blob: a37480a6e023c2f5481cd8befb1265f904c0268d [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_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_inode_item.h"
39#include "xfs_btree.h"
Christoph Hellwig8c4ed632008-10-30 16:55:13 +110040#include "xfs_btree_trace.h"
Nathan Scotta844f452005-11-02 14:38:42 +110041#include "xfs_alloc.h"
42#include "xfs_ialloc.h"
43#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include "xfs_utils.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include "xfs_quota.h"
David Chinner2a82b8b2007-07-11 11:09:12 +100047#include "xfs_filestream.h"
Christoph Hellwig739bfb22007-08-29 10:58:01 +100048#include "xfs_vnodeops.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000049#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Linus Torvalds1da177e2005-04-16 15:20:36 -070051kmem_zone_t *xfs_ifork_zone;
52kmem_zone_t *xfs_inode_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
55 * Used in xfs_itruncate(). This is the maximum number of extents
56 * freed from a file in a single transaction.
57 */
58#define XFS_ITRUNC_MAX_EXTENTS 2
59
60STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
61STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
62STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
63STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065#ifdef DEBUG
66/*
67 * Make sure that the extents in the given memory buffer
68 * are valid.
69 */
70STATIC void
71xfs_validate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +110072 xfs_ifork_t *ifp,
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 int nrecs,
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 xfs_exntfmt_t fmt)
75{
76 xfs_bmbt_irec_t irec;
Christoph Hellwiga6f64d42007-08-16 16:23:40 +100077 xfs_bmbt_rec_host_t rec;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 int i;
79
80 for (i = 0; i < nrecs; i++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +100081 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
82 rec.l0 = get_unaligned(&ep->l0);
83 rec.l1 = get_unaligned(&ep->l1);
84 xfs_bmbt_get_all(&rec, &irec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 if (fmt == XFS_EXTFMT_NOSTATE)
86 ASSERT(irec.br_state == XFS_EXT_NORM);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 }
88}
89#else /* DEBUG */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +100090#define xfs_validate_extents(ifp, nrecs, fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091#endif /* DEBUG */
92
93/*
94 * Check that none of the inode's in the buffer have a next
95 * unlinked field of 0.
96 */
97#if defined(DEBUG)
98void
99xfs_inobp_check(
100 xfs_mount_t *mp,
101 xfs_buf_t *bp)
102{
103 int i;
104 int j;
105 xfs_dinode_t *dip;
106
107 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
108
109 for (i = 0; i < j; i++) {
110 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
111 i * mp->m_sb.sb_inodesize);
112 if (!dip->di_next_unlinked) {
Dave Chinner53487782011-03-07 10:05:35 +1100113 xfs_alert(mp,
114 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 bp);
116 ASSERT(dip->di_next_unlinked);
117 }
118 }
119}
120#endif
121
122/*
David Chinner4ae29b42008-03-06 13:43:34 +1100123 * Find the buffer associated with the given inode map
124 * We do basic validation checks on the buffer once it has been
125 * retrieved from disk.
126 */
127STATIC int
128xfs_imap_to_bp(
129 xfs_mount_t *mp,
130 xfs_trans_t *tp,
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100131 struct xfs_imap *imap,
David Chinner4ae29b42008-03-06 13:43:34 +1100132 xfs_buf_t **bpp,
133 uint buf_flags,
Christoph Hellwigb48d8d62008-11-28 14:23:41 +1100134 uint iget_flags)
David Chinner4ae29b42008-03-06 13:43:34 +1100135{
136 int error;
137 int i;
138 int ni;
139 xfs_buf_t *bp;
140
141 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
David Chinnera3f74ff2008-03-06 13:43:42 +1100142 (int)imap->im_len, buf_flags, &bp);
David Chinner4ae29b42008-03-06 13:43:34 +1100143 if (error) {
David Chinnera3f74ff2008-03-06 13:43:42 +1100144 if (error != EAGAIN) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100145 xfs_warn(mp,
146 "%s: xfs_trans_read_buf() returned error %d.",
147 __func__, error);
David Chinnera3f74ff2008-03-06 13:43:42 +1100148 } else {
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000149 ASSERT(buf_flags & XBF_TRYLOCK);
David Chinnera3f74ff2008-03-06 13:43:42 +1100150 }
David Chinner4ae29b42008-03-06 13:43:34 +1100151 return error;
152 }
153
154 /*
155 * Validate the magic number and version of every inode in the buffer
156 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
157 */
158#ifdef DEBUG
159 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
160#else /* usual case */
161 ni = 1;
162#endif
163
164 for (i = 0; i < ni; i++) {
165 int di_ok;
166 xfs_dinode_t *dip;
167
168 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
169 (i << mp->m_sb.sb_inodelog));
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100170 di_ok = be16_to_cpu(dip->di_magic) == XFS_DINODE_MAGIC &&
171 XFS_DINODE_GOOD_VERSION(dip->di_version);
David Chinner4ae29b42008-03-06 13:43:34 +1100172 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
173 XFS_ERRTAG_ITOBP_INOTOBP,
174 XFS_RANDOM_ITOBP_INOTOBP))) {
Dave Chinner19207792010-06-24 11:15:47 +1000175 if (iget_flags & XFS_IGET_UNTRUSTED) {
David Chinner4ae29b42008-03-06 13:43:34 +1100176 xfs_trans_brelse(tp, bp);
177 return XFS_ERROR(EINVAL);
178 }
179 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
180 XFS_ERRLEVEL_HIGH, mp, dip);
181#ifdef DEBUG
Dave Chinner0b932cc2011-03-07 10:08:35 +1100182 xfs_emerg(mp,
183 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
David Chinner4ae29b42008-03-06 13:43:34 +1100184 (unsigned long long)imap->im_blkno, i,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100185 be16_to_cpu(dip->di_magic));
Dave Chinner0b932cc2011-03-07 10:08:35 +1100186 ASSERT(0);
David Chinner4ae29b42008-03-06 13:43:34 +1100187#endif
188 xfs_trans_brelse(tp, bp);
189 return XFS_ERROR(EFSCORRUPTED);
190 }
191 }
192
193 xfs_inobp_check(mp, bp);
194
195 /*
196 * Mark the buffer as an inode buffer now that it looks good
197 */
198 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
199
200 *bpp = bp;
201 return 0;
202}
203
204/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 * This routine is called to map an inode number within a file
206 * system to the buffer containing the on-disk version of the
207 * inode. It returns a pointer to the buffer containing the
208 * on-disk inode in the bpp parameter, and in the dip parameter
209 * it returns a pointer to the on-disk inode within that buffer.
210 *
211 * If a non-zero error is returned, then the contents of bpp and
212 * dipp are undefined.
213 *
214 * Use xfs_imap() to determine the size and location of the
215 * buffer to read from disk.
216 */
Christoph Hellwigc679eef2008-10-30 18:04:13 +1100217int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218xfs_inotobp(
219 xfs_mount_t *mp,
220 xfs_trans_t *tp,
221 xfs_ino_t ino,
222 xfs_dinode_t **dipp,
223 xfs_buf_t **bpp,
Christoph Hellwigc679eef2008-10-30 18:04:13 +1100224 int *offset,
225 uint imap_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226{
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100227 struct xfs_imap imap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 xfs_buf_t *bp;
229 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 imap.im_blkno = 0;
Christoph Hellwiga1941892008-11-28 14:23:40 +1100232 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
David Chinner4ae29b42008-03-06 13:43:34 +1100233 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000236 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XBF_LOCK, imap_flags);
David Chinner4ae29b42008-03-06 13:43:34 +1100237 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
241 *bpp = bp;
242 *offset = imap.im_boffset;
243 return 0;
244}
245
246
247/*
248 * This routine is called to map an inode to the buffer containing
249 * the on-disk version of the inode. It returns a pointer to the
250 * buffer containing the on-disk inode in the bpp parameter, and in
251 * the dip parameter it returns a pointer to the on-disk inode within
252 * that buffer.
253 *
254 * If a non-zero error is returned, then the contents of bpp and
255 * dipp are undefined.
256 *
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100257 * The inode is expected to already been mapped to its buffer and read
258 * in once, thus we can use the mapping information stored in the inode
259 * rather than calling xfs_imap(). This allows us to avoid the overhead
260 * of looking at the inode btree for small block file systems
Christoph Hellwig94e1b692008-11-28 14:23:41 +1100261 * (see xfs_imap()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 */
263int
264xfs_itobp(
265 xfs_mount_t *mp,
266 xfs_trans_t *tp,
267 xfs_inode_t *ip,
268 xfs_dinode_t **dipp,
269 xfs_buf_t **bpp,
David Chinnera3f74ff2008-03-06 13:43:42 +1100270 uint buf_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
272 xfs_buf_t *bp;
273 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100275 ASSERT(ip->i_imap.im_blkno != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100277 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
David Chinner4ae29b42008-03-06 13:43:34 +1100278 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 return error;
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000280
David Chinnera3f74ff2008-03-06 13:43:42 +1100281 if (!bp) {
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000282 ASSERT(buf_flags & XBF_TRYLOCK);
David Chinnera3f74ff2008-03-06 13:43:42 +1100283 ASSERT(tp == NULL);
284 *bpp = NULL;
285 return EAGAIN;
286 }
287
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100288 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 *bpp = bp;
290 return 0;
291}
292
293/*
294 * Move inode type and inode format specific information from the
295 * on-disk inode to the in-core inode. For fifos, devs, and sockets
296 * this means set if_rdev to the proper value. For files, directories,
297 * and symlinks this means to bring in the in-line data or extent
298 * pointers. For a file in B-tree format, only the root is immediately
299 * brought in-core. The rest will be in-lined in if_extents when it
300 * is first referenced (see xfs_iread_extents()).
301 */
302STATIC int
303xfs_iformat(
304 xfs_inode_t *ip,
305 xfs_dinode_t *dip)
306{
307 xfs_attr_shortform_t *atp;
308 int size;
309 int error;
310 xfs_fsize_t di_size;
311 ip->i_df.if_ext_max =
312 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
313 error = 0;
314
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100315 if (unlikely(be32_to_cpu(dip->di_nextents) +
316 be16_to_cpu(dip->di_anextents) >
317 be64_to_cpu(dip->di_nblocks))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100318 xfs_warn(ip->i_mount,
Nathan Scott3762ec62006-01-12 10:29:53 +1100319 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 (unsigned long long)ip->i_ino,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100321 (int)(be32_to_cpu(dip->di_nextents) +
322 be16_to_cpu(dip->di_anextents)),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 (unsigned long long)
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100324 be64_to_cpu(dip->di_nblocks));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
326 ip->i_mount, dip);
327 return XFS_ERROR(EFSCORRUPTED);
328 }
329
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100330 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
Dave Chinner65333b42011-03-07 10:03:35 +1100331 xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 (unsigned long long)ip->i_ino,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100333 dip->di_forkoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
335 ip->i_mount, dip);
336 return XFS_ERROR(EFSCORRUPTED);
337 }
338
Christoph Hellwigb89d4202009-08-10 11:32:18 -0300339 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
340 !ip->i_mount->m_rtdev_targp)) {
Dave Chinner65333b42011-03-07 10:03:35 +1100341 xfs_warn(ip->i_mount,
Christoph Hellwigb89d4202009-08-10 11:32:18 -0300342 "corrupt dinode %Lu, has realtime flag set.",
343 ip->i_ino);
344 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
345 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
346 return XFS_ERROR(EFSCORRUPTED);
347 }
348
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 switch (ip->i_d.di_mode & S_IFMT) {
350 case S_IFIFO:
351 case S_IFCHR:
352 case S_IFBLK:
353 case S_IFSOCK:
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100354 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
356 ip->i_mount, dip);
357 return XFS_ERROR(EFSCORRUPTED);
358 }
359 ip->i_d.di_size = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000360 ip->i_size = 0;
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100361 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 break;
363
364 case S_IFREG:
365 case S_IFLNK:
366 case S_IFDIR:
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100367 switch (dip->di_format) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 case XFS_DINODE_FMT_LOCAL:
369 /*
370 * no local regular files yet
371 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100372 if (unlikely((be16_to_cpu(dip->di_mode) & S_IFMT) == S_IFREG)) {
Dave Chinner65333b42011-03-07 10:03:35 +1100373 xfs_warn(ip->i_mount,
374 "corrupt inode %Lu (local format for regular file).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 (unsigned long long) ip->i_ino);
376 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
377 XFS_ERRLEVEL_LOW,
378 ip->i_mount, dip);
379 return XFS_ERROR(EFSCORRUPTED);
380 }
381
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100382 di_size = be64_to_cpu(dip->di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100384 xfs_warn(ip->i_mount,
385 "corrupt inode %Lu (bad size %Ld for local inode).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 (unsigned long long) ip->i_ino,
387 (long long) di_size);
388 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
389 XFS_ERRLEVEL_LOW,
390 ip->i_mount, dip);
391 return XFS_ERROR(EFSCORRUPTED);
392 }
393
394 size = (int)di_size;
395 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
396 break;
397 case XFS_DINODE_FMT_EXTENTS:
398 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
399 break;
400 case XFS_DINODE_FMT_BTREE:
401 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
402 break;
403 default:
404 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
405 ip->i_mount);
406 return XFS_ERROR(EFSCORRUPTED);
407 }
408 break;
409
410 default:
411 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
412 return XFS_ERROR(EFSCORRUPTED);
413 }
414 if (error) {
415 return error;
416 }
417 if (!XFS_DFORK_Q(dip))
418 return 0;
419 ASSERT(ip->i_afp == NULL);
Dave Chinner4a7eddd2010-07-20 17:53:59 +1000420 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 ip->i_afp->if_ext_max =
422 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100423 switch (dip->di_aformat) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 case XFS_DINODE_FMT_LOCAL:
425 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
Nathan Scott3b244aa2006-03-17 17:29:25 +1100426 size = be16_to_cpu(atp->hdr.totsize);
Christoph Hellwig2809f762009-01-19 02:04:16 +0100427
428 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100429 xfs_warn(ip->i_mount,
430 "corrupt inode %Lu (bad attr fork size %Ld).",
Christoph Hellwig2809f762009-01-19 02:04:16 +0100431 (unsigned long long) ip->i_ino,
432 (long long) size);
433 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
434 XFS_ERRLEVEL_LOW,
435 ip->i_mount, dip);
436 return XFS_ERROR(EFSCORRUPTED);
437 }
438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
440 break;
441 case XFS_DINODE_FMT_EXTENTS:
442 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
443 break;
444 case XFS_DINODE_FMT_BTREE:
445 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
446 break;
447 default:
448 error = XFS_ERROR(EFSCORRUPTED);
449 break;
450 }
451 if (error) {
452 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
453 ip->i_afp = NULL;
454 xfs_idestroy_fork(ip, XFS_DATA_FORK);
455 }
456 return error;
457}
458
459/*
460 * The file is in-lined in the on-disk inode.
461 * If it fits into if_inline_data, then copy
462 * it there, otherwise allocate a buffer for it
463 * and copy the data there. Either way, set
464 * if_data to point at the data.
465 * If we allocate a buffer for the data, make
466 * sure that its size is a multiple of 4 and
467 * record the real size in i_real_bytes.
468 */
469STATIC int
470xfs_iformat_local(
471 xfs_inode_t *ip,
472 xfs_dinode_t *dip,
473 int whichfork,
474 int size)
475{
476 xfs_ifork_t *ifp;
477 int real_size;
478
479 /*
480 * If the size is unreasonable, then something
481 * is wrong and we just bail out rather than crash in
482 * kmem_alloc() or memcpy() below.
483 */
484 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100485 xfs_warn(ip->i_mount,
486 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 (unsigned long long) ip->i_ino, size,
488 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
489 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
490 ip->i_mount, dip);
491 return XFS_ERROR(EFSCORRUPTED);
492 }
493 ifp = XFS_IFORK_PTR(ip, whichfork);
494 real_size = 0;
495 if (size == 0)
496 ifp->if_u1.if_data = NULL;
497 else if (size <= sizeof(ifp->if_u2.if_inline_data))
498 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
499 else {
500 real_size = roundup(size, 4);
Dave Chinner4a7eddd2010-07-20 17:53:59 +1000501 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 }
503 ifp->if_bytes = size;
504 ifp->if_real_bytes = real_size;
505 if (size)
506 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
507 ifp->if_flags &= ~XFS_IFEXTENTS;
508 ifp->if_flags |= XFS_IFINLINE;
509 return 0;
510}
511
512/*
513 * The file consists of a set of extents all
514 * of which fit into the on-disk inode.
515 * If there are few enough extents to fit into
516 * the if_inline_ext, then copy them there.
517 * Otherwise allocate a buffer for them and copy
518 * them into it. Either way, set if_extents
519 * to point at the extents.
520 */
521STATIC int
522xfs_iformat_extents(
523 xfs_inode_t *ip,
524 xfs_dinode_t *dip,
525 int whichfork)
526{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000527 xfs_bmbt_rec_t *dp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 xfs_ifork_t *ifp;
529 int nex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 int size;
531 int i;
532
533 ifp = XFS_IFORK_PTR(ip, whichfork);
534 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
535 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
536
537 /*
538 * If the number of extents is unreasonable, then something
539 * is wrong and we just bail out rather than crash in
540 * kmem_alloc() or memcpy() below.
541 */
542 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Dave Chinner65333b42011-03-07 10:03:35 +1100543 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 (unsigned long long) ip->i_ino, nex);
545 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
546 ip->i_mount, dip);
547 return XFS_ERROR(EFSCORRUPTED);
548 }
549
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100550 ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 if (nex == 0)
552 ifp->if_u1.if_extents = NULL;
553 else if (nex <= XFS_INLINE_EXTS)
554 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100555 else
556 xfs_iext_add(ifp, 0, nex);
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 ifp->if_bytes = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 if (size) {
560 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000561 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100562 for (i = 0; i < nex; i++, dp++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000563 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Harvey Harrison597bca62008-08-13 16:29:21 +1000564 ep->l0 = get_unaligned_be64(&dp->l0);
565 ep->l1 = get_unaligned_be64(&dp->l1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 }
Eric Sandeen3a59c942007-07-11 11:09:47 +1000567 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 if (whichfork != XFS_DATA_FORK ||
569 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
570 if (unlikely(xfs_check_nostate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100571 ifp, 0, nex))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
573 XFS_ERRLEVEL_LOW,
574 ip->i_mount);
575 return XFS_ERROR(EFSCORRUPTED);
576 }
577 }
578 ifp->if_flags |= XFS_IFEXTENTS;
579 return 0;
580}
581
582/*
583 * The file has too many extents to fit into
584 * the inode, so they are in B-tree format.
585 * Allocate a buffer for the root of the B-tree
586 * and copy the root into it. The i_extents
587 * field will remain NULL until all of the
588 * extents are read in (when they are needed).
589 */
590STATIC int
591xfs_iformat_btree(
592 xfs_inode_t *ip,
593 xfs_dinode_t *dip,
594 int whichfork)
595{
596 xfs_bmdr_block_t *dfp;
597 xfs_ifork_t *ifp;
598 /* REFERENCED */
599 int nrecs;
600 int size;
601
602 ifp = XFS_IFORK_PTR(ip, whichfork);
603 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
604 size = XFS_BMAP_BROOT_SPACE(dfp);
Christoph Hellwig60197e82008-10-30 17:11:19 +1100605 nrecs = be16_to_cpu(dfp->bb_numrecs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607 /*
608 * blow out if -- fork has less extents than can fit in
609 * fork (fork shouldn't be a btree format), root btree
610 * block has more records than can fit into the fork,
611 * or the number of extents is greater than the number of
612 * blocks.
613 */
614 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
615 || XFS_BMDR_SPACE_CALC(nrecs) >
616 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
617 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
Dave Chinner65333b42011-03-07 10:03:35 +1100618 xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 (unsigned long long) ip->i_ino);
Dave Chinner65333b42011-03-07 10:03:35 +1100620 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
621 ip->i_mount, dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 return XFS_ERROR(EFSCORRUPTED);
623 }
624
625 ifp->if_broot_bytes = size;
Dave Chinner4a7eddd2010-07-20 17:53:59 +1000626 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 ASSERT(ifp->if_broot != NULL);
628 /*
629 * Copy and convert from the on-disk structure
630 * to the in-memory structure.
631 */
Christoph Hellwig60197e82008-10-30 17:11:19 +1100632 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
633 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
634 ifp->if_broot, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 ifp->if_flags &= ~XFS_IFEXTENTS;
636 ifp->if_flags |= XFS_IFBROOT;
637
638 return 0;
639}
640
Eric Sandeend96f8f82009-07-02 00:09:33 -0500641STATIC void
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000642xfs_dinode_from_disk(
643 xfs_icdinode_t *to,
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100644 xfs_dinode_t *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000646 to->di_magic = be16_to_cpu(from->di_magic);
647 to->di_mode = be16_to_cpu(from->di_mode);
648 to->di_version = from ->di_version;
649 to->di_format = from->di_format;
650 to->di_onlink = be16_to_cpu(from->di_onlink);
651 to->di_uid = be32_to_cpu(from->di_uid);
652 to->di_gid = be32_to_cpu(from->di_gid);
653 to->di_nlink = be32_to_cpu(from->di_nlink);
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000654 to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
655 to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000656 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
657 to->di_flushiter = be16_to_cpu(from->di_flushiter);
658 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
659 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
660 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
661 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
662 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
663 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
664 to->di_size = be64_to_cpu(from->di_size);
665 to->di_nblocks = be64_to_cpu(from->di_nblocks);
666 to->di_extsize = be32_to_cpu(from->di_extsize);
667 to->di_nextents = be32_to_cpu(from->di_nextents);
668 to->di_anextents = be16_to_cpu(from->di_anextents);
669 to->di_forkoff = from->di_forkoff;
670 to->di_aformat = from->di_aformat;
671 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
672 to->di_dmstate = be16_to_cpu(from->di_dmstate);
673 to->di_flags = be16_to_cpu(from->di_flags);
674 to->di_gen = be32_to_cpu(from->di_gen);
675}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000677void
678xfs_dinode_to_disk(
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100679 xfs_dinode_t *to,
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000680 xfs_icdinode_t *from)
681{
682 to->di_magic = cpu_to_be16(from->di_magic);
683 to->di_mode = cpu_to_be16(from->di_mode);
684 to->di_version = from ->di_version;
685 to->di_format = from->di_format;
686 to->di_onlink = cpu_to_be16(from->di_onlink);
687 to->di_uid = cpu_to_be32(from->di_uid);
688 to->di_gid = cpu_to_be32(from->di_gid);
689 to->di_nlink = cpu_to_be32(from->di_nlink);
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000690 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
691 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000692 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
693 to->di_flushiter = cpu_to_be16(from->di_flushiter);
694 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
695 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
696 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
697 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
698 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
699 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
700 to->di_size = cpu_to_be64(from->di_size);
701 to->di_nblocks = cpu_to_be64(from->di_nblocks);
702 to->di_extsize = cpu_to_be32(from->di_extsize);
703 to->di_nextents = cpu_to_be32(from->di_nextents);
704 to->di_anextents = cpu_to_be16(from->di_anextents);
705 to->di_forkoff = from->di_forkoff;
706 to->di_aformat = from->di_aformat;
707 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
708 to->di_dmstate = cpu_to_be16(from->di_dmstate);
709 to->di_flags = cpu_to_be16(from->di_flags);
710 to->di_gen = cpu_to_be32(from->di_gen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711}
712
713STATIC uint
714_xfs_dic2xflags(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 __uint16_t di_flags)
716{
717 uint flags = 0;
718
719 if (di_flags & XFS_DIFLAG_ANY) {
720 if (di_flags & XFS_DIFLAG_REALTIME)
721 flags |= XFS_XFLAG_REALTIME;
722 if (di_flags & XFS_DIFLAG_PREALLOC)
723 flags |= XFS_XFLAG_PREALLOC;
724 if (di_flags & XFS_DIFLAG_IMMUTABLE)
725 flags |= XFS_XFLAG_IMMUTABLE;
726 if (di_flags & XFS_DIFLAG_APPEND)
727 flags |= XFS_XFLAG_APPEND;
728 if (di_flags & XFS_DIFLAG_SYNC)
729 flags |= XFS_XFLAG_SYNC;
730 if (di_flags & XFS_DIFLAG_NOATIME)
731 flags |= XFS_XFLAG_NOATIME;
732 if (di_flags & XFS_DIFLAG_NODUMP)
733 flags |= XFS_XFLAG_NODUMP;
734 if (di_flags & XFS_DIFLAG_RTINHERIT)
735 flags |= XFS_XFLAG_RTINHERIT;
736 if (di_flags & XFS_DIFLAG_PROJINHERIT)
737 flags |= XFS_XFLAG_PROJINHERIT;
738 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
739 flags |= XFS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100740 if (di_flags & XFS_DIFLAG_EXTSIZE)
741 flags |= XFS_XFLAG_EXTSIZE;
742 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
743 flags |= XFS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000744 if (di_flags & XFS_DIFLAG_NODEFRAG)
745 flags |= XFS_XFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +1000746 if (di_flags & XFS_DIFLAG_FILESTREAM)
747 flags |= XFS_XFLAG_FILESTREAM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 }
749
750 return flags;
751}
752
753uint
754xfs_ip2xflags(
755 xfs_inode_t *ip)
756{
Christoph Hellwig347d1c02007-08-28 13:57:51 +1000757 xfs_icdinode_t *dic = &ip->i_d;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Nathan Scotta916e2b2006-06-09 17:12:17 +1000759 return _xfs_dic2xflags(dic->di_flags) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100760 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
763uint
764xfs_dic2xflags(
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100765 xfs_dinode_t *dip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766{
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100767 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
Christoph Hellwig45ba5982007-12-07 14:07:20 +1100768 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769}
770
771/*
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100772 * Read the disk inode attributes into the in-core inode structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 */
774int
775xfs_iread(
776 xfs_mount_t *mp,
777 xfs_trans_t *tp,
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100778 xfs_inode_t *ip,
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100779 uint iget_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780{
781 xfs_buf_t *bp;
782 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 int error;
784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 /*
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100786 * Fill in the location information in the in-core inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 */
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100788 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100789 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100790 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
792 /*
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100793 * Get pointers to the on-disk inode and the buffer containing it.
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100794 */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100795 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
Christoph Hellwig0cadda12010-01-19 09:56:44 +0000796 XBF_LOCK, iget_flags);
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100797 if (error)
Christoph Hellwig24f211b2008-11-28 14:23:42 +1100798 return error;
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +1100799 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Christoph Hellwig76d8b272008-11-28 14:23:40 +1100800
801 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 * If we got something that isn't an inode it means someone
803 * (nfs or dmi) has a stale handle.
804 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100805 if (be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806#ifdef DEBUG
Dave Chinner53487782011-03-07 10:05:35 +1100807 xfs_alert(mp,
808 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
809 __func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810#endif /* DEBUG */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100811 error = XFS_ERROR(EINVAL);
812 goto out_brelse;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 }
814
815 /*
816 * If the on-disk inode is already linked to a directory
817 * entry, copy all of the inode into the in-core inode.
818 * xfs_iformat() handles copying in the inode format
819 * specific information.
820 * Otherwise, just get the truly permanent information.
821 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100822 if (dip->di_mode) {
823 xfs_dinode_from_disk(&ip->i_d, dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 error = xfs_iformat(ip, dip);
825 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826#ifdef DEBUG
Dave Chinner53487782011-03-07 10:05:35 +1100827 xfs_alert(mp, "%s: xfs_iformat() returned error %d",
828 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829#endif /* DEBUG */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100830 goto out_brelse;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 }
832 } else {
Christoph Hellwig81591fe2008-11-28 14:23:39 +1100833 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
834 ip->i_d.di_version = dip->di_version;
835 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
836 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 /*
838 * Make sure to pull in the mode here as well in
839 * case the inode is released without being used.
840 * This ensures that xfs_inactive() will see that
841 * the inode is already free and not try to mess
842 * with the uninitialized part of it.
843 */
844 ip->i_d.di_mode = 0;
845 /*
846 * Initialize the per-fork minima and maxima for a new
847 * inode here. xfs_iformat will do it for old inodes.
848 */
849 ip->i_df.if_ext_max =
850 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
851 }
852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 /*
854 * The inode format changed when we moved the link count and
855 * made it 32 bits long. If this is an old format inode,
856 * convert it in memory to look like a new one. If it gets
857 * flushed to disk we will convert back before flushing or
858 * logging it. We zero out the new projid field and the old link
859 * count field. We'll handle clearing the pad field (the remains
860 * of the old uuid field) when we actually convert the inode to
861 * the new format. We don't change the version number so that we
862 * can distinguish this from a real new format inode.
863 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +1100864 if (ip->i_d.di_version == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 ip->i_d.di_nlink = ip->i_d.di_onlink;
866 ip->i_d.di_onlink = 0;
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000867 xfs_set_projid(ip, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 }
869
870 ip->i_delayed_blks = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000871 ip->i_size = ip->i_d.di_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
873 /*
874 * Mark the buffer containing the inode as something to keep
875 * around for a while. This helps to keep recently accessed
876 * meta-data in-core longer.
877 */
Dave Chinner821eb212010-12-02 16:31:13 +1100878 xfs_buf_set_ref(bp, XFS_INO_REF);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
880 /*
881 * Use xfs_trans_brelse() to release the buffer containing the
882 * on-disk inode, because it was acquired with xfs_trans_read_buf()
883 * in xfs_itobp() above. If tp is NULL, this is just a normal
884 * brelse(). If we're within a transaction, then xfs_trans_brelse()
885 * will only release the buffer if it is not dirty within the
886 * transaction. It will be OK to release the buffer in this case,
887 * because inodes on disk are never destroyed and we will be
888 * locking the new in-core inode before putting it in the hash
889 * table where other processes can find it. Thus we don't have
890 * to worry about the inode being changed just because we released
891 * the buffer.
892 */
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100893 out_brelse:
894 xfs_trans_brelse(tp, bp);
Christoph Hellwig9ed04512008-10-30 18:26:04 +1100895 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
897
898/*
899 * Read in extents from a btree-format inode.
900 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
901 */
902int
903xfs_iread_extents(
904 xfs_trans_t *tp,
905 xfs_inode_t *ip,
906 int whichfork)
907{
908 int error;
909 xfs_ifork_t *ifp;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100910 xfs_extnum_t nextents;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
913 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
914 ip->i_mount);
915 return XFS_ERROR(EFSCORRUPTED);
916 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100917 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 ifp = XFS_IFORK_PTR(ip, whichfork);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 /*
921 * We know that the size is valid (it's checked in iformat_btree)
922 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 ifp->if_lastex = NULLEXTNUM;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100924 ifp->if_bytes = ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 ifp->if_flags |= XFS_IFEXTENTS;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100926 xfs_iext_add(ifp, 0, nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 error = xfs_bmap_read_extents(tp, ip, whichfork);
928 if (error) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100929 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 ifp->if_flags &= ~XFS_IFEXTENTS;
931 return error;
932 }
Christoph Hellwiga6f64d42007-08-16 16:23:40 +1000933 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 return 0;
935}
936
937/*
938 * Allocate an inode on disk and return a copy of its in-core version.
939 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
940 * appropriately within the inode. The uid and gid for the inode are
941 * set according to the contents of the given cred structure.
942 *
943 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
944 * has a free inode available, call xfs_iget()
945 * to obtain the in-core version of the allocated inode. Finally,
946 * fill in the inode and log its initial contents. In this case,
947 * ialloc_context would be set to NULL and call_again set to false.
948 *
949 * If xfs_dialloc() does not have an available inode,
950 * it will replenish its supply by doing an allocation. Since we can
951 * only do one allocation within a transaction without deadlocks, we
952 * must commit the current transaction before returning the inode itself.
953 * In this case, therefore, we will set call_again to true and return.
954 * The caller should then commit the current transaction, start a new
955 * transaction, and call xfs_ialloc() again to actually get the inode.
956 *
957 * To ensure that some other process does not grab the inode that
958 * was allocated during the first call to xfs_ialloc(), this routine
959 * also returns the [locked] bp pointing to the head of the freelist
960 * as ialloc_context. The caller should hold this buffer across
961 * the commit and pass it back into this routine on the second call.
David Chinnerb11f94d2007-07-11 11:09:33 +1000962 *
963 * If we are allocating quota inodes, we do not have a parent inode
964 * to attach to or associate with (i.e. pip == NULL) because they
965 * are not linked into the directory structure - they are attached
966 * directly to the superblock - and so have no parent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 */
968int
969xfs_ialloc(
970 xfs_trans_t *tp,
971 xfs_inode_t *pip,
972 mode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -0700973 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 xfs_dev_t rdev,
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +0000975 prid_t prid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 int okalloc,
977 xfs_buf_t **ialloc_context,
978 boolean_t *call_again,
979 xfs_inode_t **ipp)
980{
981 xfs_ino_t ino;
982 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 uint flags;
984 int error;
Christoph Hellwigdff35fd2008-08-13 16:44:15 +1000985 timespec_t tv;
David Chinnerbf904242008-10-30 17:36:14 +1100986 int filestreams = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
988 /*
989 * Call the space management code to pick
990 * the on-disk inode to be allocated.
991 */
David Chinnerb11f94d2007-07-11 11:09:33 +1000992 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 ialloc_context, call_again, &ino);
David Chinnerbf904242008-10-30 17:36:14 +1100994 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 if (*call_again || ino == NULLFSINO) {
997 *ipp = NULL;
998 return 0;
999 }
1000 ASSERT(*ialloc_context == NULL);
1001
1002 /*
1003 * Get the in-core inode with the lock held exclusively.
1004 * This is because we're setting fields here we need
1005 * to prevent others from looking at until we're done.
1006 */
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001007 error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE,
1008 XFS_ILOCK_EXCL, &ip);
David Chinnerbf904242008-10-30 17:36:14 +11001009 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 ASSERT(ip != NULL);
1012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 ip->i_d.di_mode = (__uint16_t)mode;
1014 ip->i_d.di_onlink = 0;
1015 ip->i_d.di_nlink = nlink;
1016 ASSERT(ip->i_d.di_nlink == nlink);
David Howells9e2b2dc2008-08-13 16:20:04 +01001017 ip->i_d.di_uid = current_fsuid();
1018 ip->i_d.di_gid = current_fsgid();
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00001019 xfs_set_projid(ip, prid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1021
1022 /*
1023 * If the superblock version is up to where we support new format
1024 * inodes and this is currently an old format inode, then change
1025 * the inode version number now. This way we only do the conversion
1026 * here rather than here and in the flush/logging code.
1027 */
Eric Sandeen62118702008-03-06 13:44:28 +11001028 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11001029 ip->i_d.di_version == 1) {
1030 ip->i_d.di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 /*
1032 * We've already zeroed the old link count, the projid field,
1033 * and the pad field.
1034 */
1035 }
1036
1037 /*
1038 * Project ids won't be stored on disk if we are using a version 1 inode.
1039 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11001040 if ((prid != 0) && (ip->i_d.di_version == 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 xfs_bump_ino_vers2(tp, ip);
1042
Christoph Hellwigbd186aa2007-08-30 17:21:12 +10001043 if (pip && XFS_INHERIT_GID(pip)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 ip->i_d.di_gid = pip->i_d.di_gid;
1045 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1046 ip->i_d.di_mode |= S_ISGID;
1047 }
1048 }
1049
1050 /*
1051 * If the group ID of the new file does not match the effective group
1052 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1053 * (and only if the irix_sgid_inherit compatibility variable is set).
1054 */
1055 if ((irix_sgid_inherit) &&
1056 (ip->i_d.di_mode & S_ISGID) &&
1057 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1058 ip->i_d.di_mode &= ~S_ISGID;
1059 }
1060
1061 ip->i_d.di_size = 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001062 ip->i_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 ip->i_d.di_nextents = 0;
1064 ASSERT(ip->i_d.di_nblocks == 0);
Christoph Hellwigdff35fd2008-08-13 16:44:15 +10001065
1066 nanotime(&tv);
1067 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1068 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1069 ip->i_d.di_atime = ip->i_d.di_mtime;
1070 ip->i_d.di_ctime = ip->i_d.di_mtime;
1071
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 /*
1073 * di_gen will have been taken care of in xfs_iread.
1074 */
1075 ip->i_d.di_extsize = 0;
1076 ip->i_d.di_dmevmask = 0;
1077 ip->i_d.di_dmstate = 0;
1078 ip->i_d.di_flags = 0;
1079 flags = XFS_ILOG_CORE;
1080 switch (mode & S_IFMT) {
1081 case S_IFIFO:
1082 case S_IFCHR:
1083 case S_IFBLK:
1084 case S_IFSOCK:
1085 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1086 ip->i_df.if_u2.if_rdev = rdev;
1087 ip->i_df.if_flags = 0;
1088 flags |= XFS_ILOG_DEV;
1089 break;
1090 case S_IFREG:
David Chinnerbf904242008-10-30 17:36:14 +11001091 /*
1092 * we can't set up filestreams until after the VFS inode
1093 * is set up properly.
1094 */
1095 if (pip && xfs_inode_is_filestream(pip))
1096 filestreams = 1;
David Chinner2a82b8b2007-07-11 11:09:12 +10001097 /* fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 case S_IFDIR:
David Chinnerb11f94d2007-07-11 11:09:33 +10001099 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Nathan Scott365ca832005-06-21 15:39:12 +10001100 uint di_flags = 0;
1101
1102 if ((mode & S_IFMT) == S_IFDIR) {
1103 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1104 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001105 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1106 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1107 ip->i_d.di_extsize = pip->i_d.di_extsize;
1108 }
1109 } else if ((mode & S_IFMT) == S_IFREG) {
Christoph Hellwig613d7042007-10-11 17:44:08 +10001110 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
Nathan Scott365ca832005-06-21 15:39:12 +10001111 di_flags |= XFS_DIFLAG_REALTIME;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001112 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1113 di_flags |= XFS_DIFLAG_EXTSIZE;
1114 ip->i_d.di_extsize = pip->i_d.di_extsize;
1115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 }
1117 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1118 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +10001119 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1121 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +10001122 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1124 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +10001125 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1127 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +10001128 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1129 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1130 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +10001131 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1132 xfs_inherit_nodefrag)
1133 di_flags |= XFS_DIFLAG_NODEFRAG;
David Chinner2a82b8b2007-07-11 11:09:12 +10001134 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1135 di_flags |= XFS_DIFLAG_FILESTREAM;
Nathan Scott365ca832005-06-21 15:39:12 +10001136 ip->i_d.di_flags |= di_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 }
1138 /* FALLTHROUGH */
1139 case S_IFLNK:
1140 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1141 ip->i_df.if_flags = XFS_IFEXTENTS;
1142 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1143 ip->i_df.if_u1.if_extents = NULL;
1144 break;
1145 default:
1146 ASSERT(0);
1147 }
1148 /*
1149 * Attribute fork settings for new inode.
1150 */
1151 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1152 ip->i_d.di_anextents = 0;
1153
1154 /*
1155 * Log the new values stuffed into the inode.
1156 */
Christoph Hellwigec3ba852011-02-13 13:26:42 +00001157 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 xfs_trans_log_inode(tp, ip, flags);
1159
Nathan Scottb83bd132006-06-09 16:48:30 +10001160 /* now that we have an i_mode we can setup inode ops and unlock */
Christoph Hellwig41be8be2008-08-13 16:23:13 +10001161 xfs_setup_inode(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
David Chinnerbf904242008-10-30 17:36:14 +11001163 /* now we have set up the vfs inode we can associate the filestream */
1164 if (filestreams) {
1165 error = xfs_filestream_associate(pip, ip);
1166 if (error < 0)
1167 return -error;
1168 if (!error)
1169 xfs_iflags_set(ip, XFS_IFILESTREAM);
1170 }
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 *ipp = ip;
1173 return 0;
1174}
1175
1176/*
1177 * Check to make sure that there are no blocks allocated to the
1178 * file beyond the size of the file. We don't check this for
1179 * files with fixed size extents or real time extents, but we
1180 * at least do it for regular files.
1181 */
1182#ifdef DEBUG
1183void
1184xfs_isize_check(
1185 xfs_mount_t *mp,
1186 xfs_inode_t *ip,
1187 xfs_fsize_t isize)
1188{
1189 xfs_fileoff_t map_first;
1190 int nimaps;
1191 xfs_bmbt_irec_t imaps[2];
1192
1193 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1194 return;
1195
Eric Sandeen71ddabb2007-11-23 16:29:42 +11001196 if (XFS_IS_REALTIME_INODE(ip))
1197 return;
1198
1199 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 return;
1201
1202 nimaps = 2;
1203 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1204 /*
1205 * The filesystem could be shutting down, so bmapi may return
1206 * an error.
1207 */
1208 if (xfs_bmapi(NULL, ip, map_first,
1209 (XFS_B_TO_FSB(mp,
1210 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1211 map_first),
1212 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001213 NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 return;
1215 ASSERT(nimaps == 1);
1216 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1217}
1218#endif /* DEBUG */
1219
1220/*
1221 * Calculate the last possible buffered byte in a file. This must
1222 * include data that was buffered beyond the EOF by the write code.
1223 * This also needs to deal with overflowing the xfs_fsize_t type
1224 * which can happen for sizes near the limit.
1225 *
1226 * We also need to take into account any blocks beyond the EOF. It
1227 * may be the case that they were buffered by a write which failed.
1228 * In that case the pages will still be in memory, but the inode size
1229 * will never have been updated.
1230 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05001231STATIC xfs_fsize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232xfs_file_last_byte(
1233 xfs_inode_t *ip)
1234{
1235 xfs_mount_t *mp;
1236 xfs_fsize_t last_byte;
1237 xfs_fileoff_t last_block;
1238 xfs_fileoff_t size_last_block;
1239 int error;
1240
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001241 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
1243 mp = ip->i_mount;
1244 /*
1245 * Only check for blocks beyond the EOF if the extents have
1246 * been read in. This eliminates the need for the inode lock,
1247 * and it also saves us from looking when it really isn't
1248 * necessary.
1249 */
1250 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
Lachlan McIlroyf25181f2009-04-23 22:18:00 -04001251 xfs_ilock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1253 XFS_DATA_FORK);
Lachlan McIlroyf25181f2009-04-23 22:18:00 -04001254 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 if (error) {
1256 last_block = 0;
1257 }
1258 } else {
1259 last_block = 0;
1260 }
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001261 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1263
1264 last_byte = XFS_FSB_TO_B(mp, last_block);
1265 if (last_byte < 0) {
1266 return XFS_MAXIOFFSET(mp);
1267 }
1268 last_byte += (1 << mp->m_writeio_log);
1269 if (last_byte < 0) {
1270 return XFS_MAXIOFFSET(mp);
1271 }
1272 return last_byte;
1273}
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275/*
1276 * Start the truncation of the file to new_size. The new size
1277 * must be smaller than the current size. This routine will
1278 * clear the buffer and page caches of file data in the removed
1279 * range, and xfs_itruncate_finish() will remove the underlying
1280 * disk blocks.
1281 *
1282 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1283 * must NOT have the inode lock held at all. This is because we're
1284 * calling into the buffer/page cache code and we can't hold the
1285 * inode lock when we do so.
1286 *
David Chinner38e22992006-03-22 12:47:15 +11001287 * We need to wait for any direct I/Os in flight to complete before we
1288 * proceed with the truncate. This is needed to prevent the extents
1289 * being read or written by the direct I/Os from being removed while the
1290 * I/O is in flight as there is no other method of synchronising
1291 * direct I/O with the truncate operation. Also, because we hold
1292 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1293 * started until the truncate completes and drops the lock. Essentially,
Christoph Hellwig25e41b32008-12-03 12:20:39 +01001294 * the xfs_ioend_wait() call forms an I/O barrier that provides strict
1295 * ordering between direct I/Os and the truncate operation.
David Chinner38e22992006-03-22 12:47:15 +11001296 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1298 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1299 * in the case that the caller is locking things out of order and
1300 * may not be able to call xfs_itruncate_finish() with the inode lock
1301 * held without dropping the I/O lock. If the caller must drop the
1302 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1303 * must be called again with all the same restrictions as the initial
1304 * call.
1305 */
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001306int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307xfs_itruncate_start(
1308 xfs_inode_t *ip,
1309 uint flags,
1310 xfs_fsize_t new_size)
1311{
1312 xfs_fsize_t last_byte;
1313 xfs_off_t toss_start;
1314 xfs_mount_t *mp;
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001315 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001317 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001318 ASSERT((new_size == 0) || (new_size <= ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1320 (flags == XFS_ITRUNC_MAYBE));
1321
1322 mp = ip->i_mount;
Yingping Lu9fa80462006-03-22 12:44:35 +11001323
Lachlan McIlroyc734c792007-12-18 16:17:41 +11001324 /* wait for the completion of any pending DIOs */
Lachlan McIlroyd112f292008-10-30 16:59:06 +11001325 if (new_size == 0 || new_size < ip->i_size)
Christoph Hellwig25e41b32008-12-03 12:20:39 +01001326 xfs_ioend_wait(ip);
Lachlan McIlroyc734c792007-12-18 16:17:41 +11001327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 /*
Nathan Scott67fcaa72006-06-09 17:00:52 +10001329 * Call toss_pages or flushinval_pages to get rid of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 * overlapping the region being removed. We have to use
Nathan Scott67fcaa72006-06-09 17:00:52 +10001331 * the less efficient flushinval_pages in the case that the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 * caller may not be able to finish the truncate without
1333 * dropping the inode's I/O lock. Make sure
1334 * to catch any pages brought in by buffers overlapping
1335 * the EOF by searching out beyond the isize by our
1336 * block size. We round new_size up to a block boundary
1337 * so that we don't toss things on the same block as
1338 * new_size but before it.
1339 *
Nathan Scott67fcaa72006-06-09 17:00:52 +10001340 * Before calling toss_page or flushinval_pages, make sure to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 * call remapf() over the same region if the file is mapped.
1342 * This frees up mapped file references to the pages in the
Nathan Scott67fcaa72006-06-09 17:00:52 +10001343 * given range and for the flushinval_pages case it ensures
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 * that we get the latest mapped changes flushed out.
1345 */
1346 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1347 toss_start = XFS_FSB_TO_B(mp, toss_start);
1348 if (toss_start < 0) {
1349 /*
1350 * The place to start tossing is beyond our maximum
1351 * file size, so there is no way that the data extended
1352 * out there.
1353 */
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001354 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 }
1356 last_byte = xfs_file_last_byte(ip);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001357 trace_xfs_itruncate_start(ip, flags, new_size, toss_start, last_byte);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 if (last_byte > toss_start) {
1359 if (flags & XFS_ITRUNC_DEFINITE) {
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001360 xfs_tosspages(ip, toss_start,
1361 -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 } else {
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001363 error = xfs_flushinval_pages(ip, toss_start,
1364 -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 }
1366 }
1367
1368#ifdef DEBUG
1369 if (new_size == 0) {
Christoph Hellwigdf80c932008-08-13 16:22:09 +10001370 ASSERT(VN_CACHED(VFS_I(ip)) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 }
1372#endif
Lachlan McIlroyd3cf2092007-05-08 13:49:27 +10001373 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374}
1375
1376/*
David Chinnerf6485052008-04-17 16:50:04 +10001377 * Shrink the file to the given new_size. The new size must be smaller than
1378 * the current size. This will free up the underlying blocks in the removed
1379 * range after a call to xfs_itruncate_start() or xfs_atruncate_start().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 *
David Chinnerf6485052008-04-17 16:50:04 +10001381 * The transaction passed to this routine must have made a permanent log
1382 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1383 * given transaction and start new ones, so make sure everything involved in
1384 * the transaction is tidy before calling here. Some transaction will be
1385 * returned to the caller to be committed. The incoming transaction must
1386 * already include the inode, and both inode locks must be held exclusively.
1387 * The inode must also be "held" within the transaction. On return the inode
1388 * will be "held" within the returned transaction. This routine does NOT
1389 * require any disk space to be reserved for it within the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 *
David Chinnerf6485052008-04-17 16:50:04 +10001391 * The fork parameter must be either xfs_attr_fork or xfs_data_fork, and it
1392 * indicates the fork which is to be truncated. For the attribute fork we only
1393 * support truncation to size 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 *
David Chinnerf6485052008-04-17 16:50:04 +10001395 * We use the sync parameter to indicate whether or not the first transaction
1396 * we perform might have to be synchronous. For the attr fork, it needs to be
1397 * so if the unlink of the inode is not yet known to be permanent in the log.
1398 * This keeps us from freeing and reusing the blocks of the attribute fork
1399 * before the unlink of the inode becomes permanent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 *
David Chinnerf6485052008-04-17 16:50:04 +10001401 * For the data fork, we normally have to run synchronously if we're being
1402 * called out of the inactive path or we're being called out of the create path
1403 * where we're truncating an existing file. Either way, the truncate needs to
1404 * be sync so blocks don't reappear in the file with altered data in case of a
1405 * crash. wsync filesystems can run the first case async because anything that
1406 * shrinks the inode has to run sync so by the time we're called here from
1407 * inactive, the inode size is permanently set to 0.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 *
David Chinnerf6485052008-04-17 16:50:04 +10001409 * Calls from the truncate path always need to be sync unless we're in a wsync
1410 * filesystem and the file has already been unlinked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 *
David Chinnerf6485052008-04-17 16:50:04 +10001412 * The caller is responsible for correctly setting the sync parameter. It gets
1413 * too hard for us to guess here which path we're being called out of just
1414 * based on inode state.
1415 *
1416 * If we get an error, we must return with the inode locked and linked into the
1417 * current transaction. This keeps things simple for the higher level code,
1418 * because it always knows that the inode is locked and held in the transaction
1419 * that returns to it whether errors occur or not. We don't mark the inode
1420 * dirty on error so that transactions can be easily aborted if possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 */
1422int
1423xfs_itruncate_finish(
1424 xfs_trans_t **tp,
1425 xfs_inode_t *ip,
1426 xfs_fsize_t new_size,
1427 int fork,
1428 int sync)
1429{
1430 xfs_fsblock_t first_block;
1431 xfs_fileoff_t first_unmap_block;
1432 xfs_fileoff_t last_block;
1433 xfs_filblks_t unmap_len=0;
1434 xfs_mount_t *mp;
1435 xfs_trans_t *ntp;
1436 int done;
1437 int committed;
1438 xfs_bmap_free_t free_list;
1439 int error;
1440
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10001441 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001442 ASSERT((new_size == 0) || (new_size <= ip->i_size));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 ASSERT(*tp != NULL);
1444 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1445 ASSERT(ip->i_transp == *tp);
1446 ASSERT(ip->i_itemp != NULL);
Christoph Hellwig898621d2010-06-24 11:36:58 +10001447 ASSERT(ip->i_itemp->ili_lock_flags == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
1449
1450 ntp = *tp;
1451 mp = (ntp)->t_mountp;
1452 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1453
1454 /*
1455 * We only support truncating the entire attribute fork.
1456 */
1457 if (fork == XFS_ATTR_FORK) {
1458 new_size = 0LL;
1459 }
1460 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001461 trace_xfs_itruncate_finish_start(ip, new_size);
1462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 /*
1464 * The first thing we do is set the size to new_size permanently
1465 * on disk. This way we don't have to worry about anyone ever
1466 * being able to look at the data being freed even in the face
1467 * of a crash. What we're getting around here is the case where
1468 * we free a block, it is allocated to another file, it is written
1469 * to, and then we crash. If the new data gets written to the
1470 * file but the log buffers containing the free and reallocation
1471 * don't, then we'd end up with garbage in the blocks being freed.
1472 * As long as we make the new_size permanent before actually
1473 * freeing any blocks it doesn't matter if they get writtten to.
1474 *
1475 * The callers must signal into us whether or not the size
1476 * setting here must be synchronous. There are a few cases
1477 * where it doesn't have to be synchronous. Those cases
1478 * occur if the file is unlinked and we know the unlink is
1479 * permanent or if the blocks being truncated are guaranteed
1480 * to be beyond the inode eof (regardless of the link count)
1481 * and the eof value is permanent. Both of these cases occur
1482 * only on wsync-mounted filesystems. In those cases, we're
1483 * guaranteed that no user will ever see the data in the blocks
1484 * that are being truncated so the truncate can run async.
1485 * In the free beyond eof case, the file may wind up with
1486 * more blocks allocated to it than it needs if we crash
1487 * and that won't get fixed until the next time the file
1488 * is re-opened and closed but that's ok as that shouldn't
1489 * be too many blocks.
1490 *
1491 * However, we can't just make all wsync xactions run async
1492 * because there's one call out of the create path that needs
1493 * to run sync where it's truncating an existing file to size
1494 * 0 whose size is > 0.
1495 *
1496 * It's probably possible to come up with a test in this
1497 * routine that would correctly distinguish all the above
1498 * cases from the values of the function parameters and the
1499 * inode state but for sanity's sake, I've decided to let the
1500 * layers above just tell us. It's simpler to correctly figure
1501 * out in the layer above exactly under what conditions we
1502 * can run async and I think it's easier for others read and
1503 * follow the logic in case something has to be changed.
1504 * cscope is your friend -- rcc.
1505 *
1506 * The attribute fork is much simpler.
1507 *
1508 * For the attribute fork we allow the caller to tell us whether
1509 * the unlink of the inode that led to this call is yet permanent
1510 * in the on disk log. If it is not and we will be freeing extents
1511 * in this inode then we make the first transaction synchronous
1512 * to make sure that the unlink is permanent by the time we free
1513 * the blocks.
1514 */
1515 if (fork == XFS_DATA_FORK) {
1516 if (ip->i_d.di_nextents > 0) {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001517 /*
1518 * If we are not changing the file size then do
1519 * not update the on-disk file size - we may be
1520 * called from xfs_inactive_free_eofblocks(). If we
1521 * update the on-disk file size and then the system
1522 * crashes before the contents of the file are
1523 * flushed to disk then the files may be full of
1524 * holes (ie NULL files bug).
1525 */
1526 if (ip->i_size != new_size) {
1527 ip->i_d.di_size = new_size;
1528 ip->i_size = new_size;
1529 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1530 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 }
1532 } else if (sync) {
1533 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1534 if (ip->i_d.di_anextents > 0)
1535 xfs_trans_set_sync(ntp);
1536 }
1537 ASSERT(fork == XFS_DATA_FORK ||
1538 (fork == XFS_ATTR_FORK &&
1539 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1540 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1541
1542 /*
1543 * Since it is possible for space to become allocated beyond
1544 * the end of the file (in a crash where the space is allocated
1545 * but the inode size is not yet updated), simply remove any
1546 * blocks which show up between the new EOF and the maximum
1547 * possible file size. If the first block to be removed is
1548 * beyond the maximum file size (ie it is the same as last_block),
1549 * then there is nothing to do.
1550 */
1551 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1552 ASSERT(first_unmap_block <= last_block);
1553 done = 0;
1554 if (last_block == first_unmap_block) {
1555 done = 1;
1556 } else {
1557 unmap_len = last_block - first_unmap_block + 1;
1558 }
1559 while (!done) {
1560 /*
1561 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1562 * will tell us whether it freed the entire range or
1563 * not. If this is a synchronous mount (wsync),
1564 * then we can tell bunmapi to keep all the
1565 * transactions asynchronous since the unlink
1566 * transaction that made this inode inactive has
1567 * already hit the disk. There's no danger of
1568 * the freed blocks being reused, there being a
1569 * crash, and the reused blocks suddenly reappearing
1570 * in this file with garbage in them once recovery
1571 * runs.
1572 */
Eric Sandeen9d87c312009-01-14 23:22:07 -06001573 xfs_bmap_init(&free_list, &first_block);
Lachlan McIlroy541d7d32007-10-11 17:34:33 +10001574 error = xfs_bunmapi(ntp, ip,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001575 first_unmap_block, unmap_len,
Christoph Hellwigcd8b0bb2010-06-23 18:11:15 +10001576 xfs_bmapi_aflag(fork),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001578 &first_block, &free_list,
Christoph Hellwigb4e91812010-06-23 18:11:15 +10001579 &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 if (error) {
1581 /*
1582 * If the bunmapi call encounters an error,
1583 * return to the caller where the transaction
1584 * can be properly aborted. We just need to
1585 * make sure we're not holding any resources
1586 * that we were not when we came in.
1587 */
1588 xfs_bmap_cancel(&free_list);
1589 return error;
1590 }
1591
1592 /*
1593 * Duplicate the transaction that has the permanent
1594 * reservation and commit the old transaction.
1595 */
Eric Sandeenf7c99b62007-02-10 18:37:16 +11001596 error = xfs_bmap_finish(tp, &free_list, &committed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 ntp = *tp;
Christoph Hellwig898621d2010-06-24 11:36:58 +10001598 if (committed)
1599 xfs_trans_ijoin(ntp, ip);
David Chinnerf6485052008-04-17 16:50:04 +10001600
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 if (error) {
1602 /*
David Chinnerf6485052008-04-17 16:50:04 +10001603 * If the bmap finish call encounters an error, return
1604 * to the caller where the transaction can be properly
1605 * aborted. We just need to make sure we're not
1606 * holding any resources that we were not when we came
1607 * in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 *
David Chinnerf6485052008-04-17 16:50:04 +10001609 * Aborting from this point might lose some blocks in
1610 * the file system, but oh well.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 */
1612 xfs_bmap_cancel(&free_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 return error;
1614 }
1615
1616 if (committed) {
1617 /*
David Chinnerf6485052008-04-17 16:50:04 +10001618 * Mark the inode dirty so it will be logged and
David Chinnere5720ee2008-04-10 12:21:18 +10001619 * moved forward in the log as part of every commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1622 }
David Chinnerf6485052008-04-17 16:50:04 +10001623
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 ntp = xfs_trans_dup(ntp);
David Chinnere5720ee2008-04-10 12:21:18 +10001625 error = xfs_trans_commit(*tp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 *tp = ntp;
David Chinnere5720ee2008-04-10 12:21:18 +10001627
Christoph Hellwig898621d2010-06-24 11:36:58 +10001628 xfs_trans_ijoin(ntp, ip);
David Chinnerf6485052008-04-17 16:50:04 +10001629
Dave Chinnercc09c0d2008-11-17 17:37:10 +11001630 if (error)
1631 return error;
1632 /*
1633 * transaction commit worked ok so we can drop the extra ticket
1634 * reference that we gained in xfs_trans_dup()
1635 */
1636 xfs_log_ticket_put(ntp->t_ticket);
1637 error = xfs_trans_reserve(ntp, 0,
David Chinnerf6485052008-04-17 16:50:04 +10001638 XFS_ITRUNCATE_LOG_RES(mp), 0,
1639 XFS_TRANS_PERM_LOG_RES,
1640 XFS_ITRUNCATE_LOG_COUNT);
1641 if (error)
1642 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 }
1644 /*
1645 * Only update the size in the case of the data fork, but
1646 * always re-log the inode so that our permanent transaction
1647 * can keep on rolling it forward in the log.
1648 */
1649 if (fork == XFS_DATA_FORK) {
1650 xfs_isize_check(mp, ip, new_size);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001651 /*
1652 * If we are not changing the file size then do
1653 * not update the on-disk file size - we may be
1654 * called from xfs_inactive_free_eofblocks(). If we
1655 * update the on-disk file size and then the system
1656 * crashes before the contents of the file are
1657 * flushed to disk then the files may be full of
1658 * holes (ie NULL files bug).
1659 */
1660 if (ip->i_size != new_size) {
1661 ip->i_d.di_size = new_size;
1662 ip->i_size = new_size;
1663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 }
1665 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1666 ASSERT((new_size != 0) ||
1667 (fork == XFS_ATTR_FORK) ||
1668 (ip->i_delayed_blks == 0));
1669 ASSERT((new_size != 0) ||
1670 (fork == XFS_ATTR_FORK) ||
1671 (ip->i_d.di_nextents == 0));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001672 trace_xfs_itruncate_finish_end(ip, new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 return 0;
1674}
1675
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676/*
1677 * This is called when the inode's link count goes to 0.
1678 * We place the on-disk inode on a list in the AGI. It
1679 * will be pulled from this list when the inode is freed.
1680 */
1681int
1682xfs_iunlink(
1683 xfs_trans_t *tp,
1684 xfs_inode_t *ip)
1685{
1686 xfs_mount_t *mp;
1687 xfs_agi_t *agi;
1688 xfs_dinode_t *dip;
1689 xfs_buf_t *agibp;
1690 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 xfs_agino_t agino;
1692 short bucket_index;
1693 int offset;
1694 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
1696 ASSERT(ip->i_d.di_nlink == 0);
1697 ASSERT(ip->i_d.di_mode != 0);
1698 ASSERT(ip->i_transp == tp);
1699
1700 mp = tp->t_mountp;
1701
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 /*
1703 * Get the agi buffer first. It ensures lock ordering
1704 * on the list.
1705 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001706 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
Vlad Apostolov859d7182007-10-11 17:44:18 +10001707 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 /*
1712 * Get the index into the agi hash table for the
1713 * list this inode will go on.
1714 */
1715 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1716 ASSERT(agino != 0);
1717 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1718 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001719 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
Christoph Hellwig16259e72005-11-02 15:11:25 +11001721 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 /*
1723 * There is already another inode in the bucket we need
1724 * to add ourselves to. Add us at the front of the list.
1725 * Here we put the head pointer into our next pointer,
1726 * and then we fall through to point the head at us.
1727 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001728 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
Vlad Apostolovc319b582007-11-23 16:27:51 +11001729 if (error)
1730 return error;
1731
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001732 ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 /* both on-disk, don't endian flip twice */
1734 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001735 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 offsetof(xfs_dinode_t, di_next_unlinked);
1737 xfs_trans_inode_buf(tp, ibp);
1738 xfs_trans_log_buf(tp, ibp, offset,
1739 (offset + sizeof(xfs_agino_t) - 1));
1740 xfs_inobp_check(mp, ibp);
1741 }
1742
1743 /*
1744 * Point the bucket head pointer at the inode being inserted.
1745 */
1746 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001747 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 offset = offsetof(xfs_agi_t, agi_unlinked) +
1749 (sizeof(xfs_agino_t) * bucket_index);
1750 xfs_trans_log_buf(tp, agibp, offset,
1751 (offset + sizeof(xfs_agino_t) - 1));
1752 return 0;
1753}
1754
1755/*
1756 * Pull the on-disk inode from the AGI unlinked list.
1757 */
1758STATIC int
1759xfs_iunlink_remove(
1760 xfs_trans_t *tp,
1761 xfs_inode_t *ip)
1762{
1763 xfs_ino_t next_ino;
1764 xfs_mount_t *mp;
1765 xfs_agi_t *agi;
1766 xfs_dinode_t *dip;
1767 xfs_buf_t *agibp;
1768 xfs_buf_t *ibp;
1769 xfs_agnumber_t agno;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 xfs_agino_t agino;
1771 xfs_agino_t next_agino;
1772 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001773 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001775 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 mp = tp->t_mountp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
1781 /*
1782 * Get the agi buffer first. It ensures lock ordering
1783 * on the list.
1784 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001785 error = xfs_read_agi(mp, tp, agno, &agibp);
1786 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 return error;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001788
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11001790
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 /*
1792 * Get the index into the agi hash table for the
1793 * list this inode will go on.
1794 */
1795 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1796 ASSERT(agino != 0);
1797 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001798 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 ASSERT(agi->agi_unlinked[bucket_index]);
1800
Christoph Hellwig16259e72005-11-02 15:11:25 +11001801 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 /*
1803 * We're at the head of the list. Get the inode's
1804 * on-disk buffer to see if there is anyone after us
1805 * on the list. Only modify our next pointer if it
1806 * is not already NULLAGINO. This saves us the overhead
1807 * of dealing with the buffer when there is no need to
1808 * change it.
1809 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001810 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001812 xfs_warn(mp, "%s: xfs_itobp() returned error %d.",
1813 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 return error;
1815 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001816 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 ASSERT(next_agino != 0);
1818 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001819 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001820 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 offsetof(xfs_dinode_t, di_next_unlinked);
1822 xfs_trans_inode_buf(tp, ibp);
1823 xfs_trans_log_buf(tp, ibp, offset,
1824 (offset + sizeof(xfs_agino_t) - 1));
1825 xfs_inobp_check(mp, ibp);
1826 } else {
1827 xfs_trans_brelse(tp, ibp);
1828 }
1829 /*
1830 * Point the bucket head pointer at the next inode.
1831 */
1832 ASSERT(next_agino != 0);
1833 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001834 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 offset = offsetof(xfs_agi_t, agi_unlinked) +
1836 (sizeof(xfs_agino_t) * bucket_index);
1837 xfs_trans_log_buf(tp, agibp, offset,
1838 (offset + sizeof(xfs_agino_t) - 1));
1839 } else {
1840 /*
1841 * We need to search the list for the inode being freed.
1842 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11001843 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 last_ibp = NULL;
1845 while (next_agino != agino) {
1846 /*
1847 * If the last inode wasn't the one pointing to
1848 * us, then release its buffer since we're not
1849 * going to do anything with it.
1850 */
1851 if (last_ibp != NULL) {
1852 xfs_trans_brelse(tp, last_ibp);
1853 }
1854 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1855 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
Christoph Hellwigc679eef2008-10-30 18:04:13 +11001856 &last_ibp, &last_offset, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001858 xfs_warn(mp,
1859 "%s: xfs_inotobp() returned error %d.",
1860 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 return error;
1862 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001863 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 ASSERT(next_agino != NULLAGINO);
1865 ASSERT(next_agino != 0);
1866 }
1867 /*
1868 * Now last_ibp points to the buffer previous to us on
1869 * the unlinked list. Pull us from the list.
1870 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001871 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 if (error) {
Dave Chinner0b932cc2011-03-07 10:08:35 +11001873 xfs_warn(mp, "%s: xfs_itobp(2) returned error %d.",
1874 __func__, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 return error;
1876 }
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001877 next_agino = be32_to_cpu(dip->di_next_unlinked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 ASSERT(next_agino != 0);
1879 ASSERT(next_agino != agino);
1880 if (next_agino != NULLAGINO) {
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001881 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11001882 offset = ip->i_imap.im_boffset +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 offsetof(xfs_dinode_t, di_next_unlinked);
1884 xfs_trans_inode_buf(tp, ibp);
1885 xfs_trans_log_buf(tp, ibp, offset,
1886 (offset + sizeof(xfs_agino_t) - 1));
1887 xfs_inobp_check(mp, ibp);
1888 } else {
1889 xfs_trans_brelse(tp, ibp);
1890 }
1891 /*
1892 * Point the previous inode on the list to the next inode.
1893 */
Christoph Hellwig347d1c02007-08-28 13:57:51 +10001894 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 ASSERT(next_agino != 0);
1896 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1897 xfs_trans_inode_buf(tp, last_ibp);
1898 xfs_trans_log_buf(tp, last_ibp, offset,
1899 (offset + sizeof(xfs_agino_t) - 1));
1900 xfs_inobp_check(mp, last_ibp);
1901 }
1902 return 0;
1903}
1904
Dave Chinner5b3eed72010-08-24 11:42:41 +10001905/*
1906 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1907 * inodes that are in memory - they all must be marked stale and attached to
1908 * the cluster buffer.
1909 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001910STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911xfs_ifree_cluster(
1912 xfs_inode_t *free_ip,
1913 xfs_trans_t *tp,
1914 xfs_ino_t inum)
1915{
1916 xfs_mount_t *mp = free_ip->i_mount;
1917 int blks_per_cluster;
1918 int nbufs;
1919 int ninodes;
Dave Chinner5b257b42010-06-03 16:22:29 +10001920 int i, j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 xfs_daddr_t blkno;
1922 xfs_buf_t *bp;
Dave Chinner5b257b42010-06-03 16:22:29 +10001923 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 xfs_inode_log_item_t *iip;
1925 xfs_log_item_t *lip;
Dave Chinner5017e972010-01-11 11:47:40 +00001926 struct xfs_perag *pag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
Dave Chinner5017e972010-01-11 11:47:40 +00001928 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1930 blks_per_cluster = 1;
1931 ninodes = mp->m_sb.sb_inopblock;
1932 nbufs = XFS_IALLOC_BLOCKS(mp);
1933 } else {
1934 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1935 mp->m_sb.sb_blocksize;
1936 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1937 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1938 }
1939
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 for (j = 0; j < nbufs; j++, inum += ninodes) {
1941 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1942 XFS_INO_TO_AGBNO(mp, inum));
1943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 /*
Dave Chinner5b257b42010-06-03 16:22:29 +10001945 * We obtain and lock the backing buffer first in the process
1946 * here, as we have to ensure that any dirty inode that we
1947 * can't get the flush lock on is attached to the buffer.
1948 * If we scan the in-memory inodes first, then buffer IO can
1949 * complete before we get a lock on it, and hence we may fail
1950 * to mark all the active inodes on the buffer stale.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 */
Dave Chinner5b257b42010-06-03 16:22:29 +10001952 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 mp->m_bsize * blks_per_cluster,
Christoph Hellwig0cadda12010-01-19 09:56:44 +00001954 XBF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
Dave Chinner5b257b42010-06-03 16:22:29 +10001956 /*
1957 * Walk the inodes already attached to the buffer and mark them
1958 * stale. These will all have the flush locks held, so an
Dave Chinner5b3eed72010-08-24 11:42:41 +10001959 * in-memory inode walk can't lock them. By marking them all
1960 * stale first, we will not attempt to lock them in the loop
1961 * below as the XFS_ISTALE flag will be set.
Dave Chinner5b257b42010-06-03 16:22:29 +10001962 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1964 while (lip) {
1965 if (lip->li_type == XFS_LI_INODE) {
1966 iip = (xfs_inode_log_item_t *)lip;
1967 ASSERT(iip->ili_logged == 1);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001968 lip->li_cb = xfs_istale_done;
David Chinner7b2e2a32008-10-30 17:39:12 +11001969 xfs_trans_ail_copy_lsn(mp->m_ail,
1970 &iip->ili_flush_lsn,
1971 &iip->ili_item.li_lsn);
David Chinnere5ffd2b2006-11-21 18:55:33 +11001972 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 }
1974 lip = lip->li_bio_list;
1975 }
1976
Dave Chinner5b3eed72010-08-24 11:42:41 +10001977
Dave Chinner5b257b42010-06-03 16:22:29 +10001978 /*
1979 * For each inode in memory attempt to add it to the inode
1980 * buffer and set it up for being staled on buffer IO
1981 * completion. This is safe as we've locked out tail pushing
1982 * and flushing by locking the buffer.
1983 *
1984 * We have already marked every inode that was part of a
1985 * transaction stale above, which means there is no point in
1986 * even trying to lock them.
1987 */
1988 for (i = 0; i < ninodes; i++) {
Dave Chinner5b3eed72010-08-24 11:42:41 +10001989retry:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001990 rcu_read_lock();
Dave Chinner5b257b42010-06-03 16:22:29 +10001991 ip = radix_tree_lookup(&pag->pag_ici_root,
1992 XFS_INO_TO_AGINO(mp, (inum + i)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
Dave Chinner1a3e8f32010-12-17 17:29:43 +11001994 /* Inode not in memory, nothing to do */
1995 if (!ip) {
1996 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10001997 continue;
1998 }
1999
Dave Chinner5b3eed72010-08-24 11:42:41 +10002000 /*
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002001 * because this is an RCU protected lookup, we could
2002 * find a recently freed or even reallocated inode
2003 * during the lookup. We need to check under the
2004 * i_flags_lock for a valid inode here. Skip it if it
2005 * is not valid, the wrong inode or stale.
2006 */
2007 spin_lock(&ip->i_flags_lock);
2008 if (ip->i_ino != inum + i ||
2009 __xfs_iflags_test(ip, XFS_ISTALE)) {
2010 spin_unlock(&ip->i_flags_lock);
2011 rcu_read_unlock();
2012 continue;
2013 }
2014 spin_unlock(&ip->i_flags_lock);
2015
2016 /*
Dave Chinner5b3eed72010-08-24 11:42:41 +10002017 * Don't try to lock/unlock the current inode, but we
2018 * _cannot_ skip the other inodes that we did not find
2019 * in the list attached to the buffer and are not
2020 * already marked stale. If we can't lock it, back off
2021 * and retry.
2022 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002023 if (ip != free_ip &&
2024 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002025 rcu_read_unlock();
Dave Chinner5b3eed72010-08-24 11:42:41 +10002026 delay(1);
2027 goto retry;
Dave Chinner5b257b42010-06-03 16:22:29 +10002028 }
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002029 rcu_read_unlock();
Dave Chinner5b257b42010-06-03 16:22:29 +10002030
Dave Chinner5b3eed72010-08-24 11:42:41 +10002031 xfs_iflock(ip);
Dave Chinner5b257b42010-06-03 16:22:29 +10002032 xfs_iflags_set(ip, XFS_ISTALE);
Dave Chinner5b257b42010-06-03 16:22:29 +10002033
Dave Chinner5b3eed72010-08-24 11:42:41 +10002034 /*
2035 * we don't need to attach clean inodes or those only
2036 * with unlogged changes (which we throw away, anyway).
2037 */
Dave Chinner5b257b42010-06-03 16:22:29 +10002038 iip = ip->i_itemp;
Dave Chinner5b3eed72010-08-24 11:42:41 +10002039 if (!iip || xfs_inode_clean(ip)) {
Dave Chinner5b257b42010-06-03 16:22:29 +10002040 ASSERT(ip != free_ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 ip->i_update_core = 0;
2042 xfs_ifunlock(ip);
2043 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2044 continue;
2045 }
2046
2047 iip->ili_last_fields = iip->ili_format.ilf_fields;
2048 iip->ili_format.ilf_fields = 0;
2049 iip->ili_logged = 1;
David Chinner7b2e2a32008-10-30 17:39:12 +11002050 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2051 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10002053 xfs_buf_attach_iodone(bp, xfs_istale_done,
2054 &iip->ili_item);
Dave Chinner5b257b42010-06-03 16:22:29 +10002055
2056 if (ip != free_ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 }
2059
Dave Chinner5b3eed72010-08-24 11:42:41 +10002060 xfs_trans_stale_inode_buf(tp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 xfs_trans_binval(tp, bp);
2062 }
2063
Dave Chinner5017e972010-01-11 11:47:40 +00002064 xfs_perag_put(pag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065}
2066
2067/*
2068 * This is called to return an inode to the inode free list.
2069 * The inode should already be truncated to 0 length and have
2070 * no pages associated with it. This routine also assumes that
2071 * the inode is already a part of the transaction.
2072 *
2073 * The on-disk copy of the inode will have been added to the list
2074 * of unlinked inodes in the AGI. We need to remove the inode from
2075 * that list atomically with respect to freeing it here.
2076 */
2077int
2078xfs_ifree(
2079 xfs_trans_t *tp,
2080 xfs_inode_t *ip,
2081 xfs_bmap_free_t *flist)
2082{
2083 int error;
2084 int delete;
2085 xfs_ino_t first_ino;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002086 xfs_dinode_t *dip;
2087 xfs_buf_t *ibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002089 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 ASSERT(ip->i_transp == tp);
2091 ASSERT(ip->i_d.di_nlink == 0);
2092 ASSERT(ip->i_d.di_nextents == 0);
2093 ASSERT(ip->i_d.di_anextents == 0);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10002094 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2096 ASSERT(ip->i_d.di_nblocks == 0);
2097
2098 /*
2099 * Pull the on-disk inode from the AGI unlinked list.
2100 */
2101 error = xfs_iunlink_remove(tp, ip);
2102 if (error != 0) {
2103 return error;
2104 }
2105
2106 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2107 if (error != 0) {
2108 return error;
2109 }
2110 ip->i_d.di_mode = 0; /* mark incore inode as free */
2111 ip->i_d.di_flags = 0;
2112 ip->i_d.di_dmevmask = 0;
2113 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2114 ip->i_df.if_ext_max =
2115 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2116 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2117 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2118 /*
2119 * Bump the generation count so no one will be confused
2120 * by reincarnations of this inode.
2121 */
2122 ip->i_d.di_gen++;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002123
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2125
Christoph Hellwig0cadda12010-01-19 09:56:44 +00002126 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XBF_LOCK);
Vlad Apostolovc319b582007-11-23 16:27:51 +11002127 if (error)
2128 return error;
2129
2130 /*
2131 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
2132 * from picking up this inode when it is reclaimed (its incore state
2133 * initialzed but not flushed to disk yet). The in-core di_mode is
2134 * already cleared and a corresponding transaction logged.
2135 * The hack here just synchronizes the in-core to on-disk
2136 * di_mode value in advance before the actual inode sync to disk.
2137 * This is OK because the inode is already unlinked and would never
2138 * change its di_mode again for this inode generation.
2139 * This is a temporary hack that would require a proper fix
2140 * in the future.
2141 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002142 dip->di_mode = 0;
Vlad Apostolovc319b582007-11-23 16:27:51 +11002143
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (delete) {
2145 xfs_ifree_cluster(ip, tp, first_ino);
2146 }
2147
2148 return 0;
2149}
2150
2151/*
2152 * Reallocate the space for if_broot based on the number of records
2153 * being added or deleted as indicated in rec_diff. Move the records
2154 * and pointers in if_broot to fit the new size. When shrinking this
2155 * will eliminate holes between the records and pointers created by
2156 * the caller. When growing this will create holes to be filled in
2157 * by the caller.
2158 *
2159 * The caller must not request to add more records than would fit in
2160 * the on-disk inode root. If the if_broot is currently NULL, then
2161 * if we adding records one will be allocated. The caller must also
2162 * not request that the number of records go below zero, although
2163 * it can go to zero.
2164 *
2165 * ip -- the inode whose if_broot area is changing
2166 * ext_diff -- the change in the number of records, positive or negative,
2167 * requested for the if_broot array.
2168 */
2169void
2170xfs_iroot_realloc(
2171 xfs_inode_t *ip,
2172 int rec_diff,
2173 int whichfork)
2174{
Christoph Hellwig60197e82008-10-30 17:11:19 +11002175 struct xfs_mount *mp = ip->i_mount;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 int cur_max;
2177 xfs_ifork_t *ifp;
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002178 struct xfs_btree_block *new_broot;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 int new_max;
2180 size_t new_size;
2181 char *np;
2182 char *op;
2183
2184 /*
2185 * Handle the degenerate case quietly.
2186 */
2187 if (rec_diff == 0) {
2188 return;
2189 }
2190
2191 ifp = XFS_IFORK_PTR(ip, whichfork);
2192 if (rec_diff > 0) {
2193 /*
2194 * If there wasn't any memory allocated before, just
2195 * allocate it now and get out.
2196 */
2197 if (ifp->if_broot_bytes == 0) {
2198 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002199 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 ifp->if_broot_bytes = (int)new_size;
2201 return;
2202 }
2203
2204 /*
2205 * If there is already an existing if_broot, then we need
2206 * to realloc() it and shift the pointers to their new
2207 * location. The records don't change location because
2208 * they are kept butted up against the btree block header.
2209 */
Christoph Hellwig60197e82008-10-30 17:11:19 +11002210 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 new_max = cur_max + rec_diff;
2212 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002213 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002215 KM_SLEEP | KM_NOFS);
Christoph Hellwig60197e82008-10-30 17:11:19 +11002216 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2217 ifp->if_broot_bytes);
2218 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2219 (int)new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 ifp->if_broot_bytes = (int)new_size;
2221 ASSERT(ifp->if_broot_bytes <=
2222 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2223 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2224 return;
2225 }
2226
2227 /*
2228 * rec_diff is less than 0. In this case, we are shrinking the
2229 * if_broot buffer. It must already exist. If we go to zero
2230 * records, just get rid of the root and clear the status bit.
2231 */
2232 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
Christoph Hellwig60197e82008-10-30 17:11:19 +11002233 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 new_max = cur_max + rec_diff;
2235 ASSERT(new_max >= 0);
2236 if (new_max > 0)
2237 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2238 else
2239 new_size = 0;
2240 if (new_size > 0) {
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002241 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 /*
2243 * First copy over the btree block header.
2244 */
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002245 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 } else {
2247 new_broot = NULL;
2248 ifp->if_flags &= ~XFS_IFBROOT;
2249 }
2250
2251 /*
2252 * Only copy the records and pointers if there are any.
2253 */
2254 if (new_max > 0) {
2255 /*
2256 * First copy the records.
2257 */
Christoph Hellwig136341b2008-10-30 17:11:40 +11002258 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
2259 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2261
2262 /*
2263 * Then copy the pointers.
2264 */
Christoph Hellwig60197e82008-10-30 17:11:19 +11002265 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 ifp->if_broot_bytes);
Christoph Hellwig60197e82008-10-30 17:11:19 +11002267 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 (int)new_size);
2269 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2270 }
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002271 kmem_free(ifp->if_broot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 ifp->if_broot = new_broot;
2273 ifp->if_broot_bytes = (int)new_size;
2274 ASSERT(ifp->if_broot_bytes <=
2275 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2276 return;
2277}
2278
2279
2280/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 * This is called when the amount of space needed for if_data
2282 * is increased or decreased. The change in size is indicated by
2283 * the number of bytes that need to be added or deleted in the
2284 * byte_diff parameter.
2285 *
2286 * If the amount of space needed has decreased below the size of the
2287 * inline buffer, then switch to using the inline buffer. Otherwise,
2288 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2289 * to what is needed.
2290 *
2291 * ip -- the inode whose if_data area is changing
2292 * byte_diff -- the change in the number of bytes, positive or negative,
2293 * requested for the if_data array.
2294 */
2295void
2296xfs_idata_realloc(
2297 xfs_inode_t *ip,
2298 int byte_diff,
2299 int whichfork)
2300{
2301 xfs_ifork_t *ifp;
2302 int new_size;
2303 int real_size;
2304
2305 if (byte_diff == 0) {
2306 return;
2307 }
2308
2309 ifp = XFS_IFORK_PTR(ip, whichfork);
2310 new_size = (int)ifp->if_bytes + byte_diff;
2311 ASSERT(new_size >= 0);
2312
2313 if (new_size == 0) {
2314 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002315 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 }
2317 ifp->if_u1.if_data = NULL;
2318 real_size = 0;
2319 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2320 /*
2321 * If the valid extents/data can fit in if_inline_ext/data,
2322 * copy them from the malloc'd vector and free it.
2323 */
2324 if (ifp->if_u1.if_data == NULL) {
2325 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2326 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2327 ASSERT(ifp->if_real_bytes != 0);
2328 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2329 new_size);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002330 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2332 }
2333 real_size = 0;
2334 } else {
2335 /*
2336 * Stuck with malloc/realloc.
2337 * For inline data, the underlying buffer must be
2338 * a multiple of 4 bytes in size so that it can be
2339 * logged and stay on word boundaries. We enforce
2340 * that here.
2341 */
2342 real_size = roundup(new_size, 4);
2343 if (ifp->if_u1.if_data == NULL) {
2344 ASSERT(ifp->if_real_bytes == 0);
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002345 ifp->if_u1.if_data = kmem_alloc(real_size,
2346 KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2348 /*
2349 * Only do the realloc if the underlying size
2350 * is really changing.
2351 */
2352 if (ifp->if_real_bytes != real_size) {
2353 ifp->if_u1.if_data =
2354 kmem_realloc(ifp->if_u1.if_data,
2355 real_size,
2356 ifp->if_real_bytes,
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002357 KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 }
2359 } else {
2360 ASSERT(ifp->if_real_bytes == 0);
Dave Chinner4a7eddd2010-07-20 17:53:59 +10002361 ifp->if_u1.if_data = kmem_alloc(real_size,
2362 KM_SLEEP | KM_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2364 ifp->if_bytes);
2365 }
2366 }
2367 ifp->if_real_bytes = real_size;
2368 ifp->if_bytes = new_size;
2369 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2370}
2371
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372void
2373xfs_idestroy_fork(
2374 xfs_inode_t *ip,
2375 int whichfork)
2376{
2377 xfs_ifork_t *ifp;
2378
2379 ifp = XFS_IFORK_PTR(ip, whichfork);
2380 if (ifp->if_broot != NULL) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002381 kmem_free(ifp->if_broot);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 ifp->if_broot = NULL;
2383 }
2384
2385 /*
2386 * If the format is local, then we can't have an extents
2387 * array so just look for an inline data array. If we're
2388 * not local then we may or may not have an extents list,
2389 * so check and free it up if we do.
2390 */
2391 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2392 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2393 (ifp->if_u1.if_data != NULL)) {
2394 ASSERT(ifp->if_real_bytes != 0);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002395 kmem_free(ifp->if_u1.if_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 ifp->if_u1.if_data = NULL;
2397 ifp->if_real_bytes = 0;
2398 }
2399 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11002400 ((ifp->if_flags & XFS_IFEXTIREC) ||
2401 ((ifp->if_u1.if_extents != NULL) &&
2402 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 ASSERT(ifp->if_real_bytes != 0);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002404 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 }
2406 ASSERT(ifp->if_u1.if_extents == NULL ||
2407 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2408 ASSERT(ifp->if_real_bytes == 0);
2409 if (whichfork == XFS_ATTR_FORK) {
2410 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2411 ip->i_afp = NULL;
2412 }
2413}
2414
2415/*
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002416 * This is called to unpin an inode. The caller must have the inode locked
2417 * in at least shared mode so that the buffer cannot be subsequently pinned
2418 * once someone is waiting for it to be unpinned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 */
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002420static void
2421xfs_iunpin_nowait(
2422 struct xfs_inode *ip)
David Chinnera3f74ff2008-03-06 13:43:42 +11002423{
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002424 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnera3f74ff2008-03-06 13:43:42 +11002425
Dave Chinner4aaf15d2010-03-08 11:24:07 +11002426 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2427
David Chinnera3f74ff2008-03-06 13:43:42 +11002428 /* Give the log a push to start the unpinning I/O */
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002429 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002430
David Chinnera3f74ff2008-03-06 13:43:42 +11002431}
2432
Dave Chinner777df5a2010-02-06 12:37:26 +11002433void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434xfs_iunpin_wait(
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002435 struct xfs_inode *ip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436{
Christoph Hellwig60ec6782010-02-17 19:43:56 +00002437 if (xfs_ipincount(ip)) {
2438 xfs_iunpin_nowait(ip);
2439 wait_event(ip->i_ipin_wait, (xfs_ipincount(ip) == 0));
2440 }
David Chinnera3f74ff2008-03-06 13:43:42 +11002441}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443/*
2444 * xfs_iextents_copy()
2445 *
2446 * This is called to copy the REAL extents (as opposed to the delayed
2447 * allocation extents) from the inode into the given buffer. It
2448 * returns the number of bytes copied into the buffer.
2449 *
2450 * If there are no delayed allocation extents, then we can just
2451 * memcpy() the extents into the buffer. Otherwise, we need to
2452 * examine each extent in turn and skip those which are delayed.
2453 */
2454int
2455xfs_iextents_copy(
2456 xfs_inode_t *ip,
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002457 xfs_bmbt_rec_t *dp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 int whichfork)
2459{
2460 int copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 int i;
2462 xfs_ifork_t *ifp;
2463 int nrecs;
2464 xfs_fsblock_t start_block;
2465
2466 ifp = XFS_IFORK_PTR(ip, whichfork);
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002467 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468 ASSERT(ifp->if_bytes > 0);
2469
2470 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
Eric Sandeen3a59c942007-07-11 11:09:47 +10002471 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 ASSERT(nrecs > 0);
2473
2474 /*
2475 * There are some delayed allocation extents in the
2476 * inode, so copy the extents one at a time and skip
2477 * the delayed ones. There must be at least one
2478 * non-delayed extent.
2479 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 copied = 0;
2481 for (i = 0; i < nrecs; i++) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002482 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 start_block = xfs_bmbt_get_startblock(ep);
Eric Sandeen9d87c312009-01-14 23:22:07 -06002484 if (isnullstartblock(start_block)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 /*
2486 * It's a delayed allocation extent, so skip it.
2487 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 continue;
2489 }
2490
2491 /* Translate to on disk format */
Christoph Hellwigcd8b0a92007-08-16 16:24:15 +10002492 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2493 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002494 dp++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 copied++;
2496 }
2497 ASSERT(copied != 0);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10002498 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499
2500 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2501}
2502
2503/*
2504 * Each of the following cases stores data into the same region
2505 * of the on-disk inode, so only one of them can be valid at
2506 * any given time. While it is possible to have conflicting formats
2507 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2508 * in EXTENTS format, this can only happen when the fork has
2509 * changed formats after being modified but before being flushed.
2510 * In these cases, the format always takes precedence, because the
2511 * format indicates the current state of the fork.
2512 */
2513/*ARGSUSED*/
David Chinnere4ac9672008-04-10 12:23:58 +10002514STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515xfs_iflush_fork(
2516 xfs_inode_t *ip,
2517 xfs_dinode_t *dip,
2518 xfs_inode_log_item_t *iip,
2519 int whichfork,
2520 xfs_buf_t *bp)
2521{
2522 char *cp;
2523 xfs_ifork_t *ifp;
2524 xfs_mount_t *mp;
2525#ifdef XFS_TRANS_DEBUG
2526 int first;
2527#endif
2528 static const short brootflag[2] =
2529 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2530 static const short dataflag[2] =
2531 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2532 static const short extflag[2] =
2533 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2534
David Chinnere4ac9672008-04-10 12:23:58 +10002535 if (!iip)
2536 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 ifp = XFS_IFORK_PTR(ip, whichfork);
2538 /*
2539 * This can happen if we gave up in iformat in an error path,
2540 * for the attribute fork.
2541 */
David Chinnere4ac9672008-04-10 12:23:58 +10002542 if (!ifp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 ASSERT(whichfork == XFS_ATTR_FORK);
David Chinnere4ac9672008-04-10 12:23:58 +10002544 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 }
2546 cp = XFS_DFORK_PTR(dip, whichfork);
2547 mp = ip->i_mount;
2548 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2549 case XFS_DINODE_FMT_LOCAL:
2550 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2551 (ifp->if_bytes > 0)) {
2552 ASSERT(ifp->if_u1.if_data != NULL);
2553 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2554 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 break;
2557
2558 case XFS_DINODE_FMT_EXTENTS:
2559 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2560 !(iip->ili_format.ilf_fields & extflag[whichfork]));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002561 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2562 (ifp->if_bytes == 0));
2563 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2564 (ifp->if_bytes > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2566 (ifp->if_bytes > 0)) {
2567 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2568 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2569 whichfork);
2570 }
2571 break;
2572
2573 case XFS_DINODE_FMT_BTREE:
2574 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2575 (ifp->if_broot_bytes > 0)) {
2576 ASSERT(ifp->if_broot != NULL);
2577 ASSERT(ifp->if_broot_bytes <=
2578 (XFS_IFORK_SIZE(ip, whichfork) +
2579 XFS_BROOT_SIZE_ADJ));
Christoph Hellwig60197e82008-10-30 17:11:19 +11002580 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 (xfs_bmdr_block_t *)cp,
2582 XFS_DFORK_SIZE(dip, mp, whichfork));
2583 }
2584 break;
2585
2586 case XFS_DINODE_FMT_DEV:
2587 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2588 ASSERT(whichfork == XFS_DATA_FORK);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002589 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 }
2591 break;
2592
2593 case XFS_DINODE_FMT_UUID:
2594 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2595 ASSERT(whichfork == XFS_DATA_FORK);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002596 memcpy(XFS_DFORK_DPTR(dip),
2597 &ip->i_df.if_u2.if_uuid,
2598 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 }
2600 break;
2601
2602 default:
2603 ASSERT(0);
2604 break;
2605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606}
2607
David Chinnerbad55842008-03-06 13:43:49 +11002608STATIC int
2609xfs_iflush_cluster(
2610 xfs_inode_t *ip,
2611 xfs_buf_t *bp)
2612{
2613 xfs_mount_t *mp = ip->i_mount;
Dave Chinner5017e972010-01-11 11:47:40 +00002614 struct xfs_perag *pag;
David Chinnerbad55842008-03-06 13:43:49 +11002615 unsigned long first_index, mask;
David Chinnerc8f5f122008-05-20 11:30:15 +10002616 unsigned long inodes_per_cluster;
David Chinnerbad55842008-03-06 13:43:49 +11002617 int ilist_size;
2618 xfs_inode_t **ilist;
2619 xfs_inode_t *iq;
David Chinnerbad55842008-03-06 13:43:49 +11002620 int nr_found;
2621 int clcount = 0;
2622 int bufwasdelwri;
2623 int i;
2624
Dave Chinner5017e972010-01-11 11:47:40 +00002625 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
David Chinnerbad55842008-03-06 13:43:49 +11002626
David Chinnerc8f5f122008-05-20 11:30:15 +10002627 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2628 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
David Chinner49383b02008-05-19 16:29:34 +10002629 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
David Chinnerbad55842008-03-06 13:43:49 +11002630 if (!ilist)
Dave Chinner44b56e02010-01-11 11:47:43 +00002631 goto out_put;
David Chinnerbad55842008-03-06 13:43:49 +11002632
2633 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2634 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002635 rcu_read_lock();
David Chinnerbad55842008-03-06 13:43:49 +11002636 /* really need a gang lookup range call here */
2637 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
David Chinnerc8f5f122008-05-20 11:30:15 +10002638 first_index, inodes_per_cluster);
David Chinnerbad55842008-03-06 13:43:49 +11002639 if (nr_found == 0)
2640 goto out_free;
2641
2642 for (i = 0; i < nr_found; i++) {
2643 iq = ilist[i];
2644 if (iq == ip)
2645 continue;
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002646
2647 /*
2648 * because this is an RCU protected lookup, we could find a
2649 * recently freed or even reallocated inode during the lookup.
2650 * We need to check under the i_flags_lock for a valid inode
2651 * here. Skip it if it is not valid or the wrong inode.
2652 */
2653 spin_lock(&ip->i_flags_lock);
2654 if (!ip->i_ino ||
2655 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2656 spin_unlock(&ip->i_flags_lock);
2657 continue;
2658 }
2659 spin_unlock(&ip->i_flags_lock);
2660
David Chinnerbad55842008-03-06 13:43:49 +11002661 /*
2662 * Do an un-protected check to see if the inode is dirty and
2663 * is a candidate for flushing. These checks will be repeated
2664 * later after the appropriate locks are acquired.
2665 */
David Chinner33540402008-03-06 13:43:59 +11002666 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
David Chinnerbad55842008-03-06 13:43:49 +11002667 continue;
David Chinnerbad55842008-03-06 13:43:49 +11002668
2669 /*
2670 * Try to get locks. If any are unavailable or it is pinned,
2671 * then this inode cannot be flushed and is skipped.
2672 */
2673
2674 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2675 continue;
2676 if (!xfs_iflock_nowait(iq)) {
2677 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2678 continue;
2679 }
2680 if (xfs_ipincount(iq)) {
2681 xfs_ifunlock(iq);
2682 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2683 continue;
2684 }
2685
2686 /*
2687 * arriving here means that this inode can be flushed. First
2688 * re-check that it's dirty before flushing.
2689 */
David Chinner33540402008-03-06 13:43:59 +11002690 if (!xfs_inode_clean(iq)) {
2691 int error;
David Chinnerbad55842008-03-06 13:43:49 +11002692 error = xfs_iflush_int(iq, bp);
2693 if (error) {
2694 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2695 goto cluster_corrupt_out;
2696 }
2697 clcount++;
2698 } else {
2699 xfs_ifunlock(iq);
2700 }
2701 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2702 }
2703
2704 if (clcount) {
2705 XFS_STATS_INC(xs_icluster_flushcnt);
2706 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2707 }
2708
2709out_free:
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002710 rcu_read_unlock();
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002711 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00002712out_put:
2713 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11002714 return 0;
2715
2716
2717cluster_corrupt_out:
2718 /*
2719 * Corruption detected in the clustering loop. Invalidate the
2720 * inode buffer and shut down the filesystem.
2721 */
Dave Chinner1a3e8f32010-12-17 17:29:43 +11002722 rcu_read_unlock();
David Chinnerbad55842008-03-06 13:43:49 +11002723 /*
2724 * Clean up the buffer. If it was B_DELWRI, just release it --
2725 * brelse can handle it with no problems. If not, shut down the
2726 * filesystem before releasing the buffer.
2727 */
2728 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2729 if (bufwasdelwri)
2730 xfs_buf_relse(bp);
2731
2732 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2733
2734 if (!bufwasdelwri) {
2735 /*
2736 * Just like incore_relse: if we have b_iodone functions,
2737 * mark the buffer as an error and call them. Otherwise
2738 * mark it as stale and brelse.
2739 */
2740 if (XFS_BUF_IODONE_FUNC(bp)) {
David Chinnerbad55842008-03-06 13:43:49 +11002741 XFS_BUF_UNDONE(bp);
2742 XFS_BUF_STALE(bp);
David Chinnerbad55842008-03-06 13:43:49 +11002743 XFS_BUF_ERROR(bp,EIO);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00002744 xfs_buf_ioend(bp, 0);
David Chinnerbad55842008-03-06 13:43:49 +11002745 } else {
2746 XFS_BUF_STALE(bp);
2747 xfs_buf_relse(bp);
2748 }
2749 }
2750
2751 /*
2752 * Unlocks the flush lock
2753 */
2754 xfs_iflush_abort(iq);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002755 kmem_free(ilist);
Dave Chinner44b56e02010-01-11 11:47:43 +00002756 xfs_perag_put(pag);
David Chinnerbad55842008-03-06 13:43:49 +11002757 return XFS_ERROR(EFSCORRUPTED);
2758}
2759
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760/*
2761 * xfs_iflush() will write a modified inode's changes out to the
2762 * inode's on disk home. The caller must have the inode lock held
David Chinnerc63942d2008-08-13 16:41:16 +10002763 * in at least shared mode and the inode flush completion must be
2764 * active as well. The inode lock will still be held upon return from
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 * the call and the caller is free to unlock it.
David Chinnerc63942d2008-08-13 16:41:16 +10002766 * The inode flush will be completed when the inode reaches the disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 * The flags indicate how the inode's buffer should be written out.
2768 */
2769int
2770xfs_iflush(
2771 xfs_inode_t *ip,
2772 uint flags)
2773{
2774 xfs_inode_log_item_t *iip;
2775 xfs_buf_t *bp;
2776 xfs_dinode_t *dip;
2777 xfs_mount_t *mp;
2778 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779
2780 XFS_STATS_INC(xs_iflush_count);
2781
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002782 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnerc63942d2008-08-13 16:41:16 +10002783 ASSERT(!completion_done(&ip->i_flush));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2785 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2786
2787 iip = ip->i_itemp;
2788 mp = ip->i_mount;
2789
2790 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002791 * We can't flush the inode until it is unpinned, so wait for it if we
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002792 * are allowed to block. We know no one new can pin it, because we are
David Chinnera3f74ff2008-03-06 13:43:42 +11002793 * holding the inode lock shared and you need to hold it exclusively to
2794 * pin the inode.
2795 *
2796 * If we are not allowed to block, force the log out asynchronously so
2797 * that when we come back the inode will be unpinned. If other inodes
2798 * in the same cluster are dirty, they will probably write the inode
2799 * out for us if they occur after the log force completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 */
Dave Chinnerc8543632010-02-06 12:39:36 +11002801 if (!(flags & SYNC_WAIT) && xfs_ipincount(ip)) {
David Chinnera3f74ff2008-03-06 13:43:42 +11002802 xfs_iunpin_nowait(ip);
2803 xfs_ifunlock(ip);
2804 return EAGAIN;
2805 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 xfs_iunpin_wait(ip);
2807
2808 /*
Dave Chinner4b6a4682010-01-11 11:45:21 +00002809 * For stale inodes we cannot rely on the backing buffer remaining
2810 * stale in cache for the remaining life of the stale inode and so
2811 * xfs_itobp() below may give us a buffer that no longer contains
2812 * inodes below. We have to check this after ensuring the inode is
2813 * unpinned so that it is safe to reclaim the stale inode after the
2814 * flush call.
2815 */
2816 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2817 xfs_ifunlock(ip);
2818 return 0;
2819 }
2820
2821 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 * This may have been unpinned because the filesystem is shutting
2823 * down forcibly. If that's the case we must not write this inode
2824 * to disk, because the log record didn't make it to disk!
2825 */
2826 if (XFS_FORCED_SHUTDOWN(mp)) {
2827 ip->i_update_core = 0;
2828 if (iip)
2829 iip->ili_format.ilf_fields = 0;
2830 xfs_ifunlock(ip);
2831 return XFS_ERROR(EIO);
2832 }
2833
2834 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002835 * Get the buffer containing the on-disk inode.
2836 */
Christoph Hellwig76d8b272008-11-28 14:23:40 +11002837 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
Dave Chinner1bfd8d02011-03-26 09:13:55 +11002838 (flags & SYNC_TRYLOCK) ? XBF_TRYLOCK : XBF_LOCK);
David Chinnera3f74ff2008-03-06 13:43:42 +11002839 if (error || !bp) {
2840 xfs_ifunlock(ip);
2841 return error;
2842 }
2843
2844 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 * First flush out the inode that xfs_iflush was called with.
2846 */
2847 error = xfs_iflush_int(ip, bp);
David Chinnerbad55842008-03-06 13:43:49 +11002848 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 goto corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850
2851 /*
David Chinnera3f74ff2008-03-06 13:43:42 +11002852 * If the buffer is pinned then push on the log now so we won't
2853 * get stuck waiting in the write for too long.
2854 */
2855 if (XFS_BUF_ISPINNED(bp))
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002856 xfs_log_force(mp, 0);
David Chinnera3f74ff2008-03-06 13:43:42 +11002857
2858 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 * inode clustering:
2860 * see if other inodes can be gathered into this write
2861 */
David Chinnerbad55842008-03-06 13:43:49 +11002862 error = xfs_iflush_cluster(ip, bp);
2863 if (error)
2864 goto cluster_corrupt_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865
Dave Chinnerc8543632010-02-06 12:39:36 +11002866 if (flags & SYNC_WAIT)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 error = xfs_bwrite(mp, bp);
Dave Chinnerc8543632010-02-06 12:39:36 +11002868 else
2869 xfs_bdwrite(mp, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 return error;
2871
2872corrupt_out:
2873 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10002874 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875cluster_corrupt_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 /*
2877 * Unlocks the flush lock
2878 */
David Chinnerbad55842008-03-06 13:43:49 +11002879 xfs_iflush_abort(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 return XFS_ERROR(EFSCORRUPTED);
2881}
2882
2883
2884STATIC int
2885xfs_iflush_int(
2886 xfs_inode_t *ip,
2887 xfs_buf_t *bp)
2888{
2889 xfs_inode_log_item_t *iip;
2890 xfs_dinode_t *dip;
2891 xfs_mount_t *mp;
2892#ifdef XFS_TRANS_DEBUG
2893 int first;
2894#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895
Christoph Hellwig579aa9c2008-04-22 17:34:00 +10002896 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
David Chinnerc63942d2008-08-13 16:41:16 +10002897 ASSERT(!completion_done(&ip->i_flush));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2899 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2900
2901 iip = ip->i_itemp;
2902 mp = ip->i_mount;
2903
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 /* set *dip = inode's place in the buffer */
Christoph Hellwig92bfc6e2008-11-28 14:23:41 +11002905 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906
2907 /*
2908 * Clear i_update_core before copying out the data.
2909 * This is for coordination with our timestamp updates
2910 * that don't hold the inode lock. They will always
2911 * update the timestamps BEFORE setting i_update_core,
2912 * so if we clear i_update_core after they set it we
2913 * are guaranteed to see their updates to the timestamps.
2914 * I believe that this depends on strongly ordered memory
2915 * semantics, but we have that. We use the SYNCHRONIZE
2916 * macro to make sure that the compiler does not reorder
2917 * the i_update_core access below the data copy below.
2918 */
2919 ip->i_update_core = 0;
2920 SYNCHRONIZE();
2921
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11002922 /*
Christoph Hellwigf9581b12009-10-06 20:29:26 +00002923 * Make sure to get the latest timestamps from the Linux inode.
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11002924 */
Christoph Hellwigf9581b12009-10-06 20:29:26 +00002925 xfs_synchronize_times(ip);
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11002926
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002927 if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002929 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2930 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2931 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 goto corrupt_out;
2933 }
2934 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2935 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002936 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2937 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2938 __func__, ip->i_ino, ip, ip->i_d.di_magic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 goto corrupt_out;
2940 }
2941 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
2942 if (XFS_TEST_ERROR(
2943 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2944 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2945 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002946 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2947 "%s: Bad regular inode %Lu, ptr 0x%p",
2948 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 goto corrupt_out;
2950 }
2951 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
2952 if (XFS_TEST_ERROR(
2953 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2954 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2955 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2956 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002957 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2958 "%s: Bad directory inode %Lu, ptr 0x%p",
2959 __func__, ip->i_ino, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 goto corrupt_out;
2961 }
2962 }
2963 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2964 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2965 XFS_RANDOM_IFLUSH_5)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002966 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2967 "%s: detected corrupt incore inode %Lu, "
2968 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2969 __func__, ip->i_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970 ip->i_d.di_nextents + ip->i_d.di_anextents,
Dave Chinner6a19d932011-03-07 10:02:35 +11002971 ip->i_d.di_nblocks, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 goto corrupt_out;
2973 }
2974 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2975 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
Dave Chinner6a19d932011-03-07 10:02:35 +11002976 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2977 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2978 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 goto corrupt_out;
2980 }
2981 /*
2982 * bump the flush iteration count, used to detect flushes which
2983 * postdate a log record during recovery.
2984 */
2985
2986 ip->i_d.di_flushiter++;
2987
2988 /*
2989 * Copy the dirty parts of the inode into the on-disk
2990 * inode. We always copy out the core of the inode,
2991 * because if the inode is dirty at all the core must
2992 * be.
2993 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002994 xfs_dinode_to_disk(dip, &ip->i_d);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995
2996 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2997 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2998 ip->i_d.di_flushiter = 0;
2999
3000 /*
3001 * If this is really an old format inode and the superblock version
3002 * has not been updated to support only new format inodes, then
3003 * convert back to the old inode format. If the superblock version
3004 * has been updated, then make the conversion permanent.
3005 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11003006 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3007 if (ip->i_d.di_version == 1) {
Eric Sandeen62118702008-03-06 13:44:28 +11003008 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009 /*
3010 * Convert it back.
3011 */
3012 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003013 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 } else {
3015 /*
3016 * The superblock version has already been bumped,
3017 * so just make the conversion to the new inode
3018 * format permanent.
3019 */
Christoph Hellwig51ce16d2008-11-28 14:23:39 +11003020 ip->i_d.di_version = 2;
3021 dip->di_version = 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 ip->i_d.di_onlink = 0;
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003023 dip->di_onlink = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
Christoph Hellwig81591fe2008-11-28 14:23:39 +11003025 memset(&(dip->di_pad[0]), 0,
3026 sizeof(dip->di_pad));
Arkadiusz Mi?kiewicz67430992010-09-26 06:10:18 +00003027 ASSERT(xfs_get_projid(ip) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 }
3029 }
3030
David Chinnere4ac9672008-04-10 12:23:58 +10003031 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3032 if (XFS_IFORK_Q(ip))
3033 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 xfs_inobp_check(mp, bp);
3035
3036 /*
3037 * We've recorded everything logged in the inode, so we'd
3038 * like to clear the ilf_fields bits so we don't log and
3039 * flush things unnecessarily. However, we can't stop
3040 * logging all this information until the data we've copied
3041 * into the disk buffer is written to disk. If we did we might
3042 * overwrite the copy of the inode in the log with all the
3043 * data after re-logging only part of it, and in the face of
3044 * a crash we wouldn't have all the data we need to recover.
3045 *
3046 * What we do is move the bits to the ili_last_fields field.
3047 * When logging the inode, these bits are moved back to the
3048 * ilf_fields field. In the xfs_iflush_done() routine we
3049 * clear ili_last_fields, since we know that the information
3050 * those bits represent is permanently on disk. As long as
3051 * the flush completes before the inode is logged again, then
3052 * both ilf_fields and ili_last_fields will be cleared.
3053 *
3054 * We can play with the ilf_fields bits here, because the inode
3055 * lock must be held exclusively in order to set bits there
3056 * and the flush lock protects the ili_last_fields bits.
3057 * Set ili_logged so the flush done
3058 * routine can tell whether or not to look in the AIL.
3059 * Also, store the current LSN of the inode so that we can tell
3060 * whether the item has moved in the AIL from xfs_iflush_done().
3061 * In order to read the lsn we need the AIL lock, because
3062 * it is a 64 bit value that cannot be read atomically.
3063 */
3064 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3065 iip->ili_last_fields = iip->ili_format.ilf_fields;
3066 iip->ili_format.ilf_fields = 0;
3067 iip->ili_logged = 1;
3068
David Chinner7b2e2a32008-10-30 17:39:12 +11003069 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3070 &iip->ili_item.li_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071
3072 /*
3073 * Attach the function xfs_iflush_done to the inode's
3074 * buffer. This will remove the inode from the AIL
3075 * and unlock the inode's flush lock when the inode is
3076 * completely written to disk.
3077 */
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10003078 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079
3080 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3081 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3082 } else {
3083 /*
3084 * We're flushing an inode which is not in the AIL and has
3085 * not been logged but has i_update_core set. For this
3086 * case we can use a B_DELWRI flush and immediately drop
3087 * the inode flush lock because we can avoid the whole
3088 * AIL state thing. It's OK to drop the flush lock now,
3089 * because we've already locked the buffer and to do anything
3090 * you really need both.
3091 */
3092 if (iip != NULL) {
3093 ASSERT(iip->ili_logged == 0);
3094 ASSERT(iip->ili_last_fields == 0);
3095 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3096 }
3097 xfs_ifunlock(ip);
3098 }
3099
3100 return 0;
3101
3102corrupt_out:
3103 return XFS_ERROR(EFSCORRUPTED);
3104}
3105
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003106/*
3107 * Return a pointer to the extent record at file index idx.
3108 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003109xfs_bmbt_rec_host_t *
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003110xfs_iext_get_ext(
3111 xfs_ifork_t *ifp, /* inode fork pointer */
3112 xfs_extnum_t idx) /* index of target extent */
3113{
3114 ASSERT(idx >= 0);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003115 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3116 return ifp->if_u1.if_ext_irec->er_extbuf;
3117 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3118 xfs_ext_irec_t *erp; /* irec pointer */
3119 int erp_idx = 0; /* irec index */
3120 xfs_extnum_t page_idx = idx; /* ext index in target list */
3121
3122 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3123 return &erp->er_extbuf[page_idx];
3124 } else if (ifp->if_bytes) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003125 return &ifp->if_u1.if_extents[idx];
3126 } else {
3127 return NULL;
3128 }
3129}
3130
3131/*
3132 * Insert new item(s) into the extent records for incore inode
3133 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3134 */
3135void
3136xfs_iext_insert(
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003137 xfs_inode_t *ip, /* incore inode pointer */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003138 xfs_extnum_t idx, /* starting index of new items */
3139 xfs_extnum_t count, /* number of inserted items */
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003140 xfs_bmbt_irec_t *new, /* items to insert */
3141 int state) /* type of extent conversion */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003142{
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003143 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003144 xfs_extnum_t i; /* extent record index */
3145
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003146 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
3147
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003148 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3149 xfs_iext_add(ifp, idx, count);
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003150 for (i = idx; i < idx + count; i++, new++)
3151 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003152}
3153
3154/*
3155 * This is called when the amount of space required for incore file
3156 * extents needs to be increased. The ext_diff parameter stores the
3157 * number of new extents being added and the idx parameter contains
3158 * the extent index where the new extents will be added. If the new
3159 * extents are being appended, then we just need to (re)allocate and
3160 * initialize the space. Otherwise, if the new extents are being
3161 * inserted into the middle of the existing entries, a bit more work
3162 * is required to make room for the new extents to be inserted. The
3163 * caller is responsible for filling in the new extent entries upon
3164 * return.
3165 */
3166void
3167xfs_iext_add(
3168 xfs_ifork_t *ifp, /* inode fork pointer */
3169 xfs_extnum_t idx, /* index to begin adding exts */
Nathan Scottc41564b2006-03-29 08:55:14 +10003170 int ext_diff) /* number of extents to add */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003171{
3172 int byte_diff; /* new bytes being added */
3173 int new_size; /* size of extents after adding */
3174 xfs_extnum_t nextents; /* number of extents in file */
3175
3176 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3177 ASSERT((idx >= 0) && (idx <= nextents));
3178 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3179 new_size = ifp->if_bytes + byte_diff;
3180 /*
3181 * If the new number of extents (nextents + ext_diff)
3182 * fits inside the inode, then continue to use the inline
3183 * extent buffer.
3184 */
3185 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3186 if (idx < nextents) {
3187 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3188 &ifp->if_u2.if_inline_ext[idx],
3189 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3190 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3191 }
3192 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3193 ifp->if_real_bytes = 0;
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003194 ifp->if_lastex = nextents + ext_diff;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003195 }
3196 /*
3197 * Otherwise use a linear (direct) extent list.
3198 * If the extents are currently inside the inode,
3199 * xfs_iext_realloc_direct will switch us from
3200 * inline to direct extent allocation mode.
3201 */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003202 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003203 xfs_iext_realloc_direct(ifp, new_size);
3204 if (idx < nextents) {
3205 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3206 &ifp->if_u1.if_extents[idx],
3207 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3208 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3209 }
3210 }
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003211 /* Indirection array */
3212 else {
3213 xfs_ext_irec_t *erp;
3214 int erp_idx = 0;
3215 int page_idx = idx;
3216
3217 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3218 if (ifp->if_flags & XFS_IFEXTIREC) {
3219 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3220 } else {
3221 xfs_iext_irec_init(ifp);
3222 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3223 erp = ifp->if_u1.if_ext_irec;
3224 }
3225 /* Extents fit in target extent page */
3226 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3227 if (page_idx < erp->er_extcount) {
3228 memmove(&erp->er_extbuf[page_idx + ext_diff],
3229 &erp->er_extbuf[page_idx],
3230 (erp->er_extcount - page_idx) *
3231 sizeof(xfs_bmbt_rec_t));
3232 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3233 }
3234 erp->er_extcount += ext_diff;
3235 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3236 }
3237 /* Insert a new extent page */
3238 else if (erp) {
3239 xfs_iext_add_indirect_multi(ifp,
3240 erp_idx, page_idx, ext_diff);
3241 }
3242 /*
3243 * If extent(s) are being appended to the last page in
3244 * the indirection array and the new extent(s) don't fit
3245 * in the page, then erp is NULL and erp_idx is set to
3246 * the next index needed in the indirection array.
3247 */
3248 else {
3249 int count = ext_diff;
3250
3251 while (count) {
3252 erp = xfs_iext_irec_new(ifp, erp_idx);
3253 erp->er_extcount = count;
3254 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3255 if (count) {
3256 erp_idx++;
3257 }
3258 }
3259 }
3260 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003261 ifp->if_bytes = new_size;
3262}
3263
3264/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003265 * This is called when incore extents are being added to the indirection
3266 * array and the new extents do not fit in the target extent list. The
3267 * erp_idx parameter contains the irec index for the target extent list
3268 * in the indirection array, and the idx parameter contains the extent
3269 * index within the list. The number of extents being added is stored
3270 * in the count parameter.
3271 *
3272 * |-------| |-------|
3273 * | | | | idx - number of extents before idx
3274 * | idx | | count |
3275 * | | | | count - number of extents being inserted at idx
3276 * |-------| |-------|
3277 * | count | | nex2 | nex2 - number of extents after idx + count
3278 * |-------| |-------|
3279 */
3280void
3281xfs_iext_add_indirect_multi(
3282 xfs_ifork_t *ifp, /* inode fork pointer */
3283 int erp_idx, /* target extent irec index */
3284 xfs_extnum_t idx, /* index within target list */
3285 int count) /* new extents being added */
3286{
3287 int byte_diff; /* new bytes being added */
3288 xfs_ext_irec_t *erp; /* pointer to irec entry */
3289 xfs_extnum_t ext_diff; /* number of extents to add */
3290 xfs_extnum_t ext_cnt; /* new extents still needed */
3291 xfs_extnum_t nex2; /* extents after idx + count */
3292 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3293 int nlists; /* number of irec's (lists) */
3294
3295 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3296 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3297 nex2 = erp->er_extcount - idx;
3298 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3299
3300 /*
3301 * Save second part of target extent list
3302 * (all extents past */
3303 if (nex2) {
3304 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
David Chinner67850732008-08-13 16:02:51 +10003305 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003306 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3307 erp->er_extcount -= nex2;
3308 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3309 memset(&erp->er_extbuf[idx], 0, byte_diff);
3310 }
3311
3312 /*
3313 * Add the new extents to the end of the target
3314 * list, then allocate new irec record(s) and
3315 * extent buffer(s) as needed to store the rest
3316 * of the new extents.
3317 */
3318 ext_cnt = count;
3319 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3320 if (ext_diff) {
3321 erp->er_extcount += ext_diff;
3322 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3323 ext_cnt -= ext_diff;
3324 }
3325 while (ext_cnt) {
3326 erp_idx++;
3327 erp = xfs_iext_irec_new(ifp, erp_idx);
3328 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3329 erp->er_extcount = ext_diff;
3330 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3331 ext_cnt -= ext_diff;
3332 }
3333
3334 /* Add nex2 extents back to indirection array */
3335 if (nex2) {
3336 xfs_extnum_t ext_avail;
3337 int i;
3338
3339 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3340 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3341 i = 0;
3342 /*
3343 * If nex2 extents fit in the current page, append
3344 * nex2_ep after the new extents.
3345 */
3346 if (nex2 <= ext_avail) {
3347 i = erp->er_extcount;
3348 }
3349 /*
3350 * Otherwise, check if space is available in the
3351 * next page.
3352 */
3353 else if ((erp_idx < nlists - 1) &&
3354 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3355 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3356 erp_idx++;
3357 erp++;
3358 /* Create a hole for nex2 extents */
3359 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3360 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3361 }
3362 /*
3363 * Final choice, create a new extent page for
3364 * nex2 extents.
3365 */
3366 else {
3367 erp_idx++;
3368 erp = xfs_iext_irec_new(ifp, erp_idx);
3369 }
3370 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003371 kmem_free(nex2_ep);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003372 erp->er_extcount += nex2;
3373 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3374 }
3375}
3376
3377/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003378 * This is called when the amount of space required for incore file
3379 * extents needs to be decreased. The ext_diff parameter stores the
3380 * number of extents to be removed and the idx parameter contains
3381 * the extent index where the extents will be removed from.
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003382 *
3383 * If the amount of space needed has decreased below the linear
3384 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3385 * extent array. Otherwise, use kmem_realloc() to adjust the
3386 * size to what is needed.
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003387 */
3388void
3389xfs_iext_remove(
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003390 xfs_inode_t *ip, /* incore inode pointer */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003391 xfs_extnum_t idx, /* index to begin removing exts */
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003392 int ext_diff, /* number of extents to remove */
3393 int state) /* type of extent conversion */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003394{
Christoph Hellwig6ef35542009-11-25 00:00:21 +00003395 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003396 xfs_extnum_t nextents; /* number of extents in file */
3397 int new_size; /* size of extents after removal */
3398
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003399 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3400
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003401 ASSERT(ext_diff > 0);
3402 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3403 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3404
3405 if (new_size == 0) {
3406 xfs_iext_destroy(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003407 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3408 xfs_iext_remove_indirect(ifp, idx, ext_diff);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003409 } else if (ifp->if_real_bytes) {
3410 xfs_iext_remove_direct(ifp, idx, ext_diff);
3411 } else {
3412 xfs_iext_remove_inline(ifp, idx, ext_diff);
3413 }
3414 ifp->if_bytes = new_size;
3415}
3416
3417/*
3418 * This removes ext_diff extents from the inline buffer, beginning
3419 * at extent index idx.
3420 */
3421void
3422xfs_iext_remove_inline(
3423 xfs_ifork_t *ifp, /* inode fork pointer */
3424 xfs_extnum_t idx, /* index to begin removing exts */
3425 int ext_diff) /* number of extents to remove */
3426{
3427 int nextents; /* number of extents in file */
3428
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003429 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003430 ASSERT(idx < XFS_INLINE_EXTS);
3431 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3432 ASSERT(((nextents - ext_diff) > 0) &&
3433 (nextents - ext_diff) < XFS_INLINE_EXTS);
3434
3435 if (idx + ext_diff < nextents) {
3436 memmove(&ifp->if_u2.if_inline_ext[idx],
3437 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3438 (nextents - (idx + ext_diff)) *
3439 sizeof(xfs_bmbt_rec_t));
3440 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3441 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3442 } else {
3443 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3444 ext_diff * sizeof(xfs_bmbt_rec_t));
3445 }
3446}
3447
3448/*
3449 * This removes ext_diff extents from a linear (direct) extent list,
3450 * beginning at extent index idx. If the extents are being removed
3451 * from the end of the list (ie. truncate) then we just need to re-
3452 * allocate the list to remove the extra space. Otherwise, if the
3453 * extents are being removed from the middle of the existing extent
3454 * entries, then we first need to move the extent records beginning
3455 * at idx + ext_diff up in the list to overwrite the records being
3456 * removed, then remove the extra space via kmem_realloc.
3457 */
3458void
3459xfs_iext_remove_direct(
3460 xfs_ifork_t *ifp, /* inode fork pointer */
3461 xfs_extnum_t idx, /* index to begin removing exts */
3462 int ext_diff) /* number of extents to remove */
3463{
3464 xfs_extnum_t nextents; /* number of extents in file */
3465 int new_size; /* size of extents after removal */
3466
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003467 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003468 new_size = ifp->if_bytes -
3469 (ext_diff * sizeof(xfs_bmbt_rec_t));
3470 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3471
3472 if (new_size == 0) {
3473 xfs_iext_destroy(ifp);
3474 return;
3475 }
3476 /* Move extents up in the list (if needed) */
3477 if (idx + ext_diff < nextents) {
3478 memmove(&ifp->if_u1.if_extents[idx],
3479 &ifp->if_u1.if_extents[idx + ext_diff],
3480 (nextents - (idx + ext_diff)) *
3481 sizeof(xfs_bmbt_rec_t));
3482 }
3483 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3484 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3485 /*
3486 * Reallocate the direct extent list. If the extents
3487 * will fit inside the inode then xfs_iext_realloc_direct
3488 * will switch from direct to inline extent allocation
3489 * mode for us.
3490 */
3491 xfs_iext_realloc_direct(ifp, new_size);
3492 ifp->if_bytes = new_size;
3493}
3494
3495/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003496 * This is called when incore extents are being removed from the
3497 * indirection array and the extents being removed span multiple extent
3498 * buffers. The idx parameter contains the file extent index where we
3499 * want to begin removing extents, and the count parameter contains
3500 * how many extents need to be removed.
3501 *
3502 * |-------| |-------|
3503 * | nex1 | | | nex1 - number of extents before idx
3504 * |-------| | count |
3505 * | | | | count - number of extents being removed at idx
3506 * | count | |-------|
3507 * | | | nex2 | nex2 - number of extents after idx + count
3508 * |-------| |-------|
3509 */
3510void
3511xfs_iext_remove_indirect(
3512 xfs_ifork_t *ifp, /* inode fork pointer */
3513 xfs_extnum_t idx, /* index to begin removing extents */
3514 int count) /* number of extents to remove */
3515{
3516 xfs_ext_irec_t *erp; /* indirection array pointer */
3517 int erp_idx = 0; /* indirection array index */
3518 xfs_extnum_t ext_cnt; /* extents left to remove */
3519 xfs_extnum_t ext_diff; /* extents to remove in current list */
3520 xfs_extnum_t nex1; /* number of extents before idx */
3521 xfs_extnum_t nex2; /* extents after idx + count */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003522 int page_idx = idx; /* index in target extent list */
3523
3524 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3525 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3526 ASSERT(erp != NULL);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003527 nex1 = page_idx;
3528 ext_cnt = count;
3529 while (ext_cnt) {
3530 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3531 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3532 /*
3533 * Check for deletion of entire list;
3534 * xfs_iext_irec_remove() updates extent offsets.
3535 */
3536 if (ext_diff == erp->er_extcount) {
3537 xfs_iext_irec_remove(ifp, erp_idx);
3538 ext_cnt -= ext_diff;
3539 nex1 = 0;
3540 if (ext_cnt) {
3541 ASSERT(erp_idx < ifp->if_real_bytes /
3542 XFS_IEXT_BUFSZ);
3543 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3544 nex1 = 0;
3545 continue;
3546 } else {
3547 break;
3548 }
3549 }
3550 /* Move extents up (if needed) */
3551 if (nex2) {
3552 memmove(&erp->er_extbuf[nex1],
3553 &erp->er_extbuf[nex1 + ext_diff],
3554 nex2 * sizeof(xfs_bmbt_rec_t));
3555 }
3556 /* Zero out rest of page */
3557 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3558 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3559 /* Update remaining counters */
3560 erp->er_extcount -= ext_diff;
3561 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3562 ext_cnt -= ext_diff;
3563 nex1 = 0;
3564 erp_idx++;
3565 erp++;
3566 }
3567 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3568 xfs_iext_irec_compact(ifp);
3569}
3570
3571/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003572 * Create, destroy, or resize a linear (direct) block of extents.
3573 */
3574void
3575xfs_iext_realloc_direct(
3576 xfs_ifork_t *ifp, /* inode fork pointer */
3577 int new_size) /* new size of extents */
3578{
3579 int rnew_size; /* real new size of extents */
3580
3581 rnew_size = new_size;
3582
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003583 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3584 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3585 (new_size != ifp->if_real_bytes)));
3586
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003587 /* Free extent records */
3588 if (new_size == 0) {
3589 xfs_iext_destroy(ifp);
3590 }
3591 /* Resize direct extent list and zero any new bytes */
3592 else if (ifp->if_real_bytes) {
3593 /* Check if extents will fit inside the inode */
3594 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3595 xfs_iext_direct_to_inline(ifp, new_size /
3596 (uint)sizeof(xfs_bmbt_rec_t));
3597 ifp->if_bytes = new_size;
3598 return;
3599 }
Vignesh Babu16a087d2007-06-28 16:46:37 +10003600 if (!is_power_of_2(new_size)){
Robert P. J. Day40ebd812007-11-23 16:30:51 +11003601 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003602 }
3603 if (rnew_size != ifp->if_real_bytes) {
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003604 ifp->if_u1.if_extents =
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003605 kmem_realloc(ifp->if_u1.if_extents,
3606 rnew_size,
David Chinner67850732008-08-13 16:02:51 +10003607 ifp->if_real_bytes, KM_NOFS);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003608 }
3609 if (rnew_size > ifp->if_real_bytes) {
3610 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3611 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3612 rnew_size - ifp->if_real_bytes);
3613 }
3614 }
3615 /*
3616 * Switch from the inline extent buffer to a direct
3617 * extent list. Be sure to include the inline extent
3618 * bytes in new_size.
3619 */
3620 else {
3621 new_size += ifp->if_bytes;
Vignesh Babu16a087d2007-06-28 16:46:37 +10003622 if (!is_power_of_2(new_size)) {
Robert P. J. Day40ebd812007-11-23 16:30:51 +11003623 rnew_size = roundup_pow_of_two(new_size);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003624 }
3625 xfs_iext_inline_to_direct(ifp, rnew_size);
3626 }
3627 ifp->if_real_bytes = rnew_size;
3628 ifp->if_bytes = new_size;
3629}
3630
3631/*
3632 * Switch from linear (direct) extent records to inline buffer.
3633 */
3634void
3635xfs_iext_direct_to_inline(
3636 xfs_ifork_t *ifp, /* inode fork pointer */
3637 xfs_extnum_t nextents) /* number of extents in file */
3638{
3639 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3640 ASSERT(nextents <= XFS_INLINE_EXTS);
3641 /*
3642 * The inline buffer was zeroed when we switched
3643 * from inline to direct extent allocation mode,
3644 * so we don't need to clear it here.
3645 */
3646 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3647 nextents * sizeof(xfs_bmbt_rec_t));
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003648 kmem_free(ifp->if_u1.if_extents);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003649 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3650 ifp->if_real_bytes = 0;
3651}
3652
3653/*
3654 * Switch from inline buffer to linear (direct) extent records.
3655 * new_size should already be rounded up to the next power of 2
3656 * by the caller (when appropriate), so use new_size as it is.
3657 * However, since new_size may be rounded up, we can't update
3658 * if_bytes here. It is the caller's responsibility to update
3659 * if_bytes upon return.
3660 */
3661void
3662xfs_iext_inline_to_direct(
3663 xfs_ifork_t *ifp, /* inode fork pointer */
3664 int new_size) /* number of extents in file */
3665{
David Chinner67850732008-08-13 16:02:51 +10003666 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003667 memset(ifp->if_u1.if_extents, 0, new_size);
3668 if (ifp->if_bytes) {
3669 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3670 ifp->if_bytes);
3671 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3672 sizeof(xfs_bmbt_rec_t));
3673 }
3674 ifp->if_real_bytes = new_size;
3675}
3676
3677/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003678 * Resize an extent indirection array to new_size bytes.
3679 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003680STATIC void
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003681xfs_iext_realloc_indirect(
3682 xfs_ifork_t *ifp, /* inode fork pointer */
3683 int new_size) /* new indirection array size */
3684{
3685 int nlists; /* number of irec's (ex lists) */
3686 int size; /* current indirection array size */
3687
3688 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3689 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3690 size = nlists * sizeof(xfs_ext_irec_t);
3691 ASSERT(ifp->if_real_bytes);
3692 ASSERT((new_size >= 0) && (new_size != size));
3693 if (new_size == 0) {
3694 xfs_iext_destroy(ifp);
3695 } else {
3696 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3697 kmem_realloc(ifp->if_u1.if_ext_irec,
David Chinner67850732008-08-13 16:02:51 +10003698 new_size, size, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003699 }
3700}
3701
3702/*
3703 * Switch from indirection array to linear (direct) extent allocations.
3704 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003705STATIC void
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003706xfs_iext_indirect_to_direct(
3707 xfs_ifork_t *ifp) /* inode fork pointer */
3708{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003709 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003710 xfs_extnum_t nextents; /* number of extents in file */
3711 int size; /* size of file extents */
3712
3713 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3714 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3715 ASSERT(nextents <= XFS_LINEAR_EXTS);
3716 size = nextents * sizeof(xfs_bmbt_rec_t);
3717
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10003718 xfs_iext_irec_compact_pages(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003719 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3720
3721 ep = ifp->if_u1.if_ext_irec->er_extbuf;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003722 kmem_free(ifp->if_u1.if_ext_irec);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003723 ifp->if_flags &= ~XFS_IFEXTIREC;
3724 ifp->if_u1.if_extents = ep;
3725 ifp->if_bytes = size;
3726 if (nextents < XFS_LINEAR_EXTS) {
3727 xfs_iext_realloc_direct(ifp, size);
3728 }
3729}
3730
3731/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003732 * Free incore file extents.
3733 */
3734void
3735xfs_iext_destroy(
3736 xfs_ifork_t *ifp) /* inode fork pointer */
3737{
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003738 if (ifp->if_flags & XFS_IFEXTIREC) {
3739 int erp_idx;
3740 int nlists;
3741
3742 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3743 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3744 xfs_iext_irec_remove(ifp, erp_idx);
3745 }
3746 ifp->if_flags &= ~XFS_IFEXTIREC;
3747 } else if (ifp->if_real_bytes) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003748 kmem_free(ifp->if_u1.if_extents);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003749 } else if (ifp->if_bytes) {
3750 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3751 sizeof(xfs_bmbt_rec_t));
3752 }
3753 ifp->if_u1.if_extents = NULL;
3754 ifp->if_real_bytes = 0;
3755 ifp->if_bytes = 0;
3756}
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003757
3758/*
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003759 * Return a pointer to the extent record for file system block bno.
3760 */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003761xfs_bmbt_rec_host_t * /* pointer to found extent record */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003762xfs_iext_bno_to_ext(
3763 xfs_ifork_t *ifp, /* inode fork pointer */
3764 xfs_fileoff_t bno, /* block number to search for */
3765 xfs_extnum_t *idxp) /* index of target extent */
3766{
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003767 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003768 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
Christoph Hellwiga6f64d42007-08-16 16:23:40 +10003769 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003770 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
Nathan Scottc41564b2006-03-29 08:55:14 +10003771 int high; /* upper boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003772 xfs_extnum_t idx = 0; /* index of target extent */
Nathan Scottc41564b2006-03-29 08:55:14 +10003773 int low; /* lower boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003774 xfs_extnum_t nextents; /* number of file extents */
3775 xfs_fileoff_t startoff = 0; /* start offset of extent */
3776
3777 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3778 if (nextents == 0) {
3779 *idxp = 0;
3780 return NULL;
3781 }
3782 low = 0;
3783 if (ifp->if_flags & XFS_IFEXTIREC) {
3784 /* Find target extent list */
3785 int erp_idx = 0;
3786 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3787 base = erp->er_extbuf;
3788 high = erp->er_extcount - 1;
3789 } else {
3790 base = ifp->if_u1.if_extents;
3791 high = nextents - 1;
3792 }
3793 /* Binary search extent records */
3794 while (low <= high) {
3795 idx = (low + high) >> 1;
3796 ep = base + idx;
3797 startoff = xfs_bmbt_get_startoff(ep);
3798 blockcount = xfs_bmbt_get_blockcount(ep);
3799 if (bno < startoff) {
3800 high = idx - 1;
3801 } else if (bno >= startoff + blockcount) {
3802 low = idx + 1;
3803 } else {
3804 /* Convert back to file-based extent index */
3805 if (ifp->if_flags & XFS_IFEXTIREC) {
3806 idx += erp->er_extoff;
3807 }
3808 *idxp = idx;
3809 return ep;
3810 }
3811 }
3812 /* Convert back to file-based extent index */
3813 if (ifp->if_flags & XFS_IFEXTIREC) {
3814 idx += erp->er_extoff;
3815 }
3816 if (bno >= startoff + blockcount) {
3817 if (++idx == nextents) {
3818 ep = NULL;
3819 } else {
3820 ep = xfs_iext_get_ext(ifp, idx);
3821 }
3822 }
3823 *idxp = idx;
3824 return ep;
3825}
3826
3827/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003828 * Return a pointer to the indirection array entry containing the
3829 * extent record for filesystem block bno. Store the index of the
3830 * target irec in *erp_idxp.
3831 */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003832xfs_ext_irec_t * /* pointer to found extent record */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003833xfs_iext_bno_to_irec(
3834 xfs_ifork_t *ifp, /* inode fork pointer */
3835 xfs_fileoff_t bno, /* block number to search for */
3836 int *erp_idxp) /* irec index of target ext list */
3837{
3838 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3839 xfs_ext_irec_t *erp_next; /* next indirection array entry */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11003840 int erp_idx; /* indirection array index */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003841 int nlists; /* number of extent irec's (lists) */
3842 int high; /* binary search upper limit */
3843 int low; /* binary search lower limit */
3844
3845 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3846 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3847 erp_idx = 0;
3848 low = 0;
3849 high = nlists - 1;
3850 while (low <= high) {
3851 erp_idx = (low + high) >> 1;
3852 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3853 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3854 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3855 high = erp_idx - 1;
3856 } else if (erp_next && bno >=
3857 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3858 low = erp_idx + 1;
3859 } else {
3860 break;
3861 }
3862 }
3863 *erp_idxp = erp_idx;
3864 return erp;
3865}
3866
3867/*
3868 * Return a pointer to the indirection array entry containing the
3869 * extent record at file extent index *idxp. Store the index of the
3870 * target irec in *erp_idxp and store the page index of the target
3871 * extent record in *idxp.
3872 */
3873xfs_ext_irec_t *
3874xfs_iext_idx_to_irec(
3875 xfs_ifork_t *ifp, /* inode fork pointer */
3876 xfs_extnum_t *idxp, /* extent index (file -> page) */
3877 int *erp_idxp, /* pointer to target irec */
3878 int realloc) /* new bytes were just added */
3879{
3880 xfs_ext_irec_t *prev; /* pointer to previous irec */
3881 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3882 int erp_idx; /* indirection array index */
3883 int nlists; /* number of irec's (ex lists) */
3884 int high; /* binary search upper limit */
3885 int low; /* binary search lower limit */
3886 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3887
3888 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3889 ASSERT(page_idx >= 0 && page_idx <=
3890 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
3891 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3892 erp_idx = 0;
3893 low = 0;
3894 high = nlists - 1;
3895
3896 /* Binary search extent irec's */
3897 while (low <= high) {
3898 erp_idx = (low + high) >> 1;
3899 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3900 prev = erp_idx > 0 ? erp - 1 : NULL;
3901 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3902 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3903 high = erp_idx - 1;
3904 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3905 (page_idx == erp->er_extoff + erp->er_extcount &&
3906 !realloc)) {
3907 low = erp_idx + 1;
3908 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3909 erp->er_extcount == XFS_LINEAR_EXTS) {
3910 ASSERT(realloc);
3911 page_idx = 0;
3912 erp_idx++;
3913 erp = erp_idx < nlists ? erp + 1 : NULL;
3914 break;
3915 } else {
3916 page_idx -= erp->er_extoff;
3917 break;
3918 }
3919 }
3920 *idxp = page_idx;
3921 *erp_idxp = erp_idx;
3922 return(erp);
3923}
3924
3925/*
3926 * Allocate and initialize an indirection array once the space needed
3927 * for incore extents increases above XFS_IEXT_BUFSZ.
3928 */
3929void
3930xfs_iext_irec_init(
3931 xfs_ifork_t *ifp) /* inode fork pointer */
3932{
3933 xfs_ext_irec_t *erp; /* indirection array pointer */
3934 xfs_extnum_t nextents; /* number of extents in file */
3935
3936 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3937 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3938 ASSERT(nextents <= XFS_LINEAR_EXTS);
3939
David Chinner67850732008-08-13 16:02:51 +10003940 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003941
3942 if (nextents == 0) {
David Chinner67850732008-08-13 16:02:51 +10003943 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003944 } else if (!ifp->if_real_bytes) {
3945 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3946 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3947 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3948 }
3949 erp->er_extbuf = ifp->if_u1.if_extents;
3950 erp->er_extcount = nextents;
3951 erp->er_extoff = 0;
3952
3953 ifp->if_flags |= XFS_IFEXTIREC;
3954 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3955 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3956 ifp->if_u1.if_ext_irec = erp;
3957
3958 return;
3959}
3960
3961/*
3962 * Allocate and initialize a new entry in the indirection array.
3963 */
3964xfs_ext_irec_t *
3965xfs_iext_irec_new(
3966 xfs_ifork_t *ifp, /* inode fork pointer */
3967 int erp_idx) /* index for new irec */
3968{
3969 xfs_ext_irec_t *erp; /* indirection array pointer */
3970 int i; /* loop counter */
3971 int nlists; /* number of irec's (ex lists) */
3972
3973 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3974 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3975
3976 /* Resize indirection array */
3977 xfs_iext_realloc_indirect(ifp, ++nlists *
3978 sizeof(xfs_ext_irec_t));
3979 /*
3980 * Move records down in the array so the
3981 * new page can use erp_idx.
3982 */
3983 erp = ifp->if_u1.if_ext_irec;
3984 for (i = nlists - 1; i > erp_idx; i--) {
3985 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3986 }
3987 ASSERT(i == erp_idx);
3988
3989 /* Initialize new extent record */
3990 erp = ifp->if_u1.if_ext_irec;
David Chinner67850732008-08-13 16:02:51 +10003991 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003992 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3993 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3994 erp[erp_idx].er_extcount = 0;
3995 erp[erp_idx].er_extoff = erp_idx > 0 ?
3996 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3997 return (&erp[erp_idx]);
3998}
3999
4000/*
4001 * Remove a record from the indirection array.
4002 */
4003void
4004xfs_iext_irec_remove(
4005 xfs_ifork_t *ifp, /* inode fork pointer */
4006 int erp_idx) /* irec index to remove */
4007{
4008 xfs_ext_irec_t *erp; /* indirection array pointer */
4009 int i; /* loop counter */
4010 int nlists; /* number of irec's (ex lists) */
4011
4012 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4013 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4014 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4015 if (erp->er_extbuf) {
4016 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4017 -erp->er_extcount);
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004018 kmem_free(erp->er_extbuf);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004019 }
4020 /* Compact extent records */
4021 erp = ifp->if_u1.if_ext_irec;
4022 for (i = erp_idx; i < nlists - 1; i++) {
4023 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4024 }
4025 /*
4026 * Manually free the last extent record from the indirection
4027 * array. A call to xfs_iext_realloc_indirect() with a size
4028 * of zero would result in a call to xfs_iext_destroy() which
4029 * would in turn call this function again, creating a nasty
4030 * infinite loop.
4031 */
4032 if (--nlists) {
4033 xfs_iext_realloc_indirect(ifp,
4034 nlists * sizeof(xfs_ext_irec_t));
4035 } else {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004036 kmem_free(ifp->if_u1.if_ext_irec);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004037 }
4038 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4039}
4040
4041/*
4042 * This is called to clean up large amounts of unused memory allocated
4043 * by the indirection array. Before compacting anything though, verify
4044 * that the indirection array is still needed and switch back to the
4045 * linear extent list (or even the inline buffer) if possible. The
4046 * compaction policy is as follows:
4047 *
4048 * Full Compaction: Extents fit into a single page (or inline buffer)
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10004049 * Partial Compaction: Extents occupy less than 50% of allocated space
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004050 * No Compaction: Extents occupy at least 50% of allocated space
4051 */
4052void
4053xfs_iext_irec_compact(
4054 xfs_ifork_t *ifp) /* inode fork pointer */
4055{
4056 xfs_extnum_t nextents; /* number of extents in file */
4057 int nlists; /* number of irec's (ex lists) */
4058
4059 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4060 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4061 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4062
4063 if (nextents == 0) {
4064 xfs_iext_destroy(ifp);
4065 } else if (nextents <= XFS_INLINE_EXTS) {
4066 xfs_iext_indirect_to_direct(ifp);
4067 xfs_iext_direct_to_inline(ifp, nextents);
4068 } else if (nextents <= XFS_LINEAR_EXTS) {
4069 xfs_iext_indirect_to_direct(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004070 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4071 xfs_iext_irec_compact_pages(ifp);
4072 }
4073}
4074
4075/*
4076 * Combine extents from neighboring extent pages.
4077 */
4078void
4079xfs_iext_irec_compact_pages(
4080 xfs_ifork_t *ifp) /* inode fork pointer */
4081{
4082 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4083 int erp_idx = 0; /* indirection array index */
4084 int nlists; /* number of irec's (ex lists) */
4085
4086 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4087 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4088 while (erp_idx < nlists - 1) {
4089 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4090 erp_next = erp + 1;
4091 if (erp_next->er_extcount <=
4092 (XFS_LINEAR_EXTS - erp->er_extcount)) {
Lachlan McIlroy71a8c872008-09-26 12:17:57 +10004093 memcpy(&erp->er_extbuf[erp->er_extcount],
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004094 erp_next->er_extbuf, erp_next->er_extcount *
4095 sizeof(xfs_bmbt_rec_t));
4096 erp->er_extcount += erp_next->er_extcount;
4097 /*
4098 * Free page before removing extent record
4099 * so er_extoffs don't get modified in
4100 * xfs_iext_irec_remove.
4101 */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10004102 kmem_free(erp_next->er_extbuf);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004103 erp_next->er_extbuf = NULL;
4104 xfs_iext_irec_remove(ifp, erp_idx + 1);
4105 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4106 } else {
4107 erp_idx++;
4108 }
4109 }
4110}
4111
4112/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004113 * This is called to update the er_extoff field in the indirection
4114 * array when extents have been added or removed from one of the
4115 * extent lists. erp_idx contains the irec index to begin updating
4116 * at and ext_diff contains the number of extents that were added
4117 * or removed.
4118 */
4119void
4120xfs_iext_irec_update_extoffs(
4121 xfs_ifork_t *ifp, /* inode fork pointer */
4122 int erp_idx, /* irec index to update */
4123 int ext_diff) /* number of new extents */
4124{
4125 int i; /* loop counter */
4126 int nlists; /* number of irec's (ex lists */
4127
4128 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4129 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4130 for (i = erp_idx; i < nlists; i++) {
4131 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
4132 }
4133}