blob: 5a5e035e5d388a0fa844c9c66a47cf92dd2db1e7 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_trans.h"
27#include "xfs_trans_priv.h"
28#include "xfs_sb.h"
29#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_dir2.h"
31#include "xfs_dmapi.h"
32#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110037#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110041#include "xfs_inode_item.h"
42#include "xfs_btree.h"
Christoph Hellwig8c4ed632008-10-30 16:55:13 +110043#include "xfs_btree_trace.h"
Nathan Scotta844f452005-11-02 14:38:42 +110044#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,
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100136 struct xfs_imap *imap,
David Chinner4ae29b42008-03-06 13:43:34 +1100137 xfs_buf_t **bpp,
138 uint buf_flags,
Christoph Hellwigb48d8d62008-11-28 14:23:41 +1100139 uint iget_flags)
David Chinner4ae29b42008-03-06 13:43:34 +1100140{
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));
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100176 di_ok = be16_to_cpu(dip->di_magic) == XFS_DINODE_MAGIC &&
177 XFS_DINODE_GOOD_VERSION(dip->di_version);
David Chinner4ae29b42008-03-06 13:43:34 +1100178 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
179 XFS_ERRTAG_ITOBP_INOTOBP,
180 XFS_RANDOM_ITOBP_INOTOBP))) {
Christoph Hellwigb48d8d62008-11-28 14:23:41 +1100181 if (iget_flags & XFS_IGET_BULKSTAT) {
David Chinner4ae29b42008-03-06 13:43:34 +1100182 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,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100193 be16_to_cpu(dip->di_magic));
David Chinner4ae29b42008-03-06 13:43:34 +1100194#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 Hellwigc679eef2008-10-30 18:04:13 +1100224int
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,
Christoph Hellwigc679eef2008-10-30 18:04:13 +1100231 int *offset,
232 uint imap_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100234 struct xfs_imap imap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 xfs_buf_t *bp;
236 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 imap.im_blkno = 0;
Christoph Hellwiga1941892008-11-28 14:23:40 +1100239 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
David Chinner4ae29b42008-03-06 13:43:34 +1100240 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
Christoph Hellwigc679eef2008-10-30 18:04:13 +1100243 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, imap_flags);
David Chinner4ae29b42008-03-06 13:43:34 +1100244 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
248 *bpp = bp;
249 *offset = imap.im_boffset;
250 return 0;
251}
252
253
254/*
255 * This routine is called to map an inode to the buffer containing
256 * the on-disk version of the inode. It returns a pointer to the
257 * buffer containing the on-disk inode in the bpp parameter, and in
258 * the dip parameter it returns a pointer to the on-disk inode within
259 * that buffer.
260 *
261 * If a non-zero error is returned, then the contents of bpp and
262 * dipp are undefined.
263 *
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100264 * The inode is expected to already been mapped to its buffer and read
265 * in once, thus we can use the mapping information stored in the inode
266 * rather than calling xfs_imap(). This allows us to avoid the overhead
267 * of looking at the inode btree for small block file systems
Christoph Hellwig94e1b692008-11-28 14:23:41 +1100268 * (see xfs_imap()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 */
270int
271xfs_itobp(
272 xfs_mount_t *mp,
273 xfs_trans_t *tp,
274 xfs_inode_t *ip,
275 xfs_dinode_t **dipp,
276 xfs_buf_t **bpp,
David Chinnera3f74ff2008-03-06 13:43:42 +1100277 uint buf_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278{
279 xfs_buf_t *bp;
280 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100282 ASSERT(ip->i_imap.im_blkno != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100284 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
David Chinner4ae29b42008-03-06 13:43:34 +1100285 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 return error;
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000287
David Chinnera3f74ff2008-03-06 13:43:42 +1100288 if (!bp) {
289 ASSERT(buf_flags & XFS_BUF_TRYLOCK);
290 ASSERT(tp == NULL);
291 *bpp = NULL;
292 return EAGAIN;
293 }
294
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100295 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 *bpp = bp;
297 return 0;
298}
299
300/*
301 * Move inode type and inode format specific information from the
302 * on-disk inode to the in-core inode. For fifos, devs, and sockets
303 * this means set if_rdev to the proper value. For files, directories,
304 * and symlinks this means to bring in the in-line data or extent
305 * pointers. For a file in B-tree format, only the root is immediately
306 * brought in-core. The rest will be in-lined in if_extents when it
307 * is first referenced (see xfs_iread_extents()).
308 */
309STATIC int
310xfs_iformat(
311 xfs_inode_t *ip,
312 xfs_dinode_t *dip)
313{
314 xfs_attr_shortform_t *atp;
315 int size;
316 int error;
317 xfs_fsize_t di_size;
318 ip->i_df.if_ext_max =
319 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
320 error = 0;
321
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100322 if (unlikely(be32_to_cpu(dip->di_nextents) +
323 be16_to_cpu(dip->di_anextents) >
324 be64_to_cpu(dip->di_nblocks))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100325 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
326 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 (unsigned long long)ip->i_ino,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100328 (int)(be32_to_cpu(dip->di_nextents) +
329 be16_to_cpu(dip->di_anextents)),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 (unsigned long long)
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100331 be64_to_cpu(dip->di_nblocks));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
333 ip->i_mount, dip);
334 return XFS_ERROR(EFSCORRUPTED);
335 }
336
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100337 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100338 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
339 "corrupt dinode %Lu, forkoff = 0x%x.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 (unsigned long long)ip->i_ino,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100341 dip->di_forkoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
343 ip->i_mount, dip);
344 return XFS_ERROR(EFSCORRUPTED);
345 }
346
347 switch (ip->i_d.di_mode & S_IFMT) {
348 case S_IFIFO:
349 case S_IFCHR:
350 case S_IFBLK:
351 case S_IFSOCK:
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100352 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
354 ip->i_mount, dip);
355 return XFS_ERROR(EFSCORRUPTED);
356 }
357 ip->i_d.di_size = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000358 ip->i_size = 0;
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100359 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 break;
361
362 case S_IFREG:
363 case S_IFLNK:
364 case S_IFDIR:
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100365 switch (dip->di_format) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 case XFS_DINODE_FMT_LOCAL:
367 /*
368 * no local regular files yet
369 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100370 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100371 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
372 "corrupt inode %Lu "
373 "(local format for regular file).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 (unsigned long long) ip->i_ino);
375 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
376 XFS_ERRLEVEL_LOW,
377 ip->i_mount, dip);
378 return XFS_ERROR(EFSCORRUPTED);
379 }
380
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100381 di_size = be64_to_cpu(dip->di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100383 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
384 "corrupt inode %Lu "
385 "(bad size %Ld for local inode).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 (unsigned long long) ip->i_ino,
387 (long long) di_size);
388 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
389 XFS_ERRLEVEL_LOW,
390 ip->i_mount, dip);
391 return XFS_ERROR(EFSCORRUPTED);
392 }
393
394 size = (int)di_size;
395 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
396 break;
397 case XFS_DINODE_FMT_EXTENTS:
398 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
399 break;
400 case XFS_DINODE_FMT_BTREE:
401 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
402 break;
403 default:
404 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
405 ip->i_mount);
406 return XFS_ERROR(EFSCORRUPTED);
407 }
408 break;
409
410 default:
411 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
412 return XFS_ERROR(EFSCORRUPTED);
413 }
414 if (error) {
415 return error;
416 }
417 if (!XFS_DFORK_Q(dip))
418 return 0;
419 ASSERT(ip->i_afp == NULL);
420 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
421 ip->i_afp->if_ext_max =
422 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100423 switch (dip->di_aformat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 case XFS_DINODE_FMT_LOCAL:
425 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
Nathan Scott3b244aa2006-03-17 17:29:25 +1100426 size = be16_to_cpu(atp->hdr.totsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
428 break;
429 case XFS_DINODE_FMT_EXTENTS:
430 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
431 break;
432 case XFS_DINODE_FMT_BTREE:
433 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
434 break;
435 default:
436 error = XFS_ERROR(EFSCORRUPTED);
437 break;
438 }
439 if (error) {
440 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
441 ip->i_afp = NULL;
442 xfs_idestroy_fork(ip, XFS_DATA_FORK);
443 }
444 return error;
445}
446
447/*
448 * The file is in-lined in the on-disk inode.
449 * If it fits into if_inline_data, then copy
450 * it there, otherwise allocate a buffer for it
451 * and copy the data there. Either way, set
452 * if_data to point at the data.
453 * If we allocate a buffer for the data, make
454 * sure that its size is a multiple of 4 and
455 * record the real size in i_real_bytes.
456 */
457STATIC int
458xfs_iformat_local(
459 xfs_inode_t *ip,
460 xfs_dinode_t *dip,
461 int whichfork,
462 int size)
463{
464 xfs_ifork_t *ifp;
465 int real_size;
466
467 /*
468 * If the size is unreasonable, then something
469 * is wrong and we just bail out rather than crash in
470 * kmem_alloc() or memcpy() below.
471 */
472 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100473 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
474 "corrupt inode %Lu "
475 "(bad size %d for local fork, size = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 (unsigned long long) ip->i_ino, size,
477 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
478 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
479 ip->i_mount, dip);
480 return XFS_ERROR(EFSCORRUPTED);
481 }
482 ifp = XFS_IFORK_PTR(ip, whichfork);
483 real_size = 0;
484 if (size == 0)
485 ifp->if_u1.if_data = NULL;
486 else if (size <= sizeof(ifp->if_u2.if_inline_data))
487 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
488 else {
489 real_size = roundup(size, 4);
490 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
491 }
492 ifp->if_bytes = size;
493 ifp->if_real_bytes = real_size;
494 if (size)
495 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
496 ifp->if_flags &= ~XFS_IFEXTENTS;
497 ifp->if_flags |= XFS_IFINLINE;
498 return 0;
499}
500
501/*
502 * The file consists of a set of extents all
503 * of which fit into the on-disk inode.
504 * If there are few enough extents to fit into
505 * the if_inline_ext, then copy them there.
506 * Otherwise allocate a buffer for them and copy
507 * them into it. Either way, set if_extents
508 * to point at the extents.
509 */
510STATIC int
511xfs_iformat_extents(
512 xfs_inode_t *ip,
513 xfs_dinode_t *dip,
514 int whichfork)
515{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000516 xfs_bmbt_rec_t *dp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 xfs_ifork_t *ifp;
518 int nex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 int size;
520 int i;
521
522 ifp = XFS_IFORK_PTR(ip, whichfork);
523 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
524 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
525
526 /*
527 * If the number of extents is unreasonable, then something
528 * is wrong and we just bail out rather than crash in
529 * kmem_alloc() or memcpy() below.
530 */
531 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100532 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
533 "corrupt inode %Lu ((a)extents = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 (unsigned long long) ip->i_ino, nex);
535 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
536 ip->i_mount, dip);
537 return XFS_ERROR(EFSCORRUPTED);
538 }
539
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100540 ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 if (nex == 0)
542 ifp->if_u1.if_extents = NULL;
543 else if (nex <= XFS_INLINE_EXTS)
544 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100545 else
546 xfs_iext_add(ifp, 0, nex);
547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 ifp->if_bytes = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 if (size) {
550 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000551 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100552 for (i = 0; i < nex; i++, dp++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000553 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Harvey Harrison597bca62008-08-13 16:29:21 +1000554 ep->l0 = get_unaligned_be64(&dp->l0);
555 ep->l1 = get_unaligned_be64(&dp->l1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 }
Eric Sandeen3a59c942007-07-11 11:09:47 +1000557 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 if (whichfork != XFS_DATA_FORK ||
559 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
560 if (unlikely(xfs_check_nostate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100561 ifp, 0, nex))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
563 XFS_ERRLEVEL_LOW,
564 ip->i_mount);
565 return XFS_ERROR(EFSCORRUPTED);
566 }
567 }
568 ifp->if_flags |= XFS_IFEXTENTS;
569 return 0;
570}
571
572/*
573 * The file has too many extents to fit into
574 * the inode, so they are in B-tree format.
575 * Allocate a buffer for the root of the B-tree
576 * and copy the root into it. The i_extents
577 * field will remain NULL until all of the
578 * extents are read in (when they are needed).
579 */
580STATIC int
581xfs_iformat_btree(
582 xfs_inode_t *ip,
583 xfs_dinode_t *dip,
584 int whichfork)
585{
586 xfs_bmdr_block_t *dfp;
587 xfs_ifork_t *ifp;
588 /* REFERENCED */
589 int nrecs;
590 int size;
591
592 ifp = XFS_IFORK_PTR(ip, whichfork);
593 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
594 size = XFS_BMAP_BROOT_SPACE(dfp);
Christoph Hellwig60197e82008-10-30 17:11:19 +1100595 nrecs = be16_to_cpu(dfp->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597 /*
598 * blow out if -- fork has less extents than can fit in
599 * fork (fork shouldn't be a btree format), root btree
600 * block has more records than can fit into the fork,
601 * or the number of extents is greater than the number of
602 * blocks.
603 */
604 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
605 || XFS_BMDR_SPACE_CALC(nrecs) >
606 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
607 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100608 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
609 "corrupt inode %Lu (btree).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 (unsigned long long) ip->i_ino);
611 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
612 ip->i_mount);
613 return XFS_ERROR(EFSCORRUPTED);
614 }
615
616 ifp->if_broot_bytes = size;
617 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
618 ASSERT(ifp->if_broot != NULL);
619 /*
620 * Copy and convert from the on-disk structure
621 * to the in-memory structure.
622 */
Christoph Hellwig60197e82008-10-30 17:11:19 +1100623 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
624 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
625 ifp->if_broot, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 ifp->if_flags &= ~XFS_IFEXTENTS;
627 ifp->if_flags |= XFS_IFBROOT;
628
629 return 0;
630}
631
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632void
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000633xfs_dinode_from_disk(
634 xfs_icdinode_t *to,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100635 xfs_dinode_t *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000637 to->di_magic = be16_to_cpu(from->di_magic);
638 to->di_mode = be16_to_cpu(from->di_mode);
639 to->di_version = from ->di_version;
640 to->di_format = from->di_format;
641 to->di_onlink = be16_to_cpu(from->di_onlink);
642 to->di_uid = be32_to_cpu(from->di_uid);
643 to->di_gid = be32_to_cpu(from->di_gid);
644 to->di_nlink = be32_to_cpu(from->di_nlink);
645 to->di_projid = be16_to_cpu(from->di_projid);
646 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
647 to->di_flushiter = be16_to_cpu(from->di_flushiter);
648 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
649 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
650 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
651 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
652 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
653 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
654 to->di_size = be64_to_cpu(from->di_size);
655 to->di_nblocks = be64_to_cpu(from->di_nblocks);
656 to->di_extsize = be32_to_cpu(from->di_extsize);
657 to->di_nextents = be32_to_cpu(from->di_nextents);
658 to->di_anextents = be16_to_cpu(from->di_anextents);
659 to->di_forkoff = from->di_forkoff;
660 to->di_aformat = from->di_aformat;
661 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
662 to->di_dmstate = be16_to_cpu(from->di_dmstate);
663 to->di_flags = be16_to_cpu(from->di_flags);
664 to->di_gen = be32_to_cpu(from->di_gen);
665}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000667void
668xfs_dinode_to_disk(
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100669 xfs_dinode_t *to,
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000670 xfs_icdinode_t *from)
671{
672 to->di_magic = cpu_to_be16(from->di_magic);
673 to->di_mode = cpu_to_be16(from->di_mode);
674 to->di_version = from ->di_version;
675 to->di_format = from->di_format;
676 to->di_onlink = cpu_to_be16(from->di_onlink);
677 to->di_uid = cpu_to_be32(from->di_uid);
678 to->di_gid = cpu_to_be32(from->di_gid);
679 to->di_nlink = cpu_to_be32(from->di_nlink);
680 to->di_projid = cpu_to_be16(from->di_projid);
681 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
682 to->di_flushiter = cpu_to_be16(from->di_flushiter);
683 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
684 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
685 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
686 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
687 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
688 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
689 to->di_size = cpu_to_be64(from->di_size);
690 to->di_nblocks = cpu_to_be64(from->di_nblocks);
691 to->di_extsize = cpu_to_be32(from->di_extsize);
692 to->di_nextents = cpu_to_be32(from->di_nextents);
693 to->di_anextents = cpu_to_be16(from->di_anextents);
694 to->di_forkoff = from->di_forkoff;
695 to->di_aformat = from->di_aformat;
696 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
697 to->di_dmstate = cpu_to_be16(from->di_dmstate);
698 to->di_flags = cpu_to_be16(from->di_flags);
699 to->di_gen = cpu_to_be32(from->di_gen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700}
701
702STATIC uint
703_xfs_dic2xflags(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 __uint16_t di_flags)
705{
706 uint flags = 0;
707
708 if (di_flags & XFS_DIFLAG_ANY) {
709 if (di_flags & XFS_DIFLAG_REALTIME)
710 flags |= XFS_XFLAG_REALTIME;
711 if (di_flags & XFS_DIFLAG_PREALLOC)
712 flags |= XFS_XFLAG_PREALLOC;
713 if (di_flags & XFS_DIFLAG_IMMUTABLE)
714 flags |= XFS_XFLAG_IMMUTABLE;
715 if (di_flags & XFS_DIFLAG_APPEND)
716 flags |= XFS_XFLAG_APPEND;
717 if (di_flags & XFS_DIFLAG_SYNC)
718 flags |= XFS_XFLAG_SYNC;
719 if (di_flags & XFS_DIFLAG_NOATIME)
720 flags |= XFS_XFLAG_NOATIME;
721 if (di_flags & XFS_DIFLAG_NODUMP)
722 flags |= XFS_XFLAG_NODUMP;
723 if (di_flags & XFS_DIFLAG_RTINHERIT)
724 flags |= XFS_XFLAG_RTINHERIT;
725 if (di_flags & XFS_DIFLAG_PROJINHERIT)
726 flags |= XFS_XFLAG_PROJINHERIT;
727 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
728 flags |= XFS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100729 if (di_flags & XFS_DIFLAG_EXTSIZE)
730 flags |= XFS_XFLAG_EXTSIZE;
731 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
732 flags |= XFS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000733 if (di_flags & XFS_DIFLAG_NODEFRAG)
734 flags |= XFS_XFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000735 if (di_flags & XFS_DIFLAG_FILESTREAM)
736 flags |= XFS_XFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 }
738
739 return flags;
740}
741
742uint
743xfs_ip2xflags(
744 xfs_inode_t *ip)
745{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000746 xfs_icdinode_t *dic = &ip->i_d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Nathan Scotta916e2b2006-06-09 17:12:17 +1000748 return _xfs_dic2xflags(dic->di_flags) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100749 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}
751
752uint
753xfs_dic2xflags(
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100754 xfs_dinode_t *dip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755{
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100756 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100757 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758}
759
760/*
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100761 * Read the disk inode attributes into the in-core inode structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 */
763int
764xfs_iread(
765 xfs_mount_t *mp,
766 xfs_trans_t *tp,
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100767 xfs_inode_t *ip,
Nathan Scott745b1f472006-09-28 11:02:23 +1000768 xfs_daddr_t bno,
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100769 uint iget_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
771 xfs_buf_t *bp;
772 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 int error;
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 /*
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100776 * Fill in the location information in the in-core inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100778 ip->i_imap.im_blkno = bno;
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100779 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100780 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100781 return error;
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100782 ASSERT(bno == 0 || bno == ip->i_imap.im_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
784 /*
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100785 * Get pointers to the on-disk inode and the buffer containing it.
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100786 */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100787 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100788 XFS_BUF_LOCK, iget_flags);
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100789 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100790 return error;
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100791 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100792
793 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 * If we got something that isn't an inode it means someone
795 * (nfs or dmi) has a stale handle.
796 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100797 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798#ifdef DEBUG
799 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100800 "dip->di_magic (0x%x) != "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 "XFS_DINODE_MAGIC (0x%x)",
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100802 be16_to_cpu(dip->di_magic),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 XFS_DINODE_MAGIC);
804#endif /* DEBUG */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100805 error = XFS_ERROR(EINVAL);
806 goto out_brelse;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 }
808
809 /*
810 * If the on-disk inode is already linked to a directory
811 * entry, copy all of the inode into the in-core inode.
812 * xfs_iformat() handles copying in the inode format
813 * specific information.
814 * Otherwise, just get the truly permanent information.
815 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100816 if (dip->di_mode) {
817 xfs_dinode_from_disk(&ip->i_d, dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 error = xfs_iformat(ip, dip);
819 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820#ifdef DEBUG
821 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
822 "xfs_iformat() returned error %d",
823 error);
824#endif /* DEBUG */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100825 goto out_brelse;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 }
827 } else {
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100828 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
829 ip->i_d.di_version = dip->di_version;
830 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
831 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 /*
833 * Make sure to pull in the mode here as well in
834 * case the inode is released without being used.
835 * This ensures that xfs_inactive() will see that
836 * the inode is already free and not try to mess
837 * with the uninitialized part of it.
838 */
839 ip->i_d.di_mode = 0;
840 /*
841 * Initialize the per-fork minima and maxima for a new
842 * inode here. xfs_iformat will do it for old inodes.
843 */
844 ip->i_df.if_ext_max =
845 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
846 }
847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 /*
849 * The inode format changed when we moved the link count and
850 * made it 32 bits long. If this is an old format inode,
851 * convert it in memory to look like a new one. If it gets
852 * flushed to disk we will convert back before flushing or
853 * logging it. We zero out the new projid field and the old link
854 * count field. We'll handle clearing the pad field (the remains
855 * of the old uuid field) when we actually convert the inode to
856 * the new format. We don't change the version number so that we
857 * can distinguish this from a real new format inode.
858 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +1100859 if (ip->i_d.di_version == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 ip->i_d.di_nlink = ip->i_d.di_onlink;
861 ip->i_d.di_onlink = 0;
862 ip->i_d.di_projid = 0;
863 }
864
865 ip->i_delayed_blks = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000866 ip->i_size = ip->i_d.di_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
868 /*
869 * Mark the buffer containing the inode as something to keep
870 * around for a while. This helps to keep recently accessed
871 * meta-data in-core longer.
872 */
Christoph Hellwig6d73cf12008-12-09 04:47:32 -0500873 XFS_BUF_SET_REF(bp, XFS_INO_REF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
875 /*
876 * Use xfs_trans_brelse() to release the buffer containing the
877 * on-disk inode, because it was acquired with xfs_trans_read_buf()
878 * in xfs_itobp() above. If tp is NULL, this is just a normal
879 * brelse(). If we're within a transaction, then xfs_trans_brelse()
880 * will only release the buffer if it is not dirty within the
881 * transaction. It will be OK to release the buffer in this case,
882 * because inodes on disk are never destroyed and we will be
883 * locking the new in-core inode before putting it in the hash
884 * table where other processes can find it. Thus we don't have
885 * to worry about the inode being changed just because we released
886 * the buffer.
887 */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100888 out_brelse:
889 xfs_trans_brelse(tp, bp);
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100890 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891}
892
893/*
894 * Read in extents from a btree-format inode.
895 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
896 */
897int
898xfs_iread_extents(
899 xfs_trans_t *tp,
900 xfs_inode_t *ip,
901 int whichfork)
902{
903 int error;
904 xfs_ifork_t *ifp;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100905 xfs_extnum_t nextents;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 size_t size;
907
908 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
909 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
910 ip->i_mount);
911 return XFS_ERROR(EFSCORRUPTED);
912 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100913 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
914 size = nextents * sizeof(xfs_bmbt_rec_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 ifp = XFS_IFORK_PTR(ip, whichfork);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100916
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 /*
918 * We know that the size is valid (it's checked in iformat_btree)
919 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 ifp->if_lastex = NULLEXTNUM;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100921 ifp->if_bytes = ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 ifp->if_flags |= XFS_IFEXTENTS;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100923 xfs_iext_add(ifp, 0, nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 error = xfs_bmap_read_extents(tp, ip, whichfork);
925 if (error) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100926 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 ifp->if_flags &= ~XFS_IFEXTENTS;
928 return error;
929 }
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000930 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 return 0;
932}
933
934/*
935 * Allocate an inode on disk and return a copy of its in-core version.
936 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
937 * appropriately within the inode. The uid and gid for the inode are
938 * set according to the contents of the given cred structure.
939 *
940 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
941 * has a free inode available, call xfs_iget()
942 * to obtain the in-core version of the allocated inode. Finally,
943 * fill in the inode and log its initial contents. In this case,
944 * ialloc_context would be set to NULL and call_again set to false.
945 *
946 * If xfs_dialloc() does not have an available inode,
947 * it will replenish its supply by doing an allocation. Since we can
948 * only do one allocation within a transaction without deadlocks, we
949 * must commit the current transaction before returning the inode itself.
950 * In this case, therefore, we will set call_again to true and return.
951 * The caller should then commit the current transaction, start a new
952 * transaction, and call xfs_ialloc() again to actually get the inode.
953 *
954 * To ensure that some other process does not grab the inode that
955 * was allocated during the first call to xfs_ialloc(), this routine
956 * also returns the [locked] bp pointing to the head of the freelist
957 * as ialloc_context. The caller should hold this buffer across
958 * the commit and pass it back into this routine on the second call.
David Chinnerb11f94d2007-07-11 11:09:33 +1000959 *
960 * If we are allocating quota inodes, we do not have a parent inode
961 * to attach to or associate with (i.e. pip == NULL) because they
962 * are not linked into the directory structure - they are attached
963 * directly to the superblock - and so have no parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 */
965int
966xfs_ialloc(
967 xfs_trans_t *tp,
968 xfs_inode_t *pip,
969 mode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -0700970 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 xfs_dev_t rdev,
972 cred_t *cr,
973 xfs_prid_t prid,
974 int okalloc,
975 xfs_buf_t **ialloc_context,
976 boolean_t *call_again,
977 xfs_inode_t **ipp)
978{
979 xfs_ino_t ino;
980 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 uint flags;
982 int error;
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000983 timespec_t tv;
David Chinnerbf904242008-10-30 17:36:14 +1100984 int filestreams = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986 /*
987 * Call the space management code to pick
988 * the on-disk inode to be allocated.
989 */
David Chinnerb11f94d2007-07-11 11:09:33 +1000990 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 ialloc_context, call_again, &ino);
David Chinnerbf904242008-10-30 17:36:14 +1100992 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 if (*call_again || ino == NULLFSINO) {
995 *ipp = NULL;
996 return 0;
997 }
998 ASSERT(*ialloc_context == NULL);
999
1000 /*
1001 * Get the in-core inode with the lock held exclusively.
1002 * This is because we're setting fields here we need
1003 * to prevent others from looking at until we're done.
1004 */
1005 error = xfs_trans_iget(tp->t_mountp, tp, ino,
Nathan Scott745b1f472006-09-28 11:02:23 +10001006 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
David Chinnerbf904242008-10-30 17:36:14 +11001007 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 ASSERT(ip != NULL);
1010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 ip->i_d.di_mode = (__uint16_t)mode;
1012 ip->i_d.di_onlink = 0;
1013 ip->i_d.di_nlink = nlink;
1014 ASSERT(ip->i_d.di_nlink == nlink);
David Howells9e2b2dc2008-08-13 16:20:04 +01001015 ip->i_d.di_uid = current_fsuid();
1016 ip->i_d.di_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 ip->i_d.di_projid = prid;
1018 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1019
1020 /*
1021 * If the superblock version is up to where we support new format
1022 * inodes and this is currently an old format inode, then change
1023 * the inode version number now. This way we only do the conversion
1024 * here rather than here and in the flush/logging code.
1025 */
Eric Sandeen62118702008-03-06 13:44:28 +11001026 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11001027 ip->i_d.di_version == 1) {
1028 ip->i_d.di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 /*
1030 * We've already zeroed the old link count, the projid field,
1031 * and the pad field.
1032 */
1033 }
1034
1035 /*
1036 * Project ids won't be stored on disk if we are using a version 1 inode.
1037 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11001038 if ((prid != 0) && (ip->i_d.di_version == 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 xfs_bump_ino_vers2(tp, ip);
1040
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001041 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 ip->i_d.di_gid = pip->i_d.di_gid;
1043 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1044 ip->i_d.di_mode |= S_ISGID;
1045 }
1046 }
1047
1048 /*
1049 * If the group ID of the new file does not match the effective group
1050 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1051 * (and only if the irix_sgid_inherit compatibility variable is set).
1052 */
1053 if ((irix_sgid_inherit) &&
1054 (ip->i_d.di_mode & S_ISGID) &&
1055 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1056 ip->i_d.di_mode &= ~S_ISGID;
1057 }
1058
1059 ip->i_d.di_size = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001060 ip->i_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 ip->i_d.di_nextents = 0;
1062 ASSERT(ip->i_d.di_nblocks == 0);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +10001063
1064 nanotime(&tv);
1065 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1066 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1067 ip->i_d.di_atime = ip->i_d.di_mtime;
1068 ip->i_d.di_ctime = ip->i_d.di_mtime;
1069
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 /*
1071 * di_gen will have been taken care of in xfs_iread.
1072 */
1073 ip->i_d.di_extsize = 0;
1074 ip->i_d.di_dmevmask = 0;
1075 ip->i_d.di_dmstate = 0;
1076 ip->i_d.di_flags = 0;
1077 flags = XFS_ILOG_CORE;
1078 switch (mode & S_IFMT) {
1079 case S_IFIFO:
1080 case S_IFCHR:
1081 case S_IFBLK:
1082 case S_IFSOCK:
1083 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1084 ip->i_df.if_u2.if_rdev = rdev;
1085 ip->i_df.if_flags = 0;
1086 flags |= XFS_ILOG_DEV;
1087 break;
1088 case S_IFREG:
David Chinnerbf904242008-10-30 17:36:14 +11001089 /*
1090 * we can't set up filestreams until after the VFS inode
1091 * is set up properly.
1092 */
1093 if (pip && xfs_inode_is_filestream(pip))
1094 filestreams = 1;
David Chinner2a82b8b2007-07-11 11:09:12 +10001095 /* fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +10001097 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Nathan Scott365ca832005-06-21 15:39:12 +10001098 uint di_flags = 0;
1099
1100 if ((mode & S_IFMT) == S_IFDIR) {
1101 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1102 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001103 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1104 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1105 ip->i_d.di_extsize = pip->i_d.di_extsize;
1106 }
1107 } else if ((mode & S_IFMT) == S_IFREG) {
Christoph Hellwig613d7042007-10-11 17:44:08 +10001108 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +10001109 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001110 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1111 di_flags |= XFS_DIFLAG_EXTSIZE;
1112 ip->i_d.di_extsize = pip->i_d.di_extsize;
1113 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 }
1115 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1116 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +10001117 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1119 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +10001120 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1122 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +10001123 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1125 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +10001126 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1127 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1128 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +10001129 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1130 xfs_inherit_nodefrag)
1131 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +10001132 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1133 di_flags |= XFS_DIFLAG_FILESTREAM;
Nathan Scott365ca832005-06-21 15:39:12 +10001134 ip->i_d.di_flags |= di_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 }
1136 /* FALLTHROUGH */
1137 case S_IFLNK:
1138 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1139 ip->i_df.if_flags = XFS_IFEXTENTS;
1140 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1141 ip->i_df.if_u1.if_extents = NULL;
1142 break;
1143 default:
1144 ASSERT(0);
1145 }
1146 /*
1147 * Attribute fork settings for new inode.
1148 */
1149 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1150 ip->i_d.di_anextents = 0;
1151
1152 /*
1153 * Log the new values stuffed into the inode.
1154 */
1155 xfs_trans_log_inode(tp, ip, flags);
1156
Nathan Scottb83bd132006-06-09 16:48:30 +10001157 /* now that we have an i_mode we can setup inode ops and unlock */
Christoph Hellwig41be8be2008-08-13 16:23:13 +10001158 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
David Chinnerbf904242008-10-30 17:36:14 +11001160 /* now we have set up the vfs inode we can associate the filestream */
1161 if (filestreams) {
1162 error = xfs_filestream_associate(pip, ip);
1163 if (error < 0)
1164 return -error;
1165 if (!error)
1166 xfs_iflags_set(ip, XFS_IFILESTREAM);
1167 }
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 *ipp = ip;
1170 return 0;
1171}
1172
1173/*
1174 * Check to make sure that there are no blocks allocated to the
1175 * file beyond the size of the file. We don't check this for
1176 * files with fixed size extents or real time extents, but we
1177 * at least do it for regular files.
1178 */
1179#ifdef DEBUG
1180void
1181xfs_isize_check(
1182 xfs_mount_t *mp,
1183 xfs_inode_t *ip,
1184 xfs_fsize_t isize)
1185{
1186 xfs_fileoff_t map_first;
1187 int nimaps;
1188 xfs_bmbt_irec_t imaps[2];
1189
1190 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1191 return;
1192
Eric Sandeen71ddabb2007-11-23 16:29:42 +11001193 if (XFS_IS_REALTIME_INODE(ip))
1194 return;
1195
1196 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 return;
1198
1199 nimaps = 2;
1200 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1201 /*
1202 * The filesystem could be shutting down, so bmapi may return
1203 * an error.
1204 */
1205 if (xfs_bmapi(NULL, ip, map_first,
1206 (XFS_B_TO_FSB(mp,
1207 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1208 map_first),
1209 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001210 NULL, NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 return;
1212 ASSERT(nimaps == 1);
1213 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1214}
1215#endif /* DEBUG */
1216
1217/*
1218 * Calculate the last possible buffered byte in a file. This must
1219 * include data that was buffered beyond the EOF by the write code.
1220 * This also needs to deal with overflowing the xfs_fsize_t type
1221 * which can happen for sizes near the limit.
1222 *
1223 * We also need to take into account any blocks beyond the EOF. It
1224 * may be the case that they were buffered by a write which failed.
1225 * In that case the pages will still be in memory, but the inode size
1226 * will never have been updated.
1227 */
1228xfs_fsize_t
1229xfs_file_last_byte(
1230 xfs_inode_t *ip)
1231{
1232 xfs_mount_t *mp;
1233 xfs_fsize_t last_byte;
1234 xfs_fileoff_t last_block;
1235 xfs_fileoff_t size_last_block;
1236 int error;
1237
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001238 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
1240 mp = ip->i_mount;
1241 /*
1242 * Only check for blocks beyond the EOF if the extents have
1243 * been read in. This eliminates the need for the inode lock,
1244 * and it also saves us from looking when it really isn't
1245 * necessary.
1246 */
1247 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1248 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1249 XFS_DATA_FORK);
1250 if (error) {
1251 last_block = 0;
1252 }
1253 } else {
1254 last_block = 0;
1255 }
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001256 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1258
1259 last_byte = XFS_FSB_TO_B(mp, last_block);
1260 if (last_byte < 0) {
1261 return XFS_MAXIOFFSET(mp);
1262 }
1263 last_byte += (1 << mp->m_writeio_log);
1264 if (last_byte < 0) {
1265 return XFS_MAXIOFFSET(mp);
1266 }
1267 return last_byte;
1268}
1269
1270#if defined(XFS_RW_TRACE)
1271STATIC void
1272xfs_itrunc_trace(
1273 int tag,
1274 xfs_inode_t *ip,
1275 int flag,
1276 xfs_fsize_t new_size,
1277 xfs_off_t toss_start,
1278 xfs_off_t toss_finish)
1279{
1280 if (ip->i_rwtrace == NULL) {
1281 return;
1282 }
1283
1284 ktrace_enter(ip->i_rwtrace,
1285 (void*)((long)tag),
1286 (void*)ip,
1287 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1288 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1289 (void*)((long)flag),
1290 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1291 (void*)(unsigned long)(new_size & 0xffffffff),
1292 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1293 (void*)(unsigned long)(toss_start & 0xffffffff),
1294 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1295 (void*)(unsigned long)(toss_finish & 0xffffffff),
1296 (void*)(unsigned long)current_cpu(),
Yingping Luf1fdc842006-03-22 12:44:15 +11001297 (void*)(unsigned long)current_pid(),
1298 (void*)NULL,
1299 (void*)NULL,
1300 (void*)NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301}
1302#else
1303#define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1304#endif
1305
1306/*
1307 * Start the truncation of the file to new_size. The new size
1308 * must be smaller than the current size. This routine will
1309 * clear the buffer and page caches of file data in the removed
1310 * range, and xfs_itruncate_finish() will remove the underlying
1311 * disk blocks.
1312 *
1313 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1314 * must NOT have the inode lock held at all. This is because we're
1315 * calling into the buffer/page cache code and we can't hold the
1316 * inode lock when we do so.
1317 *
David Chinner38e22992006-03-22 12:47:15 +11001318 * We need to wait for any direct I/Os in flight to complete before we
1319 * proceed with the truncate. This is needed to prevent the extents
1320 * being read or written by the direct I/Os from being removed while the
1321 * I/O is in flight as there is no other method of synchronising
1322 * direct I/O with the truncate operation. Also, because we hold
1323 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1324 * started until the truncate completes and drops the lock. Essentially,
Christoph Hellwig25e41b32008-12-03 12:20:39 +01001325 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1326 * ordering between direct I/Os and the truncate operation.
David Chinner38e22992006-03-22 12:47:15 +11001327 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1329 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1330 * in the case that the caller is locking things out of order and
1331 * may not be able to call xfs_itruncate_finish() with the inode lock
1332 * held without dropping the I/O lock. If the caller must drop the
1333 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1334 * must be called again with all the same restrictions as the initial
1335 * call.
1336 */
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001337int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338xfs_itruncate_start(
1339 xfs_inode_t *ip,
1340 uint flags,
1341 xfs_fsize_t new_size)
1342{
1343 xfs_fsize_t last_byte;
1344 xfs_off_t toss_start;
1345 xfs_mount_t *mp;
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001346 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001348 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001349 ASSERT((new_size == 0) || (new_size <= ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1351 (flags == XFS_ITRUNC_MAYBE));
1352
1353 mp = ip->i_mount;
Yingping Lu9fa80462006-03-22 12:44:35 +11001354
Lachlan McIlroyc734c792007-12-18 16:17:41 +11001355 /* wait for the completion of any pending DIOs */
Lachlan McIlroyd112f292008-10-30 16:59:06 +11001356 if (new_size == 0 || new_size < ip->i_size)
Christoph Hellwig25e41b32008-12-03 12:20:39 +01001357 xfs_ioend_wait(ip);
Lachlan McIlroyc734c792007-12-18 16:17:41 +11001358
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 /*
Nathan Scott67fcaa72006-06-09 17:00:52 +10001360 * Call toss_pages or flushinval_pages to get rid of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 * overlapping the region being removed. We have to use
Nathan Scott67fcaa72006-06-09 17:00:52 +10001362 * the less efficient flushinval_pages in the case that the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 * caller may not be able to finish the truncate without
1364 * dropping the inode's I/O lock. Make sure
1365 * to catch any pages brought in by buffers overlapping
1366 * the EOF by searching out beyond the isize by our
1367 * block size. We round new_size up to a block boundary
1368 * so that we don't toss things on the same block as
1369 * new_size but before it.
1370 *
Nathan Scott67fcaa72006-06-09 17:00:52 +10001371 * Before calling toss_page or flushinval_pages, make sure to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 * call remapf() over the same region if the file is mapped.
1373 * This frees up mapped file references to the pages in the
Nathan Scott67fcaa72006-06-09 17:00:52 +10001374 * given range and for the flushinval_pages case it ensures
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 * that we get the latest mapped changes flushed out.
1376 */
1377 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1378 toss_start = XFS_FSB_TO_B(mp, toss_start);
1379 if (toss_start < 0) {
1380 /*
1381 * The place to start tossing is beyond our maximum
1382 * file size, so there is no way that the data extended
1383 * out there.
1384 */
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001385 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 }
1387 last_byte = xfs_file_last_byte(ip);
1388 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1389 last_byte);
1390 if (last_byte > toss_start) {
1391 if (flags & XFS_ITRUNC_DEFINITE) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001392 xfs_tosspages(ip, toss_start,
1393 -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 } else {
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001395 error = xfs_flushinval_pages(ip, toss_start,
1396 -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 }
1398 }
1399
1400#ifdef DEBUG
1401 if (new_size == 0) {
Christoph Hellwigdf80c932008-08-13 16:22:09 +10001402 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 }
1404#endif
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001405 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406}
1407
1408/*
David Chinnerf6485052008-04-17 16:50:04 +10001409 * Shrink the file to the given new_size. The new size must be smaller than
1410 * the current size. This will free up the underlying blocks in the removed
1411 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 *
David Chinnerf6485052008-04-17 16:50:04 +10001413 * The transaction passed to this routine must have made a permanent log
1414 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1415 * given transaction and start new ones, so make sure everything involved in
1416 * the transaction is tidy before calling here. Some transaction will be
1417 * returned to the caller to be committed. The incoming transaction must
1418 * already include the inode, and both inode locks must be held exclusively.
1419 * The inode must also be "held" within the transaction. On return the inode
1420 * will be "held" within the returned transaction. This routine does NOT
1421 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 *
David Chinnerf6485052008-04-17 16:50:04 +10001423 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1424 * indicates the fork which is to be truncated. For the attribute fork we only
1425 * support truncation to size 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 *
David Chinnerf6485052008-04-17 16:50:04 +10001427 * We use the sync parameter to indicate whether or not the first transaction
1428 * we perform might have to be synchronous. For the attr fork, it needs to be
1429 * so if the unlink of the inode is not yet known to be permanent in the log.
1430 * This keeps us from freeing and reusing the blocks of the attribute fork
1431 * before the unlink of the inode becomes permanent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 *
David Chinnerf6485052008-04-17 16:50:04 +10001433 * For the data fork, we normally have to run synchronously if we're being
1434 * called out of the inactive path or we're being called out of the create path
1435 * where we're truncating an existing file. Either way, the truncate needs to
1436 * be sync so blocks don't reappear in the file with altered data in case of a
1437 * crash. wsync filesystems can run the first case async because anything that
1438 * shrinks the inode has to run sync so by the time we're called here from
1439 * inactive, the inode size is permanently set to 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 *
David Chinnerf6485052008-04-17 16:50:04 +10001441 * Calls from the truncate path always need to be sync unless we're in a wsync
1442 * filesystem and the file has already been unlinked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 *
David Chinnerf6485052008-04-17 16:50:04 +10001444 * The caller is responsible for correctly setting the sync parameter. It gets
1445 * too hard for us to guess here which path we're being called out of just
1446 * based on inode state.
1447 *
1448 * If we get an error, we must return with the inode locked and linked into the
1449 * current transaction. This keeps things simple for the higher level code,
1450 * because it always knows that the inode is locked and held in the transaction
1451 * that returns to it whether errors occur or not. We don't mark the inode
1452 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 */
1454int
1455xfs_itruncate_finish(
1456 xfs_trans_t **tp,
1457 xfs_inode_t *ip,
1458 xfs_fsize_t new_size,
1459 int fork,
1460 int sync)
1461{
1462 xfs_fsblock_t first_block;
1463 xfs_fileoff_t first_unmap_block;
1464 xfs_fileoff_t last_block;
1465 xfs_filblks_t unmap_len=0;
1466 xfs_mount_t *mp;
1467 xfs_trans_t *ntp;
1468 int done;
1469 int committed;
1470 xfs_bmap_free_t free_list;
1471 int error;
1472
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001473 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001474 ASSERT((new_size == 0) || (new_size <= ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 ASSERT(*tp != NULL);
1476 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1477 ASSERT(ip->i_transp == *tp);
1478 ASSERT(ip->i_itemp != NULL);
1479 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1480
1481
1482 ntp = *tp;
1483 mp = (ntp)->t_mountp;
1484 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1485
1486 /*
1487 * We only support truncating the entire attribute fork.
1488 */
1489 if (fork == XFS_ATTR_FORK) {
1490 new_size = 0LL;
1491 }
1492 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1493 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1494 /*
1495 * The first thing we do is set the size to new_size permanently
1496 * on disk. This way we don't have to worry about anyone ever
1497 * being able to look at the data being freed even in the face
1498 * of a crash. What we're getting around here is the case where
1499 * we free a block, it is allocated to another file, it is written
1500 * to, and then we crash. If the new data gets written to the
1501 * file but the log buffers containing the free and reallocation
1502 * don't, then we'd end up with garbage in the blocks being freed.
1503 * As long as we make the new_size permanent before actually
1504 * freeing any blocks it doesn't matter if they get writtten to.
1505 *
1506 * The callers must signal into us whether or not the size
1507 * setting here must be synchronous. There are a few cases
1508 * where it doesn't have to be synchronous. Those cases
1509 * occur if the file is unlinked and we know the unlink is
1510 * permanent or if the blocks being truncated are guaranteed
1511 * to be beyond the inode eof (regardless of the link count)
1512 * and the eof value is permanent. Both of these cases occur
1513 * only on wsync-mounted filesystems. In those cases, we're
1514 * guaranteed that no user will ever see the data in the blocks
1515 * that are being truncated so the truncate can run async.
1516 * In the free beyond eof case, the file may wind up with
1517 * more blocks allocated to it than it needs if we crash
1518 * and that won't get fixed until the next time the file
1519 * is re-opened and closed but that's ok as that shouldn't
1520 * be too many blocks.
1521 *
1522 * However, we can't just make all wsync xactions run async
1523 * because there's one call out of the create path that needs
1524 * to run sync where it's truncating an existing file to size
1525 * 0 whose size is > 0.
1526 *
1527 * It's probably possible to come up with a test in this
1528 * routine that would correctly distinguish all the above
1529 * cases from the values of the function parameters and the
1530 * inode state but for sanity's sake, I've decided to let the
1531 * layers above just tell us. It's simpler to correctly figure
1532 * out in the layer above exactly under what conditions we
1533 * can run async and I think it's easier for others read and
1534 * follow the logic in case something has to be changed.
1535 * cscope is your friend -- rcc.
1536 *
1537 * The attribute fork is much simpler.
1538 *
1539 * For the attribute fork we allow the caller to tell us whether
1540 * the unlink of the inode that led to this call is yet permanent
1541 * in the on disk log. If it is not and we will be freeing extents
1542 * in this inode then we make the first transaction synchronous
1543 * to make sure that the unlink is permanent by the time we free
1544 * the blocks.
1545 */
1546 if (fork == XFS_DATA_FORK) {
1547 if (ip->i_d.di_nextents > 0) {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001548 /*
1549 * If we are not changing the file size then do
1550 * not update the on-disk file size - we may be
1551 * called from xfs_inactive_free_eofblocks(). If we
1552 * update the on-disk file size and then the system
1553 * crashes before the contents of the file are
1554 * flushed to disk then the files may be full of
1555 * holes (ie NULL files bug).
1556 */
1557 if (ip->i_size != new_size) {
1558 ip->i_d.di_size = new_size;
1559 ip->i_size = new_size;
1560 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 }
1563 } else if (sync) {
1564 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1565 if (ip->i_d.di_anextents > 0)
1566 xfs_trans_set_sync(ntp);
1567 }
1568 ASSERT(fork == XFS_DATA_FORK ||
1569 (fork == XFS_ATTR_FORK &&
1570 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1571 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1572
1573 /*
1574 * Since it is possible for space to become allocated beyond
1575 * the end of the file (in a crash where the space is allocated
1576 * but the inode size is not yet updated), simply remove any
1577 * blocks which show up between the new EOF and the maximum
1578 * possible file size. If the first block to be removed is
1579 * beyond the maximum file size (ie it is the same as last_block),
1580 * then there is nothing to do.
1581 */
1582 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1583 ASSERT(first_unmap_block <= last_block);
1584 done = 0;
1585 if (last_block == first_unmap_block) {
1586 done = 1;
1587 } else {
1588 unmap_len = last_block - first_unmap_block + 1;
1589 }
1590 while (!done) {
1591 /*
1592 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1593 * will tell us whether it freed the entire range or
1594 * not. If this is a synchronous mount (wsync),
1595 * then we can tell bunmapi to keep all the
1596 * transactions asynchronous since the unlink
1597 * transaction that made this inode inactive has
1598 * already hit the disk. There's no danger of
1599 * the freed blocks being reused, there being a
1600 * crash, and the reused blocks suddenly reappearing
1601 * in this file with garbage in them once recovery
1602 * runs.
1603 */
1604 XFS_BMAP_INIT(&free_list, &first_block);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10001605 error = xfs_bunmapi(ntp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001606 first_unmap_block, unmap_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 XFS_BMAPI_AFLAG(fork) |
1608 (sync ? 0 : XFS_BMAPI_ASYNC),
1609 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001610 &first_block, &free_list,
1611 NULL, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (error) {
1613 /*
1614 * If the bunmapi call encounters an error,
1615 * return to the caller where the transaction
1616 * can be properly aborted. We just need to
1617 * make sure we're not holding any resources
1618 * that we were not when we came in.
1619 */
1620 xfs_bmap_cancel(&free_list);
1621 return error;
1622 }
1623
1624 /*
1625 * Duplicate the transaction that has the permanent
1626 * reservation and commit the old transaction.
1627 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001628 error = xfs_bmap_finish(tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 ntp = *tp;
David Chinnerf6485052008-04-17 16:50:04 +10001630 if (committed) {
1631 /* link the inode into the next xact in the chain */
1632 xfs_trans_ijoin(ntp, ip,
1633 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1634 xfs_trans_ihold(ntp, ip);
1635 }
1636
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if (error) {
1638 /*
David Chinnerf6485052008-04-17 16:50:04 +10001639 * If the bmap finish call encounters an error, return
1640 * to the caller where the transaction can be properly
1641 * aborted. We just need to make sure we're not
1642 * holding any resources that we were not when we came
1643 * in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 *
David Chinnerf6485052008-04-17 16:50:04 +10001645 * Aborting from this point might lose some blocks in
1646 * the file system, but oh well.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 */
1648 xfs_bmap_cancel(&free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 return error;
1650 }
1651
1652 if (committed) {
1653 /*
David Chinnerf6485052008-04-17 16:50:04 +10001654 * Mark the inode dirty so it will be logged and
David Chinnere5720ee2008-04-10 12:21:18 +10001655 * moved forward in the log as part of every commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1658 }
David Chinnerf6485052008-04-17 16:50:04 +10001659
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 ntp = xfs_trans_dup(ntp);
David Chinnere5720ee2008-04-10 12:21:18 +10001661 error = xfs_trans_commit(*tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 *tp = ntp;
David Chinnere5720ee2008-04-10 12:21:18 +10001663
David Chinnerf6485052008-04-17 16:50:04 +10001664 /* link the inode into the next transaction in the chain */
1665 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1666 xfs_trans_ihold(ntp, ip);
1667
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001668 if (error)
1669 return error;
1670 /*
1671 * transaction commit worked ok so we can drop the extra ticket
1672 * reference that we gained in xfs_trans_dup()
1673 */
1674 xfs_log_ticket_put(ntp->t_ticket);
1675 error = xfs_trans_reserve(ntp, 0,
David Chinnerf6485052008-04-17 16:50:04 +10001676 XFS_ITRUNCATE_LOG_RES(mp), 0,
1677 XFS_TRANS_PERM_LOG_RES,
1678 XFS_ITRUNCATE_LOG_COUNT);
1679 if (error)
1680 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 }
1682 /*
1683 * Only update the size in the case of the data fork, but
1684 * always re-log the inode so that our permanent transaction
1685 * can keep on rolling it forward in the log.
1686 */
1687 if (fork == XFS_DATA_FORK) {
1688 xfs_isize_check(mp, ip, new_size);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001689 /*
1690 * If we are not changing the file size then do
1691 * not update the on-disk file size - we may be
1692 * called from xfs_inactive_free_eofblocks(). If we
1693 * update the on-disk file size and then the system
1694 * crashes before the contents of the file are
1695 * flushed to disk then the files may be full of
1696 * holes (ie NULL files bug).
1697 */
1698 if (ip->i_size != new_size) {
1699 ip->i_d.di_size = new_size;
1700 ip->i_size = new_size;
1701 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 }
1703 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1704 ASSERT((new_size != 0) ||
1705 (fork == XFS_ATTR_FORK) ||
1706 (ip->i_delayed_blks == 0));
1707 ASSERT((new_size != 0) ||
1708 (fork == XFS_ATTR_FORK) ||
1709 (ip->i_d.di_nextents == 0));
1710 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1711 return 0;
1712}
1713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714/*
1715 * This is called when the inode's link count goes to 0.
1716 * We place the on-disk inode on a list in the AGI. It
1717 * will be pulled from this list when the inode is freed.
1718 */
1719int
1720xfs_iunlink(
1721 xfs_trans_t *tp,
1722 xfs_inode_t *ip)
1723{
1724 xfs_mount_t *mp;
1725 xfs_agi_t *agi;
1726 xfs_dinode_t *dip;
1727 xfs_buf_t *agibp;
1728 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 xfs_agino_t agino;
1730 short bucket_index;
1731 int offset;
1732 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734 ASSERT(ip->i_d.di_nlink == 0);
1735 ASSERT(ip->i_d.di_mode != 0);
1736 ASSERT(ip->i_transp == tp);
1737
1738 mp = tp->t_mountp;
1739
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 /*
1741 * Get the agi buffer first. It ensures lock ordering
1742 * on the list.
1743 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001744 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001745 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001748
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 /*
1750 * Get the index into the agi hash table for the
1751 * list this inode will go on.
1752 */
1753 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1754 ASSERT(agino != 0);
1755 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1756 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001757 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
Christoph Hellwig16259e72005-11-02 15:11:25 +11001759 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 /*
1761 * There is already another inode in the bucket we need
1762 * to add ourselves to. Add us at the front of the list.
1763 * Here we put the head pointer into our next pointer,
1764 * and then we fall through to point the head at us.
1765 */
Christoph Hellwig76d8b272008-11-28 14:23:40 +11001766 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
Vlad Apostolovc319b582007-11-23 16:27:51 +11001767 if (error)
1768 return error;
1769
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001770 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 /* both on-disk, don't endian flip twice */
1772 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001773 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 offsetof(xfs_dinode_t, di_next_unlinked);
1775 xfs_trans_inode_buf(tp, ibp);
1776 xfs_trans_log_buf(tp, ibp, offset,
1777 (offset + sizeof(xfs_agino_t) - 1));
1778 xfs_inobp_check(mp, ibp);
1779 }
1780
1781 /*
1782 * Point the bucket head pointer at the inode being inserted.
1783 */
1784 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001785 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 offset = offsetof(xfs_agi_t, agi_unlinked) +
1787 (sizeof(xfs_agino_t) * bucket_index);
1788 xfs_trans_log_buf(tp, agibp, offset,
1789 (offset + sizeof(xfs_agino_t) - 1));
1790 return 0;
1791}
1792
1793/*
1794 * Pull the on-disk inode from the AGI unlinked list.
1795 */
1796STATIC int
1797xfs_iunlink_remove(
1798 xfs_trans_t *tp,
1799 xfs_inode_t *ip)
1800{
1801 xfs_ino_t next_ino;
1802 xfs_mount_t *mp;
1803 xfs_agi_t *agi;
1804 xfs_dinode_t *dip;
1805 xfs_buf_t *agibp;
1806 xfs_buf_t *ibp;
1807 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 xfs_agino_t agino;
1809 xfs_agino_t next_agino;
1810 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001811 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001813 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
1819 /*
1820 * Get the agi buffer first. It ensures lock ordering
1821 * on the list.
1822 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001823 error = xfs_read_agi(mp, tp, agno, &agibp);
1824 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001826
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001828
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 /*
1830 * Get the index into the agi hash table for the
1831 * list this inode will go on.
1832 */
1833 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1834 ASSERT(agino != 0);
1835 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001836 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 ASSERT(agi->agi_unlinked[bucket_index]);
1838
Christoph Hellwig16259e72005-11-02 15:11:25 +11001839 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 /*
1841 * We're at the head of the list. Get the inode's
1842 * on-disk buffer to see if there is anyone after us
1843 * on the list. Only modify our next pointer if it
1844 * is not already NULLAGINO. This saves us the overhead
1845 * of dealing with the buffer when there is no need to
1846 * change it.
1847 */
Christoph Hellwig76d8b272008-11-28 14:23:40 +11001848 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 if (error) {
1850 cmn_err(CE_WARN,
1851 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1852 error, mp->m_fsname);
1853 return error;
1854 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001855 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 ASSERT(next_agino != 0);
1857 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001858 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001859 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 offsetof(xfs_dinode_t, di_next_unlinked);
1861 xfs_trans_inode_buf(tp, ibp);
1862 xfs_trans_log_buf(tp, ibp, offset,
1863 (offset + sizeof(xfs_agino_t) - 1));
1864 xfs_inobp_check(mp, ibp);
1865 } else {
1866 xfs_trans_brelse(tp, ibp);
1867 }
1868 /*
1869 * Point the bucket head pointer at the next inode.
1870 */
1871 ASSERT(next_agino != 0);
1872 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001873 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 offset = offsetof(xfs_agi_t, agi_unlinked) +
1875 (sizeof(xfs_agino_t) * bucket_index);
1876 xfs_trans_log_buf(tp, agibp, offset,
1877 (offset + sizeof(xfs_agino_t) - 1));
1878 } else {
1879 /*
1880 * We need to search the list for the inode being freed.
1881 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001882 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 last_ibp = NULL;
1884 while (next_agino != agino) {
1885 /*
1886 * If the last inode wasn't the one pointing to
1887 * us, then release its buffer since we're not
1888 * going to do anything with it.
1889 */
1890 if (last_ibp != NULL) {
1891 xfs_trans_brelse(tp, last_ibp);
1892 }
1893 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1894 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
Christoph Hellwigc679eef2008-10-30 18:04:13 +11001895 &last_ibp, &last_offset, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 if (error) {
1897 cmn_err(CE_WARN,
1898 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1899 error, mp->m_fsname);
1900 return error;
1901 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001902 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 ASSERT(next_agino != NULLAGINO);
1904 ASSERT(next_agino != 0);
1905 }
1906 /*
1907 * Now last_ibp points to the buffer previous to us on
1908 * the unlinked list. Pull us from the list.
1909 */
Christoph Hellwig76d8b272008-11-28 14:23:40 +11001910 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 if (error) {
1912 cmn_err(CE_WARN,
1913 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1914 error, mp->m_fsname);
1915 return error;
1916 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001917 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 ASSERT(next_agino != 0);
1919 ASSERT(next_agino != agino);
1920 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001921 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001922 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 offsetof(xfs_dinode_t, di_next_unlinked);
1924 xfs_trans_inode_buf(tp, ibp);
1925 xfs_trans_log_buf(tp, ibp, offset,
1926 (offset + sizeof(xfs_agino_t) - 1));
1927 xfs_inobp_check(mp, ibp);
1928 } else {
1929 xfs_trans_brelse(tp, ibp);
1930 }
1931 /*
1932 * Point the previous inode on the list to the next inode.
1933 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001934 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 ASSERT(next_agino != 0);
1936 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1937 xfs_trans_inode_buf(tp, last_ibp);
1938 xfs_trans_log_buf(tp, last_ibp, offset,
1939 (offset + sizeof(xfs_agino_t) - 1));
1940 xfs_inobp_check(mp, last_ibp);
1941 }
1942 return 0;
1943}
1944
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001945STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946xfs_ifree_cluster(
1947 xfs_inode_t *free_ip,
1948 xfs_trans_t *tp,
1949 xfs_ino_t inum)
1950{
1951 xfs_mount_t *mp = free_ip->i_mount;
1952 int blks_per_cluster;
1953 int nbufs;
1954 int ninodes;
1955 int i, j, found, pre_flushed;
1956 xfs_daddr_t blkno;
1957 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 xfs_inode_t *ip, **ip_found;
1959 xfs_inode_log_item_t *iip;
1960 xfs_log_item_t *lip;
David Chinnerda353b02007-08-28 14:00:13 +10001961 xfs_perag_t *pag = xfs_get_perag(mp, inum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962
1963 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1964 blks_per_cluster = 1;
1965 ninodes = mp->m_sb.sb_inopblock;
1966 nbufs = XFS_IALLOC_BLOCKS(mp);
1967 } else {
1968 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1969 mp->m_sb.sb_blocksize;
1970 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1971 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1972 }
1973
1974 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
1975
1976 for (j = 0; j < nbufs; j++, inum += ninodes) {
1977 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1978 XFS_INO_TO_AGBNO(mp, inum));
1979
1980
1981 /*
1982 * Look for each inode in memory and attempt to lock it,
1983 * we can be racing with flush and tail pushing here.
1984 * any inode we get the locks on, add to an array of
1985 * inode items to process later.
1986 *
1987 * The get the buffer lock, we could beat a flush
1988 * or tail pushing thread to the lock here, in which
1989 * case they will go looking for the inode buffer
1990 * and fail, we need some other form of interlock
1991 * here.
1992 */
1993 found = 0;
1994 for (i = 0; i < ninodes; i++) {
David Chinnerda353b02007-08-28 14:00:13 +10001995 read_lock(&pag->pag_ici_lock);
1996 ip = radix_tree_lookup(&pag->pag_ici_root,
1997 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
1999 /* Inode not in memory or we found it already,
2000 * nothing to do
2001 */
David Chinner7a18c382006-11-11 18:04:54 +11002002 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
David Chinnerda353b02007-08-28 14:00:13 +10002003 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 continue;
2005 }
2006
2007 if (xfs_inode_clean(ip)) {
David Chinnerda353b02007-08-28 14:00:13 +10002008 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 continue;
2010 }
2011
2012 /* If we can get the locks then add it to the
2013 * list, otherwise by the time we get the bp lock
2014 * below it will already be attached to the
2015 * inode buffer.
2016 */
2017
2018 /* This inode will already be locked - by us, lets
2019 * keep it that way.
2020 */
2021
2022 if (ip == free_ip) {
2023 if (xfs_iflock_nowait(ip)) {
David Chinner7a18c382006-11-11 18:04:54 +11002024 xfs_iflags_set(ip, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 if (xfs_inode_clean(ip)) {
2026 xfs_ifunlock(ip);
2027 } else {
2028 ip_found[found++] = ip;
2029 }
2030 }
David Chinnerda353b02007-08-28 14:00:13 +10002031 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 continue;
2033 }
2034
2035 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2036 if (xfs_iflock_nowait(ip)) {
David Chinner7a18c382006-11-11 18:04:54 +11002037 xfs_iflags_set(ip, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
2039 if (xfs_inode_clean(ip)) {
2040 xfs_ifunlock(ip);
2041 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2042 } else {
2043 ip_found[found++] = ip;
2044 }
2045 } else {
2046 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2047 }
2048 }
David Chinnerda353b02007-08-28 14:00:13 +10002049 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 }
2051
2052 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2053 mp->m_bsize * blks_per_cluster,
2054 XFS_BUF_LOCK);
2055
2056 pre_flushed = 0;
2057 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2058 while (lip) {
2059 if (lip->li_type == XFS_LI_INODE) {
2060 iip = (xfs_inode_log_item_t *)lip;
2061 ASSERT(iip->ili_logged == 1);
2062 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
David Chinner7b2e2a32008-10-30 17:39:12 +11002063 xfs_trans_ail_copy_lsn(mp->m_ail,
2064 &iip->ili_flush_lsn,
2065 &iip->ili_item.li_lsn);
David Chinnere5ffd2b2006-11-21 18:55:33 +11002066 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 pre_flushed++;
2068 }
2069 lip = lip->li_bio_list;
2070 }
2071
2072 for (i = 0; i < found; i++) {
2073 ip = ip_found[i];
2074 iip = ip->i_itemp;
2075
2076 if (!iip) {
2077 ip->i_update_core = 0;
2078 xfs_ifunlock(ip);
2079 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2080 continue;
2081 }
2082
2083 iip->ili_last_fields = iip->ili_format.ilf_fields;
2084 iip->ili_format.ilf_fields = 0;
2085 iip->ili_logged = 1;
David Chinner7b2e2a32008-10-30 17:39:12 +11002086 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2087 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
2089 xfs_buf_attach_iodone(bp,
2090 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2091 xfs_istale_done, (xfs_log_item_t *)iip);
2092 if (ip != free_ip) {
2093 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2094 }
2095 }
2096
2097 if (found || pre_flushed)
2098 xfs_trans_stale_inode_buf(tp, bp);
2099 xfs_trans_binval(tp, bp);
2100 }
2101
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002102 kmem_free(ip_found);
David Chinnerda353b02007-08-28 14:00:13 +10002103 xfs_put_perag(mp, pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104}
2105
2106/*
2107 * This is called to return an inode to the inode free list.
2108 * The inode should already be truncated to 0 length and have
2109 * no pages associated with it. This routine also assumes that
2110 * the inode is already a part of the transaction.
2111 *
2112 * The on-disk copy of the inode will have been added to the list
2113 * of unlinked inodes in the AGI. We need to remove the inode from
2114 * that list atomically with respect to freeing it here.
2115 */
2116int
2117xfs_ifree(
2118 xfs_trans_t *tp,
2119 xfs_inode_t *ip,
2120 xfs_bmap_free_t *flist)
2121{
2122 int error;
2123 int delete;
2124 xfs_ino_t first_ino;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002125 xfs_dinode_t *dip;
2126 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002128 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 ASSERT(ip->i_transp == tp);
2130 ASSERT(ip->i_d.di_nlink == 0);
2131 ASSERT(ip->i_d.di_nextents == 0);
2132 ASSERT(ip->i_d.di_anextents == 0);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002133 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2135 ASSERT(ip->i_d.di_nblocks == 0);
2136
2137 /*
2138 * Pull the on-disk inode from the AGI unlinked list.
2139 */
2140 error = xfs_iunlink_remove(tp, ip);
2141 if (error != 0) {
2142 return error;
2143 }
2144
2145 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2146 if (error != 0) {
2147 return error;
2148 }
2149 ip->i_d.di_mode = 0; /* mark incore inode as free */
2150 ip->i_d.di_flags = 0;
2151 ip->i_d.di_dmevmask = 0;
2152 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2153 ip->i_df.if_ext_max =
2154 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2155 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2156 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2157 /*
2158 * Bump the generation count so no one will be confused
2159 * by reincarnations of this inode.
2160 */
2161 ip->i_d.di_gen++;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002162
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2164
Christoph Hellwig76d8b272008-11-28 14:23:40 +11002165 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XFS_BUF_LOCK);
Vlad Apostolovc319b582007-11-23 16:27:51 +11002166 if (error)
2167 return error;
2168
2169 /*
2170 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2171 * from picking up this inode when it is reclaimed (its incore state
2172 * initialzed but not flushed to disk yet). The in-core di_mode is
2173 * already cleared and a corresponding transaction logged.
2174 * The hack here just synchronizes the in-core to on-disk
2175 * di_mode value in advance before the actual inode sync to disk.
2176 * This is OK because the inode is already unlinked and would never
2177 * change its di_mode again for this inode generation.
2178 * This is a temporary hack that would require a proper fix
2179 * in the future.
2180 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002181 dip->di_mode = 0;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002182
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 if (delete) {
2184 xfs_ifree_cluster(ip, tp, first_ino);
2185 }
2186
2187 return 0;
2188}
2189
2190/*
2191 * Reallocate the space for if_broot based on the number of records
2192 * being added or deleted as indicated in rec_diff. Move the records
2193 * and pointers in if_broot to fit the new size. When shrinking this
2194 * will eliminate holes between the records and pointers created by
2195 * the caller. When growing this will create holes to be filled in
2196 * by the caller.
2197 *
2198 * The caller must not request to add more records than would fit in
2199 * the on-disk inode root. If the if_broot is currently NULL, then
2200 * if we adding records one will be allocated. The caller must also
2201 * not request that the number of records go below zero, although
2202 * it can go to zero.
2203 *
2204 * ip -- the inode whose if_broot area is changing
2205 * ext_diff -- the change in the number of records, positive or negative,
2206 * requested for the if_broot array.
2207 */
2208void
2209xfs_iroot_realloc(
2210 xfs_inode_t *ip,
2211 int rec_diff,
2212 int whichfork)
2213{
Christoph Hellwig60197e82008-10-30 17:11:19 +11002214 struct xfs_mount *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 int cur_max;
2216 xfs_ifork_t *ifp;
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002217 struct xfs_btree_block *new_broot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 int new_max;
2219 size_t new_size;
2220 char *np;
2221 char *op;
2222
2223 /*
2224 * Handle the degenerate case quietly.
2225 */
2226 if (rec_diff == 0) {
2227 return;
2228 }
2229
2230 ifp = XFS_IFORK_PTR(ip, whichfork);
2231 if (rec_diff > 0) {
2232 /*
2233 * If there wasn't any memory allocated before, just
2234 * allocate it now and get out.
2235 */
2236 if (ifp->if_broot_bytes == 0) {
2237 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002238 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 ifp->if_broot_bytes = (int)new_size;
2240 return;
2241 }
2242
2243 /*
2244 * If there is already an existing if_broot, then we need
2245 * to realloc() it and shift the pointers to their new
2246 * location. The records don't change location because
2247 * they are kept butted up against the btree block header.
2248 */
Christoph Hellwig60197e82008-10-30 17:11:19 +11002249 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 new_max = cur_max + rec_diff;
2251 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002252 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2254 KM_SLEEP);
Christoph Hellwig60197e82008-10-30 17:11:19 +11002255 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2256 ifp->if_broot_bytes);
2257 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2258 (int)new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 ifp->if_broot_bytes = (int)new_size;
2260 ASSERT(ifp->if_broot_bytes <=
2261 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2262 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2263 return;
2264 }
2265
2266 /*
2267 * rec_diff is less than 0. In this case, we are shrinking the
2268 * if_broot buffer. It must already exist. If we go to zero
2269 * records, just get rid of the root and clear the status bit.
2270 */
2271 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
Christoph Hellwig60197e82008-10-30 17:11:19 +11002272 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 new_max = cur_max + rec_diff;
2274 ASSERT(new_max >= 0);
2275 if (new_max > 0)
2276 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2277 else
2278 new_size = 0;
2279 if (new_size > 0) {
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002280 new_broot = kmem_alloc(new_size, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 /*
2282 * First copy over the btree block header.
2283 */
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002284 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 } else {
2286 new_broot = NULL;
2287 ifp->if_flags &= ~XFS_IFBROOT;
2288 }
2289
2290 /*
2291 * Only copy the records and pointers if there are any.
2292 */
2293 if (new_max > 0) {
2294 /*
2295 * First copy the records.
2296 */
Christoph Hellwig136341b2008-10-30 17:11:40 +11002297 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2298 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2300
2301 /*
2302 * Then copy the pointers.
2303 */
Christoph Hellwig60197e82008-10-30 17:11:19 +11002304 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 ifp->if_broot_bytes);
Christoph Hellwig60197e82008-10-30 17:11:19 +11002306 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 (int)new_size);
2308 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2309 }
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002310 kmem_free(ifp->if_broot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 ifp->if_broot = new_broot;
2312 ifp->if_broot_bytes = (int)new_size;
2313 ASSERT(ifp->if_broot_bytes <=
2314 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2315 return;
2316}
2317
2318
2319/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 * This is called when the amount of space needed for if_data
2321 * is increased or decreased. The change in size is indicated by
2322 * the number of bytes that need to be added or deleted in the
2323 * byte_diff parameter.
2324 *
2325 * If the amount of space needed has decreased below the size of the
2326 * inline buffer, then switch to using the inline buffer. Otherwise,
2327 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2328 * to what is needed.
2329 *
2330 * ip -- the inode whose if_data area is changing
2331 * byte_diff -- the change in the number of bytes, positive or negative,
2332 * requested for the if_data array.
2333 */
2334void
2335xfs_idata_realloc(
2336 xfs_inode_t *ip,
2337 int byte_diff,
2338 int whichfork)
2339{
2340 xfs_ifork_t *ifp;
2341 int new_size;
2342 int real_size;
2343
2344 if (byte_diff == 0) {
2345 return;
2346 }
2347
2348 ifp = XFS_IFORK_PTR(ip, whichfork);
2349 new_size = (int)ifp->if_bytes + byte_diff;
2350 ASSERT(new_size >= 0);
2351
2352 if (new_size == 0) {
2353 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002354 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 }
2356 ifp->if_u1.if_data = NULL;
2357 real_size = 0;
2358 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2359 /*
2360 * If the valid extents/data can fit in if_inline_ext/data,
2361 * copy them from the malloc'd vector and free it.
2362 */
2363 if (ifp->if_u1.if_data == NULL) {
2364 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2365 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2366 ASSERT(ifp->if_real_bytes != 0);
2367 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2368 new_size);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002369 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2371 }
2372 real_size = 0;
2373 } else {
2374 /*
2375 * Stuck with malloc/realloc.
2376 * For inline data, the underlying buffer must be
2377 * a multiple of 4 bytes in size so that it can be
2378 * logged and stay on word boundaries. We enforce
2379 * that here.
2380 */
2381 real_size = roundup(new_size, 4);
2382 if (ifp->if_u1.if_data == NULL) {
2383 ASSERT(ifp->if_real_bytes == 0);
2384 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2385 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2386 /*
2387 * Only do the realloc if the underlying size
2388 * is really changing.
2389 */
2390 if (ifp->if_real_bytes != real_size) {
2391 ifp->if_u1.if_data =
2392 kmem_realloc(ifp->if_u1.if_data,
2393 real_size,
2394 ifp->if_real_bytes,
2395 KM_SLEEP);
2396 }
2397 } else {
2398 ASSERT(ifp->if_real_bytes == 0);
2399 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2400 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2401 ifp->if_bytes);
2402 }
2403 }
2404 ifp->if_real_bytes = real_size;
2405 ifp->if_bytes = new_size;
2406 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2407}
2408
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409void
2410xfs_idestroy_fork(
2411 xfs_inode_t *ip,
2412 int whichfork)
2413{
2414 xfs_ifork_t *ifp;
2415
2416 ifp = XFS_IFORK_PTR(ip, whichfork);
2417 if (ifp->if_broot != NULL) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002418 kmem_free(ifp->if_broot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 ifp->if_broot = NULL;
2420 }
2421
2422 /*
2423 * If the format is local, then we can't have an extents
2424 * array so just look for an inline data array. If we're
2425 * not local then we may or may not have an extents list,
2426 * so check and free it up if we do.
2427 */
2428 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2429 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2430 (ifp->if_u1.if_data != NULL)) {
2431 ASSERT(ifp->if_real_bytes != 0);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002432 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 ifp->if_u1.if_data = NULL;
2434 ifp->if_real_bytes = 0;
2435 }
2436 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11002437 ((ifp->if_flags & XFS_IFEXTIREC) ||
2438 ((ifp->if_u1.if_extents != NULL) &&
2439 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 ASSERT(ifp->if_real_bytes != 0);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002441 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 }
2443 ASSERT(ifp->if_u1.if_extents == NULL ||
2444 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2445 ASSERT(ifp->if_real_bytes == 0);
2446 if (whichfork == XFS_ATTR_FORK) {
2447 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2448 ip->i_afp = NULL;
2449 }
2450}
2451
2452/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 * Increment the pin count of the given buffer.
2454 * This value is protected by ipinlock spinlock in the mount structure.
2455 */
2456void
2457xfs_ipin(
2458 xfs_inode_t *ip)
2459{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002460 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461
2462 atomic_inc(&ip->i_pincount);
2463}
2464
2465/*
2466 * Decrement the pin count of the given inode, and wake up
2467 * anyone in xfs_iwait_unpin() if the count goes to 0. The
Nathan Scottc41564b2006-03-29 08:55:14 +10002468 * inode must have been previously pinned with a call to xfs_ipin().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 */
2470void
2471xfs_iunpin(
2472 xfs_inode_t *ip)
2473{
2474 ASSERT(atomic_read(&ip->i_pincount) > 0);
2475
David Chinner5d51eff2007-11-23 16:29:18 +11002476 if (atomic_dec_and_test(&ip->i_pincount))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 wake_up(&ip->i_ipin_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478}
2479
2480/*
David Chinnera3f74ff2008-03-06 13:43:42 +11002481 * This is called to unpin an inode. It can be directed to wait or to return
2482 * immediately without waiting for the inode to be unpinned. The caller must
2483 * have the inode locked in at least shared mode so that the buffer cannot be
2484 * subsequently pinned once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002486STATIC void
David Chinnera3f74ff2008-03-06 13:43:42 +11002487__xfs_iunpin_wait(
2488 xfs_inode_t *ip,
2489 int wait)
2490{
2491 xfs_inode_log_item_t *iip = ip->i_itemp;
2492
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002493 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002494 if (atomic_read(&ip->i_pincount) == 0)
2495 return;
2496
2497 /* Give the log a push to start the unpinning I/O */
2498 xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
2499 iip->ili_last_lsn : 0, XFS_LOG_FORCE);
2500 if (wait)
2501 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2502}
2503
2504static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505xfs_iunpin_wait(
2506 xfs_inode_t *ip)
2507{
David Chinnera3f74ff2008-03-06 13:43:42 +11002508 __xfs_iunpin_wait(ip, 1);
2509}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
David Chinnera3f74ff2008-03-06 13:43:42 +11002511static inline void
2512xfs_iunpin_nowait(
2513 xfs_inode_t *ip)
2514{
2515 __xfs_iunpin_wait(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516}
2517
2518
2519/*
2520 * xfs_iextents_copy()
2521 *
2522 * This is called to copy the REAL extents (as opposed to the delayed
2523 * allocation extents) from the inode into the given buffer. It
2524 * returns the number of bytes copied into the buffer.
2525 *
2526 * If there are no delayed allocation extents, then we can just
2527 * memcpy() the extents into the buffer. Otherwise, we need to
2528 * examine each extent in turn and skip those which are delayed.
2529 */
2530int
2531xfs_iextents_copy(
2532 xfs_inode_t *ip,
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002533 xfs_bmbt_rec_t *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 int whichfork)
2535{
2536 int copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 int i;
2538 xfs_ifork_t *ifp;
2539 int nrecs;
2540 xfs_fsblock_t start_block;
2541
2542 ifp = XFS_IFORK_PTR(ip, whichfork);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002543 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 ASSERT(ifp->if_bytes > 0);
2545
2546 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
Eric Sandeen3a59c942007-07-11 11:09:47 +10002547 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 ASSERT(nrecs > 0);
2549
2550 /*
2551 * There are some delayed allocation extents in the
2552 * inode, so copy the extents one at a time and skip
2553 * the delayed ones. There must be at least one
2554 * non-delayed extent.
2555 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 copied = 0;
2557 for (i = 0; i < nrecs; i++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002558 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 start_block = xfs_bmbt_get_startblock(ep);
2560 if (ISNULLSTARTBLOCK(start_block)) {
2561 /*
2562 * It's a delayed allocation extent, so skip it.
2563 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 continue;
2565 }
2566
2567 /* Translate to on disk format */
Christoph Hellwigcd8b0a92007-08-16 16:24:15 +10002568 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2569 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002570 dp++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 copied++;
2572 }
2573 ASSERT(copied != 0);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002574 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575
2576 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2577}
2578
2579/*
2580 * Each of the following cases stores data into the same region
2581 * of the on-disk inode, so only one of them can be valid at
2582 * any given time. While it is possible to have conflicting formats
2583 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2584 * in EXTENTS format, this can only happen when the fork has
2585 * changed formats after being modified but before being flushed.
2586 * In these cases, the format always takes precedence, because the
2587 * format indicates the current state of the fork.
2588 */
2589/*ARGSUSED*/
David Chinnere4ac9672008-04-10 12:23:58 +10002590STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591xfs_iflush_fork(
2592 xfs_inode_t *ip,
2593 xfs_dinode_t *dip,
2594 xfs_inode_log_item_t *iip,
2595 int whichfork,
2596 xfs_buf_t *bp)
2597{
2598 char *cp;
2599 xfs_ifork_t *ifp;
2600 xfs_mount_t *mp;
2601#ifdef XFS_TRANS_DEBUG
2602 int first;
2603#endif
2604 static const short brootflag[2] =
2605 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2606 static const short dataflag[2] =
2607 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2608 static const short extflag[2] =
2609 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2610
David Chinnere4ac9672008-04-10 12:23:58 +10002611 if (!iip)
2612 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 ifp = XFS_IFORK_PTR(ip, whichfork);
2614 /*
2615 * This can happen if we gave up in iformat in an error path,
2616 * for the attribute fork.
2617 */
David Chinnere4ac9672008-04-10 12:23:58 +10002618 if (!ifp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 ASSERT(whichfork == XFS_ATTR_FORK);
David Chinnere4ac9672008-04-10 12:23:58 +10002620 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 }
2622 cp = XFS_DFORK_PTR(dip, whichfork);
2623 mp = ip->i_mount;
2624 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2625 case XFS_DINODE_FMT_LOCAL:
2626 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2627 (ifp->if_bytes > 0)) {
2628 ASSERT(ifp->if_u1.if_data != NULL);
2629 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2630 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 break;
2633
2634 case XFS_DINODE_FMT_EXTENTS:
2635 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2636 !(iip->ili_format.ilf_fields & extflag[whichfork]));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002637 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2638 (ifp->if_bytes == 0));
2639 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2640 (ifp->if_bytes > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2642 (ifp->if_bytes > 0)) {
2643 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2644 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2645 whichfork);
2646 }
2647 break;
2648
2649 case XFS_DINODE_FMT_BTREE:
2650 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2651 (ifp->if_broot_bytes > 0)) {
2652 ASSERT(ifp->if_broot != NULL);
2653 ASSERT(ifp->if_broot_bytes <=
2654 (XFS_IFORK_SIZE(ip, whichfork) +
2655 XFS_BROOT_SIZE_ADJ));
Christoph Hellwig60197e82008-10-30 17:11:19 +11002656 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 (xfs_bmdr_block_t *)cp,
2658 XFS_DFORK_SIZE(dip, mp, whichfork));
2659 }
2660 break;
2661
2662 case XFS_DINODE_FMT_DEV:
2663 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2664 ASSERT(whichfork == XFS_DATA_FORK);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002665 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 }
2667 break;
2668
2669 case XFS_DINODE_FMT_UUID:
2670 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2671 ASSERT(whichfork == XFS_DATA_FORK);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002672 memcpy(XFS_DFORK_DPTR(dip),
2673 &ip->i_df.if_u2.if_uuid,
2674 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 }
2676 break;
2677
2678 default:
2679 ASSERT(0);
2680 break;
2681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682}
2683
David Chinnerbad55842008-03-06 13:43:49 +11002684STATIC int
2685xfs_iflush_cluster(
2686 xfs_inode_t *ip,
2687 xfs_buf_t *bp)
2688{
2689 xfs_mount_t *mp = ip->i_mount;
2690 xfs_perag_t *pag = xfs_get_perag(mp, ip->i_ino);
2691 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10002692 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11002693 int ilist_size;
2694 xfs_inode_t **ilist;
2695 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11002696 int nr_found;
2697 int clcount = 0;
2698 int bufwasdelwri;
2699 int i;
2700
2701 ASSERT(pag->pagi_inodeok);
2702 ASSERT(pag->pag_ici_init);
2703
David Chinnerc8f5f122008-05-20 11:30:15 +10002704 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2705 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10002706 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11002707 if (!ilist)
2708 return 0;
2709
2710 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2711 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2712 read_lock(&pag->pag_ici_lock);
2713 /* really need a gang lookup range call here */
2714 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10002715 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11002716 if (nr_found == 0)
2717 goto out_free;
2718
2719 for (i = 0; i < nr_found; i++) {
2720 iq = ilist[i];
2721 if (iq == ip)
2722 continue;
2723 /* if the inode lies outside this cluster, we're done. */
2724 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
2725 break;
2726 /*
2727 * Do an un-protected check to see if the inode is dirty and
2728 * is a candidate for flushing. These checks will be repeated
2729 * later after the appropriate locks are acquired.
2730 */
David Chinner33540402008-03-06 13:43:59 +11002731 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11002732 continue;
David Chinnerbad55842008-03-06 13:43:49 +11002733
2734 /*
2735 * Try to get locks. If any are unavailable or it is pinned,
2736 * then this inode cannot be flushed and is skipped.
2737 */
2738
2739 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2740 continue;
2741 if (!xfs_iflock_nowait(iq)) {
2742 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2743 continue;
2744 }
2745 if (xfs_ipincount(iq)) {
2746 xfs_ifunlock(iq);
2747 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2748 continue;
2749 }
2750
2751 /*
2752 * arriving here means that this inode can be flushed. First
2753 * re-check that it's dirty before flushing.
2754 */
David Chinner33540402008-03-06 13:43:59 +11002755 if (!xfs_inode_clean(iq)) {
2756 int error;
David Chinnerbad55842008-03-06 13:43:49 +11002757 error = xfs_iflush_int(iq, bp);
2758 if (error) {
2759 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2760 goto cluster_corrupt_out;
2761 }
2762 clcount++;
2763 } else {
2764 xfs_ifunlock(iq);
2765 }
2766 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2767 }
2768
2769 if (clcount) {
2770 XFS_STATS_INC(xs_icluster_flushcnt);
2771 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2772 }
2773
2774out_free:
2775 read_unlock(&pag->pag_ici_lock);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002776 kmem_free(ilist);
David Chinnerbad55842008-03-06 13:43:49 +11002777 return 0;
2778
2779
2780cluster_corrupt_out:
2781 /*
2782 * Corruption detected in the clustering loop. Invalidate the
2783 * inode buffer and shut down the filesystem.
2784 */
2785 read_unlock(&pag->pag_ici_lock);
2786 /*
2787 * Clean up the buffer. If it was B_DELWRI, just release it --
2788 * brelse can handle it with no problems. If not, shut down the
2789 * filesystem before releasing the buffer.
2790 */
2791 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2792 if (bufwasdelwri)
2793 xfs_buf_relse(bp);
2794
2795 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2796
2797 if (!bufwasdelwri) {
2798 /*
2799 * Just like incore_relse: if we have b_iodone functions,
2800 * mark the buffer as an error and call them. Otherwise
2801 * mark it as stale and brelse.
2802 */
2803 if (XFS_BUF_IODONE_FUNC(bp)) {
2804 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
2805 XFS_BUF_UNDONE(bp);
2806 XFS_BUF_STALE(bp);
David Chinnerbad55842008-03-06 13:43:49 +11002807 XFS_BUF_ERROR(bp,EIO);
2808 xfs_biodone(bp);
2809 } else {
2810 XFS_BUF_STALE(bp);
2811 xfs_buf_relse(bp);
2812 }
2813 }
2814
2815 /*
2816 * Unlocks the flush lock
2817 */
2818 xfs_iflush_abort(iq);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002819 kmem_free(ilist);
David Chinnerbad55842008-03-06 13:43:49 +11002820 return XFS_ERROR(EFSCORRUPTED);
2821}
2822
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823/*
2824 * xfs_iflush() will write a modified inode's changes out to the
2825 * inode's on disk home. The caller must have the inode lock held
David Chinnerc63942d2008-08-13 16:41:16 +10002826 * in at least shared mode and the inode flush completion must be
2827 * active as well. The inode lock will still be held upon return from
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 * the call and the caller is free to unlock it.
David Chinnerc63942d2008-08-13 16:41:16 +10002829 * The inode flush will be completed when the inode reaches the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 * The flags indicate how the inode's buffer should be written out.
2831 */
2832int
2833xfs_iflush(
2834 xfs_inode_t *ip,
2835 uint flags)
2836{
2837 xfs_inode_log_item_t *iip;
2838 xfs_buf_t *bp;
2839 xfs_dinode_t *dip;
2840 xfs_mount_t *mp;
2841 int error;
David Chinnera3f74ff2008-03-06 13:43:42 +11002842 int noblock = (flags == XFS_IFLUSH_ASYNC_NOBLOCK);
David Chinnerbad55842008-03-06 13:43:49 +11002843 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844
2845 XFS_STATS_INC(xs_iflush_count);
2846
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002847 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnerc63942d2008-08-13 16:41:16 +10002848 ASSERT(!completion_done(&ip->i_flush));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2850 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2851
2852 iip = ip->i_itemp;
2853 mp = ip->i_mount;
2854
2855 /*
2856 * If the inode isn't dirty, then just release the inode
2857 * flush lock and do nothing.
2858 */
David Chinner33540402008-03-06 13:43:59 +11002859 if (xfs_inode_clean(ip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 xfs_ifunlock(ip);
2861 return 0;
2862 }
2863
2864 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002865 * We can't flush the inode until it is unpinned, so wait for it if we
2866 * are allowed to block. We know noone new can pin it, because we are
2867 * holding the inode lock shared and you need to hold it exclusively to
2868 * pin the inode.
2869 *
2870 * If we are not allowed to block, force the log out asynchronously so
2871 * that when we come back the inode will be unpinned. If other inodes
2872 * in the same cluster are dirty, they will probably write the inode
2873 * out for us if they occur after the log force completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 */
David Chinnera3f74ff2008-03-06 13:43:42 +11002875 if (noblock && xfs_ipincount(ip)) {
2876 xfs_iunpin_nowait(ip);
2877 xfs_ifunlock(ip);
2878 return EAGAIN;
2879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 xfs_iunpin_wait(ip);
2881
2882 /*
2883 * This may have been unpinned because the filesystem is shutting
2884 * down forcibly. If that's the case we must not write this inode
2885 * to disk, because the log record didn't make it to disk!
2886 */
2887 if (XFS_FORCED_SHUTDOWN(mp)) {
2888 ip->i_update_core = 0;
2889 if (iip)
2890 iip->ili_format.ilf_fields = 0;
2891 xfs_ifunlock(ip);
2892 return XFS_ERROR(EIO);
2893 }
2894
2895 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 * Decide how buffer will be flushed out. This is done before
2897 * the call to xfs_iflush_int because this field is zeroed by it.
2898 */
2899 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
2900 /*
2901 * Flush out the inode buffer according to the directions
2902 * of the caller. In the cases where the caller has given
2903 * us a choice choose the non-delwri case. This is because
2904 * the inode is in the AIL and we need to get it out soon.
2905 */
2906 switch (flags) {
2907 case XFS_IFLUSH_SYNC:
2908 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
2909 flags = 0;
2910 break;
David Chinnera3f74ff2008-03-06 13:43:42 +11002911 case XFS_IFLUSH_ASYNC_NOBLOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 case XFS_IFLUSH_ASYNC:
2913 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
2914 flags = INT_ASYNC;
2915 break;
2916 case XFS_IFLUSH_DELWRI:
2917 flags = INT_DELWRI;
2918 break;
2919 default:
2920 ASSERT(0);
2921 flags = 0;
2922 break;
2923 }
2924 } else {
2925 switch (flags) {
2926 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
2927 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
2928 case XFS_IFLUSH_DELWRI:
2929 flags = INT_DELWRI;
2930 break;
David Chinnera3f74ff2008-03-06 13:43:42 +11002931 case XFS_IFLUSH_ASYNC_NOBLOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 case XFS_IFLUSH_ASYNC:
2933 flags = INT_ASYNC;
2934 break;
2935 case XFS_IFLUSH_SYNC:
2936 flags = 0;
2937 break;
2938 default:
2939 ASSERT(0);
2940 flags = 0;
2941 break;
2942 }
2943 }
2944
2945 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002946 * Get the buffer containing the on-disk inode.
2947 */
Christoph Hellwig76d8b272008-11-28 14:23:40 +11002948 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
David Chinnera3f74ff2008-03-06 13:43:42 +11002949 noblock ? XFS_BUF_TRYLOCK : XFS_BUF_LOCK);
2950 if (error || !bp) {
2951 xfs_ifunlock(ip);
2952 return error;
2953 }
2954
2955 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 * First flush out the inode that xfs_iflush was called with.
2957 */
2958 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11002959 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961
2962 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002963 * If the buffer is pinned then push on the log now so we won't
2964 * get stuck waiting in the write for too long.
2965 */
2966 if (XFS_BUF_ISPINNED(bp))
2967 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
2968
2969 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 * inode clustering:
2971 * see if other inodes can be gathered into this write
2972 */
David Chinnerbad55842008-03-06 13:43:49 +11002973 error = xfs_iflush_cluster(ip, bp);
2974 if (error)
2975 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 if (flags & INT_DELWRI) {
2978 xfs_bdwrite(mp, bp);
2979 } else if (flags & INT_ASYNC) {
David Chinnerdb7a19f2008-04-10 12:22:24 +10002980 error = xfs_bawrite(mp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 } else {
2982 error = xfs_bwrite(mp, bp);
2983 }
2984 return error;
2985
2986corrupt_out:
2987 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10002988 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989cluster_corrupt_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 /*
2991 * Unlocks the flush lock
2992 */
David Chinnerbad55842008-03-06 13:43:49 +11002993 xfs_iflush_abort(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 return XFS_ERROR(EFSCORRUPTED);
2995}
2996
2997
2998STATIC int
2999xfs_iflush_int(
3000 xfs_inode_t *ip,
3001 xfs_buf_t *bp)
3002{
3003 xfs_inode_log_item_t *iip;
3004 xfs_dinode_t *dip;
3005 xfs_mount_t *mp;
3006#ifdef XFS_TRANS_DEBUG
3007 int first;
3008#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10003010 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnerc63942d2008-08-13 16:41:16 +10003011 ASSERT(!completion_done(&ip->i_flush));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3013 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3014
3015 iip = ip->i_itemp;
3016 mp = ip->i_mount;
3017
3018
3019 /*
3020 * If the inode isn't dirty, then just release the inode
3021 * flush lock and do nothing.
3022 */
David Chinner33540402008-03-06 13:43:59 +11003023 if (xfs_inode_clean(ip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 xfs_ifunlock(ip);
3025 return 0;
3026 }
3027
3028 /* set *dip = inode's place in the buffer */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11003029 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030
3031 /*
3032 * Clear i_update_core before copying out the data.
3033 * This is for coordination with our timestamp updates
3034 * that don't hold the inode lock. They will always
3035 * update the timestamps BEFORE setting i_update_core,
3036 * so if we clear i_update_core after they set it we
3037 * are guaranteed to see their updates to the timestamps.
3038 * I believe that this depends on strongly ordered memory
3039 * semantics, but we have that. We use the SYNCHRONIZE
3040 * macro to make sure that the compiler does not reorder
3041 * the i_update_core access below the data copy below.
3042 */
3043 ip->i_update_core = 0;
3044 SYNCHRONIZE();
3045
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11003046 /*
3047 * Make sure to get the latest atime from the Linux inode.
3048 */
3049 xfs_synchronize_atime(ip);
3050
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003051 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3053 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3054 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003055 ip->i_ino, be16_to_cpu(dip->di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 goto corrupt_out;
3057 }
3058 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3059 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3060 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3061 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3062 ip->i_ino, ip, ip->i_d.di_magic);
3063 goto corrupt_out;
3064 }
3065 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3066 if (XFS_TEST_ERROR(
3067 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3068 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3069 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3070 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3071 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3072 ip->i_ino, ip);
3073 goto corrupt_out;
3074 }
3075 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3076 if (XFS_TEST_ERROR(
3077 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3078 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3079 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3080 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3081 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3082 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3083 ip->i_ino, ip);
3084 goto corrupt_out;
3085 }
3086 }
3087 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3088 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3089 XFS_RANDOM_IFLUSH_5)) {
3090 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3091 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3092 ip->i_ino,
3093 ip->i_d.di_nextents + ip->i_d.di_anextents,
3094 ip->i_d.di_nblocks,
3095 ip);
3096 goto corrupt_out;
3097 }
3098 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3099 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3100 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3101 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3102 ip->i_ino, ip->i_d.di_forkoff, ip);
3103 goto corrupt_out;
3104 }
3105 /*
3106 * bump the flush iteration count, used to detect flushes which
3107 * postdate a log record during recovery.
3108 */
3109
3110 ip->i_d.di_flushiter++;
3111
3112 /*
3113 * Copy the dirty parts of the inode into the on-disk
3114 * inode. We always copy out the core of the inode,
3115 * because if the inode is dirty at all the core must
3116 * be.
3117 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003118 xfs_dinode_to_disk(dip, &ip->i_d);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119
3120 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3121 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3122 ip->i_d.di_flushiter = 0;
3123
3124 /*
3125 * If this is really an old format inode and the superblock version
3126 * has not been updated to support only new format inodes, then
3127 * convert back to the old inode format. If the superblock version
3128 * has been updated, then make the conversion permanent.
3129 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11003130 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3131 if (ip->i_d.di_version == 1) {
Eric Sandeen62118702008-03-06 13:44:28 +11003132 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133 /*
3134 * Convert it back.
3135 */
3136 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003137 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138 } else {
3139 /*
3140 * The superblock version has already been bumped,
3141 * so just make the conversion to the new inode
3142 * format permanent.
3143 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11003144 ip->i_d.di_version = 2;
3145 dip->di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 ip->i_d.di_onlink = 0;
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003147 dip->di_onlink = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003149 memset(&(dip->di_pad[0]), 0,
3150 sizeof(dip->di_pad));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151 ASSERT(ip->i_d.di_projid == 0);
3152 }
3153 }
3154
David Chinnere4ac9672008-04-10 12:23:58 +10003155 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3156 if (XFS_IFORK_Q(ip))
3157 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 xfs_inobp_check(mp, bp);
3159
3160 /*
3161 * We've recorded everything logged in the inode, so we'd
3162 * like to clear the ilf_fields bits so we don't log and
3163 * flush things unnecessarily. However, we can't stop
3164 * logging all this information until the data we've copied
3165 * into the disk buffer is written to disk. If we did we might
3166 * overwrite the copy of the inode in the log with all the
3167 * data after re-logging only part of it, and in the face of
3168 * a crash we wouldn't have all the data we need to recover.
3169 *
3170 * What we do is move the bits to the ili_last_fields field.
3171 * When logging the inode, these bits are moved back to the
3172 * ilf_fields field. In the xfs_iflush_done() routine we
3173 * clear ili_last_fields, since we know that the information
3174 * those bits represent is permanently on disk. As long as
3175 * the flush completes before the inode is logged again, then
3176 * both ilf_fields and ili_last_fields will be cleared.
3177 *
3178 * We can play with the ilf_fields bits here, because the inode
3179 * lock must be held exclusively in order to set bits there
3180 * and the flush lock protects the ili_last_fields bits.
3181 * Set ili_logged so the flush done
3182 * routine can tell whether or not to look in the AIL.
3183 * Also, store the current LSN of the inode so that we can tell
3184 * whether the item has moved in the AIL from xfs_iflush_done().
3185 * In order to read the lsn we need the AIL lock, because
3186 * it is a 64 bit value that cannot be read atomically.
3187 */
3188 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3189 iip->ili_last_fields = iip->ili_format.ilf_fields;
3190 iip->ili_format.ilf_fields = 0;
3191 iip->ili_logged = 1;
3192
David Chinner7b2e2a32008-10-30 17:39:12 +11003193 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3194 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195
3196 /*
3197 * Attach the function xfs_iflush_done to the inode's
3198 * buffer. This will remove the inode from the AIL
3199 * and unlock the inode's flush lock when the inode is
3200 * completely written to disk.
3201 */
3202 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3203 xfs_iflush_done, (xfs_log_item_t *)iip);
3204
3205 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3206 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3207 } else {
3208 /*
3209 * We're flushing an inode which is not in the AIL and has
3210 * not been logged but has i_update_core set. For this
3211 * case we can use a B_DELWRI flush and immediately drop
3212 * the inode flush lock because we can avoid the whole
3213 * AIL state thing. It's OK to drop the flush lock now,
3214 * because we've already locked the buffer and to do anything
3215 * you really need both.
3216 */
3217 if (iip != NULL) {
3218 ASSERT(iip->ili_logged == 0);
3219 ASSERT(iip->ili_last_fields == 0);
3220 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3221 }
3222 xfs_ifunlock(ip);
3223 }
3224
3225 return 0;
3226
3227corrupt_out:
3228 return XFS_ERROR(EFSCORRUPTED);
3229}
3230
3231
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233#ifdef XFS_ILOCK_TRACE
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234void
3235xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3236{
3237 ktrace_enter(ip->i_lock_trace,
3238 (void *)ip,
3239 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3240 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3241 (void *)ra, /* caller of ilock */
3242 (void *)(unsigned long)current_cpu(),
3243 (void *)(unsigned long)current_pid(),
3244 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3245}
3246#endif
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003247
3248/*
3249 * Return a pointer to the extent record at file index idx.
3250 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003251xfs_bmbt_rec_host_t *
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003252xfs_iext_get_ext(
3253 xfs_ifork_t *ifp, /* inode fork pointer */
3254 xfs_extnum_t idx) /* index of target extent */
3255{
3256 ASSERT(idx >= 0);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003257 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3258 return ifp->if_u1.if_ext_irec->er_extbuf;
3259 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3260 xfs_ext_irec_t *erp; /* irec pointer */
3261 int erp_idx = 0; /* irec index */
3262 xfs_extnum_t page_idx = idx; /* ext index in target list */
3263
3264 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3265 return &erp->er_extbuf[page_idx];
3266 } else if (ifp->if_bytes) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003267 return &ifp->if_u1.if_extents[idx];
3268 } else {
3269 return NULL;
3270 }
3271}
3272
3273/*
3274 * Insert new item(s) into the extent records for incore inode
3275 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3276 */
3277void
3278xfs_iext_insert(
3279 xfs_ifork_t *ifp, /* inode fork pointer */
3280 xfs_extnum_t idx, /* starting index of new items */
3281 xfs_extnum_t count, /* number of inserted items */
3282 xfs_bmbt_irec_t *new) /* items to insert */
3283{
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003284 xfs_extnum_t i; /* extent record index */
3285
3286 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3287 xfs_iext_add(ifp, idx, count);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003288 for (i = idx; i < idx + count; i++, new++)
3289 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003290}
3291
3292/*
3293 * This is called when the amount of space required for incore file
3294 * extents needs to be increased. The ext_diff parameter stores the
3295 * number of new extents being added and the idx parameter contains
3296 * the extent index where the new extents will be added. If the new
3297 * extents are being appended, then we just need to (re)allocate and
3298 * initialize the space. Otherwise, if the new extents are being
3299 * inserted into the middle of the existing entries, a bit more work
3300 * is required to make room for the new extents to be inserted. The
3301 * caller is responsible for filling in the new extent entries upon
3302 * return.
3303 */
3304void
3305xfs_iext_add(
3306 xfs_ifork_t *ifp, /* inode fork pointer */
3307 xfs_extnum_t idx, /* index to begin adding exts */
Nathan Scottc41564b2006-03-29 08:55:14 +10003308 int ext_diff) /* number of extents to add */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003309{
3310 int byte_diff; /* new bytes being added */
3311 int new_size; /* size of extents after adding */
3312 xfs_extnum_t nextents; /* number of extents in file */
3313
3314 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3315 ASSERT((idx >= 0) && (idx <= nextents));
3316 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3317 new_size = ifp->if_bytes + byte_diff;
3318 /*
3319 * If the new number of extents (nextents + ext_diff)
3320 * fits inside the inode, then continue to use the inline
3321 * extent buffer.
3322 */
3323 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3324 if (idx < nextents) {
3325 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3326 &ifp->if_u2.if_inline_ext[idx],
3327 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3328 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3329 }
3330 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3331 ifp->if_real_bytes = 0;
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003332 ifp->if_lastex = nextents + ext_diff;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003333 }
3334 /*
3335 * Otherwise use a linear (direct) extent list.
3336 * If the extents are currently inside the inode,
3337 * xfs_iext_realloc_direct will switch us from
3338 * inline to direct extent allocation mode.
3339 */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003340 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003341 xfs_iext_realloc_direct(ifp, new_size);
3342 if (idx < nextents) {
3343 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3344 &ifp->if_u1.if_extents[idx],
3345 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3346 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3347 }
3348 }
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003349 /* Indirection array */
3350 else {
3351 xfs_ext_irec_t *erp;
3352 int erp_idx = 0;
3353 int page_idx = idx;
3354
3355 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3356 if (ifp->if_flags & XFS_IFEXTIREC) {
3357 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3358 } else {
3359 xfs_iext_irec_init(ifp);
3360 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3361 erp = ifp->if_u1.if_ext_irec;
3362 }
3363 /* Extents fit in target extent page */
3364 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3365 if (page_idx < erp->er_extcount) {
3366 memmove(&erp->er_extbuf[page_idx + ext_diff],
3367 &erp->er_extbuf[page_idx],
3368 (erp->er_extcount - page_idx) *
3369 sizeof(xfs_bmbt_rec_t));
3370 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3371 }
3372 erp->er_extcount += ext_diff;
3373 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3374 }
3375 /* Insert a new extent page */
3376 else if (erp) {
3377 xfs_iext_add_indirect_multi(ifp,
3378 erp_idx, page_idx, ext_diff);
3379 }
3380 /*
3381 * If extent(s) are being appended to the last page in
3382 * the indirection array and the new extent(s) don't fit
3383 * in the page, then erp is NULL and erp_idx is set to
3384 * the next index needed in the indirection array.
3385 */
3386 else {
3387 int count = ext_diff;
3388
3389 while (count) {
3390 erp = xfs_iext_irec_new(ifp, erp_idx);
3391 erp->er_extcount = count;
3392 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3393 if (count) {
3394 erp_idx++;
3395 }
3396 }
3397 }
3398 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003399 ifp->if_bytes = new_size;
3400}
3401
3402/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003403 * This is called when incore extents are being added to the indirection
3404 * array and the new extents do not fit in the target extent list. The
3405 * erp_idx parameter contains the irec index for the target extent list
3406 * in the indirection array, and the idx parameter contains the extent
3407 * index within the list. The number of extents being added is stored
3408 * in the count parameter.
3409 *
3410 * |-------| |-------|
3411 * | | | | idx - number of extents before idx
3412 * | idx | | count |
3413 * | | | | count - number of extents being inserted at idx
3414 * |-------| |-------|
3415 * | count | | nex2 | nex2 - number of extents after idx + count
3416 * |-------| |-------|
3417 */
3418void
3419xfs_iext_add_indirect_multi(
3420 xfs_ifork_t *ifp, /* inode fork pointer */
3421 int erp_idx, /* target extent irec index */
3422 xfs_extnum_t idx, /* index within target list */
3423 int count) /* new extents being added */
3424{
3425 int byte_diff; /* new bytes being added */
3426 xfs_ext_irec_t *erp; /* pointer to irec entry */
3427 xfs_extnum_t ext_diff; /* number of extents to add */
3428 xfs_extnum_t ext_cnt; /* new extents still needed */
3429 xfs_extnum_t nex2; /* extents after idx + count */
3430 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3431 int nlists; /* number of irec's (lists) */
3432
3433 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3434 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3435 nex2 = erp->er_extcount - idx;
3436 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3437
3438 /*
3439 * Save second part of target extent list
3440 * (all extents past */
3441 if (nex2) {
3442 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
David Chinner67850732008-08-13 16:02:51 +10003443 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003444 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3445 erp->er_extcount -= nex2;
3446 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3447 memset(&erp->er_extbuf[idx], 0, byte_diff);
3448 }
3449
3450 /*
3451 * Add the new extents to the end of the target
3452 * list, then allocate new irec record(s) and
3453 * extent buffer(s) as needed to store the rest
3454 * of the new extents.
3455 */
3456 ext_cnt = count;
3457 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3458 if (ext_diff) {
3459 erp->er_extcount += ext_diff;
3460 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3461 ext_cnt -= ext_diff;
3462 }
3463 while (ext_cnt) {
3464 erp_idx++;
3465 erp = xfs_iext_irec_new(ifp, erp_idx);
3466 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3467 erp->er_extcount = ext_diff;
3468 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3469 ext_cnt -= ext_diff;
3470 }
3471
3472 /* Add nex2 extents back to indirection array */
3473 if (nex2) {
3474 xfs_extnum_t ext_avail;
3475 int i;
3476
3477 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3478 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3479 i = 0;
3480 /*
3481 * If nex2 extents fit in the current page, append
3482 * nex2_ep after the new extents.
3483 */
3484 if (nex2 <= ext_avail) {
3485 i = erp->er_extcount;
3486 }
3487 /*
3488 * Otherwise, check if space is available in the
3489 * next page.
3490 */
3491 else if ((erp_idx < nlists - 1) &&
3492 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3493 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3494 erp_idx++;
3495 erp++;
3496 /* Create a hole for nex2 extents */
3497 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3498 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3499 }
3500 /*
3501 * Final choice, create a new extent page for
3502 * nex2 extents.
3503 */
3504 else {
3505 erp_idx++;
3506 erp = xfs_iext_irec_new(ifp, erp_idx);
3507 }
3508 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003509 kmem_free(nex2_ep);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003510 erp->er_extcount += nex2;
3511 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3512 }
3513}
3514
3515/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003516 * This is called when the amount of space required for incore file
3517 * extents needs to be decreased. The ext_diff parameter stores the
3518 * number of extents to be removed and the idx parameter contains
3519 * the extent index where the extents will be removed from.
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003520 *
3521 * If the amount of space needed has decreased below the linear
3522 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3523 * extent array. Otherwise, use kmem_realloc() to adjust the
3524 * size to what is needed.
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003525 */
3526void
3527xfs_iext_remove(
3528 xfs_ifork_t *ifp, /* inode fork pointer */
3529 xfs_extnum_t idx, /* index to begin removing exts */
3530 int ext_diff) /* number of extents to remove */
3531{
3532 xfs_extnum_t nextents; /* number of extents in file */
3533 int new_size; /* size of extents after removal */
3534
3535 ASSERT(ext_diff > 0);
3536 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3537 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3538
3539 if (new_size == 0) {
3540 xfs_iext_destroy(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003541 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3542 xfs_iext_remove_indirect(ifp, idx, ext_diff);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003543 } else if (ifp->if_real_bytes) {
3544 xfs_iext_remove_direct(ifp, idx, ext_diff);
3545 } else {
3546 xfs_iext_remove_inline(ifp, idx, ext_diff);
3547 }
3548 ifp->if_bytes = new_size;
3549}
3550
3551/*
3552 * This removes ext_diff extents from the inline buffer, beginning
3553 * at extent index idx.
3554 */
3555void
3556xfs_iext_remove_inline(
3557 xfs_ifork_t *ifp, /* inode fork pointer */
3558 xfs_extnum_t idx, /* index to begin removing exts */
3559 int ext_diff) /* number of extents to remove */
3560{
3561 int nextents; /* number of extents in file */
3562
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003563 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003564 ASSERT(idx < XFS_INLINE_EXTS);
3565 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3566 ASSERT(((nextents - ext_diff) > 0) &&
3567 (nextents - ext_diff) < XFS_INLINE_EXTS);
3568
3569 if (idx + ext_diff < nextents) {
3570 memmove(&ifp->if_u2.if_inline_ext[idx],
3571 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3572 (nextents - (idx + ext_diff)) *
3573 sizeof(xfs_bmbt_rec_t));
3574 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3575 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3576 } else {
3577 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3578 ext_diff * sizeof(xfs_bmbt_rec_t));
3579 }
3580}
3581
3582/*
3583 * This removes ext_diff extents from a linear (direct) extent list,
3584 * beginning at extent index idx. If the extents are being removed
3585 * from the end of the list (ie. truncate) then we just need to re-
3586 * allocate the list to remove the extra space. Otherwise, if the
3587 * extents are being removed from the middle of the existing extent
3588 * entries, then we first need to move the extent records beginning
3589 * at idx + ext_diff up in the list to overwrite the records being
3590 * removed, then remove the extra space via kmem_realloc.
3591 */
3592void
3593xfs_iext_remove_direct(
3594 xfs_ifork_t *ifp, /* inode fork pointer */
3595 xfs_extnum_t idx, /* index to begin removing exts */
3596 int ext_diff) /* number of extents to remove */
3597{
3598 xfs_extnum_t nextents; /* number of extents in file */
3599 int new_size; /* size of extents after removal */
3600
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003601 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003602 new_size = ifp->if_bytes -
3603 (ext_diff * sizeof(xfs_bmbt_rec_t));
3604 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3605
3606 if (new_size == 0) {
3607 xfs_iext_destroy(ifp);
3608 return;
3609 }
3610 /* Move extents up in the list (if needed) */
3611 if (idx + ext_diff < nextents) {
3612 memmove(&ifp->if_u1.if_extents[idx],
3613 &ifp->if_u1.if_extents[idx + ext_diff],
3614 (nextents - (idx + ext_diff)) *
3615 sizeof(xfs_bmbt_rec_t));
3616 }
3617 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3618 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3619 /*
3620 * Reallocate the direct extent list. If the extents
3621 * will fit inside the inode then xfs_iext_realloc_direct
3622 * will switch from direct to inline extent allocation
3623 * mode for us.
3624 */
3625 xfs_iext_realloc_direct(ifp, new_size);
3626 ifp->if_bytes = new_size;
3627}
3628
3629/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003630 * This is called when incore extents are being removed from the
3631 * indirection array and the extents being removed span multiple extent
3632 * buffers. The idx parameter contains the file extent index where we
3633 * want to begin removing extents, and the count parameter contains
3634 * how many extents need to be removed.
3635 *
3636 * |-------| |-------|
3637 * | nex1 | | | nex1 - number of extents before idx
3638 * |-------| | count |
3639 * | | | | count - number of extents being removed at idx
3640 * | count | |-------|
3641 * | | | nex2 | nex2 - number of extents after idx + count
3642 * |-------| |-------|
3643 */
3644void
3645xfs_iext_remove_indirect(
3646 xfs_ifork_t *ifp, /* inode fork pointer */
3647 xfs_extnum_t idx, /* index to begin removing extents */
3648 int count) /* number of extents to remove */
3649{
3650 xfs_ext_irec_t *erp; /* indirection array pointer */
3651 int erp_idx = 0; /* indirection array index */
3652 xfs_extnum_t ext_cnt; /* extents left to remove */
3653 xfs_extnum_t ext_diff; /* extents to remove in current list */
3654 xfs_extnum_t nex1; /* number of extents before idx */
3655 xfs_extnum_t nex2; /* extents after idx + count */
Nathan Scottc41564b2006-03-29 08:55:14 +10003656 int nlists; /* entries in indirection array */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003657 int page_idx = idx; /* index in target extent list */
3658
3659 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3660 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3661 ASSERT(erp != NULL);
3662 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3663 nex1 = page_idx;
3664 ext_cnt = count;
3665 while (ext_cnt) {
3666 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3667 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3668 /*
3669 * Check for deletion of entire list;
3670 * xfs_iext_irec_remove() updates extent offsets.
3671 */
3672 if (ext_diff == erp->er_extcount) {
3673 xfs_iext_irec_remove(ifp, erp_idx);
3674 ext_cnt -= ext_diff;
3675 nex1 = 0;
3676 if (ext_cnt) {
3677 ASSERT(erp_idx < ifp->if_real_bytes /
3678 XFS_IEXT_BUFSZ);
3679 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3680 nex1 = 0;
3681 continue;
3682 } else {
3683 break;
3684 }
3685 }
3686 /* Move extents up (if needed) */
3687 if (nex2) {
3688 memmove(&erp->er_extbuf[nex1],
3689 &erp->er_extbuf[nex1 + ext_diff],
3690 nex2 * sizeof(xfs_bmbt_rec_t));
3691 }
3692 /* Zero out rest of page */
3693 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3694 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3695 /* Update remaining counters */
3696 erp->er_extcount -= ext_diff;
3697 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3698 ext_cnt -= ext_diff;
3699 nex1 = 0;
3700 erp_idx++;
3701 erp++;
3702 }
3703 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3704 xfs_iext_irec_compact(ifp);
3705}
3706
3707/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003708 * Create, destroy, or resize a linear (direct) block of extents.
3709 */
3710void
3711xfs_iext_realloc_direct(
3712 xfs_ifork_t *ifp, /* inode fork pointer */
3713 int new_size) /* new size of extents */
3714{
3715 int rnew_size; /* real new size of extents */
3716
3717 rnew_size = new_size;
3718
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003719 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3720 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3721 (new_size != ifp->if_real_bytes)));
3722
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003723 /* Free extent records */
3724 if (new_size == 0) {
3725 xfs_iext_destroy(ifp);
3726 }
3727 /* Resize direct extent list and zero any new bytes */
3728 else if (ifp->if_real_bytes) {
3729 /* Check if extents will fit inside the inode */
3730 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3731 xfs_iext_direct_to_inline(ifp, new_size /
3732 (uint)sizeof(xfs_bmbt_rec_t));
3733 ifp->if_bytes = new_size;
3734 return;
3735 }
Vignesh Babu16a087d2007-06-28 16:46:37 +10003736 if (!is_power_of_2(new_size)){
Robert P. J. Day40ebd812007-11-23 16:30:51 +11003737 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003738 }
3739 if (rnew_size != ifp->if_real_bytes) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003740 ifp->if_u1.if_extents =
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003741 kmem_realloc(ifp->if_u1.if_extents,
3742 rnew_size,
David Chinner67850732008-08-13 16:02:51 +10003743 ifp->if_real_bytes, KM_NOFS);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003744 }
3745 if (rnew_size > ifp->if_real_bytes) {
3746 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3747 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3748 rnew_size - ifp->if_real_bytes);
3749 }
3750 }
3751 /*
3752 * Switch from the inline extent buffer to a direct
3753 * extent list. Be sure to include the inline extent
3754 * bytes in new_size.
3755 */
3756 else {
3757 new_size += ifp->if_bytes;
Vignesh Babu16a087d2007-06-28 16:46:37 +10003758 if (!is_power_of_2(new_size)) {
Robert P. J. Day40ebd812007-11-23 16:30:51 +11003759 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003760 }
3761 xfs_iext_inline_to_direct(ifp, rnew_size);
3762 }
3763 ifp->if_real_bytes = rnew_size;
3764 ifp->if_bytes = new_size;
3765}
3766
3767/*
3768 * Switch from linear (direct) extent records to inline buffer.
3769 */
3770void
3771xfs_iext_direct_to_inline(
3772 xfs_ifork_t *ifp, /* inode fork pointer */
3773 xfs_extnum_t nextents) /* number of extents in file */
3774{
3775 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3776 ASSERT(nextents <= XFS_INLINE_EXTS);
3777 /*
3778 * The inline buffer was zeroed when we switched
3779 * from inline to direct extent allocation mode,
3780 * so we don't need to clear it here.
3781 */
3782 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3783 nextents * sizeof(xfs_bmbt_rec_t));
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003784 kmem_free(ifp->if_u1.if_extents);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003785 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3786 ifp->if_real_bytes = 0;
3787}
3788
3789/*
3790 * Switch from inline buffer to linear (direct) extent records.
3791 * new_size should already be rounded up to the next power of 2
3792 * by the caller (when appropriate), so use new_size as it is.
3793 * However, since new_size may be rounded up, we can't update
3794 * if_bytes here. It is the caller's responsibility to update
3795 * if_bytes upon return.
3796 */
3797void
3798xfs_iext_inline_to_direct(
3799 xfs_ifork_t *ifp, /* inode fork pointer */
3800 int new_size) /* number of extents in file */
3801{
David Chinner67850732008-08-13 16:02:51 +10003802 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003803 memset(ifp->if_u1.if_extents, 0, new_size);
3804 if (ifp->if_bytes) {
3805 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3806 ifp->if_bytes);
3807 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3808 sizeof(xfs_bmbt_rec_t));
3809 }
3810 ifp->if_real_bytes = new_size;
3811}
3812
3813/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003814 * Resize an extent indirection array to new_size bytes.
3815 */
3816void
3817xfs_iext_realloc_indirect(
3818 xfs_ifork_t *ifp, /* inode fork pointer */
3819 int new_size) /* new indirection array size */
3820{
3821 int nlists; /* number of irec's (ex lists) */
3822 int size; /* current indirection array size */
3823
3824 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3825 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3826 size = nlists * sizeof(xfs_ext_irec_t);
3827 ASSERT(ifp->if_real_bytes);
3828 ASSERT((new_size >= 0) && (new_size != size));
3829 if (new_size == 0) {
3830 xfs_iext_destroy(ifp);
3831 } else {
3832 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3833 kmem_realloc(ifp->if_u1.if_ext_irec,
David Chinner67850732008-08-13 16:02:51 +10003834 new_size, size, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003835 }
3836}
3837
3838/*
3839 * Switch from indirection array to linear (direct) extent allocations.
3840 */
3841void
3842xfs_iext_indirect_to_direct(
3843 xfs_ifork_t *ifp) /* inode fork pointer */
3844{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003845 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003846 xfs_extnum_t nextents; /* number of extents in file */
3847 int size; /* size of file extents */
3848
3849 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3850 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3851 ASSERT(nextents <= XFS_LINEAR_EXTS);
3852 size = nextents * sizeof(xfs_bmbt_rec_t);
3853
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10003854 xfs_iext_irec_compact_pages(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003855 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3856
3857 ep = ifp->if_u1.if_ext_irec->er_extbuf;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003858 kmem_free(ifp->if_u1.if_ext_irec);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003859 ifp->if_flags &= ~XFS_IFEXTIREC;
3860 ifp->if_u1.if_extents = ep;
3861 ifp->if_bytes = size;
3862 if (nextents < XFS_LINEAR_EXTS) {
3863 xfs_iext_realloc_direct(ifp, size);
3864 }
3865}
3866
3867/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003868 * Free incore file extents.
3869 */
3870void
3871xfs_iext_destroy(
3872 xfs_ifork_t *ifp) /* inode fork pointer */
3873{
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003874 if (ifp->if_flags & XFS_IFEXTIREC) {
3875 int erp_idx;
3876 int nlists;
3877
3878 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3879 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3880 xfs_iext_irec_remove(ifp, erp_idx);
3881 }
3882 ifp->if_flags &= ~XFS_IFEXTIREC;
3883 } else if (ifp->if_real_bytes) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003884 kmem_free(ifp->if_u1.if_extents);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003885 } else if (ifp->if_bytes) {
3886 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3887 sizeof(xfs_bmbt_rec_t));
3888 }
3889 ifp->if_u1.if_extents = NULL;
3890 ifp->if_real_bytes = 0;
3891 ifp->if_bytes = 0;
3892}
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003893
3894/*
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003895 * Return a pointer to the extent record for file system block bno.
3896 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003897xfs_bmbt_rec_host_t * /* pointer to found extent record */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003898xfs_iext_bno_to_ext(
3899 xfs_ifork_t *ifp, /* inode fork pointer */
3900 xfs_fileoff_t bno, /* block number to search for */
3901 xfs_extnum_t *idxp) /* index of target extent */
3902{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003903 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003904 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003905 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003906 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
Nathan Scottc41564b2006-03-29 08:55:14 +10003907 int high; /* upper boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003908 xfs_extnum_t idx = 0; /* index of target extent */
Nathan Scottc41564b2006-03-29 08:55:14 +10003909 int low; /* lower boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003910 xfs_extnum_t nextents; /* number of file extents */
3911 xfs_fileoff_t startoff = 0; /* start offset of extent */
3912
3913 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3914 if (nextents == 0) {
3915 *idxp = 0;
3916 return NULL;
3917 }
3918 low = 0;
3919 if (ifp->if_flags & XFS_IFEXTIREC) {
3920 /* Find target extent list */
3921 int erp_idx = 0;
3922 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3923 base = erp->er_extbuf;
3924 high = erp->er_extcount - 1;
3925 } else {
3926 base = ifp->if_u1.if_extents;
3927 high = nextents - 1;
3928 }
3929 /* Binary search extent records */
3930 while (low <= high) {
3931 idx = (low + high) >> 1;
3932 ep = base + idx;
3933 startoff = xfs_bmbt_get_startoff(ep);
3934 blockcount = xfs_bmbt_get_blockcount(ep);
3935 if (bno < startoff) {
3936 high = idx - 1;
3937 } else if (bno >= startoff + blockcount) {
3938 low = idx + 1;
3939 } else {
3940 /* Convert back to file-based extent index */
3941 if (ifp->if_flags & XFS_IFEXTIREC) {
3942 idx += erp->er_extoff;
3943 }
3944 *idxp = idx;
3945 return ep;
3946 }
3947 }
3948 /* Convert back to file-based extent index */
3949 if (ifp->if_flags & XFS_IFEXTIREC) {
3950 idx += erp->er_extoff;
3951 }
3952 if (bno >= startoff + blockcount) {
3953 if (++idx == nextents) {
3954 ep = NULL;
3955 } else {
3956 ep = xfs_iext_get_ext(ifp, idx);
3957 }
3958 }
3959 *idxp = idx;
3960 return ep;
3961}
3962
3963/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003964 * Return a pointer to the indirection array entry containing the
3965 * extent record for filesystem block bno. Store the index of the
3966 * target irec in *erp_idxp.
3967 */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003968xfs_ext_irec_t * /* pointer to found extent record */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003969xfs_iext_bno_to_irec(
3970 xfs_ifork_t *ifp, /* inode fork pointer */
3971 xfs_fileoff_t bno, /* block number to search for */
3972 int *erp_idxp) /* irec index of target ext list */
3973{
3974 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3975 xfs_ext_irec_t *erp_next; /* next indirection array entry */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003976 int erp_idx; /* indirection array index */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003977 int nlists; /* number of extent irec's (lists) */
3978 int high; /* binary search upper limit */
3979 int low; /* binary search lower limit */
3980
3981 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3982 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3983 erp_idx = 0;
3984 low = 0;
3985 high = nlists - 1;
3986 while (low <= high) {
3987 erp_idx = (low + high) >> 1;
3988 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3989 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3990 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3991 high = erp_idx - 1;
3992 } else if (erp_next && bno >=
3993 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3994 low = erp_idx + 1;
3995 } else {
3996 break;
3997 }
3998 }
3999 *erp_idxp = erp_idx;
4000 return erp;
4001}
4002
4003/*
4004 * Return a pointer to the indirection array entry containing the
4005 * extent record at file extent index *idxp. Store the index of the
4006 * target irec in *erp_idxp and store the page index of the target
4007 * extent record in *idxp.
4008 */
4009xfs_ext_irec_t *
4010xfs_iext_idx_to_irec(
4011 xfs_ifork_t *ifp, /* inode fork pointer */
4012 xfs_extnum_t *idxp, /* extent index (file -> page) */
4013 int *erp_idxp, /* pointer to target irec */
4014 int realloc) /* new bytes were just added */
4015{
4016 xfs_ext_irec_t *prev; /* pointer to previous irec */
4017 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4018 int erp_idx; /* indirection array index */
4019 int nlists; /* number of irec's (ex lists) */
4020 int high; /* binary search upper limit */
4021 int low; /* binary search lower limit */
4022 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4023
4024 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4025 ASSERT(page_idx >= 0 && page_idx <=
4026 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4027 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4028 erp_idx = 0;
4029 low = 0;
4030 high = nlists - 1;
4031
4032 /* Binary search extent irec's */
4033 while (low <= high) {
4034 erp_idx = (low + high) >> 1;
4035 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4036 prev = erp_idx > 0 ? erp - 1 : NULL;
4037 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4038 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4039 high = erp_idx - 1;
4040 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4041 (page_idx == erp->er_extoff + erp->er_extcount &&
4042 !realloc)) {
4043 low = erp_idx + 1;
4044 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4045 erp->er_extcount == XFS_LINEAR_EXTS) {
4046 ASSERT(realloc);
4047 page_idx = 0;
4048 erp_idx++;
4049 erp = erp_idx < nlists ? erp + 1 : NULL;
4050 break;
4051 } else {
4052 page_idx -= erp->er_extoff;
4053 break;
4054 }
4055 }
4056 *idxp = page_idx;
4057 *erp_idxp = erp_idx;
4058 return(erp);
4059}
4060
4061/*
4062 * Allocate and initialize an indirection array once the space needed
4063 * for incore extents increases above XFS_IEXT_BUFSZ.
4064 */
4065void
4066xfs_iext_irec_init(
4067 xfs_ifork_t *ifp) /* inode fork pointer */
4068{
4069 xfs_ext_irec_t *erp; /* indirection array pointer */
4070 xfs_extnum_t nextents; /* number of extents in file */
4071
4072 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4073 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4074 ASSERT(nextents <= XFS_LINEAR_EXTS);
4075
David Chinner67850732008-08-13 16:02:51 +10004076 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004077
4078 if (nextents == 0) {
David Chinner67850732008-08-13 16:02:51 +10004079 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004080 } else if (!ifp->if_real_bytes) {
4081 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4082 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4083 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4084 }
4085 erp->er_extbuf = ifp->if_u1.if_extents;
4086 erp->er_extcount = nextents;
4087 erp->er_extoff = 0;
4088
4089 ifp->if_flags |= XFS_IFEXTIREC;
4090 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4091 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4092 ifp->if_u1.if_ext_irec = erp;
4093
4094 return;
4095}
4096
4097/*
4098 * Allocate and initialize a new entry in the indirection array.
4099 */
4100xfs_ext_irec_t *
4101xfs_iext_irec_new(
4102 xfs_ifork_t *ifp, /* inode fork pointer */
4103 int erp_idx) /* index for new irec */
4104{
4105 xfs_ext_irec_t *erp; /* indirection array pointer */
4106 int i; /* loop counter */
4107 int nlists; /* number of irec's (ex lists) */
4108
4109 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4110 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4111
4112 /* Resize indirection array */
4113 xfs_iext_realloc_indirect(ifp, ++nlists *
4114 sizeof(xfs_ext_irec_t));
4115 /*
4116 * Move records down in the array so the
4117 * new page can use erp_idx.
4118 */
4119 erp = ifp->if_u1.if_ext_irec;
4120 for (i = nlists - 1; i > erp_idx; i--) {
4121 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4122 }
4123 ASSERT(i == erp_idx);
4124
4125 /* Initialize new extent record */
4126 erp = ifp->if_u1.if_ext_irec;
David Chinner67850732008-08-13 16:02:51 +10004127 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004128 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4129 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4130 erp[erp_idx].er_extcount = 0;
4131 erp[erp_idx].er_extoff = erp_idx > 0 ?
4132 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4133 return (&erp[erp_idx]);
4134}
4135
4136/*
4137 * Remove a record from the indirection array.
4138 */
4139void
4140xfs_iext_irec_remove(
4141 xfs_ifork_t *ifp, /* inode fork pointer */
4142 int erp_idx) /* irec index to remove */
4143{
4144 xfs_ext_irec_t *erp; /* indirection array pointer */
4145 int i; /* loop counter */
4146 int nlists; /* number of irec's (ex lists) */
4147
4148 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4149 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4150 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4151 if (erp->er_extbuf) {
4152 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4153 -erp->er_extcount);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004154 kmem_free(erp->er_extbuf);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004155 }
4156 /* Compact extent records */
4157 erp = ifp->if_u1.if_ext_irec;
4158 for (i = erp_idx; i < nlists - 1; i++) {
4159 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4160 }
4161 /*
4162 * Manually free the last extent record from the indirection
4163 * array. A call to xfs_iext_realloc_indirect() with a size
4164 * of zero would result in a call to xfs_iext_destroy() which
4165 * would in turn call this function again, creating a nasty
4166 * infinite loop.
4167 */
4168 if (--nlists) {
4169 xfs_iext_realloc_indirect(ifp,
4170 nlists * sizeof(xfs_ext_irec_t));
4171 } else {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004172 kmem_free(ifp->if_u1.if_ext_irec);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004173 }
4174 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4175}
4176
4177/*
4178 * This is called to clean up large amounts of unused memory allocated
4179 * by the indirection array. Before compacting anything though, verify
4180 * that the indirection array is still needed and switch back to the
4181 * linear extent list (or even the inline buffer) if possible. The
4182 * compaction policy is as follows:
4183 *
4184 * Full Compaction: Extents fit into a single page (or inline buffer)
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10004185 * Partial Compaction: Extents occupy less than 50% of allocated space
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004186 * No Compaction: Extents occupy at least 50% of allocated space
4187 */
4188void
4189xfs_iext_irec_compact(
4190 xfs_ifork_t *ifp) /* inode fork pointer */
4191{
4192 xfs_extnum_t nextents; /* number of extents in file */
4193 int nlists; /* number of irec's (ex lists) */
4194
4195 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4196 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4197 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4198
4199 if (nextents == 0) {
4200 xfs_iext_destroy(ifp);
4201 } else if (nextents <= XFS_INLINE_EXTS) {
4202 xfs_iext_indirect_to_direct(ifp);
4203 xfs_iext_direct_to_inline(ifp, nextents);
4204 } else if (nextents <= XFS_LINEAR_EXTS) {
4205 xfs_iext_indirect_to_direct(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004206 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4207 xfs_iext_irec_compact_pages(ifp);
4208 }
4209}
4210
4211/*
4212 * Combine extents from neighboring extent pages.
4213 */
4214void
4215xfs_iext_irec_compact_pages(
4216 xfs_ifork_t *ifp) /* inode fork pointer */
4217{
4218 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4219 int erp_idx = 0; /* indirection array index */
4220 int nlists; /* number of irec's (ex lists) */
4221
4222 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4223 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4224 while (erp_idx < nlists - 1) {
4225 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4226 erp_next = erp + 1;
4227 if (erp_next->er_extcount <=
4228 (XFS_LINEAR_EXTS - erp->er_extcount)) {
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10004229 memcpy(&erp->er_extbuf[erp->er_extcount],
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004230 erp_next->er_extbuf, erp_next->er_extcount *
4231 sizeof(xfs_bmbt_rec_t));
4232 erp->er_extcount += erp_next->er_extcount;
4233 /*
4234 * Free page before removing extent record
4235 * so er_extoffs don't get modified in
4236 * xfs_iext_irec_remove.
4237 */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004238 kmem_free(erp_next->er_extbuf);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004239 erp_next->er_extbuf = NULL;
4240 xfs_iext_irec_remove(ifp, erp_idx + 1);
4241 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4242 } else {
4243 erp_idx++;
4244 }
4245 }
4246}
4247
4248/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004249 * This is called to update the er_extoff field in the indirection
4250 * array when extents have been added or removed from one of the
4251 * extent lists. erp_idx contains the irec index to begin updating
4252 * at and ext_diff contains the number of extents that were added
4253 * or removed.
4254 */
4255void
4256xfs_iext_irec_update_extoffs(
4257 xfs_ifork_t *ifp, /* inode fork pointer */
4258 int erp_idx, /* irec index to update */
4259 int ext_diff) /* number of new extents */
4260{
4261 int i; /* loop counter */
4262 int nlists; /* number of irec's (ex lists */
4263
4264 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4265 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4266 for (i = erp_idx; i < nlists; i++) {
4267 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
4268 }
4269}