blob: 8cd6e8d8fe9c452e1d7c995d216a4f5745c61f53 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include "xfs_quota.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100051#include "xfs_filestream.h"
Christoph Hellwig739bfb22007-08-29 10:58:01 +100052#include "xfs_vnodeops.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000053#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055kmem_zone_t *xfs_ifork_zone;
56kmem_zone_t *xfs_inode_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
58/*
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
61 */
62#define XFS_ITRUNC_MAX_EXTENTS 2
63
64STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
65STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
66STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
67STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
68
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#ifdef DEBUG
70/*
71 * Make sure that the extents in the given memory buffer
72 * are valid.
73 */
74STATIC void
75xfs_validate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +110076 xfs_ifork_t *ifp,
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 int nrecs,
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 xfs_exntfmt_t fmt)
79{
80 xfs_bmbt_irec_t irec;
Christoph Hellwiga6f64d42007-08-16 16:23:40 +100081 xfs_bmbt_rec_host_t rec;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 int i;
83
84 for (i = 0; i < nrecs; i++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +100085 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
86 rec.l0 = get_unaligned(&ep->l0);
87 rec.l1 = get_unaligned(&ep->l1);
88 xfs_bmbt_get_all(&rec, &irec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 if (fmt == XFS_EXTFMT_NOSTATE)
90 ASSERT(irec.br_state == XFS_EXT_NORM);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 }
92}
93#else /* DEBUG */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +100094#define xfs_validate_extents(ifp, nrecs, fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095#endif /* DEBUG */
96
97/*
98 * Check that none of the inode's in the buffer have a next
99 * unlinked field of 0.
100 */
101#if defined(DEBUG)
102void
103xfs_inobp_check(
104 xfs_mount_t *mp,
105 xfs_buf_t *bp)
106{
107 int i;
108 int j;
109 xfs_dinode_t *dip;
110
111 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
112
113 for (i = 0; i < j; i++) {
114 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
115 i * mp->m_sb.sb_inodesize);
116 if (!dip->di_next_unlinked) {
117 xfs_fs_cmn_err(CE_ALERT, mp,
118 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
119 bp);
120 ASSERT(dip->di_next_unlinked);
121 }
122 }
123}
124#endif
125
126/*
David Chinner4ae29b42008-03-06 13:43:34 +1100127 * Find the buffer associated with the given inode map
128 * We do basic validation checks on the buffer once it has been
129 * retrieved from disk.
130 */
131STATIC int
132xfs_imap_to_bp(
133 xfs_mount_t *mp,
134 xfs_trans_t *tp,
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100135 struct xfs_imap *imap,
David Chinner4ae29b42008-03-06 13:43:34 +1100136 xfs_buf_t **bpp,
137 uint buf_flags,
Christoph Hellwigb48d8d62008-11-28 14:23:41 +1100138 uint iget_flags)
David Chinner4ae29b42008-03-06 13:43:34 +1100139{
140 int error;
141 int i;
142 int ni;
143 xfs_buf_t *bp;
144
145 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
David Chinnera3f74ff2008-03-06 13:43:42 +1100146 (int)imap->im_len, buf_flags, &bp);
David Chinner4ae29b42008-03-06 13:43:34 +1100147 if (error) {
David Chinnera3f74ff2008-03-06 13:43:42 +1100148 if (error != EAGAIN) {
149 cmn_err(CE_WARN,
150 "xfs_imap_to_bp: xfs_trans_read_buf()returned "
David Chinner4ae29b42008-03-06 13:43:34 +1100151 "an error %d on %s. Returning error.",
152 error, mp->m_fsname);
David Chinnera3f74ff2008-03-06 13:43:42 +1100153 } else {
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000154 ASSERT(buf_flags & XBF_TRYLOCK);
David Chinnera3f74ff2008-03-06 13:43:42 +1100155 }
David Chinner4ae29b42008-03-06 13:43:34 +1100156 return error;
157 }
158
159 /*
160 * Validate the magic number and version of every inode in the buffer
161 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
162 */
163#ifdef DEBUG
164 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
165#else /* usual case */
166 ni = 1;
167#endif
168
169 for (i = 0; i < ni; i++) {
170 int di_ok;
171 xfs_dinode_t *dip;
172
173 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
174 (i << mp->m_sb.sb_inodelog));
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100175 di_ok = be16_to_cpu(dip->di_magic) == XFS_DINODE_MAGIC &&
176 XFS_DINODE_GOOD_VERSION(dip->di_version);
David Chinner4ae29b42008-03-06 13:43:34 +1100177 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
178 XFS_ERRTAG_ITOBP_INOTOBP,
179 XFS_RANDOM_ITOBP_INOTOBP))) {
Christoph Hellwigb48d8d62008-11-28 14:23:41 +1100180 if (iget_flags & XFS_IGET_BULKSTAT) {
David Chinner4ae29b42008-03-06 13:43:34 +1100181 xfs_trans_brelse(tp, bp);
182 return XFS_ERROR(EINVAL);
183 }
184 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
185 XFS_ERRLEVEL_HIGH, mp, dip);
186#ifdef DEBUG
187 cmn_err(CE_PANIC,
188 "Device %s - bad inode magic/vsn "
189 "daddr %lld #%d (magic=%x)",
190 XFS_BUFTARG_NAME(mp->m_ddev_targp),
191 (unsigned long long)imap->im_blkno, i,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100192 be16_to_cpu(dip->di_magic));
David Chinner4ae29b42008-03-06 13:43:34 +1100193#endif
194 xfs_trans_brelse(tp, bp);
195 return XFS_ERROR(EFSCORRUPTED);
196 }
197 }
198
199 xfs_inobp_check(mp, bp);
200
201 /*
202 * Mark the buffer as an inode buffer now that it looks good
203 */
204 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
205
206 *bpp = bp;
207 return 0;
208}
209
210/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 * This routine is called to map an inode number within a file
212 * system to the buffer containing the on-disk version of the
213 * inode. It returns a pointer to the buffer containing the
214 * on-disk inode in the bpp parameter, and in the dip parameter
215 * it returns a pointer to the on-disk inode within that buffer.
216 *
217 * If a non-zero error is returned, then the contents of bpp and
218 * dipp are undefined.
219 *
220 * Use xfs_imap() to determine the size and location of the
221 * buffer to read from disk.
222 */
Christoph Hellwigc679eef2008-10-30 18:04:13 +1100223int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224xfs_inotobp(
225 xfs_mount_t *mp,
226 xfs_trans_t *tp,
227 xfs_ino_t ino,
228 xfs_dinode_t **dipp,
229 xfs_buf_t **bpp,
Christoph Hellwigc679eef2008-10-30 18:04:13 +1100230 int *offset,
231 uint imap_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100233 struct xfs_imap imap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 xfs_buf_t *bp;
235 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 imap.im_blkno = 0;
Christoph Hellwiga1941892008-11-28 14:23:40 +1100238 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
David Chinner4ae29b42008-03-06 13:43:34 +1100239 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000242 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XBF_LOCK, imap_flags);
David Chinner4ae29b42008-03-06 13:43:34 +1100243 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
247 *bpp = bp;
248 *offset = imap.im_boffset;
249 return 0;
250}
251
252
253/*
254 * This routine is called to map an inode to the buffer containing
255 * the on-disk version of the inode. It returns a pointer to the
256 * buffer containing the on-disk inode in the bpp parameter, and in
257 * the dip parameter it returns a pointer to the on-disk inode within
258 * that buffer.
259 *
260 * If a non-zero error is returned, then the contents of bpp and
261 * dipp are undefined.
262 *
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100263 * The inode is expected to already been mapped to its buffer and read
264 * in once, thus we can use the mapping information stored in the inode
265 * rather than calling xfs_imap(). This allows us to avoid the overhead
266 * of looking at the inode btree for small block file systems
Christoph Hellwig94e1b692008-11-28 14:23:41 +1100267 * (see xfs_imap()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 */
269int
270xfs_itobp(
271 xfs_mount_t *mp,
272 xfs_trans_t *tp,
273 xfs_inode_t *ip,
274 xfs_dinode_t **dipp,
275 xfs_buf_t **bpp,
David Chinnera3f74ff2008-03-06 13:43:42 +1100276 uint buf_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
278 xfs_buf_t *bp;
279 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100281 ASSERT(ip->i_imap.im_blkno != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100283 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
David Chinner4ae29b42008-03-06 13:43:34 +1100284 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 return error;
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000286
David Chinnera3f74ff2008-03-06 13:43:42 +1100287 if (!bp) {
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000288 ASSERT(buf_flags & XBF_TRYLOCK);
David Chinnera3f74ff2008-03-06 13:43:42 +1100289 ASSERT(tp == NULL);
290 *bpp = NULL;
291 return EAGAIN;
292 }
293
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100294 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 *bpp = bp;
296 return 0;
297}
298
299/*
300 * Move inode type and inode format specific information from the
301 * on-disk inode to the in-core inode. For fifos, devs, and sockets
302 * this means set if_rdev to the proper value. For files, directories,
303 * and symlinks this means to bring in the in-line data or extent
304 * pointers. For a file in B-tree format, only the root is immediately
305 * brought in-core. The rest will be in-lined in if_extents when it
306 * is first referenced (see xfs_iread_extents()).
307 */
308STATIC int
309xfs_iformat(
310 xfs_inode_t *ip,
311 xfs_dinode_t *dip)
312{
313 xfs_attr_shortform_t *atp;
314 int size;
315 int error;
316 xfs_fsize_t di_size;
317 ip->i_df.if_ext_max =
318 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
319 error = 0;
320
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100321 if (unlikely(be32_to_cpu(dip->di_nextents) +
322 be16_to_cpu(dip->di_anextents) >
323 be64_to_cpu(dip->di_nblocks))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100324 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
325 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 (unsigned long long)ip->i_ino,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100327 (int)(be32_to_cpu(dip->di_nextents) +
328 be16_to_cpu(dip->di_anextents)),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 (unsigned long long)
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100330 be64_to_cpu(dip->di_nblocks));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
332 ip->i_mount, dip);
333 return XFS_ERROR(EFSCORRUPTED);
334 }
335
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100336 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100337 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
338 "corrupt dinode %Lu, forkoff = 0x%x.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 (unsigned long long)ip->i_ino,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100340 dip->di_forkoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
342 ip->i_mount, dip);
343 return XFS_ERROR(EFSCORRUPTED);
344 }
345
Christoph Hellwigb89d4202009-08-10 11:32:18 -0300346 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
347 !ip->i_mount->m_rtdev_targp)) {
348 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
349 "corrupt dinode %Lu, has realtime flag set.",
350 ip->i_ino);
351 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
352 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
353 return XFS_ERROR(EFSCORRUPTED);
354 }
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 switch (ip->i_d.di_mode & S_IFMT) {
357 case S_IFIFO:
358 case S_IFCHR:
359 case S_IFBLK:
360 case S_IFSOCK:
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100361 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
363 ip->i_mount, dip);
364 return XFS_ERROR(EFSCORRUPTED);
365 }
366 ip->i_d.di_size = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000367 ip->i_size = 0;
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100368 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 break;
370
371 case S_IFREG:
372 case S_IFLNK:
373 case S_IFDIR:
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100374 switch (dip->di_format) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 case XFS_DINODE_FMT_LOCAL:
376 /*
377 * no local regular files yet
378 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100379 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100380 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
381 "corrupt inode %Lu "
382 "(local format for regular file).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 (unsigned long long) ip->i_ino);
384 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
385 XFS_ERRLEVEL_LOW,
386 ip->i_mount, dip);
387 return XFS_ERROR(EFSCORRUPTED);
388 }
389
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100390 di_size = be64_to_cpu(dip->di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100392 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
393 "corrupt inode %Lu "
394 "(bad size %Ld for local inode).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 (unsigned long long) ip->i_ino,
396 (long long) di_size);
397 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
398 XFS_ERRLEVEL_LOW,
399 ip->i_mount, dip);
400 return XFS_ERROR(EFSCORRUPTED);
401 }
402
403 size = (int)di_size;
404 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
405 break;
406 case XFS_DINODE_FMT_EXTENTS:
407 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
408 break;
409 case XFS_DINODE_FMT_BTREE:
410 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
411 break;
412 default:
413 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
414 ip->i_mount);
415 return XFS_ERROR(EFSCORRUPTED);
416 }
417 break;
418
419 default:
420 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
421 return XFS_ERROR(EFSCORRUPTED);
422 }
423 if (error) {
424 return error;
425 }
426 if (!XFS_DFORK_Q(dip))
427 return 0;
428 ASSERT(ip->i_afp == NULL);
429 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
430 ip->i_afp->if_ext_max =
431 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100432 switch (dip->di_aformat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 case XFS_DINODE_FMT_LOCAL:
434 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
Nathan Scott3b244aa2006-03-17 17:29:25 +1100435 size = be16_to_cpu(atp->hdr.totsize);
Christoph Hellwig2809f762009-01-19 02:04:16 +0100436
437 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
438 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
439 "corrupt inode %Lu "
440 "(bad attr fork size %Ld).",
441 (unsigned long long) ip->i_ino,
442 (long long) size);
443 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
444 XFS_ERRLEVEL_LOW,
445 ip->i_mount, dip);
446 return XFS_ERROR(EFSCORRUPTED);
447 }
448
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
450 break;
451 case XFS_DINODE_FMT_EXTENTS:
452 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
453 break;
454 case XFS_DINODE_FMT_BTREE:
455 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
456 break;
457 default:
458 error = XFS_ERROR(EFSCORRUPTED);
459 break;
460 }
461 if (error) {
462 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
463 ip->i_afp = NULL;
464 xfs_idestroy_fork(ip, XFS_DATA_FORK);
465 }
466 return error;
467}
468
469/*
470 * The file is in-lined in the on-disk inode.
471 * If it fits into if_inline_data, then copy
472 * it there, otherwise allocate a buffer for it
473 * and copy the data there. Either way, set
474 * if_data to point at the data.
475 * If we allocate a buffer for the data, make
476 * sure that its size is a multiple of 4 and
477 * record the real size in i_real_bytes.
478 */
479STATIC int
480xfs_iformat_local(
481 xfs_inode_t *ip,
482 xfs_dinode_t *dip,
483 int whichfork,
484 int size)
485{
486 xfs_ifork_t *ifp;
487 int real_size;
488
489 /*
490 * If the size is unreasonable, then something
491 * is wrong and we just bail out rather than crash in
492 * kmem_alloc() or memcpy() below.
493 */
494 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100495 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
496 "corrupt inode %Lu "
497 "(bad size %d for local fork, size = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 (unsigned long long) ip->i_ino, size,
499 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
500 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
501 ip->i_mount, dip);
502 return XFS_ERROR(EFSCORRUPTED);
503 }
504 ifp = XFS_IFORK_PTR(ip, whichfork);
505 real_size = 0;
506 if (size == 0)
507 ifp->if_u1.if_data = NULL;
508 else if (size <= sizeof(ifp->if_u2.if_inline_data))
509 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
510 else {
511 real_size = roundup(size, 4);
512 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
513 }
514 ifp->if_bytes = size;
515 ifp->if_real_bytes = real_size;
516 if (size)
517 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
518 ifp->if_flags &= ~XFS_IFEXTENTS;
519 ifp->if_flags |= XFS_IFINLINE;
520 return 0;
521}
522
523/*
524 * The file consists of a set of extents all
525 * of which fit into the on-disk inode.
526 * If there are few enough extents to fit into
527 * the if_inline_ext, then copy them there.
528 * Otherwise allocate a buffer for them and copy
529 * them into it. Either way, set if_extents
530 * to point at the extents.
531 */
532STATIC int
533xfs_iformat_extents(
534 xfs_inode_t *ip,
535 xfs_dinode_t *dip,
536 int whichfork)
537{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000538 xfs_bmbt_rec_t *dp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 xfs_ifork_t *ifp;
540 int nex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 int size;
542 int i;
543
544 ifp = XFS_IFORK_PTR(ip, whichfork);
545 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
546 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
547
548 /*
549 * If the number of extents is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
552 */
553 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100554 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
555 "corrupt inode %Lu ((a)extents = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 (unsigned long long) ip->i_ino, nex);
557 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
558 ip->i_mount, dip);
559 return XFS_ERROR(EFSCORRUPTED);
560 }
561
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100562 ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 if (nex == 0)
564 ifp->if_u1.if_extents = NULL;
565 else if (nex <= XFS_INLINE_EXTS)
566 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100567 else
568 xfs_iext_add(ifp, 0, nex);
569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 ifp->if_bytes = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 if (size) {
572 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000573 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100574 for (i = 0; i < nex; i++, dp++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000575 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Harvey Harrison597bca62008-08-13 16:29:21 +1000576 ep->l0 = get_unaligned_be64(&dp->l0);
577 ep->l1 = get_unaligned_be64(&dp->l1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
Eric Sandeen3a59c942007-07-11 11:09:47 +1000579 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 if (whichfork != XFS_DATA_FORK ||
581 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
582 if (unlikely(xfs_check_nostate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100583 ifp, 0, nex))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
585 XFS_ERRLEVEL_LOW,
586 ip->i_mount);
587 return XFS_ERROR(EFSCORRUPTED);
588 }
589 }
590 ifp->if_flags |= XFS_IFEXTENTS;
591 return 0;
592}
593
594/*
595 * The file has too many extents to fit into
596 * the inode, so they are in B-tree format.
597 * Allocate a buffer for the root of the B-tree
598 * and copy the root into it. The i_extents
599 * field will remain NULL until all of the
600 * extents are read in (when they are needed).
601 */
602STATIC int
603xfs_iformat_btree(
604 xfs_inode_t *ip,
605 xfs_dinode_t *dip,
606 int whichfork)
607{
608 xfs_bmdr_block_t *dfp;
609 xfs_ifork_t *ifp;
610 /* REFERENCED */
611 int nrecs;
612 int size;
613
614 ifp = XFS_IFORK_PTR(ip, whichfork);
615 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
616 size = XFS_BMAP_BROOT_SPACE(dfp);
Christoph Hellwig60197e82008-10-30 17:11:19 +1100617 nrecs = be16_to_cpu(dfp->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 /*
620 * blow out if -- fork has less extents than can fit in
621 * fork (fork shouldn't be a btree format), root btree
622 * block has more records than can fit into the fork,
623 * or the number of extents is greater than the number of
624 * blocks.
625 */
626 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
627 || XFS_BMDR_SPACE_CALC(nrecs) >
628 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
629 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100630 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
631 "corrupt inode %Lu (btree).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 (unsigned long long) ip->i_ino);
633 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
634 ip->i_mount);
635 return XFS_ERROR(EFSCORRUPTED);
636 }
637
638 ifp->if_broot_bytes = size;
639 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
640 ASSERT(ifp->if_broot != NULL);
641 /*
642 * Copy and convert from the on-disk structure
643 * to the in-memory structure.
644 */
Christoph Hellwig60197e82008-10-30 17:11:19 +1100645 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
646 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
647 ifp->if_broot, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 ifp->if_flags &= ~XFS_IFEXTENTS;
649 ifp->if_flags |= XFS_IFBROOT;
650
651 return 0;
652}
653
Eric Sandeend96f8f82009-07-02 00:09:33 -0500654STATIC void
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000655xfs_dinode_from_disk(
656 xfs_icdinode_t *to,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100657 xfs_dinode_t *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000659 to->di_magic = be16_to_cpu(from->di_magic);
660 to->di_mode = be16_to_cpu(from->di_mode);
661 to->di_version = from ->di_version;
662 to->di_format = from->di_format;
663 to->di_onlink = be16_to_cpu(from->di_onlink);
664 to->di_uid = be32_to_cpu(from->di_uid);
665 to->di_gid = be32_to_cpu(from->di_gid);
666 to->di_nlink = be32_to_cpu(from->di_nlink);
667 to->di_projid = be16_to_cpu(from->di_projid);
668 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
669 to->di_flushiter = be16_to_cpu(from->di_flushiter);
670 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
671 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
672 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
673 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
674 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
675 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
676 to->di_size = be64_to_cpu(from->di_size);
677 to->di_nblocks = be64_to_cpu(from->di_nblocks);
678 to->di_extsize = be32_to_cpu(from->di_extsize);
679 to->di_nextents = be32_to_cpu(from->di_nextents);
680 to->di_anextents = be16_to_cpu(from->di_anextents);
681 to->di_forkoff = from->di_forkoff;
682 to->di_aformat = from->di_aformat;
683 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
684 to->di_dmstate = be16_to_cpu(from->di_dmstate);
685 to->di_flags = be16_to_cpu(from->di_flags);
686 to->di_gen = be32_to_cpu(from->di_gen);
687}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000689void
690xfs_dinode_to_disk(
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100691 xfs_dinode_t *to,
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000692 xfs_icdinode_t *from)
693{
694 to->di_magic = cpu_to_be16(from->di_magic);
695 to->di_mode = cpu_to_be16(from->di_mode);
696 to->di_version = from ->di_version;
697 to->di_format = from->di_format;
698 to->di_onlink = cpu_to_be16(from->di_onlink);
699 to->di_uid = cpu_to_be32(from->di_uid);
700 to->di_gid = cpu_to_be32(from->di_gid);
701 to->di_nlink = cpu_to_be32(from->di_nlink);
702 to->di_projid = cpu_to_be16(from->di_projid);
703 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
704 to->di_flushiter = cpu_to_be16(from->di_flushiter);
705 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
706 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
707 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
708 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
709 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
710 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
711 to->di_size = cpu_to_be64(from->di_size);
712 to->di_nblocks = cpu_to_be64(from->di_nblocks);
713 to->di_extsize = cpu_to_be32(from->di_extsize);
714 to->di_nextents = cpu_to_be32(from->di_nextents);
715 to->di_anextents = cpu_to_be16(from->di_anextents);
716 to->di_forkoff = from->di_forkoff;
717 to->di_aformat = from->di_aformat;
718 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
719 to->di_dmstate = cpu_to_be16(from->di_dmstate);
720 to->di_flags = cpu_to_be16(from->di_flags);
721 to->di_gen = cpu_to_be32(from->di_gen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722}
723
724STATIC uint
725_xfs_dic2xflags(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 __uint16_t di_flags)
727{
728 uint flags = 0;
729
730 if (di_flags & XFS_DIFLAG_ANY) {
731 if (di_flags & XFS_DIFLAG_REALTIME)
732 flags |= XFS_XFLAG_REALTIME;
733 if (di_flags & XFS_DIFLAG_PREALLOC)
734 flags |= XFS_XFLAG_PREALLOC;
735 if (di_flags & XFS_DIFLAG_IMMUTABLE)
736 flags |= XFS_XFLAG_IMMUTABLE;
737 if (di_flags & XFS_DIFLAG_APPEND)
738 flags |= XFS_XFLAG_APPEND;
739 if (di_flags & XFS_DIFLAG_SYNC)
740 flags |= XFS_XFLAG_SYNC;
741 if (di_flags & XFS_DIFLAG_NOATIME)
742 flags |= XFS_XFLAG_NOATIME;
743 if (di_flags & XFS_DIFLAG_NODUMP)
744 flags |= XFS_XFLAG_NODUMP;
745 if (di_flags & XFS_DIFLAG_RTINHERIT)
746 flags |= XFS_XFLAG_RTINHERIT;
747 if (di_flags & XFS_DIFLAG_PROJINHERIT)
748 flags |= XFS_XFLAG_PROJINHERIT;
749 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
750 flags |= XFS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100751 if (di_flags & XFS_DIFLAG_EXTSIZE)
752 flags |= XFS_XFLAG_EXTSIZE;
753 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
754 flags |= XFS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000755 if (di_flags & XFS_DIFLAG_NODEFRAG)
756 flags |= XFS_XFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000757 if (di_flags & XFS_DIFLAG_FILESTREAM)
758 flags |= XFS_XFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
760
761 return flags;
762}
763
764uint
765xfs_ip2xflags(
766 xfs_inode_t *ip)
767{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000768 xfs_icdinode_t *dic = &ip->i_d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Nathan Scotta916e2b2006-06-09 17:12:17 +1000770 return _xfs_dic2xflags(dic->di_flags) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100771 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772}
773
774uint
775xfs_dic2xflags(
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100776 xfs_dinode_t *dip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100778 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100779 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780}
781
782/*
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100783 * Read the disk inode attributes into the in-core inode structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 */
785int
786xfs_iread(
787 xfs_mount_t *mp,
788 xfs_trans_t *tp,
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100789 xfs_inode_t *ip,
Nathan Scott745b1f472006-09-28 11:02:23 +1000790 xfs_daddr_t bno,
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100791 uint iget_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
793 xfs_buf_t *bp;
794 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 int error;
796
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 /*
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100798 * Fill in the location information in the in-core inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100800 ip->i_imap.im_blkno = bno;
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100801 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100802 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100803 return error;
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100804 ASSERT(bno == 0 || bno == ip->i_imap.im_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
806 /*
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100807 * Get pointers to the on-disk inode and the buffer containing it.
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100808 */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100809 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000810 XBF_LOCK, iget_flags);
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100811 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100812 return error;
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100813 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100814
815 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 * If we got something that isn't an inode it means someone
817 * (nfs or dmi) has a stale handle.
818 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100819 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820#ifdef DEBUG
821 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100822 "dip->di_magic (0x%x) != "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 "XFS_DINODE_MAGIC (0x%x)",
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100824 be16_to_cpu(dip->di_magic),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 XFS_DINODE_MAGIC);
826#endif /* DEBUG */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100827 error = XFS_ERROR(EINVAL);
828 goto out_brelse;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 }
830
831 /*
832 * If the on-disk inode is already linked to a directory
833 * entry, copy all of the inode into the in-core inode.
834 * xfs_iformat() handles copying in the inode format
835 * specific information.
836 * Otherwise, just get the truly permanent information.
837 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100838 if (dip->di_mode) {
839 xfs_dinode_from_disk(&ip->i_d, dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 error = xfs_iformat(ip, dip);
841 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842#ifdef DEBUG
843 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
844 "xfs_iformat() returned error %d",
845 error);
846#endif /* DEBUG */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100847 goto out_brelse;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 }
849 } else {
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100850 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
851 ip->i_d.di_version = dip->di_version;
852 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
853 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 /*
855 * Make sure to pull in the mode here as well in
856 * case the inode is released without being used.
857 * This ensures that xfs_inactive() will see that
858 * the inode is already free and not try to mess
859 * with the uninitialized part of it.
860 */
861 ip->i_d.di_mode = 0;
862 /*
863 * Initialize the per-fork minima and maxima for a new
864 * inode here. xfs_iformat will do it for old inodes.
865 */
866 ip->i_df.if_ext_max =
867 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
868 }
869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 /*
871 * The inode format changed when we moved the link count and
872 * made it 32 bits long. If this is an old format inode,
873 * convert it in memory to look like a new one. If it gets
874 * flushed to disk we will convert back before flushing or
875 * logging it. We zero out the new projid field and the old link
876 * count field. We'll handle clearing the pad field (the remains
877 * of the old uuid field) when we actually convert the inode to
878 * the new format. We don't change the version number so that we
879 * can distinguish this from a real new format inode.
880 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +1100881 if (ip->i_d.di_version == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 ip->i_d.di_nlink = ip->i_d.di_onlink;
883 ip->i_d.di_onlink = 0;
884 ip->i_d.di_projid = 0;
885 }
886
887 ip->i_delayed_blks = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000888 ip->i_size = ip->i_d.di_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
890 /*
891 * Mark the buffer containing the inode as something to keep
892 * around for a while. This helps to keep recently accessed
893 * meta-data in-core longer.
894 */
Christoph Hellwig6d73cf12008-12-09 04:47:32 -0500895 XFS_BUF_SET_REF(bp, XFS_INO_REF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 /*
898 * Use xfs_trans_brelse() to release the buffer containing the
899 * on-disk inode, because it was acquired with xfs_trans_read_buf()
900 * in xfs_itobp() above. If tp is NULL, this is just a normal
901 * brelse(). If we're within a transaction, then xfs_trans_brelse()
902 * will only release the buffer if it is not dirty within the
903 * transaction. It will be OK to release the buffer in this case,
904 * because inodes on disk are never destroyed and we will be
905 * locking the new in-core inode before putting it in the hash
906 * table where other processes can find it. Thus we don't have
907 * to worry about the inode being changed just because we released
908 * the buffer.
909 */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100910 out_brelse:
911 xfs_trans_brelse(tp, bp);
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100912 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913}
914
915/*
916 * Read in extents from a btree-format inode.
917 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
918 */
919int
920xfs_iread_extents(
921 xfs_trans_t *tp,
922 xfs_inode_t *ip,
923 int whichfork)
924{
925 int error;
926 xfs_ifork_t *ifp;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100927 xfs_extnum_t nextents;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 size_t size;
929
930 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
931 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
932 ip->i_mount);
933 return XFS_ERROR(EFSCORRUPTED);
934 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100935 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
936 size = nextents * sizeof(xfs_bmbt_rec_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 ifp = XFS_IFORK_PTR(ip, whichfork);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 /*
940 * We know that the size is valid (it's checked in iformat_btree)
941 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 ifp->if_lastex = NULLEXTNUM;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100943 ifp->if_bytes = ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 ifp->if_flags |= XFS_IFEXTENTS;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100945 xfs_iext_add(ifp, 0, nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 error = xfs_bmap_read_extents(tp, ip, whichfork);
947 if (error) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100948 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 ifp->if_flags &= ~XFS_IFEXTENTS;
950 return error;
951 }
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000952 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 return 0;
954}
955
956/*
957 * Allocate an inode on disk and return a copy of its in-core version.
958 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
959 * appropriately within the inode. The uid and gid for the inode are
960 * set according to the contents of the given cred structure.
961 *
962 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
963 * has a free inode available, call xfs_iget()
964 * to obtain the in-core version of the allocated inode. Finally,
965 * fill in the inode and log its initial contents. In this case,
966 * ialloc_context would be set to NULL and call_again set to false.
967 *
968 * If xfs_dialloc() does not have an available inode,
969 * it will replenish its supply by doing an allocation. Since we can
970 * only do one allocation within a transaction without deadlocks, we
971 * must commit the current transaction before returning the inode itself.
972 * In this case, therefore, we will set call_again to true and return.
973 * The caller should then commit the current transaction, start a new
974 * transaction, and call xfs_ialloc() again to actually get the inode.
975 *
976 * To ensure that some other process does not grab the inode that
977 * was allocated during the first call to xfs_ialloc(), this routine
978 * also returns the [locked] bp pointing to the head of the freelist
979 * as ialloc_context. The caller should hold this buffer across
980 * the commit and pass it back into this routine on the second call.
David Chinnerb11f94d2007-07-11 11:09:33 +1000981 *
982 * If we are allocating quota inodes, we do not have a parent inode
983 * to attach to or associate with (i.e. pip == NULL) because they
984 * are not linked into the directory structure - they are attached
985 * directly to the superblock - and so have no parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 */
987int
988xfs_ialloc(
989 xfs_trans_t *tp,
990 xfs_inode_t *pip,
991 mode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -0700992 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 xfs_dev_t rdev,
994 cred_t *cr,
995 xfs_prid_t prid,
996 int okalloc,
997 xfs_buf_t **ialloc_context,
998 boolean_t *call_again,
999 xfs_inode_t **ipp)
1000{
1001 xfs_ino_t ino;
1002 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 uint flags;
1004 int error;
Christoph Hellwigdff35fd2008-08-13 16:44:15 +10001005 timespec_t tv;
David Chinnerbf904242008-10-30 17:36:14 +11001006 int filestreams = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
1008 /*
1009 * Call the space management code to pick
1010 * the on-disk inode to be allocated.
1011 */
David Chinnerb11f94d2007-07-11 11:09:33 +10001012 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 ialloc_context, call_again, &ino);
David Chinnerbf904242008-10-30 17:36:14 +11001014 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 if (*call_again || ino == NULLFSINO) {
1017 *ipp = NULL;
1018 return 0;
1019 }
1020 ASSERT(*ialloc_context == NULL);
1021
1022 /*
1023 * Get the in-core inode with the lock held exclusively.
1024 * This is because we're setting fields here we need
1025 * to prevent others from looking at until we're done.
1026 */
1027 error = xfs_trans_iget(tp->t_mountp, tp, ino,
Nathan Scott745b1f472006-09-28 11:02:23 +10001028 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
David Chinnerbf904242008-10-30 17:36:14 +11001029 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 ASSERT(ip != NULL);
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 ip->i_d.di_mode = (__uint16_t)mode;
1034 ip->i_d.di_onlink = 0;
1035 ip->i_d.di_nlink = nlink;
1036 ASSERT(ip->i_d.di_nlink == nlink);
David Howells9e2b2dc2008-08-13 16:20:04 +01001037 ip->i_d.di_uid = current_fsuid();
1038 ip->i_d.di_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 ip->i_d.di_projid = prid;
1040 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1041
1042 /*
1043 * If the superblock version is up to where we support new format
1044 * inodes and this is currently an old format inode, then change
1045 * the inode version number now. This way we only do the conversion
1046 * here rather than here and in the flush/logging code.
1047 */
Eric Sandeen62118702008-03-06 13:44:28 +11001048 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11001049 ip->i_d.di_version == 1) {
1050 ip->i_d.di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 /*
1052 * We've already zeroed the old link count, the projid field,
1053 * and the pad field.
1054 */
1055 }
1056
1057 /*
1058 * Project ids won't be stored on disk if we are using a version 1 inode.
1059 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11001060 if ((prid != 0) && (ip->i_d.di_version == 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 xfs_bump_ino_vers2(tp, ip);
1062
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001063 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 ip->i_d.di_gid = pip->i_d.di_gid;
1065 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1066 ip->i_d.di_mode |= S_ISGID;
1067 }
1068 }
1069
1070 /*
1071 * If the group ID of the new file does not match the effective group
1072 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1073 * (and only if the irix_sgid_inherit compatibility variable is set).
1074 */
1075 if ((irix_sgid_inherit) &&
1076 (ip->i_d.di_mode & S_ISGID) &&
1077 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1078 ip->i_d.di_mode &= ~S_ISGID;
1079 }
1080
1081 ip->i_d.di_size = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001082 ip->i_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 ip->i_d.di_nextents = 0;
1084 ASSERT(ip->i_d.di_nblocks == 0);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +10001085
1086 nanotime(&tv);
1087 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1088 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1089 ip->i_d.di_atime = ip->i_d.di_mtime;
1090 ip->i_d.di_ctime = ip->i_d.di_mtime;
1091
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 /*
1093 * di_gen will have been taken care of in xfs_iread.
1094 */
1095 ip->i_d.di_extsize = 0;
1096 ip->i_d.di_dmevmask = 0;
1097 ip->i_d.di_dmstate = 0;
1098 ip->i_d.di_flags = 0;
1099 flags = XFS_ILOG_CORE;
1100 switch (mode & S_IFMT) {
1101 case S_IFIFO:
1102 case S_IFCHR:
1103 case S_IFBLK:
1104 case S_IFSOCK:
1105 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1106 ip->i_df.if_u2.if_rdev = rdev;
1107 ip->i_df.if_flags = 0;
1108 flags |= XFS_ILOG_DEV;
1109 break;
1110 case S_IFREG:
David Chinnerbf904242008-10-30 17:36:14 +11001111 /*
1112 * we can't set up filestreams until after the VFS inode
1113 * is set up properly.
1114 */
1115 if (pip && xfs_inode_is_filestream(pip))
1116 filestreams = 1;
David Chinner2a82b8b2007-07-11 11:09:12 +10001117 /* fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +10001119 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Nathan Scott365ca832005-06-21 15:39:12 +10001120 uint di_flags = 0;
1121
1122 if ((mode & S_IFMT) == S_IFDIR) {
1123 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1124 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001125 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1126 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1127 ip->i_d.di_extsize = pip->i_d.di_extsize;
1128 }
1129 } else if ((mode & S_IFMT) == S_IFREG) {
Christoph Hellwig613d7042007-10-11 17:44:08 +10001130 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +10001131 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001132 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1133 di_flags |= XFS_DIFLAG_EXTSIZE;
1134 ip->i_d.di_extsize = pip->i_d.di_extsize;
1135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 }
1137 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1138 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +10001139 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1141 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +10001142 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1144 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +10001145 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1147 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +10001148 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1149 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1150 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +10001151 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1152 xfs_inherit_nodefrag)
1153 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +10001154 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1155 di_flags |= XFS_DIFLAG_FILESTREAM;
Nathan Scott365ca832005-06-21 15:39:12 +10001156 ip->i_d.di_flags |= di_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 }
1158 /* FALLTHROUGH */
1159 case S_IFLNK:
1160 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1161 ip->i_df.if_flags = XFS_IFEXTENTS;
1162 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1163 ip->i_df.if_u1.if_extents = NULL;
1164 break;
1165 default:
1166 ASSERT(0);
1167 }
1168 /*
1169 * Attribute fork settings for new inode.
1170 */
1171 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1172 ip->i_d.di_anextents = 0;
1173
1174 /*
1175 * Log the new values stuffed into the inode.
1176 */
1177 xfs_trans_log_inode(tp, ip, flags);
1178
Nathan Scottb83bd132006-06-09 16:48:30 +10001179 /* now that we have an i_mode we can setup inode ops and unlock */
Christoph Hellwig41be8be2008-08-13 16:23:13 +10001180 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
David Chinnerbf904242008-10-30 17:36:14 +11001182 /* now we have set up the vfs inode we can associate the filestream */
1183 if (filestreams) {
1184 error = xfs_filestream_associate(pip, ip);
1185 if (error < 0)
1186 return -error;
1187 if (!error)
1188 xfs_iflags_set(ip, XFS_IFILESTREAM);
1189 }
1190
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 *ipp = ip;
1192 return 0;
1193}
1194
1195/*
1196 * Check to make sure that there are no blocks allocated to the
1197 * file beyond the size of the file. We don't check this for
1198 * files with fixed size extents or real time extents, but we
1199 * at least do it for regular files.
1200 */
1201#ifdef DEBUG
1202void
1203xfs_isize_check(
1204 xfs_mount_t *mp,
1205 xfs_inode_t *ip,
1206 xfs_fsize_t isize)
1207{
1208 xfs_fileoff_t map_first;
1209 int nimaps;
1210 xfs_bmbt_irec_t imaps[2];
1211
1212 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1213 return;
1214
Eric Sandeen71ddabb2007-11-23 16:29:42 +11001215 if (XFS_IS_REALTIME_INODE(ip))
1216 return;
1217
1218 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 return;
1220
1221 nimaps = 2;
1222 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1223 /*
1224 * The filesystem could be shutting down, so bmapi may return
1225 * an error.
1226 */
1227 if (xfs_bmapi(NULL, ip, map_first,
1228 (XFS_B_TO_FSB(mp,
1229 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1230 map_first),
1231 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001232 NULL, NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 return;
1234 ASSERT(nimaps == 1);
1235 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1236}
1237#endif /* DEBUG */
1238
1239/*
1240 * Calculate the last possible buffered byte in a file. This must
1241 * include data that was buffered beyond the EOF by the write code.
1242 * This also needs to deal with overflowing the xfs_fsize_t type
1243 * which can happen for sizes near the limit.
1244 *
1245 * We also need to take into account any blocks beyond the EOF. It
1246 * may be the case that they were buffered by a write which failed.
1247 * In that case the pages will still be in memory, but the inode size
1248 * will never have been updated.
1249 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05001250STATIC xfs_fsize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251xfs_file_last_byte(
1252 xfs_inode_t *ip)
1253{
1254 xfs_mount_t *mp;
1255 xfs_fsize_t last_byte;
1256 xfs_fileoff_t last_block;
1257 xfs_fileoff_t size_last_block;
1258 int error;
1259
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001260 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
1262 mp = ip->i_mount;
1263 /*
1264 * Only check for blocks beyond the EOF if the extents have
1265 * been read in. This eliminates the need for the inode lock,
1266 * and it also saves us from looking when it really isn't
1267 * necessary.
1268 */
1269 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
Lachlan McIlroyf25181f2009-04-23 22:18:00 -04001270 xfs_ilock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1272 XFS_DATA_FORK);
Lachlan McIlroyf25181f2009-04-23 22:18:00 -04001273 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 if (error) {
1275 last_block = 0;
1276 }
1277 } else {
1278 last_block = 0;
1279 }
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001280 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1282
1283 last_byte = XFS_FSB_TO_B(mp, last_block);
1284 if (last_byte < 0) {
1285 return XFS_MAXIOFFSET(mp);
1286 }
1287 last_byte += (1 << mp->m_writeio_log);
1288 if (last_byte < 0) {
1289 return XFS_MAXIOFFSET(mp);
1290 }
1291 return last_byte;
1292}
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294/*
1295 * Start the truncation of the file to new_size. The new size
1296 * must be smaller than the current size. This routine will
1297 * clear the buffer and page caches of file data in the removed
1298 * range, and xfs_itruncate_finish() will remove the underlying
1299 * disk blocks.
1300 *
1301 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1302 * must NOT have the inode lock held at all. This is because we're
1303 * calling into the buffer/page cache code and we can't hold the
1304 * inode lock when we do so.
1305 *
David Chinner38e22992006-03-22 12:47:15 +11001306 * We need to wait for any direct I/Os in flight to complete before we
1307 * proceed with the truncate. This is needed to prevent the extents
1308 * being read or written by the direct I/Os from being removed while the
1309 * I/O is in flight as there is no other method of synchronising
1310 * direct I/O with the truncate operation. Also, because we hold
1311 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1312 * started until the truncate completes and drops the lock. Essentially,
Christoph Hellwig25e41b32008-12-03 12:20:39 +01001313 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1314 * ordering between direct I/Os and the truncate operation.
David Chinner38e22992006-03-22 12:47:15 +11001315 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1317 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1318 * in the case that the caller is locking things out of order and
1319 * may not be able to call xfs_itruncate_finish() with the inode lock
1320 * held without dropping the I/O lock. If the caller must drop the
1321 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1322 * must be called again with all the same restrictions as the initial
1323 * call.
1324 */
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001325int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326xfs_itruncate_start(
1327 xfs_inode_t *ip,
1328 uint flags,
1329 xfs_fsize_t new_size)
1330{
1331 xfs_fsize_t last_byte;
1332 xfs_off_t toss_start;
1333 xfs_mount_t *mp;
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001334 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001336 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001337 ASSERT((new_size == 0) || (new_size <= ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1339 (flags == XFS_ITRUNC_MAYBE));
1340
1341 mp = ip->i_mount;
Yingping Lu9fa80462006-03-22 12:44:35 +11001342
Lachlan McIlroyc734c792007-12-18 16:17:41 +11001343 /* wait for the completion of any pending DIOs */
Lachlan McIlroyd112f292008-10-30 16:59:06 +11001344 if (new_size == 0 || new_size < ip->i_size)
Christoph Hellwig25e41b32008-12-03 12:20:39 +01001345 xfs_ioend_wait(ip);
Lachlan McIlroyc734c792007-12-18 16:17:41 +11001346
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 /*
Nathan Scott67fcaa72006-06-09 17:00:52 +10001348 * Call toss_pages or flushinval_pages to get rid of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 * overlapping the region being removed. We have to use
Nathan Scott67fcaa72006-06-09 17:00:52 +10001350 * the less efficient flushinval_pages in the case that the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 * caller may not be able to finish the truncate without
1352 * dropping the inode's I/O lock. Make sure
1353 * to catch any pages brought in by buffers overlapping
1354 * the EOF by searching out beyond the isize by our
1355 * block size. We round new_size up to a block boundary
1356 * so that we don't toss things on the same block as
1357 * new_size but before it.
1358 *
Nathan Scott67fcaa72006-06-09 17:00:52 +10001359 * Before calling toss_page or flushinval_pages, make sure to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 * call remapf() over the same region if the file is mapped.
1361 * This frees up mapped file references to the pages in the
Nathan Scott67fcaa72006-06-09 17:00:52 +10001362 * given range and for the flushinval_pages case it ensures
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 * that we get the latest mapped changes flushed out.
1364 */
1365 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1366 toss_start = XFS_FSB_TO_B(mp, toss_start);
1367 if (toss_start < 0) {
1368 /*
1369 * The place to start tossing is beyond our maximum
1370 * file size, so there is no way that the data extended
1371 * out there.
1372 */
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001373 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 }
1375 last_byte = xfs_file_last_byte(ip);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001376 trace_xfs_itruncate_start(ip, flags, new_size, toss_start, last_byte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 if (last_byte > toss_start) {
1378 if (flags & XFS_ITRUNC_DEFINITE) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001379 xfs_tosspages(ip, toss_start,
1380 -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 } else {
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001382 error = xfs_flushinval_pages(ip, toss_start,
1383 -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 }
1385 }
1386
1387#ifdef DEBUG
1388 if (new_size == 0) {
Christoph Hellwigdf80c932008-08-13 16:22:09 +10001389 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 }
1391#endif
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001392 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393}
1394
1395/*
David Chinnerf6485052008-04-17 16:50:04 +10001396 * Shrink the file to the given new_size. The new size must be smaller than
1397 * the current size. This will free up the underlying blocks in the removed
1398 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 *
David Chinnerf6485052008-04-17 16:50:04 +10001400 * The transaction passed to this routine must have made a permanent log
1401 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1402 * given transaction and start new ones, so make sure everything involved in
1403 * the transaction is tidy before calling here. Some transaction will be
1404 * returned to the caller to be committed. The incoming transaction must
1405 * already include the inode, and both inode locks must be held exclusively.
1406 * The inode must also be "held" within the transaction. On return the inode
1407 * will be "held" within the returned transaction. This routine does NOT
1408 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 *
David Chinnerf6485052008-04-17 16:50:04 +10001410 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1411 * indicates the fork which is to be truncated. For the attribute fork we only
1412 * support truncation to size 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 *
David Chinnerf6485052008-04-17 16:50:04 +10001414 * We use the sync parameter to indicate whether or not the first transaction
1415 * we perform might have to be synchronous. For the attr fork, it needs to be
1416 * so if the unlink of the inode is not yet known to be permanent in the log.
1417 * This keeps us from freeing and reusing the blocks of the attribute fork
1418 * before the unlink of the inode becomes permanent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 *
David Chinnerf6485052008-04-17 16:50:04 +10001420 * For the data fork, we normally have to run synchronously if we're being
1421 * called out of the inactive path or we're being called out of the create path
1422 * where we're truncating an existing file. Either way, the truncate needs to
1423 * be sync so blocks don't reappear in the file with altered data in case of a
1424 * crash. wsync filesystems can run the first case async because anything that
1425 * shrinks the inode has to run sync so by the time we're called here from
1426 * inactive, the inode size is permanently set to 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 *
David Chinnerf6485052008-04-17 16:50:04 +10001428 * Calls from the truncate path always need to be sync unless we're in a wsync
1429 * filesystem and the file has already been unlinked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 *
David Chinnerf6485052008-04-17 16:50:04 +10001431 * The caller is responsible for correctly setting the sync parameter. It gets
1432 * too hard for us to guess here which path we're being called out of just
1433 * based on inode state.
1434 *
1435 * If we get an error, we must return with the inode locked and linked into the
1436 * current transaction. This keeps things simple for the higher level code,
1437 * because it always knows that the inode is locked and held in the transaction
1438 * that returns to it whether errors occur or not. We don't mark the inode
1439 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 */
1441int
1442xfs_itruncate_finish(
1443 xfs_trans_t **tp,
1444 xfs_inode_t *ip,
1445 xfs_fsize_t new_size,
1446 int fork,
1447 int sync)
1448{
1449 xfs_fsblock_t first_block;
1450 xfs_fileoff_t first_unmap_block;
1451 xfs_fileoff_t last_block;
1452 xfs_filblks_t unmap_len=0;
1453 xfs_mount_t *mp;
1454 xfs_trans_t *ntp;
1455 int done;
1456 int committed;
1457 xfs_bmap_free_t free_list;
1458 int error;
1459
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001460 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001461 ASSERT((new_size == 0) || (new_size <= ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 ASSERT(*tp != NULL);
1463 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1464 ASSERT(ip->i_transp == *tp);
1465 ASSERT(ip->i_itemp != NULL);
1466 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1467
1468
1469 ntp = *tp;
1470 mp = (ntp)->t_mountp;
1471 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1472
1473 /*
1474 * We only support truncating the entire attribute fork.
1475 */
1476 if (fork == XFS_ATTR_FORK) {
1477 new_size = 0LL;
1478 }
1479 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001480 trace_xfs_itruncate_finish_start(ip, new_size);
1481
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 /*
1483 * The first thing we do is set the size to new_size permanently
1484 * on disk. This way we don't have to worry about anyone ever
1485 * being able to look at the data being freed even in the face
1486 * of a crash. What we're getting around here is the case where
1487 * we free a block, it is allocated to another file, it is written
1488 * to, and then we crash. If the new data gets written to the
1489 * file but the log buffers containing the free and reallocation
1490 * don't, then we'd end up with garbage in the blocks being freed.
1491 * As long as we make the new_size permanent before actually
1492 * freeing any blocks it doesn't matter if they get writtten to.
1493 *
1494 * The callers must signal into us whether or not the size
1495 * setting here must be synchronous. There are a few cases
1496 * where it doesn't have to be synchronous. Those cases
1497 * occur if the file is unlinked and we know the unlink is
1498 * permanent or if the blocks being truncated are guaranteed
1499 * to be beyond the inode eof (regardless of the link count)
1500 * and the eof value is permanent. Both of these cases occur
1501 * only on wsync-mounted filesystems. In those cases, we're
1502 * guaranteed that no user will ever see the data in the blocks
1503 * that are being truncated so the truncate can run async.
1504 * In the free beyond eof case, the file may wind up with
1505 * more blocks allocated to it than it needs if we crash
1506 * and that won't get fixed until the next time the file
1507 * is re-opened and closed but that's ok as that shouldn't
1508 * be too many blocks.
1509 *
1510 * However, we can't just make all wsync xactions run async
1511 * because there's one call out of the create path that needs
1512 * to run sync where it's truncating an existing file to size
1513 * 0 whose size is > 0.
1514 *
1515 * It's probably possible to come up with a test in this
1516 * routine that would correctly distinguish all the above
1517 * cases from the values of the function parameters and the
1518 * inode state but for sanity's sake, I've decided to let the
1519 * layers above just tell us. It's simpler to correctly figure
1520 * out in the layer above exactly under what conditions we
1521 * can run async and I think it's easier for others read and
1522 * follow the logic in case something has to be changed.
1523 * cscope is your friend -- rcc.
1524 *
1525 * The attribute fork is much simpler.
1526 *
1527 * For the attribute fork we allow the caller to tell us whether
1528 * the unlink of the inode that led to this call is yet permanent
1529 * in the on disk log. If it is not and we will be freeing extents
1530 * in this inode then we make the first transaction synchronous
1531 * to make sure that the unlink is permanent by the time we free
1532 * the blocks.
1533 */
1534 if (fork == XFS_DATA_FORK) {
1535 if (ip->i_d.di_nextents > 0) {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001536 /*
1537 * If we are not changing the file size then do
1538 * not update the on-disk file size - we may be
1539 * called from xfs_inactive_free_eofblocks(). If we
1540 * update the on-disk file size and then the system
1541 * crashes before the contents of the file are
1542 * flushed to disk then the files may be full of
1543 * holes (ie NULL files bug).
1544 */
1545 if (ip->i_size != new_size) {
1546 ip->i_d.di_size = new_size;
1547 ip->i_size = new_size;
1548 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 }
1551 } else if (sync) {
1552 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1553 if (ip->i_d.di_anextents > 0)
1554 xfs_trans_set_sync(ntp);
1555 }
1556 ASSERT(fork == XFS_DATA_FORK ||
1557 (fork == XFS_ATTR_FORK &&
1558 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1559 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1560
1561 /*
1562 * Since it is possible for space to become allocated beyond
1563 * the end of the file (in a crash where the space is allocated
1564 * but the inode size is not yet updated), simply remove any
1565 * blocks which show up between the new EOF and the maximum
1566 * possible file size. If the first block to be removed is
1567 * beyond the maximum file size (ie it is the same as last_block),
1568 * then there is nothing to do.
1569 */
1570 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1571 ASSERT(first_unmap_block <= last_block);
1572 done = 0;
1573 if (last_block == first_unmap_block) {
1574 done = 1;
1575 } else {
1576 unmap_len = last_block - first_unmap_block + 1;
1577 }
1578 while (!done) {
1579 /*
1580 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1581 * will tell us whether it freed the entire range or
1582 * not. If this is a synchronous mount (wsync),
1583 * then we can tell bunmapi to keep all the
1584 * transactions asynchronous since the unlink
1585 * transaction that made this inode inactive has
1586 * already hit the disk. There's no danger of
1587 * the freed blocks being reused, there being a
1588 * crash, and the reused blocks suddenly reappearing
1589 * in this file with garbage in them once recovery
1590 * runs.
1591 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06001592 xfs_bmap_init(&free_list, &first_block);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10001593 error = xfs_bunmapi(ntp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001594 first_unmap_block, unmap_len,
Eric Sandeen9d87c312009-01-14 23:22:07 -06001595 xfs_bmapi_aflag(fork) |
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 (sync ? 0 : XFS_BMAPI_ASYNC),
1597 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001598 &first_block, &free_list,
1599 NULL, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 if (error) {
1601 /*
1602 * If the bunmapi call encounters an error,
1603 * return to the caller where the transaction
1604 * can be properly aborted. We just need to
1605 * make sure we're not holding any resources
1606 * that we were not when we came in.
1607 */
1608 xfs_bmap_cancel(&free_list);
1609 return error;
1610 }
1611
1612 /*
1613 * Duplicate the transaction that has the permanent
1614 * reservation and commit the old transaction.
1615 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001616 error = xfs_bmap_finish(tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 ntp = *tp;
David Chinnerf6485052008-04-17 16:50:04 +10001618 if (committed) {
1619 /* link the inode into the next xact in the chain */
1620 xfs_trans_ijoin(ntp, ip,
1621 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1622 xfs_trans_ihold(ntp, ip);
1623 }
1624
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 if (error) {
1626 /*
David Chinnerf6485052008-04-17 16:50:04 +10001627 * If the bmap finish call encounters an error, return
1628 * to the caller where the transaction can be properly
1629 * aborted. We just need to make sure we're not
1630 * holding any resources that we were not when we came
1631 * in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 *
David Chinnerf6485052008-04-17 16:50:04 +10001633 * Aborting from this point might lose some blocks in
1634 * the file system, but oh well.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 */
1636 xfs_bmap_cancel(&free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 return error;
1638 }
1639
1640 if (committed) {
1641 /*
David Chinnerf6485052008-04-17 16:50:04 +10001642 * Mark the inode dirty so it will be logged and
David Chinnere5720ee2008-04-10 12:21:18 +10001643 * moved forward in the log as part of every commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1646 }
David Chinnerf6485052008-04-17 16:50:04 +10001647
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 ntp = xfs_trans_dup(ntp);
David Chinnere5720ee2008-04-10 12:21:18 +10001649 error = xfs_trans_commit(*tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 *tp = ntp;
David Chinnere5720ee2008-04-10 12:21:18 +10001651
David Chinnerf6485052008-04-17 16:50:04 +10001652 /* link the inode into the next transaction in the chain */
1653 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1654 xfs_trans_ihold(ntp, ip);
1655
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001656 if (error)
1657 return error;
1658 /*
1659 * transaction commit worked ok so we can drop the extra ticket
1660 * reference that we gained in xfs_trans_dup()
1661 */
1662 xfs_log_ticket_put(ntp->t_ticket);
1663 error = xfs_trans_reserve(ntp, 0,
David Chinnerf6485052008-04-17 16:50:04 +10001664 XFS_ITRUNCATE_LOG_RES(mp), 0,
1665 XFS_TRANS_PERM_LOG_RES,
1666 XFS_ITRUNCATE_LOG_COUNT);
1667 if (error)
1668 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 }
1670 /*
1671 * Only update the size in the case of the data fork, but
1672 * always re-log the inode so that our permanent transaction
1673 * can keep on rolling it forward in the log.
1674 */
1675 if (fork == XFS_DATA_FORK) {
1676 xfs_isize_check(mp, ip, new_size);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001677 /*
1678 * If we are not changing the file size then do
1679 * not update the on-disk file size - we may be
1680 * called from xfs_inactive_free_eofblocks(). If we
1681 * update the on-disk file size and then the system
1682 * crashes before the contents of the file are
1683 * flushed to disk then the files may be full of
1684 * holes (ie NULL files bug).
1685 */
1686 if (ip->i_size != new_size) {
1687 ip->i_d.di_size = new_size;
1688 ip->i_size = new_size;
1689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 }
1691 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1692 ASSERT((new_size != 0) ||
1693 (fork == XFS_ATTR_FORK) ||
1694 (ip->i_delayed_blks == 0));
1695 ASSERT((new_size != 0) ||
1696 (fork == XFS_ATTR_FORK) ||
1697 (ip->i_d.di_nextents == 0));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001698 trace_xfs_itruncate_finish_end(ip, new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 return 0;
1700}
1701
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702/*
1703 * This is called when the inode's link count goes to 0.
1704 * We place the on-disk inode on a list in the AGI. It
1705 * will be pulled from this list when the inode is freed.
1706 */
1707int
1708xfs_iunlink(
1709 xfs_trans_t *tp,
1710 xfs_inode_t *ip)
1711{
1712 xfs_mount_t *mp;
1713 xfs_agi_t *agi;
1714 xfs_dinode_t *dip;
1715 xfs_buf_t *agibp;
1716 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 xfs_agino_t agino;
1718 short bucket_index;
1719 int offset;
1720 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
1722 ASSERT(ip->i_d.di_nlink == 0);
1723 ASSERT(ip->i_d.di_mode != 0);
1724 ASSERT(ip->i_transp == tp);
1725
1726 mp = tp->t_mountp;
1727
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 /*
1729 * Get the agi buffer first. It ensures lock ordering
1730 * on the list.
1731 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001732 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001733 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001736
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 /*
1738 * Get the index into the agi hash table for the
1739 * list this inode will go on.
1740 */
1741 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1742 ASSERT(agino != 0);
1743 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1744 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001745 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
Christoph Hellwig16259e72005-11-02 15:11:25 +11001747 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 /*
1749 * There is already another inode in the bucket we need
1750 * to add ourselves to. Add us at the front of the list.
1751 * Here we put the head pointer into our next pointer,
1752 * and then we fall through to point the head at us.
1753 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001754 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
Vlad Apostolovc319b582007-11-23 16:27:51 +11001755 if (error)
1756 return error;
1757
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001758 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 /* both on-disk, don't endian flip twice */
1760 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001761 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 offsetof(xfs_dinode_t, di_next_unlinked);
1763 xfs_trans_inode_buf(tp, ibp);
1764 xfs_trans_log_buf(tp, ibp, offset,
1765 (offset + sizeof(xfs_agino_t) - 1));
1766 xfs_inobp_check(mp, ibp);
1767 }
1768
1769 /*
1770 * Point the bucket head pointer at the inode being inserted.
1771 */
1772 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001773 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 offset = offsetof(xfs_agi_t, agi_unlinked) +
1775 (sizeof(xfs_agino_t) * bucket_index);
1776 xfs_trans_log_buf(tp, agibp, offset,
1777 (offset + sizeof(xfs_agino_t) - 1));
1778 return 0;
1779}
1780
1781/*
1782 * Pull the on-disk inode from the AGI unlinked list.
1783 */
1784STATIC int
1785xfs_iunlink_remove(
1786 xfs_trans_t *tp,
1787 xfs_inode_t *ip)
1788{
1789 xfs_ino_t next_ino;
1790 xfs_mount_t *mp;
1791 xfs_agi_t *agi;
1792 xfs_dinode_t *dip;
1793 xfs_buf_t *agibp;
1794 xfs_buf_t *ibp;
1795 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 xfs_agino_t agino;
1797 xfs_agino_t next_agino;
1798 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001799 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001801 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
1807 /*
1808 * Get the agi buffer first. It ensures lock ordering
1809 * on the list.
1810 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001811 error = xfs_read_agi(mp, tp, agno, &agibp);
1812 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001814
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001816
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 /*
1818 * Get the index into the agi hash table for the
1819 * list this inode will go on.
1820 */
1821 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1822 ASSERT(agino != 0);
1823 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001824 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 ASSERT(agi->agi_unlinked[bucket_index]);
1826
Christoph Hellwig16259e72005-11-02 15:11:25 +11001827 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 /*
1829 * We're at the head of the list. Get the inode's
1830 * on-disk buffer to see if there is anyone after us
1831 * on the list. Only modify our next pointer if it
1832 * is not already NULLAGINO. This saves us the overhead
1833 * of dealing with the buffer when there is no need to
1834 * change it.
1835 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001836 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 if (error) {
1838 cmn_err(CE_WARN,
1839 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1840 error, mp->m_fsname);
1841 return error;
1842 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001843 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 ASSERT(next_agino != 0);
1845 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001846 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001847 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 offsetof(xfs_dinode_t, di_next_unlinked);
1849 xfs_trans_inode_buf(tp, ibp);
1850 xfs_trans_log_buf(tp, ibp, offset,
1851 (offset + sizeof(xfs_agino_t) - 1));
1852 xfs_inobp_check(mp, ibp);
1853 } else {
1854 xfs_trans_brelse(tp, ibp);
1855 }
1856 /*
1857 * Point the bucket head pointer at the next inode.
1858 */
1859 ASSERT(next_agino != 0);
1860 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001861 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 offset = offsetof(xfs_agi_t, agi_unlinked) +
1863 (sizeof(xfs_agino_t) * bucket_index);
1864 xfs_trans_log_buf(tp, agibp, offset,
1865 (offset + sizeof(xfs_agino_t) - 1));
1866 } else {
1867 /*
1868 * We need to search the list for the inode being freed.
1869 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001870 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 last_ibp = NULL;
1872 while (next_agino != agino) {
1873 /*
1874 * If the last inode wasn't the one pointing to
1875 * us, then release its buffer since we're not
1876 * going to do anything with it.
1877 */
1878 if (last_ibp != NULL) {
1879 xfs_trans_brelse(tp, last_ibp);
1880 }
1881 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1882 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
Christoph Hellwigc679eef2008-10-30 18:04:13 +11001883 &last_ibp, &last_offset, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 if (error) {
1885 cmn_err(CE_WARN,
1886 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
1887 error, mp->m_fsname);
1888 return error;
1889 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001890 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 ASSERT(next_agino != NULLAGINO);
1892 ASSERT(next_agino != 0);
1893 }
1894 /*
1895 * Now last_ibp points to the buffer previous to us on
1896 * the unlinked list. Pull us from the list.
1897 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001898 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 if (error) {
1900 cmn_err(CE_WARN,
1901 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
1902 error, mp->m_fsname);
1903 return error;
1904 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001905 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 ASSERT(next_agino != 0);
1907 ASSERT(next_agino != agino);
1908 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001909 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001910 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 offsetof(xfs_dinode_t, di_next_unlinked);
1912 xfs_trans_inode_buf(tp, ibp);
1913 xfs_trans_log_buf(tp, ibp, offset,
1914 (offset + sizeof(xfs_agino_t) - 1));
1915 xfs_inobp_check(mp, ibp);
1916 } else {
1917 xfs_trans_brelse(tp, ibp);
1918 }
1919 /*
1920 * Point the previous inode on the list to the next inode.
1921 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001922 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 ASSERT(next_agino != 0);
1924 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1925 xfs_trans_inode_buf(tp, last_ibp);
1926 xfs_trans_log_buf(tp, last_ibp, offset,
1927 (offset + sizeof(xfs_agino_t) - 1));
1928 xfs_inobp_check(mp, last_ibp);
1929 }
1930 return 0;
1931}
1932
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001933STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934xfs_ifree_cluster(
1935 xfs_inode_t *free_ip,
1936 xfs_trans_t *tp,
1937 xfs_ino_t inum)
1938{
1939 xfs_mount_t *mp = free_ip->i_mount;
1940 int blks_per_cluster;
1941 int nbufs;
1942 int ninodes;
1943 int i, j, found, pre_flushed;
1944 xfs_daddr_t blkno;
1945 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 xfs_inode_t *ip, **ip_found;
1947 xfs_inode_log_item_t *iip;
1948 xfs_log_item_t *lip;
Dave Chinner5017e972010-01-11 11:47:40 +00001949 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
Dave Chinner5017e972010-01-11 11:47:40 +00001951 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1953 blks_per_cluster = 1;
1954 ninodes = mp->m_sb.sb_inopblock;
1955 nbufs = XFS_IALLOC_BLOCKS(mp);
1956 } else {
1957 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1958 mp->m_sb.sb_blocksize;
1959 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1960 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1961 }
1962
1963 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
1964
1965 for (j = 0; j < nbufs; j++, inum += ninodes) {
1966 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1967 XFS_INO_TO_AGBNO(mp, inum));
1968
1969
1970 /*
1971 * Look for each inode in memory and attempt to lock it,
1972 * we can be racing with flush and tail pushing here.
1973 * any inode we get the locks on, add to an array of
1974 * inode items to process later.
1975 *
1976 * The get the buffer lock, we could beat a flush
1977 * or tail pushing thread to the lock here, in which
1978 * case they will go looking for the inode buffer
1979 * and fail, we need some other form of interlock
1980 * here.
1981 */
1982 found = 0;
1983 for (i = 0; i < ninodes; i++) {
David Chinnerda353b02007-08-28 14:00:13 +10001984 read_lock(&pag->pag_ici_lock);
1985 ip = radix_tree_lookup(&pag->pag_ici_root,
1986 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
1988 /* Inode not in memory or we found it already,
1989 * nothing to do
1990 */
David Chinner7a18c382006-11-11 18:04:54 +11001991 if (!ip || xfs_iflags_test(ip, XFS_ISTALE)) {
David Chinnerda353b02007-08-28 14:00:13 +10001992 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 continue;
1994 }
1995
1996 if (xfs_inode_clean(ip)) {
David Chinnerda353b02007-08-28 14:00:13 +10001997 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 continue;
1999 }
2000
2001 /* If we can get the locks then add it to the
2002 * list, otherwise by the time we get the bp lock
2003 * below it will already be attached to the
2004 * inode buffer.
2005 */
2006
2007 /* This inode will already be locked - by us, lets
2008 * keep it that way.
2009 */
2010
2011 if (ip == free_ip) {
2012 if (xfs_iflock_nowait(ip)) {
David Chinner7a18c382006-11-11 18:04:54 +11002013 xfs_iflags_set(ip, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 if (xfs_inode_clean(ip)) {
2015 xfs_ifunlock(ip);
2016 } else {
2017 ip_found[found++] = ip;
2018 }
2019 }
David Chinnerda353b02007-08-28 14:00:13 +10002020 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 continue;
2022 }
2023
2024 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2025 if (xfs_iflock_nowait(ip)) {
David Chinner7a18c382006-11-11 18:04:54 +11002026 xfs_iflags_set(ip, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
2028 if (xfs_inode_clean(ip)) {
2029 xfs_ifunlock(ip);
2030 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2031 } else {
2032 ip_found[found++] = ip;
2033 }
2034 } else {
2035 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2036 }
2037 }
David Chinnerda353b02007-08-28 14:00:13 +10002038 read_unlock(&pag->pag_ici_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 }
2040
2041 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2042 mp->m_bsize * blks_per_cluster,
Christoph Hellwig0cadda12010-01-19 09:56:44 +00002043 XBF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
2045 pre_flushed = 0;
2046 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2047 while (lip) {
2048 if (lip->li_type == XFS_LI_INODE) {
2049 iip = (xfs_inode_log_item_t *)lip;
2050 ASSERT(iip->ili_logged == 1);
2051 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
David Chinner7b2e2a32008-10-30 17:39:12 +11002052 xfs_trans_ail_copy_lsn(mp->m_ail,
2053 &iip->ili_flush_lsn,
2054 &iip->ili_item.li_lsn);
David Chinnere5ffd2b2006-11-21 18:55:33 +11002055 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 pre_flushed++;
2057 }
2058 lip = lip->li_bio_list;
2059 }
2060
2061 for (i = 0; i < found; i++) {
2062 ip = ip_found[i];
2063 iip = ip->i_itemp;
2064
2065 if (!iip) {
2066 ip->i_update_core = 0;
2067 xfs_ifunlock(ip);
2068 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2069 continue;
2070 }
2071
2072 iip->ili_last_fields = iip->ili_format.ilf_fields;
2073 iip->ili_format.ilf_fields = 0;
2074 iip->ili_logged = 1;
David Chinner7b2e2a32008-10-30 17:39:12 +11002075 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2076 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
2078 xfs_buf_attach_iodone(bp,
2079 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2080 xfs_istale_done, (xfs_log_item_t *)iip);
2081 if (ip != free_ip) {
2082 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2083 }
2084 }
2085
2086 if (found || pre_flushed)
2087 xfs_trans_stale_inode_buf(tp, bp);
2088 xfs_trans_binval(tp, bp);
2089 }
2090
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002091 kmem_free(ip_found);
Dave Chinner5017e972010-01-11 11:47:40 +00002092 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093}
2094
2095/*
2096 * This is called to return an inode to the inode free list.
2097 * The inode should already be truncated to 0 length and have
2098 * no pages associated with it. This routine also assumes that
2099 * the inode is already a part of the transaction.
2100 *
2101 * The on-disk copy of the inode will have been added to the list
2102 * of unlinked inodes in the AGI. We need to remove the inode from
2103 * that list atomically with respect to freeing it here.
2104 */
2105int
2106xfs_ifree(
2107 xfs_trans_t *tp,
2108 xfs_inode_t *ip,
2109 xfs_bmap_free_t *flist)
2110{
2111 int error;
2112 int delete;
2113 xfs_ino_t first_ino;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002114 xfs_dinode_t *dip;
2115 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002117 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 ASSERT(ip->i_transp == tp);
2119 ASSERT(ip->i_d.di_nlink == 0);
2120 ASSERT(ip->i_d.di_nextents == 0);
2121 ASSERT(ip->i_d.di_anextents == 0);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002122 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2124 ASSERT(ip->i_d.di_nblocks == 0);
2125
2126 /*
2127 * Pull the on-disk inode from the AGI unlinked list.
2128 */
2129 error = xfs_iunlink_remove(tp, ip);
2130 if (error != 0) {
2131 return error;
2132 }
2133
2134 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2135 if (error != 0) {
2136 return error;
2137 }
2138 ip->i_d.di_mode = 0; /* mark incore inode as free */
2139 ip->i_d.di_flags = 0;
2140 ip->i_d.di_dmevmask = 0;
2141 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2142 ip->i_df.if_ext_max =
2143 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2144 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2145 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2146 /*
2147 * Bump the generation count so no one will be confused
2148 * by reincarnations of this inode.
2149 */
2150 ip->i_d.di_gen++;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002151
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2153
Christoph Hellwig0cadda12010-01-19 09:56:44 +00002154 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XBF_LOCK);
Vlad Apostolovc319b582007-11-23 16:27:51 +11002155 if (error)
2156 return error;
2157
2158 /*
2159 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2160 * from picking up this inode when it is reclaimed (its incore state
2161 * initialzed but not flushed to disk yet). The in-core di_mode is
2162 * already cleared and a corresponding transaction logged.
2163 * The hack here just synchronizes the in-core to on-disk
2164 * di_mode value in advance before the actual inode sync to disk.
2165 * This is OK because the inode is already unlinked and would never
2166 * change its di_mode again for this inode generation.
2167 * This is a temporary hack that would require a proper fix
2168 * in the future.
2169 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002170 dip->di_mode = 0;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002171
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 if (delete) {
2173 xfs_ifree_cluster(ip, tp, first_ino);
2174 }
2175
2176 return 0;
2177}
2178
2179/*
2180 * Reallocate the space for if_broot based on the number of records
2181 * being added or deleted as indicated in rec_diff. Move the records
2182 * and pointers in if_broot to fit the new size. When shrinking this
2183 * will eliminate holes between the records and pointers created by
2184 * the caller. When growing this will create holes to be filled in
2185 * by the caller.
2186 *
2187 * The caller must not request to add more records than would fit in
2188 * the on-disk inode root. If the if_broot is currently NULL, then
2189 * if we adding records one will be allocated. The caller must also
2190 * not request that the number of records go below zero, although
2191 * it can go to zero.
2192 *
2193 * ip -- the inode whose if_broot area is changing
2194 * ext_diff -- the change in the number of records, positive or negative,
2195 * requested for the if_broot array.
2196 */
2197void
2198xfs_iroot_realloc(
2199 xfs_inode_t *ip,
2200 int rec_diff,
2201 int whichfork)
2202{
Christoph Hellwig60197e82008-10-30 17:11:19 +11002203 struct xfs_mount *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 int cur_max;
2205 xfs_ifork_t *ifp;
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002206 struct xfs_btree_block *new_broot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 int new_max;
2208 size_t new_size;
2209 char *np;
2210 char *op;
2211
2212 /*
2213 * Handle the degenerate case quietly.
2214 */
2215 if (rec_diff == 0) {
2216 return;
2217 }
2218
2219 ifp = XFS_IFORK_PTR(ip, whichfork);
2220 if (rec_diff > 0) {
2221 /*
2222 * If there wasn't any memory allocated before, just
2223 * allocate it now and get out.
2224 */
2225 if (ifp->if_broot_bytes == 0) {
2226 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002227 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 ifp->if_broot_bytes = (int)new_size;
2229 return;
2230 }
2231
2232 /*
2233 * If there is already an existing if_broot, then we need
2234 * to realloc() it and shift the pointers to their new
2235 * location. The records don't change location because
2236 * they are kept butted up against the btree block header.
2237 */
Christoph Hellwig60197e82008-10-30 17:11:19 +11002238 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 new_max = cur_max + rec_diff;
2240 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002241 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2243 KM_SLEEP);
Christoph Hellwig60197e82008-10-30 17:11:19 +11002244 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2245 ifp->if_broot_bytes);
2246 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2247 (int)new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 ifp->if_broot_bytes = (int)new_size;
2249 ASSERT(ifp->if_broot_bytes <=
2250 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2251 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2252 return;
2253 }
2254
2255 /*
2256 * rec_diff is less than 0. In this case, we are shrinking the
2257 * if_broot buffer. It must already exist. If we go to zero
2258 * records, just get rid of the root and clear the status bit.
2259 */
2260 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
Christoph Hellwig60197e82008-10-30 17:11:19 +11002261 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 new_max = cur_max + rec_diff;
2263 ASSERT(new_max >= 0);
2264 if (new_max > 0)
2265 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2266 else
2267 new_size = 0;
2268 if (new_size > 0) {
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002269 new_broot = kmem_alloc(new_size, KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 /*
2271 * First copy over the btree block header.
2272 */
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002273 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 } else {
2275 new_broot = NULL;
2276 ifp->if_flags &= ~XFS_IFBROOT;
2277 }
2278
2279 /*
2280 * Only copy the records and pointers if there are any.
2281 */
2282 if (new_max > 0) {
2283 /*
2284 * First copy the records.
2285 */
Christoph Hellwig136341b2008-10-30 17:11:40 +11002286 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2287 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2289
2290 /*
2291 * Then copy the pointers.
2292 */
Christoph Hellwig60197e82008-10-30 17:11:19 +11002293 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 ifp->if_broot_bytes);
Christoph Hellwig60197e82008-10-30 17:11:19 +11002295 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 (int)new_size);
2297 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2298 }
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002299 kmem_free(ifp->if_broot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 ifp->if_broot = new_broot;
2301 ifp->if_broot_bytes = (int)new_size;
2302 ASSERT(ifp->if_broot_bytes <=
2303 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2304 return;
2305}
2306
2307
2308/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 * This is called when the amount of space needed for if_data
2310 * is increased or decreased. The change in size is indicated by
2311 * the number of bytes that need to be added or deleted in the
2312 * byte_diff parameter.
2313 *
2314 * If the amount of space needed has decreased below the size of the
2315 * inline buffer, then switch to using the inline buffer. Otherwise,
2316 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2317 * to what is needed.
2318 *
2319 * ip -- the inode whose if_data area is changing
2320 * byte_diff -- the change in the number of bytes, positive or negative,
2321 * requested for the if_data array.
2322 */
2323void
2324xfs_idata_realloc(
2325 xfs_inode_t *ip,
2326 int byte_diff,
2327 int whichfork)
2328{
2329 xfs_ifork_t *ifp;
2330 int new_size;
2331 int real_size;
2332
2333 if (byte_diff == 0) {
2334 return;
2335 }
2336
2337 ifp = XFS_IFORK_PTR(ip, whichfork);
2338 new_size = (int)ifp->if_bytes + byte_diff;
2339 ASSERT(new_size >= 0);
2340
2341 if (new_size == 0) {
2342 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002343 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 }
2345 ifp->if_u1.if_data = NULL;
2346 real_size = 0;
2347 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2348 /*
2349 * If the valid extents/data can fit in if_inline_ext/data,
2350 * copy them from the malloc'd vector and free it.
2351 */
2352 if (ifp->if_u1.if_data == NULL) {
2353 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2354 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2355 ASSERT(ifp->if_real_bytes != 0);
2356 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2357 new_size);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002358 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2360 }
2361 real_size = 0;
2362 } else {
2363 /*
2364 * Stuck with malloc/realloc.
2365 * For inline data, the underlying buffer must be
2366 * a multiple of 4 bytes in size so that it can be
2367 * logged and stay on word boundaries. We enforce
2368 * that here.
2369 */
2370 real_size = roundup(new_size, 4);
2371 if (ifp->if_u1.if_data == NULL) {
2372 ASSERT(ifp->if_real_bytes == 0);
2373 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2374 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2375 /*
2376 * Only do the realloc if the underlying size
2377 * is really changing.
2378 */
2379 if (ifp->if_real_bytes != real_size) {
2380 ifp->if_u1.if_data =
2381 kmem_realloc(ifp->if_u1.if_data,
2382 real_size,
2383 ifp->if_real_bytes,
2384 KM_SLEEP);
2385 }
2386 } else {
2387 ASSERT(ifp->if_real_bytes == 0);
2388 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2389 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2390 ifp->if_bytes);
2391 }
2392 }
2393 ifp->if_real_bytes = real_size;
2394 ifp->if_bytes = new_size;
2395 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2396}
2397
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398void
2399xfs_idestroy_fork(
2400 xfs_inode_t *ip,
2401 int whichfork)
2402{
2403 xfs_ifork_t *ifp;
2404
2405 ifp = XFS_IFORK_PTR(ip, whichfork);
2406 if (ifp->if_broot != NULL) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002407 kmem_free(ifp->if_broot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 ifp->if_broot = NULL;
2409 }
2410
2411 /*
2412 * If the format is local, then we can't have an extents
2413 * array so just look for an inline data array. If we're
2414 * not local then we may or may not have an extents list,
2415 * so check and free it up if we do.
2416 */
2417 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2418 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2419 (ifp->if_u1.if_data != NULL)) {
2420 ASSERT(ifp->if_real_bytes != 0);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002421 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 ifp->if_u1.if_data = NULL;
2423 ifp->if_real_bytes = 0;
2424 }
2425 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11002426 ((ifp->if_flags & XFS_IFEXTIREC) ||
2427 ((ifp->if_u1.if_extents != NULL) &&
2428 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 ASSERT(ifp->if_real_bytes != 0);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002430 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 }
2432 ASSERT(ifp->if_u1.if_extents == NULL ||
2433 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2434 ASSERT(ifp->if_real_bytes == 0);
2435 if (whichfork == XFS_ATTR_FORK) {
2436 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2437 ip->i_afp = NULL;
2438 }
2439}
2440
2441/*
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002442 * This is called to unpin an inode. The caller must have the inode locked
2443 * in at least shared mode so that the buffer cannot be subsequently pinned
2444 * once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 */
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002446static void
2447xfs_iunpin_nowait(
2448 struct xfs_inode *ip)
David Chinnera3f74ff2008-03-06 13:43:42 +11002449{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002450 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002451
Dave Chinner4aaf15d2010-03-08 11:24:07 +11002452 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2453
David Chinnera3f74ff2008-03-06 13:43:42 +11002454 /* Give the log a push to start the unpinning I/O */
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002455 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002456
David Chinnera3f74ff2008-03-06 13:43:42 +11002457}
2458
Dave Chinner777df5a2010-02-06 12:37:26 +11002459void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460xfs_iunpin_wait(
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002461 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462{
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002463 if (xfs_ipincount(ip)) {
2464 xfs_iunpin_nowait(ip);
2465 wait_event(ip->i_ipin_wait, (xfs_ipincount(ip) == 0));
2466 }
David Chinnera3f74ff2008-03-06 13:43:42 +11002467}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469/*
2470 * xfs_iextents_copy()
2471 *
2472 * This is called to copy the REAL extents (as opposed to the delayed
2473 * allocation extents) from the inode into the given buffer. It
2474 * returns the number of bytes copied into the buffer.
2475 *
2476 * If there are no delayed allocation extents, then we can just
2477 * memcpy() the extents into the buffer. Otherwise, we need to
2478 * examine each extent in turn and skip those which are delayed.
2479 */
2480int
2481xfs_iextents_copy(
2482 xfs_inode_t *ip,
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002483 xfs_bmbt_rec_t *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 int whichfork)
2485{
2486 int copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 int i;
2488 xfs_ifork_t *ifp;
2489 int nrecs;
2490 xfs_fsblock_t start_block;
2491
2492 ifp = XFS_IFORK_PTR(ip, whichfork);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002493 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 ASSERT(ifp->if_bytes > 0);
2495
2496 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
Eric Sandeen3a59c942007-07-11 11:09:47 +10002497 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 ASSERT(nrecs > 0);
2499
2500 /*
2501 * There are some delayed allocation extents in the
2502 * inode, so copy the extents one at a time and skip
2503 * the delayed ones. There must be at least one
2504 * non-delayed extent.
2505 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 copied = 0;
2507 for (i = 0; i < nrecs; i++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002508 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 start_block = xfs_bmbt_get_startblock(ep);
Eric Sandeen9d87c312009-01-14 23:22:07 -06002510 if (isnullstartblock(start_block)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 /*
2512 * It's a delayed allocation extent, so skip it.
2513 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 continue;
2515 }
2516
2517 /* Translate to on disk format */
Christoph Hellwigcd8b0a92007-08-16 16:24:15 +10002518 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2519 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002520 dp++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 copied++;
2522 }
2523 ASSERT(copied != 0);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002524 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
2526 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2527}
2528
2529/*
2530 * Each of the following cases stores data into the same region
2531 * of the on-disk inode, so only one of them can be valid at
2532 * any given time. While it is possible to have conflicting formats
2533 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2534 * in EXTENTS format, this can only happen when the fork has
2535 * changed formats after being modified but before being flushed.
2536 * In these cases, the format always takes precedence, because the
2537 * format indicates the current state of the fork.
2538 */
2539/*ARGSUSED*/
David Chinnere4ac9672008-04-10 12:23:58 +10002540STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541xfs_iflush_fork(
2542 xfs_inode_t *ip,
2543 xfs_dinode_t *dip,
2544 xfs_inode_log_item_t *iip,
2545 int whichfork,
2546 xfs_buf_t *bp)
2547{
2548 char *cp;
2549 xfs_ifork_t *ifp;
2550 xfs_mount_t *mp;
2551#ifdef XFS_TRANS_DEBUG
2552 int first;
2553#endif
2554 static const short brootflag[2] =
2555 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2556 static const short dataflag[2] =
2557 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2558 static const short extflag[2] =
2559 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2560
David Chinnere4ac9672008-04-10 12:23:58 +10002561 if (!iip)
2562 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 ifp = XFS_IFORK_PTR(ip, whichfork);
2564 /*
2565 * This can happen if we gave up in iformat in an error path,
2566 * for the attribute fork.
2567 */
David Chinnere4ac9672008-04-10 12:23:58 +10002568 if (!ifp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 ASSERT(whichfork == XFS_ATTR_FORK);
David Chinnere4ac9672008-04-10 12:23:58 +10002570 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 }
2572 cp = XFS_DFORK_PTR(dip, whichfork);
2573 mp = ip->i_mount;
2574 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2575 case XFS_DINODE_FMT_LOCAL:
2576 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2577 (ifp->if_bytes > 0)) {
2578 ASSERT(ifp->if_u1.if_data != NULL);
2579 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2580 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 break;
2583
2584 case XFS_DINODE_FMT_EXTENTS:
2585 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2586 !(iip->ili_format.ilf_fields & extflag[whichfork]));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002587 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2588 (ifp->if_bytes == 0));
2589 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2590 (ifp->if_bytes > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2592 (ifp->if_bytes > 0)) {
2593 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2594 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2595 whichfork);
2596 }
2597 break;
2598
2599 case XFS_DINODE_FMT_BTREE:
2600 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2601 (ifp->if_broot_bytes > 0)) {
2602 ASSERT(ifp->if_broot != NULL);
2603 ASSERT(ifp->if_broot_bytes <=
2604 (XFS_IFORK_SIZE(ip, whichfork) +
2605 XFS_BROOT_SIZE_ADJ));
Christoph Hellwig60197e82008-10-30 17:11:19 +11002606 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 (xfs_bmdr_block_t *)cp,
2608 XFS_DFORK_SIZE(dip, mp, whichfork));
2609 }
2610 break;
2611
2612 case XFS_DINODE_FMT_DEV:
2613 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2614 ASSERT(whichfork == XFS_DATA_FORK);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002615 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 }
2617 break;
2618
2619 case XFS_DINODE_FMT_UUID:
2620 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2621 ASSERT(whichfork == XFS_DATA_FORK);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002622 memcpy(XFS_DFORK_DPTR(dip),
2623 &ip->i_df.if_u2.if_uuid,
2624 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 }
2626 break;
2627
2628 default:
2629 ASSERT(0);
2630 break;
2631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632}
2633
David Chinnerbad55842008-03-06 13:43:49 +11002634STATIC int
2635xfs_iflush_cluster(
2636 xfs_inode_t *ip,
2637 xfs_buf_t *bp)
2638{
2639 xfs_mount_t *mp = ip->i_mount;
Dave Chinner5017e972010-01-11 11:47:40 +00002640 struct xfs_perag *pag;
David Chinnerbad55842008-03-06 13:43:49 +11002641 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10002642 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11002643 int ilist_size;
2644 xfs_inode_t **ilist;
2645 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11002646 int nr_found;
2647 int clcount = 0;
2648 int bufwasdelwri;
2649 int i;
2650
Dave Chinner5017e972010-01-11 11:47:40 +00002651 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
David Chinnerbad55842008-03-06 13:43:49 +11002652 ASSERT(pag->pagi_inodeok);
2653 ASSERT(pag->pag_ici_init);
2654
David Chinnerc8f5f122008-05-20 11:30:15 +10002655 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2656 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10002657 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11002658 if (!ilist)
Dave Chinner44b56e02010-01-11 11:47:43 +00002659 goto out_put;
David Chinnerbad55842008-03-06 13:43:49 +11002660
2661 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2662 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2663 read_lock(&pag->pag_ici_lock);
2664 /* really need a gang lookup range call here */
2665 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10002666 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11002667 if (nr_found == 0)
2668 goto out_free;
2669
2670 for (i = 0; i < nr_found; i++) {
2671 iq = ilist[i];
2672 if (iq == ip)
2673 continue;
2674 /* if the inode lies outside this cluster, we're done. */
2675 if ((XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index)
2676 break;
2677 /*
2678 * Do an un-protected check to see if the inode is dirty and
2679 * is a candidate for flushing. These checks will be repeated
2680 * later after the appropriate locks are acquired.
2681 */
David Chinner33540402008-03-06 13:43:59 +11002682 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11002683 continue;
David Chinnerbad55842008-03-06 13:43:49 +11002684
2685 /*
2686 * Try to get locks. If any are unavailable or it is pinned,
2687 * then this inode cannot be flushed and is skipped.
2688 */
2689
2690 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2691 continue;
2692 if (!xfs_iflock_nowait(iq)) {
2693 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2694 continue;
2695 }
2696 if (xfs_ipincount(iq)) {
2697 xfs_ifunlock(iq);
2698 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2699 continue;
2700 }
2701
2702 /*
2703 * arriving here means that this inode can be flushed. First
2704 * re-check that it's dirty before flushing.
2705 */
David Chinner33540402008-03-06 13:43:59 +11002706 if (!xfs_inode_clean(iq)) {
2707 int error;
David Chinnerbad55842008-03-06 13:43:49 +11002708 error = xfs_iflush_int(iq, bp);
2709 if (error) {
2710 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2711 goto cluster_corrupt_out;
2712 }
2713 clcount++;
2714 } else {
2715 xfs_ifunlock(iq);
2716 }
2717 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2718 }
2719
2720 if (clcount) {
2721 XFS_STATS_INC(xs_icluster_flushcnt);
2722 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2723 }
2724
2725out_free:
2726 read_unlock(&pag->pag_ici_lock);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002727 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00002728out_put:
2729 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11002730 return 0;
2731
2732
2733cluster_corrupt_out:
2734 /*
2735 * Corruption detected in the clustering loop. Invalidate the
2736 * inode buffer and shut down the filesystem.
2737 */
2738 read_unlock(&pag->pag_ici_lock);
2739 /*
2740 * Clean up the buffer. If it was B_DELWRI, just release it --
2741 * brelse can handle it with no problems. If not, shut down the
2742 * filesystem before releasing the buffer.
2743 */
2744 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2745 if (bufwasdelwri)
2746 xfs_buf_relse(bp);
2747
2748 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2749
2750 if (!bufwasdelwri) {
2751 /*
2752 * Just like incore_relse: if we have b_iodone functions,
2753 * mark the buffer as an error and call them. Otherwise
2754 * mark it as stale and brelse.
2755 */
2756 if (XFS_BUF_IODONE_FUNC(bp)) {
2757 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
2758 XFS_BUF_UNDONE(bp);
2759 XFS_BUF_STALE(bp);
David Chinnerbad55842008-03-06 13:43:49 +11002760 XFS_BUF_ERROR(bp,EIO);
2761 xfs_biodone(bp);
2762 } else {
2763 XFS_BUF_STALE(bp);
2764 xfs_buf_relse(bp);
2765 }
2766 }
2767
2768 /*
2769 * Unlocks the flush lock
2770 */
2771 xfs_iflush_abort(iq);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002772 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00002773 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11002774 return XFS_ERROR(EFSCORRUPTED);
2775}
2776
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777/*
2778 * xfs_iflush() will write a modified inode's changes out to the
2779 * inode's on disk home. The caller must have the inode lock held
David Chinnerc63942d2008-08-13 16:41:16 +10002780 * in at least shared mode and the inode flush completion must be
2781 * active as well. The inode lock will still be held upon return from
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 * the call and the caller is free to unlock it.
David Chinnerc63942d2008-08-13 16:41:16 +10002783 * The inode flush will be completed when the inode reaches the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 * The flags indicate how the inode's buffer should be written out.
2785 */
2786int
2787xfs_iflush(
2788 xfs_inode_t *ip,
2789 uint flags)
2790{
2791 xfs_inode_log_item_t *iip;
2792 xfs_buf_t *bp;
2793 xfs_dinode_t *dip;
2794 xfs_mount_t *mp;
2795 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796
2797 XFS_STATS_INC(xs_iflush_count);
2798
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002799 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnerc63942d2008-08-13 16:41:16 +10002800 ASSERT(!completion_done(&ip->i_flush));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2802 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2803
2804 iip = ip->i_itemp;
2805 mp = ip->i_mount;
2806
2807 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002808 * We can't flush the inode until it is unpinned, so wait for it if we
2809 * are allowed to block. We know noone new can pin it, because we are
2810 * holding the inode lock shared and you need to hold it exclusively to
2811 * pin the inode.
2812 *
2813 * If we are not allowed to block, force the log out asynchronously so
2814 * that when we come back the inode will be unpinned. If other inodes
2815 * in the same cluster are dirty, they will probably write the inode
2816 * out for us if they occur after the log force completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 */
Dave Chinnerc8543632010-02-06 12:39:36 +11002818 if (!(flags & SYNC_WAIT) && xfs_ipincount(ip)) {
David Chinnera3f74ff2008-03-06 13:43:42 +11002819 xfs_iunpin_nowait(ip);
2820 xfs_ifunlock(ip);
2821 return EAGAIN;
2822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 xfs_iunpin_wait(ip);
2824
2825 /*
Dave Chinner4b6a4682010-01-11 11:45:21 +00002826 * For stale inodes we cannot rely on the backing buffer remaining
2827 * stale in cache for the remaining life of the stale inode and so
2828 * xfs_itobp() below may give us a buffer that no longer contains
2829 * inodes below. We have to check this after ensuring the inode is
2830 * unpinned so that it is safe to reclaim the stale inode after the
2831 * flush call.
2832 */
2833 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2834 xfs_ifunlock(ip);
2835 return 0;
2836 }
2837
2838 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 * This may have been unpinned because the filesystem is shutting
2840 * down forcibly. If that's the case we must not write this inode
2841 * to disk, because the log record didn't make it to disk!
2842 */
2843 if (XFS_FORCED_SHUTDOWN(mp)) {
2844 ip->i_update_core = 0;
2845 if (iip)
2846 iip->ili_format.ilf_fields = 0;
2847 xfs_ifunlock(ip);
2848 return XFS_ERROR(EIO);
2849 }
2850
2851 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002852 * Get the buffer containing the on-disk inode.
2853 */
Christoph Hellwig76d8b272008-11-28 14:23:40 +11002854 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
Dave Chinnerc8543632010-02-06 12:39:36 +11002855 (flags & SYNC_WAIT) ? XBF_LOCK : XBF_TRYLOCK);
David Chinnera3f74ff2008-03-06 13:43:42 +11002856 if (error || !bp) {
2857 xfs_ifunlock(ip);
2858 return error;
2859 }
2860
2861 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 * First flush out the inode that xfs_iflush was called with.
2863 */
2864 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11002865 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
2868 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002869 * If the buffer is pinned then push on the log now so we won't
2870 * get stuck waiting in the write for too long.
2871 */
2872 if (XFS_BUF_ISPINNED(bp))
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002873 xfs_log_force(mp, 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11002874
2875 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 * inode clustering:
2877 * see if other inodes can be gathered into this write
2878 */
David Chinnerbad55842008-03-06 13:43:49 +11002879 error = xfs_iflush_cluster(ip, bp);
2880 if (error)
2881 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882
Dave Chinnerc8543632010-02-06 12:39:36 +11002883 if (flags & SYNC_WAIT)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 error = xfs_bwrite(mp, bp);
Dave Chinnerc8543632010-02-06 12:39:36 +11002885 else
2886 xfs_bdwrite(mp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 return error;
2888
2889corrupt_out:
2890 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10002891 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892cluster_corrupt_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 /*
2894 * Unlocks the flush lock
2895 */
David Chinnerbad55842008-03-06 13:43:49 +11002896 xfs_iflush_abort(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 return XFS_ERROR(EFSCORRUPTED);
2898}
2899
2900
2901STATIC int
2902xfs_iflush_int(
2903 xfs_inode_t *ip,
2904 xfs_buf_t *bp)
2905{
2906 xfs_inode_log_item_t *iip;
2907 xfs_dinode_t *dip;
2908 xfs_mount_t *mp;
2909#ifdef XFS_TRANS_DEBUG
2910 int first;
2911#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002913 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnerc63942d2008-08-13 16:41:16 +10002914 ASSERT(!completion_done(&ip->i_flush));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2916 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2917
2918 iip = ip->i_itemp;
2919 mp = ip->i_mount;
2920
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 /* set *dip = inode's place in the buffer */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002922 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923
2924 /*
2925 * Clear i_update_core before copying out the data.
2926 * This is for coordination with our timestamp updates
2927 * that don't hold the inode lock. They will always
2928 * update the timestamps BEFORE setting i_update_core,
2929 * so if we clear i_update_core after they set it we
2930 * are guaranteed to see their updates to the timestamps.
2931 * I believe that this depends on strongly ordered memory
2932 * semantics, but we have that. We use the SYNCHRONIZE
2933 * macro to make sure that the compiler does not reorder
2934 * the i_update_core access below the data copy below.
2935 */
2936 ip->i_update_core = 0;
2937 SYNCHRONIZE();
2938
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11002939 /*
Christoph Hellwigf9581b12009-10-06 20:29:26 +00002940 * Make sure to get the latest timestamps from the Linux inode.
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11002941 */
Christoph Hellwigf9581b12009-10-06 20:29:26 +00002942 xfs_synchronize_times(ip);
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11002943
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002944 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2946 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2947 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002948 ip->i_ino, be16_to_cpu(dip->di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 goto corrupt_out;
2950 }
2951 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2952 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2953 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2954 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2955 ip->i_ino, ip, ip->i_d.di_magic);
2956 goto corrupt_out;
2957 }
2958 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
2959 if (XFS_TEST_ERROR(
2960 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2961 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2962 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2963 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2964 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
2965 ip->i_ino, ip);
2966 goto corrupt_out;
2967 }
2968 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
2969 if (XFS_TEST_ERROR(
2970 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2971 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2972 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2973 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2974 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2975 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
2976 ip->i_ino, ip);
2977 goto corrupt_out;
2978 }
2979 }
2980 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2981 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2982 XFS_RANDOM_IFLUSH_5)) {
2983 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2984 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
2985 ip->i_ino,
2986 ip->i_d.di_nextents + ip->i_d.di_anextents,
2987 ip->i_d.di_nblocks,
2988 ip);
2989 goto corrupt_out;
2990 }
2991 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2992 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2993 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
2994 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2995 ip->i_ino, ip->i_d.di_forkoff, ip);
2996 goto corrupt_out;
2997 }
2998 /*
2999 * bump the flush iteration count, used to detect flushes which
3000 * postdate a log record during recovery.
3001 */
3002
3003 ip->i_d.di_flushiter++;
3004
3005 /*
3006 * Copy the dirty parts of the inode into the on-disk
3007 * inode. We always copy out the core of the inode,
3008 * because if the inode is dirty at all the core must
3009 * be.
3010 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003011 xfs_dinode_to_disk(dip, &ip->i_d);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012
3013 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3014 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3015 ip->i_d.di_flushiter = 0;
3016
3017 /*
3018 * If this is really an old format inode and the superblock version
3019 * has not been updated to support only new format inodes, then
3020 * convert back to the old inode format. If the superblock version
3021 * has been updated, then make the conversion permanent.
3022 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11003023 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3024 if (ip->i_d.di_version == 1) {
Eric Sandeen62118702008-03-06 13:44:28 +11003025 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 /*
3027 * Convert it back.
3028 */
3029 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003030 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 } else {
3032 /*
3033 * The superblock version has already been bumped,
3034 * so just make the conversion to the new inode
3035 * format permanent.
3036 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11003037 ip->i_d.di_version = 2;
3038 dip->di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 ip->i_d.di_onlink = 0;
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003040 dip->di_onlink = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003042 memset(&(dip->di_pad[0]), 0,
3043 sizeof(dip->di_pad));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 ASSERT(ip->i_d.di_projid == 0);
3045 }
3046 }
3047
David Chinnere4ac9672008-04-10 12:23:58 +10003048 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3049 if (XFS_IFORK_Q(ip))
3050 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 xfs_inobp_check(mp, bp);
3052
3053 /*
3054 * We've recorded everything logged in the inode, so we'd
3055 * like to clear the ilf_fields bits so we don't log and
3056 * flush things unnecessarily. However, we can't stop
3057 * logging all this information until the data we've copied
3058 * into the disk buffer is written to disk. If we did we might
3059 * overwrite the copy of the inode in the log with all the
3060 * data after re-logging only part of it, and in the face of
3061 * a crash we wouldn't have all the data we need to recover.
3062 *
3063 * What we do is move the bits to the ili_last_fields field.
3064 * When logging the inode, these bits are moved back to the
3065 * ilf_fields field. In the xfs_iflush_done() routine we
3066 * clear ili_last_fields, since we know that the information
3067 * those bits represent is permanently on disk. As long as
3068 * the flush completes before the inode is logged again, then
3069 * both ilf_fields and ili_last_fields will be cleared.
3070 *
3071 * We can play with the ilf_fields bits here, because the inode
3072 * lock must be held exclusively in order to set bits there
3073 * and the flush lock protects the ili_last_fields bits.
3074 * Set ili_logged so the flush done
3075 * routine can tell whether or not to look in the AIL.
3076 * Also, store the current LSN of the inode so that we can tell
3077 * whether the item has moved in the AIL from xfs_iflush_done().
3078 * In order to read the lsn we need the AIL lock, because
3079 * it is a 64 bit value that cannot be read atomically.
3080 */
3081 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3082 iip->ili_last_fields = iip->ili_format.ilf_fields;
3083 iip->ili_format.ilf_fields = 0;
3084 iip->ili_logged = 1;
3085
David Chinner7b2e2a32008-10-30 17:39:12 +11003086 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3087 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088
3089 /*
3090 * Attach the function xfs_iflush_done to the inode's
3091 * buffer. This will remove the inode from the AIL
3092 * and unlock the inode's flush lock when the inode is
3093 * completely written to disk.
3094 */
3095 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3096 xfs_iflush_done, (xfs_log_item_t *)iip);
3097
3098 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3099 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3100 } else {
3101 /*
3102 * We're flushing an inode which is not in the AIL and has
3103 * not been logged but has i_update_core set. For this
3104 * case we can use a B_DELWRI flush and immediately drop
3105 * the inode flush lock because we can avoid the whole
3106 * AIL state thing. It's OK to drop the flush lock now,
3107 * because we've already locked the buffer and to do anything
3108 * you really need both.
3109 */
3110 if (iip != NULL) {
3111 ASSERT(iip->ili_logged == 0);
3112 ASSERT(iip->ili_last_fields == 0);
3113 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3114 }
3115 xfs_ifunlock(ip);
3116 }
3117
3118 return 0;
3119
3120corrupt_out:
3121 return XFS_ERROR(EFSCORRUPTED);
3122}
3123
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003124/*
3125 * Return a pointer to the extent record at file index idx.
3126 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003127xfs_bmbt_rec_host_t *
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003128xfs_iext_get_ext(
3129 xfs_ifork_t *ifp, /* inode fork pointer */
3130 xfs_extnum_t idx) /* index of target extent */
3131{
3132 ASSERT(idx >= 0);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003133 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3134 return ifp->if_u1.if_ext_irec->er_extbuf;
3135 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3136 xfs_ext_irec_t *erp; /* irec pointer */
3137 int erp_idx = 0; /* irec index */
3138 xfs_extnum_t page_idx = idx; /* ext index in target list */
3139
3140 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3141 return &erp->er_extbuf[page_idx];
3142 } else if (ifp->if_bytes) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003143 return &ifp->if_u1.if_extents[idx];
3144 } else {
3145 return NULL;
3146 }
3147}
3148
3149/*
3150 * Insert new item(s) into the extent records for incore inode
3151 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3152 */
3153void
3154xfs_iext_insert(
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003155 xfs_inode_t *ip, /* incore inode pointer */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003156 xfs_extnum_t idx, /* starting index of new items */
3157 xfs_extnum_t count, /* number of inserted items */
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003158 xfs_bmbt_irec_t *new, /* items to insert */
3159 int state) /* type of extent conversion */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003160{
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003161 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003162 xfs_extnum_t i; /* extent record index */
3163
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003164 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
3165
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003166 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3167 xfs_iext_add(ifp, idx, count);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003168 for (i = idx; i < idx + count; i++, new++)
3169 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003170}
3171
3172/*
3173 * This is called when the amount of space required for incore file
3174 * extents needs to be increased. The ext_diff parameter stores the
3175 * number of new extents being added and the idx parameter contains
3176 * the extent index where the new extents will be added. If the new
3177 * extents are being appended, then we just need to (re)allocate and
3178 * initialize the space. Otherwise, if the new extents are being
3179 * inserted into the middle of the existing entries, a bit more work
3180 * is required to make room for the new extents to be inserted. The
3181 * caller is responsible for filling in the new extent entries upon
3182 * return.
3183 */
3184void
3185xfs_iext_add(
3186 xfs_ifork_t *ifp, /* inode fork pointer */
3187 xfs_extnum_t idx, /* index to begin adding exts */
Nathan Scottc41564b2006-03-29 08:55:14 +10003188 int ext_diff) /* number of extents to add */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003189{
3190 int byte_diff; /* new bytes being added */
3191 int new_size; /* size of extents after adding */
3192 xfs_extnum_t nextents; /* number of extents in file */
3193
3194 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3195 ASSERT((idx >= 0) && (idx <= nextents));
3196 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3197 new_size = ifp->if_bytes + byte_diff;
3198 /*
3199 * If the new number of extents (nextents + ext_diff)
3200 * fits inside the inode, then continue to use the inline
3201 * extent buffer.
3202 */
3203 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3204 if (idx < nextents) {
3205 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3206 &ifp->if_u2.if_inline_ext[idx],
3207 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3208 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3209 }
3210 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3211 ifp->if_real_bytes = 0;
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003212 ifp->if_lastex = nextents + ext_diff;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003213 }
3214 /*
3215 * Otherwise use a linear (direct) extent list.
3216 * If the extents are currently inside the inode,
3217 * xfs_iext_realloc_direct will switch us from
3218 * inline to direct extent allocation mode.
3219 */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003220 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003221 xfs_iext_realloc_direct(ifp, new_size);
3222 if (idx < nextents) {
3223 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3224 &ifp->if_u1.if_extents[idx],
3225 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3226 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3227 }
3228 }
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003229 /* Indirection array */
3230 else {
3231 xfs_ext_irec_t *erp;
3232 int erp_idx = 0;
3233 int page_idx = idx;
3234
3235 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3236 if (ifp->if_flags & XFS_IFEXTIREC) {
3237 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3238 } else {
3239 xfs_iext_irec_init(ifp);
3240 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3241 erp = ifp->if_u1.if_ext_irec;
3242 }
3243 /* Extents fit in target extent page */
3244 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3245 if (page_idx < erp->er_extcount) {
3246 memmove(&erp->er_extbuf[page_idx + ext_diff],
3247 &erp->er_extbuf[page_idx],
3248 (erp->er_extcount - page_idx) *
3249 sizeof(xfs_bmbt_rec_t));
3250 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3251 }
3252 erp->er_extcount += ext_diff;
3253 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3254 }
3255 /* Insert a new extent page */
3256 else if (erp) {
3257 xfs_iext_add_indirect_multi(ifp,
3258 erp_idx, page_idx, ext_diff);
3259 }
3260 /*
3261 * If extent(s) are being appended to the last page in
3262 * the indirection array and the new extent(s) don't fit
3263 * in the page, then erp is NULL and erp_idx is set to
3264 * the next index needed in the indirection array.
3265 */
3266 else {
3267 int count = ext_diff;
3268
3269 while (count) {
3270 erp = xfs_iext_irec_new(ifp, erp_idx);
3271 erp->er_extcount = count;
3272 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3273 if (count) {
3274 erp_idx++;
3275 }
3276 }
3277 }
3278 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003279 ifp->if_bytes = new_size;
3280}
3281
3282/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003283 * This is called when incore extents are being added to the indirection
3284 * array and the new extents do not fit in the target extent list. The
3285 * erp_idx parameter contains the irec index for the target extent list
3286 * in the indirection array, and the idx parameter contains the extent
3287 * index within the list. The number of extents being added is stored
3288 * in the count parameter.
3289 *
3290 * |-------| |-------|
3291 * | | | | idx - number of extents before idx
3292 * | idx | | count |
3293 * | | | | count - number of extents being inserted at idx
3294 * |-------| |-------|
3295 * | count | | nex2 | nex2 - number of extents after idx + count
3296 * |-------| |-------|
3297 */
3298void
3299xfs_iext_add_indirect_multi(
3300 xfs_ifork_t *ifp, /* inode fork pointer */
3301 int erp_idx, /* target extent irec index */
3302 xfs_extnum_t idx, /* index within target list */
3303 int count) /* new extents being added */
3304{
3305 int byte_diff; /* new bytes being added */
3306 xfs_ext_irec_t *erp; /* pointer to irec entry */
3307 xfs_extnum_t ext_diff; /* number of extents to add */
3308 xfs_extnum_t ext_cnt; /* new extents still needed */
3309 xfs_extnum_t nex2; /* extents after idx + count */
3310 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3311 int nlists; /* number of irec's (lists) */
3312
3313 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3314 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3315 nex2 = erp->er_extcount - idx;
3316 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3317
3318 /*
3319 * Save second part of target extent list
3320 * (all extents past */
3321 if (nex2) {
3322 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
David Chinner67850732008-08-13 16:02:51 +10003323 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003324 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3325 erp->er_extcount -= nex2;
3326 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3327 memset(&erp->er_extbuf[idx], 0, byte_diff);
3328 }
3329
3330 /*
3331 * Add the new extents to the end of the target
3332 * list, then allocate new irec record(s) and
3333 * extent buffer(s) as needed to store the rest
3334 * of the new extents.
3335 */
3336 ext_cnt = count;
3337 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3338 if (ext_diff) {
3339 erp->er_extcount += ext_diff;
3340 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3341 ext_cnt -= ext_diff;
3342 }
3343 while (ext_cnt) {
3344 erp_idx++;
3345 erp = xfs_iext_irec_new(ifp, erp_idx);
3346 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3347 erp->er_extcount = ext_diff;
3348 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3349 ext_cnt -= ext_diff;
3350 }
3351
3352 /* Add nex2 extents back to indirection array */
3353 if (nex2) {
3354 xfs_extnum_t ext_avail;
3355 int i;
3356
3357 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3358 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3359 i = 0;
3360 /*
3361 * If nex2 extents fit in the current page, append
3362 * nex2_ep after the new extents.
3363 */
3364 if (nex2 <= ext_avail) {
3365 i = erp->er_extcount;
3366 }
3367 /*
3368 * Otherwise, check if space is available in the
3369 * next page.
3370 */
3371 else if ((erp_idx < nlists - 1) &&
3372 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3373 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3374 erp_idx++;
3375 erp++;
3376 /* Create a hole for nex2 extents */
3377 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3378 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3379 }
3380 /*
3381 * Final choice, create a new extent page for
3382 * nex2 extents.
3383 */
3384 else {
3385 erp_idx++;
3386 erp = xfs_iext_irec_new(ifp, erp_idx);
3387 }
3388 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003389 kmem_free(nex2_ep);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003390 erp->er_extcount += nex2;
3391 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3392 }
3393}
3394
3395/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003396 * This is called when the amount of space required for incore file
3397 * extents needs to be decreased. The ext_diff parameter stores the
3398 * number of extents to be removed and the idx parameter contains
3399 * the extent index where the extents will be removed from.
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003400 *
3401 * If the amount of space needed has decreased below the linear
3402 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3403 * extent array. Otherwise, use kmem_realloc() to adjust the
3404 * size to what is needed.
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003405 */
3406void
3407xfs_iext_remove(
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003408 xfs_inode_t *ip, /* incore inode pointer */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003409 xfs_extnum_t idx, /* index to begin removing exts */
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003410 int ext_diff, /* number of extents to remove */
3411 int state) /* type of extent conversion */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003412{
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003413 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003414 xfs_extnum_t nextents; /* number of extents in file */
3415 int new_size; /* size of extents after removal */
3416
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003417 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3418
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003419 ASSERT(ext_diff > 0);
3420 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3421 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3422
3423 if (new_size == 0) {
3424 xfs_iext_destroy(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003425 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3426 xfs_iext_remove_indirect(ifp, idx, ext_diff);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003427 } else if (ifp->if_real_bytes) {
3428 xfs_iext_remove_direct(ifp, idx, ext_diff);
3429 } else {
3430 xfs_iext_remove_inline(ifp, idx, ext_diff);
3431 }
3432 ifp->if_bytes = new_size;
3433}
3434
3435/*
3436 * This removes ext_diff extents from the inline buffer, beginning
3437 * at extent index idx.
3438 */
3439void
3440xfs_iext_remove_inline(
3441 xfs_ifork_t *ifp, /* inode fork pointer */
3442 xfs_extnum_t idx, /* index to begin removing exts */
3443 int ext_diff) /* number of extents to remove */
3444{
3445 int nextents; /* number of extents in file */
3446
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003447 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003448 ASSERT(idx < XFS_INLINE_EXTS);
3449 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3450 ASSERT(((nextents - ext_diff) > 0) &&
3451 (nextents - ext_diff) < XFS_INLINE_EXTS);
3452
3453 if (idx + ext_diff < nextents) {
3454 memmove(&ifp->if_u2.if_inline_ext[idx],
3455 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3456 (nextents - (idx + ext_diff)) *
3457 sizeof(xfs_bmbt_rec_t));
3458 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3459 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3460 } else {
3461 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3462 ext_diff * sizeof(xfs_bmbt_rec_t));
3463 }
3464}
3465
3466/*
3467 * This removes ext_diff extents from a linear (direct) extent list,
3468 * beginning at extent index idx. If the extents are being removed
3469 * from the end of the list (ie. truncate) then we just need to re-
3470 * allocate the list to remove the extra space. Otherwise, if the
3471 * extents are being removed from the middle of the existing extent
3472 * entries, then we first need to move the extent records beginning
3473 * at idx + ext_diff up in the list to overwrite the records being
3474 * removed, then remove the extra space via kmem_realloc.
3475 */
3476void
3477xfs_iext_remove_direct(
3478 xfs_ifork_t *ifp, /* inode fork pointer */
3479 xfs_extnum_t idx, /* index to begin removing exts */
3480 int ext_diff) /* number of extents to remove */
3481{
3482 xfs_extnum_t nextents; /* number of extents in file */
3483 int new_size; /* size of extents after removal */
3484
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003485 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003486 new_size = ifp->if_bytes -
3487 (ext_diff * sizeof(xfs_bmbt_rec_t));
3488 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3489
3490 if (new_size == 0) {
3491 xfs_iext_destroy(ifp);
3492 return;
3493 }
3494 /* Move extents up in the list (if needed) */
3495 if (idx + ext_diff < nextents) {
3496 memmove(&ifp->if_u1.if_extents[idx],
3497 &ifp->if_u1.if_extents[idx + ext_diff],
3498 (nextents - (idx + ext_diff)) *
3499 sizeof(xfs_bmbt_rec_t));
3500 }
3501 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3502 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3503 /*
3504 * Reallocate the direct extent list. If the extents
3505 * will fit inside the inode then xfs_iext_realloc_direct
3506 * will switch from direct to inline extent allocation
3507 * mode for us.
3508 */
3509 xfs_iext_realloc_direct(ifp, new_size);
3510 ifp->if_bytes = new_size;
3511}
3512
3513/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003514 * This is called when incore extents are being removed from the
3515 * indirection array and the extents being removed span multiple extent
3516 * buffers. The idx parameter contains the file extent index where we
3517 * want to begin removing extents, and the count parameter contains
3518 * how many extents need to be removed.
3519 *
3520 * |-------| |-------|
3521 * | nex1 | | | nex1 - number of extents before idx
3522 * |-------| | count |
3523 * | | | | count - number of extents being removed at idx
3524 * | count | |-------|
3525 * | | | nex2 | nex2 - number of extents after idx + count
3526 * |-------| |-------|
3527 */
3528void
3529xfs_iext_remove_indirect(
3530 xfs_ifork_t *ifp, /* inode fork pointer */
3531 xfs_extnum_t idx, /* index to begin removing extents */
3532 int count) /* number of extents to remove */
3533{
3534 xfs_ext_irec_t *erp; /* indirection array pointer */
3535 int erp_idx = 0; /* indirection array index */
3536 xfs_extnum_t ext_cnt; /* extents left to remove */
3537 xfs_extnum_t ext_diff; /* extents to remove in current list */
3538 xfs_extnum_t nex1; /* number of extents before idx */
3539 xfs_extnum_t nex2; /* extents after idx + count */
Nathan Scottc41564b2006-03-29 08:55:14 +10003540 int nlists; /* entries in indirection array */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003541 int page_idx = idx; /* index in target extent list */
3542
3543 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3544 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3545 ASSERT(erp != NULL);
3546 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3547 nex1 = page_idx;
3548 ext_cnt = count;
3549 while (ext_cnt) {
3550 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3551 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3552 /*
3553 * Check for deletion of entire list;
3554 * xfs_iext_irec_remove() updates extent offsets.
3555 */
3556 if (ext_diff == erp->er_extcount) {
3557 xfs_iext_irec_remove(ifp, erp_idx);
3558 ext_cnt -= ext_diff;
3559 nex1 = 0;
3560 if (ext_cnt) {
3561 ASSERT(erp_idx < ifp->if_real_bytes /
3562 XFS_IEXT_BUFSZ);
3563 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3564 nex1 = 0;
3565 continue;
3566 } else {
3567 break;
3568 }
3569 }
3570 /* Move extents up (if needed) */
3571 if (nex2) {
3572 memmove(&erp->er_extbuf[nex1],
3573 &erp->er_extbuf[nex1 + ext_diff],
3574 nex2 * sizeof(xfs_bmbt_rec_t));
3575 }
3576 /* Zero out rest of page */
3577 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3578 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3579 /* Update remaining counters */
3580 erp->er_extcount -= ext_diff;
3581 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3582 ext_cnt -= ext_diff;
3583 nex1 = 0;
3584 erp_idx++;
3585 erp++;
3586 }
3587 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3588 xfs_iext_irec_compact(ifp);
3589}
3590
3591/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003592 * Create, destroy, or resize a linear (direct) block of extents.
3593 */
3594void
3595xfs_iext_realloc_direct(
3596 xfs_ifork_t *ifp, /* inode fork pointer */
3597 int new_size) /* new size of extents */
3598{
3599 int rnew_size; /* real new size of extents */
3600
3601 rnew_size = new_size;
3602
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003603 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3604 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3605 (new_size != ifp->if_real_bytes)));
3606
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003607 /* Free extent records */
3608 if (new_size == 0) {
3609 xfs_iext_destroy(ifp);
3610 }
3611 /* Resize direct extent list and zero any new bytes */
3612 else if (ifp->if_real_bytes) {
3613 /* Check if extents will fit inside the inode */
3614 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3615 xfs_iext_direct_to_inline(ifp, new_size /
3616 (uint)sizeof(xfs_bmbt_rec_t));
3617 ifp->if_bytes = new_size;
3618 return;
3619 }
Vignesh Babu16a087d2007-06-28 16:46:37 +10003620 if (!is_power_of_2(new_size)){
Robert P. J. Day40ebd812007-11-23 16:30:51 +11003621 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003622 }
3623 if (rnew_size != ifp->if_real_bytes) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003624 ifp->if_u1.if_extents =
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003625 kmem_realloc(ifp->if_u1.if_extents,
3626 rnew_size,
David Chinner67850732008-08-13 16:02:51 +10003627 ifp->if_real_bytes, KM_NOFS);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003628 }
3629 if (rnew_size > ifp->if_real_bytes) {
3630 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3631 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3632 rnew_size - ifp->if_real_bytes);
3633 }
3634 }
3635 /*
3636 * Switch from the inline extent buffer to a direct
3637 * extent list. Be sure to include the inline extent
3638 * bytes in new_size.
3639 */
3640 else {
3641 new_size += ifp->if_bytes;
Vignesh Babu16a087d2007-06-28 16:46:37 +10003642 if (!is_power_of_2(new_size)) {
Robert P. J. Day40ebd812007-11-23 16:30:51 +11003643 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003644 }
3645 xfs_iext_inline_to_direct(ifp, rnew_size);
3646 }
3647 ifp->if_real_bytes = rnew_size;
3648 ifp->if_bytes = new_size;
3649}
3650
3651/*
3652 * Switch from linear (direct) extent records to inline buffer.
3653 */
3654void
3655xfs_iext_direct_to_inline(
3656 xfs_ifork_t *ifp, /* inode fork pointer */
3657 xfs_extnum_t nextents) /* number of extents in file */
3658{
3659 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3660 ASSERT(nextents <= XFS_INLINE_EXTS);
3661 /*
3662 * The inline buffer was zeroed when we switched
3663 * from inline to direct extent allocation mode,
3664 * so we don't need to clear it here.
3665 */
3666 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3667 nextents * sizeof(xfs_bmbt_rec_t));
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003668 kmem_free(ifp->if_u1.if_extents);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003669 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3670 ifp->if_real_bytes = 0;
3671}
3672
3673/*
3674 * Switch from inline buffer to linear (direct) extent records.
3675 * new_size should already be rounded up to the next power of 2
3676 * by the caller (when appropriate), so use new_size as it is.
3677 * However, since new_size may be rounded up, we can't update
3678 * if_bytes here. It is the caller's responsibility to update
3679 * if_bytes upon return.
3680 */
3681void
3682xfs_iext_inline_to_direct(
3683 xfs_ifork_t *ifp, /* inode fork pointer */
3684 int new_size) /* number of extents in file */
3685{
David Chinner67850732008-08-13 16:02:51 +10003686 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003687 memset(ifp->if_u1.if_extents, 0, new_size);
3688 if (ifp->if_bytes) {
3689 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3690 ifp->if_bytes);
3691 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3692 sizeof(xfs_bmbt_rec_t));
3693 }
3694 ifp->if_real_bytes = new_size;
3695}
3696
3697/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003698 * Resize an extent indirection array to new_size bytes.
3699 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003700STATIC void
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003701xfs_iext_realloc_indirect(
3702 xfs_ifork_t *ifp, /* inode fork pointer */
3703 int new_size) /* new indirection array size */
3704{
3705 int nlists; /* number of irec's (ex lists) */
3706 int size; /* current indirection array size */
3707
3708 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3709 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3710 size = nlists * sizeof(xfs_ext_irec_t);
3711 ASSERT(ifp->if_real_bytes);
3712 ASSERT((new_size >= 0) && (new_size != size));
3713 if (new_size == 0) {
3714 xfs_iext_destroy(ifp);
3715 } else {
3716 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3717 kmem_realloc(ifp->if_u1.if_ext_irec,
David Chinner67850732008-08-13 16:02:51 +10003718 new_size, size, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003719 }
3720}
3721
3722/*
3723 * Switch from indirection array to linear (direct) extent allocations.
3724 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003725STATIC void
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003726xfs_iext_indirect_to_direct(
3727 xfs_ifork_t *ifp) /* inode fork pointer */
3728{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003729 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003730 xfs_extnum_t nextents; /* number of extents in file */
3731 int size; /* size of file extents */
3732
3733 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3734 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3735 ASSERT(nextents <= XFS_LINEAR_EXTS);
3736 size = nextents * sizeof(xfs_bmbt_rec_t);
3737
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10003738 xfs_iext_irec_compact_pages(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003739 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3740
3741 ep = ifp->if_u1.if_ext_irec->er_extbuf;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003742 kmem_free(ifp->if_u1.if_ext_irec);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003743 ifp->if_flags &= ~XFS_IFEXTIREC;
3744 ifp->if_u1.if_extents = ep;
3745 ifp->if_bytes = size;
3746 if (nextents < XFS_LINEAR_EXTS) {
3747 xfs_iext_realloc_direct(ifp, size);
3748 }
3749}
3750
3751/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003752 * Free incore file extents.
3753 */
3754void
3755xfs_iext_destroy(
3756 xfs_ifork_t *ifp) /* inode fork pointer */
3757{
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003758 if (ifp->if_flags & XFS_IFEXTIREC) {
3759 int erp_idx;
3760 int nlists;
3761
3762 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3763 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3764 xfs_iext_irec_remove(ifp, erp_idx);
3765 }
3766 ifp->if_flags &= ~XFS_IFEXTIREC;
3767 } else if (ifp->if_real_bytes) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003768 kmem_free(ifp->if_u1.if_extents);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003769 } else if (ifp->if_bytes) {
3770 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3771 sizeof(xfs_bmbt_rec_t));
3772 }
3773 ifp->if_u1.if_extents = NULL;
3774 ifp->if_real_bytes = 0;
3775 ifp->if_bytes = 0;
3776}
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003777
3778/*
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003779 * Return a pointer to the extent record for file system block bno.
3780 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003781xfs_bmbt_rec_host_t * /* pointer to found extent record */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003782xfs_iext_bno_to_ext(
3783 xfs_ifork_t *ifp, /* inode fork pointer */
3784 xfs_fileoff_t bno, /* block number to search for */
3785 xfs_extnum_t *idxp) /* index of target extent */
3786{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003787 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003788 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003789 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003790 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
Nathan Scottc41564b2006-03-29 08:55:14 +10003791 int high; /* upper boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003792 xfs_extnum_t idx = 0; /* index of target extent */
Nathan Scottc41564b2006-03-29 08:55:14 +10003793 int low; /* lower boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003794 xfs_extnum_t nextents; /* number of file extents */
3795 xfs_fileoff_t startoff = 0; /* start offset of extent */
3796
3797 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3798 if (nextents == 0) {
3799 *idxp = 0;
3800 return NULL;
3801 }
3802 low = 0;
3803 if (ifp->if_flags & XFS_IFEXTIREC) {
3804 /* Find target extent list */
3805 int erp_idx = 0;
3806 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3807 base = erp->er_extbuf;
3808 high = erp->er_extcount - 1;
3809 } else {
3810 base = ifp->if_u1.if_extents;
3811 high = nextents - 1;
3812 }
3813 /* Binary search extent records */
3814 while (low <= high) {
3815 idx = (low + high) >> 1;
3816 ep = base + idx;
3817 startoff = xfs_bmbt_get_startoff(ep);
3818 blockcount = xfs_bmbt_get_blockcount(ep);
3819 if (bno < startoff) {
3820 high = idx - 1;
3821 } else if (bno >= startoff + blockcount) {
3822 low = idx + 1;
3823 } else {
3824 /* Convert back to file-based extent index */
3825 if (ifp->if_flags & XFS_IFEXTIREC) {
3826 idx += erp->er_extoff;
3827 }
3828 *idxp = idx;
3829 return ep;
3830 }
3831 }
3832 /* Convert back to file-based extent index */
3833 if (ifp->if_flags & XFS_IFEXTIREC) {
3834 idx += erp->er_extoff;
3835 }
3836 if (bno >= startoff + blockcount) {
3837 if (++idx == nextents) {
3838 ep = NULL;
3839 } else {
3840 ep = xfs_iext_get_ext(ifp, idx);
3841 }
3842 }
3843 *idxp = idx;
3844 return ep;
3845}
3846
3847/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003848 * Return a pointer to the indirection array entry containing the
3849 * extent record for filesystem block bno. Store the index of the
3850 * target irec in *erp_idxp.
3851 */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003852xfs_ext_irec_t * /* pointer to found extent record */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003853xfs_iext_bno_to_irec(
3854 xfs_ifork_t *ifp, /* inode fork pointer */
3855 xfs_fileoff_t bno, /* block number to search for */
3856 int *erp_idxp) /* irec index of target ext list */
3857{
3858 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3859 xfs_ext_irec_t *erp_next; /* next indirection array entry */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003860 int erp_idx; /* indirection array index */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003861 int nlists; /* number of extent irec's (lists) */
3862 int high; /* binary search upper limit */
3863 int low; /* binary search lower limit */
3864
3865 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3866 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3867 erp_idx = 0;
3868 low = 0;
3869 high = nlists - 1;
3870 while (low <= high) {
3871 erp_idx = (low + high) >> 1;
3872 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3873 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3874 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3875 high = erp_idx - 1;
3876 } else if (erp_next && bno >=
3877 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3878 low = erp_idx + 1;
3879 } else {
3880 break;
3881 }
3882 }
3883 *erp_idxp = erp_idx;
3884 return erp;
3885}
3886
3887/*
3888 * Return a pointer to the indirection array entry containing the
3889 * extent record at file extent index *idxp. Store the index of the
3890 * target irec in *erp_idxp and store the page index of the target
3891 * extent record in *idxp.
3892 */
3893xfs_ext_irec_t *
3894xfs_iext_idx_to_irec(
3895 xfs_ifork_t *ifp, /* inode fork pointer */
3896 xfs_extnum_t *idxp, /* extent index (file -> page) */
3897 int *erp_idxp, /* pointer to target irec */
3898 int realloc) /* new bytes were just added */
3899{
3900 xfs_ext_irec_t *prev; /* pointer to previous irec */
3901 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3902 int erp_idx; /* indirection array index */
3903 int nlists; /* number of irec's (ex lists) */
3904 int high; /* binary search upper limit */
3905 int low; /* binary search lower limit */
3906 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3907
3908 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3909 ASSERT(page_idx >= 0 && page_idx <=
3910 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
3911 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3912 erp_idx = 0;
3913 low = 0;
3914 high = nlists - 1;
3915
3916 /* Binary search extent irec's */
3917 while (low <= high) {
3918 erp_idx = (low + high) >> 1;
3919 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3920 prev = erp_idx > 0 ? erp - 1 : NULL;
3921 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3922 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3923 high = erp_idx - 1;
3924 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3925 (page_idx == erp->er_extoff + erp->er_extcount &&
3926 !realloc)) {
3927 low = erp_idx + 1;
3928 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3929 erp->er_extcount == XFS_LINEAR_EXTS) {
3930 ASSERT(realloc);
3931 page_idx = 0;
3932 erp_idx++;
3933 erp = erp_idx < nlists ? erp + 1 : NULL;
3934 break;
3935 } else {
3936 page_idx -= erp->er_extoff;
3937 break;
3938 }
3939 }
3940 *idxp = page_idx;
3941 *erp_idxp = erp_idx;
3942 return(erp);
3943}
3944
3945/*
3946 * Allocate and initialize an indirection array once the space needed
3947 * for incore extents increases above XFS_IEXT_BUFSZ.
3948 */
3949void
3950xfs_iext_irec_init(
3951 xfs_ifork_t *ifp) /* inode fork pointer */
3952{
3953 xfs_ext_irec_t *erp; /* indirection array pointer */
3954 xfs_extnum_t nextents; /* number of extents in file */
3955
3956 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3957 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3958 ASSERT(nextents <= XFS_LINEAR_EXTS);
3959
David Chinner67850732008-08-13 16:02:51 +10003960 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003961
3962 if (nextents == 0) {
David Chinner67850732008-08-13 16:02:51 +10003963 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003964 } else if (!ifp->if_real_bytes) {
3965 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3966 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3967 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3968 }
3969 erp->er_extbuf = ifp->if_u1.if_extents;
3970 erp->er_extcount = nextents;
3971 erp->er_extoff = 0;
3972
3973 ifp->if_flags |= XFS_IFEXTIREC;
3974 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3975 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3976 ifp->if_u1.if_ext_irec = erp;
3977
3978 return;
3979}
3980
3981/*
3982 * Allocate and initialize a new entry in the indirection array.
3983 */
3984xfs_ext_irec_t *
3985xfs_iext_irec_new(
3986 xfs_ifork_t *ifp, /* inode fork pointer */
3987 int erp_idx) /* index for new irec */
3988{
3989 xfs_ext_irec_t *erp; /* indirection array pointer */
3990 int i; /* loop counter */
3991 int nlists; /* number of irec's (ex lists) */
3992
3993 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3994 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3995
3996 /* Resize indirection array */
3997 xfs_iext_realloc_indirect(ifp, ++nlists *
3998 sizeof(xfs_ext_irec_t));
3999 /*
4000 * Move records down in the array so the
4001 * new page can use erp_idx.
4002 */
4003 erp = ifp->if_u1.if_ext_irec;
4004 for (i = nlists - 1; i > erp_idx; i--) {
4005 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4006 }
4007 ASSERT(i == erp_idx);
4008
4009 /* Initialize new extent record */
4010 erp = ifp->if_u1.if_ext_irec;
David Chinner67850732008-08-13 16:02:51 +10004011 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004012 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4013 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4014 erp[erp_idx].er_extcount = 0;
4015 erp[erp_idx].er_extoff = erp_idx > 0 ?
4016 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4017 return (&erp[erp_idx]);
4018}
4019
4020/*
4021 * Remove a record from the indirection array.
4022 */
4023void
4024xfs_iext_irec_remove(
4025 xfs_ifork_t *ifp, /* inode fork pointer */
4026 int erp_idx) /* irec index to remove */
4027{
4028 xfs_ext_irec_t *erp; /* indirection array pointer */
4029 int i; /* loop counter */
4030 int nlists; /* number of irec's (ex lists) */
4031
4032 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4033 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4034 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4035 if (erp->er_extbuf) {
4036 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4037 -erp->er_extcount);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004038 kmem_free(erp->er_extbuf);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004039 }
4040 /* Compact extent records */
4041 erp = ifp->if_u1.if_ext_irec;
4042 for (i = erp_idx; i < nlists - 1; i++) {
4043 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4044 }
4045 /*
4046 * Manually free the last extent record from the indirection
4047 * array. A call to xfs_iext_realloc_indirect() with a size
4048 * of zero would result in a call to xfs_iext_destroy() which
4049 * would in turn call this function again, creating a nasty
4050 * infinite loop.
4051 */
4052 if (--nlists) {
4053 xfs_iext_realloc_indirect(ifp,
4054 nlists * sizeof(xfs_ext_irec_t));
4055 } else {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004056 kmem_free(ifp->if_u1.if_ext_irec);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004057 }
4058 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4059}
4060
4061/*
4062 * This is called to clean up large amounts of unused memory allocated
4063 * by the indirection array. Before compacting anything though, verify
4064 * that the indirection array is still needed and switch back to the
4065 * linear extent list (or even the inline buffer) if possible. The
4066 * compaction policy is as follows:
4067 *
4068 * Full Compaction: Extents fit into a single page (or inline buffer)
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10004069 * Partial Compaction: Extents occupy less than 50% of allocated space
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004070 * No Compaction: Extents occupy at least 50% of allocated space
4071 */
4072void
4073xfs_iext_irec_compact(
4074 xfs_ifork_t *ifp) /* inode fork pointer */
4075{
4076 xfs_extnum_t nextents; /* number of extents in file */
4077 int nlists; /* number of irec's (ex lists) */
4078
4079 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4080 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4081 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4082
4083 if (nextents == 0) {
4084 xfs_iext_destroy(ifp);
4085 } else if (nextents <= XFS_INLINE_EXTS) {
4086 xfs_iext_indirect_to_direct(ifp);
4087 xfs_iext_direct_to_inline(ifp, nextents);
4088 } else if (nextents <= XFS_LINEAR_EXTS) {
4089 xfs_iext_indirect_to_direct(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004090 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4091 xfs_iext_irec_compact_pages(ifp);
4092 }
4093}
4094
4095/*
4096 * Combine extents from neighboring extent pages.
4097 */
4098void
4099xfs_iext_irec_compact_pages(
4100 xfs_ifork_t *ifp) /* inode fork pointer */
4101{
4102 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4103 int erp_idx = 0; /* indirection array index */
4104 int nlists; /* number of irec's (ex lists) */
4105
4106 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4107 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4108 while (erp_idx < nlists - 1) {
4109 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4110 erp_next = erp + 1;
4111 if (erp_next->er_extcount <=
4112 (XFS_LINEAR_EXTS - erp->er_extcount)) {
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10004113 memcpy(&erp->er_extbuf[erp->er_extcount],
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004114 erp_next->er_extbuf, erp_next->er_extcount *
4115 sizeof(xfs_bmbt_rec_t));
4116 erp->er_extcount += erp_next->er_extcount;
4117 /*
4118 * Free page before removing extent record
4119 * so er_extoffs don't get modified in
4120 * xfs_iext_irec_remove.
4121 */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004122 kmem_free(erp_next->er_extbuf);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004123 erp_next->er_extbuf = NULL;
4124 xfs_iext_irec_remove(ifp, erp_idx + 1);
4125 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4126 } else {
4127 erp_idx++;
4128 }
4129 }
4130}
4131
4132/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004133 * This is called to update the er_extoff field in the indirection
4134 * array when extents have been added or removed from one of the
4135 * extent lists. erp_idx contains the irec index to begin updating
4136 * at and ext_diff contains the number of extents that were added
4137 * or removed.
4138 */
4139void
4140xfs_iext_irec_update_extoffs(
4141 xfs_ifork_t *ifp, /* inode fork pointer */
4142 int erp_idx, /* irec index to update */
4143 int ext_diff) /* number of new extents */
4144{
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 for (i = erp_idx; i < nlists; i++) {
4151 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
4152 }
4153}