28 14:00:13 +10002161 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 continue;
2163 }
2164
2165 if (xfs_inode_clean(ip)) {
David Chinnerda353b02007-08-28 14:00:13 +10002166 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 continue;
2168 }
2169
2170 /* If we can get the locks then add it to the
2171 * list, otherwise by the time we get the bp lock
2172 * below it will already be attached to the
2173 * inode buffer.
2174 */
2175
2176 /* This inode will already be locked - by us, lets
2177 * keep it that way.
2178 */
2179
2180 if (ip == free_ip) {
2181 if (xfs_iflock_nowait(ip)) {
David Chinner7a18c382006-11-11 18:04:54 +11002182 xfs_iflags_set(ip, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 if (xfs_inode_clean(ip)) {
2184 xfs_ifunlock(ip);
2185 } else {
2186 ip_found[found++] = ip;
2187 }
2188 }
David Chinnerda353b02007-08-28 14:00:13 +10002189 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 continue;
2191 }
2192
2193 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2194 if (xfs_iflock_nowait(ip)) {
David Chinner7a18c382006-11-11 18:04:54 +11002195 xfs_iflags_set(ip, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196
2197 if (xfs_inode_clean(ip)) {
2198 xfs_ifunlock(ip);
2199 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2200 } else {
2201 ip_found[found++] = ip;
2202 }
2203 } else {
2204 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2205 }
2206 }
David Chinnerda353b02007-08-28 14:00:13 +10002207 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 }
2209
2210 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2211 mp->m_bsize * blks_per_cluster,
2212 XFS_BUF_LOCK);
2213
2214 pre_flushed = 0;
2215 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2216 while (lip) {
2217 if (lip->li_type == XFS_LI_INODE) {
2218 iip = (xfs_inode_log_item_t *)lip;
2219 ASSERT(iip->ili_logged == 1);
2220 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
Donald Douwsma287f3da2007-10-11 17:36:05 +10002221 spin_lock(&mp->m_ail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 iip->ili_flush_lsn = iip->ili_item.li_lsn;
Donald Douwsma287f3da2007-10-11 17:36:05 +10002223 spin_unlock(&mp->m_ail_lock);
David Chinnere5ffd2b2006-11-21 18:55:33 +11002224 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 pre_flushed++;
2226 }
2227 lip = lip->li_bio_list;
2228 }
2229
2230 for (i = 0; i < found; i++) {
2231 ip = ip_found[i];
2232 iip = ip->i_itemp;
2233
2234 if (!iip) {
2235 ip->i_update_core = 0;
2236 xfs_ifunlock(ip);
2237 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2238 continue;
2239 }
2240
2241 iip->ili_last_fields = iip->ili_format.ilf_fields;
2242 iip->ili_format.ilf_fields = 0;
2243 iip->ili_logged = 1;
Donald Douwsma287f3da2007-10-11 17:36:05 +10002244 spin_lock(&mp->m_ail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 iip->ili_flush_lsn = iip->ili_item.li_lsn;
Donald Douwsma287f3da2007-10-11 17:36:05 +10002246 spin_unlock(&mp->m_ail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248 xfs_buf_attach_iodone(bp,
2249 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2250 xfs_istale_done, (xfs_log_item_t *)iip);
2251 if (ip != free_ip) {
2252 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2253 }
2254 }
2255
2256 if (found || pre_flushed)
2257 xfs_trans_stale_inode_buf(tp, bp);
2258 xfs_trans_binval(tp, bp);
2259 }
2260
2261 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
David Chinnerda353b02007-08-28 14:00:13 +10002262 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263}
2264
2265/*
2266 * This is called to return an inode to the inode free list.
2267 * The inode should already be truncated to 0 length and have
2268 * no pages associated with it. This routine also assumes that
2269 * the inode is already a part of the transaction.
2270 *
2271 * The on-disk copy of the inode will have been added to the list
2272 * of unlinked inodes in the AGI. We need to remove the inode from
2273 * that list atomically with respect to freeing it here.
2274 */
2275int
2276xfs_ifree(
2277 xfs_trans_t *tp,
2278 xfs_inode_t *ip,
2279 xfs_bmap_free_t *flist)
2280{
2281 int error;
2282 int delete;
2283 xfs_ino_t first_ino;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002284 xfs_dinode_t *dip;
2285 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002287 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 ASSERT(ip->i_transp == tp);
2289 ASSERT(ip->i_d.di_nlink == 0);
2290 ASSERT(ip->i_d.di_nextents == 0);
2291 ASSERT(ip->i_d.di_anextents == 0);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002292 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2294 ASSERT(ip->i_d.di_nblocks == 0);
2295
2296 /*
2297 * Pull the on-disk inode from the AGI unlinked list.
2298 */
2299 error = xfs_iunlink_remove(tp, ip);
2300 if (error != 0) {
2301 return error;
2302 }
2303
2304 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2305 if (error != 0) {
2306 return error;
2307 }
2308 ip->i_d.di_mode = 0; /* mark incore inode as free */
2309 ip->i_d.di_flags = 0;
2310 ip->i_d.di_dmevmask = 0;
2311 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2312 ip->i_df.if_ext_max =
2313 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2314 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2315 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2316 /*
2317 * Bump the generation count so no one will be confused
2318 * by reincarnations of this inode.
2319 */
2320 ip->i_d.di_gen++;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002321
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2323
David Chinnera3f74ff2008-03-06 13:43:42 +11002324 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, 0, 0, XFS_BUF_LOCK);
Vlad Apostolovc319b582007-11-23 16:27:51 +11002325 if (error)
2326 return error;
2327
2328 /*
2329 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2330 * from picking up this inode when it is reclaimed (its incore state
2331 * initialzed but not flushed to disk yet). The in-core di_mode is
2332 * already cleared and a corresponding transaction logged.
2333 * The hack here just synchronizes the in-core to on-disk
2334 * di_mode value in advance before the actual inode sync to disk.
2335 * This is OK because the inode is already unlinked and would never
2336 * change its di_mode again for this inode generation.
2337 * This is a temporary hack that would require a proper fix
2338 * in the future.
2339 */
2340 dip->di_core.di_mode = 0;
2341
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 if (delete) {
2343 xfs_ifree_cluster(ip, tp, first_ino);
2344 }
2345
2346 return 0;
2347}
2348
2349/*
2350 * Reallocate the space for if_broot based on the number of records
2351 * being added or deleted as indicated in rec_diff. Move the records
2352 * and pointers in if_broot to fit the new size. When shrinking this
2353 * will eliminate holes between the records and pointers created by
2354 * the caller. When growing this will create holes to be filled in
2355 * by the caller.
2356 *
2357 * The caller must not request to add more records than would fit in
2358 * the on-disk inode root. If the if_broot is currently NULL, then
2359 * if we adding records one will be allocated. The caller must also
2360 * not request that the number of records go below zero, although
2361 * it can go to zero.
2362 *
2363 * ip -- the inode whose if_broot area is changing
2364 * ext_diff -- the change in the number of records, positive or negative,
2365 * requested for the if_broot array.
2366 */
2367void
2368xfs_iroot_realloc(
2369 xfs_inode_t *ip,
2370 int rec_diff,
2371 int whichfork)
2372{
2373 int cur_max;
2374 xfs_ifork_t *ifp;
2375 xfs_bmbt_block_t *new_broot;
2376 int new_max;
2377 size_t new_size;
2378 char *np;
2379 char *op;
2380
2381 /*
2382 * Handle the degenerate case quietly.
2383 */
2384 if (rec_diff == 0) {
2385 return;
2386 }
2387
2388 ifp = XFS_IFORK_PTR(ip, whichfork);
2389 if (rec_diff > 0) {
2390 /*
2391 * If there wasn't any memory allocated before, just
2392 * allocate it now and get out.
2393 */
2394 if (ifp->if_broot_bytes == 0) {
2395 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2396 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2397 KM_SLEEP);
2398 ifp->if_broot_bytes = (int)new_size;
2399 return;
2400 }
2401
2402 /*
2403 * If there is already an existing if_broot, then we need
2404 * to realloc() it and shift the pointers to their new
2405 * location. The records don't change location because
2406 * they are kept butted up against the btree block header.
2407 */
2408 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2409 new_max = cur_max + rec_diff;
2410 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2411 ifp->if_broot = (xfs_bmbt_block_t *)
2412 kmem_realloc(ifp->if_broot,
2413 new_size,
2414 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2415 KM_SLEEP);
2416 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2417 ifp->if_broot_bytes);
2418 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2419 (int)new_size);
2420 ifp->if_broot_bytes = (int)new_size;
2421 ASSERT(ifp->if_broot_bytes <=
2422 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2423 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2424 return;
2425 }
2426
2427 /*
2428 * rec_diff is less than 0. In this case, we are shrinking the
2429 * if_broot buffer. It must already exist. If we go to zero
2430 * records, just get rid of the root and clear the status bit.
2431 */
2432 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2433 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2434 new_max = cur_max + rec_diff;
2435 ASSERT(new_max >= 0);
2436 if (new_max > 0)
2437 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2438 else
2439 new_size = 0;
2440 if (new_size > 0) {
2441 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2442 /*
2443 * First copy over the btree block header.
2444 */
2445 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2446 } else {
2447 new_broot = NULL;
2448 ifp->if_flags &= ~XFS_IFBROOT;
2449 }
2450
2451 /*
2452 * Only copy the records and pointers if there are any.
2453 */
2454 if (new_max > 0) {
2455 /*
2456 * First copy the records.
2457 */
2458 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2459 ifp->if_broot_bytes);
2460 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2461 (int)new_size);
2462 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2463
2464 /*
2465 * Then copy the pointers.
2466 */
2467 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2468 ifp->if_broot_bytes);
2469 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2470 (int)new_size);
2471 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2472 }
2473 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2474 ifp->if_broot = new_broot;
2475 ifp->if_broot_bytes = (int)new_size;
2476 ASSERT(ifp->if_broot_bytes <=
2477 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2478 return;
2479}
2480
2481
2482/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 * This is called when the amount of space needed for if_data
2484 * is increased or decreased. The change in size is indicated by
2485 * the number of bytes that need to be added or deleted in the
2486 * byte_diff parameter.
2487 *
2488 * If the amount of space needed has decreased below the size of the
2489 * inline buffer, then switch to using the inline buffer. Otherwise,
2490 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2491 * to what is needed.
2492 *
2493 * ip -- the inode whose if_data area is changing
2494 * byte_diff -- the change in the number of bytes, positive or negative,
2495 * requested for the if_data array.
2496 */
2497void
2498xfs_idata_realloc(
2499 xfs_inode_t *ip,
2500 int byte_diff,
2501 int whichfork)
2502{
2503 xfs_ifork_t *ifp;
2504 int new_size;
2505 int real_size;
2506
2507 if (byte_diff == 0) {
2508 return;
2509 }
2510
2511 ifp = XFS_IFORK_PTR(ip, whichfork);
2512 new_size = (int)ifp->if_bytes + byte_diff;
2513 ASSERT(new_size >= 0);
2514
2515 if (new_size == 0) {
2516 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2517 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2518 }
2519 ifp->if_u1.if_data = NULL;
2520 real_size = 0;
2521 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2522 /*
2523 * If the valid extents/data can fit in if_inline_ext/data,
2524 * copy them from the malloc'd vector and free it.
2525 */
2526 if (ifp->if_u1.if_data == NULL) {
2527 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2528 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2529 ASSERT(ifp->if_real_bytes != 0);
2530 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2531 new_size);
2532 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2533 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2534 }
2535 real_size = 0;
2536 } else {
2537 /*
2538 * Stuck with malloc/realloc.
2539 * For inline data, the underlying buffer must be
2540 * a multiple of 4 bytes in size so that it can be
2541 * logged and stay on word boundaries. We enforce
2542 * that here.
2543 */
2544 real_size = roundup(new_size, 4);
2545 if (ifp->if_u1.if_data == NULL) {
2546 ASSERT(ifp->if_real_bytes == 0);
2547 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2548 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2549 /*
2550 * Only do the realloc if the underlying size
2551 * is really changing.
2552 */
2553 if (ifp->if_real_bytes != real_size) {
2554 ifp->if_u1.if_data =
2555 kmem_realloc(ifp->if_u1.if_data,
2556 real_size,
2557 ifp->if_real_bytes,
2558 KM_SLEEP);
2559 }
2560 } else {
2561 ASSERT(ifp->if_real_bytes == 0);
2562 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2563 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2564 ifp->if_bytes);
2565 }
2566 }
2567 ifp->if_real_bytes = real_size;
2568 ifp->if_bytes = new_size;
2569 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2570}
2571
2572
2573
2574
2575/*
2576 * Map inode to disk block and offset.
2577 *
2578 * mp -- the mount point structure for the current file system
2579 * tp -- the current transaction
2580 * ino -- the inode number of the inode to be located
2581 * imap -- this structure is filled in with the information necessary
2582 * to retrieve the given inode from disk
2583 * flags -- flags to pass to xfs_dilocate indicating whether or not
2584 * lookups in the inode btree were OK or not
2585 */
2586int
2587xfs_imap(
2588 xfs_mount_t *mp,
2589 xfs_trans_t *tp,
2590 xfs_ino_t ino,
2591 xfs_imap_t *imap,
2592 uint flags)
2593{
2594 xfs_fsblock_t fsbno;
2595 int len;
2596 int off;
2597 int error;
2598
2599 fsbno = imap->im_blkno ?
2600 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2601 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
David Chinner4ae29b42008-03-06 13:43:34 +11002602 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 return error;
David Chinner4ae29b42008-03-06 13:43:34 +11002604
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2606 imap->im_len = XFS_FSB_TO_BB(mp, len);
2607 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2608 imap->im_ioffset = (ushort)off;
2609 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
David Chinner4ae29b42008-03-06 13:43:34 +11002610
2611 /*
2612 * If the inode number maps to a block outside the bounds
2613 * of the file system then return NULL rather than calling
2614 * read_buf and panicing when we get an error from the
2615 * driver.
2616 */
2617 if ((imap->im_blkno + imap->im_len) >
2618 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2619 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
2620 "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
2621 " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
2622 (unsigned long long) imap->im_blkno,
2623 (unsigned long long) imap->im_len,
2624 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2625 return EINVAL;
2626 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 return 0;
2628}
2629
2630void
2631xfs_idestroy_fork(
2632 xfs_inode_t *ip,
2633 int whichfork)
2634{
2635 xfs_ifork_t *ifp;
2636
2637 ifp = XFS_IFORK_PTR(ip, whichfork);
2638 if (ifp->if_broot != NULL) {
2639 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2640 ifp->if_broot = NULL;
2641 }
2642
2643 /*
2644 * If the format is local, then we can't have an extents
2645 * array so just look for an inline data array. If we're
2646 * not local then we may or may not have an extents list,
2647 * so check and free it up if we do.
2648 */
2649 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2650 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2651 (ifp->if_u1.if_data != NULL)) {
2652 ASSERT(ifp->if_real_bytes != 0);
2653 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2654 ifp->if_u1.if_data = NULL;
2655 ifp->if_real_bytes = 0;
2656 }
2657 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11002658 ((ifp->if_flags & XFS_IFEXTIREC) ||
2659 ((ifp->if_u1.if_extents != NULL) &&
2660 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 ASSERT(ifp->if_real_bytes != 0);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002662 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 }
2664 ASSERT(ifp->if_u1.if_extents == NULL ||
2665 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2666 ASSERT(ifp->if_real_bytes == 0);
2667 if (whichfork == XFS_ATTR_FORK) {
2668 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2669 ip->i_afp = NULL;
2670 }
2671}
2672
2673/*
2674 * This is called free all the memory associated with an inode.
2675 * It must free the inode itself and any buffers allocated for
2676 * if_extents/if_data and if_broot. It must also free the lock
2677 * associated with the inode.
2678 */
2679void
2680xfs_idestroy(
2681 xfs_inode_t *ip)
2682{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 switch (ip->i_d.di_mode & S_IFMT) {
2684 case S_IFREG:
2685 case S_IFDIR:
2686 case S_IFLNK:
2687 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2688 break;
2689 }
2690 if (ip->i_afp)
2691 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2692 mrfree(&ip->i_lock);
2693 mrfree(&ip->i_iolock);
2694 freesema(&ip->i_flock);
Christoph Hellwig1543d792007-08-29 11:46:47 +10002695
Lachlan McIlroycf441ee2008-02-07 16:42:19 +11002696#ifdef XFS_INODE_TRACE
Christoph Hellwig1543d792007-08-29 11:46:47 +10002697 ktrace_free(ip->i_trace);
2698#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699#ifdef XFS_BMAP_TRACE
2700 ktrace_free(ip->i_xtrace);
2701#endif
2702#ifdef XFS_BMBT_TRACE
2703 ktrace_free(ip->i_btrace);
2704#endif
2705#ifdef XFS_RW_TRACE
2706 ktrace_free(ip->i_rwtrace);
2707#endif
2708#ifdef XFS_ILOCK_TRACE
2709 ktrace_free(ip->i_lock_trace);
2710#endif
2711#ifdef XFS_DIR2_TRACE
2712 ktrace_free(ip->i_dir_trace);
2713#endif
2714 if (ip->i_itemp) {
David Chinnerf74eaf52007-02-10 18:36:04 +11002715 /*
2716 * Only if we are shutting down the fs will we see an
2717 * inode still in the AIL. If it is there, we should remove
2718 * it to prevent a use-after-free from occurring.
2719 */
2720 xfs_mount_t *mp = ip->i_mount;
2721 xfs_log_item_t *lip = &ip->i_itemp->ili_item;
David Chinnerf74eaf52007-02-10 18:36:04 +11002722
2723 ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
2724 XFS_FORCED_SHUTDOWN(ip->i_mount));
2725 if (lip->li_flags & XFS_LI_IN_AIL) {
Donald Douwsma287f3da2007-10-11 17:36:05 +10002726 spin_lock(&mp->m_ail_lock);
David Chinnerf74eaf52007-02-10 18:36:04 +11002727 if (lip->li_flags & XFS_LI_IN_AIL)
Donald Douwsma287f3da2007-10-11 17:36:05 +10002728 xfs_trans_delete_ail(mp, lip);
David Chinnerf74eaf52007-02-10 18:36:04 +11002729 else
Donald Douwsma287f3da2007-10-11 17:36:05 +10002730 spin_unlock(&mp->m_ail_lock);
David Chinnerf74eaf52007-02-10 18:36:04 +11002731 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 xfs_inode_item_destroy(ip);
2733 }
2734 kmem_zone_free(xfs_inode_zone, ip);
2735}
2736
2737
2738/*
2739 * Increment the pin count of the given buffer.
2740 * This value is protected by ipinlock spinlock in the mount structure.
2741 */
2742void
2743xfs_ipin(
2744 xfs_inode_t *ip)
2745{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002746 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747
2748 atomic_inc(&ip->i_pincount);
2749}
2750
2751/*
2752 * Decrement the pin count of the given inode, and wake up
2753 * anyone in xfs_iwait_unpin() if the count goes to 0. The
Nathan Scottc41564b2006-03-29 08:55:14 +10002754 * inode must have been previously pinned with a call to xfs_ipin().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 */
2756void
2757xfs_iunpin(
2758 xfs_inode_t *ip)
2759{
2760 ASSERT(atomic_read(&ip->i_pincount) > 0);
2761
David Chinner5d51eff2007-11-23 16:29:18 +11002762 if (atomic_dec_and_test(&ip->i_pincount))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 wake_up(&ip->i_ipin_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764}
2765
2766/*
David Chinnera3f74ff2008-03-06 13:43:42 +11002767 * This is called to unpin an inode. It can be directed to wait or to return
2768 * immediately without waiting for the inode to be unpinned. The caller must
2769 * have the inode locked in at least shared mode so that the buffer cannot be
2770 * subsequently pinned once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002772STATIC void
David Chinnera3f74ff2008-03-06 13:43:42 +11002773__xfs_iunpin_wait(
2774 xfs_inode_t *ip,
2775 int wait)
2776{
2777 xfs_inode_log_item_t *iip = ip->i_itemp;
2778
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002779 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002780 if (atomic_read(&ip->i_pincount) == 0)
2781 return;
2782
2783 /* Give the log a push to start the unpinning I/O */
2784 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2785 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2786 if (wait)
2787 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2788}
2789
2790static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791xfs_iunpin_wait(
2792 xfs_inode_t *ip)
2793{
David Chinnera3f74ff2008-03-06 13:43:42 +11002794 __xfs_iunpin_wait(ip, 1);
2795}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796
David Chinnera3f74ff2008-03-06 13:43:42 +11002797static inline void
2798xfs_iunpin_nowait(
2799 xfs_inode_t *ip)
2800{
2801 __xfs_iunpin_wait(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802}
2803
2804
2805/*
2806 * xfs_iextents_copy()
2807 *
2808 * This is called to copy the REAL extents (as opposed to the delayed
2809 * allocation extents) from the inode into the given buffer. It
2810 * returns the number of bytes copied into the buffer.
2811 *
2812 * If there are no delayed allocation extents, then we can just
2813 * memcpy() the extents into the buffer. Otherwise, we need to
2814 * examine each extent in turn and skip those which are delayed.
2815 */
2816int
2817xfs_iextents_copy(
2818 xfs_inode_t *ip,
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002819 xfs_bmbt_rec_t *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 int whichfork)
2821{
2822 int copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 int i;
2824 xfs_ifork_t *ifp;
2825 int nrecs;
2826 xfs_fsblock_t start_block;
2827
2828 ifp = XFS_IFORK_PTR(ip, whichfork);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002829 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 ASSERT(ifp->if_bytes > 0);
2831
2832 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
Eric Sandeen3a59c942007-07-11 11:09:47 +10002833 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 ASSERT(nrecs > 0);
2835
2836 /*
2837 * There are some delayed allocation extents in the
2838 * inode, so copy the extents one at a time and skip
2839 * the delayed ones. There must be at least one
2840 * non-delayed extent.
2841 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 copied = 0;
2843 for (i = 0; i < nrecs; i++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002844 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 start_block = xfs_bmbt_get_startblock(ep);
2846 if (ISNULLSTARTBLOCK(start_block)) {
2847 /*
2848 * It's a delayed allocation extent, so skip it.
2849 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 continue;
2851 }
2852
2853 /* Translate to on disk format */
Christoph Hellwigcd8b0a92007-08-16 16:24:15 +10002854 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2855 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002856 dp++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 copied++;
2858 }
2859 ASSERT(copied != 0);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002860 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861
2862 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2863}
2864
2865/*
2866 * Each of the following cases stores data into the same region
2867 * of the on-disk inode, so only one of them can be valid at
2868 * any given time. While it is possible to have conflicting formats
2869 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2870 * in EXTENTS format, this can only happen when the fork has
2871 * changed formats after being modified but before being flushed.
2872 * In these cases, the format always takes precedence, because the
2873 * format indicates the current state of the fork.
2874 */
2875/*ARGSUSED*/
David Chinnere4ac9672008-04-10 12:23:58 +10002876STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877xfs_iflush_fork(
2878 xfs_inode_t *ip,
2879 xfs_dinode_t *dip,
2880 xfs_inode_log_item_t *iip,
2881 int whichfork,
2882 xfs_buf_t *bp)
2883{
2884 char *cp;
2885 xfs_ifork_t *ifp;
2886 xfs_mount_t *mp;
2887#ifdef XFS_TRANS_DEBUG
2888 int first;
2889#endif
2890 static const short brootflag[2] =
2891 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2892 static const short dataflag[2] =
2893 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2894 static const short extflag[2] =
2895 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2896
David Chinnere4ac9672008-04-10 12:23:58 +10002897 if (!iip)
2898 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 ifp = XFS_IFORK_PTR(ip, whichfork);
2900 /*
2901 * This can happen if we gave up in iformat in an error path,
2902 * for the attribute fork.
2903 */
David Chinnere4ac9672008-04-10 12:23:58 +10002904 if (!ifp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 ASSERT(whichfork == XFS_ATTR_FORK);
David Chinnere4ac9672008-04-10 12:23:58 +10002906 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 }
2908 cp = XFS_DFORK_PTR(dip, whichfork);
2909 mp = ip->i_mount;
2910 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2911 case XFS_DINODE_FMT_LOCAL:
2912 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2913 (ifp->if_bytes > 0)) {
2914 ASSERT(ifp->if_u1.if_data != NULL);
2915 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2916 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2917 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 break;
2919
2920 case XFS_DINODE_FMT_EXTENTS:
2921 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2922 !(iip->ili_format.ilf_fields & extflag[whichfork]));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002923 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2924 (ifp->if_bytes == 0));
2925 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2926 (ifp->if_bytes > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2928 (ifp->if_bytes > 0)) {
2929 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2930 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2931 whichfork);
2932 }
2933 break;
2934
2935 case XFS_DINODE_FMT_BTREE:
2936 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2937 (ifp->if_broot_bytes > 0)) {
2938 ASSERT(ifp->if_broot != NULL);
2939 ASSERT(ifp->if_broot_bytes <=
2940 (XFS_IFORK_SIZE(ip, whichfork) +
2941 XFS_BROOT_SIZE_ADJ));
2942 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2943 (xfs_bmdr_block_t *)cp,
2944 XFS_DFORK_SIZE(dip, mp, whichfork));
2945 }
2946 break;
2947
2948 case XFS_DINODE_FMT_DEV:
2949 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2950 ASSERT(whichfork == XFS_DATA_FORK);
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002951 dip->di_u.di_dev = cpu_to_be32(ip->i_df.if_u2.if_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 }
2953 break;
2954
2955 case XFS_DINODE_FMT_UUID:
2956 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2957 ASSERT(whichfork == XFS_DATA_FORK);
2958 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2959 sizeof(uuid_t));
2960 }
2961 break;
2962
2963 default:
2964 ASSERT(0);
2965 break;
2966 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967}
2968
David Chinnerbad55842008-03-06 13:43:49 +11002969STATIC int
2970xfs_iflush_cluster(
2971 xfs_inode_t *ip,
2972 xfs_buf_t *bp)
2973{
2974 xfs_mount_t *mp = ip->i_mount;
2975 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
2976 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10002977 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11002978 int ilist_size;
2979 xfs_inode_t **ilist;
2980 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11002981 int nr_found;
2982 int clcount = 0;
2983 int bufwasdelwri;
2984 int i;
2985
2986 ASSERT(pag->pagi_inodeok);
2987 ASSERT(pag->pag_ici_init);
2988
David Chinnerc8f5f122008-05-20 11:30:15 +10002989 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2990 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10002991 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11002992 if (!ilist)
2993 return 0;
2994
2995 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2996 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2997 read_lock(&pag->pag_ici_lock);
2998 /* really need a gang lookup range call here */
2999 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10003000 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11003001 if (nr_found == 0)
3002 goto out_free;
3003
3004 for (i = 0; i < nr_found; i++) {
3005 iq = ilist[i];
3006 if (iq == ip)
3007 continue;
3008 /* if the inode lies outside this cluster, we're done. */
3009 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
3010 break;
3011 /*
3012 * Do an un-protected check to see if the inode is dirty and
3013 * is a candidate for flushing. These checks will be repeated
3014 * later after the appropriate locks are acquired.
3015 */
David Chinner33540402008-03-06 13:43:59 +11003016 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11003017 continue;
David Chinnerbad55842008-03-06 13:43:49 +11003018
3019 /*
3020 * Try to get locks. If any are unavailable or it is pinned,
3021 * then this inode cannot be flushed and is skipped.
3022 */
3023
3024 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
3025 continue;
3026 if (!xfs_iflock_nowait(iq)) {
3027 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3028 continue;
3029 }
3030 if (xfs_ipincount(iq)) {
3031 xfs_ifunlock(iq);
3032 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3033 continue;
3034 }
3035
3036 /*
3037 * arriving here means that this inode can be flushed. First
3038 * re-check that it's dirty before flushing.
3039 */
David Chinner33540402008-03-06 13:43:59 +11003040 if (!xfs_inode_clean(iq)) {
3041 int error;
David Chinnerbad55842008-03-06 13:43:49 +11003042 error = xfs_iflush_int(iq, bp);
3043 if (error) {
3044 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3045 goto cluster_corrupt_out;
3046 }
3047 clcount++;
3048 } else {
3049 xfs_ifunlock(iq);
3050 }
3051 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3052 }
3053
3054 if (clcount) {
3055 XFS_STATS_INC(xs_icluster_flushcnt);
3056 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3057 }
3058
3059out_free:
3060 read_unlock(&pag->pag_ici_lock);
3061 kmem_free(ilist, ilist_size);
3062 return 0;
3063
3064
3065cluster_corrupt_out:
3066 /*
3067 * Corruption detected in the clustering loop. Invalidate the
3068 * inode buffer and shut down the filesystem.
3069 */
3070 read_unlock(&pag->pag_ici_lock);
3071 /*
3072 * Clean up the buffer. If it was B_DELWRI, just release it --
3073 * brelse can handle it with no problems. If not, shut down the
3074 * filesystem before releasing the buffer.
3075 */
3076 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
3077 if (bufwasdelwri)
3078 xfs_buf_relse(bp);
3079
3080 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3081
3082 if (!bufwasdelwri) {
3083 /*
3084 * Just like incore_relse: if we have b_iodone functions,
3085 * mark the buffer as an error and call them. Otherwise
3086 * mark it as stale and brelse.
3087 */
3088 if (XFS_BUF_IODONE_FUNC(bp)) {
3089 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3090 XFS_BUF_UNDONE(bp);
3091 XFS_BUF_STALE(bp);
3092 XFS_BUF_SHUT(bp);
3093 XFS_BUF_ERROR(bp,EIO);
3094 xfs_biodone(bp);
3095 } else {
3096 XFS_BUF_STALE(bp);
3097 xfs_buf_relse(bp);
3098 }
3099 }
3100
3101 /*
3102 * Unlocks the flush lock
3103 */
3104 xfs_iflush_abort(iq);
3105 kmem_free(ilist, ilist_size);
3106 return XFS_ERROR(EFSCORRUPTED);
3107}
3108
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109/*
3110 * xfs_iflush() will write a modified inode's changes out to the
3111 * inode's on disk home. The caller must have the inode lock held
3112 * in at least shared mode and the inode flush semaphore must be
3113 * held as well. The inode lock will still be held upon return from
3114 * the call and the caller is free to unlock it.
3115 * The inode flush lock will be unlocked when the inode reaches the disk.
3116 * The flags indicate how the inode's buffer should be written out.
3117 */
3118int
3119xfs_iflush(
3120 xfs_inode_t *ip,
3121 uint flags)
3122{
3123 xfs_inode_log_item_t *iip;
3124 xfs_buf_t *bp;
3125 xfs_dinode_t *dip;
3126 xfs_mount_t *mp;
3127 int error;
David Chinnera3f74ff2008-03-06 13:43:42 +11003128 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
David Chinnerbad55842008-03-06 13:43:49 +11003129 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130
3131 XFS_STATS_INC(xs_iflush_count);
3132
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003133 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Al Viro0d8fee32006-06-19 08:41:30 +10003134 ASSERT(issemalocked(&(ip->i_flock)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3136 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3137
3138 iip = ip->i_itemp;
3139 mp = ip->i_mount;
3140
3141 /*
3142 * If the inode isn't dirty, then just release the inode
3143 * flush lock and do nothing.
3144 */
David Chinner33540402008-03-06 13:43:59 +11003145 if (xfs_inode_clean(ip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 ASSERT((iip != NULL) ?
3147 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3148 xfs_ifunlock(ip);
3149 return 0;
3150 }
3151
3152 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003153 * We can't flush the inode until it is unpinned, so wait for it if we
3154 * are allowed to block. We know noone new can pin it, because we are
3155 * holding the inode lock shared and you need to hold it exclusively to
3156 * pin the inode.
3157 *
3158 * If we are not allowed to block, force the log out asynchronously so
3159 * that when we come back the inode will be unpinned. If other inodes
3160 * in the same cluster are dirty, they will probably write the inode
3161 * out for us if they occur after the log force completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 */
David Chinnera3f74ff2008-03-06 13:43:42 +11003163 if (noblock && xfs_ipincount(ip)) {
3164 xfs_iunpin_nowait(ip);
3165 xfs_ifunlock(ip);
3166 return EAGAIN;
3167 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168 xfs_iunpin_wait(ip);
3169
3170 /*
3171 * This may have been unpinned because the filesystem is shutting
3172 * down forcibly. If that's the case we must not write this inode
3173 * to disk, because the log record didn't make it to disk!
3174 */
3175 if (XFS_FORCED_SHUTDOWN(mp)) {
3176 ip->i_update_core = 0;
3177 if (iip)
3178 iip->ili_format.ilf_fields = 0;
3179 xfs_ifunlock(ip);
3180 return XFS_ERROR(EIO);
3181 }
3182
3183 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 * Decide how buffer will be flushed out. This is done before
3185 * the call to xfs_iflush_int because this field is zeroed by it.
3186 */
3187 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3188 /*
3189 * Flush out the inode buffer according to the directions
3190 * of the caller. In the cases where the caller has given
3191 * us a choice choose the non-delwri case. This is because
3192 * the inode is in the AIL and we need to get it out soon.
3193 */
3194 switch (flags) {
3195 case XFS_IFLUSH_SYNC:
3196 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3197 flags = 0;
3198 break;
David Chinnera3f74ff2008-03-06 13:43:42 +11003199 case XFS_IFLUSH_ASYNC_NOBLOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200 case XFS_IFLUSH_ASYNC:
3201 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3202 flags = INT_ASYNC;
3203 break;
3204 case XFS_IFLUSH_DELWRI:
3205 flags = INT_DELWRI;
3206 break;
3207 default:
3208 ASSERT(0);
3209 flags = 0;
3210 break;
3211 }
3212 } else {
3213 switch (flags) {
3214 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3215 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3216 case XFS_IFLUSH_DELWRI:
3217 flags = INT_DELWRI;
3218 break;
David Chinnera3f74ff2008-03-06 13:43:42 +11003219 case XFS_IFLUSH_ASYNC_NOBLOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220 case XFS_IFLUSH_ASYNC:
3221 flags = INT_ASYNC;
3222 break;
3223 case XFS_IFLUSH_SYNC:
3224 flags = 0;
3225 break;
3226 default:
3227 ASSERT(0);
3228 flags = 0;
3229 break;
3230 }
3231 }
3232
3233 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003234 * Get the buffer containing the on-disk inode.
3235 */
3236 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0,
3237 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
3238 if (error || !bp) {
3239 xfs_ifunlock(ip);
3240 return error;
3241 }
3242
3243 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 * First flush out the inode that xfs_iflush was called with.
3245 */
3246 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11003247 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249
3250 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11003251 * If the buffer is pinned then push on the log now so we won't
3252 * get stuck waiting in the write for too long.
3253 */
3254 if (XFS_BUF_ISPINNED(bp))
3255 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3256
3257 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 * inode clustering:
3259 * see if other inodes can be gathered into this write
3260 */
David Chinnerbad55842008-03-06 13:43:49 +11003261 error = xfs_iflush_cluster(ip, bp);
3262 if (error)
3263 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 if (flags & INT_DELWRI) {
3266 xfs_bdwrite(mp, bp);
3267 } else if (flags & INT_ASYNC) {
David Chinnerdb7a19f2008-04-10 12:22:24 +10003268 error = xfs_bawrite(mp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 } else {
3270 error = xfs_bwrite(mp, bp);
3271 }
3272 return error;
3273
3274corrupt_out:
3275 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10003276 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277cluster_corrupt_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 /*
3279 * Unlocks the flush lock
3280 */
David Chinnerbad55842008-03-06 13:43:49 +11003281 xfs_iflush_abort(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282 return XFS_ERROR(EFSCORRUPTED);
3283}
3284
3285
3286STATIC int
3287xfs_iflush_int(
3288 xfs_inode_t *ip,
3289 xfs_buf_t *bp)
3290{
3291 xfs_inode_log_item_t *iip;
3292 xfs_dinode_t *dip;
3293 xfs_mount_t *mp;
3294#ifdef XFS_TRANS_DEBUG
3295 int first;
3296#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003298 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Al Viro0d8fee32006-06-19 08:41:30 +10003299 ASSERT(issemalocked(&(ip->i_flock)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3301 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3302
3303 iip = ip->i_itemp;
3304 mp = ip->i_mount;
3305
3306
3307 /*
3308 * If the inode isn't dirty, then just release the inode
3309 * flush lock and do nothing.
3310 */
David Chinner33540402008-03-06 13:43:59 +11003311 if (xfs_inode_clean(ip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312 xfs_ifunlock(ip);
3313 return 0;
3314 }
3315
3316 /* set *dip = inode's place in the buffer */
3317 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3318
3319 /*
3320 * Clear i_update_core before copying out the data.
3321 * This is for coordination with our timestamp updates
3322 * that don't hold the inode lock. They will always
3323 * update the timestamps BEFORE setting i_update_core,
3324 * so if we clear i_update_core after they set it we
3325 * are guaranteed to see their updates to the timestamps.
3326 * I believe that this depends on strongly ordered memory
3327 * semantics, but we have that. We use the SYNCHRONIZE
3328 * macro to make sure that the compiler does not reorder
3329 * the i_update_core access below the data copy below.
3330 */
3331 ip->i_update_core = 0;
3332 SYNCHRONIZE();
3333
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11003334 /*
3335 * Make sure to get the latest atime from the Linux inode.
3336 */
3337 xfs_synchronize_atime(ip);
3338
Christoph Hellwig347d1c02007-08-28 13:57:51 +10003339 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_core.di_magic) != XFS_DINODE_MAGIC,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3341 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3342 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
Christoph Hellwig347d1c02007-08-28 13:57:51 +10003343 ip->i_ino, be16_to_cpu(dip->di_core.di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344 goto corrupt_out;
3345 }
3346 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3347 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3348 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3349 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3350 ip->i_ino, ip, ip->i_d.di_magic);
3351 goto corrupt_out;
3352 }
3353 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3354 if (XFS_TEST_ERROR(
3355 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3356 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3357 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3358 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3359 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3360 ip->i_ino, ip);
3361 goto corrupt_out;
3362 }
3363 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3364 if (XFS_TEST_ERROR(
3365 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3366 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3367 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3368 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3369 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3370 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3371 ip->i_ino, ip);
3372 goto corrupt_out;
3373 }
3374 }
3375 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3376 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3377 XFS_RANDOM_IFLUSH_5)) {
3378 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3379 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3380 ip->i_ino,
3381 ip->i_d.di_nextents + ip->i_d.di_anextents,
3382 ip->i_d.di_nblocks,
3383 ip);
3384 goto corrupt_out;
3385 }
3386 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3387 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3388 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3389 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3390 ip->i_ino, ip->i_d.di_forkoff, ip);
3391 goto corrupt_out;
3392 }
3393 /*
3394 * bump the flush iteration count, used to detect flushes which
3395 * postdate a log record during recovery.
3396 */
3397
3398 ip->i_d.di_flushiter++;
3399
3400 /*
3401 * Copy the dirty parts of the inode into the on-disk
3402 * inode. We always copy out the core of the inode,
3403 * because if the inode is dirty at all the core must
3404 * be.
3405 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10003406 xfs_dinode_to_disk(&dip->di_core, &ip->i_d);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407
3408 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3409 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3410 ip->i_d.di_flushiter = 0;
3411
3412 /*
3413 * If this is really an old format inode and the superblock version
3414 * has not been updated to support only new format inodes, then
3415 * convert back to the old inode format. If the superblock version
3416 * has been updated, then make the conversion permanent.
3417 */
3418 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
Eric Sandeen62118702008-03-06 13:44:28 +11003419 xfs_sb_version_hasnlink(&mp->m_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
Eric Sandeen62118702008-03-06 13:44:28 +11003421 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422 /*
3423 * Convert it back.
3424 */
3425 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
Christoph Hellwig347d1c02007-08-28 13:57:51 +10003426 dip->di_core.di_onlink = cpu_to_be16(ip->i_d.di_nlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 } else {
3428 /*
3429 * The superblock version has already been bumped,
3430 * so just make the conversion to the new inode
3431 * format permanent.
3432 */
3433 ip->i_d.di_version = XFS_DINODE_VERSION_2;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10003434 dip->di_core.di_version = XFS_DINODE_VERSION_2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435 ip->i_d.di_onlink = 0;
3436 dip->di_core.di_onlink = 0;
3437 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3438 memset(&(dip->di_core.di_pad[0]), 0,
3439 sizeof(dip->di_core.di_pad));
3440 ASSERT(ip->i_d.di_projid == 0);
3441 }
3442 }
3443
David Chinnere4ac9672008-04-10 12:23:58 +10003444 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3445 if (XFS_IFORK_Q(ip))
3446 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447 xfs_inobp_check(mp, bp);
3448
3449 /*
3450 * We've recorded everything logged in the inode, so we'd
3451 * like to clear the ilf_fields bits so we don't log and
3452 * flush things unnecessarily. However, we can't stop
3453 * logging all this information until the data we've copied
3454 * into the disk buffer is written to disk. If we did we might
3455 * overwrite the copy of the inode in the log with all the
3456 * data after re-logging only part of it, and in the face of
3457 * a crash we wouldn't have all the data we need to recover.
3458 *
3459 * What we do is move the bits to the ili_last_fields field.
3460 * When logging the inode, these bits are moved back to the
3461 * ilf_fields field. In the xfs_iflush_done() routine we
3462 * clear ili_last_fields, since we know that the information
3463 * those bits represent is permanently on disk. As long as
3464 * the flush completes before the inode is logged again, then
3465 * both ilf_fields and ili_last_fields will be cleared.
3466 *
3467 * We can play with the ilf_fields bits here, because the inode
3468 * lock must be held exclusively in order to set bits there
3469 * and the flush lock protects the ili_last_fields bits.
3470 * Set ili_logged so the flush done
3471 * routine can tell whether or not to look in the AIL.
3472 * Also, store the current LSN of the inode so that we can tell
3473 * whether the item has moved in the AIL from xfs_iflush_done().
3474 * In order to read the lsn we need the AIL lock, because
3475 * it is a 64 bit value that cannot be read atomically.
3476 */
3477 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3478 iip->ili_last_fields = iip->ili_format.ilf_fields;
3479 iip->ili_format.ilf_fields = 0;
3480 iip->ili_logged = 1;
3481
3482 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
Donald Douwsma287f3da2007-10-11 17:36:05 +10003483 spin_lock(&mp->m_ail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 iip->ili_flush_lsn = iip->ili_item.li_lsn;
Donald Douwsma287f3da2007-10-11 17:36:05 +10003485 spin_unlock(&mp->m_ail_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003486
3487 /*
3488 * Attach the function xfs_iflush_done to the inode's
3489 * buffer. This will remove the inode from the AIL
3490 * and unlock the inode's flush lock when the inode is
3491 * completely written to disk.
3492 */
3493 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3494 xfs_iflush_done, (xfs_log_item_t *)iip);
3495
3496 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3497 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3498 } else {
3499 /*
3500 * We're flushing an inode which is not in the AIL and has
3501 * not been logged but has i_update_core set. For this
3502 * case we can use a B_DELWRI flush and immediately drop
3503 * the inode flush lock because we can avoid the whole
3504 * AIL state thing. It's OK to drop the flush lock now,
3505 * because we've already locked the buffer and to do anything
3506 * you really need both.
3507 */
3508 if (iip != NULL) {
3509 ASSERT(iip->ili_logged == 0);
3510 ASSERT(iip->ili_last_fields == 0);
3511 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3512 }
3513 xfs_ifunlock(ip);
3514 }
3515
3516 return 0;
3517
3518corrupt_out:
3519 return XFS_ERROR(EFSCORRUPTED);
3520}
3521
3522
3523/*
Christoph Hellwigefa80272005-06-21 15:37:17 +10003524 * Flush all inactive inodes in mp.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525 */
Christoph Hellwigefa80272005-06-21 15:37:17 +10003526void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527xfs_iflush_all(
Christoph Hellwigefa80272005-06-21 15:37:17 +10003528 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10003531 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532
Christoph Hellwigefa80272005-06-21 15:37:17 +10003533 again:
3534 XFS_MOUNT_ILOCK(mp);
3535 ip = mp->m_inodes;
3536 if (ip == NULL)
3537 goto out;
3538
3539 do {
3540 /* Make sure we skip markers inserted by sync */
3541 if (ip->i_mount == NULL) {
3542 ip = ip->i_mnext;
3543 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545
Christoph Hellwigefa80272005-06-21 15:37:17 +10003546 vp = XFS_ITOV_NULL(ip);
3547 if (!vp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 XFS_MOUNT_IUNLOCK(mp);
Christoph Hellwigefa80272005-06-21 15:37:17 +10003549 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3550 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552
Christoph Hellwigefa80272005-06-21 15:37:17 +10003553 ASSERT(vn_count(vp) == 0);
3554
3555 ip = ip->i_mnext;
3556 } while (ip != mp->m_inodes);
3557 out:
3558 XFS_MOUNT_IUNLOCK(mp);
3559}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561#ifdef XFS_ILOCK_TRACE
3562ktrace_t *xfs_ilock_trace_buf;
3563
3564void
3565xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3566{
3567 ktrace_enter(ip->i_lock_trace,
3568 (void *)ip,
3569 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3570 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3571 (void *)ra, /* caller of ilock */
3572 (void *)(unsigned long)current_cpu(),
3573 (void *)(unsigned long)current_pid(),
3574 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3575}
3576#endif
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003577
3578/*
3579 * Return a pointer to the extent record at file index idx.
3580 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003581xfs_bmbt_rec_host_t *
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003582xfs_iext_get_ext(
3583 xfs_ifork_t *ifp, /* inode fork pointer */
3584 xfs_extnum_t idx) /* index of target extent */
3585{
3586 ASSERT(idx >= 0);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003587 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3588 return ifp->if_u1.if_ext_irec->er_extbuf;
3589 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3590 xfs_ext_irec_t *erp; /* irec pointer */
3591 int erp_idx = 0; /* irec index */
3592 xfs_extnum_t page_idx = idx; /* ext index in target list */
3593
3594 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3595 return &erp->er_extbuf[page_idx];
3596 } else if (ifp->if_bytes) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003597 return &ifp->if_u1.if_extents[idx];
3598 } else {
3599 return NULL;
3600 }
3601}
3602
3603/*
3604 * Insert new item(s) into the extent records for incore inode
3605 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3606 */
3607void
3608xfs_iext_insert(
3609 xfs_ifork_t *ifp, /* inode fork pointer */
3610 xfs_extnum_t idx, /* starting index of new items */
3611 xfs_extnum_t count, /* number of inserted items */
3612 xfs_bmbt_irec_t *new) /* items to insert */
3613{
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003614 xfs_extnum_t i; /* extent record index */
3615
3616 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3617 xfs_iext_add(ifp, idx, count);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003618 for (i = idx; i < idx + count; i++, new++)
3619 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003620}
3621
3622/*
3623 * This is called when the amount of space required for incore file
3624 * extents needs to be increased. The ext_diff parameter stores the
3625 * number of new extents being added and the idx parameter contains
3626 * the extent index where the new extents will be added. If the new
3627 * extents are being appended, then we just need to (re)allocate and
3628 * initialize the space. Otherwise, if the new extents are being
3629 * inserted into the middle of the existing entries, a bit more work
3630 * is required to make room for the new extents to be inserted. The
3631 * caller is responsible for filling in the new extent entries upon
3632 * return.
3633 */
3634void
3635xfs_iext_add(
3636 xfs_ifork_t *ifp, /* inode fork pointer */
3637 xfs_extnum_t idx, /* index to begin adding exts */
Nathan Scottc41564b2006-03-29 08:55:14 +10003638 int ext_diff) /* number of extents to add */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003639{
3640 int byte_diff; /* new bytes being added */
3641 int new_size; /* size of extents after adding */
3642 xfs_extnum_t nextents; /* number of extents in file */
3643
3644 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3645 ASSERT((idx >= 0) && (idx <= nextents));
3646 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3647 new_size = ifp->if_bytes + byte_diff;
3648 /*
3649 * If the new number of extents (nextents + ext_diff)
3650 * fits inside the inode, then continue to use the inline
3651 * extent buffer.
3652 */
3653 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3654 if (idx < nextents) {
3655 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3656 &ifp->if_u2.if_inline_ext[idx],
3657 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3658 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3659 }
3660 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3661 ifp->if_real_bytes = 0;
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003662 ifp->if_lastex = nextents + ext_diff;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003663 }
3664 /*
3665 * Otherwise use a linear (direct) extent list.
3666 * If the extents are currently inside the inode,
3667 * xfs_iext_realloc_direct will switch us from
3668 * inline to direct extent allocation mode.
3669 */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003670 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003671 xfs_iext_realloc_direct(ifp, new_size);
3672 if (idx < nextents) {
3673 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3674 &ifp->if_u1.if_extents[idx],
3675 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3676 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3677 }
3678 }
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003679 /* Indirection array */
3680 else {
3681 xfs_ext_irec_t *erp;
3682 int erp_idx = 0;
3683 int page_idx = idx;
3684
3685 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3686 if (ifp->if_flags & XFS_IFEXTIREC) {
3687 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3688 } else {
3689 xfs_iext_irec_init(ifp);
3690 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3691 erp = ifp->if_u1.if_ext_irec;
3692 }
3693 /* Extents fit in target extent page */
3694 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3695 if (page_idx < erp->er_extcount) {
3696 memmove(&erp->er_extbuf[page_idx + ext_diff],
3697 &erp->er_extbuf[page_idx],
3698 (erp->er_extcount - page_idx) *
3699 sizeof(xfs_bmbt_rec_t));
3700 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3701 }
3702 erp->er_extcount += ext_diff;
3703 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3704 }
3705 /* Insert a new extent page */
3706 else if (erp) {
3707 xfs_iext_add_indirect_multi(ifp,
3708 erp_idx, page_idx, ext_diff);
3709 }
3710 /*
3711 * If extent(s) are being appended to the last page in
3712 * the indirection array and the new extent(s) don't fit
3713 * in the page, then erp is NULL and erp_idx is set to
3714 * the next index needed in the indirection array.
3715 */
3716 else {
3717 int count = ext_diff;
3718
3719 while (count) {
3720 erp = xfs_iext_irec_new(ifp, erp_idx);
3721 erp->er_extcount = count;
3722 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3723 if (count) {
3724 erp_idx++;
3725 }
3726 }
3727 }
3728 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003729 ifp->if_bytes = new_size;
3730}
3731
3732/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003733 * This is called when incore extents are being added to the indirection
3734 * array and the new extents do not fit in the target extent list. The
3735 * erp_idx parameter contains the irec index for the target extent list
3736 * in the indirection array, and the idx parameter contains the extent
3737 * index within the list. The number of extents being added is stored
3738 * in the count parameter.
3739 *
3740 * |-------| |-------|
3741 * | | | | idx - number of extents before idx
3742 * | idx | | count |
3743 * | | | | count - number of extents being inserted at idx
3744 * |-------| |-------|
3745 * | count | | nex2 | nex2 - number of extents after idx + count
3746 * |-------| |-------|
3747 */
3748void
3749xfs_iext_add_indirect_multi(
3750 xfs_ifork_t *ifp, /* inode fork pointer */
3751 int erp_idx, /* target extent irec index */
3752 xfs_extnum_t idx, /* index within target list */
3753 int count) /* new extents being added */
3754{
3755 int byte_diff; /* new bytes being added */
3756 xfs_ext_irec_t *erp; /* pointer to irec entry */
3757 xfs_extnum_t ext_diff; /* number of extents to add */
3758 xfs_extnum_t ext_cnt; /* new extents still needed */
3759 xfs_extnum_t nex2; /* extents after idx + count */
3760 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3761 int nlists; /* number of irec's (lists) */
3762
3763 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3764 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3765 nex2 = erp->er_extcount - idx;
3766 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3767
3768 /*
3769 * Save second part of target extent list
3770 * (all extents past */
3771 if (nex2) {
3772 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3773 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3774 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3775 erp->er_extcount -= nex2;
3776 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3777 memset(&erp->er_extbuf[idx], 0, byte_diff);
3778 }
3779
3780 /*
3781 * Add the new extents to the end of the target
3782 * list, then allocate new irec record(s) and
3783 * extent buffer(s) as needed to store the rest
3784 * of the new extents.
3785 */
3786 ext_cnt = count;
3787 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3788 if (ext_diff) {
3789 erp->er_extcount += ext_diff;
3790 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3791 ext_cnt -= ext_diff;
3792 }
3793 while (ext_cnt) {
3794 erp_idx++;
3795 erp = xfs_iext_irec_new(ifp, erp_idx);
3796 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3797 erp->er_extcount = ext_diff;
3798 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3799 ext_cnt -= ext_diff;
3800 }
3801
3802 /* Add nex2 extents back to indirection array */
3803 if (nex2) {
3804 xfs_extnum_t ext_avail;
3805 int i;
3806
3807 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3808 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3809 i = 0;
3810 /*
3811 * If nex2 extents fit in the current page, append
3812 * nex2_ep after the new extents.
3813 */
3814 if (nex2 <= ext_avail) {
3815 i = erp->er_extcount;
3816 }
3817 /*
3818 * Otherwise, check if space is available in the
3819 * next page.
3820 */
3821 else if ((erp_idx < nlists - 1) &&
3822 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3823 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3824 erp_idx++;
3825 erp++;
3826 /* Create a hole for nex2 extents */
3827 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3828 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3829 }
3830 /*
3831 * Final choice, create a new extent page for
3832 * nex2 extents.
3833 */
3834 else {
3835 erp_idx++;
3836 erp = xfs_iext_irec_new(ifp, erp_idx);
3837 }
3838 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3839 kmem_free(nex2_ep, byte_diff);
3840 erp->er_extcount += nex2;
3841 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3842 }
3843}
3844
3845/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003846 * This is called when the amount of space required for incore file
3847 * extents needs to be decreased. The ext_diff parameter stores the
3848 * number of extents to be removed and the idx parameter contains
3849 * the extent index where the extents will be removed from.
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003850 *
3851 * If the amount of space needed has decreased below the linear
3852 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3853 * extent array. Otherwise, use kmem_realloc() to adjust the
3854 * size to what is needed.
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003855 */
3856void
3857xfs_iext_remove(
3858 xfs_ifork_t *ifp, /* inode fork pointer */
3859 xfs_extnum_t idx, /* index to begin removing exts */
3860 int ext_diff) /* number of extents to remove */
3861{
3862 xfs_extnum_t nextents; /* number of extents in file */
3863 int new_size; /* size of extents after removal */
3864
3865 ASSERT(ext_diff > 0);
3866 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3867 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3868
3869 if (new_size == 0) {
3870 xfs_iext_destroy(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003871 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3872 xfs_iext_remove_indirect(ifp, idx, ext_diff);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003873 } else if (ifp->if_real_bytes) {
3874 xfs_iext_remove_direct(ifp, idx, ext_diff);
3875 } else {
3876 xfs_iext_remove_inline(ifp, idx, ext_diff);
3877 }
3878 ifp->if_bytes = new_size;
3879}
3880
3881/*
3882 * This removes ext_diff extents from the inline buffer, beginning
3883 * at extent index idx.
3884 */
3885void
3886xfs_iext_remove_inline(
3887 xfs_ifork_t *ifp, /* inode fork pointer */
3888 xfs_extnum_t idx, /* index to begin removing exts */
3889 int ext_diff) /* number of extents to remove */
3890{
3891 int nextents; /* number of extents in file */
3892
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003893 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003894 ASSERT(idx < XFS_INLINE_EXTS);
3895 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3896 ASSERT(((nextents - ext_diff) > 0) &&
3897 (nextents - ext_diff) < XFS_INLINE_EXTS);
3898
3899 if (idx + ext_diff < nextents) {
3900 memmove(&ifp->if_u2.if_inline_ext[idx],
3901 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3902 (nextents - (idx + ext_diff)) *
3903 sizeof(xfs_bmbt_rec_t));
3904 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3905 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3906 } else {
3907 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3908 ext_diff * sizeof(xfs_bmbt_rec_t));
3909 }
3910}
3911
3912/*
3913 * This removes ext_diff extents from a linear (direct) extent list,
3914 * beginning at extent index idx. If the extents are being removed
3915 * from the end of the list (ie. truncate) then we just need to re-
3916 * allocate the list to remove the extra space. Otherwise, if the
3917 * extents are being removed from the middle of the existing extent
3918 * entries, then we first need to move the extent records beginning
3919 * at idx + ext_diff up in the list to overwrite the records being
3920 * removed, then remove the extra space via kmem_realloc.
3921 */
3922void
3923xfs_iext_remove_direct(
3924 xfs_ifork_t *ifp, /* inode fork pointer */
3925 xfs_extnum_t idx, /* index to begin removing exts */
3926 int ext_diff) /* number of extents to remove */
3927{
3928 xfs_extnum_t nextents; /* number of extents in file */
3929 int new_size; /* size of extents after removal */
3930
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003931 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003932 new_size = ifp->if_bytes -
3933 (ext_diff * sizeof(xfs_bmbt_rec_t));
3934 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3935
3936 if (new_size == 0) {
3937 xfs_iext_destroy(ifp);
3938 return;
3939 }
3940 /* Move extents up in the list (if needed) */
3941 if (idx + ext_diff < nextents) {
3942 memmove(&ifp->if_u1.if_extents[idx],
3943 &ifp->if_u1.if_extents[idx + ext_diff],
3944 (nextents - (idx + ext_diff)) *
3945 sizeof(xfs_bmbt_rec_t));
3946 }
3947 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3948 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3949 /*
3950 * Reallocate the direct extent list. If the extents
3951 * will fit inside the inode then xfs_iext_realloc_direct
3952 * will switch from direct to inline extent allocation
3953 * mode for us.
3954 */
3955 xfs_iext_realloc_direct(ifp, new_size);
3956 ifp->if_bytes = new_size;
3957}
3958
3959/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003960 * This is called when incore extents are being removed from the
3961 * indirection array and the extents being removed span multiple extent
3962 * buffers. The idx parameter contains the file extent index where we
3963 * want to begin removing extents, and the count parameter contains
3964 * how many extents need to be removed.
3965 *
3966 * |-------| |-------|
3967 * | nex1 | | | nex1 - number of extents before idx
3968 * |-------| | count |
3969 * | | | | count - number of extents being removed at idx
3970 * | count | |-------|
3971 * | | | nex2 | nex2 - number of extents after idx + count
3972 * |-------| |-------|
3973 */
3974void
3975xfs_iext_remove_indirect(
3976 xfs_ifork_t *ifp, /* inode fork pointer */
3977 xfs_extnum_t idx, /* index to begin removing extents */
3978 int count) /* number of extents to remove */
3979{
3980 xfs_ext_irec_t *erp; /* indirection array pointer */
3981 int erp_idx = 0; /* indirection array index */
3982 xfs_extnum_t ext_cnt; /* extents left to remove */
3983 xfs_extnum_t ext_diff; /* extents to remove in current list */
3984 xfs_extnum_t nex1; /* number of extents before idx */
3985 xfs_extnum_t nex2; /* extents after idx + count */
Nathan Scottc41564b2006-03-29 08:55:14 +10003986 int nlists; /* entries in indirection array */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003987 int page_idx = idx; /* index in target extent list */
3988
3989 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3990 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3991 ASSERT(erp != NULL);
3992 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3993 nex1 = page_idx;
3994 ext_cnt = count;
3995 while (ext_cnt) {
3996 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3997 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3998 /*
3999 * Check for deletion of entire list;
4000 * xfs_iext_irec_remove() updates extent offsets.
4001 */
4002 if (ext_diff == erp->er_extcount) {
4003 xfs_iext_irec_remove(ifp, erp_idx);
4004 ext_cnt -= ext_diff;
4005 nex1 = 0;
4006 if (ext_cnt) {
4007 ASSERT(erp_idx < ifp->if_real_bytes /
4008 XFS_IEXT_BUFSZ);
4009 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4010 nex1 = 0;
4011 continue;
4012 } else {
4013 break;
4014 }
4015 }
4016 /* Move extents up (if needed) */
4017 if (nex2) {
4018 memmove(&erp->er_extbuf[nex1],
4019 &erp->er_extbuf[nex1 + ext_diff],
4020 nex2 * sizeof(xfs_bmbt_rec_t));
4021 }
4022 /* Zero out rest of page */
4023 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4024 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4025 /* Update remaining counters */
4026 erp->er_extcount -= ext_diff;
4027 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4028 ext_cnt -= ext_diff;
4029 nex1 = 0;
4030 erp_idx++;
4031 erp++;
4032 }
4033 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4034 xfs_iext_irec_compact(ifp);
4035}
4036
4037/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004038 * Create, destroy, or resize a linear (direct) block of extents.
4039 */
4040void
4041xfs_iext_realloc_direct(
4042 xfs_ifork_t *ifp, /* inode fork pointer */
4043 int new_size) /* new size of extents */
4044{
4045 int rnew_size; /* real new size of extents */
4046
4047 rnew_size = new_size;
4048
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004049 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4050 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4051 (new_size != ifp->if_real_bytes)));
4052
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004053 /* Free extent records */
4054 if (new_size == 0) {
4055 xfs_iext_destroy(ifp);
4056 }
4057 /* Resize direct extent list and zero any new bytes */
4058 else if (ifp->if_real_bytes) {
4059 /* Check if extents will fit inside the inode */
4060 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4061 xfs_iext_direct_to_inline(ifp, new_size /
4062 (uint)sizeof(xfs_bmbt_rec_t));
4063 ifp->if_bytes = new_size;
4064 return;
4065 }
Vignesh Babu16a087d2007-06-28 16:46:37 +10004066 if (!is_power_of_2(new_size)){
Robert P. J. Day40ebd812007-11-23 16:30:51 +11004067 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004068 }
4069 if (rnew_size != ifp->if_real_bytes) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10004070 ifp->if_u1.if_extents =
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004071 kmem_realloc(ifp->if_u1.if_extents,
4072 rnew_size,
4073 ifp->if_real_bytes,
4074 KM_SLEEP);
4075 }
4076 if (rnew_size > ifp->if_real_bytes) {
4077 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4078 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4079 rnew_size - ifp->if_real_bytes);
4080 }
4081 }
4082 /*
4083 * Switch from the inline extent buffer to a direct
4084 * extent list. Be sure to include the inline extent
4085 * bytes in new_size.
4086 */
4087 else {
4088 new_size += ifp->if_bytes;
Vignesh Babu16a087d2007-06-28 16:46:37 +10004089 if (!is_power_of_2(new_size)) {
Robert P. J. Day40ebd812007-11-23 16:30:51 +11004090 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004091 }
4092 xfs_iext_inline_to_direct(ifp, rnew_size);
4093 }
4094 ifp->if_real_bytes = rnew_size;
4095 ifp->if_bytes = new_size;
4096}
4097
4098/*
4099 * Switch from linear (direct) extent records to inline buffer.
4100 */
4101void
4102xfs_iext_direct_to_inline(
4103 xfs_ifork_t *ifp, /* inode fork pointer */
4104 xfs_extnum_t nextents) /* number of extents in file */
4105{
4106 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4107 ASSERT(nextents <= XFS_INLINE_EXTS);
4108 /*
4109 * The inline buffer was zeroed when we switched
4110 * from inline to direct extent allocation mode,
4111 * so we don't need to clear it here.
4112 */
4113 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4114 nextents * sizeof(xfs_bmbt_rec_t));
Mandy Kirkconnellfe6c1e72006-06-09 14:51:25 +10004115 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004116 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4117 ifp->if_real_bytes = 0;
4118}
4119
4120/*
4121 * Switch from inline buffer to linear (direct) extent records.
4122 * new_size should already be rounded up to the next power of 2
4123 * by the caller (when appropriate), so use new_size as it is.
4124 * However, since new_size may be rounded up, we can't update
4125 * if_bytes here. It is the caller's responsibility to update
4126 * if_bytes upon return.
4127 */
4128void
4129xfs_iext_inline_to_direct(
4130 xfs_ifork_t *ifp, /* inode fork pointer */
4131 int new_size) /* number of extents in file */
4132{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10004133 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_SLEEP);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004134 memset(ifp->if_u1.if_extents, 0, new_size);
4135 if (ifp->if_bytes) {
4136 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4137 ifp->if_bytes);
4138 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4139 sizeof(xfs_bmbt_rec_t));
4140 }
4141 ifp->if_real_bytes = new_size;
4142}
4143
4144/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004145 * Resize an extent indirection array to new_size bytes.
4146 */
4147void
4148xfs_iext_realloc_indirect(
4149 xfs_ifork_t *ifp, /* inode fork pointer */
4150 int new_size) /* new indirection array size */
4151{
4152 int nlists; /* number of irec's (ex lists) */
4153 int size; /* current indirection array size */
4154
4155 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4156 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4157 size = nlists * sizeof(xfs_ext_irec_t);
4158 ASSERT(ifp->if_real_bytes);
4159 ASSERT((new_size >= 0) && (new_size != size));
4160 if (new_size == 0) {
4161 xfs_iext_destroy(ifp);
4162 } else {
4163 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4164 kmem_realloc(ifp->if_u1.if_ext_irec,
4165 new_size, size, KM_SLEEP);
4166 }
4167}
4168
4169/*
4170 * Switch from indirection array to linear (direct) extent allocations.
4171 */
4172void
4173xfs_iext_indirect_to_direct(
4174 xfs_ifork_t *ifp) /* inode fork pointer */
4175{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10004176 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004177 xfs_extnum_t nextents; /* number of extents in file */
4178 int size; /* size of file extents */
4179
4180 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4181 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4182 ASSERT(nextents <= XFS_LINEAR_EXTS);
4183 size = nextents * sizeof(xfs_bmbt_rec_t);
4184
4185 xfs_iext_irec_compact_full(ifp);
4186 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4187
4188 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4189 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4190 ifp->if_flags &= ~XFS_IFEXTIREC;
4191 ifp->if_u1.if_extents = ep;
4192 ifp->if_bytes = size;
4193 if (nextents < XFS_LINEAR_EXTS) {
4194 xfs_iext_realloc_direct(ifp, size);
4195 }
4196}
4197
4198/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004199 * Free incore file extents.
4200 */
4201void
4202xfs_iext_destroy(
4203 xfs_ifork_t *ifp) /* inode fork pointer */
4204{
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004205 if (ifp->if_flags & XFS_IFEXTIREC) {
4206 int erp_idx;
4207 int nlists;
4208
4209 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4210 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4211 xfs_iext_irec_remove(ifp, erp_idx);
4212 }
4213 ifp->if_flags &= ~XFS_IFEXTIREC;
4214 } else if (ifp->if_real_bytes) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004215 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4216 } else if (ifp->if_bytes) {
4217 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4218 sizeof(xfs_bmbt_rec_t));
4219 }
4220 ifp->if_u1.if_extents = NULL;
4221 ifp->if_real_bytes = 0;
4222 ifp->if_bytes = 0;
4223}
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004224
4225/*
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004226 * Return a pointer to the extent record for file system block bno.
4227 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10004228xfs_bmbt_rec_host_t * /* pointer to found extent record */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004229xfs_iext_bno_to_ext(
4230 xfs_ifork_t *ifp, /* inode fork pointer */
4231 xfs_fileoff_t bno, /* block number to search for */
4232 xfs_extnum_t *idxp) /* index of target extent */
4233{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10004234 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004235 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10004236 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004237 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
Nathan Scottc41564b2006-03-29 08:55:14 +10004238 int high; /* upper boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004239 xfs_extnum_t idx = 0; /* index of target extent */
Nathan Scottc41564b2006-03-29 08:55:14 +10004240 int low; /* lower boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004241 xfs_extnum_t nextents; /* number of file extents */
4242 xfs_fileoff_t startoff = 0; /* start offset of extent */
4243
4244 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4245 if (nextents == 0) {
4246 *idxp = 0;
4247 return NULL;
4248 }
4249 low = 0;
4250 if (ifp->if_flags & XFS_IFEXTIREC) {
4251 /* Find target extent list */
4252 int erp_idx = 0;
4253 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4254 base = erp->er_extbuf;
4255 high = erp->er_extcount - 1;
4256 } else {
4257 base = ifp->if_u1.if_extents;
4258 high = nextents - 1;
4259 }
4260 /* Binary search extent records */
4261 while (low <= high) {
4262 idx = (low + high) >> 1;
4263 ep = base + idx;
4264 startoff = xfs_bmbt_get_startoff(ep);
4265 blockcount = xfs_bmbt_get_blockcount(ep);
4266 if (bno < startoff) {
4267 high = idx - 1;
4268 } else if (bno >= startoff + blockcount) {
4269 low = idx + 1;
4270 } else {
4271 /* Convert back to file-based extent index */
4272 if (ifp->if_flags & XFS_IFEXTIREC) {
4273 idx += erp->er_extoff;
4274 }
4275 *idxp = idx;
4276 return ep;
4277 }
4278 }
4279 /* Convert back to file-based extent index */
4280 if (ifp->if_flags & XFS_IFEXTIREC) {
4281 idx += erp->er_extoff;
4282 }
4283 if (bno >= startoff + blockcount) {
4284 if (++idx == nextents) {
4285 ep = NULL;
4286 } else {
4287 ep = xfs_iext_get_ext(ifp, idx);
4288 }
4289 }
4290 *idxp = idx;
4291 return ep;
4292}
4293
4294/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004295 * Return a pointer to the indirection array entry containing the
4296 * extent record for filesystem block bno. Store the index of the
4297 * target irec in *erp_idxp.
4298 */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004299xfs_ext_irec_t * /* pointer to found extent record */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004300xfs_iext_bno_to_irec(
4301 xfs_ifork_t *ifp, /* inode fork pointer */
4302 xfs_fileoff_t bno, /* block number to search for */
4303 int *erp_idxp) /* irec index of target ext list */
4304{
4305 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4306 xfs_ext_irec_t *erp_next; /* next indirection array entry */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004307 int erp_idx; /* indirection array index */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004308 int nlists; /* number of extent irec's (lists) */
4309 int high; /* binary search upper limit */
4310 int low; /* binary search lower limit */
4311
4312 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4313 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4314 erp_idx = 0;
4315 low = 0;
4316 high = nlists - 1;
4317 while (low <= high) {
4318 erp_idx = (low + high) >> 1;
4319 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4320 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4321 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4322 high = erp_idx - 1;
4323 } else if (erp_next && bno >=
4324 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4325 low = erp_idx + 1;
4326 } else {
4327 break;
4328 }
4329 }
4330 *erp_idxp = erp_idx;
4331 return erp;
4332}
4333
4334/*
4335 * Return a pointer to the indirection array entry containing the
4336 * extent record at file extent index *idxp. Store the index of the
4337 * target irec in *erp_idxp and store the page index of the target
4338 * extent record in *idxp.
4339 */
4340xfs_ext_irec_t *
4341xfs_iext_idx_to_irec(
4342 xfs_ifork_t *ifp, /* inode fork pointer */
4343 xfs_extnum_t *idxp, /* extent index (file -> page) */
4344 int *erp_idxp, /* pointer to target irec */
4345 int realloc) /* new bytes were just added */
4346{
4347 xfs_ext_irec_t *prev; /* pointer to previous irec */
4348 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4349 int erp_idx; /* indirection array index */
4350 int nlists; /* number of irec's (ex lists) */
4351 int high; /* binary search upper limit */
4352 int low; /* binary search lower limit */
4353 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4354
4355 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4356 ASSERT(page_idx >= 0 && page_idx <=
4357 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4358 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4359 erp_idx = 0;
4360 low = 0;
4361 high = nlists - 1;
4362
4363 /* Binary search extent irec's */
4364 while (low <= high) {
4365 erp_idx = (low + high) >> 1;
4366 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4367 prev = erp_idx > 0 ? erp - 1 : NULL;
4368 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4369 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4370 high = erp_idx - 1;
4371 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4372 (page_idx == erp->er_extoff + erp->er_extcount &&
4373 !realloc)) {
4374 low = erp_idx + 1;
4375 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4376 erp->er_extcount == XFS_LINEAR_EXTS) {
4377 ASSERT(realloc);
4378 page_idx = 0;
4379 erp_idx++;
4380 erp = erp_idx < nlists ? erp + 1 : NULL;
4381 break;
4382 } else {
4383 page_idx -= erp->er_extoff;
4384 break;
4385 }
4386 }
4387 *idxp = page_idx;
4388 *erp_idxp = erp_idx;
4389 return(erp);
4390}
4391
4392/*
4393 * Allocate and initialize an indirection array once the space needed
4394 * for incore extents increases above XFS_IEXT_BUFSZ.
4395 */
4396void
4397xfs_iext_irec_init(
4398 xfs_ifork_t *ifp) /* inode fork pointer */
4399{
4400 xfs_ext_irec_t *erp; /* indirection array pointer */
4401 xfs_extnum_t nextents; /* number of extents in file */
4402
4403 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4404 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4405 ASSERT(nextents <= XFS_LINEAR_EXTS);
4406
4407 erp = (xfs_ext_irec_t *)
4408 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4409
4410 if (nextents == 0) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10004411 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004412 } else if (!ifp->if_real_bytes) {
4413 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4414 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4415 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4416 }
4417 erp->er_extbuf = ifp->if_u1.if_extents;
4418 erp->er_extcount = nextents;
4419 erp->er_extoff = 0;
4420
4421 ifp->if_flags |= XFS_IFEXTIREC;
4422 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4423 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4424 ifp->if_u1.if_ext_irec = erp;
4425
4426 return;
4427}
4428
4429/*
4430 * Allocate and initialize a new entry in the indirection array.
4431 */
4432xfs_ext_irec_t *
4433xfs_iext_irec_new(
4434 xfs_ifork_t *ifp, /* inode fork pointer */
4435 int erp_idx) /* index for new irec */
4436{
4437 xfs_ext_irec_t *erp; /* indirection array pointer */
4438 int i; /* loop counter */
4439 int nlists; /* number of irec's (ex lists) */
4440
4441 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4442 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4443
4444 /* Resize indirection array */
4445 xfs_iext_realloc_indirect(ifp, ++nlists *
4446 sizeof(xfs_ext_irec_t));
4447 /*
4448 * Move records down in the array so the
4449 * new page can use erp_idx.
4450 */
4451 erp = ifp->if_u1.if_ext_irec;
4452 for (i = nlists - 1; i > erp_idx; i--) {
4453 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4454 }
4455 ASSERT(i == erp_idx);
4456
4457 /* Initialize new extent record */
4458 erp = ifp->if_u1.if_ext_irec;
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10004459 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004460 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4461 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4462 erp[erp_idx].er_extcount = 0;
4463 erp[erp_idx].er_extoff = erp_idx > 0 ?
4464 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4465 return (&erp[erp_idx]);
4466}
4467
4468/*
4469 * Remove a record from the indirection array.
4470 */
4471void
4472xfs_iext_irec_remove(
4473 xfs_ifork_t *ifp, /* inode fork pointer */
4474 int erp_idx) /* irec index to remove */
4475{
4476 xfs_ext_irec_t *erp; /* indirection array pointer */
4477 int i; /* loop counter */
4478 int nlists; /* number of irec's (ex lists) */
4479
4480 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4481 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4482 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4483 if (erp->er_extbuf) {
4484 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4485 -erp->er_extcount);
4486 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4487 }
4488 /* Compact extent records */
4489 erp = ifp->if_u1.if_ext_irec;
4490 for (i = erp_idx; i < nlists - 1; i++) {
4491 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4492 }
4493 /*
4494 * Manually free the last extent record from the indirection
4495 * array. A call to xfs_iext_realloc_indirect() with a size
4496 * of zero would result in a call to xfs_iext_destroy() which
4497 * would in turn call this function again, creating a nasty
4498 * infinite loop.
4499 */
4500 if (--nlists) {
4501 xfs_iext_realloc_indirect(ifp,
4502 nlists * sizeof(xfs_ext_irec_t));
4503 } else {
4504 kmem_free(ifp->if_u1.if_ext_irec,
4505 sizeof(xfs_ext_irec_t));
4506 }
4507 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4508}
4509
4510/*
4511 * This is called to clean up large amounts of unused memory allocated
4512 * by the indirection array. Before compacting anything though, verify
4513 * that the indirection array is still needed and switch back to the
4514 * linear extent list (or even the inline buffer) if possible. The
4515 * compaction policy is as follows:
4516 *
4517 * Full Compaction: Extents fit into a single page (or inline buffer)
4518 * Full Compaction: Extents occupy less than 10% of allocated space
4519 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4520 * No Compaction: Extents occupy at least 50% of allocated space
4521 */
4522void
4523xfs_iext_irec_compact(
4524 xfs_ifork_t *ifp) /* inode fork pointer */
4525{
4526 xfs_extnum_t nextents; /* number of extents in file */
4527 int nlists; /* number of irec's (ex lists) */
4528
4529 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4530 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4531 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4532
4533 if (nextents == 0) {
4534 xfs_iext_destroy(ifp);
4535 } else if (nextents <= XFS_INLINE_EXTS) {
4536 xfs_iext_indirect_to_direct(ifp);
4537 xfs_iext_direct_to_inline(ifp, nextents);
4538 } else if (nextents <= XFS_LINEAR_EXTS) {
4539 xfs_iext_indirect_to_direct(ifp);
4540 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4541 xfs_iext_irec_compact_full(ifp);
4542 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4543 xfs_iext_irec_compact_pages(ifp);
4544 }
4545}
4546
4547/*
4548 * Combine extents from neighboring extent pages.
4549 */
4550void
4551xfs_iext_irec_compact_pages(
4552 xfs_ifork_t *ifp) /* inode fork pointer */
4553{
4554 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4555 int erp_idx = 0; /* indirection array index */
4556 int nlists; /* number of irec's (ex lists) */
4557
4558 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4559 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4560 while (erp_idx < nlists - 1) {
4561 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4562 erp_next = erp + 1;
4563 if (erp_next->er_extcount <=
4564 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4565 memmove(&erp->er_extbuf[erp->er_extcount],
4566 erp_next->er_extbuf, erp_next->er_extcount *
4567 sizeof(xfs_bmbt_rec_t));
4568 erp->er_extcount += erp_next->er_extcount;
4569 /*
4570 * Free page before removing extent record
4571 * so er_extoffs don't get modified in
4572 * xfs_iext_irec_remove.
4573 */
4574 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4575 erp_next->er_extbuf = NULL;
4576 xfs_iext_irec_remove(ifp, erp_idx + 1);
4577 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4578 } else {
4579 erp_idx++;
4580 }
4581 }
4582}
4583
4584/*
4585 * Fully compact the extent records managed by the indirection array.
4586 */
4587void
4588xfs_iext_irec_compact_full(
4589 xfs_ifork_t *ifp) /* inode fork pointer */
4590{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10004591 xfs_bmbt_rec_host_t *ep, *ep_next; /* extent record pointers */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004592 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4593 int erp_idx = 0; /* extent irec index */
4594 int ext_avail; /* empty entries in ex list */
4595 int ext_diff; /* number of exts to add */
4596 int nlists; /* number of irec's (ex lists) */
4597
4598 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4599 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4600 erp = ifp->if_u1.if_ext_irec;
4601 ep = &erp->er_extbuf[erp->er_extcount];
4602 erp_next = erp + 1;
4603 ep_next = erp_next->er_extbuf;
4604 while (erp_idx < nlists - 1) {
4605 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4606 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4607 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4608 erp->er_extcount += ext_diff;
4609 erp_next->er_extcount -= ext_diff;
4610 /* Remove next page */
4611 if (erp_next->er_extcount == 0) {
4612 /*
4613 * Free page before removing extent record
4614 * so er_extoffs don't get modified in
4615 * xfs_iext_irec_remove.
4616 */
4617 kmem_free(erp_next->er_extbuf,
4618 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4619 erp_next->er_extbuf = NULL;
4620 xfs_iext_irec_remove(ifp, erp_idx + 1);
4621 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4622 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4623 /* Update next page */
4624 } else {
4625 /* Move rest of page up to become next new page */
4626 memmove(erp_next->er_extbuf, ep_next,
4627 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4628 ep_next = erp_next->er_extbuf;
4629 memset(&ep_next[erp_next->er_extcount], 0,
4630 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4631 sizeof(xfs_bmbt_rec_t));
4632 }
4633 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4634 erp_idx++;
4635 if (erp_idx < nlists)
4636 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4637 else
4638 break;
4639 }
4640 ep = &erp->er_extbuf[erp->er_extcount];
4641 erp_next = erp + 1;
4642 ep_next = erp_next->er_extbuf;
4643 }
4644}
4645
4646/*
4647 * This is called to update the er_extoff field in the indirection
4648 * array when extents have been added or removed from one of the
4649 * extent lists. erp_idx contains the irec index to begin updating
4650 * at and ext_diff contains the number of extents that were added
4651 * or removed.
4652 */
4653void
4654xfs_iext_irec_update_extoffs(
4655 xfs_ifork_t *ifp, /* inode fork pointer */
4656 int erp_idx, /* irec index to update */
4657 int ext_diff) /* number of new extents */
4658{
4659 int i; /* loop counter */
4660 int nlists; /* number of irec's (ex lists */
4661
4662 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4663 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4664 for (i = erp_idx; i < nlists; i++) {
4665 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
4666 }
4667}