blob: 6b7ac8bdcac24331d8629e9ba7d8e83ffb35a6fa [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * 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"
29#include "xfs_dir.h"
30#include "xfs_dir2.h"
31#include "xfs_dmapi.h"
32#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_alloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_dir_sf.h"
37#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include "xfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110042#include "xfs_inode_item.h"
43#include "xfs_btree.h"
44#include "xfs_alloc.h"
45#include "xfs_ialloc.h"
46#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include "xfs_rw.h"
48#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include "xfs_utils.h"
50#include "xfs_dir2_trace.h"
51#include "xfs_quota.h"
52#include "xfs_mac.h"
53#include "xfs_acl.h"
54
55
56kmem_zone_t *xfs_ifork_zone;
57kmem_zone_t *xfs_inode_zone;
58kmem_zone_t *xfs_chashlist_zone;
59
60/*
61 * Used in xfs_itruncate(). This is the maximum number of extents
62 * freed from a file in a single transaction.
63 */
64#define XFS_ITRUNC_MAX_EXTENTS 2
65
66STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
67STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
68STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
69STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
70
71
72#ifdef DEBUG
73/*
74 * Make sure that the extents in the given memory buffer
75 * are valid.
76 */
77STATIC void
78xfs_validate_extents(
79 xfs_bmbt_rec_t *ep,
80 int nrecs,
81 int disk,
82 xfs_exntfmt_t fmt)
83{
84 xfs_bmbt_irec_t irec;
85 xfs_bmbt_rec_t rec;
86 int i;
87
88 for (i = 0; i < nrecs; i++) {
89 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);
97 ep++;
98 }
99}
100#else /* DEBUG */
101#define xfs_validate_extents(ep, nrecs, disk, fmt)
102#endif /* DEBUG */
103
104/*
105 * Check that none of the inode's in the buffer have a next
106 * unlinked field of 0.
107 */
108#if defined(DEBUG)
109void
110xfs_inobp_check(
111 xfs_mount_t *mp,
112 xfs_buf_t *bp)
113{
114 int i;
115 int j;
116 xfs_dinode_t *dip;
117
118 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
119
120 for (i = 0; i < j; i++) {
121 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
122 i * mp->m_sb.sb_inodesize);
123 if (!dip->di_next_unlinked) {
124 xfs_fs_cmn_err(CE_ALERT, mp,
125 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
126 bp);
127 ASSERT(dip->di_next_unlinked);
128 }
129 }
130}
131#endif
132
133/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 * This routine is called to map an inode number within a file
135 * system to the buffer containing the on-disk version of the
136 * inode. It returns a pointer to the buffer containing the
137 * on-disk inode in the bpp parameter, and in the dip parameter
138 * it returns a pointer to the on-disk inode within that buffer.
139 *
140 * If a non-zero error is returned, then the contents of bpp and
141 * dipp are undefined.
142 *
143 * Use xfs_imap() to determine the size and location of the
144 * buffer to read from disk.
145 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000146STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147xfs_inotobp(
148 xfs_mount_t *mp,
149 xfs_trans_t *tp,
150 xfs_ino_t ino,
151 xfs_dinode_t **dipp,
152 xfs_buf_t **bpp,
153 int *offset)
154{
155 int di_ok;
156 xfs_imap_t imap;
157 xfs_buf_t *bp;
158 int error;
159 xfs_dinode_t *dip;
160
161 /*
162 * Call the space managment code to find the location of the
163 * inode on disk.
164 */
165 imap.im_blkno = 0;
166 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
167 if (error != 0) {
168 cmn_err(CE_WARN,
169 "xfs_inotobp: xfs_imap() returned an "
170 "error %d on %s. Returning error.", error, mp->m_fsname);
171 return error;
172 }
173
174 /*
175 * If the inode number maps to a block outside the bounds of the
176 * file system then return NULL rather than calling read_buf
177 * and panicing when we get an error from the driver.
178 */
179 if ((imap.im_blkno + imap.im_len) >
180 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
181 cmn_err(CE_WARN,
Christoph Hellwigda1650a2005-11-02 10:21:35 +1100182 "xfs_inotobp: inode number (%llu + %d) maps to a block outside the bounds "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 "of the file system %s. Returning EINVAL.",
Christoph Hellwigda1650a2005-11-02 10:21:35 +1100184 (unsigned long long)imap.im_blkno,
185 imap.im_len, mp->m_fsname);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 return XFS_ERROR(EINVAL);
187 }
188
189 /*
190 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
191 * default to just a read_buf() call.
192 */
193 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
194 (int)imap.im_len, XFS_BUF_LOCK, &bp);
195
196 if (error) {
197 cmn_err(CE_WARN,
198 "xfs_inotobp: xfs_trans_read_buf() returned an "
199 "error %d on %s. Returning error.", error, mp->m_fsname);
200 return error;
201 }
202 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
203 di_ok =
204 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
205 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
206 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
207 XFS_RANDOM_ITOBP_INOTOBP))) {
208 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
209 xfs_trans_brelse(tp, bp);
210 cmn_err(CE_WARN,
211 "xfs_inotobp: XFS_TEST_ERROR() returned an "
212 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
213 return XFS_ERROR(EFSCORRUPTED);
214 }
215
216 xfs_inobp_check(mp, bp);
217
218 /*
219 * Set *dipp to point to the on-disk inode in the buffer.
220 */
221 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
222 *bpp = bp;
223 *offset = imap.im_boffset;
224 return 0;
225}
226
227
228/*
229 * This routine is called to map an inode to the buffer containing
230 * the on-disk version of the inode. It returns a pointer to the
231 * buffer containing the on-disk inode in the bpp parameter, and in
232 * the dip parameter it returns a pointer to the on-disk inode within
233 * that buffer.
234 *
235 * If a non-zero error is returned, then the contents of bpp and
236 * dipp are undefined.
237 *
238 * If the inode is new and has not yet been initialized, use xfs_imap()
239 * to determine the size and location of the buffer to read from disk.
240 * If the inode has already been mapped to its buffer and read in once,
241 * then use the mapping information stored in the inode rather than
242 * calling xfs_imap(). This allows us to avoid the overhead of looking
243 * at the inode btree for small block file systems (see xfs_dilocate()).
244 * We can tell whether the inode has been mapped in before by comparing
245 * its disk block address to 0. Only uninitialized inodes will have
246 * 0 for the disk block address.
247 */
248int
249xfs_itobp(
250 xfs_mount_t *mp,
251 xfs_trans_t *tp,
252 xfs_inode_t *ip,
253 xfs_dinode_t **dipp,
254 xfs_buf_t **bpp,
255 xfs_daddr_t bno)
256{
257 xfs_buf_t *bp;
258 int error;
259 xfs_imap_t imap;
260#ifdef __KERNEL__
261 int i;
262 int ni;
263#endif
264
265 if (ip->i_blkno == (xfs_daddr_t)0) {
266 /*
267 * Call the space management code to find the location of the
268 * inode on disk.
269 */
270 imap.im_blkno = bno;
271 error = xfs_imap(mp, tp, ip->i_ino, &imap, XFS_IMAP_LOOKUP);
272 if (error != 0) {
273 return error;
274 }
275
276 /*
277 * If the inode number maps to a block outside the bounds
278 * of the file system then return NULL rather than calling
279 * read_buf and panicing when we get an error from the
280 * driver.
281 */
282 if ((imap.im_blkno + imap.im_len) >
283 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
284#ifdef DEBUG
285 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
286 "(imap.im_blkno (0x%llx) "
287 "+ imap.im_len (0x%llx)) > "
288 " XFS_FSB_TO_BB(mp, "
289 "mp->m_sb.sb_dblocks) (0x%llx)",
290 (unsigned long long) imap.im_blkno,
291 (unsigned long long) imap.im_len,
292 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
293#endif /* DEBUG */
294 return XFS_ERROR(EINVAL);
295 }
296
297 /*
298 * Fill in the fields in the inode that will be used to
299 * map the inode to its buffer from now on.
300 */
301 ip->i_blkno = imap.im_blkno;
302 ip->i_len = imap.im_len;
303 ip->i_boffset = imap.im_boffset;
304 } else {
305 /*
306 * We've already mapped the inode once, so just use the
307 * mapping that we saved the first time.
308 */
309 imap.im_blkno = ip->i_blkno;
310 imap.im_len = ip->i_len;
311 imap.im_boffset = ip->i_boffset;
312 }
313 ASSERT(bno == 0 || bno == imap.im_blkno);
314
315 /*
316 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
317 * default to just a read_buf() call.
318 */
319 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
320 (int)imap.im_len, XFS_BUF_LOCK, &bp);
321
322 if (error) {
323#ifdef DEBUG
324 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
325 "xfs_trans_read_buf() returned error %d, "
326 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
327 error, (unsigned long long) imap.im_blkno,
328 (unsigned long long) imap.im_len);
329#endif /* DEBUG */
330 return error;
331 }
332#ifdef __KERNEL__
333 /*
334 * Validate the magic number and version of every inode in the buffer
335 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
336 */
337#ifdef DEBUG
338 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
339#else
340 ni = 1;
341#endif
342 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));
350 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
351 XFS_RANDOM_ITOBP_INOTOBP))) {
352#ifdef DEBUG
353 prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
354 mp->m_ddev_targp,
355 (unsigned long long)imap.im_blkno, i,
356 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
357#endif
358 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
359 mp, dip);
360 xfs_trans_brelse(tp, bp);
361 return XFS_ERROR(EFSCORRUPTED);
362 }
363 }
364#endif /* __KERNEL__ */
365
366 xfs_inobp_check(mp, bp);
367
368 /*
369 * Mark the buffer as an inode buffer now that it looks good
370 */
371 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
372
373 /*
374 * Set *dipp to point to the on-disk inode in the buffer.
375 */
376 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
377 *bpp = bp;
378 return 0;
379}
380
381/*
382 * Move inode type and inode format specific information from the
383 * on-disk inode to the in-core inode. For fifos, devs, and sockets
384 * this means set if_rdev to the proper value. For files, directories,
385 * and symlinks this means to bring in the in-line data or extent
386 * pointers. For a file in B-tree format, only the root is immediately
387 * brought in-core. The rest will be in-lined in if_extents when it
388 * is first referenced (see xfs_iread_extents()).
389 */
390STATIC int
391xfs_iformat(
392 xfs_inode_t *ip,
393 xfs_dinode_t *dip)
394{
395 xfs_attr_shortform_t *atp;
396 int size;
397 int error;
398 xfs_fsize_t di_size;
399 ip->i_df.if_ext_max =
400 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
401 error = 0;
402
403 if (unlikely(
404 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
405 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
406 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
407 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
408 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu."
409 " Unmount and run xfs_repair.",
410 (unsigned long long)ip->i_ino,
411 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
412 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
413 (unsigned long long)
414 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
415 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
416 ip->i_mount, dip);
417 return XFS_ERROR(EFSCORRUPTED);
418 }
419
420 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
421 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
422 "corrupt dinode %Lu, forkoff = 0x%x."
423 " Unmount and run xfs_repair.",
424 (unsigned long long)ip->i_ino,
425 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
426 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
427 ip->i_mount, dip);
428 return XFS_ERROR(EFSCORRUPTED);
429 }
430
431 switch (ip->i_d.di_mode & S_IFMT) {
432 case S_IFIFO:
433 case S_IFCHR:
434 case S_IFBLK:
435 case S_IFSOCK:
436 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
437 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
438 ip->i_mount, dip);
439 return XFS_ERROR(EFSCORRUPTED);
440 }
441 ip->i_d.di_size = 0;
442 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
443 break;
444
445 case S_IFREG:
446 case S_IFLNK:
447 case S_IFDIR:
448 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
449 case XFS_DINODE_FMT_LOCAL:
450 /*
451 * no local regular files yet
452 */
453 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
454 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
455 "corrupt inode (local format for regular file) %Lu. Unmount and run xfs_repair.",
456 (unsigned long long) ip->i_ino);
457 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
458 XFS_ERRLEVEL_LOW,
459 ip->i_mount, dip);
460 return XFS_ERROR(EFSCORRUPTED);
461 }
462
463 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
464 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
465 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
466 "corrupt inode %Lu (bad size %Ld for local inode). Unmount and run xfs_repair.",
467 (unsigned long long) ip->i_ino,
468 (long long) di_size);
469 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
470 XFS_ERRLEVEL_LOW,
471 ip->i_mount, dip);
472 return XFS_ERROR(EFSCORRUPTED);
473 }
474
475 size = (int)di_size;
476 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
477 break;
478 case XFS_DINODE_FMT_EXTENTS:
479 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
480 break;
481 case XFS_DINODE_FMT_BTREE:
482 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
483 break;
484 default:
485 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
486 ip->i_mount);
487 return XFS_ERROR(EFSCORRUPTED);
488 }
489 break;
490
491 default:
492 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
493 return XFS_ERROR(EFSCORRUPTED);
494 }
495 if (error) {
496 return error;
497 }
498 if (!XFS_DFORK_Q(dip))
499 return 0;
500 ASSERT(ip->i_afp == NULL);
501 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
502 ip->i_afp->if_ext_max =
503 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
504 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
505 case XFS_DINODE_FMT_LOCAL:
506 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
507 size = (int)INT_GET(atp->hdr.totsize, ARCH_CONVERT);
508 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
509 break;
510 case XFS_DINODE_FMT_EXTENTS:
511 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
512 break;
513 case XFS_DINODE_FMT_BTREE:
514 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
515 break;
516 default:
517 error = XFS_ERROR(EFSCORRUPTED);
518 break;
519 }
520 if (error) {
521 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
522 ip->i_afp = NULL;
523 xfs_idestroy_fork(ip, XFS_DATA_FORK);
524 }
525 return error;
526}
527
528/*
529 * The file is in-lined in the on-disk inode.
530 * If it fits into if_inline_data, then copy
531 * it there, otherwise allocate a buffer for it
532 * and copy the data there. Either way, set
533 * if_data to point at the data.
534 * If we allocate a buffer for the data, make
535 * sure that its size is a multiple of 4 and
536 * record the real size in i_real_bytes.
537 */
538STATIC int
539xfs_iformat_local(
540 xfs_inode_t *ip,
541 xfs_dinode_t *dip,
542 int whichfork,
543 int size)
544{
545 xfs_ifork_t *ifp;
546 int real_size;
547
548 /*
549 * If the size is unreasonable, then something
550 * is wrong and we just bail out rather than crash in
551 * kmem_alloc() or memcpy() below.
552 */
553 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
554 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
555 "corrupt inode %Lu (bad size %d for local fork, size = %d). Unmount and run xfs_repair.",
556 (unsigned long long) ip->i_ino, size,
557 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
558 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
559 ip->i_mount, dip);
560 return XFS_ERROR(EFSCORRUPTED);
561 }
562 ifp = XFS_IFORK_PTR(ip, whichfork);
563 real_size = 0;
564 if (size == 0)
565 ifp->if_u1.if_data = NULL;
566 else if (size <= sizeof(ifp->if_u2.if_inline_data))
567 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
568 else {
569 real_size = roundup(size, 4);
570 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
571 }
572 ifp->if_bytes = size;
573 ifp->if_real_bytes = real_size;
574 if (size)
575 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
576 ifp->if_flags &= ~XFS_IFEXTENTS;
577 ifp->if_flags |= XFS_IFINLINE;
578 return 0;
579}
580
581/*
582 * The file consists of a set of extents all
583 * of which fit into the on-disk inode.
584 * If there are few enough extents to fit into
585 * the if_inline_ext, then copy them there.
586 * Otherwise allocate a buffer for them and copy
587 * them into it. Either way, set if_extents
588 * to point at the extents.
589 */
590STATIC int
591xfs_iformat_extents(
592 xfs_inode_t *ip,
593 xfs_dinode_t *dip,
594 int whichfork)
595{
596 xfs_bmbt_rec_t *ep, *dp;
597 xfs_ifork_t *ifp;
598 int nex;
599 int real_size;
600 int size;
601 int i;
602
603 ifp = XFS_IFORK_PTR(ip, whichfork);
604 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
605 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
606
607 /*
608 * If the number of extents is unreasonable, then something
609 * is wrong and we just bail out rather than crash in
610 * kmem_alloc() or memcpy() below.
611 */
612 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
613 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
614 "corrupt inode %Lu ((a)extents = %d). Unmount and run xfs_repair.",
615 (unsigned long long) ip->i_ino, nex);
616 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
617 ip->i_mount, dip);
618 return XFS_ERROR(EFSCORRUPTED);
619 }
620
621 real_size = 0;
622 if (nex == 0)
623 ifp->if_u1.if_extents = NULL;
624 else if (nex <= XFS_INLINE_EXTS)
625 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
626 else {
627 ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP);
628 ASSERT(ifp->if_u1.if_extents != NULL);
629 real_size = size;
630 }
631 ifp->if_bytes = size;
632 ifp->if_real_bytes = real_size;
633 if (size) {
634 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
635 xfs_validate_extents(dp, nex, 1, XFS_EXTFMT_INODE(ip));
636 ep = ifp->if_u1.if_extents;
637 for (i = 0; i < nex; i++, ep++, dp++) {
638 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
639 ARCH_CONVERT);
640 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
641 ARCH_CONVERT);
642 }
643 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
644 whichfork);
645 if (whichfork != XFS_DATA_FORK ||
646 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
647 if (unlikely(xfs_check_nostate_extents(
648 ifp->if_u1.if_extents, nex))) {
649 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
650 XFS_ERRLEVEL_LOW,
651 ip->i_mount);
652 return XFS_ERROR(EFSCORRUPTED);
653 }
654 }
655 ifp->if_flags |= XFS_IFEXTENTS;
656 return 0;
657}
658
659/*
660 * The file has too many extents to fit into
661 * the inode, so they are in B-tree format.
662 * Allocate a buffer for the root of the B-tree
663 * and copy the root into it. The i_extents
664 * field will remain NULL until all of the
665 * extents are read in (when they are needed).
666 */
667STATIC int
668xfs_iformat_btree(
669 xfs_inode_t *ip,
670 xfs_dinode_t *dip,
671 int whichfork)
672{
673 xfs_bmdr_block_t *dfp;
674 xfs_ifork_t *ifp;
675 /* REFERENCED */
676 int nrecs;
677 int size;
678
679 ifp = XFS_IFORK_PTR(ip, whichfork);
680 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
681 size = XFS_BMAP_BROOT_SPACE(dfp);
682 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
683
684 /*
685 * blow out if -- fork has less extents than can fit in
686 * fork (fork shouldn't be a btree format), root btree
687 * block has more records than can fit into the fork,
688 * or the number of extents is greater than the number of
689 * blocks.
690 */
691 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
692 || XFS_BMDR_SPACE_CALC(nrecs) >
693 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
694 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
695 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
696 "corrupt inode %Lu (btree). Unmount and run xfs_repair.",
697 (unsigned long long) ip->i_ino);
698 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
699 ip->i_mount);
700 return XFS_ERROR(EFSCORRUPTED);
701 }
702
703 ifp->if_broot_bytes = size;
704 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
705 ASSERT(ifp->if_broot != NULL);
706 /*
707 * Copy and convert from the on-disk structure
708 * to the in-memory structure.
709 */
710 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
711 ifp->if_broot, size);
712 ifp->if_flags &= ~XFS_IFEXTENTS;
713 ifp->if_flags |= XFS_IFBROOT;
714
715 return 0;
716}
717
718/*
719 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
720 * and native format
721 *
722 * buf = on-disk representation
723 * dip = native representation
724 * dir = direction - +ve -> disk to native
725 * -ve -> native to disk
726 */
727void
728xfs_xlate_dinode_core(
729 xfs_caddr_t buf,
730 xfs_dinode_core_t *dip,
731 int dir)
732{
733 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
734 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
735 xfs_arch_t arch = ARCH_CONVERT;
736
737 ASSERT(dir);
738
739 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
740 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
741 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
742 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
743 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
744 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
745 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
746 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
747 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
748
749 if (dir > 0) {
750 memcpy(mem_core->di_pad, buf_core->di_pad,
751 sizeof(buf_core->di_pad));
752 } else {
753 memcpy(buf_core->di_pad, mem_core->di_pad,
754 sizeof(buf_core->di_pad));
755 }
756
757 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
758
759 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
760 dir, arch);
761 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
762 dir, arch);
763 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
764 dir, arch);
765 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
766 dir, arch);
767 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
768 dir, arch);
769 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
770 dir, arch);
771 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
772 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
773 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
774 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
775 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
776 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
777 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
778 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
779 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
780 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
781 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
782}
783
784STATIC uint
785_xfs_dic2xflags(
786 xfs_dinode_core_t *dic,
787 __uint16_t di_flags)
788{
789 uint flags = 0;
790
791 if (di_flags & XFS_DIFLAG_ANY) {
792 if (di_flags & XFS_DIFLAG_REALTIME)
793 flags |= XFS_XFLAG_REALTIME;
794 if (di_flags & XFS_DIFLAG_PREALLOC)
795 flags |= XFS_XFLAG_PREALLOC;
796 if (di_flags & XFS_DIFLAG_IMMUTABLE)
797 flags |= XFS_XFLAG_IMMUTABLE;
798 if (di_flags & XFS_DIFLAG_APPEND)
799 flags |= XFS_XFLAG_APPEND;
800 if (di_flags & XFS_DIFLAG_SYNC)
801 flags |= XFS_XFLAG_SYNC;
802 if (di_flags & XFS_DIFLAG_NOATIME)
803 flags |= XFS_XFLAG_NOATIME;
804 if (di_flags & XFS_DIFLAG_NODUMP)
805 flags |= XFS_XFLAG_NODUMP;
806 if (di_flags & XFS_DIFLAG_RTINHERIT)
807 flags |= XFS_XFLAG_RTINHERIT;
808 if (di_flags & XFS_DIFLAG_PROJINHERIT)
809 flags |= XFS_XFLAG_PROJINHERIT;
810 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
811 flags |= XFS_XFLAG_NOSYMLINKS;
Nathan Scottdd9f4382006-01-11 15:28:28 +1100812 if (di_flags & XFS_DIFLAG_EXTSIZE)
813 flags |= XFS_XFLAG_EXTSIZE;
814 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
815 flags |= XFS_XFLAG_EXTSZINHERIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 }
817
818 return flags;
819}
820
821uint
822xfs_ip2xflags(
823 xfs_inode_t *ip)
824{
825 xfs_dinode_core_t *dic = &ip->i_d;
826
827 return _xfs_dic2xflags(dic, dic->di_flags) |
828 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
829}
830
831uint
832xfs_dic2xflags(
833 xfs_dinode_core_t *dic)
834{
835 return _xfs_dic2xflags(dic, INT_GET(dic->di_flags, ARCH_CONVERT)) |
836 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
837}
838
839/*
840 * Given a mount structure and an inode number, return a pointer
841 * to a newly allocated in-core inode coresponding to the given
842 * inode number.
843 *
844 * Initialize the inode's attributes and extent pointers if it
845 * already has them (it will not if the inode has no links).
846 */
847int
848xfs_iread(
849 xfs_mount_t *mp,
850 xfs_trans_t *tp,
851 xfs_ino_t ino,
852 xfs_inode_t **ipp,
853 xfs_daddr_t bno)
854{
855 xfs_buf_t *bp;
856 xfs_dinode_t *dip;
857 xfs_inode_t *ip;
858 int error;
859
860 ASSERT(xfs_inode_zone != NULL);
861
862 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
863 ip->i_ino = ino;
864 ip->i_mount = mp;
865
866 /*
867 * Get pointer's to the on-disk inode and the buffer containing it.
868 * If the inode number refers to a block outside the file system
869 * then xfs_itobp() will return NULL. In this case we should
870 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
871 * know that this is a new incore inode.
872 */
873 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno);
874
875 if (error != 0) {
876 kmem_zone_free(xfs_inode_zone, ip);
877 return error;
878 }
879
880 /*
881 * Initialize inode's trace buffers.
882 * Do this before xfs_iformat in case it adds entries.
883 */
884#ifdef XFS_BMAP_TRACE
885 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
886#endif
887#ifdef XFS_BMBT_TRACE
888 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
889#endif
890#ifdef XFS_RW_TRACE
891 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
892#endif
893#ifdef XFS_ILOCK_TRACE
894 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
895#endif
896#ifdef XFS_DIR2_TRACE
897 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
898#endif
899
900 /*
901 * If we got something that isn't an inode it means someone
902 * (nfs or dmi) has a stale handle.
903 */
904 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
905 kmem_zone_free(xfs_inode_zone, ip);
906 xfs_trans_brelse(tp, bp);
907#ifdef DEBUG
908 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
909 "dip->di_core.di_magic (0x%x) != "
910 "XFS_DINODE_MAGIC (0x%x)",
911 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
912 XFS_DINODE_MAGIC);
913#endif /* DEBUG */
914 return XFS_ERROR(EINVAL);
915 }
916
917 /*
918 * If the on-disk inode is already linked to a directory
919 * entry, copy all of the inode into the in-core inode.
920 * xfs_iformat() handles copying in the inode format
921 * specific information.
922 * Otherwise, just get the truly permanent information.
923 */
924 if (dip->di_core.di_mode) {
925 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
926 &(ip->i_d), 1);
927 error = xfs_iformat(ip, dip);
928 if (error) {
929 kmem_zone_free(xfs_inode_zone, ip);
930 xfs_trans_brelse(tp, bp);
931#ifdef DEBUG
932 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
933 "xfs_iformat() returned error %d",
934 error);
935#endif /* DEBUG */
936 return error;
937 }
938 } else {
939 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
940 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
941 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
942 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
943 /*
944 * Make sure to pull in the mode here as well in
945 * case the inode is released without being used.
946 * This ensures that xfs_inactive() will see that
947 * the inode is already free and not try to mess
948 * with the uninitialized part of it.
949 */
950 ip->i_d.di_mode = 0;
951 /*
952 * Initialize the per-fork minima and maxima for a new
953 * inode here. xfs_iformat will do it for old inodes.
954 */
955 ip->i_df.if_ext_max =
956 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
957 }
958
959 INIT_LIST_HEAD(&ip->i_reclaim);
960
961 /*
962 * The inode format changed when we moved the link count and
963 * made it 32 bits long. If this is an old format inode,
964 * convert it in memory to look like a new one. If it gets
965 * flushed to disk we will convert back before flushing or
966 * logging it. We zero out the new projid field and the old link
967 * count field. We'll handle clearing the pad field (the remains
968 * of the old uuid field) when we actually convert the inode to
969 * the new format. We don't change the version number so that we
970 * can distinguish this from a real new format inode.
971 */
972 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
973 ip->i_d.di_nlink = ip->i_d.di_onlink;
974 ip->i_d.di_onlink = 0;
975 ip->i_d.di_projid = 0;
976 }
977
978 ip->i_delayed_blks = 0;
979
980 /*
981 * Mark the buffer containing the inode as something to keep
982 * around for a while. This helps to keep recently accessed
983 * meta-data in-core longer.
984 */
985 XFS_BUF_SET_REF(bp, XFS_INO_REF);
986
987 /*
988 * Use xfs_trans_brelse() to release the buffer containing the
989 * on-disk inode, because it was acquired with xfs_trans_read_buf()
990 * in xfs_itobp() above. If tp is NULL, this is just a normal
991 * brelse(). If we're within a transaction, then xfs_trans_brelse()
992 * will only release the buffer if it is not dirty within the
993 * transaction. It will be OK to release the buffer in this case,
994 * because inodes on disk are never destroyed and we will be
995 * locking the new in-core inode before putting it in the hash
996 * table where other processes can find it. Thus we don't have
997 * to worry about the inode being changed just because we released
998 * the buffer.
999 */
1000 xfs_trans_brelse(tp, bp);
1001 *ipp = ip;
1002 return 0;
1003}
1004
1005/*
1006 * Read in extents from a btree-format inode.
1007 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1008 */
1009int
1010xfs_iread_extents(
1011 xfs_trans_t *tp,
1012 xfs_inode_t *ip,
1013 int whichfork)
1014{
1015 int error;
1016 xfs_ifork_t *ifp;
1017 size_t size;
1018
1019 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1020 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1021 ip->i_mount);
1022 return XFS_ERROR(EFSCORRUPTED);
1023 }
1024 size = XFS_IFORK_NEXTENTS(ip, whichfork) * (uint)sizeof(xfs_bmbt_rec_t);
1025 ifp = XFS_IFORK_PTR(ip, whichfork);
1026 /*
1027 * We know that the size is valid (it's checked in iformat_btree)
1028 */
1029 ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP);
1030 ASSERT(ifp->if_u1.if_extents != NULL);
1031 ifp->if_lastex = NULLEXTNUM;
1032 ifp->if_bytes = ifp->if_real_bytes = (int)size;
1033 ifp->if_flags |= XFS_IFEXTENTS;
1034 error = xfs_bmap_read_extents(tp, ip, whichfork);
1035 if (error) {
1036 kmem_free(ifp->if_u1.if_extents, size);
1037 ifp->if_u1.if_extents = NULL;
1038 ifp->if_bytes = ifp->if_real_bytes = 0;
1039 ifp->if_flags &= ~XFS_IFEXTENTS;
1040 return error;
1041 }
1042 xfs_validate_extents((xfs_bmbt_rec_t *)ifp->if_u1.if_extents,
1043 XFS_IFORK_NEXTENTS(ip, whichfork), 0, XFS_EXTFMT_INODE(ip));
1044 return 0;
1045}
1046
1047/*
1048 * Allocate an inode on disk and return a copy of its in-core version.
1049 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1050 * appropriately within the inode. The uid and gid for the inode are
1051 * set according to the contents of the given cred structure.
1052 *
1053 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1054 * has a free inode available, call xfs_iget()
1055 * to obtain the in-core version of the allocated inode. Finally,
1056 * fill in the inode and log its initial contents. In this case,
1057 * ialloc_context would be set to NULL and call_again set to false.
1058 *
1059 * If xfs_dialloc() does not have an available inode,
1060 * it will replenish its supply by doing an allocation. Since we can
1061 * only do one allocation within a transaction without deadlocks, we
1062 * must commit the current transaction before returning the inode itself.
1063 * In this case, therefore, we will set call_again to true and return.
1064 * The caller should then commit the current transaction, start a new
1065 * transaction, and call xfs_ialloc() again to actually get the inode.
1066 *
1067 * To ensure that some other process does not grab the inode that
1068 * was allocated during the first call to xfs_ialloc(), this routine
1069 * also returns the [locked] bp pointing to the head of the freelist
1070 * as ialloc_context. The caller should hold this buffer across
1071 * the commit and pass it back into this routine on the second call.
1072 */
1073int
1074xfs_ialloc(
1075 xfs_trans_t *tp,
1076 xfs_inode_t *pip,
1077 mode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -07001078 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 xfs_dev_t rdev,
1080 cred_t *cr,
1081 xfs_prid_t prid,
1082 int okalloc,
1083 xfs_buf_t **ialloc_context,
1084 boolean_t *call_again,
1085 xfs_inode_t **ipp)
1086{
1087 xfs_ino_t ino;
1088 xfs_inode_t *ip;
1089 vnode_t *vp;
1090 uint flags;
1091 int error;
1092
1093 /*
1094 * Call the space management code to pick
1095 * the on-disk inode to be allocated.
1096 */
1097 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1098 ialloc_context, call_again, &ino);
1099 if (error != 0) {
1100 return error;
1101 }
1102 if (*call_again || ino == NULLFSINO) {
1103 *ipp = NULL;
1104 return 0;
1105 }
1106 ASSERT(*ialloc_context == NULL);
1107
1108 /*
1109 * Get the in-core inode with the lock held exclusively.
1110 * This is because we're setting fields here we need
1111 * to prevent others from looking at until we're done.
1112 */
1113 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1114 IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1115 if (error != 0) {
1116 return error;
1117 }
1118 ASSERT(ip != NULL);
1119
1120 vp = XFS_ITOV(ip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 ip->i_d.di_mode = (__uint16_t)mode;
1122 ip->i_d.di_onlink = 0;
1123 ip->i_d.di_nlink = nlink;
1124 ASSERT(ip->i_d.di_nlink == nlink);
1125 ip->i_d.di_uid = current_fsuid(cr);
1126 ip->i_d.di_gid = current_fsgid(cr);
1127 ip->i_d.di_projid = prid;
1128 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1129
1130 /*
1131 * If the superblock version is up to where we support new format
1132 * inodes and this is currently an old format inode, then change
1133 * the inode version number now. This way we only do the conversion
1134 * here rather than here and in the flush/logging code.
1135 */
1136 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1137 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1138 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1139 /*
1140 * We've already zeroed the old link count, the projid field,
1141 * and the pad field.
1142 */
1143 }
1144
1145 /*
1146 * Project ids won't be stored on disk if we are using a version 1 inode.
1147 */
1148 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1149 xfs_bump_ino_vers2(tp, ip);
1150
1151 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1152 ip->i_d.di_gid = pip->i_d.di_gid;
1153 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1154 ip->i_d.di_mode |= S_ISGID;
1155 }
1156 }
1157
1158 /*
1159 * If the group ID of the new file does not match the effective group
1160 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1161 * (and only if the irix_sgid_inherit compatibility variable is set).
1162 */
1163 if ((irix_sgid_inherit) &&
1164 (ip->i_d.di_mode & S_ISGID) &&
1165 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1166 ip->i_d.di_mode &= ~S_ISGID;
1167 }
1168
1169 ip->i_d.di_size = 0;
1170 ip->i_d.di_nextents = 0;
1171 ASSERT(ip->i_d.di_nblocks == 0);
1172 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1173 /*
1174 * di_gen will have been taken care of in xfs_iread.
1175 */
1176 ip->i_d.di_extsize = 0;
1177 ip->i_d.di_dmevmask = 0;
1178 ip->i_d.di_dmstate = 0;
1179 ip->i_d.di_flags = 0;
1180 flags = XFS_ILOG_CORE;
1181 switch (mode & S_IFMT) {
1182 case S_IFIFO:
1183 case S_IFCHR:
1184 case S_IFBLK:
1185 case S_IFSOCK:
1186 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1187 ip->i_df.if_u2.if_rdev = rdev;
1188 ip->i_df.if_flags = 0;
1189 flags |= XFS_ILOG_DEV;
1190 break;
1191 case S_IFREG:
1192 case S_IFDIR:
1193 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
Nathan Scott365ca832005-06-21 15:39:12 +10001194 uint di_flags = 0;
1195
1196 if ((mode & S_IFMT) == S_IFDIR) {
1197 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1198 di_flags |= XFS_DIFLAG_RTINHERIT;
Nathan Scottdd9f4382006-01-11 15:28:28 +11001199 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1200 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1201 ip->i_d.di_extsize = pip->i_d.di_extsize;
1202 }
1203 } else if ((mode & S_IFMT) == S_IFREG) {
Nathan Scott365ca832005-06-21 15:39:12 +10001204 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1205 di_flags |= XFS_DIFLAG_REALTIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1207 }
Nathan Scottdd9f4382006-01-11 15:28:28 +11001208 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1209 di_flags |= XFS_DIFLAG_EXTSIZE;
1210 ip->i_d.di_extsize = pip->i_d.di_extsize;
1211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 }
1213 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1214 xfs_inherit_noatime)
Nathan Scott365ca832005-06-21 15:39:12 +10001215 di_flags |= XFS_DIFLAG_NOATIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1217 xfs_inherit_nodump)
Nathan Scott365ca832005-06-21 15:39:12 +10001218 di_flags |= XFS_DIFLAG_NODUMP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1220 xfs_inherit_sync)
Nathan Scott365ca832005-06-21 15:39:12 +10001221 di_flags |= XFS_DIFLAG_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1223 xfs_inherit_nosymlinks)
Nathan Scott365ca832005-06-21 15:39:12 +10001224 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1225 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1226 di_flags |= XFS_DIFLAG_PROJINHERIT;
1227 ip->i_d.di_flags |= di_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 }
1229 /* FALLTHROUGH */
1230 case S_IFLNK:
1231 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1232 ip->i_df.if_flags = XFS_IFEXTENTS;
1233 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1234 ip->i_df.if_u1.if_extents = NULL;
1235 break;
1236 default:
1237 ASSERT(0);
1238 }
1239 /*
1240 * Attribute fork settings for new inode.
1241 */
1242 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1243 ip->i_d.di_anextents = 0;
1244
1245 /*
1246 * Log the new values stuffed into the inode.
1247 */
1248 xfs_trans_log_inode(tp, ip, flags);
1249
Christoph Hellwig0432dab2005-09-02 16:46:51 +10001250 /* now that we have an i_mode we can set Linux inode ops (& unlock) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 VFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1252
1253 *ipp = ip;
1254 return 0;
1255}
1256
1257/*
1258 * Check to make sure that there are no blocks allocated to the
1259 * file beyond the size of the file. We don't check this for
1260 * files with fixed size extents or real time extents, but we
1261 * at least do it for regular files.
1262 */
1263#ifdef DEBUG
1264void
1265xfs_isize_check(
1266 xfs_mount_t *mp,
1267 xfs_inode_t *ip,
1268 xfs_fsize_t isize)
1269{
1270 xfs_fileoff_t map_first;
1271 int nimaps;
1272 xfs_bmbt_irec_t imaps[2];
1273
1274 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1275 return;
1276
Nathan Scottdd9f4382006-01-11 15:28:28 +11001277 if (ip->i_d.di_flags & (XFS_DIFLAG_REALTIME | XFS_DIFLAG_EXTSIZE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 return;
1279
1280 nimaps = 2;
1281 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1282 /*
1283 * The filesystem could be shutting down, so bmapi may return
1284 * an error.
1285 */
1286 if (xfs_bmapi(NULL, ip, map_first,
1287 (XFS_B_TO_FSB(mp,
1288 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1289 map_first),
1290 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1291 NULL))
1292 return;
1293 ASSERT(nimaps == 1);
1294 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1295}
1296#endif /* DEBUG */
1297
1298/*
1299 * Calculate the last possible buffered byte in a file. This must
1300 * include data that was buffered beyond the EOF by the write code.
1301 * This also needs to deal with overflowing the xfs_fsize_t type
1302 * which can happen for sizes near the limit.
1303 *
1304 * We also need to take into account any blocks beyond the EOF. It
1305 * may be the case that they were buffered by a write which failed.
1306 * In that case the pages will still be in memory, but the inode size
1307 * will never have been updated.
1308 */
1309xfs_fsize_t
1310xfs_file_last_byte(
1311 xfs_inode_t *ip)
1312{
1313 xfs_mount_t *mp;
1314 xfs_fsize_t last_byte;
1315 xfs_fileoff_t last_block;
1316 xfs_fileoff_t size_last_block;
1317 int error;
1318
1319 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1320
1321 mp = ip->i_mount;
1322 /*
1323 * Only check for blocks beyond the EOF if the extents have
1324 * been read in. This eliminates the need for the inode lock,
1325 * and it also saves us from looking when it really isn't
1326 * necessary.
1327 */
1328 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1329 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1330 XFS_DATA_FORK);
1331 if (error) {
1332 last_block = 0;
1333 }
1334 } else {
1335 last_block = 0;
1336 }
1337 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1338 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1339
1340 last_byte = XFS_FSB_TO_B(mp, last_block);
1341 if (last_byte < 0) {
1342 return XFS_MAXIOFFSET(mp);
1343 }
1344 last_byte += (1 << mp->m_writeio_log);
1345 if (last_byte < 0) {
1346 return XFS_MAXIOFFSET(mp);
1347 }
1348 return last_byte;
1349}
1350
1351#if defined(XFS_RW_TRACE)
1352STATIC void
1353xfs_itrunc_trace(
1354 int tag,
1355 xfs_inode_t *ip,
1356 int flag,
1357 xfs_fsize_t new_size,
1358 xfs_off_t toss_start,
1359 xfs_off_t toss_finish)
1360{
1361 if (ip->i_rwtrace == NULL) {
1362 return;
1363 }
1364
1365 ktrace_enter(ip->i_rwtrace,
1366 (void*)((long)tag),
1367 (void*)ip,
1368 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1369 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1370 (void*)((long)flag),
1371 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1372 (void*)(unsigned long)(new_size & 0xffffffff),
1373 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1374 (void*)(unsigned long)(toss_start & 0xffffffff),
1375 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1376 (void*)(unsigned long)(toss_finish & 0xffffffff),
1377 (void*)(unsigned long)current_cpu(),
1378 (void*)0,
1379 (void*)0,
1380 (void*)0,
1381 (void*)0);
1382}
1383#else
1384#define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1385#endif
1386
1387/*
1388 * Start the truncation of the file to new_size. The new size
1389 * must be smaller than the current size. This routine will
1390 * clear the buffer and page caches of file data in the removed
1391 * range, and xfs_itruncate_finish() will remove the underlying
1392 * disk blocks.
1393 *
1394 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1395 * must NOT have the inode lock held at all. This is because we're
1396 * calling into the buffer/page cache code and we can't hold the
1397 * inode lock when we do so.
1398 *
1399 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1400 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1401 * in the case that the caller is locking things out of order and
1402 * may not be able to call xfs_itruncate_finish() with the inode lock
1403 * held without dropping the I/O lock. If the caller must drop the
1404 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1405 * must be called again with all the same restrictions as the initial
1406 * call.
1407 */
1408void
1409xfs_itruncate_start(
1410 xfs_inode_t *ip,
1411 uint flags,
1412 xfs_fsize_t new_size)
1413{
1414 xfs_fsize_t last_byte;
1415 xfs_off_t toss_start;
1416 xfs_mount_t *mp;
1417 vnode_t *vp;
1418
1419 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1420 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1421 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1422 (flags == XFS_ITRUNC_MAYBE));
1423
1424 mp = ip->i_mount;
1425 vp = XFS_ITOV(ip);
1426 /*
1427 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1428 * overlapping the region being removed. We have to use
1429 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1430 * caller may not be able to finish the truncate without
1431 * dropping the inode's I/O lock. Make sure
1432 * to catch any pages brought in by buffers overlapping
1433 * the EOF by searching out beyond the isize by our
1434 * block size. We round new_size up to a block boundary
1435 * so that we don't toss things on the same block as
1436 * new_size but before it.
1437 *
1438 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1439 * call remapf() over the same region if the file is mapped.
1440 * This frees up mapped file references to the pages in the
1441 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1442 * that we get the latest mapped changes flushed out.
1443 */
1444 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1445 toss_start = XFS_FSB_TO_B(mp, toss_start);
1446 if (toss_start < 0) {
1447 /*
1448 * The place to start tossing is beyond our maximum
1449 * file size, so there is no way that the data extended
1450 * out there.
1451 */
1452 return;
1453 }
1454 last_byte = xfs_file_last_byte(ip);
1455 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1456 last_byte);
1457 if (last_byte > toss_start) {
1458 if (flags & XFS_ITRUNC_DEFINITE) {
1459 VOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1460 } else {
1461 VOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1462 }
1463 }
1464
1465#ifdef DEBUG
1466 if (new_size == 0) {
1467 ASSERT(VN_CACHED(vp) == 0);
1468 }
1469#endif
1470}
1471
1472/*
1473 * Shrink the file to the given new_size. The new
1474 * size must be smaller than the current size.
1475 * This will free up the underlying blocks
1476 * in the removed range after a call to xfs_itruncate_start()
1477 * or xfs_atruncate_start().
1478 *
1479 * The transaction passed to this routine must have made
1480 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1481 * This routine may commit the given transaction and
1482 * start new ones, so make sure everything involved in
1483 * the transaction is tidy before calling here.
1484 * Some transaction will be returned to the caller to be
1485 * committed. The incoming transaction must already include
1486 * the inode, and both inode locks must be held exclusively.
1487 * The inode must also be "held" within the transaction. On
1488 * return the inode will be "held" within the returned transaction.
1489 * This routine does NOT require any disk space to be reserved
1490 * for it within the transaction.
1491 *
1492 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1493 * and it indicates the fork which is to be truncated. For the
1494 * attribute fork we only support truncation to size 0.
1495 *
1496 * We use the sync parameter to indicate whether or not the first
1497 * transaction we perform might have to be synchronous. For the attr fork,
1498 * it needs to be so if the unlink of the inode is not yet known to be
1499 * permanent in the log. This keeps us from freeing and reusing the
1500 * blocks of the attribute fork before the unlink of the inode becomes
1501 * permanent.
1502 *
1503 * For the data fork, we normally have to run synchronously if we're
1504 * being called out of the inactive path or we're being called
1505 * out of the create path where we're truncating an existing file.
1506 * Either way, the truncate needs to be sync so blocks don't reappear
1507 * in the file with altered data in case of a crash. wsync filesystems
1508 * can run the first case async because anything that shrinks the inode
1509 * has to run sync so by the time we're called here from inactive, the
1510 * inode size is permanently set to 0.
1511 *
1512 * Calls from the truncate path always need to be sync unless we're
1513 * in a wsync filesystem and the file has already been unlinked.
1514 *
1515 * The caller is responsible for correctly setting the sync parameter.
1516 * It gets too hard for us to guess here which path we're being called
1517 * out of just based on inode state.
1518 */
1519int
1520xfs_itruncate_finish(
1521 xfs_trans_t **tp,
1522 xfs_inode_t *ip,
1523 xfs_fsize_t new_size,
1524 int fork,
1525 int sync)
1526{
1527 xfs_fsblock_t first_block;
1528 xfs_fileoff_t first_unmap_block;
1529 xfs_fileoff_t last_block;
1530 xfs_filblks_t unmap_len=0;
1531 xfs_mount_t *mp;
1532 xfs_trans_t *ntp;
1533 int done;
1534 int committed;
1535 xfs_bmap_free_t free_list;
1536 int error;
1537
1538 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1539 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1540 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1541 ASSERT(*tp != NULL);
1542 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1543 ASSERT(ip->i_transp == *tp);
1544 ASSERT(ip->i_itemp != NULL);
1545 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1546
1547
1548 ntp = *tp;
1549 mp = (ntp)->t_mountp;
1550 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1551
1552 /*
1553 * We only support truncating the entire attribute fork.
1554 */
1555 if (fork == XFS_ATTR_FORK) {
1556 new_size = 0LL;
1557 }
1558 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1559 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1560 /*
1561 * The first thing we do is set the size to new_size permanently
1562 * on disk. This way we don't have to worry about anyone ever
1563 * being able to look at the data being freed even in the face
1564 * of a crash. What we're getting around here is the case where
1565 * we free a block, it is allocated to another file, it is written
1566 * to, and then we crash. If the new data gets written to the
1567 * file but the log buffers containing the free and reallocation
1568 * don't, then we'd end up with garbage in the blocks being freed.
1569 * As long as we make the new_size permanent before actually
1570 * freeing any blocks it doesn't matter if they get writtten to.
1571 *
1572 * The callers must signal into us whether or not the size
1573 * setting here must be synchronous. There are a few cases
1574 * where it doesn't have to be synchronous. Those cases
1575 * occur if the file is unlinked and we know the unlink is
1576 * permanent or if the blocks being truncated are guaranteed
1577 * to be beyond the inode eof (regardless of the link count)
1578 * and the eof value is permanent. Both of these cases occur
1579 * only on wsync-mounted filesystems. In those cases, we're
1580 * guaranteed that no user will ever see the data in the blocks
1581 * that are being truncated so the truncate can run async.
1582 * In the free beyond eof case, the file may wind up with
1583 * more blocks allocated to it than it needs if we crash
1584 * and that won't get fixed until the next time the file
1585 * is re-opened and closed but that's ok as that shouldn't
1586 * be too many blocks.
1587 *
1588 * However, we can't just make all wsync xactions run async
1589 * because there's one call out of the create path that needs
1590 * to run sync where it's truncating an existing file to size
1591 * 0 whose size is > 0.
1592 *
1593 * It's probably possible to come up with a test in this
1594 * routine that would correctly distinguish all the above
1595 * cases from the values of the function parameters and the
1596 * inode state but for sanity's sake, I've decided to let the
1597 * layers above just tell us. It's simpler to correctly figure
1598 * out in the layer above exactly under what conditions we
1599 * can run async and I think it's easier for others read and
1600 * follow the logic in case something has to be changed.
1601 * cscope is your friend -- rcc.
1602 *
1603 * The attribute fork is much simpler.
1604 *
1605 * For the attribute fork we allow the caller to tell us whether
1606 * the unlink of the inode that led to this call is yet permanent
1607 * in the on disk log. If it is not and we will be freeing extents
1608 * in this inode then we make the first transaction synchronous
1609 * to make sure that the unlink is permanent by the time we free
1610 * the blocks.
1611 */
1612 if (fork == XFS_DATA_FORK) {
1613 if (ip->i_d.di_nextents > 0) {
1614 ip->i_d.di_size = new_size;
1615 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1616 }
1617 } else if (sync) {
1618 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1619 if (ip->i_d.di_anextents > 0)
1620 xfs_trans_set_sync(ntp);
1621 }
1622 ASSERT(fork == XFS_DATA_FORK ||
1623 (fork == XFS_ATTR_FORK &&
1624 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1625 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1626
1627 /*
1628 * Since it is possible for space to become allocated beyond
1629 * the end of the file (in a crash where the space is allocated
1630 * but the inode size is not yet updated), simply remove any
1631 * blocks which show up between the new EOF and the maximum
1632 * possible file size. If the first block to be removed is
1633 * beyond the maximum file size (ie it is the same as last_block),
1634 * then there is nothing to do.
1635 */
1636 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1637 ASSERT(first_unmap_block <= last_block);
1638 done = 0;
1639 if (last_block == first_unmap_block) {
1640 done = 1;
1641 } else {
1642 unmap_len = last_block - first_unmap_block + 1;
1643 }
1644 while (!done) {
1645 /*
1646 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1647 * will tell us whether it freed the entire range or
1648 * not. If this is a synchronous mount (wsync),
1649 * then we can tell bunmapi to keep all the
1650 * transactions asynchronous since the unlink
1651 * transaction that made this inode inactive has
1652 * already hit the disk. There's no danger of
1653 * the freed blocks being reused, there being a
1654 * crash, and the reused blocks suddenly reappearing
1655 * in this file with garbage in them once recovery
1656 * runs.
1657 */
1658 XFS_BMAP_INIT(&free_list, &first_block);
1659 error = xfs_bunmapi(ntp, ip, first_unmap_block,
1660 unmap_len,
1661 XFS_BMAPI_AFLAG(fork) |
1662 (sync ? 0 : XFS_BMAPI_ASYNC),
1663 XFS_ITRUNC_MAX_EXTENTS,
1664 &first_block, &free_list, &done);
1665 if (error) {
1666 /*
1667 * If the bunmapi call encounters an error,
1668 * return to the caller where the transaction
1669 * can be properly aborted. We just need to
1670 * make sure we're not holding any resources
1671 * that we were not when we came in.
1672 */
1673 xfs_bmap_cancel(&free_list);
1674 return error;
1675 }
1676
1677 /*
1678 * Duplicate the transaction that has the permanent
1679 * reservation and commit the old transaction.
1680 */
1681 error = xfs_bmap_finish(tp, &free_list, first_block,
1682 &committed);
1683 ntp = *tp;
1684 if (error) {
1685 /*
1686 * If the bmap finish call encounters an error,
1687 * return to the caller where the transaction
1688 * can be properly aborted. We just need to
1689 * make sure we're not holding any resources
1690 * that we were not when we came in.
1691 *
1692 * Aborting from this point might lose some
1693 * blocks in the file system, but oh well.
1694 */
1695 xfs_bmap_cancel(&free_list);
1696 if (committed) {
1697 /*
1698 * If the passed in transaction committed
1699 * in xfs_bmap_finish(), then we want to
1700 * add the inode to this one before returning.
1701 * This keeps things simple for the higher
1702 * level code, because it always knows that
1703 * the inode is locked and held in the
1704 * transaction that returns to it whether
1705 * errors occur or not. We don't mark the
1706 * inode dirty so that this transaction can
1707 * be easily aborted if possible.
1708 */
1709 xfs_trans_ijoin(ntp, ip,
1710 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1711 xfs_trans_ihold(ntp, ip);
1712 }
1713 return error;
1714 }
1715
1716 if (committed) {
1717 /*
1718 * The first xact was committed,
1719 * so add the inode to the new one.
1720 * Mark it dirty so it will be logged
1721 * and moved forward in the log as
1722 * part of every commit.
1723 */
1724 xfs_trans_ijoin(ntp, ip,
1725 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1726 xfs_trans_ihold(ntp, ip);
1727 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1728 }
1729 ntp = xfs_trans_dup(ntp);
1730 (void) xfs_trans_commit(*tp, 0, NULL);
1731 *tp = ntp;
1732 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1733 XFS_TRANS_PERM_LOG_RES,
1734 XFS_ITRUNCATE_LOG_COUNT);
1735 /*
1736 * Add the inode being truncated to the next chained
1737 * transaction.
1738 */
1739 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1740 xfs_trans_ihold(ntp, ip);
1741 if (error)
1742 return (error);
1743 }
1744 /*
1745 * Only update the size in the case of the data fork, but
1746 * always re-log the inode so that our permanent transaction
1747 * can keep on rolling it forward in the log.
1748 */
1749 if (fork == XFS_DATA_FORK) {
1750 xfs_isize_check(mp, ip, new_size);
1751 ip->i_d.di_size = new_size;
1752 }
1753 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1754 ASSERT((new_size != 0) ||
1755 (fork == XFS_ATTR_FORK) ||
1756 (ip->i_delayed_blks == 0));
1757 ASSERT((new_size != 0) ||
1758 (fork == XFS_ATTR_FORK) ||
1759 (ip->i_d.di_nextents == 0));
1760 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1761 return 0;
1762}
1763
1764
1765/*
1766 * xfs_igrow_start
1767 *
1768 * Do the first part of growing a file: zero any data in the last
1769 * block that is beyond the old EOF. We need to do this before
1770 * the inode is joined to the transaction to modify the i_size.
1771 * That way we can drop the inode lock and call into the buffer
1772 * cache to get the buffer mapping the EOF.
1773 */
1774int
1775xfs_igrow_start(
1776 xfs_inode_t *ip,
1777 xfs_fsize_t new_size,
1778 cred_t *credp)
1779{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 int error;
1781
1782 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1783 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1784 ASSERT(new_size > ip->i_d.di_size);
1785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 /*
1787 * Zero any pages that may have been created by
1788 * xfs_write_file() beyond the end of the file
1789 * and any blocks between the old and new file sizes.
1790 */
Eric Sandeen24ee8082006-01-11 15:34:32 +11001791 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size,
1792 ip->i_d.di_size, new_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 return error;
1794}
1795
1796/*
1797 * xfs_igrow_finish
1798 *
1799 * This routine is called to extend the size of a file.
1800 * The inode must have both the iolock and the ilock locked
1801 * for update and it must be a part of the current transaction.
1802 * The xfs_igrow_start() function must have been called previously.
1803 * If the change_flag is not zero, the inode change timestamp will
1804 * be updated.
1805 */
1806void
1807xfs_igrow_finish(
1808 xfs_trans_t *tp,
1809 xfs_inode_t *ip,
1810 xfs_fsize_t new_size,
1811 int change_flag)
1812{
1813 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1814 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1815 ASSERT(ip->i_transp == tp);
1816 ASSERT(new_size > ip->i_d.di_size);
1817
1818 /*
1819 * Update the file size. Update the inode change timestamp
1820 * if change_flag set.
1821 */
1822 ip->i_d.di_size = new_size;
1823 if (change_flag)
1824 xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
1825 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1826
1827}
1828
1829
1830/*
1831 * This is called when the inode's link count goes to 0.
1832 * We place the on-disk inode on a list in the AGI. It
1833 * will be pulled from this list when the inode is freed.
1834 */
1835int
1836xfs_iunlink(
1837 xfs_trans_t *tp,
1838 xfs_inode_t *ip)
1839{
1840 xfs_mount_t *mp;
1841 xfs_agi_t *agi;
1842 xfs_dinode_t *dip;
1843 xfs_buf_t *agibp;
1844 xfs_buf_t *ibp;
1845 xfs_agnumber_t agno;
1846 xfs_daddr_t agdaddr;
1847 xfs_agino_t agino;
1848 short bucket_index;
1849 int offset;
1850 int error;
1851 int agi_ok;
1852
1853 ASSERT(ip->i_d.di_nlink == 0);
1854 ASSERT(ip->i_d.di_mode != 0);
1855 ASSERT(ip->i_transp == tp);
1856
1857 mp = tp->t_mountp;
1858
1859 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1860 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1861
1862 /*
1863 * Get the agi buffer first. It ensures lock ordering
1864 * on the list.
1865 */
1866 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1867 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1868 if (error) {
1869 return error;
1870 }
1871 /*
1872 * Validate the magic number of the agi block.
1873 */
1874 agi = XFS_BUF_TO_AGI(agibp);
1875 agi_ok =
Christoph Hellwig16259e72005-11-02 15:11:25 +11001876 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1877 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK,
1879 XFS_RANDOM_IUNLINK))) {
1880 XFS_CORRUPTION_ERROR("xfs_iunlink", XFS_ERRLEVEL_LOW, mp, agi);
1881 xfs_trans_brelse(tp, agibp);
1882 return XFS_ERROR(EFSCORRUPTED);
1883 }
1884 /*
1885 * Get the index into the agi hash table for the
1886 * list this inode will go on.
1887 */
1888 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1889 ASSERT(agino != 0);
1890 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1891 ASSERT(agi->agi_unlinked[bucket_index]);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001892 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
Christoph Hellwig16259e72005-11-02 15:11:25 +11001894 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 /*
1896 * There is already another inode in the bucket we need
1897 * to add ourselves to. Add us at the front of the list.
1898 * Here we put the head pointer into our next pointer,
1899 * and then we fall through to point the head at us.
1900 */
1901 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
1902 if (error) {
1903 return error;
1904 }
1905 ASSERT(INT_GET(dip->di_next_unlinked, ARCH_CONVERT) == NULLAGINO);
1906 ASSERT(dip->di_next_unlinked);
1907 /* both on-disk, don't endian flip twice */
1908 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1909 offset = ip->i_boffset +
1910 offsetof(xfs_dinode_t, di_next_unlinked);
1911 xfs_trans_inode_buf(tp, ibp);
1912 xfs_trans_log_buf(tp, ibp, offset,
1913 (offset + sizeof(xfs_agino_t) - 1));
1914 xfs_inobp_check(mp, ibp);
1915 }
1916
1917 /*
1918 * Point the bucket head pointer at the inode being inserted.
1919 */
1920 ASSERT(agino != 0);
Christoph Hellwig16259e72005-11-02 15:11:25 +11001921 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 offset = offsetof(xfs_agi_t, agi_unlinked) +
1923 (sizeof(xfs_agino_t) * bucket_index);
1924 xfs_trans_log_buf(tp, agibp, offset,
1925 (offset + sizeof(xfs_agino_t) - 1));
1926 return 0;
1927}
1928
1929/*
1930 * Pull the on-disk inode from the AGI unlinked list.
1931 */
1932STATIC int
1933xfs_iunlink_remove(
1934 xfs_trans_t *tp,
1935 xfs_inode_t *ip)
1936{
1937 xfs_ino_t next_ino;
1938 xfs_mount_t *mp;
1939 xfs_agi_t *agi;
1940 xfs_dinode_t *dip;
1941 xfs_buf_t *agibp;
1942 xfs_buf_t *ibp;
1943 xfs_agnumber_t agno;
1944 xfs_daddr_t agdaddr;
1945 xfs_agino_t agino;
1946 xfs_agino_t next_agino;
1947 xfs_buf_t *last_ibp;
1948 xfs_dinode_t *last_dip;
1949 short bucket_index;
1950 int offset, last_offset;
1951 int error;
1952 int agi_ok;
1953
1954 /*
1955 * First pull the on-disk inode from the AGI unlinked list.
1956 */
1957 mp = tp->t_mountp;
1958
1959 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1960 agdaddr = XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp));
1961
1962 /*
1963 * Get the agi buffer first. It ensures lock ordering
1964 * on the list.
1965 */
1966 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, agdaddr,
1967 XFS_FSS_TO_BB(mp, 1), 0, &agibp);
1968 if (error) {
1969 cmn_err(CE_WARN,
1970 "xfs_iunlink_remove: xfs_trans_read_buf() returned an error %d on %s. Returning error.",
1971 error, mp->m_fsname);
1972 return error;
1973 }
1974 /*
1975 * Validate the magic number of the agi block.
1976 */
1977 agi = XFS_BUF_TO_AGI(agibp);
1978 agi_ok =
Christoph Hellwig16259e72005-11-02 15:11:25 +11001979 be32_to_cpu(agi->agi_magicnum) == XFS_AGI_MAGIC &&
1980 XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 if (unlikely(XFS_TEST_ERROR(!agi_ok, mp, XFS_ERRTAG_IUNLINK_REMOVE,
1982 XFS_RANDOM_IUNLINK_REMOVE))) {
1983 XFS_CORRUPTION_ERROR("xfs_iunlink_remove", XFS_ERRLEVEL_LOW,
1984 mp, agi);
1985 xfs_trans_brelse(tp, agibp);
1986 cmn_err(CE_WARN,
1987 "xfs_iunlink_remove: XFS_TEST_ERROR() returned an error on %s. Returning EFSCORRUPTED.",
1988 mp->m_fsname);
1989 return XFS_ERROR(EFSCORRUPTED);
1990 }
1991 /*
1992 * Get the index into the agi hash table for the
1993 * list this inode will go on.
1994 */
1995 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1996 ASSERT(agino != 0);
1997 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
Christoph Hellwig16259e72005-11-02 15:11:25 +11001998 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 ASSERT(agi->agi_unlinked[bucket_index]);
2000
Christoph Hellwig16259e72005-11-02 15:11:25 +11002001 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 /*
2003 * We're at the head of the list. Get the inode's
2004 * on-disk buffer to see if there is anyone after us
2005 * on the list. Only modify our next pointer if it
2006 * is not already NULLAGINO. This saves us the overhead
2007 * of dealing with the buffer when there is no need to
2008 * change it.
2009 */
2010 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
2011 if (error) {
2012 cmn_err(CE_WARN,
2013 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2014 error, mp->m_fsname);
2015 return error;
2016 }
2017 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2018 ASSERT(next_agino != 0);
2019 if (next_agino != NULLAGINO) {
2020 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2021 offset = ip->i_boffset +
2022 offsetof(xfs_dinode_t, di_next_unlinked);
2023 xfs_trans_inode_buf(tp, ibp);
2024 xfs_trans_log_buf(tp, ibp, offset,
2025 (offset + sizeof(xfs_agino_t) - 1));
2026 xfs_inobp_check(mp, ibp);
2027 } else {
2028 xfs_trans_brelse(tp, ibp);
2029 }
2030 /*
2031 * Point the bucket head pointer at the next inode.
2032 */
2033 ASSERT(next_agino != 0);
2034 ASSERT(next_agino != agino);
Christoph Hellwig16259e72005-11-02 15:11:25 +11002035 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 offset = offsetof(xfs_agi_t, agi_unlinked) +
2037 (sizeof(xfs_agino_t) * bucket_index);
2038 xfs_trans_log_buf(tp, agibp, offset,
2039 (offset + sizeof(xfs_agino_t) - 1));
2040 } else {
2041 /*
2042 * We need to search the list for the inode being freed.
2043 */
Christoph Hellwig16259e72005-11-02 15:11:25 +11002044 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 last_ibp = NULL;
2046 while (next_agino != agino) {
2047 /*
2048 * If the last inode wasn't the one pointing to
2049 * us, then release its buffer since we're not
2050 * going to do anything with it.
2051 */
2052 if (last_ibp != NULL) {
2053 xfs_trans_brelse(tp, last_ibp);
2054 }
2055 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2056 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
2057 &last_ibp, &last_offset);
2058 if (error) {
2059 cmn_err(CE_WARN,
2060 "xfs_iunlink_remove: xfs_inotobp() returned an error %d on %s. Returning error.",
2061 error, mp->m_fsname);
2062 return error;
2063 }
2064 next_agino = INT_GET(last_dip->di_next_unlinked, ARCH_CONVERT);
2065 ASSERT(next_agino != NULLAGINO);
2066 ASSERT(next_agino != 0);
2067 }
2068 /*
2069 * Now last_ibp points to the buffer previous to us on
2070 * the unlinked list. Pull us from the list.
2071 */
2072 error = xfs_itobp(mp, tp, ip, &dip, &ibp, 0);
2073 if (error) {
2074 cmn_err(CE_WARN,
2075 "xfs_iunlink_remove: xfs_itobp() returned an error %d on %s. Returning error.",
2076 error, mp->m_fsname);
2077 return error;
2078 }
2079 next_agino = INT_GET(dip->di_next_unlinked, ARCH_CONVERT);
2080 ASSERT(next_agino != 0);
2081 ASSERT(next_agino != agino);
2082 if (next_agino != NULLAGINO) {
2083 INT_SET(dip->di_next_unlinked, ARCH_CONVERT, NULLAGINO);
2084 offset = ip->i_boffset +
2085 offsetof(xfs_dinode_t, di_next_unlinked);
2086 xfs_trans_inode_buf(tp, ibp);
2087 xfs_trans_log_buf(tp, ibp, offset,
2088 (offset + sizeof(xfs_agino_t) - 1));
2089 xfs_inobp_check(mp, ibp);
2090 } else {
2091 xfs_trans_brelse(tp, ibp);
2092 }
2093 /*
2094 * Point the previous inode on the list to the next inode.
2095 */
2096 INT_SET(last_dip->di_next_unlinked, ARCH_CONVERT, next_agino);
2097 ASSERT(next_agino != 0);
2098 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2099 xfs_trans_inode_buf(tp, last_ibp);
2100 xfs_trans_log_buf(tp, last_ibp, offset,
2101 (offset + sizeof(xfs_agino_t) - 1));
2102 xfs_inobp_check(mp, last_ibp);
2103 }
2104 return 0;
2105}
2106
2107static __inline__ int xfs_inode_clean(xfs_inode_t *ip)
2108{
2109 return (((ip->i_itemp == NULL) ||
2110 !(ip->i_itemp->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
2111 (ip->i_update_core == 0));
2112}
2113
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002114STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115xfs_ifree_cluster(
2116 xfs_inode_t *free_ip,
2117 xfs_trans_t *tp,
2118 xfs_ino_t inum)
2119{
2120 xfs_mount_t *mp = free_ip->i_mount;
2121 int blks_per_cluster;
2122 int nbufs;
2123 int ninodes;
2124 int i, j, found, pre_flushed;
2125 xfs_daddr_t blkno;
2126 xfs_buf_t *bp;
2127 xfs_ihash_t *ih;
2128 xfs_inode_t *ip, **ip_found;
2129 xfs_inode_log_item_t *iip;
2130 xfs_log_item_t *lip;
2131 SPLDECL(s);
2132
2133 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2134 blks_per_cluster = 1;
2135 ninodes = mp->m_sb.sb_inopblock;
2136 nbufs = XFS_IALLOC_BLOCKS(mp);
2137 } else {
2138 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2139 mp->m_sb.sb_blocksize;
2140 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2141 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2142 }
2143
2144 ip_found = kmem_alloc(ninodes * sizeof(xfs_inode_t *), KM_NOFS);
2145
2146 for (j = 0; j < nbufs; j++, inum += ninodes) {
2147 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2148 XFS_INO_TO_AGBNO(mp, inum));
2149
2150
2151 /*
2152 * Look for each inode in memory and attempt to lock it,
2153 * we can be racing with flush and tail pushing here.
2154 * any inode we get the locks on, add to an array of
2155 * inode items to process later.
2156 *
2157 * The get the buffer lock, we could beat a flush
2158 * or tail pushing thread to the lock here, in which
2159 * case they will go looking for the inode buffer
2160 * and fail, we need some other form of interlock
2161 * here.
2162 */
2163 found = 0;
2164 for (i = 0; i < ninodes; i++) {
2165 ih = XFS_IHASH(mp, inum + i);
2166 read_lock(&ih->ih_lock);
2167 for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
2168 if (ip->i_ino == inum + i)
2169 break;
2170 }
2171
2172 /* Inode not in memory or we found it already,
2173 * nothing to do
2174 */
2175 if (!ip || (ip->i_flags & XFS_ISTALE)) {
2176 read_unlock(&ih->ih_lock);
2177 continue;
2178 }
2179
2180 if (xfs_inode_clean(ip)) {
2181 read_unlock(&ih->ih_lock);
2182 continue;
2183 }
2184
2185 /* If we can get the locks then add it to the
2186 * list, otherwise by the time we get the bp lock
2187 * below it will already be attached to the
2188 * inode buffer.
2189 */
2190
2191 /* This inode will already be locked - by us, lets
2192 * keep it that way.
2193 */
2194
2195 if (ip == free_ip) {
2196 if (xfs_iflock_nowait(ip)) {
2197 ip->i_flags |= XFS_ISTALE;
2198
2199 if (xfs_inode_clean(ip)) {
2200 xfs_ifunlock(ip);
2201 } else {
2202 ip_found[found++] = ip;
2203 }
2204 }
2205 read_unlock(&ih->ih_lock);
2206 continue;
2207 }
2208
2209 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2210 if (xfs_iflock_nowait(ip)) {
2211 ip->i_flags |= XFS_ISTALE;
2212
2213 if (xfs_inode_clean(ip)) {
2214 xfs_ifunlock(ip);
2215 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2216 } else {
2217 ip_found[found++] = ip;
2218 }
2219 } else {
2220 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2221 }
2222 }
2223
2224 read_unlock(&ih->ih_lock);
2225 }
2226
2227 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2228 mp->m_bsize * blks_per_cluster,
2229 XFS_BUF_LOCK);
2230
2231 pre_flushed = 0;
2232 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
2233 while (lip) {
2234 if (lip->li_type == XFS_LI_INODE) {
2235 iip = (xfs_inode_log_item_t *)lip;
2236 ASSERT(iip->ili_logged == 1);
2237 lip->li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*)) xfs_istale_done;
2238 AIL_LOCK(mp,s);
2239 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2240 AIL_UNLOCK(mp, s);
2241 iip->ili_inode->i_flags |= XFS_ISTALE;
2242 pre_flushed++;
2243 }
2244 lip = lip->li_bio_list;
2245 }
2246
2247 for (i = 0; i < found; i++) {
2248 ip = ip_found[i];
2249 iip = ip->i_itemp;
2250
2251 if (!iip) {
2252 ip->i_update_core = 0;
2253 xfs_ifunlock(ip);
2254 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2255 continue;
2256 }
2257
2258 iip->ili_last_fields = iip->ili_format.ilf_fields;
2259 iip->ili_format.ilf_fields = 0;
2260 iip->ili_logged = 1;
2261 AIL_LOCK(mp,s);
2262 iip->ili_flush_lsn = iip->ili_item.li_lsn;
2263 AIL_UNLOCK(mp, s);
2264
2265 xfs_buf_attach_iodone(bp,
2266 (void(*)(xfs_buf_t*,xfs_log_item_t*))
2267 xfs_istale_done, (xfs_log_item_t *)iip);
2268 if (ip != free_ip) {
2269 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2270 }
2271 }
2272
2273 if (found || pre_flushed)
2274 xfs_trans_stale_inode_buf(tp, bp);
2275 xfs_trans_binval(tp, bp);
2276 }
2277
2278 kmem_free(ip_found, ninodes * sizeof(xfs_inode_t *));
2279}
2280
2281/*
2282 * This is called to return an inode to the inode free list.
2283 * The inode should already be truncated to 0 length and have
2284 * no pages associated with it. This routine also assumes that
2285 * the inode is already a part of the transaction.
2286 *
2287 * The on-disk copy of the inode will have been added to the list
2288 * of unlinked inodes in the AGI. We need to remove the inode from
2289 * that list atomically with respect to freeing it here.
2290 */
2291int
2292xfs_ifree(
2293 xfs_trans_t *tp,
2294 xfs_inode_t *ip,
2295 xfs_bmap_free_t *flist)
2296{
2297 int error;
2298 int delete;
2299 xfs_ino_t first_ino;
2300
2301 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2302 ASSERT(ip->i_transp == tp);
2303 ASSERT(ip->i_d.di_nlink == 0);
2304 ASSERT(ip->i_d.di_nextents == 0);
2305 ASSERT(ip->i_d.di_anextents == 0);
2306 ASSERT((ip->i_d.di_size == 0) ||
2307 ((ip->i_d.di_mode & S_IFMT) != S_IFREG));
2308 ASSERT(ip->i_d.di_nblocks == 0);
2309
2310 /*
2311 * Pull the on-disk inode from the AGI unlinked list.
2312 */
2313 error = xfs_iunlink_remove(tp, ip);
2314 if (error != 0) {
2315 return error;
2316 }
2317
2318 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2319 if (error != 0) {
2320 return error;
2321 }
2322 ip->i_d.di_mode = 0; /* mark incore inode as free */
2323 ip->i_d.di_flags = 0;
2324 ip->i_d.di_dmevmask = 0;
2325 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2326 ip->i_df.if_ext_max =
2327 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
2328 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2329 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2330 /*
2331 * Bump the generation count so no one will be confused
2332 * by reincarnations of this inode.
2333 */
2334 ip->i_d.di_gen++;
2335 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2336
2337 if (delete) {
2338 xfs_ifree_cluster(ip, tp, first_ino);
2339 }
2340
2341 return 0;
2342}
2343
2344/*
2345 * Reallocate the space for if_broot based on the number of records
2346 * being added or deleted as indicated in rec_diff. Move the records
2347 * and pointers in if_broot to fit the new size. When shrinking this
2348 * will eliminate holes between the records and pointers created by
2349 * the caller. When growing this will create holes to be filled in
2350 * by the caller.
2351 *
2352 * The caller must not request to add more records than would fit in
2353 * the on-disk inode root. If the if_broot is currently NULL, then
2354 * if we adding records one will be allocated. The caller must also
2355 * not request that the number of records go below zero, although
2356 * it can go to zero.
2357 *
2358 * ip -- the inode whose if_broot area is changing
2359 * ext_diff -- the change in the number of records, positive or negative,
2360 * requested for the if_broot array.
2361 */
2362void
2363xfs_iroot_realloc(
2364 xfs_inode_t *ip,
2365 int rec_diff,
2366 int whichfork)
2367{
2368 int cur_max;
2369 xfs_ifork_t *ifp;
2370 xfs_bmbt_block_t *new_broot;
2371 int new_max;
2372 size_t new_size;
2373 char *np;
2374 char *op;
2375
2376 /*
2377 * Handle the degenerate case quietly.
2378 */
2379 if (rec_diff == 0) {
2380 return;
2381 }
2382
2383 ifp = XFS_IFORK_PTR(ip, whichfork);
2384 if (rec_diff > 0) {
2385 /*
2386 * If there wasn't any memory allocated before, just
2387 * allocate it now and get out.
2388 */
2389 if (ifp->if_broot_bytes == 0) {
2390 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
2391 ifp->if_broot = (xfs_bmbt_block_t*)kmem_alloc(new_size,
2392 KM_SLEEP);
2393 ifp->if_broot_bytes = (int)new_size;
2394 return;
2395 }
2396
2397 /*
2398 * If there is already an existing if_broot, then we need
2399 * to realloc() it and shift the pointers to their new
2400 * location. The records don't change location because
2401 * they are kept butted up against the btree block header.
2402 */
2403 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2404 new_max = cur_max + rec_diff;
2405 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2406 ifp->if_broot = (xfs_bmbt_block_t *)
2407 kmem_realloc(ifp->if_broot,
2408 new_size,
2409 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
2410 KM_SLEEP);
2411 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2412 ifp->if_broot_bytes);
2413 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2414 (int)new_size);
2415 ifp->if_broot_bytes = (int)new_size;
2416 ASSERT(ifp->if_broot_bytes <=
2417 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2418 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
2419 return;
2420 }
2421
2422 /*
2423 * rec_diff is less than 0. In this case, we are shrinking the
2424 * if_broot buffer. It must already exist. If we go to zero
2425 * records, just get rid of the root and clear the status bit.
2426 */
2427 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
2428 cur_max = XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes);
2429 new_max = cur_max + rec_diff;
2430 ASSERT(new_max >= 0);
2431 if (new_max > 0)
2432 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
2433 else
2434 new_size = 0;
2435 if (new_size > 0) {
2436 new_broot = (xfs_bmbt_block_t *)kmem_alloc(new_size, KM_SLEEP);
2437 /*
2438 * First copy over the btree block header.
2439 */
2440 memcpy(new_broot, ifp->if_broot, sizeof(xfs_bmbt_block_t));
2441 } else {
2442 new_broot = NULL;
2443 ifp->if_flags &= ~XFS_IFBROOT;
2444 }
2445
2446 /*
2447 * Only copy the records and pointers if there are any.
2448 */
2449 if (new_max > 0) {
2450 /*
2451 * First copy the records.
2452 */
2453 op = (char *)XFS_BMAP_BROOT_REC_ADDR(ifp->if_broot, 1,
2454 ifp->if_broot_bytes);
2455 np = (char *)XFS_BMAP_BROOT_REC_ADDR(new_broot, 1,
2456 (int)new_size);
2457 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
2458
2459 /*
2460 * Then copy the pointers.
2461 */
2462 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(ifp->if_broot, 1,
2463 ifp->if_broot_bytes);
2464 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(new_broot, 1,
2465 (int)new_size);
2466 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2467 }
2468 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2469 ifp->if_broot = new_broot;
2470 ifp->if_broot_bytes = (int)new_size;
2471 ASSERT(ifp->if_broot_bytes <=
2472 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2473 return;
2474}
2475
2476
2477/*
2478 * This is called when the amount of space needed for if_extents
2479 * is increased or decreased. The change in size is indicated by
2480 * the number of extents that need to be added or deleted in the
2481 * ext_diff parameter.
2482 *
2483 * If the amount of space needed has decreased below the size of the
2484 * inline buffer, then switch to using the inline buffer. Otherwise,
2485 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2486 * to what is needed.
2487 *
2488 * ip -- the inode whose if_extents area is changing
2489 * ext_diff -- the change in the number of extents, positive or negative,
2490 * requested for the if_extents array.
2491 */
2492void
2493xfs_iext_realloc(
2494 xfs_inode_t *ip,
2495 int ext_diff,
2496 int whichfork)
2497{
2498 int byte_diff;
2499 xfs_ifork_t *ifp;
2500 int new_size;
2501 uint rnew_size;
2502
2503 if (ext_diff == 0) {
2504 return;
2505 }
2506
2507 ifp = XFS_IFORK_PTR(ip, whichfork);
2508 byte_diff = ext_diff * (uint)sizeof(xfs_bmbt_rec_t);
2509 new_size = (int)ifp->if_bytes + byte_diff;
2510 ASSERT(new_size >= 0);
2511
2512 if (new_size == 0) {
2513 if (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext) {
2514 ASSERT(ifp->if_real_bytes != 0);
2515 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
2516 }
2517 ifp->if_u1.if_extents = NULL;
2518 rnew_size = 0;
2519 } else if (new_size <= sizeof(ifp->if_u2.if_inline_ext)) {
2520 /*
2521 * If the valid extents can fit in if_inline_ext,
2522 * copy them from the malloc'd vector and free it.
2523 */
2524 if (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext) {
2525 /*
2526 * For now, empty files are format EXTENTS,
2527 * so the if_extents pointer is null.
2528 */
2529 if (ifp->if_u1.if_extents) {
2530 memcpy(ifp->if_u2.if_inline_ext,
2531 ifp->if_u1.if_extents, new_size);
2532 kmem_free(ifp->if_u1.if_extents,
2533 ifp->if_real_bytes);
2534 }
2535 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2536 }
2537 rnew_size = 0;
2538 } else {
2539 rnew_size = new_size;
2540 if ((rnew_size & (rnew_size - 1)) != 0)
2541 rnew_size = xfs_iroundup(rnew_size);
2542 /*
2543 * Stuck with malloc/realloc.
2544 */
2545 if (ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext) {
2546 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
2547 kmem_alloc(rnew_size, KM_SLEEP);
2548 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
2549 sizeof(ifp->if_u2.if_inline_ext));
2550 } else if (rnew_size != ifp->if_real_bytes) {
2551 ifp->if_u1.if_extents = (xfs_bmbt_rec_t *)
2552 kmem_realloc(ifp->if_u1.if_extents,
2553 rnew_size,
2554 ifp->if_real_bytes,
2555 KM_NOFS);
2556 }
2557 }
2558 ifp->if_real_bytes = rnew_size;
2559 ifp->if_bytes = new_size;
2560}
2561
2562
2563/*
2564 * This is called when the amount of space needed for if_data
2565 * is increased or decreased. The change in size is indicated by
2566 * the number of bytes that need to be added or deleted in the
2567 * byte_diff parameter.
2568 *
2569 * If the amount of space needed has decreased below the size of the
2570 * inline buffer, then switch to using the inline buffer. Otherwise,
2571 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2572 * to what is needed.
2573 *
2574 * ip -- the inode whose if_data area is changing
2575 * byte_diff -- the change in the number of bytes, positive or negative,
2576 * requested for the if_data array.
2577 */
2578void
2579xfs_idata_realloc(
2580 xfs_inode_t *ip,
2581 int byte_diff,
2582 int whichfork)
2583{
2584 xfs_ifork_t *ifp;
2585 int new_size;
2586 int real_size;
2587
2588 if (byte_diff == 0) {
2589 return;
2590 }
2591
2592 ifp = XFS_IFORK_PTR(ip, whichfork);
2593 new_size = (int)ifp->if_bytes + byte_diff;
2594 ASSERT(new_size >= 0);
2595
2596 if (new_size == 0) {
2597 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2598 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2599 }
2600 ifp->if_u1.if_data = NULL;
2601 real_size = 0;
2602 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2603 /*
2604 * If the valid extents/data can fit in if_inline_ext/data,
2605 * copy them from the malloc'd vector and free it.
2606 */
2607 if (ifp->if_u1.if_data == NULL) {
2608 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2609 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2610 ASSERT(ifp->if_real_bytes != 0);
2611 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2612 new_size);
2613 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2614 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2615 }
2616 real_size = 0;
2617 } else {
2618 /*
2619 * Stuck with malloc/realloc.
2620 * For inline data, the underlying buffer must be
2621 * a multiple of 4 bytes in size so that it can be
2622 * logged and stay on word boundaries. We enforce
2623 * that here.
2624 */
2625 real_size = roundup(new_size, 4);
2626 if (ifp->if_u1.if_data == NULL) {
2627 ASSERT(ifp->if_real_bytes == 0);
2628 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2629 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2630 /*
2631 * Only do the realloc if the underlying size
2632 * is really changing.
2633 */
2634 if (ifp->if_real_bytes != real_size) {
2635 ifp->if_u1.if_data =
2636 kmem_realloc(ifp->if_u1.if_data,
2637 real_size,
2638 ifp->if_real_bytes,
2639 KM_SLEEP);
2640 }
2641 } else {
2642 ASSERT(ifp->if_real_bytes == 0);
2643 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
2644 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2645 ifp->if_bytes);
2646 }
2647 }
2648 ifp->if_real_bytes = real_size;
2649 ifp->if_bytes = new_size;
2650 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2651}
2652
2653
2654
2655
2656/*
2657 * Map inode to disk block and offset.
2658 *
2659 * mp -- the mount point structure for the current file system
2660 * tp -- the current transaction
2661 * ino -- the inode number of the inode to be located
2662 * imap -- this structure is filled in with the information necessary
2663 * to retrieve the given inode from disk
2664 * flags -- flags to pass to xfs_dilocate indicating whether or not
2665 * lookups in the inode btree were OK or not
2666 */
2667int
2668xfs_imap(
2669 xfs_mount_t *mp,
2670 xfs_trans_t *tp,
2671 xfs_ino_t ino,
2672 xfs_imap_t *imap,
2673 uint flags)
2674{
2675 xfs_fsblock_t fsbno;
2676 int len;
2677 int off;
2678 int error;
2679
2680 fsbno = imap->im_blkno ?
2681 XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
2682 error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
2683 if (error != 0) {
2684 return error;
2685 }
2686 imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
2687 imap->im_len = XFS_FSB_TO_BB(mp, len);
2688 imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
2689 imap->im_ioffset = (ushort)off;
2690 imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
2691 return 0;
2692}
2693
2694void
2695xfs_idestroy_fork(
2696 xfs_inode_t *ip,
2697 int whichfork)
2698{
2699 xfs_ifork_t *ifp;
2700
2701 ifp = XFS_IFORK_PTR(ip, whichfork);
2702 if (ifp->if_broot != NULL) {
2703 kmem_free(ifp->if_broot, ifp->if_broot_bytes);
2704 ifp->if_broot = NULL;
2705 }
2706
2707 /*
2708 * If the format is local, then we can't have an extents
2709 * array so just look for an inline data array. If we're
2710 * not local then we may or may not have an extents list,
2711 * so check and free it up if we do.
2712 */
2713 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2714 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2715 (ifp->if_u1.if_data != NULL)) {
2716 ASSERT(ifp->if_real_bytes != 0);
2717 kmem_free(ifp->if_u1.if_data, ifp->if_real_bytes);
2718 ifp->if_u1.if_data = NULL;
2719 ifp->if_real_bytes = 0;
2720 }
2721 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2722 (ifp->if_u1.if_extents != NULL) &&
2723 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)) {
2724 ASSERT(ifp->if_real_bytes != 0);
2725 kmem_free(ifp->if_u1.if_extents, ifp->if_real_bytes);
2726 ifp->if_u1.if_extents = NULL;
2727 ifp->if_real_bytes = 0;
2728 }
2729 ASSERT(ifp->if_u1.if_extents == NULL ||
2730 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2731 ASSERT(ifp->if_real_bytes == 0);
2732 if (whichfork == XFS_ATTR_FORK) {
2733 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2734 ip->i_afp = NULL;
2735 }
2736}
2737
2738/*
2739 * This is called free all the memory associated with an inode.
2740 * It must free the inode itself and any buffers allocated for
2741 * if_extents/if_data and if_broot. It must also free the lock
2742 * associated with the inode.
2743 */
2744void
2745xfs_idestroy(
2746 xfs_inode_t *ip)
2747{
2748
2749 switch (ip->i_d.di_mode & S_IFMT) {
2750 case S_IFREG:
2751 case S_IFDIR:
2752 case S_IFLNK:
2753 xfs_idestroy_fork(ip, XFS_DATA_FORK);
2754 break;
2755 }
2756 if (ip->i_afp)
2757 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
2758 mrfree(&ip->i_lock);
2759 mrfree(&ip->i_iolock);
2760 freesema(&ip->i_flock);
2761#ifdef XFS_BMAP_TRACE
2762 ktrace_free(ip->i_xtrace);
2763#endif
2764#ifdef XFS_BMBT_TRACE
2765 ktrace_free(ip->i_btrace);
2766#endif
2767#ifdef XFS_RW_TRACE
2768 ktrace_free(ip->i_rwtrace);
2769#endif
2770#ifdef XFS_ILOCK_TRACE
2771 ktrace_free(ip->i_lock_trace);
2772#endif
2773#ifdef XFS_DIR2_TRACE
2774 ktrace_free(ip->i_dir_trace);
2775#endif
2776 if (ip->i_itemp) {
2777 /* XXXdpd should be able to assert this but shutdown
2778 * is leaving the AIL behind. */
2779 ASSERT(((ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL) == 0) ||
2780 XFS_FORCED_SHUTDOWN(ip->i_mount));
2781 xfs_inode_item_destroy(ip);
2782 }
2783 kmem_zone_free(xfs_inode_zone, ip);
2784}
2785
2786
2787/*
2788 * Increment the pin count of the given buffer.
2789 * This value is protected by ipinlock spinlock in the mount structure.
2790 */
2791void
2792xfs_ipin(
2793 xfs_inode_t *ip)
2794{
2795 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE));
2796
2797 atomic_inc(&ip->i_pincount);
2798}
2799
2800/*
2801 * Decrement the pin count of the given inode, and wake up
2802 * anyone in xfs_iwait_unpin() if the count goes to 0. The
2803 * inode must have been previoulsy pinned with a call to xfs_ipin().
2804 */
2805void
2806xfs_iunpin(
2807 xfs_inode_t *ip)
2808{
2809 ASSERT(atomic_read(&ip->i_pincount) > 0);
2810
2811 if (atomic_dec_and_test(&ip->i_pincount)) {
2812 vnode_t *vp = XFS_ITOV_NULL(ip);
2813
2814 /* make sync come back and flush this inode */
2815 if (vp) {
2816 struct inode *inode = LINVFS_GET_IP(vp);
2817
2818 if (!(inode->i_state & I_NEW))
2819 mark_inode_dirty_sync(inode);
2820 }
2821
2822 wake_up(&ip->i_ipin_wait);
2823 }
2824}
2825
2826/*
2827 * This is called to wait for the given inode to be unpinned.
2828 * It will sleep until this happens. The caller must have the
2829 * inode locked in at least shared mode so that the buffer cannot
2830 * be subsequently pinned once someone is waiting for it to be
2831 * unpinned.
2832 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002833STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834xfs_iunpin_wait(
2835 xfs_inode_t *ip)
2836{
2837 xfs_inode_log_item_t *iip;
2838 xfs_lsn_t lsn;
2839
2840 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS));
2841
2842 if (atomic_read(&ip->i_pincount) == 0) {
2843 return;
2844 }
2845
2846 iip = ip->i_itemp;
2847 if (iip && iip->ili_last_lsn) {
2848 lsn = iip->ili_last_lsn;
2849 } else {
2850 lsn = (xfs_lsn_t)0;
2851 }
2852
2853 /*
2854 * Give the log a push so we don't wait here too long.
2855 */
2856 xfs_log_force(ip->i_mount, lsn, XFS_LOG_FORCE);
2857
2858 wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
2859}
2860
2861
2862/*
2863 * xfs_iextents_copy()
2864 *
2865 * This is called to copy the REAL extents (as opposed to the delayed
2866 * allocation extents) from the inode into the given buffer. It
2867 * returns the number of bytes copied into the buffer.
2868 *
2869 * If there are no delayed allocation extents, then we can just
2870 * memcpy() the extents into the buffer. Otherwise, we need to
2871 * examine each extent in turn and skip those which are delayed.
2872 */
2873int
2874xfs_iextents_copy(
2875 xfs_inode_t *ip,
2876 xfs_bmbt_rec_t *buffer,
2877 int whichfork)
2878{
2879 int copied;
2880 xfs_bmbt_rec_t *dest_ep;
2881 xfs_bmbt_rec_t *ep;
2882#ifdef XFS_BMAP_TRACE
2883 static char fname[] = "xfs_iextents_copy";
2884#endif
2885 int i;
2886 xfs_ifork_t *ifp;
2887 int nrecs;
2888 xfs_fsblock_t start_block;
2889
2890 ifp = XFS_IFORK_PTR(ip, whichfork);
2891 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
2892 ASSERT(ifp->if_bytes > 0);
2893
2894 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2895 xfs_bmap_trace_exlist(fname, ip, nrecs, whichfork);
2896 ASSERT(nrecs > 0);
2897
2898 /*
2899 * There are some delayed allocation extents in the
2900 * inode, so copy the extents one at a time and skip
2901 * the delayed ones. There must be at least one
2902 * non-delayed extent.
2903 */
2904 ep = ifp->if_u1.if_extents;
2905 dest_ep = buffer;
2906 copied = 0;
2907 for (i = 0; i < nrecs; i++) {
2908 start_block = xfs_bmbt_get_startblock(ep);
2909 if (ISNULLSTARTBLOCK(start_block)) {
2910 /*
2911 * It's a delayed allocation extent, so skip it.
2912 */
2913 ep++;
2914 continue;
2915 }
2916
2917 /* Translate to on disk format */
2918 put_unaligned(INT_GET(ep->l0, ARCH_CONVERT),
2919 (__uint64_t*)&dest_ep->l0);
2920 put_unaligned(INT_GET(ep->l1, ARCH_CONVERT),
2921 (__uint64_t*)&dest_ep->l1);
2922 dest_ep++;
2923 ep++;
2924 copied++;
2925 }
2926 ASSERT(copied != 0);
2927 xfs_validate_extents(buffer, copied, 1, XFS_EXTFMT_INODE(ip));
2928
2929 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2930}
2931
2932/*
2933 * Each of the following cases stores data into the same region
2934 * of the on-disk inode, so only one of them can be valid at
2935 * any given time. While it is possible to have conflicting formats
2936 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2937 * in EXTENTS format, this can only happen when the fork has
2938 * changed formats after being modified but before being flushed.
2939 * In these cases, the format always takes precedence, because the
2940 * format indicates the current state of the fork.
2941 */
2942/*ARGSUSED*/
2943STATIC int
2944xfs_iflush_fork(
2945 xfs_inode_t *ip,
2946 xfs_dinode_t *dip,
2947 xfs_inode_log_item_t *iip,
2948 int whichfork,
2949 xfs_buf_t *bp)
2950{
2951 char *cp;
2952 xfs_ifork_t *ifp;
2953 xfs_mount_t *mp;
2954#ifdef XFS_TRANS_DEBUG
2955 int first;
2956#endif
2957 static const short brootflag[2] =
2958 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2959 static const short dataflag[2] =
2960 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2961 static const short extflag[2] =
2962 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2963
2964 if (iip == NULL)
2965 return 0;
2966 ifp = XFS_IFORK_PTR(ip, whichfork);
2967 /*
2968 * This can happen if we gave up in iformat in an error path,
2969 * for the attribute fork.
2970 */
2971 if (ifp == NULL) {
2972 ASSERT(whichfork == XFS_ATTR_FORK);
2973 return 0;
2974 }
2975 cp = XFS_DFORK_PTR(dip, whichfork);
2976 mp = ip->i_mount;
2977 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2978 case XFS_DINODE_FMT_LOCAL:
2979 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2980 (ifp->if_bytes > 0)) {
2981 ASSERT(ifp->if_u1.if_data != NULL);
2982 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2983 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2984 }
2985 if (whichfork == XFS_DATA_FORK) {
2986 if (unlikely(XFS_DIR_SHORTFORM_VALIDATE_ONDISK(mp, dip))) {
2987 XFS_ERROR_REPORT("xfs_iflush_fork",
2988 XFS_ERRLEVEL_LOW, mp);
2989 return XFS_ERROR(EFSCORRUPTED);
2990 }
2991 }
2992 break;
2993
2994 case XFS_DINODE_FMT_EXTENTS:
2995 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2996 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2997 ASSERT((ifp->if_u1.if_extents != NULL) || (ifp->if_bytes == 0));
2998 ASSERT((ifp->if_u1.if_extents == NULL) || (ifp->if_bytes > 0));
2999 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
3000 (ifp->if_bytes > 0)) {
3001 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
3002 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
3003 whichfork);
3004 }
3005 break;
3006
3007 case XFS_DINODE_FMT_BTREE:
3008 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
3009 (ifp->if_broot_bytes > 0)) {
3010 ASSERT(ifp->if_broot != NULL);
3011 ASSERT(ifp->if_broot_bytes <=
3012 (XFS_IFORK_SIZE(ip, whichfork) +
3013 XFS_BROOT_SIZE_ADJ));
3014 xfs_bmbt_to_bmdr(ifp->if_broot, ifp->if_broot_bytes,
3015 (xfs_bmdr_block_t *)cp,
3016 XFS_DFORK_SIZE(dip, mp, whichfork));
3017 }
3018 break;
3019
3020 case XFS_DINODE_FMT_DEV:
3021 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
3022 ASSERT(whichfork == XFS_DATA_FORK);
3023 INT_SET(dip->di_u.di_dev, ARCH_CONVERT, ip->i_df.if_u2.if_rdev);
3024 }
3025 break;
3026
3027 case XFS_DINODE_FMT_UUID:
3028 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
3029 ASSERT(whichfork == XFS_DATA_FORK);
3030 memcpy(&dip->di_u.di_muuid, &ip->i_df.if_u2.if_uuid,
3031 sizeof(uuid_t));
3032 }
3033 break;
3034
3035 default:
3036 ASSERT(0);
3037 break;
3038 }
3039
3040 return 0;
3041}
3042
3043/*
3044 * xfs_iflush() will write a modified inode's changes out to the
3045 * inode's on disk home. The caller must have the inode lock held
3046 * in at least shared mode and the inode flush semaphore must be
3047 * held as well. The inode lock will still be held upon return from
3048 * the call and the caller is free to unlock it.
3049 * The inode flush lock will be unlocked when the inode reaches the disk.
3050 * The flags indicate how the inode's buffer should be written out.
3051 */
3052int
3053xfs_iflush(
3054 xfs_inode_t *ip,
3055 uint flags)
3056{
3057 xfs_inode_log_item_t *iip;
3058 xfs_buf_t *bp;
3059 xfs_dinode_t *dip;
3060 xfs_mount_t *mp;
3061 int error;
3062 /* REFERENCED */
3063 xfs_chash_t *ch;
3064 xfs_inode_t *iq;
3065 int clcount; /* count of inodes clustered */
3066 int bufwasdelwri;
3067 enum { INT_DELWRI = (1 << 0), INT_ASYNC = (1 << 1) };
3068 SPLDECL(s);
3069
3070 XFS_STATS_INC(xs_iflush_count);
3071
3072 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3073 ASSERT(valusema(&ip->i_flock) <= 0);
3074 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3075 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3076
3077 iip = ip->i_itemp;
3078 mp = ip->i_mount;
3079
3080 /*
3081 * If the inode isn't dirty, then just release the inode
3082 * flush lock and do nothing.
3083 */
3084 if ((ip->i_update_core == 0) &&
3085 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3086 ASSERT((iip != NULL) ?
3087 !(iip->ili_item.li_flags & XFS_LI_IN_AIL) : 1);
3088 xfs_ifunlock(ip);
3089 return 0;
3090 }
3091
3092 /*
3093 * We can't flush the inode until it is unpinned, so
3094 * wait for it. We know noone new can pin it, because
3095 * we are holding the inode lock shared and you need
3096 * to hold it exclusively to pin the inode.
3097 */
3098 xfs_iunpin_wait(ip);
3099
3100 /*
3101 * This may have been unpinned because the filesystem is shutting
3102 * down forcibly. If that's the case we must not write this inode
3103 * to disk, because the log record didn't make it to disk!
3104 */
3105 if (XFS_FORCED_SHUTDOWN(mp)) {
3106 ip->i_update_core = 0;
3107 if (iip)
3108 iip->ili_format.ilf_fields = 0;
3109 xfs_ifunlock(ip);
3110 return XFS_ERROR(EIO);
3111 }
3112
3113 /*
3114 * Get the buffer containing the on-disk inode.
3115 */
3116 error = xfs_itobp(mp, NULL, ip, &dip, &bp, 0);
3117 if (error != 0) {
3118 xfs_ifunlock(ip);
3119 return error;
3120 }
3121
3122 /*
3123 * Decide how buffer will be flushed out. This is done before
3124 * the call to xfs_iflush_int because this field is zeroed by it.
3125 */
3126 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3127 /*
3128 * Flush out the inode buffer according to the directions
3129 * of the caller. In the cases where the caller has given
3130 * us a choice choose the non-delwri case. This is because
3131 * the inode is in the AIL and we need to get it out soon.
3132 */
3133 switch (flags) {
3134 case XFS_IFLUSH_SYNC:
3135 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3136 flags = 0;
3137 break;
3138 case XFS_IFLUSH_ASYNC:
3139 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3140 flags = INT_ASYNC;
3141 break;
3142 case XFS_IFLUSH_DELWRI:
3143 flags = INT_DELWRI;
3144 break;
3145 default:
3146 ASSERT(0);
3147 flags = 0;
3148 break;
3149 }
3150 } else {
3151 switch (flags) {
3152 case XFS_IFLUSH_DELWRI_ELSE_SYNC:
3153 case XFS_IFLUSH_DELWRI_ELSE_ASYNC:
3154 case XFS_IFLUSH_DELWRI:
3155 flags = INT_DELWRI;
3156 break;
3157 case XFS_IFLUSH_ASYNC:
3158 flags = INT_ASYNC;
3159 break;
3160 case XFS_IFLUSH_SYNC:
3161 flags = 0;
3162 break;
3163 default:
3164 ASSERT(0);
3165 flags = 0;
3166 break;
3167 }
3168 }
3169
3170 /*
3171 * First flush out the inode that xfs_iflush was called with.
3172 */
3173 error = xfs_iflush_int(ip, bp);
3174 if (error) {
3175 goto corrupt_out;
3176 }
3177
3178 /*
3179 * inode clustering:
3180 * see if other inodes can be gathered into this write
3181 */
3182
3183 ip->i_chash->chl_buf = bp;
3184
3185 ch = XFS_CHASH(mp, ip->i_blkno);
3186 s = mutex_spinlock(&ch->ch_lock);
3187
3188 clcount = 0;
3189 for (iq = ip->i_cnext; iq != ip; iq = iq->i_cnext) {
3190 /*
3191 * Do an un-protected check to see if the inode is dirty and
3192 * is a candidate for flushing. These checks will be repeated
3193 * later after the appropriate locks are acquired.
3194 */
3195 iip = iq->i_itemp;
3196 if ((iq->i_update_core == 0) &&
3197 ((iip == NULL) ||
3198 !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)) &&
3199 xfs_ipincount(iq) == 0) {
3200 continue;
3201 }
3202
3203 /*
3204 * Try to get locks. If any are unavailable,
3205 * then this inode cannot be flushed and is skipped.
3206 */
3207
3208 /* get inode locks (just i_lock) */
3209 if (xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) {
3210 /* get inode flush lock */
3211 if (xfs_iflock_nowait(iq)) {
3212 /* check if pinned */
3213 if (xfs_ipincount(iq) == 0) {
3214 /* arriving here means that
3215 * this inode can be flushed.
3216 * first re-check that it's
3217 * dirty
3218 */
3219 iip = iq->i_itemp;
3220 if ((iq->i_update_core != 0)||
3221 ((iip != NULL) &&
3222 (iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3223 clcount++;
3224 error = xfs_iflush_int(iq, bp);
3225 if (error) {
3226 xfs_iunlock(iq,
3227 XFS_ILOCK_SHARED);
3228 goto cluster_corrupt_out;
3229 }
3230 } else {
3231 xfs_ifunlock(iq);
3232 }
3233 } else {
3234 xfs_ifunlock(iq);
3235 }
3236 }
3237 xfs_iunlock(iq, XFS_ILOCK_SHARED);
3238 }
3239 }
3240 mutex_spinunlock(&ch->ch_lock, s);
3241
3242 if (clcount) {
3243 XFS_STATS_INC(xs_icluster_flushcnt);
3244 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
3245 }
3246
3247 /*
3248 * If the buffer is pinned then push on the log so we won't
3249 * get stuck waiting in the write for too long.
3250 */
3251 if (XFS_BUF_ISPINNED(bp)){
3252 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
3253 }
3254
3255 if (flags & INT_DELWRI) {
3256 xfs_bdwrite(mp, bp);
3257 } else if (flags & INT_ASYNC) {
3258 xfs_bawrite(mp, bp);
3259 } else {
3260 error = xfs_bwrite(mp, bp);
3261 }
3262 return error;
3263
3264corrupt_out:
3265 xfs_buf_relse(bp);
3266 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3267 xfs_iflush_abort(ip);
3268 /*
3269 * Unlocks the flush lock
3270 */
3271 return XFS_ERROR(EFSCORRUPTED);
3272
3273cluster_corrupt_out:
3274 /* Corruption detected in the clustering loop. Invalidate the
3275 * inode buffer and shut down the filesystem.
3276 */
3277 mutex_spinunlock(&ch->ch_lock, s);
3278
3279 /*
3280 * Clean up the buffer. If it was B_DELWRI, just release it --
3281 * brelse can handle it with no problems. If not, shut down the
3282 * filesystem before releasing the buffer.
3283 */
3284 if ((bufwasdelwri= XFS_BUF_ISDELAYWRITE(bp))) {
3285 xfs_buf_relse(bp);
3286 }
3287
3288 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
3289
3290 if(!bufwasdelwri) {
3291 /*
3292 * Just like incore_relse: if we have b_iodone functions,
3293 * mark the buffer as an error and call them. Otherwise
3294 * mark it as stale and brelse.
3295 */
3296 if (XFS_BUF_IODONE_FUNC(bp)) {
3297 XFS_BUF_CLR_BDSTRAT_FUNC(bp);
3298 XFS_BUF_UNDONE(bp);
3299 XFS_BUF_STALE(bp);
3300 XFS_BUF_SHUT(bp);
3301 XFS_BUF_ERROR(bp,EIO);
3302 xfs_biodone(bp);
3303 } else {
3304 XFS_BUF_STALE(bp);
3305 xfs_buf_relse(bp);
3306 }
3307 }
3308
3309 xfs_iflush_abort(iq);
3310 /*
3311 * Unlocks the flush lock
3312 */
3313 return XFS_ERROR(EFSCORRUPTED);
3314}
3315
3316
3317STATIC int
3318xfs_iflush_int(
3319 xfs_inode_t *ip,
3320 xfs_buf_t *bp)
3321{
3322 xfs_inode_log_item_t *iip;
3323 xfs_dinode_t *dip;
3324 xfs_mount_t *mp;
3325#ifdef XFS_TRANS_DEBUG
3326 int first;
3327#endif
3328 SPLDECL(s);
3329
3330 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE|MR_ACCESS));
3331 ASSERT(valusema(&ip->i_flock) <= 0);
3332 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3333 ip->i_d.di_nextents > ip->i_df.if_ext_max);
3334
3335 iip = ip->i_itemp;
3336 mp = ip->i_mount;
3337
3338
3339 /*
3340 * If the inode isn't dirty, then just release the inode
3341 * flush lock and do nothing.
3342 */
3343 if ((ip->i_update_core == 0) &&
3344 ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL))) {
3345 xfs_ifunlock(ip);
3346 return 0;
3347 }
3348
3349 /* set *dip = inode's place in the buffer */
3350 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
3351
3352 /*
3353 * Clear i_update_core before copying out the data.
3354 * This is for coordination with our timestamp updates
3355 * that don't hold the inode lock. They will always
3356 * update the timestamps BEFORE setting i_update_core,
3357 * so if we clear i_update_core after they set it we
3358 * are guaranteed to see their updates to the timestamps.
3359 * I believe that this depends on strongly ordered memory
3360 * semantics, but we have that. We use the SYNCHRONIZE
3361 * macro to make sure that the compiler does not reorder
3362 * the i_update_core access below the data copy below.
3363 */
3364 ip->i_update_core = 0;
3365 SYNCHRONIZE();
3366
Christoph Hellwig42fe2b12006-01-11 15:35:17 +11003367 /*
3368 * Make sure to get the latest atime from the Linux inode.
3369 */
3370 xfs_synchronize_atime(ip);
3371
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3373 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3374 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3375 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3376 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3377 goto corrupt_out;
3378 }
3379 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3380 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3381 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3382 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3383 ip->i_ino, ip, ip->i_d.di_magic);
3384 goto corrupt_out;
3385 }
3386 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3387 if (XFS_TEST_ERROR(
3388 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3389 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3390 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3391 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3392 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3393 ip->i_ino, ip);
3394 goto corrupt_out;
3395 }
3396 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3397 if (XFS_TEST_ERROR(
3398 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3399 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3400 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3401 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3402 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3403 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3404 ip->i_ino, ip);
3405 goto corrupt_out;
3406 }
3407 }
3408 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3409 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3410 XFS_RANDOM_IFLUSH_5)) {
3411 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3412 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3413 ip->i_ino,
3414 ip->i_d.di_nextents + ip->i_d.di_anextents,
3415 ip->i_d.di_nblocks,
3416 ip);
3417 goto corrupt_out;
3418 }
3419 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3420 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3421 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3422 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3423 ip->i_ino, ip->i_d.di_forkoff, ip);
3424 goto corrupt_out;
3425 }
3426 /*
3427 * bump the flush iteration count, used to detect flushes which
3428 * postdate a log record during recovery.
3429 */
3430
3431 ip->i_d.di_flushiter++;
3432
3433 /*
3434 * Copy the dirty parts of the inode into the on-disk
3435 * inode. We always copy out the core of the inode,
3436 * because if the inode is dirty at all the core must
3437 * be.
3438 */
3439 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3440
3441 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3442 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3443 ip->i_d.di_flushiter = 0;
3444
3445 /*
3446 * If this is really an old format inode and the superblock version
3447 * has not been updated to support only new format inodes, then
3448 * convert back to the old inode format. If the superblock version
3449 * has been updated, then make the conversion permanent.
3450 */
3451 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3452 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3453 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3454 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3455 /*
3456 * Convert it back.
3457 */
3458 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3459 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3460 } else {
3461 /*
3462 * The superblock version has already been bumped,
3463 * so just make the conversion to the new inode
3464 * format permanent.
3465 */
3466 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3467 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3468 ip->i_d.di_onlink = 0;
3469 dip->di_core.di_onlink = 0;
3470 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3471 memset(&(dip->di_core.di_pad[0]), 0,
3472 sizeof(dip->di_core.di_pad));
3473 ASSERT(ip->i_d.di_projid == 0);
3474 }
3475 }
3476
3477 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3478 goto corrupt_out;
3479 }
3480
3481 if (XFS_IFORK_Q(ip)) {
3482 /*
3483 * The only error from xfs_iflush_fork is on the data fork.
3484 */
3485 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3486 }
3487 xfs_inobp_check(mp, bp);
3488
3489 /*
3490 * We've recorded everything logged in the inode, so we'd
3491 * like to clear the ilf_fields bits so we don't log and
3492 * flush things unnecessarily. However, we can't stop
3493 * logging all this information until the data we've copied
3494 * into the disk buffer is written to disk. If we did we might
3495 * overwrite the copy of the inode in the log with all the
3496 * data after re-logging only part of it, and in the face of
3497 * a crash we wouldn't have all the data we need to recover.
3498 *
3499 * What we do is move the bits to the ili_last_fields field.
3500 * When logging the inode, these bits are moved back to the
3501 * ilf_fields field. In the xfs_iflush_done() routine we
3502 * clear ili_last_fields, since we know that the information
3503 * those bits represent is permanently on disk. As long as
3504 * the flush completes before the inode is logged again, then
3505 * both ilf_fields and ili_last_fields will be cleared.
3506 *
3507 * We can play with the ilf_fields bits here, because the inode
3508 * lock must be held exclusively in order to set bits there
3509 * and the flush lock protects the ili_last_fields bits.
3510 * Set ili_logged so the flush done
3511 * routine can tell whether or not to look in the AIL.
3512 * Also, store the current LSN of the inode so that we can tell
3513 * whether the item has moved in the AIL from xfs_iflush_done().
3514 * In order to read the lsn we need the AIL lock, because
3515 * it is a 64 bit value that cannot be read atomically.
3516 */
3517 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3518 iip->ili_last_fields = iip->ili_format.ilf_fields;
3519 iip->ili_format.ilf_fields = 0;
3520 iip->ili_logged = 1;
3521
3522 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3523 AIL_LOCK(mp,s);
3524 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3525 AIL_UNLOCK(mp, s);
3526
3527 /*
3528 * Attach the function xfs_iflush_done to the inode's
3529 * buffer. This will remove the inode from the AIL
3530 * and unlock the inode's flush lock when the inode is
3531 * completely written to disk.
3532 */
3533 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3534 xfs_iflush_done, (xfs_log_item_t *)iip);
3535
3536 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3537 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3538 } else {
3539 /*
3540 * We're flushing an inode which is not in the AIL and has
3541 * not been logged but has i_update_core set. For this
3542 * case we can use a B_DELWRI flush and immediately drop
3543 * the inode flush lock because we can avoid the whole
3544 * AIL state thing. It's OK to drop the flush lock now,
3545 * because we've already locked the buffer and to do anything
3546 * you really need both.
3547 */
3548 if (iip != NULL) {
3549 ASSERT(iip->ili_logged == 0);
3550 ASSERT(iip->ili_last_fields == 0);
3551 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3552 }
3553 xfs_ifunlock(ip);
3554 }
3555
3556 return 0;
3557
3558corrupt_out:
3559 return XFS_ERROR(EFSCORRUPTED);
3560}
3561
3562
3563/*
Christoph Hellwigefa80272005-06-21 15:37:17 +10003564 * Flush all inactive inodes in mp.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565 */
Christoph Hellwigefa80272005-06-21 15:37:17 +10003566void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567xfs_iflush_all(
Christoph Hellwigefa80272005-06-21 15:37:17 +10003568 xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 xfs_inode_t *ip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571 vnode_t *vp;
3572
Christoph Hellwigefa80272005-06-21 15:37:17 +10003573 again:
3574 XFS_MOUNT_ILOCK(mp);
3575 ip = mp->m_inodes;
3576 if (ip == NULL)
3577 goto out;
3578
3579 do {
3580 /* Make sure we skip markers inserted by sync */
3581 if (ip->i_mount == NULL) {
3582 ip = ip->i_mnext;
3583 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585
Christoph Hellwigefa80272005-06-21 15:37:17 +10003586 vp = XFS_ITOV_NULL(ip);
3587 if (!vp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588 XFS_MOUNT_IUNLOCK(mp);
Christoph Hellwigefa80272005-06-21 15:37:17 +10003589 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3590 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003591 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592
Christoph Hellwigefa80272005-06-21 15:37:17 +10003593 ASSERT(vn_count(vp) == 0);
3594
3595 ip = ip->i_mnext;
3596 } while (ip != mp->m_inodes);
3597 out:
3598 XFS_MOUNT_IUNLOCK(mp);
3599}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600
3601/*
3602 * xfs_iaccess: check accessibility of inode for mode.
3603 */
3604int
3605xfs_iaccess(
3606 xfs_inode_t *ip,
3607 mode_t mode,
3608 cred_t *cr)
3609{
3610 int error;
3611 mode_t orgmode = mode;
3612 struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip));
3613
3614 if (mode & S_IWUSR) {
3615 umode_t imode = inode->i_mode;
3616
3617 if (IS_RDONLY(inode) &&
3618 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3619 return XFS_ERROR(EROFS);
3620
3621 if (IS_IMMUTABLE(inode))
3622 return XFS_ERROR(EACCES);
3623 }
3624
3625 /*
3626 * If there's an Access Control List it's used instead of
3627 * the mode bits.
3628 */
3629 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3630 return error ? XFS_ERROR(error) : 0;
3631
3632 if (current_fsuid(cr) != ip->i_d.di_uid) {
3633 mode >>= 3;
3634 if (!in_group_p((gid_t)ip->i_d.di_gid))
3635 mode >>= 3;
3636 }
3637
3638 /*
3639 * If the DACs are ok we don't need any capability check.
3640 */
3641 if ((ip->i_d.di_mode & mode) == mode)
3642 return 0;
3643 /*
3644 * Read/write DACs are always overridable.
3645 * Executable DACs are overridable if at least one exec bit is set.
3646 */
3647 if (!(orgmode & S_IXUSR) ||
3648 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3649 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3650 return 0;
3651
3652 if ((orgmode == S_IRUSR) ||
3653 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3654 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3655 return 0;
3656#ifdef NOISE
3657 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3658#endif /* NOISE */
3659 return XFS_ERROR(EACCES);
3660 }
3661 return XFS_ERROR(EACCES);
3662}
3663
3664/*
3665 * xfs_iroundup: round up argument to next power of two
3666 */
3667uint
3668xfs_iroundup(
3669 uint v)
3670{
3671 int i;
3672 uint m;
3673
3674 if ((v & (v - 1)) == 0)
3675 return v;
3676 ASSERT((v & 0x80000000) == 0);
3677 if ((v & (v + 1)) == 0)
3678 return v + 1;
3679 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3680 if (v & m)
3681 continue;
3682 v |= m;
3683 if ((v & (v + 1)) == 0)
3684 return v + 1;
3685 }
3686 ASSERT(0);
3687 return( 0 );
3688}
3689
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690#ifdef XFS_ILOCK_TRACE
3691ktrace_t *xfs_ilock_trace_buf;
3692
3693void
3694xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3695{
3696 ktrace_enter(ip->i_lock_trace,
3697 (void *)ip,
3698 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3699 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3700 (void *)ra, /* caller of ilock */
3701 (void *)(unsigned long)current_cpu(),
3702 (void *)(unsigned long)current_pid(),
3703 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3704}
3705#endif