blob: c27d7d495aa0f498ade09aa6a5db5a67a8fa01ce [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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
24#include "xfs_imap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_trans.h"
26#include "xfs_trans_priv.h"
27#include "xfs_sb.h"
28#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_dir2.h"
30#include "xfs_dmapi.h"
31#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110036#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110040#include "xfs_inode_item.h"
41#include "xfs_btree.h"
42#include "xfs_alloc.h"
43#include "xfs_ialloc.h"
44#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include "xfs_rw.h"
46#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "xfs_utils.h"
48#include "xfs_dir2_trace.h"
49#include "xfs_quota.h"
50#include "xfs_mac.h"
51#include "xfs_acl.h"
52
53
54kmem_zone_t *xfs_ifork_zone;
55kmem_zone_t *xfs_inode_zone;
56kmem_zone_t *xfs_chashlist_zone;
57
58/*
59 * Used in xfs_itruncate(). This is the maximum number of extents
60 * freed from a file in a single transaction.
61 */
62#define XFS_ITRUNC_MAX_EXTENTS 2
63
64STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
65STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
66STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
67STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
68
69
70#ifdef DEBUG
71/*
72 * Make sure that the extents in the given memory buffer
73 * are valid.
74 */
75STATIC void
76xfs_validate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +110077 xfs_ifork_t *ifp,
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 int nrecs,
79 int disk,
80 xfs_exntfmt_t fmt)
81{
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +110082 xfs_bmbt_rec_t *ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 xfs_bmbt_irec_t irec;
84 xfs_bmbt_rec_t rec;
85 int i;
86
87 for (i = 0; i < nrecs; i++) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +110088 ep = xfs_iext_get_ext(ifp, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
90 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
91 if (disk)
92 xfs_bmbt_disk_get_all(&rec, &irec);
93 else
94 xfs_bmbt_get_all(&rec, &irec);
95 if (fmt == XFS_EXTFMT_NOSTATE)
96 ASSERT(irec.br_state == XFS_EXT_NORM);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 }
98}
99#else /* DEBUG */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100100#define xfs_validate_extents(ifp, nrecs, disk, fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101#endif /* DEBUG */
102
103/*
104 * Check that none of the inode's in the buffer have a next
105 * unlinked field of 0.
106 */
107#if defined(DEBUG)
108void
109xfs_inobp_check(
110 xfs_mount_t *mp,
111 xfs_buf_t *bp)
112{
113 int i;
114 int j;
115 xfs_dinode_t *dip;
116
117 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
118
119 for (i = 0; i < j; i++) {
120 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
121 i * mp->m_sb.sb_inodesize);
122 if (!dip->di_next_unlinked) {
123 xfs_fs_cmn_err(CE_ALERT, mp,
124 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
125 bp);
126 ASSERT(dip->di_next_unlinked);
127 }
128 }
129}
130#endif
131
132/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 * This routine is called to map an inode number within a file
134 * system to the buffer containing the on-disk version of the
135 * inode. It returns a pointer to the buffer containing the
136 * on-disk inode in the bpp parameter, and in the dip parameter
137 * it returns a pointer to the on-disk inode within that buffer.
138 *
139 * If a non-zero error is returned, then the contents of bpp and
140 * dipp are undefined.
141 *
142 * Use xfs_imap() to determine the size and location of the
143 * buffer to read from disk.
144 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000145STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146xfs_inotobp(
147 xfs_mount_t *mp,
148 xfs_trans_t *tp,
149 xfs_ino_t ino,
150 xfs_dinode_t **dipp,
151 xfs_buf_t **bpp,
152 int *offset)
153{
154 int di_ok;
155 xfs_imap_t imap;
156 xfs_buf_t *bp;
157 int error;
158 xfs_dinode_t *dip;
159
160 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000161 * Call the space management code to find the location of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 * inode on disk.
163 */
164 imap.im_blkno = 0;
165 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
166 if (error != 0) {
167 cmn_err(CE_WARN,
168 "xfs_inotobp: xfs_imap() returned an "
169 "error %d on %s. Returning error.", error, mp->m_fsname);
170 return error;
171 }
172
173 /*
174 * If the inode number maps to a block outside the bounds of the
175 * file system then return NULL rather than calling read_buf
176 * and panicing when we get an error from the driver.
177 */
178 if ((imap.im_blkno + imap.im_len) >
179 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
180 cmn_err(CE_WARN,
Christoph Hellwigda1650a2005-11-02 10:21:35 +1100181 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 "of the file system %s. Returning EINVAL.",
Christoph Hellwigda1650a2005-11-02 10:21:35 +1100183 (unsigned long long)imap.im_blkno,
184 imap.im_len, mp->m_fsname);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 return XFS_ERROR(EINVAL);
186 }
187
188 /*
189 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
190 * default to just a read_buf() call.
191 */
192 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
193 (int)imap.im_len, XFS_BUF_LOCK, &bp);
194
195 if (error) {
196 cmn_err(CE_WARN,
197 "xfs_inotobp: xfs_trans_read_buf() returned an "
198 "error %d on %s. Returning error.", error, mp->m_fsname);
199 return error;
200 }
201 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
202 di_ok =
203 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
204 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
205 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
206 XFS_RANDOM_ITOBP_INOTOBP))) {
207 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
208 xfs_trans_brelse(tp, bp);
209 cmn_err(CE_WARN,
210 "xfs_inotobp: XFS_TEST_ERROR() returned an "
211 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
212 return XFS_ERROR(EFSCORRUPTED);
213 }
214
215 xfs_inobp_check(mp, bp);
216
217 /*
218 * Set *dipp to point to the on-disk inode in the buffer.
219 */
220 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
221 *bpp = bp;
222 *offset = imap.im_boffset;
223 return 0;
224}
225
226
227/*
228 * This routine is called to map an inode to the buffer containing
229 * the on-disk version of the inode. It returns a pointer to the
230 * buffer containing the on-disk inode in the bpp parameter, and in
231 * the dip parameter it returns a pointer to the on-disk inode within
232 * that buffer.
233 *
234 * If a non-zero error is returned, then the contents of bpp and
235 * dipp are undefined.
236 *
237 * If the inode is new and has not yet been initialized, use xfs_imap()
238 * to determine the size and location of the buffer to read from disk.
239 * If the inode has already been mapped to its buffer and read in once,
240 * then use the mapping information stored in the inode rather than
241 * calling xfs_imap(). This allows us to avoid the overhead of looking
242 * at the inode btree for small block file systems (see xfs_dilocate()).
243 * We can tell whether the inode has been mapped in before by comparing
244 * its disk block address to 0. Only uninitialized inodes will have
245 * 0 for the disk block address.
246 */
247int
248xfs_itobp(
249 xfs_mount_t *mp,
250 xfs_trans_t *tp,
251 xfs_inode_t *ip,
252 xfs_dinode_t **dipp,
253 xfs_buf_t **bpp,
Nathan Scottb12dd342006-03-17 17:26:04 +1100254 xfs_daddr_t bno,
255 uint imap_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256{
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000257 xfs_imap_t imap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 xfs_buf_t *bp;
259 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 int i;
261 int ni;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
263 if (ip->i_blkno == (xfs_daddr_t)0) {
264 /*
265 * Call the space management code to find the location of the
266 * inode on disk.
267 */
268 imap.im_blkno = bno;
Nathan Scottb12dd342006-03-17 17:26:04 +1100269 if ((error = xfs_imap(mp, tp, ip->i_ino, &imap,
270 XFS_IMAP_LOOKUP | imap_flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
273 /*
274 * If the inode number maps to a block outside the bounds
275 * of the file system then return NULL rather than calling
276 * read_buf and panicing when we get an error from the
277 * driver.
278 */
279 if ((imap.im_blkno + imap.im_len) >
280 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
281#ifdef DEBUG
282 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
283 "(imap.im_blkno (0x%llx) "
284 "+ imap.im_len (0x%llx)) > "
285 " XFS_FSB_TO_BB(mp, "
286 "mp->m_sb.sb_dblocks) (0x%llx)",
287 (unsigned long long) imap.im_blkno,
288 (unsigned long long) imap.im_len,
289 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
290#endif /* DEBUG */
291 return XFS_ERROR(EINVAL);
292 }
293
294 /*
295 * Fill in the fields in the inode that will be used to
296 * map the inode to its buffer from now on.
297 */
298 ip->i_blkno = imap.im_blkno;
299 ip->i_len = imap.im_len;
300 ip->i_boffset = imap.im_boffset;
301 } else {
302 /*
303 * We've already mapped the inode once, so just use the
304 * mapping that we saved the first time.
305 */
306 imap.im_blkno = ip->i_blkno;
307 imap.im_len = ip->i_len;
308 imap.im_boffset = ip->i_boffset;
309 }
310 ASSERT(bno == 0 || bno == imap.im_blkno);
311
312 /*
313 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
314 * default to just a read_buf() call.
315 */
316 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
317 (int)imap.im_len, XFS_BUF_LOCK, &bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 if (error) {
319#ifdef DEBUG
320 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
321 "xfs_trans_read_buf() returned error %d, "
322 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
323 error, (unsigned long long) imap.im_blkno,
324 (unsigned long long) imap.im_len);
325#endif /* DEBUG */
326 return error;
327 }
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 /*
330 * Validate the magic number and version of every inode in the buffer
331 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000332 * No validation is done here in userspace (xfs_repair).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 */
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000334#if !defined(__KERNEL__)
335 ni = 0;
336#elif defined(DEBUG)
Nathan Scott41ff7152006-07-28 17:05:51 +1000337 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000338#else /* usual case */
Nathan Scott41ff7152006-07-28 17:05:51 +1000339 ni = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340#endif
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000341
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 for (i = 0; i < ni; i++) {
343 int di_ok;
344 xfs_dinode_t *dip;
345
346 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
347 (i << mp->m_sb.sb_inodelog));
348 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
349 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
Nathan Scott41ff7152006-07-28 17:05:51 +1000350 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
351 XFS_ERRTAG_ITOBP_INOTOBP,
352 XFS_RANDOM_ITOBP_INOTOBP))) {
353 if (imap_flags & XFS_IMAP_BULKSTAT) {
354 xfs_trans_brelse(tp, bp);
355 return XFS_ERROR(EINVAL);
356 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357#ifdef DEBUG
Nathan Scott41ff7152006-07-28 17:05:51 +1000358 cmn_err(CE_ALERT,
Nathan Scott4d1a2ed2006-06-09 17:12:28 +1000359 "Device %s - bad inode magic/vsn "
360 "daddr %lld #%d (magic=%x)",
Nathan Scottb6574522006-06-09 15:29:40 +1000361 XFS_BUFTARG_NAME(mp->m_ddev_targp),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 (unsigned long long)imap.im_blkno, i,
363 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
364#endif
365 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
366 mp, dip);
367 xfs_trans_brelse(tp, bp);
368 return XFS_ERROR(EFSCORRUPTED);
369 }
370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372 xfs_inobp_check(mp, bp);
373
374 /*
375 * Mark the buffer as an inode buffer now that it looks good
376 */
377 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
378
379 /*
380 * Set *dipp to point to the on-disk inode in the buffer.
381 */
382 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
383 *bpp = bp;
384 return 0;
385}
386
387/*
388 * Move inode type and inode format specific information from the
389 * on-disk inode to the in-core inode. For fifos, devs, and sockets
390 * this means set if_rdev to the proper value. For files, directories,
391 * and symlinks this means to bring in the in-line data or extent
392 * pointers. For a file in B-tree format, only the root is immediately
393 * brought in-core. The rest will be in-lined in if_extents when it
394 * is first referenced (see xfs_iread_extents()).
395 */
396STATIC int
397xfs_iformat(
398 xfs_inode_t *ip,
399 xfs_dinode_t *dip)
400{
401 xfs_attr_shortform_t *atp;
402 int size;
403 int error;
404 xfs_fsize_t di_size;
405 ip->i_df.if_ext_max =
406 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
407 error = 0;
408
409 if (unlikely(
410 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
411 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
412 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100413 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
414 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 (unsigned long long)ip->i_ino,
416 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
417 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
418 (unsigned long long)
419 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
420 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
421 ip->i_mount, dip);
422 return XFS_ERROR(EFSCORRUPTED);
423 }
424
425 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100426 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
427 "corrupt dinode %Lu, forkoff = 0x%x.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 (unsigned long long)ip->i_ino,
429 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
430 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
431 ip->i_mount, dip);
432 return XFS_ERROR(EFSCORRUPTED);
433 }
434
435 switch (ip->i_d.di_mode & S_IFMT) {
436 case S_IFIFO:
437 case S_IFCHR:
438 case S_IFBLK:
439 case S_IFSOCK:
440 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
441 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
442 ip->i_mount, dip);
443 return XFS_ERROR(EFSCORRUPTED);
444 }
445 ip->i_d.di_size = 0;
446 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
447 break;
448
449 case S_IFREG:
450 case S_IFLNK:
451 case S_IFDIR:
452 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
453 case XFS_DINODE_FMT_LOCAL:
454 /*
455 * no local regular files yet
456 */
457 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100458 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
459 "corrupt inode %Lu "
460 "(local format for regular file).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 (unsigned long long) ip->i_ino);
462 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
463 XFS_ERRLEVEL_LOW,
464 ip->i_mount, dip);
465 return XFS_ERROR(EFSCORRUPTED);
466 }
467
468 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
469 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100470 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
471 "corrupt inode %Lu "
472 "(bad size %Ld for local inode).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 (unsigned long long) ip->i_ino,
474 (long long) di_size);
475 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
476 XFS_ERRLEVEL_LOW,
477 ip->i_mount, dip);
478 return XFS_ERROR(EFSCORRUPTED);
479 }
480
481 size = (int)di_size;
482 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
483 break;
484 case XFS_DINODE_FMT_EXTENTS:
485 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
486 break;
487 case XFS_DINODE_FMT_BTREE:
488 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
489 break;
490 default:
491 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
492 ip->i_mount);
493 return XFS_ERROR(EFSCORRUPTED);
494 }
495 break;
496
497 default:
498 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
499 return XFS_ERROR(EFSCORRUPTED);
500 }
501 if (error) {
502 return error;
503 }
504 if (!XFS_DFORK_Q(dip))
505 return 0;
506 ASSERT(ip->i_afp == NULL);
507 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
508 ip->i_afp->if_ext_max =
509 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
510 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
511 case XFS_DINODE_FMT_LOCAL:
512 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
Nathan Scott3b244aa2006-03-17 17:29:25 +1100513 size = be16_to_cpu(atp->hdr.totsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
515 break;
516 case XFS_DINODE_FMT_EXTENTS:
517 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
518 break;
519 case XFS_DINODE_FMT_BTREE:
520 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
521 break;
522 default:
523 error = XFS_ERROR(EFSCORRUPTED);
524 break;
525 }
526 if (error) {
527 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
528 ip->i_afp = NULL;
529 xfs_idestroy_fork(ip, XFS_DATA_FORK);
530 }
531 return error;
532}
533
534/*
535 * The file is in-lined in the on-disk inode.
536 * If it fits into if_inline_data, then copy
537 * it there, otherwise allocate a buffer for it
538 * and copy the data there. Either way, set
539 * if_data to point at the data.
540 * If we allocate a buffer for the data, make
541 * sure that its size is a multiple of 4 and
542 * record the real size in i_real_bytes.
543 */
544STATIC int
545xfs_iformat_local(
546 xfs_inode_t *ip,
547 xfs_dinode_t *dip,
548 int whichfork,
549 int size)
550{
551 xfs_ifork_t *ifp;
552 int real_size;
553
554 /*
555 * If the size is unreasonable, then something
556 * is wrong and we just bail out rather than crash in
557 * kmem_alloc() or memcpy() below.
558 */
559 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100560 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
561 "corrupt inode %Lu "
562 "(bad size %d for local fork, size = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 (unsigned long long) ip->i_ino, size,
564 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
565 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
566 ip->i_mount, dip);
567 return XFS_ERROR(EFSCORRUPTED);
568 }
569 ifp = XFS_IFORK_PTR(ip, whichfork);
570 real_size = 0;
571 if (size == 0)
572 ifp->if_u1.if_data = NULL;
573 else if (size <= sizeof(ifp->if_u2.if_inline_data))
574 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
575 else {
576 real_size = roundup(size, 4);
577 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
578 }
579 ifp->if_bytes = size;
580 ifp->if_real_bytes = real_size;
581 if (size)
582 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
583 ifp->if_flags &= ~XFS_IFEXTENTS;
584 ifp->if_flags |= XFS_IFINLINE;
585 return 0;
586}
587
588/*
589 * The file consists of a set of extents all
590 * of which fit into the on-disk inode.
591 * If there are few enough extents to fit into
592 * the if_inline_ext, then copy them there.
593 * Otherwise allocate a buffer for them and copy
594 * them into it. Either way, set if_extents
595 * to point at the extents.
596 */
597STATIC int
598xfs_iformat_extents(
599 xfs_inode_t *ip,
600 xfs_dinode_t *dip,
601 int whichfork)
602{
603 xfs_bmbt_rec_t *ep, *dp;
604 xfs_ifork_t *ifp;
605 int nex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 int size;
607 int i;
608
609 ifp = XFS_IFORK_PTR(ip, whichfork);
610 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
611 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
612
613 /*
614 * If the number of extents is unreasonable, then something
615 * is wrong and we just bail out rather than crash in
616 * kmem_alloc() or memcpy() below.
617 */
618 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100619 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
620 "corrupt inode %Lu ((a)extents = %d).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 (unsigned long long) ip->i_ino, nex);
622 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
623 ip->i_mount, dip);
624 return XFS_ERROR(EFSCORRUPTED);
625 }
626
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100627 ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (nex == 0)
629 ifp->if_u1.if_extents = NULL;
630 else if (nex <= XFS_INLINE_EXTS)
631 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100632 else
633 xfs_iext_add(ifp, 0, nex);
634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 ifp->if_bytes = size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (size) {
637 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100638 xfs_validate_extents(ifp, nex, 1, XFS_EXTFMT_INODE(ip));
639 for (i = 0; i < nex; i++, dp++) {
640 ep = xfs_iext_get_ext(ifp, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
642 ARCH_CONVERT);
643 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
644 ARCH_CONVERT);
645 }
646 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
647 whichfork);
648 if (whichfork != XFS_DATA_FORK ||
649 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
650 if (unlikely(xfs_check_nostate_extents(
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +1100651 ifp, 0, nex))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
653 XFS_ERRLEVEL_LOW,
654 ip->i_mount);
655 return XFS_ERROR(EFSCORRUPTED);
656 }
657 }
658 ifp->if_flags |= XFS_IFEXTENTS;
659 return 0;
660}
661
662/*
663 * The file has too many extents to fit into
664 * the inode, so they are in B-tree format.
665 * Allocate a buffer for the root of the B-tree
666 * and copy the root into it. The i_extents
667 * field will remain NULL until all of the
668 * extents are read in (when they are needed).
669 */
670STATIC int
671xfs_iformat_btree(
672 xfs_inode_t *ip,
673 xfs_dinode_t *dip,
674 int whichfork)
675{
676 xfs_bmdr_block_t *dfp;
677 xfs_ifork_t *ifp;
678 /* REFERENCED */
679 int nrecs;
680 int size;
681
682 ifp = XFS_IFORK_PTR(ip, whichfork);
683 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
684 size = XFS_BMAP_BROOT_SPACE(dfp);
685 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
686
687 /*
688 * blow out if -- fork has less extents than can fit in
689 * fork (fork shouldn't be a btree format), root btree
690 * block has more records than can fit into the fork,
691 * or the number of extents is greater than the number of
692 * blocks.
693 */
694 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
695 || XFS_BMDR_SPACE_CALC(nrecs) >
696 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
697 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
Nathan Scott3762ec62006-01-12 10:29:53 +1100698 xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
699 "corrupt inode %Lu (btree).",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 (unsigned long long) ip->i_ino);
701 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
702 ip->i_mount);
703 return XFS_ERROR(EFSCORRUPTED);
704 }
705
706 ifp->if_broot_bytes = size;
707 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
708 ASSERT(ifp->if_broot != NULL);
709 /*
710 * Copy and convert from the on-disk structure
711 * to the in-memory structure.
712 */
713 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
714 ifp->if_broot, size);
715 ifp->if_flags &= ~XFS_IFEXTENTS;
716 ifp->if_flags |= XFS_IFBROOT;
717
718 return 0;
719}
720
721/*
722 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
723 * and native format
724 *
725 * buf = on-disk representation
726 * dip = native representation
727 * dir = direction - +ve -> disk to native
728 * -ve -> native to disk
729 */
730void
731xfs_xlate_dinode_core(
732 xfs_caddr_t buf,
733 xfs_dinode_core_t *dip,
734 int dir)
735{
736 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
737 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
738 xfs_arch_t arch = ARCH_CONVERT;
739
740 ASSERT(dir);
741
742 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
743 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
744 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
745 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
746 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
747 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
748 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
749 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
750 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
751
752 if (dir > 0) {
753 memcpy(mem_core->di_pad, buf_core->di_pad,
754 sizeof(buf_core->di_pad));
755 } else {
756 memcpy(buf_core->di_pad, mem_core->di_pad,
757 sizeof(buf_core->di_pad));
758 }
759
760 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
761
762 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
763 dir, arch);
764 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
765 dir, arch);
766 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
767 dir, arch);
768 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
769 dir, arch);
770 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
771 dir, arch);
772 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
773 dir, arch);
774 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
775 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
776 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
777 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
778 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
779 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
780 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
781 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
782 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
783 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
784 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
785}
786
787STATIC uint
788_xfs_dic2xflags(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 __uint16_t di_flags)
790{
791 uint flags = 0;
792
793 if (di_flags & XFS_DIFLAG_ANY) {
794 if (di_flags & XFS_DIFLAG_REALTIME)
795 flags |= XFS_XFLAG_REALTIME;
796 if (di_flags & XFS_DIFLAG_PREALLOC)
797 flags |= XFS_XFLAG_PREALLOC;
798 if (di_flags & XFS_DIFLAG_IMMUTABLE)
799 flags |= XFS_XFLAG_IMMUTABLE;
800 if (di_flags & XFS_DIFLAG_APPEND)
801 flags |= XFS_XFLAG_APPEND;
802 if (di_flags & XFS_DIFLAG_SYNC)
803 flags |= XFS_XFLAG_SYNC;
804 if (di_flags & XFS_DIFLAG_NOATIME)
805 flags |= XFS_XFLAG_NOATIME;
806 if (di_flags & XFS_DIFLAG_NODUMP)
807 flags |= XFS_XFLAG_NODUMP;
808 if (di_flags & XFS_DIFLAG_RTINHERIT)
809 flags |= XFS_XFLAG_RTINHERIT;
810 if (di_flags & XFS_DIFLAG_PROJINHERIT)
811 flags |= XFS_XFLAG_PROJINHERIT;
812 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
813 flags |= XFS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100814 if (di_flags & XFS_DIFLAG_EXTSIZE)
815 flags |= XFS_XFLAG_EXTSIZE;
816 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
817 flags |= XFS_XFLAG_EXTSZINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +1000818 if (di_flags & XFS_DIFLAG_NODEFRAG)
819 flags |= XFS_XFLAG_NODEFRAG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 }
821
822 return flags;
823}
824
825uint
826xfs_ip2xflags(
827 xfs_inode_t *ip)
828{
829 xfs_dinode_core_t *dic = &ip->i_d;
830
Nathan Scotta916e2b2006-06-09 17:12:17 +1000831 return _xfs_dic2xflags(dic->di_flags) |
832 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
835uint
836xfs_dic2xflags(
837 xfs_dinode_core_t *dic)
838{
Nathan Scotta916e2b2006-06-09 17:12:17 +1000839 return _xfs_dic2xflags(INT_GET(dic->di_flags, ARCH_CONVERT)) |
840 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841}
842
843/*
844 * Given a mount structure and an inode number, return a pointer
Nathan Scottc41564b2006-03-29 08:55:14 +1000845 * to a newly allocated in-core inode corresponding to the given
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 * inode number.
847 *
848 * Initialize the inode's attributes and extent pointers if it
849 * already has them (it will not if the inode has no links).
850 */
851int
852xfs_iread(
853 xfs_mount_t *mp,
854 xfs_trans_t *tp,
855 xfs_ino_t ino,
856 xfs_inode_t **ipp,
Nathan Scott745b1f472006-09-28 11:02:23 +1000857 xfs_daddr_t bno,
858 uint imap_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
860 xfs_buf_t *bp;
861 xfs_dinode_t *dip;
862 xfs_inode_t *ip;
863 int error;
864
865 ASSERT(xfs_inode_zone != NULL);
866
867 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
868 ip->i_ino = ino;
869 ip->i_mount = mp;
David Chinnerf273ab82006-09-28 11:06:03 +1000870 spin_lock_init(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
872 /*
873 * Get pointer's to the on-disk inode and the buffer containing it.
874 * If the inode number refers to a block outside the file system
875 * then xfs_itobp() will return NULL. In this case we should
876 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
877 * know that this is a new incore inode.
878 */
Nathan Scott745b1f472006-09-28 11:02:23 +1000879 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno, imap_flags);
Nathan Scottb12dd342006-03-17 17:26:04 +1100880 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 kmem_zone_free(xfs_inode_zone, ip);
882 return error;
883 }
884
885 /*
886 * Initialize inode's trace buffers.
887 * Do this before xfs_iformat in case it adds entries.
888 */
889#ifdef XFS_BMAP_TRACE
890 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
891#endif
892#ifdef XFS_BMBT_TRACE
893 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
894#endif
895#ifdef XFS_RW_TRACE
896 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
897#endif
898#ifdef XFS_ILOCK_TRACE
899 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
900#endif
901#ifdef XFS_DIR2_TRACE
902 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
903#endif
904
905 /*
906 * If we got something that isn't an inode it means someone
907 * (nfs or dmi) has a stale handle.
908 */
909 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
910 kmem_zone_free(xfs_inode_zone, ip);
911 xfs_trans_brelse(tp, bp);
912#ifdef DEBUG
913 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
914 "dip->di_core.di_magic (0x%x) != "
915 "XFS_DINODE_MAGIC (0x%x)",
916 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
917 XFS_DINODE_MAGIC);
918#endif /* DEBUG */
919 return XFS_ERROR(EINVAL);
920 }
921
922 /*
923 * If the on-disk inode is already linked to a directory
924 * entry, copy all of the inode into the in-core inode.
925 * xfs_iformat() handles copying in the inode format
926 * specific information.
927 * Otherwise, just get the truly permanent information.
928 */
929 if (dip->di_core.di_mode) {
930 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
931 &(ip->i_d), 1);
932 error = xfs_iformat(ip, dip);
933 if (error) {
934 kmem_zone_free(xfs_inode_zone, ip);
935 xfs_trans_brelse(tp, bp);
936#ifdef DEBUG
937 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
938 "xfs_iformat() returned error %d",
939 error);
940#endif /* DEBUG */
941 return error;
942 }
943 } else {
944 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
945 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
946 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
947 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
948 /*
949 * Make sure to pull in the mode here as well in
950 * case the inode is released without being used.
951 * This ensures that xfs_inactive() will see that
952 * the inode is already free and not try to mess
953 * with the uninitialized part of it.
954 */
955 ip->i_d.di_mode = 0;
956 /*
957 * Initialize the per-fork minima and maxima for a new
958 * inode here. xfs_iformat will do it for old inodes.
959 */
960 ip->i_df.if_ext_max =
961 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
962 }
963
964 INIT_LIST_HEAD(&ip->i_reclaim);
965
966 /*
967 * The inode format changed when we moved the link count and
968 * made it 32 bits long. If this is an old format inode,
969 * convert it in memory to look like a new one. If it gets
970 * flushed to disk we will convert back before flushing or
971 * logging it. We zero out the new projid field and the old link
972 * count field. We'll handle clearing the pad field (the remains
973 * of the old uuid field) when we actually convert the inode to
974 * the new format. We don't change the version number so that we
975 * can distinguish this from a real new format inode.
976 */
977 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
978 ip->i_d.di_nlink = ip->i_d.di_onlink;
979 ip->i_d.di_onlink = 0;
980 ip->i_d.di_projid = 0;
981 }
982
983 ip->i_delayed_blks = 0;
984
985 /*
986 * Mark the buffer containing the inode as something to keep
987 * around for a while. This helps to keep recently accessed
988 * meta-data in-core longer.
989 */
990 XFS_BUF_SET_REF(bp, XFS_INO_REF);
991
992 /*
993 * Use xfs_trans_brelse() to release the buffer containing the
994 * on-disk inode, because it was acquired with xfs_trans_read_buf()
995 * in xfs_itobp() above. If tp is NULL, this is just a normal
996 * brelse(). If we're within a transaction, then xfs_trans_brelse()
997 * will only release the buffer if it is not dirty within the
998 * transaction. It will be OK to release the buffer in this case,
999 * because inodes on disk are never destroyed and we will be
1000 * locking the new in-core inode before putting it in the hash
1001 * table where other processes can find it. Thus we don't have
1002 * to worry about the inode being changed just because we released
1003 * the buffer.
1004 */
1005 xfs_trans_brelse(tp, bp);
1006 *ipp = ip;
1007 return 0;
1008}
1009
1010/*
1011 * Read in extents from a btree-format inode.
1012 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1013 */
1014int
1015xfs_iread_extents(
1016 xfs_trans_t *tp,
1017 xfs_inode_t *ip,
1018 int whichfork)
1019{
1020 int error;
1021 xfs_ifork_t *ifp;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001022 xfs_extnum_t nextents;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 size_t size;
1024
1025 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1026 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1027 ip->i_mount);
1028 return XFS_ERROR(EFSCORRUPTED);
1029 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001030 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
1031 size = nextents * sizeof(xfs_bmbt_rec_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 ifp = XFS_IFORK_PTR(ip, whichfork);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 /*
1035 * We know that the size is valid (it's checked in iformat_btree)
1036 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 ifp->if_lastex = NULLEXTNUM;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001038 ifp->if_bytes = ifp->if_real_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 ifp->if_flags |= XFS_IFEXTENTS;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001040 xfs_iext_add(ifp, 0, nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 error = xfs_bmap_read_extents(tp, ip, whichfork);
1042 if (error) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001043 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 ifp->if_flags &= ~XFS_IFEXTENTS;
1045 return error;
1046 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11001047 xfs_validate_extents(ifp, nextents, 0, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 return 0;
1049}
1050
1051/*
1052 * Allocate an inode on disk and return a copy of its in-core version.
1053 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1054 * appropriately within the inode. The uid and gid for the inode are
1055 * set according to the contents of the given cred structure.
1056 *
1057 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1058 * has a free inode available, call xfs_iget()
1059 * to obtain the in-core version of the allocated inode. Finally,
1060 * fill in the inode and log its initial contents. In this case,
1061 * ialloc_context would be set to NULL and call_again set to false.
1062 *
1063 * If xfs_dialloc() does not have an available inode,
1064 * it will replenish its supply by doing an allocation. Since we can
1065 * only do one allocation within a transaction without deadlocks, we
1066 * must commit the current transaction before returning the inode itself.
1067 * In this case, therefore, we will set call_again to true and return.
1068 * The caller should then commit the current transaction, start a new
1069 * transaction, and call xfs_ialloc() again to actually get the inode.
1070 *
1071 * To ensure that some other process does not grab the inode that
1072 * was allocated during the first call to xfs_ialloc(), this routine
1073 * also returns the [locked] bp pointing to the head of the freelist
1074 * as ialloc_context. The caller should hold this buffer across
1075 * the commit and pass it back into this routine on the second call.
1076 */
1077int
1078xfs_ialloc(
1079 xfs_trans_t *tp,
1080 xfs_inode_t *pip,
1081 mode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -07001082 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 xfs_dev_t rdev,
1084 cred_t *cr,
1085 xfs_prid_t prid,
1086 int okalloc,
1087 xfs_buf_t **ialloc_context,
1088 boolean_t *call_again,
1089 xfs_inode_t **ipp)
1090{
1091 xfs_ino_t ino;
1092 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001093 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 uint flags;
1095 int error;
1096
1097 /*
1098 * Call the space management code to pick
1099 * the on-disk inode to be allocated.
1100 */
1101 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1102 ialloc_context, call_again, &ino);
1103 if (error != 0) {
1104 return error;
1105 }
1106 if (*call_again || ino == NULLFSINO) {
1107 *ipp = NULL;
1108 return 0;
1109 }
1110 ASSERT(*ialloc_context == NULL);
1111
1112 /*
1113 * Get the in-core inode with the lock held exclusively.
1114 * This is because we're setting fields here we need
1115 * to prevent others from looking at until we're done.
1116 */
1117 error = xfs_trans_iget(tp->t_mountp, tp, ino,
Nathan Scott745b1f472006-09-28 11:02:23 +10001118 XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 if (error != 0) {
1120 return error;
1121 }
1122 ASSERT(ip != NULL);
1123
1124 vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 ip->i_d.di_mode = (__uint16_t)mode;
1126 ip->i_d.di_onlink = 0;
1127 ip->i_d.di_nlink = nlink;
1128 ASSERT(ip->i_d.di_nlink == nlink);
1129 ip->i_d.di_uid = current_fsuid(cr);
1130 ip->i_d.di_gid = current_fsgid(cr);
1131 ip->i_d.di_projid = prid;
1132 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1133
1134 /*
1135 * If the superblock version is up to where we support new format
1136 * inodes and this is currently an old format inode, then change
1137 * the inode version number now. This way we only do the conversion
1138 * here rather than here and in the flush/logging code.
1139 */
1140 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1141 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1142 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1143 /*
1144 * We've already zeroed the old link count, the projid field,
1145 * and the pad field.
1146 */
1147 }
1148
1149 /*
1150 * Project ids won't be stored on disk if we are using a version 1 inode.
1151 */
1152 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1153 xfs_bump_ino_vers2(tp, ip);
1154
1155 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1156 ip->i_d.di_gid = pip->i_d.di_gid;
1157 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1158 ip->i_d.di_mode |= S_ISGID;
1159 }
1160 }
1161
1162 /*
1163 * If the group ID of the new file does not match the effective group
1164 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1165 * (and only if the irix_sgid_inherit compatibility variable is set).
1166 */
1167 if ((irix_sgid_inherit) &&
1168 (ip->i_d.di_mode & S_ISGID) &&
1169 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1170 ip->i_d.di_mode &= ~S_ISGID;
1171 }
1172
1173 ip->i_d.di_size = 0;
1174 ip->i_d.di_nextents = 0;
1175 ASSERT(ip->i_d.di_nblocks == 0);
1176 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1177 /*
1178 * di_gen will have been taken care of in xfs_iread.
1179 */
1180 ip->i_d.di_extsize = 0;
1181 ip->i_d.di_dmevmask = 0;
1182 ip->i_d.di_dmstate = 0;
1183 ip->i_d.di_flags = 0;
1184 flags = XFS_ILOG_CORE;
1185 switch (mode & S_IFMT) {
1186 case S_IFIFO:
1187 case S_IFCHR:
1188 case S_IFBLK:
1189 case S_IFSOCK:
1190 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1191 ip->i_df.if_u2.if_rdev = rdev;
1192 ip->i_df.if_flags = 0;
1193 flags |= XFS_ILOG_DEV;
1194 break;
1195 case S_IFREG:
1196 case S_IFDIR:
1197 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Nathan Scott365ca832005-06-21 15:39:12 +10001198 uint di_flags = 0;
1199
1200 if ((mode & S_IFMT) == S_IFDIR) {
1201 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1202 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001203 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1204 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1205 ip->i_d.di_extsize = pip->i_d.di_extsize;
1206 }
1207 } else if ((mode & S_IFMT) == S_IFREG) {
Nathan Scott365ca832005-06-21 15:39:12 +10001208 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1209 di_flags |= XFS_DIFLAG_REALTIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1211 }
Nathan Scottdd9f4382006-01-11 15:28:28 +11001212 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1213 di_flags |= XFS_DIFLAG_EXTSIZE;
1214 ip->i_d.di_extsize = pip->i_d.di_extsize;
1215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 }
1217 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1218 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +10001219 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1221 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +10001222 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1224 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +10001225 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1227 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +10001228 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1229 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1230 di_flags |= XFS_DIFLAG_PROJINHERIT;
Barry Naujokd3446ea2006-06-09 14:54:19 +10001231 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1232 xfs_inherit_nodefrag)
1233 di_flags |= XFS_DIFLAG_NODEFRAG;
Nathan Scott365ca832005-06-21 15:39:12 +10001234 ip->i_d.di_flags |= di_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 }
1236 /* FALLTHROUGH */
1237 case S_IFLNK:
1238 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1239 ip->i_df.if_flags = XFS_IFEXTENTS;
1240 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1241 ip->i_df.if_u1.if_extents = NULL;
1242 break;
1243 default:
1244 ASSERT(0);
1245 }
1246 /*
1247 * Attribute fork settings for new inode.
1248 */
1249 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1250 ip->i_d.di_anextents = 0;
1251
1252 /*
1253 * Log the new values stuffed into the inode.
1254 */
1255 xfs_trans_log_inode(tp, ip, flags);
1256
Nathan Scottb83bd132006-06-09 16:48:30 +10001257 /* now that we have an i_mode we can setup inode ops and unlock */
1258 bhv_vfs_init_vnode(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
1260 *ipp = ip;
1261 return 0;
1262}
1263
1264/*
1265 * Check to make sure that there are no blocks allocated to the
1266 * file beyond the size of the file. We don't check this for
1267 * files with fixed size extents or real time extents, but we
1268 * at least do it for regular files.
1269 */
1270#ifdef DEBUG
1271void
1272xfs_isize_check(
1273 xfs_mount_t *mp,
1274 xfs_inode_t *ip,
1275 xfs_fsize_t isize)
1276{
1277 xfs_fileoff_t map_first;
1278 int nimaps;
1279 xfs_bmbt_irec_t imaps[2];
1280
1281 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1282 return;
1283
Nathan Scottdd9f4382006-01-11 15:28:28 +11001284 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 return;
1286
1287 nimaps = 2;
1288 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1289 /*
1290 * The filesystem could be shutting down, so bmapi may return
1291 * an error.
1292 */
1293 if (xfs_bmapi(NULL, ip, map_first,
1294 (XFS_B_TO_FSB(mp,
1295 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1296 map_first),
1297 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001298 NULL, NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 return;
1300 ASSERT(nimaps == 1);
1301 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1302}
1303#endif /* DEBUG */
1304
1305/*
1306 * Calculate the last possible buffered byte in a file. This must
1307 * include data that was buffered beyond the EOF by the write code.
1308 * This also needs to deal with overflowing the xfs_fsize_t type
1309 * which can happen for sizes near the limit.
1310 *
1311 * We also need to take into account any blocks beyond the EOF. It
1312 * may be the case that they were buffered by a write which failed.
1313 * In that case the pages will still be in memory, but the inode size
1314 * will never have been updated.
1315 */
1316xfs_fsize_t
1317xfs_file_last_byte(
1318 xfs_inode_t *ip)
1319{
1320 xfs_mount_t *mp;
1321 xfs_fsize_t last_byte;
1322 xfs_fileoff_t last_block;
1323 xfs_fileoff_t size_last_block;
1324 int error;
1325
1326 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1327
1328 mp = ip->i_mount;
1329 /*
1330 * Only check for blocks beyond the EOF if the extents have
1331 * been read in. This eliminates the need for the inode lock,
1332 * and it also saves us from looking when it really isn't
1333 * necessary.
1334 */
1335 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1336 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1337 XFS_DATA_FORK);
1338 if (error) {
1339 last_block = 0;
1340 }
1341 } else {
1342 last_block = 0;
1343 }
1344 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1345 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1346
1347 last_byte = XFS_FSB_TO_B(mp, last_block);
1348 if (last_byte < 0) {
1349 return XFS_MAXIOFFSET(mp);
1350 }
1351 last_byte += (1 << mp->m_writeio_log);
1352 if (last_byte < 0) {
1353 return XFS_MAXIOFFSET(mp);
1354 }
1355 return last_byte;
1356}
1357
1358#if defined(XFS_RW_TRACE)
1359STATIC void
1360xfs_itrunc_trace(
1361 int tag,
1362 xfs_inode_t *ip,
1363 int flag,
1364 xfs_fsize_t new_size,
1365 xfs_off_t toss_start,
1366 xfs_off_t toss_finish)
1367{
1368 if (ip->i_rwtrace == NULL) {
1369 return;
1370 }
1371
1372 ktrace_enter(ip->i_rwtrace,
1373 (void*)((long)tag),
1374 (void*)ip,
1375 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1376 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1377 (void*)((long)flag),
1378 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1379 (void*)(unsigned long)(new_size & 0xffffffff),
1380 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1381 (void*)(unsigned long)(toss_start & 0xffffffff),
1382 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1383 (void*)(unsigned long)(toss_finish & 0xffffffff),
1384 (void*)(unsigned long)current_cpu(),
Yingping Luf1fdc842006-03-22 12:44:15 +11001385 (void*)(unsigned long)current_pid(),
1386 (void*)NULL,
1387 (void*)NULL,
1388 (void*)NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389}
1390#else
1391#define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1392#endif
1393
1394/*
1395 * Start the truncation of the file to new_size. The new size
1396 * must be smaller than the current size. This routine will
1397 * clear the buffer and page caches of file data in the removed
1398 * range, and xfs_itruncate_finish() will remove the underlying
1399 * disk blocks.
1400 *
1401 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1402 * must NOT have the inode lock held at all. This is because we're
1403 * calling into the buffer/page cache code and we can't hold the
1404 * inode lock when we do so.
1405 *
David Chinner38e22992006-03-22 12:47:15 +11001406 * We need to wait for any direct I/Os in flight to complete before we
1407 * proceed with the truncate. This is needed to prevent the extents
1408 * being read or written by the direct I/Os from being removed while the
1409 * I/O is in flight as there is no other method of synchronising
1410 * direct I/O with the truncate operation. Also, because we hold
1411 * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
1412 * started until the truncate completes and drops the lock. Essentially,
1413 * the vn_iowait() call forms an I/O barrier that provides strict ordering
1414 * between direct I/Os and the truncate operation.
1415 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1417 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1418 * in the case that the caller is locking things out of order and
1419 * may not be able to call xfs_itruncate_finish() with the inode lock
1420 * held without dropping the I/O lock. If the caller must drop the
1421 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1422 * must be called again with all the same restrictions as the initial
1423 * call.
1424 */
1425void
1426xfs_itruncate_start(
1427 xfs_inode_t *ip,
1428 uint flags,
1429 xfs_fsize_t new_size)
1430{
1431 xfs_fsize_t last_byte;
1432 xfs_off_t toss_start;
1433 xfs_mount_t *mp;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001434 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1437 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1438 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1439 (flags == XFS_ITRUNC_MAYBE));
1440
1441 mp = ip->i_mount;
1442 vp = XFS_ITOV(ip);
Yingping Lu9fa80462006-03-22 12:44:35 +11001443
1444 vn_iowait(vp); /* wait for the completion of any pending DIOs */
1445
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 /*
Nathan Scott67fcaa72006-06-09 17:00:52 +10001447 * Call toss_pages or flushinval_pages to get rid of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 * overlapping the region being removed. We have to use
Nathan Scott67fcaa72006-06-09 17:00:52 +10001449 * the less efficient flushinval_pages in the case that the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 * caller may not be able to finish the truncate without
1451 * dropping the inode's I/O lock. Make sure
1452 * to catch any pages brought in by buffers overlapping
1453 * the EOF by searching out beyond the isize by our
1454 * block size. We round new_size up to a block boundary
1455 * so that we don't toss things on the same block as
1456 * new_size but before it.
1457 *
Nathan Scott67fcaa72006-06-09 17:00:52 +10001458 * Before calling toss_page or flushinval_pages, make sure to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 * call remapf() over the same region if the file is mapped.
1460 * This frees up mapped file references to the pages in the
Nathan Scott67fcaa72006-06-09 17:00:52 +10001461 * given range and for the flushinval_pages case it ensures
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 * that we get the latest mapped changes flushed out.
1463 */
1464 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1465 toss_start = XFS_FSB_TO_B(mp, toss_start);
1466 if (toss_start < 0) {
1467 /*
1468 * The place to start tossing is beyond our maximum
1469 * file size, so there is no way that the data extended
1470 * out there.
1471 */
1472 return;
1473 }
1474 last_byte = xfs_file_last_byte(ip);
1475 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1476 last_byte);
1477 if (last_byte > toss_start) {
1478 if (flags & XFS_ITRUNC_DEFINITE) {
Nathan Scott67fcaa72006-06-09 17:00:52 +10001479 bhv_vop_toss_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 } else {
Nathan Scott67fcaa72006-06-09 17:00:52 +10001481 bhv_vop_flushinval_pages(vp, toss_start, -1, FI_REMAPF_LOCKED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 }
1483 }
1484
1485#ifdef DEBUG
1486 if (new_size == 0) {
1487 ASSERT(VN_CACHED(vp) == 0);
1488 }
1489#endif
1490}
1491
1492/*
1493 * Shrink the file to the given new_size. The new
1494 * size must be smaller than the current size.
1495 * This will free up the underlying blocks
1496 * in the removed range after a call to xfs_itruncate_start()
1497 * or xfs_atruncate_start().
1498 *
1499 * The transaction passed to this routine must have made
1500 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1501 * This routine may commit the given transaction and
1502 * start new ones, so make sure everything involved in
1503 * the transaction is tidy before calling here.
1504 * Some transaction will be returned to the caller to be
1505 * committed. The incoming transaction must already include
1506 * the inode, and both inode locks must be held exclusively.
1507 * The inode must also be "held" within the transaction. On
1508 * return the inode will be "held" within the returned transaction.
1509 * This routine does NOT require any disk space to be reserved
1510 * for it within the transaction.
1511 *
1512 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1513 * and it indicates the fork which is to be truncated. For the
1514 * attribute fork we only support truncation to size 0.
1515 *
1516 * We use the sync parameter to indicate whether or not the first
1517 * transaction we perform might have to be synchronous. For the attr fork,
1518 * it needs to be so if the unlink of the inode is not yet known to be
1519 * permanent in the log. This keeps us from freeing and reusing the
1520 * blocks of the attribute fork before the unlink of the inode becomes
1521 * permanent.
1522 *
1523 * For the data fork, we normally have to run synchronously if we're
1524 * being called out of the inactive path or we're being called
1525 * out of the create path where we're truncating an existing file.
1526 * Either way, the truncate needs to be sync so blocks don't reappear
1527 * in the file with altered data in case of a crash. wsync filesystems
1528 * can run the first case async because anything that shrinks the inode
1529 * has to run sync so by the time we're called here from inactive, the
1530 * inode size is permanently set to 0.
1531 *
1532 * Calls from the truncate path always need to be sync unless we're
1533 * in a wsync filesystem and the file has already been unlinked.
1534 *
1535 * The caller is responsible for correctly setting the sync parameter.
1536 * It gets too hard for us to guess here which path we're being called
1537 * out of just based on inode state.
1538 */
1539int
1540xfs_itruncate_finish(
1541 xfs_trans_t **tp,
1542 xfs_inode_t *ip,
1543 xfs_fsize_t new_size,
1544 int fork,
1545 int sync)
1546{
1547 xfs_fsblock_t first_block;
1548 xfs_fileoff_t first_unmap_block;
1549 xfs_fileoff_t last_block;
1550 xfs_filblks_t unmap_len=0;
1551 xfs_mount_t *mp;
1552 xfs_trans_t *ntp;
1553 int done;
1554 int committed;
1555 xfs_bmap_free_t free_list;
1556 int error;
1557
1558 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1559 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1560 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1561 ASSERT(*tp != NULL);
1562 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1563 ASSERT(ip->i_transp == *tp);
1564 ASSERT(ip->i_itemp != NULL);
1565 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1566
1567
1568 ntp = *tp;
1569 mp = (ntp)->t_mountp;
1570 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1571
1572 /*
1573 * We only support truncating the entire attribute fork.
1574 */
1575 if (fork == XFS_ATTR_FORK) {
1576 new_size = 0LL;
1577 }
1578 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1579 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1580 /*
1581 * The first thing we do is set the size to new_size permanently
1582 * on disk. This way we don't have to worry about anyone ever
1583 * being able to look at the data being freed even in the face
1584 * of a crash. What we're getting around here is the case where
1585 * we free a block, it is allocated to another file, it is written
1586 * to, and then we crash. If the new data gets written to the
1587 * file but the log buffers containing the free and reallocation
1588 * don't, then we'd end up with garbage in the blocks being freed.
1589 * As long as we make the new_size permanent before actually
1590 * freeing any blocks it doesn't matter if they get writtten to.
1591 *
1592 * The callers must signal into us whether or not the size
1593 * setting here must be synchronous. There are a few cases
1594 * where it doesn't have to be synchronous. Those cases
1595 * occur if the file is unlinked and we know the unlink is
1596 * permanent or if the blocks being truncated are guaranteed
1597 * to be beyond the inode eof (regardless of the link count)
1598 * and the eof value is permanent. Both of these cases occur
1599 * only on wsync-mounted filesystems. In those cases, we're
1600 * guaranteed that no user will ever see the data in the blocks
1601 * that are being truncated so the truncate can run async.
1602 * In the free beyond eof case, the file may wind up with
1603 * more blocks allocated to it than it needs if we crash
1604 * and that won't get fixed until the next time the file
1605 * is re-opened and closed but that's ok as that shouldn't
1606 * be too many blocks.
1607 *
1608 * However, we can't just make all wsync xactions run async
1609 * because there's one call out of the create path that needs
1610 * to run sync where it's truncating an existing file to size
1611 * 0 whose size is > 0.
1612 *
1613 * It's probably possible to come up with a test in this
1614 * routine that would correctly distinguish all the above
1615 * cases from the values of the function parameters and the
1616 * inode state but for sanity's sake, I've decided to let the
1617 * layers above just tell us. It's simpler to correctly figure
1618 * out in the layer above exactly under what conditions we
1619 * can run async and I think it's easier for others read and
1620 * follow the logic in case something has to be changed.
1621 * cscope is your friend -- rcc.
1622 *
1623 * The attribute fork is much simpler.
1624 *
1625 * For the attribute fork we allow the caller to tell us whether
1626 * the unlink of the inode that led to this call is yet permanent
1627 * in the on disk log. If it is not and we will be freeing extents
1628 * in this inode then we make the first transaction synchronous
1629 * to make sure that the unlink is permanent by the time we free
1630 * the blocks.
1631 */
1632 if (fork == XFS_DATA_FORK) {
1633 if (ip->i_d.di_nextents > 0) {
1634 ip->i_d.di_size = new_size;
1635 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1636 }
1637 } else if (sync) {
1638 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1639 if (ip->i_d.di_anextents > 0)
1640 xfs_trans_set_sync(ntp);
1641 }
1642 ASSERT(fork == XFS_DATA_FORK ||
1643 (fork == XFS_ATTR_FORK &&
1644 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1645 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1646
1647 /*
1648 * Since it is possible for space to become allocated beyond
1649 * the end of the file (in a crash where the space is allocated
1650 * but the inode size is not yet updated), simply remove any
1651 * blocks which show up between the new EOF and the maximum
1652 * possible file size. If the first block to be removed is
1653 * beyond the maximum file size (ie it is the same as last_block),
1654 * then there is nothing to do.
1655 */
1656 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1657 ASSERT(first_unmap_block <= last_block);
1658 done = 0;
1659 if (last_block == first_unmap_block) {
1660 done = 1;
1661 } else {
1662 unmap_len = last_block - first_unmap_block + 1;
1663 }
1664 while (!done) {
1665 /*
1666 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1667 * will tell us whether it freed the entire range or
1668 * not. If this is a synchronous mount (wsync),
1669 * then we can tell bunmapi to keep all the
1670 * transactions asynchronous since the unlink
1671 * transaction that made this inode inactive has
1672 * already hit the disk. There's no danger of
1673 * the freed blocks being reused, there being a
1674 * crash, and the reused blocks suddenly reappearing
1675 * in this file with garbage in them once recovery
1676 * runs.
1677 */
1678 XFS_BMAP_INIT(&free_list, &first_block);
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001679 error = XFS_BUNMAPI(mp, ntp, &ip->i_iocore,
1680 first_unmap_block, unmap_len,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 XFS_BMAPI_AFLAG(fork) |
1682 (sync ? 0 : XFS_BMAPI_ASYNC),
1683 XFS_ITRUNC_MAX_EXTENTS,
Olaf Weber3e57ecf2006-06-09 14:48:12 +10001684 &first_block, &free_list,
1685 NULL, &done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 if (error) {
1687 /*
1688 * If the bunmapi call encounters an error,
1689 * return to the caller where the transaction
1690 * can be properly aborted. We just need to
1691 * make sure we're not holding any resources
1692 * that we were not when we came in.
1693 */
1694 xfs_bmap_cancel(&free_list);
1695 return error;
1696 }
1697
1698 /*
1699 * Duplicate the transaction that has the permanent
1700 * reservation and commit the old transaction.
1701 */
1702 error = xfs_bmap_finish(tp, &free_list, first_block,
1703 &committed);
1704 ntp = *tp;
1705 if (error) {
1706 /*
1707 * If the bmap finish call encounters an error,
1708 * return to the caller where the transaction
1709 * can be properly aborted. We just need to
1710 * make sure we're not holding any resources
1711 * that we were not when we came in.
1712 *
1713 * Aborting from this point might lose some
1714 * blocks in the file system, but oh well.
1715 */
1716 xfs_bmap_cancel(&free_list);
1717 if (committed) {
1718 /*
1719 * If the passed in transaction committed
1720 * in xfs_bmap_finish(), then we want to
1721 * add the inode to this one before returning.
1722 * This keeps things simple for the higher
1723 * level code, because it always knows that
1724 * the inode is locked and held in the
1725 * transaction that returns to it whether
1726 * errors occur or not. We don't mark the
1727 * inode dirty so that this transaction can
1728 * be easily aborted if possible.
1729 */
1730 xfs_trans_ijoin(ntp, ip,
1731 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1732 xfs_trans_ihold(ntp, ip);
1733 }
1734 return error;
1735 }
1736
1737 if (committed) {
1738 /*
1739 * The first xact was committed,
1740 * so add the inode to the new one.
1741 * Mark it dirty so it will be logged
1742 * and moved forward in the log as
1743 * part of every commit.
1744 */
1745 xfs_trans_ijoin(ntp, ip,
1746 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1747 xfs_trans_ihold(ntp, ip);
1748 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1749 }
1750 ntp = xfs_trans_dup(ntp);
1751 (void) xfs_trans_commit(*tp, 0, NULL);
1752 *tp = ntp;
1753 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1754 XFS_TRANS_PERM_LOG_RES,
1755 XFS_ITRUNCATE_LOG_COUNT);
1756 /*
1757 * Add the inode being truncated to the next chained
1758 * transaction.
1759 */
1760 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1761 xfs_trans_ihold(ntp, ip);
1762 if (error)
1763 return (error);
1764 }
1765 /*
1766 * Only update the size in the case of the data fork, but
1767 * always re-log the inode so that our permanent transaction
1768 * can keep on rolling it forward in the log.
1769 */
1770 if (fork == XFS_DATA_FORK) {
1771 xfs_isize_check(mp, ip, new_size);
1772 ip->i_d.di_size = new_size;
1773 }
1774 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1775 ASSERT((new_size != 0) ||
1776 (fork == XFS_ATTR_FORK) ||
1777 (ip->i_delayed_blks == 0));
1778 ASSERT((new_size != 0) ||
1779 (fork == XFS_ATTR_FORK) ||
1780 (ip->i_d.di_nextents == 0));
1781 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1782 return 0;
1783}
1784
1785
1786/*
1787 * xfs_igrow_start
1788 *
1789 * Do the first part of growing a file: zero any data in the last
1790 * block that is beyond the old EOF. We need to do this before
1791 * the inode is joined to the transaction to modify the i_size.
1792 * That way we can drop the inode lock and call into the buffer
1793 * cache to get the buffer mapping the EOF.
1794 */
1795int
1796xfs_igrow_start(
1797 xfs_inode_t *ip,
1798 xfs_fsize_t new_size,
1799 cred_t *credp)
1800{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 int error;
1802
1803 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1804 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1805 ASSERT(new_size > ip->i_d.di_size);
1806
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 /*
1808 * Zero any pages that may have been created by
1809 * xfs_write_file() beyond the end of the file
1810 * and any blocks between the old and new file sizes.
1811 */
Eric Sandeen24ee8082006-01-11 15:34:32 +11001812 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1813 ip->i_d.di_size, new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 return error;
1815}
1816
1817/*
1818 * xfs_igrow_finish
1819 *
1820 * This routine is called to extend the size of a file.
1821 * The inode must have both the iolock and the ilock locked
1822 * for update and it must be a part of the current transaction.
1823 * The xfs_igrow_start() function must have been called previously.
1824 * If the change_flag is not zero, the inode change timestamp will
1825 * be updated.
1826 */
1827void
1828xfs_igrow_finish(
1829 xfs_trans_t *tp,
1830 xfs_inode_t *ip,
1831 xfs_fsize_t new_size,
1832 int change_flag)
1833{
1834 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1835 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1836 ASSERT(ip->i_transp == tp);
1837 ASSERT(new_size > ip->i_d.di_size);
1838
1839 /*
1840 * Update the file size. Update the inode change timestamp
1841 * if change_flag set.
1842 */
1843 ip->i_d.di_size = new_size;
1844 if (change_flag)
1845 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1846 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1847
1848}
1849
1850
1851/*
1852 * This is called when the inode's link count goes to 0.
1853 * We place the on-disk inode on a list in the AGI. It
1854 * will be pulled from this list when the inode is freed.
1855 */
1856int
1857xfs_iunlink(
1858 xfs_trans_t *tp,
1859 xfs_inode_t *ip)
1860{
1861 xfs_mount_t *mp;
1862 xfs_agi_t *agi;
1863 xfs_dinode_t *dip;
1864 xfs_buf_t *agibp;
1865 xfs_buf_t *ibp;
1866 xfs_agnumber_t agno;
1867 xfs_daddr_t agdaddr;
1868 xfs_agino_t agino;
1869 short bucket_index;
1870 int offset;
1871 int error;
1872 int agi_ok;
1873
1874 ASSERT(ip->i_d.di_nlink == 0);
1875 ASSERT(ip->i_d.di_mode != 0);
1876 ASSERT(ip->i_transp == tp);
1877
1878 mp = tp->t_mountp;
1879
1880 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1881 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1882
1883 /*
1884 * Get the agi buffer first. It ensures lock ordering
1885 * on the list.
1886 */
1887 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1888 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1889 if (error) {
1890 return error;
1891 }
1892 /*
1893 * Validate the magic number of the agi block.
1894 */
1895 agi = XFS_BUF_TO_AGI(agibp);
1896 agi_ok =
Christoph Hellwig16259e72005-11-02 15:11:25 +11001897 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1898 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1900 XFS_RANDOM_IUNLINK))) {
1901 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1902 xfs_trans_brelse(tp, agibp);
1903 return XFS_ERROR(EFSCORRUPTED);
1904 }
1905 /*
1906 * Get the index into the agi hash table for the
1907 * list this inode will go on.
1908 */
1909 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1910 ASSERT(agino != 0);
1911 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1912 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001913 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
Christoph Hellwig16259e72005-11-02 15:11:25 +11001915 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 /*
1917 * There is already another inode in the bucket we need
1918 * to add ourselves to. Add us at the front of the list.
1919 * Here we put the head pointer into our next pointer,
1920 * and then we fall through to point the head at us.
1921 */
Nathan Scottb12dd342006-03-17 17:26:04 +11001922 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 if (error) {
1924 return error;
1925 }
1926 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1927 ASSERT(dip->di_next_unlinked);
1928 /* both on-disk, don't endian flip twice */
1929 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1930 offset = ip->i_boffset +
1931 offsetof(xfs_dinode_t, di_next_unlinked);
1932 xfs_trans_inode_buf(tp, ibp);
1933 xfs_trans_log_buf(tp, ibp, offset,
1934 (offset + sizeof(xfs_agino_t) - 1));
1935 xfs_inobp_check(mp, ibp);
1936 }
1937
1938 /*
1939 * Point the bucket head pointer at the inode being inserted.
1940 */
1941 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001942 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 offset = offsetof(xfs_agi_t, agi_unlinked) +
1944 (sizeof(xfs_agino_t) * bucket_index);
1945 xfs_trans_log_buf(tp, agibp, offset,
1946 (offset + sizeof(xfs_agino_t) - 1));
1947 return 0;
1948}
1949
1950/*
1951 * Pull the on-disk inode from the AGI unlinked list.
1952 */
1953STATIC int
1954xfs_iunlink_remove(
1955 xfs_trans_t *tp,
1956 xfs_inode_t *ip)
1957{
1958 xfs_ino_t next_ino;
1959 xfs_mount_t *mp;
1960 xfs_agi_t *agi;
1961 xfs_dinode_t *dip;
1962 xfs_buf_t *agibp;
1963 xfs_buf_t *ibp;
1964 xfs_agnumber_t agno;
1965 xfs_daddr_t agdaddr;
1966 xfs_agino_t agino;
1967 xfs_agino_t next_agino;
1968 xfs_buf_t *last_ibp;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001969 xfs_dinode_t *last_dip = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 short bucket_index;
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001971 int offset, last_offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 int error;
1973 int agi_ok;
1974
1975 /*
1976 * First pull the on-disk inode from the AGI unlinked list.
1977 */
1978 mp = tp->t_mountp;
1979
1980 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1981 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1982
1983 /*
1984 * Get the agi buffer first. It ensures lock ordering
1985 * on the list.
1986 */
1987 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1988 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1989 if (error) {
1990 cmn_err(CE_WARN,
1991 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1992 error, mp->m_fsname);
1993 return error;
1994 }
1995 /*
1996 * Validate the magic number of the agi block.
1997 */
1998 agi = XFS_BUF_TO_AGI(agibp);
1999 agi_ok =
Christoph Hellwig16259e72005-11-02 15:11:25 +11002000 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
2001 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
2003 XFS_RANDOM_IUNLINK_REMOVE))) {
2004 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
2005 mp, agi);
2006 xfs_trans_brelse(tp, agibp);
2007 cmn_err(CE_WARN,
2008 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
2009 mp->m_fsname);
2010 return XFS_ERROR(EFSCORRUPTED);
2011 }
2012 /*
2013 * Get the index into the agi hash table for the
2014 * list this inode will go on.
2015 */
2016 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2017 ASSERT(agino != 0);
2018 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig16259e72005-11-02 15:11:25 +11002019 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 ASSERT(agi->agi_unlinked[bucket_index]);
2021
Christoph Hellwig16259e72005-11-02 15:11:25 +11002022 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 /*
2024 * We're at the head of the list. Get the inode's
2025 * on-disk buffer to see if there is anyone after us
2026 * on the list. Only modify our next pointer if it
2027 * is not already NULLAGINO. This saves us the overhead
2028 * of dealing with the buffer when there is no need to
2029 * change it.
2030 */
Nathan Scottb12dd342006-03-17 17:26:04 +11002031 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 if (error) {
2033 cmn_err(CE_WARN,
2034 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2035 error, mp->m_fsname);
2036 return error;
2037 }
2038 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2039 ASSERT(next_agino != 0);
2040 if (next_agino != NULLAGINO) {
2041 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2042 offset = ip->i_boffset +
2043 offsetof(xfs_dinode_t, di_next_unlinked);
2044 xfs_trans_inode_buf(tp, ibp);
2045 xfs_trans_log_buf(tp, ibp, offset,
2046 (offset + sizeof(xfs_agino_t) - 1));
2047 xfs_inobp_check(mp, ibp);
2048 } else {
2049 xfs_trans_brelse(tp, ibp);
2050 }
2051 /*
2052 * Point the bucket head pointer at the next inode.
2053 */
2054 ASSERT(next_agino != 0);
2055 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002056 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 offset = offsetof(xfs_agi_t, agi_unlinked) +
2058 (sizeof(xfs_agino_t) * bucket_index);
2059 xfs_trans_log_buf(tp, agibp, offset,
2060 (offset + sizeof(xfs_agino_t) - 1));
2061 } else {
2062 /*
2063 * We need to search the list for the inode being freed.
2064 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002065 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 last_ibp = NULL;
2067 while (next_agino != agino) {
2068 /*
2069 * If the last inode wasn't the one pointing to
2070 * us, then release its buffer since we're not
2071 * going to do anything with it.
2072 */
2073 if (last_ibp != NULL) {
2074 xfs_trans_brelse(tp, last_ibp);
2075 }
2076 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2077 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2078 &last_ibp, &last_offset);
2079 if (error) {
2080 cmn_err(CE_WARN,
2081 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2082 error, mp->m_fsname);
2083 return error;
2084 }
2085 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2086 ASSERT(next_agino != NULLAGINO);
2087 ASSERT(next_agino != 0);
2088 }
2089 /*
2090 * Now last_ibp points to the buffer previous to us on
2091 * the unlinked list. Pull us from the list.
2092 */
Nathan Scottb12dd342006-03-17 17:26:04 +11002093 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 if (error) {
2095 cmn_err(CE_WARN,
2096 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2097 error, mp->m_fsname);
2098 return error;
2099 }
2100 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2101 ASSERT(next_agino != 0);
2102 ASSERT(next_agino != agino);
2103 if (next_agino != NULLAGINO) {
2104 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2105 offset = ip->i_boffset +
2106 offsetof(xfs_dinode_t, di_next_unlinked);
2107 xfs_trans_inode_buf(tp, ibp);
2108 xfs_trans_log_buf(tp, ibp, offset,
2109 (offset + sizeof(xfs_agino_t) - 1));
2110 xfs_inobp_check(mp, ibp);
2111 } else {
2112 xfs_trans_brelse(tp, ibp);
2113 }
2114 /*
2115 * Point the previous inode on the list to the next inode.
2116 */
2117 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2118 ASSERT(next_agino != 0);
2119 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2120 xfs_trans_inode_buf(tp, last_ibp);
2121 xfs_trans_log_buf(tp, last_ibp, offset,
2122 (offset + sizeof(xfs_agino_t) - 1));
2123 xfs_inobp_check(mp, last_ibp);
2124 }
2125 return 0;
2126}
2127
2128static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2129{
2130 return (((ip->i_itemp == NULL) ||
2131 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2132 (ip->i_update_core == 0));
2133}
2134
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002135STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136xfs_ifree_cluster(
2137 xfs_inode_t *free_ip,
2138 xfs_trans_t *tp,
2139 xfs_ino_t inum)
2140{
2141 xfs_mount_t *mp = free_ip->i_mount;
2142 int blks_per_cluster;
2143 int nbufs;
2144 int ninodes;
2145 int i, j, found, pre_flushed;
2146 xfs_daddr_t blkno;
2147 xfs_buf_t *bp;
2148 xfs_ihash_t *ih;
2149 xfs_inode_t *ip, **ip_found;
2150 xfs_inode_log_item_t *iip;
2151 xfs_log_item_t *lip;
2152 SPLDECL(s);
2153
2154 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2155 blks_per_cluster = 1;
2156 ninodes = mp->m_sb.sb_inopblock;
2157 nbufs = XFS_IALLOC_BLOCKS(mp);
2158 } else {
2159 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2160 mp->m_sb.sb_blocksize;
2161 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2162 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2163 }
2164
2165 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2166
2167 for (j = 0; j < nbufs; j++, inum += ninodes) {
2168 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2169 XFS_INO_TO_AGBNO(mp, inum));
2170
2171
2172 /*
2173 * Look for each inode in memory and attempt to lock it,
2174 * we can be racing with flush and tail pushing here.
2175 * any inode we get the locks on, add to an array of
2176 * inode items to process later.
2177 *
2178 * The get the buffer lock, we could beat a flush
2179 * or tail pushing thread to the lock here, in which
2180 * case they will go looking for the inode buffer
2181 * and fail, we need some other form of interlock
2182 * here.
2183 */
2184 found = 0;
2185 for (i = 0; i < ninodes; i++) {
2186 ih = XFS_IHASH(mp, inum + i);
2187 read_lock(&ih->ih_lock);
2188 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2189 if (ip->i_ino == inum + i)
2190 break;
2191 }
2192
2193 /* Inode not in memory or we found it already,
2194 * nothing to do
2195 */
2196 if (!ip || (ip->i_flags & XFS_ISTALE)) {
2197 read_unlock(&ih->ih_lock);
2198 continue;
2199 }
2200
2201 if (xfs_inode_clean(ip)) {
2202 read_unlock(&ih->ih_lock);
2203 continue;
2204 }
2205
2206 /* If we can get the locks then add it to the
2207 * list, otherwise by the time we get the bp lock
2208 * below it will already be attached to the
2209 * inode buffer.
2210 */
2211
2212 /* This inode will already be locked - by us, lets
2213 * keep it that way.
2214 */
2215
2216 if (ip == free_ip) {
2217 if (xfs_iflock_nowait(ip)) {
David Chinnerf273ab82006-09-28 11:06:03 +10002218 spin_lock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 ip->i_flags |= XFS_ISTALE;
David Chinnerf273ab82006-09-28 11:06:03 +10002220 spin_unlock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
2222 if (xfs_inode_clean(ip)) {
2223 xfs_ifunlock(ip);
2224 } else {
2225 ip_found[found++] = ip;
2226 }
2227 }
2228 read_unlock(&ih->ih_lock);
2229 continue;
2230 }
2231
2232 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2233 if (xfs_iflock_nowait(ip)) {
David Chinnerf273ab82006-09-28 11:06:03 +10002234 spin_lock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 ip->i_flags |= XFS_ISTALE;
David Chinnerf273ab82006-09-28 11:06:03 +10002236 spin_unlock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
2238 if (xfs_inode_clean(ip)) {
2239 xfs_ifunlock(ip);
2240 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2241 } else {
2242 ip_found[found++] = ip;
2243 }
2244 } else {
2245 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2246 }
2247 }
2248
2249 read_unlock(&ih->ih_lock);
2250 }
2251
2252 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2253 mp->m_bsize * blks_per_cluster,
2254 XFS_BUF_LOCK);
2255
2256 pre_flushed = 0;
2257 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2258 while (lip) {
2259 if (lip->li_type == XFS_LI_INODE) {
2260 iip = (xfs_inode_log_item_t *)lip;
2261 ASSERT(iip->ili_logged == 1);
2262 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2263 AIL_LOCK(mp,s);
2264 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2265 AIL_UNLOCK(mp, s);
David Chinnerf273ab82006-09-28 11:06:03 +10002266 spin_lock(&iip->ili_inode->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 iip->ili_inode->i_flags |= XFS_ISTALE;
David Chinnerf273ab82006-09-28 11:06:03 +10002268 spin_unlock(&iip->ili_inode->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 pre_flushed++;
2270 }
2271 lip = lip->li_bio_list;
2272 }
2273
2274 for (i = 0; i < found; i++) {
2275 ip = ip_found[i];
2276 iip = ip->i_itemp;
2277
2278 if (!iip) {
2279 ip->i_update_core = 0;
2280 xfs_ifunlock(ip);
2281 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2282 continue;
2283 }
2284
2285 iip->ili_last_fields = iip->ili_format.ilf_fields;
2286 iip->ili_format.ilf_fields = 0;
2287 iip->ili_logged = 1;
2288 AIL_LOCK(mp,s);
2289 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2290 AIL_UNLOCK(mp, s);
2291
2292 xfs_buf_attach_iodone(bp,
2293 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2294 xfs_istale_done, (xfs_log_item_t *)iip);
2295 if (ip != free_ip) {
2296 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2297 }
2298 }
2299
2300 if (found || pre_flushed)
2301 xfs_trans_stale_inode_buf(tp, bp);
2302 xfs_trans_binval(tp, bp);
2303 }
2304
2305 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2306}
2307
2308/*
2309 * This is called to return an inode to the inode free list.
2310 * The inode should already be truncated to 0 length and have
2311 * no pages associated with it. This routine also assumes that
2312 * the inode is already a part of the transaction.
2313 *
2314 * The on-disk copy of the inode will have been added to the list
2315 * of unlinked inodes in the AGI. We need to remove the inode from
2316 * that list atomically with respect to freeing it here.
2317 */
2318int
2319xfs_ifree(
2320 xfs_trans_t *tp,
2321 xfs_inode_t *ip,
2322 xfs_bmap_free_t *flist)
2323{
2324 int error;
2325 int delete;
2326 xfs_ino_t first_ino;
2327
2328 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2329 ASSERT(ip->i_transp == tp);
2330 ASSERT(ip->i_d.di_nlink == 0);
2331 ASSERT(ip->i_d.di_nextents == 0);
2332 ASSERT(ip->i_d.di_anextents == 0);
2333 ASSERT((ip->i_d.di_size == 0) ||
2334 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2335 ASSERT(ip->i_d.di_nblocks == 0);
2336
2337 /*
2338 * Pull the on-disk inode from the AGI unlinked list.
2339 */
2340 error = xfs_iunlink_remove(tp, ip);
2341 if (error != 0) {
2342 return error;
2343 }
2344
2345 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2346 if (error != 0) {
2347 return error;
2348 }
2349 ip->i_d.di_mode = 0; /* mark incore inode as free */
2350 ip->i_d.di_flags = 0;
2351 ip->i_d.di_dmevmask = 0;
2352 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2353 ip->i_df.if_ext_max =
2354 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2355 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2356 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2357 /*
2358 * Bump the generation count so no one will be confused
2359 * by reincarnations of this inode.
2360 */
2361 ip->i_d.di_gen++;
2362 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2363
2364 if (delete) {
2365 xfs_ifree_cluster(ip, tp, first_ino);
2366 }
2367
2368 return 0;
2369}
2370
2371/*
2372 * Reallocate the space for if_broot based on the number of records
2373 * being added or deleted as indicated in rec_diff. Move the records
2374 * and pointers in if_broot to fit the new size. When shrinking this
2375 * will eliminate holes between the records and pointers created by
2376 * the caller. When growing this will create holes to be filled in
2377 * by the caller.
2378 *
2379 * The caller must not request to add more records than would fit in
2380 * the on-disk inode root. If the if_broot is currently NULL, then
2381 * if we adding records one will be allocated. The caller must also
2382 * not request that the number of records go below zero, although
2383 * it can go to zero.
2384 *
2385 * ip -- the inode whose if_broot area is changing
2386 * ext_diff -- the change in the number of records, positive or negative,
2387 * requested for the if_broot array.
2388 */
2389void
2390xfs_iroot_realloc(
2391 xfs_inode_t *ip,
2392 int rec_diff,
2393 int whichfork)
2394{
2395 int cur_max;
2396 xfs_ifork_t *ifp;
2397 xfs_bmbt_block_t *new_broot;
2398 int new_max;
2399 size_t new_size;
2400 char *np;
2401 char *op;
2402
2403 /*
2404 * Handle the degenerate case quietly.
2405 */
2406 if (rec_diff == 0) {
2407 return;
2408 }
2409
2410 ifp = XFS_IFORK_PTR(ip, whichfork);
2411 if (rec_diff > 0) {
2412 /*
2413 * If there wasn't any memory allocated before, just
2414 * allocate it now and get out.
2415 */
2416 if (ifp->if_broot_bytes == 0) {
2417 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2418 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2419 KM_SLEEP);
2420 ifp->if_broot_bytes = (int)new_size;
2421 return;
2422 }
2423
2424 /*
2425 * If there is already an existing if_broot, then we need
2426 * to realloc() it and shift the pointers to their new
2427 * location. The records don't change location because
2428 * they are kept butted up against the btree block header.
2429 */
2430 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2431 new_max = cur_max + rec_diff;
2432 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2433 ifp->if_broot = (xfs_bmbt_block_t *)
2434 kmem_realloc(ifp->if_broot,
2435 new_size,
2436 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2437 KM_SLEEP);
2438 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2439 ifp->if_broot_bytes);
2440 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2441 (int)new_size);
2442 ifp->if_broot_bytes = (int)new_size;
2443 ASSERT(ifp->if_broot_bytes <=
2444 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2445 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2446 return;
2447 }
2448
2449 /*
2450 * rec_diff is less than 0. In this case, we are shrinking the
2451 * if_broot buffer. It must already exist. If we go to zero
2452 * records, just get rid of the root and clear the status bit.
2453 */
2454 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2455 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2456 new_max = cur_max + rec_diff;
2457 ASSERT(new_max >= 0);
2458 if (new_max > 0)
2459 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2460 else
2461 new_size = 0;
2462 if (new_size > 0) {
2463 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2464 /*
2465 * First copy over the btree block header.
2466 */
2467 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2468 } else {
2469 new_broot = NULL;
2470 ifp->if_flags &= ~XFS_IFBROOT;
2471 }
2472
2473 /*
2474 * Only copy the records and pointers if there are any.
2475 */
2476 if (new_max > 0) {
2477 /*
2478 * First copy the records.
2479 */
2480 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2481 ifp->if_broot_bytes);
2482 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2483 (int)new_size);
2484 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2485
2486 /*
2487 * Then copy the pointers.
2488 */
2489 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2490 ifp->if_broot_bytes);
2491 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2492 (int)new_size);
2493 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2494 }
2495 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2496 ifp->if_broot = new_broot;
2497 ifp->if_broot_bytes = (int)new_size;
2498 ASSERT(ifp->if_broot_bytes <=
2499 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2500 return;
2501}
2502
2503
2504/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 * This is called when the amount of space needed for if_data
2506 * is increased or decreased. The change in size is indicated by
2507 * the number of bytes that need to be added or deleted in the
2508 * byte_diff parameter.
2509 *
2510 * If the amount of space needed has decreased below the size of the
2511 * inline buffer, then switch to using the inline buffer. Otherwise,
2512 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2513 * to what is needed.
2514 *
2515 * ip -- the inode whose if_data area is changing
2516 * byte_diff -- the change in the number of bytes, positive or negative,
2517 * requested for the if_data array.
2518 */
2519void
2520xfs_idata_realloc(
2521 xfs_inode_t *ip,
2522 int byte_diff,
2523 int whichfork)
2524{
2525 xfs_ifork_t *ifp;
2526 int new_size;
2527 int real_size;
2528
2529 if (byte_diff == 0) {
2530 return;
2531 }
2532
2533 ifp = XFS_IFORK_PTR(ip, whichfork);
2534 new_size = (int)ifp->if_bytes + byte_diff;
2535 ASSERT(new_size >= 0);
2536
2537 if (new_size == 0) {
2538 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2539 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2540 }
2541 ifp->if_u1.if_data = NULL;
2542 real_size = 0;
2543 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2544 /*
2545 * If the valid extents/data can fit in if_inline_ext/data,
2546 * copy them from the malloc'd vector and free it.
2547 */
2548 if (ifp->if_u1.if_data == NULL) {
2549 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2550 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2551 ASSERT(ifp->if_real_bytes != 0);
2552 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2553 new_size);
2554 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2555 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2556 }
2557 real_size = 0;
2558 } else {
2559 /*
2560 * Stuck with malloc/realloc.
2561 * For inline data, the underlying buffer must be
2562 * a multiple of 4 bytes in size so that it can be
2563 * logged and stay on word boundaries. We enforce
2564 * that here.
2565 */
2566 real_size = roundup(new_size, 4);
2567 if (ifp->if_u1.if_data == NULL) {
2568 ASSERT(ifp->if_real_bytes == 0);
2569 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2570 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2571 /*
2572 * Only do the realloc if the underlying size
2573 * is really changing.
2574 */
2575 if (ifp->if_real_bytes != real_size) {
2576 ifp->if_u1.if_data =
2577 kmem_realloc(ifp->if_u1.if_data,
2578 real_size,
2579 ifp->if_real_bytes,
2580 KM_SLEEP);
2581 }
2582 } else {
2583 ASSERT(ifp->if_real_bytes == 0);
2584 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2585 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2586 ifp->if_bytes);
2587 }
2588 }
2589 ifp->if_real_bytes = real_size;
2590 ifp->if_bytes = new_size;
2591 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2592}
2593
2594
2595
2596
2597/*
2598 * Map inode to disk block and offset.
2599 *
2600 * mp -- the mount point structure for the current file system
2601 * tp -- the current transaction
2602 * ino -- the inode number of the inode to be located
2603 * imap -- this structure is filled in with the information necessary
2604 * to retrieve the given inode from disk
2605 * flags -- flags to pass to xfs_dilocate indicating whether or not
2606 * lookups in the inode btree were OK or not
2607 */
2608int
2609xfs_imap(
2610 xfs_mount_t *mp,
2611 xfs_trans_t *tp,
2612 xfs_ino_t ino,
2613 xfs_imap_t *imap,
2614 uint flags)
2615{
2616 xfs_fsblock_t fsbno;
2617 int len;
2618 int off;
2619 int error;
2620
2621 fsbno = imap->im_blkno ?
2622 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2623 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2624 if (error != 0) {
2625 return error;
2626 }
2627 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2628 imap->im_len = XFS_FSB_TO_BB(mp, len);
2629 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2630 imap->im_ioffset = (ushort)off;
2631 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2632 return 0;
2633}
2634
2635void
2636xfs_idestroy_fork(
2637 xfs_inode_t *ip,
2638 int whichfork)
2639{
2640 xfs_ifork_t *ifp;
2641
2642 ifp = XFS_IFORK_PTR(ip, whichfork);
2643 if (ifp->if_broot != NULL) {
2644 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2645 ifp->if_broot = NULL;
2646 }
2647
2648 /*
2649 * If the format is local, then we can't have an extents
2650 * array so just look for an inline data array. If we're
2651 * not local then we may or may not have an extents list,
2652 * so check and free it up if we do.
2653 */
2654 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2655 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2656 (ifp->if_u1.if_data != NULL)) {
2657 ASSERT(ifp->if_real_bytes != 0);
2658 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2659 ifp->if_u1.if_data = NULL;
2660 ifp->if_real_bytes = 0;
2661 }
2662 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11002663 ((ifp->if_flags & XFS_IFEXTIREC) ||
2664 ((ifp->if_u1.if_extents != NULL) &&
2665 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 ASSERT(ifp->if_real_bytes != 0);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002667 xfs_iext_destroy(ifp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 }
2669 ASSERT(ifp->if_u1.if_extents == NULL ||
2670 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2671 ASSERT(ifp->if_real_bytes == 0);
2672 if (whichfork == XFS_ATTR_FORK) {
2673 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2674 ip->i_afp = NULL;
2675 }
2676}
2677
2678/*
2679 * This is called free all the memory associated with an inode.
2680 * It must free the inode itself and any buffers allocated for
2681 * if_extents/if_data and if_broot. It must also free the lock
2682 * associated with the inode.
2683 */
2684void
2685xfs_idestroy(
2686 xfs_inode_t *ip)
2687{
2688
2689 switch (ip->i_d.di_mode & S_IFMT) {
2690 case S_IFREG:
2691 case S_IFDIR:
2692 case S_IFLNK:
2693 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2694 break;
2695 }
2696 if (ip->i_afp)
2697 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2698 mrfree(&ip->i_lock);
2699 mrfree(&ip->i_iolock);
2700 freesema(&ip->i_flock);
2701#ifdef XFS_BMAP_TRACE
2702 ktrace_free(ip->i_xtrace);
2703#endif
2704#ifdef XFS_BMBT_TRACE
2705 ktrace_free(ip->i_btrace);
2706#endif
2707#ifdef XFS_RW_TRACE
2708 ktrace_free(ip->i_rwtrace);
2709#endif
2710#ifdef XFS_ILOCK_TRACE
2711 ktrace_free(ip->i_lock_trace);
2712#endif
2713#ifdef XFS_DIR2_TRACE
2714 ktrace_free(ip->i_dir_trace);
2715#endif
2716 if (ip->i_itemp) {
2717 /* XXXdpd should be able to assert this but shutdown
2718 * is leaving the AIL behind. */
2719 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2720 XFS_FORCED_SHUTDOWN(ip->i_mount));
2721 xfs_inode_item_destroy(ip);
2722 }
2723 kmem_zone_free(xfs_inode_zone, ip);
2724}
2725
2726
2727/*
2728 * Increment the pin count of the given buffer.
2729 * This value is protected by ipinlock spinlock in the mount structure.
2730 */
2731void
2732xfs_ipin(
2733 xfs_inode_t *ip)
2734{
2735 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2736
2737 atomic_inc(&ip->i_pincount);
2738}
2739
2740/*
2741 * Decrement the pin count of the given inode, and wake up
2742 * anyone in xfs_iwait_unpin() if the count goes to 0. The
Nathan Scottc41564b2006-03-29 08:55:14 +10002743 * inode must have been previously pinned with a call to xfs_ipin().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 */
2745void
2746xfs_iunpin(
2747 xfs_inode_t *ip)
2748{
2749 ASSERT(atomic_read(&ip->i_pincount) > 0);
2750
2751 if (atomic_dec_and_test(&ip->i_pincount)) {
David Chinner58829e42006-04-11 15:11:20 +10002752 /*
2753 * If the inode is currently being reclaimed, the
2754 * linux inode _and_ the xfs vnode may have been
2755 * freed so we cannot reference either of them safely.
2756 * Hence we should not try to do anything to them
2757 * if the xfs inode is currently in the reclaim
2758 * path.
2759 *
2760 * However, we still need to issue the unpin wakeup
2761 * call as the inode reclaim may be blocked waiting for
2762 * the inode to become unpinned.
2763 */
David Chinnerf273ab82006-09-28 11:06:03 +10002764 struct inode *inode = NULL;
2765
2766 spin_lock(&ip->i_flags_lock);
David Chinner58829e42006-04-11 15:11:20 +10002767 if (!(ip->i_flags & (XFS_IRECLAIM|XFS_IRECLAIMABLE))) {
Nathan Scott67fcaa72006-06-09 17:00:52 +10002768 bhv_vnode_t *vp = XFS_ITOV_NULL(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769
David Chinner58829e42006-04-11 15:11:20 +10002770 /* make sync come back and flush this inode */
2771 if (vp) {
David Chinnerf273ab82006-09-28 11:06:03 +10002772 inode = vn_to_inode(vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
David Chinner71425082006-06-09 14:55:52 +10002774 if (!(inode->i_state &
David Chinnerf273ab82006-09-28 11:06:03 +10002775 (I_NEW|I_FREEING|I_CLEAR))) {
2776 inode = igrab(inode);
2777 if (inode)
2778 mark_inode_dirty_sync(inode);
2779 } else
2780 inode = NULL;
David Chinner58829e42006-04-11 15:11:20 +10002781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 }
David Chinnerf273ab82006-09-28 11:06:03 +10002783 spin_unlock(&ip->i_flags_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 wake_up(&ip->i_ipin_wait);
David Chinnerf273ab82006-09-28 11:06:03 +10002785 if (inode)
2786 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 }
2788}
2789
2790/*
2791 * This is called to wait for the given inode to be unpinned.
2792 * It will sleep until this happens. The caller must have the
2793 * inode locked in at least shared mode so that the buffer cannot
2794 * be subsequently pinned once someone is waiting for it to be
2795 * unpinned.
2796 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002797STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798xfs_iunpin_wait(
2799 xfs_inode_t *ip)
2800{
2801 xfs_inode_log_item_t *iip;
2802 xfs_lsn_t lsn;
2803
2804 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2805
2806 if (atomic_read(&ip->i_pincount) == 0) {
2807 return;
2808 }
2809
2810 iip = ip->i_itemp;
2811 if (iip && iip->ili_last_lsn) {
2812 lsn = iip->ili_last_lsn;
2813 } else {
2814 lsn = (xfs_lsn_t)0;
2815 }
2816
2817 /*
2818 * Give the log a push so we don't wait here too long.
2819 */
2820 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2821
2822 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2823}
2824
2825
2826/*
2827 * xfs_iextents_copy()
2828 *
2829 * This is called to copy the REAL extents (as opposed to the delayed
2830 * allocation extents) from the inode into the given buffer. It
2831 * returns the number of bytes copied into the buffer.
2832 *
2833 * If there are no delayed allocation extents, then we can just
2834 * memcpy() the extents into the buffer. Otherwise, we need to
2835 * examine each extent in turn and skip those which are delayed.
2836 */
2837int
2838xfs_iextents_copy(
2839 xfs_inode_t *ip,
2840 xfs_bmbt_rec_t *buffer,
2841 int whichfork)
2842{
2843 int copied;
2844 xfs_bmbt_rec_t *dest_ep;
2845 xfs_bmbt_rec_t *ep;
2846#ifdef XFS_BMAP_TRACE
2847 static char fname[] = "xfs_iextents_copy";
2848#endif
2849 int i;
2850 xfs_ifork_t *ifp;
2851 int nrecs;
2852 xfs_fsblock_t start_block;
2853
2854 ifp = XFS_IFORK_PTR(ip, whichfork);
2855 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2856 ASSERT(ifp->if_bytes > 0);
2857
2858 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2859 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2860 ASSERT(nrecs > 0);
2861
2862 /*
2863 * There are some delayed allocation extents in the
2864 * inode, so copy the extents one at a time and skip
2865 * the delayed ones. There must be at least one
2866 * non-delayed extent.
2867 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 dest_ep = buffer;
2869 copied = 0;
2870 for (i = 0; i < nrecs; i++) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002871 ep = xfs_iext_get_ext(ifp, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 start_block = xfs_bmbt_get_startblock(ep);
2873 if (ISNULLSTARTBLOCK(start_block)) {
2874 /*
2875 * It's a delayed allocation extent, so skip it.
2876 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 continue;
2878 }
2879
2880 /* Translate to on disk format */
2881 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2882 (__uint64_t*)&dest_ep->l0);
2883 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2884 (__uint64_t*)&dest_ep->l1);
2885 dest_ep++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 copied++;
2887 }
2888 ASSERT(copied != 0);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002889 xfs_validate_extents(ifp, copied, 1, XFS_EXTFMT_INODE(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890
2891 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2892}
2893
2894/*
2895 * Each of the following cases stores data into the same region
2896 * of the on-disk inode, so only one of them can be valid at
2897 * any given time. While it is possible to have conflicting formats
2898 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2899 * in EXTENTS format, this can only happen when the fork has
2900 * changed formats after being modified but before being flushed.
2901 * In these cases, the format always takes precedence, because the
2902 * format indicates the current state of the fork.
2903 */
2904/*ARGSUSED*/
2905STATIC int
2906xfs_iflush_fork(
2907 xfs_inode_t *ip,
2908 xfs_dinode_t *dip,
2909 xfs_inode_log_item_t *iip,
2910 int whichfork,
2911 xfs_buf_t *bp)
2912{
2913 char *cp;
2914 xfs_ifork_t *ifp;
2915 xfs_mount_t *mp;
2916#ifdef XFS_TRANS_DEBUG
2917 int first;
2918#endif
2919 static const short brootflag[2] =
2920 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2921 static const short dataflag[2] =
2922 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2923 static const short extflag[2] =
2924 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2925
2926 if (iip == NULL)
2927 return 0;
2928 ifp = XFS_IFORK_PTR(ip, whichfork);
2929 /*
2930 * This can happen if we gave up in iformat in an error path,
2931 * for the attribute fork.
2932 */
2933 if (ifp == NULL) {
2934 ASSERT(whichfork == XFS_ATTR_FORK);
2935 return 0;
2936 }
2937 cp = XFS_DFORK_PTR(dip, whichfork);
2938 mp = ip->i_mount;
2939 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2940 case XFS_DINODE_FMT_LOCAL:
2941 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2942 (ifp->if_bytes > 0)) {
2943 ASSERT(ifp->if_u1.if_data != NULL);
2944 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2945 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2946 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 break;
2948
2949 case XFS_DINODE_FMT_EXTENTS:
2950 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2951 !(iip->ili_format.ilf_fields & extflag[whichfork]));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11002952 ASSERT((xfs_iext_get_ext(ifp, 0) != NULL) ||
2953 (ifp->if_bytes == 0));
2954 ASSERT((xfs_iext_get_ext(ifp, 0) == NULL) ||
2955 (ifp->if_bytes > 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2957 (ifp->if_bytes > 0)) {
2958 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2959 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2960 whichfork);
2961 }
2962 break;
2963
2964 case XFS_DINODE_FMT_BTREE:
2965 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2966 (ifp->if_broot_bytes > 0)) {
2967 ASSERT(ifp->if_broot != NULL);
2968 ASSERT(ifp->if_broot_bytes <=
2969 (XFS_IFORK_SIZE(ip, whichfork) +
2970 XFS_BROOT_SIZE_ADJ));
2971 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
2972 (xfs_bmdr_block_t *)cp,
2973 XFS_DFORK_SIZE(dip, mp, whichfork));
2974 }
2975 break;
2976
2977 case XFS_DINODE_FMT_DEV:
2978 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2979 ASSERT(whichfork == XFS_DATA_FORK);
2980 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
2981 }
2982 break;
2983
2984 case XFS_DINODE_FMT_UUID:
2985 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2986 ASSERT(whichfork == XFS_DATA_FORK);
2987 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
2988 sizeof(uuid_t));
2989 }
2990 break;
2991
2992 default:
2993 ASSERT(0);
2994 break;
2995 }
2996
2997 return 0;
2998}
2999
3000/*
3001 * xfs_iflush() will write a modified inode's changes out to the
3002 * inode's on disk home. The caller must have the inode lock held
3003 * in at least shared mode and the inode flush semaphore must be
3004 * held as well. The inode lock will still be held upon return from
3005 * the call and the caller is free to unlock it.
3006 * The inode flush lock will be unlocked when the inode reaches the disk.
3007 * The flags indicate how the inode's buffer should be written out.
3008 */
3009int
3010xfs_iflush(
3011 xfs_inode_t *ip,
3012 uint flags)
3013{
3014 xfs_inode_log_item_t *iip;
3015 xfs_buf_t *bp;
3016 xfs_dinode_t *dip;
3017 xfs_mount_t *mp;
3018 int error;
3019 /* REFERENCED */
3020 xfs_chash_t *ch;
3021 xfs_inode_t *iq;
3022 int clcount; /* count of inodes clustered */
3023 int bufwasdelwri;
3024 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3025 SPLDECL(s);
3026
3027 XFS_STATS_INC(xs_iflush_count);
3028
3029 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
Al Viro0d8fee32006-06-19 08:41:30 +10003030 ASSERT(issemalocked(&(ip->i_flock)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3032 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3033
3034 iip = ip->i_itemp;
3035 mp = ip->i_mount;
3036
3037 /*
3038 * If the inode isn't dirty, then just release the inode
3039 * flush lock and do nothing.
3040 */
3041 if ((ip->i_update_core == 0) &&
3042 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3043 ASSERT((iip != NULL) ?
3044 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3045 xfs_ifunlock(ip);
3046 return 0;
3047 }
3048
3049 /*
3050 * We can't flush the inode until it is unpinned, so
3051 * wait for it. We know noone new can pin it, because
3052 * we are holding the inode lock shared and you need
3053 * to hold it exclusively to pin the inode.
3054 */
3055 xfs_iunpin_wait(ip);
3056
3057 /*
3058 * This may have been unpinned because the filesystem is shutting
3059 * down forcibly. If that's the case we must not write this inode
3060 * to disk, because the log record didn't make it to disk!
3061 */
3062 if (XFS_FORCED_SHUTDOWN(mp)) {
3063 ip->i_update_core = 0;
3064 if (iip)
3065 iip->ili_format.ilf_fields = 0;
3066 xfs_ifunlock(ip);
3067 return XFS_ERROR(EIO);
3068 }
3069
3070 /*
3071 * Get the buffer containing the on-disk inode.
3072 */
Nathan Scottb12dd342006-03-17 17:26:04 +11003073 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0, 0);
3074 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 xfs_ifunlock(ip);
3076 return error;
3077 }
3078
3079 /*
3080 * Decide how buffer will be flushed out. This is done before
3081 * the call to xfs_iflush_int because this field is zeroed by it.
3082 */
3083 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3084 /*
3085 * Flush out the inode buffer according to the directions
3086 * of the caller. In the cases where the caller has given
3087 * us a choice choose the non-delwri case. This is because
3088 * the inode is in the AIL and we need to get it out soon.
3089 */
3090 switch (flags) {
3091 case XFS_IFLUSH_SYNC:
3092 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3093 flags = 0;
3094 break;
3095 case XFS_IFLUSH_ASYNC:
3096 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3097 flags = INT_ASYNC;
3098 break;
3099 case XFS_IFLUSH_DELWRI:
3100 flags = INT_DELWRI;
3101 break;
3102 default:
3103 ASSERT(0);
3104 flags = 0;
3105 break;
3106 }
3107 } else {
3108 switch (flags) {
3109 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3110 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3111 case XFS_IFLUSH_DELWRI:
3112 flags = INT_DELWRI;
3113 break;
3114 case XFS_IFLUSH_ASYNC:
3115 flags = INT_ASYNC;
3116 break;
3117 case XFS_IFLUSH_SYNC:
3118 flags = 0;
3119 break;
3120 default:
3121 ASSERT(0);
3122 flags = 0;
3123 break;
3124 }
3125 }
3126
3127 /*
3128 * First flush out the inode that xfs_iflush was called with.
3129 */
3130 error = xfs_iflush_int(ip, bp);
3131 if (error) {
3132 goto corrupt_out;
3133 }
3134
3135 /*
3136 * inode clustering:
3137 * see if other inodes can be gathered into this write
3138 */
3139
3140 ip->i_chash->chl_buf = bp;
3141
3142 ch = XFS_CHASH(mp, ip->i_blkno);
3143 s = mutex_spinlock(&ch->ch_lock);
3144
3145 clcount = 0;
3146 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3147 /*
3148 * Do an un-protected check to see if the inode is dirty and
3149 * is a candidate for flushing. These checks will be repeated
3150 * later after the appropriate locks are acquired.
3151 */
3152 iip = iq->i_itemp;
3153 if ((iq->i_update_core == 0) &&
3154 ((iip == NULL) ||
3155 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3156 xfs_ipincount(iq) == 0) {
3157 continue;
3158 }
3159
3160 /*
3161 * Try to get locks. If any are unavailable,
3162 * then this inode cannot be flushed and is skipped.
3163 */
3164
3165 /* get inode locks (just i_lock) */
3166 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3167 /* get inode flush lock */
3168 if (xfs_iflock_nowait(iq)) {
3169 /* check if pinned */
3170 if (xfs_ipincount(iq) == 0) {
3171 /* arriving here means that
3172 * this inode can be flushed.
3173 * first re-check that it's
3174 * dirty
3175 */
3176 iip = iq->i_itemp;
3177 if ((iq->i_update_core != 0)||
3178 ((iip != NULL) &&
3179 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3180 clcount++;
3181 error = xfs_iflush_int(iq, bp);
3182 if (error) {
3183 xfs_iunlock(iq,
3184 XFS_ILOCK_SHARED);
3185 goto cluster_corrupt_out;
3186 }
3187 } else {
3188 xfs_ifunlock(iq);
3189 }
3190 } else {
3191 xfs_ifunlock(iq);
3192 }
3193 }
3194 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3195 }
3196 }
3197 mutex_spinunlock(&ch->ch_lock, s);
3198
3199 if (clcount) {
3200 XFS_STATS_INC(xs_icluster_flushcnt);
3201 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3202 }
3203
3204 /*
3205 * If the buffer is pinned then push on the log so we won't
3206 * get stuck waiting in the write for too long.
3207 */
3208 if (XFS_BUF_ISPINNED(bp)){
3209 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3210 }
3211
3212 if (flags & INT_DELWRI) {
3213 xfs_bdwrite(mp, bp);
3214 } else if (flags & INT_ASYNC) {
3215 xfs_bawrite(mp, bp);
3216 } else {
3217 error = xfs_bwrite(mp, bp);
3218 }
3219 return error;
3220
3221corrupt_out:
3222 xfs_buf_relse(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10003223 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 xfs_iflush_abort(ip);
3225 /*
3226 * Unlocks the flush lock
3227 */
3228 return XFS_ERROR(EFSCORRUPTED);
3229
3230cluster_corrupt_out:
3231 /* Corruption detected in the clustering loop. Invalidate the
3232 * inode buffer and shut down the filesystem.
3233 */
3234 mutex_spinunlock(&ch->ch_lock, s);
3235
3236 /*
3237 * Clean up the buffer. If it was B_DELWRI, just release it --
3238 * brelse can handle it with no problems. If not, shut down the
3239 * filesystem before releasing the buffer.
3240 */
3241 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3242 xfs_buf_relse(bp);
3243 }
3244
Nathan Scott7d04a332006-06-09 14:58:38 +10003245 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246
3247 if(!bufwasdelwri) {
3248 /*
3249 * Just like incore_relse: if we have b_iodone functions,
3250 * mark the buffer as an error and call them. Otherwise
3251 * mark it as stale and brelse.
3252 */
3253 if (XFS_BUF_IODONE_FUNC(bp)) {
3254 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3255 XFS_BUF_UNDONE(bp);
3256 XFS_BUF_STALE(bp);
3257 XFS_BUF_SHUT(bp);
3258 XFS_BUF_ERROR(bp,EIO);
3259 xfs_biodone(bp);
3260 } else {
3261 XFS_BUF_STALE(bp);
3262 xfs_buf_relse(bp);
3263 }
3264 }
3265
3266 xfs_iflush_abort(iq);
3267 /*
3268 * Unlocks the flush lock
3269 */
3270 return XFS_ERROR(EFSCORRUPTED);
3271}
3272
3273
3274STATIC int
3275xfs_iflush_int(
3276 xfs_inode_t *ip,
3277 xfs_buf_t *bp)
3278{
3279 xfs_inode_log_item_t *iip;
3280 xfs_dinode_t *dip;
3281 xfs_mount_t *mp;
3282#ifdef XFS_TRANS_DEBUG
3283 int first;
3284#endif
3285 SPLDECL(s);
3286
3287 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
Al Viro0d8fee32006-06-19 08:41:30 +10003288 ASSERT(issemalocked(&(ip->i_flock)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3290 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3291
3292 iip = ip->i_itemp;
3293 mp = ip->i_mount;
3294
3295
3296 /*
3297 * If the inode isn't dirty, then just release the inode
3298 * flush lock and do nothing.
3299 */
3300 if ((ip->i_update_core == 0) &&
3301 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3302 xfs_ifunlock(ip);
3303 return 0;
3304 }
3305
3306 /* set *dip = inode's place in the buffer */
3307 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3308
3309 /*
3310 * Clear i_update_core before copying out the data.
3311 * This is for coordination with our timestamp updates
3312 * that don't hold the inode lock. They will always
3313 * update the timestamps BEFORE setting i_update_core,
3314 * so if we clear i_update_core after they set it we
3315 * are guaranteed to see their updates to the timestamps.
3316 * I believe that this depends on strongly ordered memory
3317 * semantics, but we have that. We use the SYNCHRONIZE
3318 * macro to make sure that the compiler does not reorder
3319 * the i_update_core access below the data copy below.
3320 */
3321 ip->i_update_core = 0;
3322 SYNCHRONIZE();
3323
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11003324 /*
3325 * Make sure to get the latest atime from the Linux inode.
3326 */
3327 xfs_synchronize_atime(ip);
3328
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3330 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3331 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3332 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3333 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3334 goto corrupt_out;
3335 }
3336 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3337 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3338 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3339 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3340 ip->i_ino, ip, ip->i_d.di_magic);
3341 goto corrupt_out;
3342 }
3343 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3344 if (XFS_TEST_ERROR(
3345 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3346 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3347 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3348 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3349 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3350 ip->i_ino, ip);
3351 goto corrupt_out;
3352 }
3353 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3354 if (XFS_TEST_ERROR(
3355 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3356 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3357 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3358 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3359 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3360 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3361 ip->i_ino, ip);
3362 goto corrupt_out;
3363 }
3364 }
3365 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3366 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3367 XFS_RANDOM_IFLUSH_5)) {
3368 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3369 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3370 ip->i_ino,
3371 ip->i_d.di_nextents + ip->i_d.di_anextents,
3372 ip->i_d.di_nblocks,
3373 ip);
3374 goto corrupt_out;
3375 }
3376 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3377 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3378 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3379 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3380 ip->i_ino, ip->i_d.di_forkoff, ip);
3381 goto corrupt_out;
3382 }
3383 /*
3384 * bump the flush iteration count, used to detect flushes which
3385 * postdate a log record during recovery.
3386 */
3387
3388 ip->i_d.di_flushiter++;
3389
3390 /*
3391 * Copy the dirty parts of the inode into the on-disk
3392 * inode. We always copy out the core of the inode,
3393 * because if the inode is dirty at all the core must
3394 * be.
3395 */
3396 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3397
3398 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3399 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3400 ip->i_d.di_flushiter = 0;
3401
3402 /*
3403 * If this is really an old format inode and the superblock version
3404 * has not been updated to support only new format inodes, then
3405 * convert back to the old inode format. If the superblock version
3406 * has been updated, then make the conversion permanent.
3407 */
3408 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3409 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3410 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3411 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3412 /*
3413 * Convert it back.
3414 */
3415 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3416 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3417 } else {
3418 /*
3419 * The superblock version has already been bumped,
3420 * so just make the conversion to the new inode
3421 * format permanent.
3422 */
3423 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3424 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3425 ip->i_d.di_onlink = 0;
3426 dip->di_core.di_onlink = 0;
3427 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3428 memset(&(dip->di_core.di_pad[0]), 0,
3429 sizeof(dip->di_core.di_pad));
3430 ASSERT(ip->i_d.di_projid == 0);
3431 }
3432 }
3433
3434 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3435 goto corrupt_out;
3436 }
3437
3438 if (XFS_IFORK_Q(ip)) {
3439 /*
3440 * The only error from xfs_iflush_fork is on the data fork.
3441 */
3442 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3443 }
3444 xfs_inobp_check(mp, bp);
3445
3446 /*
3447 * We've recorded everything logged in the inode, so we'd
3448 * like to clear the ilf_fields bits so we don't log and
3449 * flush things unnecessarily. However, we can't stop
3450 * logging all this information until the data we've copied
3451 * into the disk buffer is written to disk. If we did we might
3452 * overwrite the copy of the inode in the log with all the
3453 * data after re-logging only part of it, and in the face of
3454 * a crash we wouldn't have all the data we need to recover.
3455 *
3456 * What we do is move the bits to the ili_last_fields field.
3457 * When logging the inode, these bits are moved back to the
3458 * ilf_fields field. In the xfs_iflush_done() routine we
3459 * clear ili_last_fields, since we know that the information
3460 * those bits represent is permanently on disk. As long as
3461 * the flush completes before the inode is logged again, then
3462 * both ilf_fields and ili_last_fields will be cleared.
3463 *
3464 * We can play with the ilf_fields bits here, because the inode
3465 * lock must be held exclusively in order to set bits there
3466 * and the flush lock protects the ili_last_fields bits.
3467 * Set ili_logged so the flush done
3468 * routine can tell whether or not to look in the AIL.
3469 * Also, store the current LSN of the inode so that we can tell
3470 * whether the item has moved in the AIL from xfs_iflush_done().
3471 * In order to read the lsn we need the AIL lock, because
3472 * it is a 64 bit value that cannot be read atomically.
3473 */
3474 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3475 iip->ili_last_fields = iip->ili_format.ilf_fields;
3476 iip->ili_format.ilf_fields = 0;
3477 iip->ili_logged = 1;
3478
3479 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3480 AIL_LOCK(mp,s);
3481 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3482 AIL_UNLOCK(mp, s);
3483
3484 /*
3485 * Attach the function xfs_iflush_done to the inode's
3486 * buffer. This will remove the inode from the AIL
3487 * and unlock the inode's flush lock when the inode is
3488 * completely written to disk.
3489 */
3490 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3491 xfs_iflush_done, (xfs_log_item_t *)iip);
3492
3493 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3494 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3495 } else {
3496 /*
3497 * We're flushing an inode which is not in the AIL and has
3498 * not been logged but has i_update_core set. For this
3499 * case we can use a B_DELWRI flush and immediately drop
3500 * the inode flush lock because we can avoid the whole
3501 * AIL state thing. It's OK to drop the flush lock now,
3502 * because we've already locked the buffer and to do anything
3503 * you really need both.
3504 */
3505 if (iip != NULL) {
3506 ASSERT(iip->ili_logged == 0);
3507 ASSERT(iip->ili_last_fields == 0);
3508 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3509 }
3510 xfs_ifunlock(ip);
3511 }
3512
3513 return 0;
3514
3515corrupt_out:
3516 return XFS_ERROR(EFSCORRUPTED);
3517}
3518
3519
3520/*
Christoph Hellwigefa80272005-06-21 15:37:17 +10003521 * Flush all inactive inodes in mp.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 */
Christoph Hellwigefa80272005-06-21 15:37:17 +10003523void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524xfs_iflush_all(
Christoph Hellwigefa80272005-06-21 15:37:17 +10003525 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +10003528 bhv_vnode_t *vp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529
Christoph Hellwigefa80272005-06-21 15:37:17 +10003530 again:
3531 XFS_MOUNT_ILOCK(mp);
3532 ip = mp->m_inodes;
3533 if (ip == NULL)
3534 goto out;
3535
3536 do {
3537 /* Make sure we skip markers inserted by sync */
3538 if (ip->i_mount == NULL) {
3539 ip = ip->i_mnext;
3540 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542
Christoph Hellwigefa80272005-06-21 15:37:17 +10003543 vp = XFS_ITOV_NULL(ip);
3544 if (!vp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 XFS_MOUNT_IUNLOCK(mp);
Christoph Hellwigefa80272005-06-21 15:37:17 +10003546 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3547 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003549
Christoph Hellwigefa80272005-06-21 15:37:17 +10003550 ASSERT(vn_count(vp) == 0);
3551
3552 ip = ip->i_mnext;
3553 } while (ip != mp->m_inodes);
3554 out:
3555 XFS_MOUNT_IUNLOCK(mp);
3556}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557
3558/*
3559 * xfs_iaccess: check accessibility of inode for mode.
3560 */
3561int
3562xfs_iaccess(
3563 xfs_inode_t *ip,
3564 mode_t mode,
3565 cred_t *cr)
3566{
3567 int error;
3568 mode_t orgmode = mode;
Nathan Scottec86dc02006-03-17 17:25:36 +11003569 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570
3571 if (mode & S_IWUSR) {
3572 umode_t imode = inode->i_mode;
3573
3574 if (IS_RDONLY(inode) &&
3575 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3576 return XFS_ERROR(EROFS);
3577
3578 if (IS_IMMUTABLE(inode))
3579 return XFS_ERROR(EACCES);
3580 }
3581
3582 /*
3583 * If there's an Access Control List it's used instead of
3584 * the mode bits.
3585 */
3586 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3587 return error ? XFS_ERROR(error) : 0;
3588
3589 if (current_fsuid(cr) != ip->i_d.di_uid) {
3590 mode >>= 3;
3591 if (!in_group_p((gid_t)ip->i_d.di_gid))
3592 mode >>= 3;
3593 }
3594
3595 /*
3596 * If the DACs are ok we don't need any capability check.
3597 */
3598 if ((ip->i_d.di_mode & mode) == mode)
3599 return 0;
3600 /*
3601 * Read/write DACs are always overridable.
3602 * Executable DACs are overridable if at least one exec bit is set.
3603 */
3604 if (!(orgmode & S_IXUSR) ||
3605 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3606 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3607 return 0;
3608
3609 if ((orgmode == S_IRUSR) ||
3610 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3611 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3612 return 0;
3613#ifdef NOISE
3614 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3615#endif /* NOISE */
3616 return XFS_ERROR(EACCES);
3617 }
3618 return XFS_ERROR(EACCES);
3619}
3620
3621/*
3622 * xfs_iroundup: round up argument to next power of two
3623 */
3624uint
3625xfs_iroundup(
3626 uint v)
3627{
3628 int i;
3629 uint m;
3630
3631 if ((v & (v - 1)) == 0)
3632 return v;
3633 ASSERT((v & 0x80000000) == 0);
3634 if ((v & (v + 1)) == 0)
3635 return v + 1;
3636 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3637 if (v & m)
3638 continue;
3639 v |= m;
3640 if ((v & (v + 1)) == 0)
3641 return v + 1;
3642 }
3643 ASSERT(0);
3644 return( 0 );
3645}
3646
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647#ifdef XFS_ILOCK_TRACE
3648ktrace_t *xfs_ilock_trace_buf;
3649
3650void
3651xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3652{
3653 ktrace_enter(ip->i_lock_trace,
3654 (void *)ip,
3655 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3656 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3657 (void *)ra, /* caller of ilock */
3658 (void *)(unsigned long)current_cpu(),
3659 (void *)(unsigned long)current_pid(),
3660 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3661}
3662#endif
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003663
3664/*
3665 * Return a pointer to the extent record at file index idx.
3666 */
3667xfs_bmbt_rec_t *
3668xfs_iext_get_ext(
3669 xfs_ifork_t *ifp, /* inode fork pointer */
3670 xfs_extnum_t idx) /* index of target extent */
3671{
3672 ASSERT(idx >= 0);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003673 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
3674 return ifp->if_u1.if_ext_irec->er_extbuf;
3675 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3676 xfs_ext_irec_t *erp; /* irec pointer */
3677 int erp_idx = 0; /* irec index */
3678 xfs_extnum_t page_idx = idx; /* ext index in target list */
3679
3680 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3681 return &erp->er_extbuf[page_idx];
3682 } else if (ifp->if_bytes) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003683 return &ifp->if_u1.if_extents[idx];
3684 } else {
3685 return NULL;
3686 }
3687}
3688
3689/*
3690 * Insert new item(s) into the extent records for incore inode
3691 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
3692 */
3693void
3694xfs_iext_insert(
3695 xfs_ifork_t *ifp, /* inode fork pointer */
3696 xfs_extnum_t idx, /* starting index of new items */
3697 xfs_extnum_t count, /* number of inserted items */
3698 xfs_bmbt_irec_t *new) /* items to insert */
3699{
3700 xfs_bmbt_rec_t *ep; /* extent record pointer */
3701 xfs_extnum_t i; /* extent record index */
3702
3703 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3704 xfs_iext_add(ifp, idx, count);
3705 for (i = idx; i < idx + count; i++, new++) {
3706 ep = xfs_iext_get_ext(ifp, i);
3707 xfs_bmbt_set_all(ep, new);
3708 }
3709}
3710
3711/*
3712 * This is called when the amount of space required for incore file
3713 * extents needs to be increased. The ext_diff parameter stores the
3714 * number of new extents being added and the idx parameter contains
3715 * the extent index where the new extents will be added. If the new
3716 * extents are being appended, then we just need to (re)allocate and
3717 * initialize the space. Otherwise, if the new extents are being
3718 * inserted into the middle of the existing entries, a bit more work
3719 * is required to make room for the new extents to be inserted. The
3720 * caller is responsible for filling in the new extent entries upon
3721 * return.
3722 */
3723void
3724xfs_iext_add(
3725 xfs_ifork_t *ifp, /* inode fork pointer */
3726 xfs_extnum_t idx, /* index to begin adding exts */
Nathan Scottc41564b2006-03-29 08:55:14 +10003727 int ext_diff) /* number of extents to add */
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003728{
3729 int byte_diff; /* new bytes being added */
3730 int new_size; /* size of extents after adding */
3731 xfs_extnum_t nextents; /* number of extents in file */
3732
3733 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3734 ASSERT((idx >= 0) && (idx <= nextents));
3735 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
3736 new_size = ifp->if_bytes + byte_diff;
3737 /*
3738 * If the new number of extents (nextents + ext_diff)
3739 * fits inside the inode, then continue to use the inline
3740 * extent buffer.
3741 */
3742 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
3743 if (idx < nextents) {
3744 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
3745 &ifp->if_u2.if_inline_ext[idx],
3746 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3747 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
3748 }
3749 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3750 ifp->if_real_bytes = 0;
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003751 ifp->if_lastex = nextents + ext_diff;
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003752 }
3753 /*
3754 * Otherwise use a linear (direct) extent list.
3755 * If the extents are currently inside the inode,
3756 * xfs_iext_realloc_direct will switch us from
3757 * inline to direct extent allocation mode.
3758 */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003759 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003760 xfs_iext_realloc_direct(ifp, new_size);
3761 if (idx < nextents) {
3762 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
3763 &ifp->if_u1.if_extents[idx],
3764 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
3765 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
3766 }
3767 }
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003768 /* Indirection array */
3769 else {
3770 xfs_ext_irec_t *erp;
3771 int erp_idx = 0;
3772 int page_idx = idx;
3773
3774 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
3775 if (ifp->if_flags & XFS_IFEXTIREC) {
3776 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
3777 } else {
3778 xfs_iext_irec_init(ifp);
3779 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3780 erp = ifp->if_u1.if_ext_irec;
3781 }
3782 /* Extents fit in target extent page */
3783 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
3784 if (page_idx < erp->er_extcount) {
3785 memmove(&erp->er_extbuf[page_idx + ext_diff],
3786 &erp->er_extbuf[page_idx],
3787 (erp->er_extcount - page_idx) *
3788 sizeof(xfs_bmbt_rec_t));
3789 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
3790 }
3791 erp->er_extcount += ext_diff;
3792 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3793 }
3794 /* Insert a new extent page */
3795 else if (erp) {
3796 xfs_iext_add_indirect_multi(ifp,
3797 erp_idx, page_idx, ext_diff);
3798 }
3799 /*
3800 * If extent(s) are being appended to the last page in
3801 * the indirection array and the new extent(s) don't fit
3802 * in the page, then erp is NULL and erp_idx is set to
3803 * the next index needed in the indirection array.
3804 */
3805 else {
3806 int count = ext_diff;
3807
3808 while (count) {
3809 erp = xfs_iext_irec_new(ifp, erp_idx);
3810 erp->er_extcount = count;
3811 count -= MIN(count, (int)XFS_LINEAR_EXTS);
3812 if (count) {
3813 erp_idx++;
3814 }
3815 }
3816 }
3817 }
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003818 ifp->if_bytes = new_size;
3819}
3820
3821/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003822 * This is called when incore extents are being added to the indirection
3823 * array and the new extents do not fit in the target extent list. The
3824 * erp_idx parameter contains the irec index for the target extent list
3825 * in the indirection array, and the idx parameter contains the extent
3826 * index within the list. The number of extents being added is stored
3827 * in the count parameter.
3828 *
3829 * |-------| |-------|
3830 * | | | | idx - number of extents before idx
3831 * | idx | | count |
3832 * | | | | count - number of extents being inserted at idx
3833 * |-------| |-------|
3834 * | count | | nex2 | nex2 - number of extents after idx + count
3835 * |-------| |-------|
3836 */
3837void
3838xfs_iext_add_indirect_multi(
3839 xfs_ifork_t *ifp, /* inode fork pointer */
3840 int erp_idx, /* target extent irec index */
3841 xfs_extnum_t idx, /* index within target list */
3842 int count) /* new extents being added */
3843{
3844 int byte_diff; /* new bytes being added */
3845 xfs_ext_irec_t *erp; /* pointer to irec entry */
3846 xfs_extnum_t ext_diff; /* number of extents to add */
3847 xfs_extnum_t ext_cnt; /* new extents still needed */
3848 xfs_extnum_t nex2; /* extents after idx + count */
3849 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3850 int nlists; /* number of irec's (lists) */
3851
3852 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3853 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3854 nex2 = erp->er_extcount - idx;
3855 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3856
3857 /*
3858 * Save second part of target extent list
3859 * (all extents past */
3860 if (nex2) {
3861 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3862 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
3863 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3864 erp->er_extcount -= nex2;
3865 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3866 memset(&erp->er_extbuf[idx], 0, byte_diff);
3867 }
3868
3869 /*
3870 * Add the new extents to the end of the target
3871 * list, then allocate new irec record(s) and
3872 * extent buffer(s) as needed to store the rest
3873 * of the new extents.
3874 */
3875 ext_cnt = count;
3876 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3877 if (ext_diff) {
3878 erp->er_extcount += ext_diff;
3879 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3880 ext_cnt -= ext_diff;
3881 }
3882 while (ext_cnt) {
3883 erp_idx++;
3884 erp = xfs_iext_irec_new(ifp, erp_idx);
3885 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3886 erp->er_extcount = ext_diff;
3887 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3888 ext_cnt -= ext_diff;
3889 }
3890
3891 /* Add nex2 extents back to indirection array */
3892 if (nex2) {
3893 xfs_extnum_t ext_avail;
3894 int i;
3895
3896 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3897 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3898 i = 0;
3899 /*
3900 * If nex2 extents fit in the current page, append
3901 * nex2_ep after the new extents.
3902 */
3903 if (nex2 <= ext_avail) {
3904 i = erp->er_extcount;
3905 }
3906 /*
3907 * Otherwise, check if space is available in the
3908 * next page.
3909 */
3910 else if ((erp_idx < nlists - 1) &&
3911 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3912 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3913 erp_idx++;
3914 erp++;
3915 /* Create a hole for nex2 extents */
3916 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3917 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3918 }
3919 /*
3920 * Final choice, create a new extent page for
3921 * nex2 extents.
3922 */
3923 else {
3924 erp_idx++;
3925 erp = xfs_iext_irec_new(ifp, erp_idx);
3926 }
3927 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3928 kmem_free(nex2_ep, byte_diff);
3929 erp->er_extcount += nex2;
3930 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3931 }
3932}
3933
3934/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003935 * This is called when the amount of space required for incore file
3936 * extents needs to be decreased. The ext_diff parameter stores the
3937 * number of extents to be removed and the idx parameter contains
3938 * the extent index where the extents will be removed from.
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003939 *
3940 * If the amount of space needed has decreased below the linear
3941 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3942 * extent array. Otherwise, use kmem_realloc() to adjust the
3943 * size to what is needed.
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003944 */
3945void
3946xfs_iext_remove(
3947 xfs_ifork_t *ifp, /* inode fork pointer */
3948 xfs_extnum_t idx, /* index to begin removing exts */
3949 int ext_diff) /* number of extents to remove */
3950{
3951 xfs_extnum_t nextents; /* number of extents in file */
3952 int new_size; /* size of extents after removal */
3953
3954 ASSERT(ext_diff > 0);
3955 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3956 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3957
3958 if (new_size == 0) {
3959 xfs_iext_destroy(ifp);
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003960 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3961 xfs_iext_remove_indirect(ifp, idx, ext_diff);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003962 } else if (ifp->if_real_bytes) {
3963 xfs_iext_remove_direct(ifp, idx, ext_diff);
3964 } else {
3965 xfs_iext_remove_inline(ifp, idx, ext_diff);
3966 }
3967 ifp->if_bytes = new_size;
3968}
3969
3970/*
3971 * This removes ext_diff extents from the inline buffer, beginning
3972 * at extent index idx.
3973 */
3974void
3975xfs_iext_remove_inline(
3976 xfs_ifork_t *ifp, /* inode fork pointer */
3977 xfs_extnum_t idx, /* index to begin removing exts */
3978 int ext_diff) /* number of extents to remove */
3979{
3980 int nextents; /* number of extents in file */
3981
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11003982 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11003983 ASSERT(idx < XFS_INLINE_EXTS);
3984 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3985 ASSERT(((nextents - ext_diff) > 0) &&
3986 (nextents - ext_diff) < XFS_INLINE_EXTS);
3987
3988 if (idx + ext_diff < nextents) {
3989 memmove(&ifp->if_u2.if_inline_ext[idx],
3990 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3991 (nextents - (idx + ext_diff)) *
3992 sizeof(xfs_bmbt_rec_t));
3993 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3994 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3995 } else {
3996 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3997 ext_diff * sizeof(xfs_bmbt_rec_t));
3998 }
3999}
4000
4001/*
4002 * This removes ext_diff extents from a linear (direct) extent list,
4003 * beginning at extent index idx. If the extents are being removed
4004 * from the end of the list (ie. truncate) then we just need to re-
4005 * allocate the list to remove the extra space. Otherwise, if the
4006 * extents are being removed from the middle of the existing extent
4007 * entries, then we first need to move the extent records beginning
4008 * at idx + ext_diff up in the list to overwrite the records being
4009 * removed, then remove the extra space via kmem_realloc.
4010 */
4011void
4012xfs_iext_remove_direct(
4013 xfs_ifork_t *ifp, /* inode fork pointer */
4014 xfs_extnum_t idx, /* index to begin removing exts */
4015 int ext_diff) /* number of extents to remove */
4016{
4017 xfs_extnum_t nextents; /* number of extents in file */
4018 int new_size; /* size of extents after removal */
4019
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004020 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004021 new_size = ifp->if_bytes -
4022 (ext_diff * sizeof(xfs_bmbt_rec_t));
4023 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4024
4025 if (new_size == 0) {
4026 xfs_iext_destroy(ifp);
4027 return;
4028 }
4029 /* Move extents up in the list (if needed) */
4030 if (idx + ext_diff < nextents) {
4031 memmove(&ifp->if_u1.if_extents[idx],
4032 &ifp->if_u1.if_extents[idx + ext_diff],
4033 (nextents - (idx + ext_diff)) *
4034 sizeof(xfs_bmbt_rec_t));
4035 }
4036 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
4037 0, ext_diff * sizeof(xfs_bmbt_rec_t));
4038 /*
4039 * Reallocate the direct extent list. If the extents
4040 * will fit inside the inode then xfs_iext_realloc_direct
4041 * will switch from direct to inline extent allocation
4042 * mode for us.
4043 */
4044 xfs_iext_realloc_direct(ifp, new_size);
4045 ifp->if_bytes = new_size;
4046}
4047
4048/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004049 * This is called when incore extents are being removed from the
4050 * indirection array and the extents being removed span multiple extent
4051 * buffers. The idx parameter contains the file extent index where we
4052 * want to begin removing extents, and the count parameter contains
4053 * how many extents need to be removed.
4054 *
4055 * |-------| |-------|
4056 * | nex1 | | | nex1 - number of extents before idx
4057 * |-------| | count |
4058 * | | | | count - number of extents being removed at idx
4059 * | count | |-------|
4060 * | | | nex2 | nex2 - number of extents after idx + count
4061 * |-------| |-------|
4062 */
4063void
4064xfs_iext_remove_indirect(
4065 xfs_ifork_t *ifp, /* inode fork pointer */
4066 xfs_extnum_t idx, /* index to begin removing extents */
4067 int count) /* number of extents to remove */
4068{
4069 xfs_ext_irec_t *erp; /* indirection array pointer */
4070 int erp_idx = 0; /* indirection array index */
4071 xfs_extnum_t ext_cnt; /* extents left to remove */
4072 xfs_extnum_t ext_diff; /* extents to remove in current list */
4073 xfs_extnum_t nex1; /* number of extents before idx */
4074 xfs_extnum_t nex2; /* extents after idx + count */
Nathan Scottc41564b2006-03-29 08:55:14 +10004075 int nlists; /* entries in indirection array */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004076 int page_idx = idx; /* index in target extent list */
4077
4078 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4079 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
4080 ASSERT(erp != NULL);
4081 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4082 nex1 = page_idx;
4083 ext_cnt = count;
4084 while (ext_cnt) {
4085 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
4086 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
4087 /*
4088 * Check for deletion of entire list;
4089 * xfs_iext_irec_remove() updates extent offsets.
4090 */
4091 if (ext_diff == erp->er_extcount) {
4092 xfs_iext_irec_remove(ifp, erp_idx);
4093 ext_cnt -= ext_diff;
4094 nex1 = 0;
4095 if (ext_cnt) {
4096 ASSERT(erp_idx < ifp->if_real_bytes /
4097 XFS_IEXT_BUFSZ);
4098 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4099 nex1 = 0;
4100 continue;
4101 } else {
4102 break;
4103 }
4104 }
4105 /* Move extents up (if needed) */
4106 if (nex2) {
4107 memmove(&erp->er_extbuf[nex1],
4108 &erp->er_extbuf[nex1 + ext_diff],
4109 nex2 * sizeof(xfs_bmbt_rec_t));
4110 }
4111 /* Zero out rest of page */
4112 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
4113 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
4114 /* Update remaining counters */
4115 erp->er_extcount -= ext_diff;
4116 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
4117 ext_cnt -= ext_diff;
4118 nex1 = 0;
4119 erp_idx++;
4120 erp++;
4121 }
4122 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
4123 xfs_iext_irec_compact(ifp);
4124}
4125
4126/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004127 * Create, destroy, or resize a linear (direct) block of extents.
4128 */
4129void
4130xfs_iext_realloc_direct(
4131 xfs_ifork_t *ifp, /* inode fork pointer */
4132 int new_size) /* new size of extents */
4133{
4134 int rnew_size; /* real new size of extents */
4135
4136 rnew_size = new_size;
4137
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004138 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
4139 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
4140 (new_size != ifp->if_real_bytes)));
4141
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004142 /* Free extent records */
4143 if (new_size == 0) {
4144 xfs_iext_destroy(ifp);
4145 }
4146 /* Resize direct extent list and zero any new bytes */
4147 else if (ifp->if_real_bytes) {
4148 /* Check if extents will fit inside the inode */
4149 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
4150 xfs_iext_direct_to_inline(ifp, new_size /
4151 (uint)sizeof(xfs_bmbt_rec_t));
4152 ifp->if_bytes = new_size;
4153 return;
4154 }
4155 if ((new_size & (new_size - 1)) != 0) {
4156 rnew_size = xfs_iroundup(new_size);
4157 }
4158 if (rnew_size != ifp->if_real_bytes) {
4159 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4160 kmem_realloc(ifp->if_u1.if_extents,
4161 rnew_size,
4162 ifp->if_real_bytes,
4163 KM_SLEEP);
4164 }
4165 if (rnew_size > ifp->if_real_bytes) {
4166 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
4167 (uint)sizeof(xfs_bmbt_rec_t)], 0,
4168 rnew_size - ifp->if_real_bytes);
4169 }
4170 }
4171 /*
4172 * Switch from the inline extent buffer to a direct
4173 * extent list. Be sure to include the inline extent
4174 * bytes in new_size.
4175 */
4176 else {
4177 new_size += ifp->if_bytes;
4178 if ((new_size & (new_size - 1)) != 0) {
4179 rnew_size = xfs_iroundup(new_size);
4180 }
4181 xfs_iext_inline_to_direct(ifp, rnew_size);
4182 }
4183 ifp->if_real_bytes = rnew_size;
4184 ifp->if_bytes = new_size;
4185}
4186
4187/*
4188 * Switch from linear (direct) extent records to inline buffer.
4189 */
4190void
4191xfs_iext_direct_to_inline(
4192 xfs_ifork_t *ifp, /* inode fork pointer */
4193 xfs_extnum_t nextents) /* number of extents in file */
4194{
4195 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4196 ASSERT(nextents <= XFS_INLINE_EXTS);
4197 /*
4198 * The inline buffer was zeroed when we switched
4199 * from inline to direct extent allocation mode,
4200 * so we don't need to clear it here.
4201 */
4202 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
4203 nextents * sizeof(xfs_bmbt_rec_t));
Mandy Kirkconnellfe6c1e72006-06-09 14:51:25 +10004204 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004205 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
4206 ifp->if_real_bytes = 0;
4207}
4208
4209/*
4210 * Switch from inline buffer to linear (direct) extent records.
4211 * new_size should already be rounded up to the next power of 2
4212 * by the caller (when appropriate), so use new_size as it is.
4213 * However, since new_size may be rounded up, we can't update
4214 * if_bytes here. It is the caller's responsibility to update
4215 * if_bytes upon return.
4216 */
4217void
4218xfs_iext_inline_to_direct(
4219 xfs_ifork_t *ifp, /* inode fork pointer */
4220 int new_size) /* number of extents in file */
4221{
4222 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4223 kmem_alloc(new_size, KM_SLEEP);
4224 memset(ifp->if_u1.if_extents, 0, new_size);
4225 if (ifp->if_bytes) {
4226 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
4227 ifp->if_bytes);
4228 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4229 sizeof(xfs_bmbt_rec_t));
4230 }
4231 ifp->if_real_bytes = new_size;
4232}
4233
4234/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004235 * Resize an extent indirection array to new_size bytes.
4236 */
4237void
4238xfs_iext_realloc_indirect(
4239 xfs_ifork_t *ifp, /* inode fork pointer */
4240 int new_size) /* new indirection array size */
4241{
4242 int nlists; /* number of irec's (ex lists) */
4243 int size; /* current indirection array size */
4244
4245 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4246 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4247 size = nlists * sizeof(xfs_ext_irec_t);
4248 ASSERT(ifp->if_real_bytes);
4249 ASSERT((new_size >= 0) && (new_size != size));
4250 if (new_size == 0) {
4251 xfs_iext_destroy(ifp);
4252 } else {
4253 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
4254 kmem_realloc(ifp->if_u1.if_ext_irec,
4255 new_size, size, KM_SLEEP);
4256 }
4257}
4258
4259/*
4260 * Switch from indirection array to linear (direct) extent allocations.
4261 */
4262void
4263xfs_iext_indirect_to_direct(
4264 xfs_ifork_t *ifp) /* inode fork pointer */
4265{
4266 xfs_bmbt_rec_t *ep; /* extent record pointer */
4267 xfs_extnum_t nextents; /* number of extents in file */
4268 int size; /* size of file extents */
4269
4270 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4271 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4272 ASSERT(nextents <= XFS_LINEAR_EXTS);
4273 size = nextents * sizeof(xfs_bmbt_rec_t);
4274
4275 xfs_iext_irec_compact_full(ifp);
4276 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
4277
4278 ep = ifp->if_u1.if_ext_irec->er_extbuf;
4279 kmem_free(ifp->if_u1.if_ext_irec, sizeof(xfs_ext_irec_t));
4280 ifp->if_flags &= ~XFS_IFEXTIREC;
4281 ifp->if_u1.if_extents = ep;
4282 ifp->if_bytes = size;
4283 if (nextents < XFS_LINEAR_EXTS) {
4284 xfs_iext_realloc_direct(ifp, size);
4285 }
4286}
4287
4288/*
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004289 * Free incore file extents.
4290 */
4291void
4292xfs_iext_destroy(
4293 xfs_ifork_t *ifp) /* inode fork pointer */
4294{
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004295 if (ifp->if_flags & XFS_IFEXTIREC) {
4296 int erp_idx;
4297 int nlists;
4298
4299 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4300 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
4301 xfs_iext_irec_remove(ifp, erp_idx);
4302 }
4303 ifp->if_flags &= ~XFS_IFEXTIREC;
4304 } else if (ifp->if_real_bytes) {
Mandy Kirkconnell4eea22f2006-03-14 13:29:52 +11004305 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
4306 } else if (ifp->if_bytes) {
4307 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
4308 sizeof(xfs_bmbt_rec_t));
4309 }
4310 ifp->if_u1.if_extents = NULL;
4311 ifp->if_real_bytes = 0;
4312 ifp->if_bytes = 0;
4313}
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004314
4315/*
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004316 * Return a pointer to the extent record for file system block bno.
4317 */
4318xfs_bmbt_rec_t * /* pointer to found extent record */
4319xfs_iext_bno_to_ext(
4320 xfs_ifork_t *ifp, /* inode fork pointer */
4321 xfs_fileoff_t bno, /* block number to search for */
4322 xfs_extnum_t *idxp) /* index of target extent */
4323{
4324 xfs_bmbt_rec_t *base; /* pointer to first extent */
4325 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
4326 xfs_bmbt_rec_t *ep = NULL; /* pointer to target extent */
4327 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
Nathan Scottc41564b2006-03-29 08:55:14 +10004328 int high; /* upper boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004329 xfs_extnum_t idx = 0; /* index of target extent */
Nathan Scottc41564b2006-03-29 08:55:14 +10004330 int low; /* lower boundary in search */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004331 xfs_extnum_t nextents; /* number of file extents */
4332 xfs_fileoff_t startoff = 0; /* start offset of extent */
4333
4334 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4335 if (nextents == 0) {
4336 *idxp = 0;
4337 return NULL;
4338 }
4339 low = 0;
4340 if (ifp->if_flags & XFS_IFEXTIREC) {
4341 /* Find target extent list */
4342 int erp_idx = 0;
4343 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
4344 base = erp->er_extbuf;
4345 high = erp->er_extcount - 1;
4346 } else {
4347 base = ifp->if_u1.if_extents;
4348 high = nextents - 1;
4349 }
4350 /* Binary search extent records */
4351 while (low <= high) {
4352 idx = (low + high) >> 1;
4353 ep = base + idx;
4354 startoff = xfs_bmbt_get_startoff(ep);
4355 blockcount = xfs_bmbt_get_blockcount(ep);
4356 if (bno < startoff) {
4357 high = idx - 1;
4358 } else if (bno >= startoff + blockcount) {
4359 low = idx + 1;
4360 } else {
4361 /* Convert back to file-based extent index */
4362 if (ifp->if_flags & XFS_IFEXTIREC) {
4363 idx += erp->er_extoff;
4364 }
4365 *idxp = idx;
4366 return ep;
4367 }
4368 }
4369 /* Convert back to file-based extent index */
4370 if (ifp->if_flags & XFS_IFEXTIREC) {
4371 idx += erp->er_extoff;
4372 }
4373 if (bno >= startoff + blockcount) {
4374 if (++idx == nextents) {
4375 ep = NULL;
4376 } else {
4377 ep = xfs_iext_get_ext(ifp, idx);
4378 }
4379 }
4380 *idxp = idx;
4381 return ep;
4382}
4383
4384/*
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004385 * Return a pointer to the indirection array entry containing the
4386 * extent record for filesystem block bno. Store the index of the
4387 * target irec in *erp_idxp.
4388 */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004389xfs_ext_irec_t * /* pointer to found extent record */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004390xfs_iext_bno_to_irec(
4391 xfs_ifork_t *ifp, /* inode fork pointer */
4392 xfs_fileoff_t bno, /* block number to search for */
4393 int *erp_idxp) /* irec index of target ext list */
4394{
4395 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
4396 xfs_ext_irec_t *erp_next; /* next indirection array entry */
Mandy Kirkconnell8867bc92006-03-17 17:25:04 +11004397 int erp_idx; /* indirection array index */
Mandy Kirkconnell0293ce32006-03-14 13:30:23 +11004398 int nlists; /* number of extent irec's (lists) */
4399 int high; /* binary search upper limit */
4400 int low; /* binary search lower limit */
4401
4402 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4403 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4404 erp_idx = 0;
4405 low = 0;
4406 high = nlists - 1;
4407 while (low <= high) {
4408 erp_idx = (low + high) >> 1;
4409 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4410 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
4411 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
4412 high = erp_idx - 1;
4413 } else if (erp_next && bno >=
4414 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
4415 low = erp_idx + 1;
4416 } else {
4417 break;
4418 }
4419 }
4420 *erp_idxp = erp_idx;
4421 return erp;
4422}
4423
4424/*
4425 * Return a pointer to the indirection array entry containing the
4426 * extent record at file extent index *idxp. Store the index of the
4427 * target irec in *erp_idxp and store the page index of the target
4428 * extent record in *idxp.
4429 */
4430xfs_ext_irec_t *
4431xfs_iext_idx_to_irec(
4432 xfs_ifork_t *ifp, /* inode fork pointer */
4433 xfs_extnum_t *idxp, /* extent index (file -> page) */
4434 int *erp_idxp, /* pointer to target irec */
4435 int realloc) /* new bytes were just added */
4436{
4437 xfs_ext_irec_t *prev; /* pointer to previous irec */
4438 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
4439 int erp_idx; /* indirection array index */
4440 int nlists; /* number of irec's (ex lists) */
4441 int high; /* binary search upper limit */
4442 int low; /* binary search lower limit */
4443 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
4444
4445 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4446 ASSERT(page_idx >= 0 && page_idx <=
4447 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
4448 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4449 erp_idx = 0;
4450 low = 0;
4451 high = nlists - 1;
4452
4453 /* Binary search extent irec's */
4454 while (low <= high) {
4455 erp_idx = (low + high) >> 1;
4456 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4457 prev = erp_idx > 0 ? erp - 1 : NULL;
4458 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
4459 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
4460 high = erp_idx - 1;
4461 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
4462 (page_idx == erp->er_extoff + erp->er_extcount &&
4463 !realloc)) {
4464 low = erp_idx + 1;
4465 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
4466 erp->er_extcount == XFS_LINEAR_EXTS) {
4467 ASSERT(realloc);
4468 page_idx = 0;
4469 erp_idx++;
4470 erp = erp_idx < nlists ? erp + 1 : NULL;
4471 break;
4472 } else {
4473 page_idx -= erp->er_extoff;
4474 break;
4475 }
4476 }
4477 *idxp = page_idx;
4478 *erp_idxp = erp_idx;
4479 return(erp);
4480}
4481
4482/*
4483 * Allocate and initialize an indirection array once the space needed
4484 * for incore extents increases above XFS_IEXT_BUFSZ.
4485 */
4486void
4487xfs_iext_irec_init(
4488 xfs_ifork_t *ifp) /* inode fork pointer */
4489{
4490 xfs_ext_irec_t *erp; /* indirection array pointer */
4491 xfs_extnum_t nextents; /* number of extents in file */
4492
4493 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
4494 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4495 ASSERT(nextents <= XFS_LINEAR_EXTS);
4496
4497 erp = (xfs_ext_irec_t *)
4498 kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
4499
4500 if (nextents == 0) {
4501 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
4502 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4503 } else if (!ifp->if_real_bytes) {
4504 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
4505 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
4506 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
4507 }
4508 erp->er_extbuf = ifp->if_u1.if_extents;
4509 erp->er_extcount = nextents;
4510 erp->er_extoff = 0;
4511
4512 ifp->if_flags |= XFS_IFEXTIREC;
4513 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
4514 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
4515 ifp->if_u1.if_ext_irec = erp;
4516
4517 return;
4518}
4519
4520/*
4521 * Allocate and initialize a new entry in the indirection array.
4522 */
4523xfs_ext_irec_t *
4524xfs_iext_irec_new(
4525 xfs_ifork_t *ifp, /* inode fork pointer */
4526 int erp_idx) /* index for new irec */
4527{
4528 xfs_ext_irec_t *erp; /* indirection array pointer */
4529 int i; /* loop counter */
4530 int nlists; /* number of irec's (ex lists) */
4531
4532 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4533 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4534
4535 /* Resize indirection array */
4536 xfs_iext_realloc_indirect(ifp, ++nlists *
4537 sizeof(xfs_ext_irec_t));
4538 /*
4539 * Move records down in the array so the
4540 * new page can use erp_idx.
4541 */
4542 erp = ifp->if_u1.if_ext_irec;
4543 for (i = nlists - 1; i > erp_idx; i--) {
4544 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
4545 }
4546 ASSERT(i == erp_idx);
4547
4548 /* Initialize new extent record */
4549 erp = ifp->if_u1.if_ext_irec;
4550 erp[erp_idx].er_extbuf = (xfs_bmbt_rec_t *)
4551 kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
4552 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4553 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
4554 erp[erp_idx].er_extcount = 0;
4555 erp[erp_idx].er_extoff = erp_idx > 0 ?
4556 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
4557 return (&erp[erp_idx]);
4558}
4559
4560/*
4561 * Remove a record from the indirection array.
4562 */
4563void
4564xfs_iext_irec_remove(
4565 xfs_ifork_t *ifp, /* inode fork pointer */
4566 int erp_idx) /* irec index to remove */
4567{
4568 xfs_ext_irec_t *erp; /* indirection array pointer */
4569 int i; /* loop counter */
4570 int nlists; /* number of irec's (ex lists) */
4571
4572 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4573 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4574 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4575 if (erp->er_extbuf) {
4576 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
4577 -erp->er_extcount);
4578 kmem_free(erp->er_extbuf, XFS_IEXT_BUFSZ);
4579 }
4580 /* Compact extent records */
4581 erp = ifp->if_u1.if_ext_irec;
4582 for (i = erp_idx; i < nlists - 1; i++) {
4583 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
4584 }
4585 /*
4586 * Manually free the last extent record from the indirection
4587 * array. A call to xfs_iext_realloc_indirect() with a size
4588 * of zero would result in a call to xfs_iext_destroy() which
4589 * would in turn call this function again, creating a nasty
4590 * infinite loop.
4591 */
4592 if (--nlists) {
4593 xfs_iext_realloc_indirect(ifp,
4594 nlists * sizeof(xfs_ext_irec_t));
4595 } else {
4596 kmem_free(ifp->if_u1.if_ext_irec,
4597 sizeof(xfs_ext_irec_t));
4598 }
4599 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
4600}
4601
4602/*
4603 * This is called to clean up large amounts of unused memory allocated
4604 * by the indirection array. Before compacting anything though, verify
4605 * that the indirection array is still needed and switch back to the
4606 * linear extent list (or even the inline buffer) if possible. The
4607 * compaction policy is as follows:
4608 *
4609 * Full Compaction: Extents fit into a single page (or inline buffer)
4610 * Full Compaction: Extents occupy less than 10% of allocated space
4611 * Partial Compaction: Extents occupy > 10% and < 50% of allocated space
4612 * No Compaction: Extents occupy at least 50% of allocated space
4613 */
4614void
4615xfs_iext_irec_compact(
4616 xfs_ifork_t *ifp) /* inode fork pointer */
4617{
4618 xfs_extnum_t nextents; /* number of extents in file */
4619 int nlists; /* number of irec's (ex lists) */
4620
4621 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4622 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4623 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4624
4625 if (nextents == 0) {
4626 xfs_iext_destroy(ifp);
4627 } else if (nextents <= XFS_INLINE_EXTS) {
4628 xfs_iext_indirect_to_direct(ifp);
4629 xfs_iext_direct_to_inline(ifp, nextents);
4630 } else if (nextents <= XFS_LINEAR_EXTS) {
4631 xfs_iext_indirect_to_direct(ifp);
4632 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 3) {
4633 xfs_iext_irec_compact_full(ifp);
4634 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
4635 xfs_iext_irec_compact_pages(ifp);
4636 }
4637}
4638
4639/*
4640 * Combine extents from neighboring extent pages.
4641 */
4642void
4643xfs_iext_irec_compact_pages(
4644 xfs_ifork_t *ifp) /* inode fork pointer */
4645{
4646 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
4647 int erp_idx = 0; /* indirection array index */
4648 int nlists; /* number of irec's (ex lists) */
4649
4650 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4651 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4652 while (erp_idx < nlists - 1) {
4653 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4654 erp_next = erp + 1;
4655 if (erp_next->er_extcount <=
4656 (XFS_LINEAR_EXTS - erp->er_extcount)) {
4657 memmove(&erp->er_extbuf[erp->er_extcount],
4658 erp_next->er_extbuf, erp_next->er_extcount *
4659 sizeof(xfs_bmbt_rec_t));
4660 erp->er_extcount += erp_next->er_extcount;
4661 /*
4662 * Free page before removing extent record
4663 * so er_extoffs don't get modified in
4664 * xfs_iext_irec_remove.
4665 */
4666 kmem_free(erp_next->er_extbuf, XFS_IEXT_BUFSZ);
4667 erp_next->er_extbuf = NULL;
4668 xfs_iext_irec_remove(ifp, erp_idx + 1);
4669 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4670 } else {
4671 erp_idx++;
4672 }
4673 }
4674}
4675
4676/*
4677 * Fully compact the extent records managed by the indirection array.
4678 */
4679void
4680xfs_iext_irec_compact_full(
4681 xfs_ifork_t *ifp) /* inode fork pointer */
4682{
4683 xfs_bmbt_rec_t *ep, *ep_next; /* extent record pointers */
4684 xfs_ext_irec_t *erp, *erp_next; /* extent irec pointers */
4685 int erp_idx = 0; /* extent irec index */
4686 int ext_avail; /* empty entries in ex list */
4687 int ext_diff; /* number of exts to add */
4688 int nlists; /* number of irec's (ex lists) */
4689
4690 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4691 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4692 erp = ifp->if_u1.if_ext_irec;
4693 ep = &erp->er_extbuf[erp->er_extcount];
4694 erp_next = erp + 1;
4695 ep_next = erp_next->er_extbuf;
4696 while (erp_idx < nlists - 1) {
4697 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
4698 ext_diff = MIN(ext_avail, erp_next->er_extcount);
4699 memcpy(ep, ep_next, ext_diff * sizeof(xfs_bmbt_rec_t));
4700 erp->er_extcount += ext_diff;
4701 erp_next->er_extcount -= ext_diff;
4702 /* Remove next page */
4703 if (erp_next->er_extcount == 0) {
4704 /*
4705 * Free page before removing extent record
4706 * so er_extoffs don't get modified in
4707 * xfs_iext_irec_remove.
4708 */
4709 kmem_free(erp_next->er_extbuf,
4710 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4711 erp_next->er_extbuf = NULL;
4712 xfs_iext_irec_remove(ifp, erp_idx + 1);
4713 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4714 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4715 /* Update next page */
4716 } else {
4717 /* Move rest of page up to become next new page */
4718 memmove(erp_next->er_extbuf, ep_next,
4719 erp_next->er_extcount * sizeof(xfs_bmbt_rec_t));
4720 ep_next = erp_next->er_extbuf;
4721 memset(&ep_next[erp_next->er_extcount], 0,
4722 (XFS_LINEAR_EXTS - erp_next->er_extcount) *
4723 sizeof(xfs_bmbt_rec_t));
4724 }
4725 if (erp->er_extcount == XFS_LINEAR_EXTS) {
4726 erp_idx++;
4727 if (erp_idx < nlists)
4728 erp = &ifp->if_u1.if_ext_irec[erp_idx];
4729 else
4730 break;
4731 }
4732 ep = &erp->er_extbuf[erp->er_extcount];
4733 erp_next = erp + 1;
4734 ep_next = erp_next->er_extbuf;
4735 }
4736}
4737
4738/*
4739 * This is called to update the er_extoff field in the indirection
4740 * array when extents have been added or removed from one of the
4741 * extent lists. erp_idx contains the irec index to begin updating
4742 * at and ext_diff contains the number of extents that were added
4743 * or removed.
4744 */
4745void
4746xfs_iext_irec_update_extoffs(
4747 xfs_ifork_t *ifp, /* inode fork pointer */
4748 int erp_idx, /* irec index to update */
4749 int ext_diff) /* number of new extents */
4750{
4751 int i; /* loop counter */
4752 int nlists; /* number of irec's (ex lists */
4753
4754 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
4755 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
4756 for (i = erp_idx; i < nlists; i++) {
4757 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
4758 }
4759}