blob: 99421638e860c87d76659939dd0284cf5748a158 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (c) 2000-2003 Silicon Graphics, Inc. All Rights Reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11 *
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
18 *
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22 *
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
25 *
26 * http://www.sgi.com
27 *
28 * For further information regarding this notice, see:
29 *
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31 */
32
33#include "xfs.h"
34#include "xfs_macros.h"
35#include "xfs_types.h"
36#include "xfs_inum.h"
37#include "xfs_log.h"
38#include "xfs_trans.h"
39#include "xfs_trans_priv.h"
40#include "xfs_sb.h"
41#include "xfs_ag.h"
42#include "xfs_dir.h"
43#include "xfs_dir2.h"
44#include "xfs_dmapi.h"
45#include "xfs_mount.h"
46#include "xfs_alloc_btree.h"
47#include "xfs_bmap_btree.h"
48#include "xfs_ialloc_btree.h"
49#include "xfs_btree.h"
50#include "xfs_imap.h"
51#include "xfs_alloc.h"
52#include "xfs_ialloc.h"
53#include "xfs_attr_sf.h"
54#include "xfs_dir_sf.h"
55#include "xfs_dir2_sf.h"
56#include "xfs_dinode.h"
57#include "xfs_inode_item.h"
58#include "xfs_inode.h"
59#include "xfs_bmap.h"
60#include "xfs_buf_item.h"
61#include "xfs_rw.h"
62#include "xfs_error.h"
63#include "xfs_bit.h"
64#include "xfs_utils.h"
65#include "xfs_dir2_trace.h"
66#include "xfs_quota.h"
67#include "xfs_mac.h"
68#include "xfs_acl.h"
69
70
71kmem_zone_t *xfs_ifork_zone;
72kmem_zone_t *xfs_inode_zone;
73kmem_zone_t *xfs_chashlist_zone;
74
75/*
76 * Used in xfs_itruncate(). This is the maximum number of extents
77 * freed from a file in a single transaction.
78 */
79#define XFS_ITRUNC_MAX_EXTENTS 2
80
81STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
82STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
83STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
84STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
85
86
87#ifdef DEBUG
88/*
89 * Make sure that the extents in the given memory buffer
90 * are valid.
91 */
92STATIC void
93xfs_validate_extents(
94 xfs_bmbt_rec_t *ep,
95 int nrecs,
96 int disk,
97 xfs_exntfmt_t fmt)
98{
99 xfs_bmbt_irec_t irec;
100 xfs_bmbt_rec_t rec;
101 int i;
102
103 for (i = 0; i < nrecs; i++) {
104 rec.l0 = get_unaligned((__uint64_t*)&ep->l0);
105 rec.l1 = get_unaligned((__uint64_t*)&ep->l1);
106 if (disk)
107 xfs_bmbt_disk_get_all(&rec, &irec);
108 else
109 xfs_bmbt_get_all(&rec, &irec);
110 if (fmt == XFS_EXTFMT_NOSTATE)
111 ASSERT(irec.br_state == XFS_EXT_NORM);
112 ep++;
113 }
114}
115#else /* DEBUG */
116#define xfs_validate_extents(ep, nrecs, disk, fmt)
117#endif /* DEBUG */
118
119/*
120 * Check that none of the inode's in the buffer have a next
121 * unlinked field of 0.
122 */
123#if defined(DEBUG)
124void
125xfs_inobp_check(
126 xfs_mount_t *mp,
127 xfs_buf_t *bp)
128{
129 int i;
130 int j;
131 xfs_dinode_t *dip;
132
133 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
134
135 for (i = 0; i < j; i++) {
136 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
137 i * mp->m_sb.sb_inodesize);
138 if (!dip->di_next_unlinked) {
139 xfs_fs_cmn_err(CE_ALERT, mp,
140 "Detected a bogus zero next_unlinked field in incore inode buffer 0x%p. About to pop an ASSERT.",
141 bp);
142 ASSERT(dip->di_next_unlinked);
143 }
144 }
145}
146#endif
147
148/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 * This routine is called to map an inode number within a file
150 * system to the buffer containing the on-disk version of the
151 * inode. It returns a pointer to the buffer containing the
152 * on-disk inode in the bpp parameter, and in the dip parameter
153 * it returns a pointer to the on-disk inode within that buffer.
154 *
155 * If a non-zero error is returned, then the contents of bpp and
156 * dipp are undefined.
157 *
158 * Use xfs_imap() to determine the size and location of the
159 * buffer to read from disk.
160 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000161STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162xfs_inotobp(
163 xfs_mount_t *mp,
164 xfs_trans_t *tp,
165 xfs_ino_t ino,
166 xfs_dinode_t **dipp,
167 xfs_buf_t **bpp,
168 int *offset)
169{
170 int di_ok;
171 xfs_imap_t imap;
172 xfs_buf_t *bp;
173 int error;
174 xfs_dinode_t *dip;
175
176 /*
177 * Call the space managment code to find the location of the
178 * inode on disk.
179 */
180 imap.im_blkno = 0;
181 error = xfs_imap(mp, tp, ino, &imap, XFS_IMAP_LOOKUP);
182 if (error != 0) {
183 cmn_err(CE_WARN,
184 "xfs_inotobp: xfs_imap() returned an "
185 "error %d on %s. Returning error.", error, mp->m_fsname);
186 return error;
187 }
188
189 /*
190 * If the inode number maps to a block outside the bounds of the
191 * file system then return NULL rather than calling read_buf
192 * and panicing when we get an error from the driver.
193 */
194 if ((imap.im_blkno + imap.im_len) >
195 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
196 cmn_err(CE_WARN,
197 "xfs_inotobp: inode number (%d + %d) maps to a block outside the bounds "
198 "of the file system %s. Returning EINVAL.",
199 imap.im_blkno, imap.im_len,mp->m_fsname);
200 return XFS_ERROR(EINVAL);
201 }
202
203 /*
204 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
205 * default to just a read_buf() call.
206 */
207 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
208 (int)imap.im_len, XFS_BUF_LOCK, &bp);
209
210 if (error) {
211 cmn_err(CE_WARN,
212 "xfs_inotobp: xfs_trans_read_buf() returned an "
213 "error %d on %s. Returning error.", error, mp->m_fsname);
214 return error;
215 }
216 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 0);
217 di_ok =
218 INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
219 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
220 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
221 XFS_RANDOM_ITOBP_INOTOBP))) {
222 XFS_CORRUPTION_ERROR("xfs_inotobp", XFS_ERRLEVEL_LOW, mp, dip);
223 xfs_trans_brelse(tp, bp);
224 cmn_err(CE_WARN,
225 "xfs_inotobp: XFS_TEST_ERROR() returned an "
226 "error on %s. Returning EFSCORRUPTED.", mp->m_fsname);
227 return XFS_ERROR(EFSCORRUPTED);
228 }
229
230 xfs_inobp_check(mp, bp);
231
232 /*
233 * Set *dipp to point to the on-disk inode in the buffer.
234 */
235 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
236 *bpp = bp;
237 *offset = imap.im_boffset;
238 return 0;
239}
240
241
242/*
243 * This routine is called to map an inode to the buffer containing
244 * the on-disk version of the inode. It returns a pointer to the
245 * buffer containing the on-disk inode in the bpp parameter, and in
246 * the dip parameter it returns a pointer to the on-disk inode within
247 * that buffer.
248 *
249 * If a non-zero error is returned, then the contents of bpp and
250 * dipp are undefined.
251 *
252 * If the inode is new and has not yet been initialized, use xfs_imap()
253 * to determine the size and location of the buffer to read from disk.
254 * If the inode has already been mapped to its buffer and read in once,
255 * then use the mapping information stored in the inode rather than
256 * calling xfs_imap(). This allows us to avoid the overhead of looking
257 * at the inode btree for small block file systems (see xfs_dilocate()).
258 * We can tell whether the inode has been mapped in before by comparing
259 * its disk block address to 0. Only uninitialized inodes will have
260 * 0 for the disk block address.
261 */
262int
263xfs_itobp(
264 xfs_mount_t *mp,
265 xfs_trans_t *tp,
266 xfs_inode_t *ip,
267 xfs_dinode_t **dipp,
268 xfs_buf_t **bpp,
269 xfs_daddr_t bno)
270{
271 xfs_buf_t *bp;
272 int error;
273 xfs_imap_t imap;
274#ifdef __KERNEL__
275 int i;
276 int ni;
277#endif
278
279 if (ip->i_blkno == (xfs_daddr_t)0) {
280 /*
281 * Call the space management code to find the location of the
282 * inode on disk.
283 */
284 imap.im_blkno = bno;
285 error = xfs_imap(mp, tp, ip->i_ino, &imap, XFS_IMAP_LOOKUP);
286 if (error != 0) {
287 return error;
288 }
289
290 /*
291 * If the inode number maps to a block outside the bounds
292 * of the file system then return NULL rather than calling
293 * read_buf and panicing when we get an error from the
294 * driver.
295 */
296 if ((imap.im_blkno + imap.im_len) >
297 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
298#ifdef DEBUG
299 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
300 "(imap.im_blkno (0x%llx) "
301 "+ imap.im_len (0x%llx)) > "
302 " XFS_FSB_TO_BB(mp, "
303 "mp->m_sb.sb_dblocks) (0x%llx)",
304 (unsigned long long) imap.im_blkno,
305 (unsigned long long) imap.im_len,
306 XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
307#endif /* DEBUG */
308 return XFS_ERROR(EINVAL);
309 }
310
311 /*
312 * Fill in the fields in the inode that will be used to
313 * map the inode to its buffer from now on.
314 */
315 ip->i_blkno = imap.im_blkno;
316 ip->i_len = imap.im_len;
317 ip->i_boffset = imap.im_boffset;
318 } else {
319 /*
320 * We've already mapped the inode once, so just use the
321 * mapping that we saved the first time.
322 */
323 imap.im_blkno = ip->i_blkno;
324 imap.im_len = ip->i_len;
325 imap.im_boffset = ip->i_boffset;
326 }
327 ASSERT(bno == 0 || bno == imap.im_blkno);
328
329 /*
330 * Read in the buffer. If tp is NULL, xfs_trans_read_buf() will
331 * default to just a read_buf() call.
332 */
333 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap.im_blkno,
334 (int)imap.im_len, XFS_BUF_LOCK, &bp);
335
336 if (error) {
337#ifdef DEBUG
338 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_itobp: "
339 "xfs_trans_read_buf() returned error %d, "
340 "imap.im_blkno 0x%llx, imap.im_len 0x%llx",
341 error, (unsigned long long) imap.im_blkno,
342 (unsigned long long) imap.im_len);
343#endif /* DEBUG */
344 return error;
345 }
346#ifdef __KERNEL__
347 /*
348 * Validate the magic number and version of every inode in the buffer
349 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
350 */
351#ifdef DEBUG
352 ni = BBTOB(imap.im_len) >> mp->m_sb.sb_inodelog;
353#else
354 ni = 1;
355#endif
356 for (i = 0; i < ni; i++) {
357 int di_ok;
358 xfs_dinode_t *dip;
359
360 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
361 (i << mp->m_sb.sb_inodelog));
362 di_ok = INT_GET(dip->di_core.di_magic, ARCH_CONVERT) == XFS_DINODE_MAGIC &&
363 XFS_DINODE_GOOD_VERSION(INT_GET(dip->di_core.di_version, ARCH_CONVERT));
364 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, XFS_ERRTAG_ITOBP_INOTOBP,
365 XFS_RANDOM_ITOBP_INOTOBP))) {
366#ifdef DEBUG
367 prdev("bad inode magic/vsn daddr %lld #%d (magic=%x)",
368 mp->m_ddev_targp,
369 (unsigned long long)imap.im_blkno, i,
370 INT_GET(dip->di_core.di_magic, ARCH_CONVERT));
371#endif
372 XFS_CORRUPTION_ERROR("xfs_itobp", XFS_ERRLEVEL_HIGH,
373 mp, dip);
374 xfs_trans_brelse(tp, bp);
375 return XFS_ERROR(EFSCORRUPTED);
376 }
377 }
378#endif /* __KERNEL__ */
379
380 xfs_inobp_check(mp, bp);
381
382 /*
383 * Mark the buffer as an inode buffer now that it looks good
384 */
385 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
386
387 /*
388 * Set *dipp to point to the on-disk inode in the buffer.
389 */
390 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
391 *bpp = bp;
392 return 0;
393}
394
395/*
396 * Move inode type and inode format specific information from the
397 * on-disk inode to the in-core inode. For fifos, devs, and sockets
398 * this means set if_rdev to the proper value. For files, directories,
399 * and symlinks this means to bring in the in-line data or extent
400 * pointers. For a file in B-tree format, only the root is immediately
401 * brought in-core. The rest will be in-lined in if_extents when it
402 * is first referenced (see xfs_iread_extents()).
403 */
404STATIC int
405xfs_iformat(
406 xfs_inode_t *ip,
407 xfs_dinode_t *dip)
408{
409 xfs_attr_shortform_t *atp;
410 int size;
411 int error;
412 xfs_fsize_t di_size;
413 ip->i_df.if_ext_max =
414 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
415 error = 0;
416
417 if (unlikely(
418 INT_GET(dip->di_core.di_nextents, ARCH_CONVERT) +
419 INT_GET(dip->di_core.di_anextents, ARCH_CONVERT) >
420 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT))) {
421 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
422 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu."
423 " Unmount and run xfs_repair.",
424 (unsigned long long)ip->i_ino,
425 (int)(INT_GET(dip->di_core.di_nextents, ARCH_CONVERT)
426 + INT_GET(dip->di_core.di_anextents, ARCH_CONVERT)),
427 (unsigned long long)
428 INT_GET(dip->di_core.di_nblocks, ARCH_CONVERT));
429 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
430 ip->i_mount, dip);
431 return XFS_ERROR(EFSCORRUPTED);
432 }
433
434 if (unlikely(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT) > ip->i_mount->m_sb.sb_inodesize)) {
435 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
436 "corrupt dinode %Lu, forkoff = 0x%x."
437 " Unmount and run xfs_repair.",
438 (unsigned long long)ip->i_ino,
439 (int)(INT_GET(dip->di_core.di_forkoff, ARCH_CONVERT)));
440 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
441 ip->i_mount, dip);
442 return XFS_ERROR(EFSCORRUPTED);
443 }
444
445 switch (ip->i_d.di_mode & S_IFMT) {
446 case S_IFIFO:
447 case S_IFCHR:
448 case S_IFBLK:
449 case S_IFSOCK:
450 if (unlikely(INT_GET(dip->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_DEV)) {
451 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
452 ip->i_mount, dip);
453 return XFS_ERROR(EFSCORRUPTED);
454 }
455 ip->i_d.di_size = 0;
456 ip->i_df.if_u2.if_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
457 break;
458
459 case S_IFREG:
460 case S_IFLNK:
461 case S_IFDIR:
462 switch (INT_GET(dip->di_core.di_format, ARCH_CONVERT)) {
463 case XFS_DINODE_FMT_LOCAL:
464 /*
465 * no local regular files yet
466 */
467 if (unlikely((INT_GET(dip->di_core.di_mode, ARCH_CONVERT) & S_IFMT) == S_IFREG)) {
468 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
469 "corrupt inode (local format for regular file) %Lu. Unmount and run xfs_repair.",
470 (unsigned long long) ip->i_ino);
471 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
472 XFS_ERRLEVEL_LOW,
473 ip->i_mount, dip);
474 return XFS_ERROR(EFSCORRUPTED);
475 }
476
477 di_size = INT_GET(dip->di_core.di_size, ARCH_CONVERT);
478 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
479 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
480 "corrupt inode %Lu (bad size %Ld for local inode). Unmount and run xfs_repair.",
481 (unsigned long long) ip->i_ino,
482 (long long) di_size);
483 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
484 XFS_ERRLEVEL_LOW,
485 ip->i_mount, dip);
486 return XFS_ERROR(EFSCORRUPTED);
487 }
488
489 size = (int)di_size;
490 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
491 break;
492 case XFS_DINODE_FMT_EXTENTS:
493 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
494 break;
495 case XFS_DINODE_FMT_BTREE:
496 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
497 break;
498 default:
499 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
500 ip->i_mount);
501 return XFS_ERROR(EFSCORRUPTED);
502 }
503 break;
504
505 default:
506 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
507 return XFS_ERROR(EFSCORRUPTED);
508 }
509 if (error) {
510 return error;
511 }
512 if (!XFS_DFORK_Q(dip))
513 return 0;
514 ASSERT(ip->i_afp == NULL);
515 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
516 ip->i_afp->if_ext_max =
517 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
518 switch (INT_GET(dip->di_core.di_aformat, ARCH_CONVERT)) {
519 case XFS_DINODE_FMT_LOCAL:
520 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
521 size = (int)INT_GET(atp->hdr.totsize, ARCH_CONVERT);
522 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
523 break;
524 case XFS_DINODE_FMT_EXTENTS:
525 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
526 break;
527 case XFS_DINODE_FMT_BTREE:
528 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
529 break;
530 default:
531 error = XFS_ERROR(EFSCORRUPTED);
532 break;
533 }
534 if (error) {
535 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
536 ip->i_afp = NULL;
537 xfs_idestroy_fork(ip, XFS_DATA_FORK);
538 }
539 return error;
540}
541
542/*
543 * The file is in-lined in the on-disk inode.
544 * If it fits into if_inline_data, then copy
545 * it there, otherwise allocate a buffer for it
546 * and copy the data there. Either way, set
547 * if_data to point at the data.
548 * If we allocate a buffer for the data, make
549 * sure that its size is a multiple of 4 and
550 * record the real size in i_real_bytes.
551 */
552STATIC int
553xfs_iformat_local(
554 xfs_inode_t *ip,
555 xfs_dinode_t *dip,
556 int whichfork,
557 int size)
558{
559 xfs_ifork_t *ifp;
560 int real_size;
561
562 /*
563 * If the size is unreasonable, then something
564 * is wrong and we just bail out rather than crash in
565 * kmem_alloc() or memcpy() below.
566 */
567 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
568 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
569 "corrupt inode %Lu (bad size %d for local fork, size = %d). Unmount and run xfs_repair.",
570 (unsigned long long) ip->i_ino, size,
571 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
572 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
573 ip->i_mount, dip);
574 return XFS_ERROR(EFSCORRUPTED);
575 }
576 ifp = XFS_IFORK_PTR(ip, whichfork);
577 real_size = 0;
578 if (size == 0)
579 ifp->if_u1.if_data = NULL;
580 else if (size <= sizeof(ifp->if_u2.if_inline_data))
581 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
582 else {
583 real_size = roundup(size, 4);
584 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP);
585 }
586 ifp->if_bytes = size;
587 ifp->if_real_bytes = real_size;
588 if (size)
589 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
590 ifp->if_flags &= ~XFS_IFEXTENTS;
591 ifp->if_flags |= XFS_IFINLINE;
592 return 0;
593}
594
595/*
596 * The file consists of a set of extents all
597 * of which fit into the on-disk inode.
598 * If there are few enough extents to fit into
599 * the if_inline_ext, then copy them there.
600 * Otherwise allocate a buffer for them and copy
601 * them into it. Either way, set if_extents
602 * to point at the extents.
603 */
604STATIC int
605xfs_iformat_extents(
606 xfs_inode_t *ip,
607 xfs_dinode_t *dip,
608 int whichfork)
609{
610 xfs_bmbt_rec_t *ep, *dp;
611 xfs_ifork_t *ifp;
612 int nex;
613 int real_size;
614 int size;
615 int i;
616
617 ifp = XFS_IFORK_PTR(ip, whichfork);
618 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
619 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
620
621 /*
622 * If the number of extents is unreasonable, then something
623 * is wrong and we just bail out rather than crash in
624 * kmem_alloc() or memcpy() below.
625 */
626 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
627 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
628 "corrupt inode %Lu ((a)extents = %d). Unmount and run xfs_repair.",
629 (unsigned long long) ip->i_ino, nex);
630 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
631 ip->i_mount, dip);
632 return XFS_ERROR(EFSCORRUPTED);
633 }
634
635 real_size = 0;
636 if (nex == 0)
637 ifp->if_u1.if_extents = NULL;
638 else if (nex <= XFS_INLINE_EXTS)
639 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
640 else {
641 ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP);
642 ASSERT(ifp->if_u1.if_extents != NULL);
643 real_size = size;
644 }
645 ifp->if_bytes = size;
646 ifp->if_real_bytes = real_size;
647 if (size) {
648 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
649 xfs_validate_extents(dp, nex, 1, XFS_EXTFMT_INODE(ip));
650 ep = ifp->if_u1.if_extents;
651 for (i = 0; i < nex; i++, ep++, dp++) {
652 ep->l0 = INT_GET(get_unaligned((__uint64_t*)&dp->l0),
653 ARCH_CONVERT);
654 ep->l1 = INT_GET(get_unaligned((__uint64_t*)&dp->l1),
655 ARCH_CONVERT);
656 }
657 xfs_bmap_trace_exlist("xfs_iformat_extents", ip, nex,
658 whichfork);
659 if (whichfork != XFS_DATA_FORK ||
660 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
661 if (unlikely(xfs_check_nostate_extents(
662 ifp->if_u1.if_extents, nex))) {
663 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
664 XFS_ERRLEVEL_LOW,
665 ip->i_mount);
666 return XFS_ERROR(EFSCORRUPTED);
667 }
668 }
669 ifp->if_flags |= XFS_IFEXTENTS;
670 return 0;
671}
672
673/*
674 * The file has too many extents to fit into
675 * the inode, so they are in B-tree format.
676 * Allocate a buffer for the root of the B-tree
677 * and copy the root into it. The i_extents
678 * field will remain NULL until all of the
679 * extents are read in (when they are needed).
680 */
681STATIC int
682xfs_iformat_btree(
683 xfs_inode_t *ip,
684 xfs_dinode_t *dip,
685 int whichfork)
686{
687 xfs_bmdr_block_t *dfp;
688 xfs_ifork_t *ifp;
689 /* REFERENCED */
690 int nrecs;
691 int size;
692
693 ifp = XFS_IFORK_PTR(ip, whichfork);
694 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
695 size = XFS_BMAP_BROOT_SPACE(dfp);
696 nrecs = XFS_BMAP_BROOT_NUMRECS(dfp);
697
698 /*
699 * blow out if -- fork has less extents than can fit in
700 * fork (fork shouldn't be a btree format), root btree
701 * block has more records than can fit into the fork,
702 * or the number of extents is greater than the number of
703 * blocks.
704 */
705 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
706 || XFS_BMDR_SPACE_CALC(nrecs) >
707 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
708 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
709 xfs_fs_cmn_err(CE_WARN, ip->i_mount,
710 "corrupt inode %Lu (btree). Unmount and run xfs_repair.",
711 (unsigned long long) ip->i_ino);
712 XFS_ERROR_REPORT("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
713 ip->i_mount);
714 return XFS_ERROR(EFSCORRUPTED);
715 }
716
717 ifp->if_broot_bytes = size;
718 ifp->if_broot = kmem_alloc(size, KM_SLEEP);
719 ASSERT(ifp->if_broot != NULL);
720 /*
721 * Copy and convert from the on-disk structure
722 * to the in-memory structure.
723 */
724 xfs_bmdr_to_bmbt(dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
725 ifp->if_broot, size);
726 ifp->if_flags &= ~XFS_IFEXTENTS;
727 ifp->if_flags |= XFS_IFBROOT;
728
729 return 0;
730}
731
732/*
733 * xfs_xlate_dinode_core - translate an xfs_inode_core_t between ondisk
734 * and native format
735 *
736 * buf = on-disk representation
737 * dip = native representation
738 * dir = direction - +ve -> disk to native
739 * -ve -> native to disk
740 */
741void
742xfs_xlate_dinode_core(
743 xfs_caddr_t buf,
744 xfs_dinode_core_t *dip,
745 int dir)
746{
747 xfs_dinode_core_t *buf_core = (xfs_dinode_core_t *)buf;
748 xfs_dinode_core_t *mem_core = (xfs_dinode_core_t *)dip;
749 xfs_arch_t arch = ARCH_CONVERT;
750
751 ASSERT(dir);
752
753 INT_XLATE(buf_core->di_magic, mem_core->di_magic, dir, arch);
754 INT_XLATE(buf_core->di_mode, mem_core->di_mode, dir, arch);
755 INT_XLATE(buf_core->di_version, mem_core->di_version, dir, arch);
756 INT_XLATE(buf_core->di_format, mem_core->di_format, dir, arch);
757 INT_XLATE(buf_core->di_onlink, mem_core->di_onlink, dir, arch);
758 INT_XLATE(buf_core->di_uid, mem_core->di_uid, dir, arch);
759 INT_XLATE(buf_core->di_gid, mem_core->di_gid, dir, arch);
760 INT_XLATE(buf_core->di_nlink, mem_core->di_nlink, dir, arch);
761 INT_XLATE(buf_core->di_projid, mem_core->di_projid, dir, arch);
762
763 if (dir > 0) {
764 memcpy(mem_core->di_pad, buf_core->di_pad,
765 sizeof(buf_core->di_pad));
766 } else {
767 memcpy(buf_core->di_pad, mem_core->di_pad,
768 sizeof(buf_core->di_pad));
769 }
770
771 INT_XLATE(buf_core->di_flushiter, mem_core->di_flushiter, dir, arch);
772
773 INT_XLATE(buf_core->di_atime.t_sec, mem_core->di_atime.t_sec,
774 dir, arch);
775 INT_XLATE(buf_core->di_atime.t_nsec, mem_core->di_atime.t_nsec,
776 dir, arch);
777 INT_XLATE(buf_core->di_mtime.t_sec, mem_core->di_mtime.t_sec,
778 dir, arch);
779 INT_XLATE(buf_core->di_mtime.t_nsec, mem_core->di_mtime.t_nsec,
780 dir, arch);
781 INT_XLATE(buf_core->di_ctime.t_sec, mem_core->di_ctime.t_sec,
782 dir, arch);
783 INT_XLATE(buf_core->di_ctime.t_nsec, mem_core->di_ctime.t_nsec,
784 dir, arch);
785 INT_XLATE(buf_core->di_size, mem_core->di_size, dir, arch);
786 INT_XLATE(buf_core->di_nblocks, mem_core->di_nblocks, dir, arch);
787 INT_XLATE(buf_core->di_extsize, mem_core->di_extsize, dir, arch);
788 INT_XLATE(buf_core->di_nextents, mem_core->di_nextents, dir, arch);
789 INT_XLATE(buf_core->di_anextents, mem_core->di_anextents, dir, arch);
790 INT_XLATE(buf_core->di_forkoff, mem_core->di_forkoff, dir, arch);
791 INT_XLATE(buf_core->di_aformat, mem_core->di_aformat, dir, arch);
792 INT_XLATE(buf_core->di_dmevmask, mem_core->di_dmevmask, dir, arch);
793 INT_XLATE(buf_core->di_dmstate, mem_core->di_dmstate, dir, arch);
794 INT_XLATE(buf_core->di_flags, mem_core->di_flags, dir, arch);
795 INT_XLATE(buf_core->di_gen, mem_core->di_gen, dir, arch);
796}
797
798STATIC uint
799_xfs_dic2xflags(
800 xfs_dinode_core_t *dic,
801 __uint16_t di_flags)
802{
803 uint flags = 0;
804
805 if (di_flags & XFS_DIFLAG_ANY) {
806 if (di_flags & XFS_DIFLAG_REALTIME)
807 flags |= XFS_XFLAG_REALTIME;
808 if (di_flags & XFS_DIFLAG_PREALLOC)
809 flags |= XFS_XFLAG_PREALLOC;
810 if (di_flags & XFS_DIFLAG_IMMUTABLE)
811 flags |= XFS_XFLAG_IMMUTABLE;
812 if (di_flags & XFS_DIFLAG_APPEND)
813 flags |= XFS_XFLAG_APPEND;
814 if (di_flags & XFS_DIFLAG_SYNC)
815 flags |= XFS_XFLAG_SYNC;
816 if (di_flags & XFS_DIFLAG_NOATIME)
817 flags |= XFS_XFLAG_NOATIME;
818 if (di_flags & XFS_DIFLAG_NODUMP)
819 flags |= XFS_XFLAG_NODUMP;
820 if (di_flags & XFS_DIFLAG_RTINHERIT)
821 flags |= XFS_XFLAG_RTINHERIT;
822 if (di_flags & XFS_DIFLAG_PROJINHERIT)
823 flags |= XFS_XFLAG_PROJINHERIT;
824 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
825 flags |= XFS_XFLAG_NOSYMLINKS;
826 }
827
828 return flags;
829}
830
831uint
832xfs_ip2xflags(
833 xfs_inode_t *ip)
834{
835 xfs_dinode_core_t *dic = &ip->i_d;
836
837 return _xfs_dic2xflags(dic, dic->di_flags) |
838 (XFS_CFORK_Q(dic) ? XFS_XFLAG_HASATTR : 0);
839}
840
841uint
842xfs_dic2xflags(
843 xfs_dinode_core_t *dic)
844{
845 return _xfs_dic2xflags(dic, INT_GET(dic->di_flags, ARCH_CONVERT)) |
846 (XFS_CFORK_Q_DISK(dic) ? XFS_XFLAG_HASATTR : 0);
847}
848
849/*
850 * Given a mount structure and an inode number, return a pointer
851 * to a newly allocated in-core inode coresponding to the given
852 * inode number.
853 *
854 * Initialize the inode's attributes and extent pointers if it
855 * already has them (it will not if the inode has no links).
856 */
857int
858xfs_iread(
859 xfs_mount_t *mp,
860 xfs_trans_t *tp,
861 xfs_ino_t ino,
862 xfs_inode_t **ipp,
863 xfs_daddr_t bno)
864{
865 xfs_buf_t *bp;
866 xfs_dinode_t *dip;
867 xfs_inode_t *ip;
868 int error;
869
870 ASSERT(xfs_inode_zone != NULL);
871
872 ip = kmem_zone_zalloc(xfs_inode_zone, KM_SLEEP);
873 ip->i_ino = ino;
874 ip->i_mount = mp;
875
876 /*
877 * Get pointer's to the on-disk inode and the buffer containing it.
878 * If the inode number refers to a block outside the file system
879 * then xfs_itobp() will return NULL. In this case we should
880 * return NULL as well. Set i_blkno to 0 so that xfs_itobp() will
881 * know that this is a new incore inode.
882 */
883 error = xfs_itobp(mp, tp, ip, &dip, &bp, bno);
884
885 if (error != 0) {
886 kmem_zone_free(xfs_inode_zone, ip);
887 return error;
888 }
889
890 /*
891 * Initialize inode's trace buffers.
892 * Do this before xfs_iformat in case it adds entries.
893 */
894#ifdef XFS_BMAP_TRACE
895 ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_SLEEP);
896#endif
897#ifdef XFS_BMBT_TRACE
898 ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_SLEEP);
899#endif
900#ifdef XFS_RW_TRACE
901 ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_SLEEP);
902#endif
903#ifdef XFS_ILOCK_TRACE
904 ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_SLEEP);
905#endif
906#ifdef XFS_DIR2_TRACE
907 ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_SLEEP);
908#endif
909
910 /*
911 * If we got something that isn't an inode it means someone
912 * (nfs or dmi) has a stale handle.
913 */
914 if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC) {
915 kmem_zone_free(xfs_inode_zone, ip);
916 xfs_trans_brelse(tp, bp);
917#ifdef DEBUG
918 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
919 "dip->di_core.di_magic (0x%x) != "
920 "XFS_DINODE_MAGIC (0x%x)",
921 INT_GET(dip->di_core.di_magic, ARCH_CONVERT),
922 XFS_DINODE_MAGIC);
923#endif /* DEBUG */
924 return XFS_ERROR(EINVAL);
925 }
926
927 /*
928 * If the on-disk inode is already linked to a directory
929 * entry, copy all of the inode into the in-core inode.
930 * xfs_iformat() handles copying in the inode format
931 * specific information.
932 * Otherwise, just get the truly permanent information.
933 */
934 if (dip->di_core.di_mode) {
935 xfs_xlate_dinode_core((xfs_caddr_t)&dip->di_core,
936 &(ip->i_d), 1);
937 error = xfs_iformat(ip, dip);
938 if (error) {
939 kmem_zone_free(xfs_inode_zone, ip);
940 xfs_trans_brelse(tp, bp);
941#ifdef DEBUG
942 xfs_fs_cmn_err(CE_ALERT, mp, "xfs_iread: "
943 "xfs_iformat() returned error %d",
944 error);
945#endif /* DEBUG */
946 return error;
947 }
948 } else {
949 ip->i_d.di_magic = INT_GET(dip->di_core.di_magic, ARCH_CONVERT);
950 ip->i_d.di_version = INT_GET(dip->di_core.di_version, ARCH_CONVERT);
951 ip->i_d.di_gen = INT_GET(dip->di_core.di_gen, ARCH_CONVERT);
952 ip->i_d.di_flushiter = INT_GET(dip->di_core.di_flushiter, ARCH_CONVERT);
953 /*
954 * Make sure to pull in the mode here as well in
955 * case the inode is released without being used.
956 * This ensures that xfs_inactive() will see that
957 * the inode is already free and not try to mess
958 * with the uninitialized part of it.
959 */
960 ip->i_d.di_mode = 0;
961 /*
962 * Initialize the per-fork minima and maxima for a new
963 * inode here. xfs_iformat will do it for old inodes.
964 */
965 ip->i_df.if_ext_max =
966 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
967 }
968
969 INIT_LIST_HEAD(&ip->i_reclaim);
970
971 /*
972 * The inode format changed when we moved the link count and
973 * made it 32 bits long. If this is an old format inode,
974 * convert it in memory to look like a new one. If it gets
975 * flushed to disk we will convert back before flushing or
976 * logging it. We zero out the new projid field and the old link
977 * count field. We'll handle clearing the pad field (the remains
978 * of the old uuid field) when we actually convert the inode to
979 * the new format. We don't change the version number so that we
980 * can distinguish this from a real new format inode.
981 */
982 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
983 ip->i_d.di_nlink = ip->i_d.di_onlink;
984 ip->i_d.di_onlink = 0;
985 ip->i_d.di_projid = 0;
986 }
987
988 ip->i_delayed_blks = 0;
989
990 /*
991 * Mark the buffer containing the inode as something to keep
992 * around for a while. This helps to keep recently accessed
993 * meta-data in-core longer.
994 */
995 XFS_BUF_SET_REF(bp, XFS_INO_REF);
996
997 /*
998 * Use xfs_trans_brelse() to release the buffer containing the
999 * on-disk inode, because it was acquired with xfs_trans_read_buf()
1000 * in xfs_itobp() above. If tp is NULL, this is just a normal
1001 * brelse(). If we're within a transaction, then xfs_trans_brelse()
1002 * will only release the buffer if it is not dirty within the
1003 * transaction. It will be OK to release the buffer in this case,
1004 * because inodes on disk are never destroyed and we will be
1005 * locking the new in-core inode before putting it in the hash
1006 * table where other processes can find it. Thus we don't have
1007 * to worry about the inode being changed just because we released
1008 * the buffer.
1009 */
1010 xfs_trans_brelse(tp, bp);
1011 *ipp = ip;
1012 return 0;
1013}
1014
1015/*
1016 * Read in extents from a btree-format inode.
1017 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
1018 */
1019int
1020xfs_iread_extents(
1021 xfs_trans_t *tp,
1022 xfs_inode_t *ip,
1023 int whichfork)
1024{
1025 int error;
1026 xfs_ifork_t *ifp;
1027 size_t size;
1028
1029 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
1030 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
1031 ip->i_mount);
1032 return XFS_ERROR(EFSCORRUPTED);
1033 }
1034 size = XFS_IFORK_NEXTENTS(ip, whichfork) * (uint)sizeof(xfs_bmbt_rec_t);
1035 ifp = XFS_IFORK_PTR(ip, whichfork);
1036 /*
1037 * We know that the size is valid (it's checked in iformat_btree)
1038 */
1039 ifp->if_u1.if_extents = kmem_alloc(size, KM_SLEEP);
1040 ASSERT(ifp->if_u1.if_extents != NULL);
1041 ifp->if_lastex = NULLEXTNUM;
1042 ifp->if_bytes = ifp->if_real_bytes = (int)size;
1043 ifp->if_flags |= XFS_IFEXTENTS;
1044 error = xfs_bmap_read_extents(tp, ip, whichfork);
1045 if (error) {
1046 kmem_free(ifp->if_u1.if_extents, size);
1047 ifp->if_u1.if_extents = NULL;
1048 ifp->if_bytes = ifp->if_real_bytes = 0;
1049 ifp->if_flags &= ~XFS_IFEXTENTS;
1050 return error;
1051 }
1052 xfs_validate_extents((xfs_bmbt_rec_t *)ifp->if_u1.if_extents,
1053 XFS_IFORK_NEXTENTS(ip, whichfork), 0, XFS_EXTFMT_INODE(ip));
1054 return 0;
1055}
1056
1057/*
1058 * Allocate an inode on disk and return a copy of its in-core version.
1059 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
1060 * appropriately within the inode. The uid and gid for the inode are
1061 * set according to the contents of the given cred structure.
1062 *
1063 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
1064 * has a free inode available, call xfs_iget()
1065 * to obtain the in-core version of the allocated inode. Finally,
1066 * fill in the inode and log its initial contents. In this case,
1067 * ialloc_context would be set to NULL and call_again set to false.
1068 *
1069 * If xfs_dialloc() does not have an available inode,
1070 * it will replenish its supply by doing an allocation. Since we can
1071 * only do one allocation within a transaction without deadlocks, we
1072 * must commit the current transaction before returning the inode itself.
1073 * In this case, therefore, we will set call_again to true and return.
1074 * The caller should then commit the current transaction, start a new
1075 * transaction, and call xfs_ialloc() again to actually get the inode.
1076 *
1077 * To ensure that some other process does not grab the inode that
1078 * was allocated during the first call to xfs_ialloc(), this routine
1079 * also returns the [locked] bp pointing to the head of the freelist
1080 * as ialloc_context. The caller should hold this buffer across
1081 * the commit and pass it back into this routine on the second call.
1082 */
1083int
1084xfs_ialloc(
1085 xfs_trans_t *tp,
1086 xfs_inode_t *pip,
1087 mode_t mode,
Nathan Scott31b084a2005-05-05 13:25:00 -07001088 xfs_nlink_t nlink,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 xfs_dev_t rdev,
1090 cred_t *cr,
1091 xfs_prid_t prid,
1092 int okalloc,
1093 xfs_buf_t **ialloc_context,
1094 boolean_t *call_again,
1095 xfs_inode_t **ipp)
1096{
1097 xfs_ino_t ino;
1098 xfs_inode_t *ip;
1099 vnode_t *vp;
1100 uint flags;
1101 int error;
1102
1103 /*
1104 * Call the space management code to pick
1105 * the on-disk inode to be allocated.
1106 */
1107 error = xfs_dialloc(tp, pip->i_ino, mode, okalloc,
1108 ialloc_context, call_again, &ino);
1109 if (error != 0) {
1110 return error;
1111 }
1112 if (*call_again || ino == NULLFSINO) {
1113 *ipp = NULL;
1114 return 0;
1115 }
1116 ASSERT(*ialloc_context == NULL);
1117
1118 /*
1119 * Get the in-core inode with the lock held exclusively.
1120 * This is because we're setting fields here we need
1121 * to prevent others from looking at until we're done.
1122 */
1123 error = xfs_trans_iget(tp->t_mountp, tp, ino,
1124 IGET_CREATE, XFS_ILOCK_EXCL, &ip);
1125 if (error != 0) {
1126 return error;
1127 }
1128 ASSERT(ip != NULL);
1129
1130 vp = XFS_ITOV(ip);
1131 vp->v_type = IFTOVT(mode);
1132 ip->i_d.di_mode = (__uint16_t)mode;
1133 ip->i_d.di_onlink = 0;
1134 ip->i_d.di_nlink = nlink;
1135 ASSERT(ip->i_d.di_nlink == nlink);
1136 ip->i_d.di_uid = current_fsuid(cr);
1137 ip->i_d.di_gid = current_fsgid(cr);
1138 ip->i_d.di_projid = prid;
1139 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1140
1141 /*
1142 * If the superblock version is up to where we support new format
1143 * inodes and this is currently an old format inode, then change
1144 * the inode version number now. This way we only do the conversion
1145 * here rather than here and in the flush/logging code.
1146 */
1147 if (XFS_SB_VERSION_HASNLINK(&tp->t_mountp->m_sb) &&
1148 ip->i_d.di_version == XFS_DINODE_VERSION_1) {
1149 ip->i_d.di_version = XFS_DINODE_VERSION_2;
1150 /*
1151 * We've already zeroed the old link count, the projid field,
1152 * and the pad field.
1153 */
1154 }
1155
1156 /*
1157 * Project ids won't be stored on disk if we are using a version 1 inode.
1158 */
1159 if ( (prid != 0) && (ip->i_d.di_version == XFS_DINODE_VERSION_1))
1160 xfs_bump_ino_vers2(tp, ip);
1161
1162 if (XFS_INHERIT_GID(pip, vp->v_vfsp)) {
1163 ip->i_d.di_gid = pip->i_d.di_gid;
1164 if ((pip->i_d.di_mode & S_ISGID) && (mode & S_IFMT) == S_IFDIR) {
1165 ip->i_d.di_mode |= S_ISGID;
1166 }
1167 }
1168
1169 /*
1170 * If the group ID of the new file does not match the effective group
1171 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1172 * (and only if the irix_sgid_inherit compatibility variable is set).
1173 */
1174 if ((irix_sgid_inherit) &&
1175 (ip->i_d.di_mode & S_ISGID) &&
1176 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1177 ip->i_d.di_mode &= ~S_ISGID;
1178 }
1179
1180 ip->i_d.di_size = 0;
1181 ip->i_d.di_nextents = 0;
1182 ASSERT(ip->i_d.di_nblocks == 0);
1183 xfs_ichgtime(ip, XFS_ICHGTIME_CHG|XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD);
1184 /*
1185 * di_gen will have been taken care of in xfs_iread.
1186 */
1187 ip->i_d.di_extsize = 0;
1188 ip->i_d.di_dmevmask = 0;
1189 ip->i_d.di_dmstate = 0;
1190 ip->i_d.di_flags = 0;
1191 flags = XFS_ILOG_CORE;
1192 switch (mode & S_IFMT) {
1193 case S_IFIFO:
1194 case S_IFCHR:
1195 case S_IFBLK:
1196 case S_IFSOCK:
1197 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1198 ip->i_df.if_u2.if_rdev = rdev;
1199 ip->i_df.if_flags = 0;
1200 flags |= XFS_ILOG_DEV;
1201 break;
1202 case S_IFREG:
1203 case S_IFDIR:
1204 if (unlikely(pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1205 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) {
1206 if ((mode & S_IFMT) == S_IFDIR) {
1207 ip->i_d.di_flags |= XFS_DIFLAG_RTINHERIT;
1208 } else {
1209 ip->i_d.di_flags |= XFS_DIFLAG_REALTIME;
1210 ip->i_iocore.io_flags |= XFS_IOCORE_RT;
1211 }
1212 }
1213 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1214 xfs_inherit_noatime)
1215 ip->i_d.di_flags |= XFS_DIFLAG_NOATIME;
1216 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1217 xfs_inherit_nodump)
1218 ip->i_d.di_flags |= XFS_DIFLAG_NODUMP;
1219 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1220 xfs_inherit_sync)
1221 ip->i_d.di_flags |= XFS_DIFLAG_SYNC;
1222 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1223 xfs_inherit_nosymlinks)
1224 ip->i_d.di_flags |= XFS_DIFLAG_NOSYMLINKS;
1225 }
1226 /* FALLTHROUGH */
1227 case S_IFLNK:
1228 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1229 ip->i_df.if_flags = XFS_IFEXTENTS;
1230 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1231 ip->i_df.if_u1.if_extents = NULL;
1232 break;
1233 default:
1234 ASSERT(0);
1235 }
1236 /*
1237 * Attribute fork settings for new inode.
1238 */
1239 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1240 ip->i_d.di_anextents = 0;
1241
1242 /*
1243 * Log the new values stuffed into the inode.
1244 */
1245 xfs_trans_log_inode(tp, ip, flags);
1246
1247 /* now that we have a v_type we can set Linux inode ops (& unlock) */
1248 VFS_INIT_VNODE(XFS_MTOVFS(tp->t_mountp), vp, XFS_ITOBHV(ip), 1);
1249
1250 *ipp = ip;
1251 return 0;
1252}
1253
1254/*
1255 * Check to make sure that there are no blocks allocated to the
1256 * file beyond the size of the file. We don't check this for
1257 * files with fixed size extents or real time extents, but we
1258 * at least do it for regular files.
1259 */
1260#ifdef DEBUG
1261void
1262xfs_isize_check(
1263 xfs_mount_t *mp,
1264 xfs_inode_t *ip,
1265 xfs_fsize_t isize)
1266{
1267 xfs_fileoff_t map_first;
1268 int nimaps;
1269 xfs_bmbt_irec_t imaps[2];
1270
1271 if ((ip->i_d.di_mode & S_IFMT) != S_IFREG)
1272 return;
1273
1274 if ( ip->i_d.di_flags & XFS_DIFLAG_REALTIME )
1275 return;
1276
1277 nimaps = 2;
1278 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1279 /*
1280 * The filesystem could be shutting down, so bmapi may return
1281 * an error.
1282 */
1283 if (xfs_bmapi(NULL, ip, map_first,
1284 (XFS_B_TO_FSB(mp,
1285 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1286 map_first),
1287 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1288 NULL))
1289 return;
1290 ASSERT(nimaps == 1);
1291 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1292}
1293#endif /* DEBUG */
1294
1295/*
1296 * Calculate the last possible buffered byte in a file. This must
1297 * include data that was buffered beyond the EOF by the write code.
1298 * This also needs to deal with overflowing the xfs_fsize_t type
1299 * which can happen for sizes near the limit.
1300 *
1301 * We also need to take into account any blocks beyond the EOF. It
1302 * may be the case that they were buffered by a write which failed.
1303 * In that case the pages will still be in memory, but the inode size
1304 * will never have been updated.
1305 */
1306xfs_fsize_t
1307xfs_file_last_byte(
1308 xfs_inode_t *ip)
1309{
1310 xfs_mount_t *mp;
1311 xfs_fsize_t last_byte;
1312 xfs_fileoff_t last_block;
1313 xfs_fileoff_t size_last_block;
1314 int error;
1315
1316 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE | MR_ACCESS));
1317
1318 mp = ip->i_mount;
1319 /*
1320 * Only check for blocks beyond the EOF if the extents have
1321 * been read in. This eliminates the need for the inode lock,
1322 * and it also saves us from looking when it really isn't
1323 * necessary.
1324 */
1325 if (ip->i_df.if_flags & XFS_IFEXTENTS) {
1326 error = xfs_bmap_last_offset(NULL, ip, &last_block,
1327 XFS_DATA_FORK);
1328 if (error) {
1329 last_block = 0;
1330 }
1331 } else {
1332 last_block = 0;
1333 }
1334 size_last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)ip->i_d.di_size);
1335 last_block = XFS_FILEOFF_MAX(last_block, size_last_block);
1336
1337 last_byte = XFS_FSB_TO_B(mp, last_block);
1338 if (last_byte < 0) {
1339 return XFS_MAXIOFFSET(mp);
1340 }
1341 last_byte += (1 << mp->m_writeio_log);
1342 if (last_byte < 0) {
1343 return XFS_MAXIOFFSET(mp);
1344 }
1345 return last_byte;
1346}
1347
1348#if defined(XFS_RW_TRACE)
1349STATIC void
1350xfs_itrunc_trace(
1351 int tag,
1352 xfs_inode_t *ip,
1353 int flag,
1354 xfs_fsize_t new_size,
1355 xfs_off_t toss_start,
1356 xfs_off_t toss_finish)
1357{
1358 if (ip->i_rwtrace == NULL) {
1359 return;
1360 }
1361
1362 ktrace_enter(ip->i_rwtrace,
1363 (void*)((long)tag),
1364 (void*)ip,
1365 (void*)(unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff),
1366 (void*)(unsigned long)(ip->i_d.di_size & 0xffffffff),
1367 (void*)((long)flag),
1368 (void*)(unsigned long)((new_size >> 32) & 0xffffffff),
1369 (void*)(unsigned long)(new_size & 0xffffffff),
1370 (void*)(unsigned long)((toss_start >> 32) & 0xffffffff),
1371 (void*)(unsigned long)(toss_start & 0xffffffff),
1372 (void*)(unsigned long)((toss_finish >> 32) & 0xffffffff),
1373 (void*)(unsigned long)(toss_finish & 0xffffffff),
1374 (void*)(unsigned long)current_cpu(),
1375 (void*)0,
1376 (void*)0,
1377 (void*)0,
1378 (void*)0);
1379}
1380#else
1381#define xfs_itrunc_trace(tag, ip, flag, new_size, toss_start, toss_finish)
1382#endif
1383
1384/*
1385 * Start the truncation of the file to new_size. The new size
1386 * must be smaller than the current size. This routine will
1387 * clear the buffer and page caches of file data in the removed
1388 * range, and xfs_itruncate_finish() will remove the underlying
1389 * disk blocks.
1390 *
1391 * The inode must have its I/O lock locked EXCLUSIVELY, and it
1392 * must NOT have the inode lock held at all. This is because we're
1393 * calling into the buffer/page cache code and we can't hold the
1394 * inode lock when we do so.
1395 *
1396 * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
1397 * or XFS_ITRUNC_MAYBE. The XFS_ITRUNC_MAYBE value should be used
1398 * in the case that the caller is locking things out of order and
1399 * may not be able to call xfs_itruncate_finish() with the inode lock
1400 * held without dropping the I/O lock. If the caller must drop the
1401 * I/O lock before calling xfs_itruncate_finish(), then xfs_itruncate_start()
1402 * must be called again with all the same restrictions as the initial
1403 * call.
1404 */
1405void
1406xfs_itruncate_start(
1407 xfs_inode_t *ip,
1408 uint flags,
1409 xfs_fsize_t new_size)
1410{
1411 xfs_fsize_t last_byte;
1412 xfs_off_t toss_start;
1413 xfs_mount_t *mp;
1414 vnode_t *vp;
1415
1416 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1417 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1418 ASSERT((flags == XFS_ITRUNC_DEFINITE) ||
1419 (flags == XFS_ITRUNC_MAYBE));
1420
1421 mp = ip->i_mount;
1422 vp = XFS_ITOV(ip);
1423 /*
1424 * Call VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES() to get rid of pages and buffers
1425 * overlapping the region being removed. We have to use
1426 * the less efficient VOP_FLUSHINVAL_PAGES() in the case that the
1427 * caller may not be able to finish the truncate without
1428 * dropping the inode's I/O lock. Make sure
1429 * to catch any pages brought in by buffers overlapping
1430 * the EOF by searching out beyond the isize by our
1431 * block size. We round new_size up to a block boundary
1432 * so that we don't toss things on the same block as
1433 * new_size but before it.
1434 *
1435 * Before calling VOP_TOSS_PAGES() or VOP_FLUSHINVAL_PAGES(), make sure to
1436 * call remapf() over the same region if the file is mapped.
1437 * This frees up mapped file references to the pages in the
1438 * given range and for the VOP_FLUSHINVAL_PAGES() case it ensures
1439 * that we get the latest mapped changes flushed out.
1440 */
1441 toss_start = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1442 toss_start = XFS_FSB_TO_B(mp, toss_start);
1443 if (toss_start < 0) {
1444 /*
1445 * The place to start tossing is beyond our maximum
1446 * file size, so there is no way that the data extended
1447 * out there.
1448 */
1449 return;
1450 }
1451 last_byte = xfs_file_last_byte(ip);
1452 xfs_itrunc_trace(XFS_ITRUNC_START, ip, flags, new_size, toss_start,
1453 last_byte);
1454 if (last_byte > toss_start) {
1455 if (flags & XFS_ITRUNC_DEFINITE) {
1456 VOP_TOSS_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1457 } else {
1458 VOP_FLUSHINVAL_PAGES(vp, toss_start, -1, FI_REMAPF_LOCKED);
1459 }
1460 }
1461
1462#ifdef DEBUG
1463 if (new_size == 0) {
1464 ASSERT(VN_CACHED(vp) == 0);
1465 }
1466#endif
1467}
1468
1469/*
1470 * Shrink the file to the given new_size. The new
1471 * size must be smaller than the current size.
1472 * This will free up the underlying blocks
1473 * in the removed range after a call to xfs_itruncate_start()
1474 * or xfs_atruncate_start().
1475 *
1476 * The transaction passed to this routine must have made
1477 * a permanent log reservation of at least XFS_ITRUNCATE_LOG_RES.
1478 * This routine may commit the given transaction and
1479 * start new ones, so make sure everything involved in
1480 * the transaction is tidy before calling here.
1481 * Some transaction will be returned to the caller to be
1482 * committed. The incoming transaction must already include
1483 * the inode, and both inode locks must be held exclusively.
1484 * The inode must also be "held" within the transaction. On
1485 * return the inode will be "held" within the returned transaction.
1486 * This routine does NOT require any disk space to be reserved
1487 * for it within the transaction.
1488 *
1489 * The fork parameter must be either xfs_attr_fork or xfs_data_fork,
1490 * and it indicates the fork which is to be truncated. For the
1491 * attribute fork we only support truncation to size 0.
1492 *
1493 * We use the sync parameter to indicate whether or not the first
1494 * transaction we perform might have to be synchronous. For the attr fork,
1495 * it needs to be so if the unlink of the inode is not yet known to be
1496 * permanent in the log. This keeps us from freeing and reusing the
1497 * blocks of the attribute fork before the unlink of the inode becomes
1498 * permanent.
1499 *
1500 * For the data fork, we normally have to run synchronously if we're
1501 * being called out of the inactive path or we're being called
1502 * out of the create path where we're truncating an existing file.
1503 * Either way, the truncate needs to be sync so blocks don't reappear
1504 * in the file with altered data in case of a crash. wsync filesystems
1505 * can run the first case async because anything that shrinks the inode
1506 * has to run sync so by the time we're called here from inactive, the
1507 * inode size is permanently set to 0.
1508 *
1509 * Calls from the truncate path always need to be sync unless we're
1510 * in a wsync filesystem and the file has already been unlinked.
1511 *
1512 * The caller is responsible for correctly setting the sync parameter.
1513 * It gets too hard for us to guess here which path we're being called
1514 * out of just based on inode state.
1515 */
1516int
1517xfs_itruncate_finish(
1518 xfs_trans_t **tp,
1519 xfs_inode_t *ip,
1520 xfs_fsize_t new_size,
1521 int fork,
1522 int sync)
1523{
1524 xfs_fsblock_t first_block;
1525 xfs_fileoff_t first_unmap_block;
1526 xfs_fileoff_t last_block;
1527 xfs_filblks_t unmap_len=0;
1528 xfs_mount_t *mp;
1529 xfs_trans_t *ntp;
1530 int done;
1531 int committed;
1532 xfs_bmap_free_t free_list;
1533 int error;
1534
1535 ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
1536 ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
1537 ASSERT((new_size == 0) || (new_size <= ip->i_d.di_size));
1538 ASSERT(*tp != NULL);
1539 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
1540 ASSERT(ip->i_transp == *tp);
1541 ASSERT(ip->i_itemp != NULL);
1542 ASSERT(ip->i_itemp->ili_flags & XFS_ILI_HOLD);
1543
1544
1545 ntp = *tp;
1546 mp = (ntp)->t_mountp;
1547 ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
1548
1549 /*
1550 * We only support truncating the entire attribute fork.
1551 */
1552 if (fork == XFS_ATTR_FORK) {
1553 new_size = 0LL;
1554 }
1555 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1556 xfs_itrunc_trace(XFS_ITRUNC_FINISH1, ip, 0, new_size, 0, 0);
1557 /*
1558 * The first thing we do is set the size to new_size permanently
1559 * on disk. This way we don't have to worry about anyone ever
1560 * being able to look at the data being freed even in the face
1561 * of a crash. What we're getting around here is the case where
1562 * we free a block, it is allocated to another file, it is written
1563 * to, and then we crash. If the new data gets written to the
1564 * file but the log buffers containing the free and reallocation
1565 * don't, then we'd end up with garbage in the blocks being freed.
1566 * As long as we make the new_size permanent before actually
1567 * freeing any blocks it doesn't matter if they get writtten to.
1568 *
1569 * The callers must signal into us whether or not the size
1570 * setting here must be synchronous. There are a few cases
1571 * where it doesn't have to be synchronous. Those cases
1572 * occur if the file is unlinked and we know the unlink is
1573 * permanent or if the blocks being truncated are guaranteed
1574 * to be beyond the inode eof (regardless of the link count)
1575 * and the eof value is permanent. Both of these cases occur
1576 * only on wsync-mounted filesystems. In those cases, we're
1577 * guaranteed that no user will ever see the data in the blocks
1578 * that are being truncated so the truncate can run async.
1579 * In the free beyond eof case, the file may wind up with
1580 * more blocks allocated to it than it needs if we crash
1581 * and that won't get fixed until the next time the file
1582 * is re-opened and closed but that's ok as that shouldn't
1583 * be too many blocks.
1584 *
1585 * However, we can't just make all wsync xactions run async
1586 * because there's one call out of the create path that needs
1587 * to run sync where it's truncating an existing file to size
1588 * 0 whose size is > 0.
1589 *
1590 * It's probably possible to come up with a test in this
1591 * routine that would correctly distinguish all the above
1592 * cases from the values of the function parameters and the
1593 * inode state but for sanity's sake, I've decided to let the
1594 * layers above just tell us. It's simpler to correctly figure
1595 * out in the layer above exactly under what conditions we
1596 * can run async and I think it's easier for others read and
1597 * follow the logic in case something has to be changed.
1598 * cscope is your friend -- rcc.
1599 *
1600 * The attribute fork is much simpler.
1601 *
1602 * For the attribute fork we allow the caller to tell us whether
1603 * the unlink of the inode that led to this call is yet permanent
1604 * in the on disk log. If it is not and we will be freeing extents
1605 * in this inode then we make the first transaction synchronous
1606 * to make sure that the unlink is permanent by the time we free
1607 * the blocks.
1608 */
1609 if (fork == XFS_DATA_FORK) {
1610 if (ip->i_d.di_nextents > 0) {
1611 ip->i_d.di_size = new_size;
1612 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1613 }
1614 } else if (sync) {
1615 ASSERT(!(mp->m_flags & XFS_MOUNT_WSYNC));
1616 if (ip->i_d.di_anextents > 0)
1617 xfs_trans_set_sync(ntp);
1618 }
1619 ASSERT(fork == XFS_DATA_FORK ||
1620 (fork == XFS_ATTR_FORK &&
1621 ((sync && !(mp->m_flags & XFS_MOUNT_WSYNC)) ||
1622 (sync == 0 && (mp->m_flags & XFS_MOUNT_WSYNC)))));
1623
1624 /*
1625 * Since it is possible for space to become allocated beyond
1626 * the end of the file (in a crash where the space is allocated
1627 * but the inode size is not yet updated), simply remove any
1628 * blocks which show up between the new EOF and the maximum
1629 * possible file size. If the first block to be removed is
1630 * beyond the maximum file size (ie it is the same as last_block),
1631 * then there is nothing to do.
1632 */
1633 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1634 ASSERT(first_unmap_block <= last_block);
1635 done = 0;
1636 if (last_block == first_unmap_block) {
1637 done = 1;
1638 } else {
1639 unmap_len = last_block - first_unmap_block + 1;
1640 }
1641 while (!done) {
1642 /*
1643 * Free up up to XFS_ITRUNC_MAX_EXTENTS. xfs_bunmapi()
1644 * will tell us whether it freed the entire range or
1645 * not. If this is a synchronous mount (wsync),
1646 * then we can tell bunmapi to keep all the
1647 * transactions asynchronous since the unlink
1648 * transaction that made this inode inactive has
1649 * already hit the disk. There's no danger of
1650 * the freed blocks being reused, there being a
1651 * crash, and the reused blocks suddenly reappearing
1652 * in this file with garbage in them once recovery
1653 * runs.
1654 */
1655 XFS_BMAP_INIT(&free_list, &first_block);
1656 error = xfs_bunmapi(ntp, ip, first_unmap_block,
1657 unmap_len,
1658 XFS_BMAPI_AFLAG(fork) |
1659 (sync ? 0 : XFS_BMAPI_ASYNC),
1660 XFS_ITRUNC_MAX_EXTENTS,
1661 &first_block, &free_list, &done);
1662 if (error) {
1663 /*
1664 * If the bunmapi call encounters an error,
1665 * return to the caller where the transaction
1666 * can be properly aborted. We just need to
1667 * make sure we're not holding any resources
1668 * that we were not when we came in.
1669 */
1670 xfs_bmap_cancel(&free_list);
1671 return error;
1672 }
1673
1674 /*
1675 * Duplicate the transaction that has the permanent
1676 * reservation and commit the old transaction.
1677 */
1678 error = xfs_bmap_finish(tp, &free_list, first_block,
1679 &committed);
1680 ntp = *tp;
1681 if (error) {
1682 /*
1683 * If the bmap finish call encounters an error,
1684 * return to the caller where the transaction
1685 * can be properly aborted. We just need to
1686 * make sure we're not holding any resources
1687 * that we were not when we came in.
1688 *
1689 * Aborting from this point might lose some
1690 * blocks in the file system, but oh well.
1691 */
1692 xfs_bmap_cancel(&free_list);
1693 if (committed) {
1694 /*
1695 * If the passed in transaction committed
1696 * in xfs_bmap_finish(), then we want to
1697 * add the inode to this one before returning.
1698 * This keeps things simple for the higher
1699 * level code, because it always knows that
1700 * the inode is locked and held in the
1701 * transaction that returns to it whether
1702 * errors occur or not. We don't mark the
1703 * inode dirty so that this transaction can
1704 * be easily aborted if possible.
1705 */
1706 xfs_trans_ijoin(ntp, ip,
1707 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1708 xfs_trans_ihold(ntp, ip);
1709 }
1710 return error;
1711 }
1712
1713 if (committed) {
1714 /*
1715 * The first xact was committed,
1716 * so add the inode to the new one.
1717 * Mark it dirty so it will be logged
1718 * and moved forward in the log as
1719 * part of every commit.
1720 */
1721 xfs_trans_ijoin(ntp, ip,
1722 XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1723 xfs_trans_ihold(ntp, ip);
1724 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1725 }
1726 ntp = xfs_trans_dup(ntp);
1727 (void) xfs_trans_commit(*tp, 0, NULL);
1728 *tp = ntp;
1729 error = xfs_trans_reserve(ntp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1730 XFS_TRANS_PERM_LOG_RES,
1731 XFS_ITRUNCATE_LOG_COUNT);
1732 /*
1733 * Add the inode being truncated to the next chained
1734 * transaction.
1735 */
1736 xfs_trans_ijoin(ntp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1737 xfs_trans_ihold(ntp, ip);
1738 if (error)
1739 return (error);
1740 }
1741 /*
1742 * Only update the size in the case of the data fork, but
1743 * always re-log the inode so that our permanent transaction
1744 * can keep on rolling it forward in the log.
1745 */
1746 if (fork == XFS_DATA_FORK) {
1747 xfs_isize_check(mp, ip, new_size);
1748 ip->i_d.di_size = new_size;
1749 }
1750 xfs_trans_log_inode(ntp, ip, XFS_ILOG_CORE);
1751 ASSERT((new_size != 0) ||
1752 (fork == XFS_ATTR_FORK) ||
1753 (ip->i_delayed_blks == 0));
1754 ASSERT((new_size != 0) ||
1755 (fork == XFS_ATTR_FORK) ||
1756 (ip->i_d.di_nextents == 0));
1757 xfs_itrunc_trace(XFS_ITRUNC_FINISH2, ip, 0, new_size, 0, 0);
1758 return 0;
1759}
1760
1761
1762/*
1763 * xfs_igrow_start
1764 *
1765 * Do the first part of growing a file: zero any data in the last
1766 * block that is beyond the old EOF. We need to do this before
1767 * the inode is joined to the transaction to modify the i_size.
1768 * That way we can drop the inode lock and call into the buffer
1769 * cache to get the buffer mapping the EOF.
1770 */
1771int
1772xfs_igrow_start(
1773 xfs_inode_t *ip,
1774 xfs_fsize_t new_size,
1775 cred_t *credp)
1776{
1777 xfs_fsize_t isize;
1778 int error;
1779
1780 ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
1781 ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE) != 0);
1782 ASSERT(new_size > ip->i_d.di_size);
1783
1784 error = 0;
1785 isize = ip->i_d.di_size;
1786 /*
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 */
1791 error = xfs_zero_eof(XFS_ITOV(ip), &ip->i_iocore, new_size, isize,
1792 new_size);
1793 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 =
1876 INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC &&
1877 XFS_AGI_GOOD_VERSION(INT_GET(agi->agi_versionnum, ARCH_CONVERT));
1878 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]);
1892 ASSERT(INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != agino);
1893
1894 if (INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != NULLAGINO) {
1895 /*
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);
1921 INT_SET(agi->agi_unlinked[bucket_index], ARCH_CONVERT, agino);
1922 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 =
1979 INT_GET(agi->agi_magicnum, ARCH_CONVERT) == XFS_AGI_MAGIC &&
1980 XFS_AGI_GOOD_VERSION(INT_GET(agi->agi_versionnum, ARCH_CONVERT));
1981 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;
1998 ASSERT(INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) != NULLAGINO);
1999 ASSERT(agi->agi_unlinked[bucket_index]);
2000
2001 if (INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT) == agino) {
2002 /*
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);
2035 INT_SET(agi->agi_unlinked[bucket_index], ARCH_CONVERT, next_agino);
2036 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 */
2044 next_agino = INT_GET(agi->agi_unlinked[bucket_index], ARCH_CONVERT);
2045 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
3367 if (XFS_TEST_ERROR(INT_GET(dip->di_core.di_magic,ARCH_CONVERT) != XFS_DINODE_MAGIC,
3368 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3369 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3370 "xfs_iflush: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3371 ip->i_ino, (int) INT_GET(dip->di_core.di_magic, ARCH_CONVERT), dip);
3372 goto corrupt_out;
3373 }
3374 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3375 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3376 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3377 "xfs_iflush: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3378 ip->i_ino, ip, ip->i_d.di_magic);
3379 goto corrupt_out;
3380 }
3381 if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3382 if (XFS_TEST_ERROR(
3383 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3384 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3385 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3386 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3387 "xfs_iflush: Bad regular inode %Lu, ptr 0x%p",
3388 ip->i_ino, ip);
3389 goto corrupt_out;
3390 }
3391 } else if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
3392 if (XFS_TEST_ERROR(
3393 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3394 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3395 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3396 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3397 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3398 "xfs_iflush: Bad directory inode %Lu, ptr 0x%p",
3399 ip->i_ino, ip);
3400 goto corrupt_out;
3401 }
3402 }
3403 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3404 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3405 XFS_RANDOM_IFLUSH_5)) {
3406 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3407 "xfs_iflush: detected corrupt incore inode %Lu, total extents = %d, nblocks = %Ld, ptr 0x%p",
3408 ip->i_ino,
3409 ip->i_d.di_nextents + ip->i_d.di_anextents,
3410 ip->i_d.di_nblocks,
3411 ip);
3412 goto corrupt_out;
3413 }
3414 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3415 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3416 xfs_cmn_err(XFS_PTAG_IFLUSH, CE_ALERT, mp,
3417 "xfs_iflush: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3418 ip->i_ino, ip->i_d.di_forkoff, ip);
3419 goto corrupt_out;
3420 }
3421 /*
3422 * bump the flush iteration count, used to detect flushes which
3423 * postdate a log record during recovery.
3424 */
3425
3426 ip->i_d.di_flushiter++;
3427
3428 /*
3429 * Copy the dirty parts of the inode into the on-disk
3430 * inode. We always copy out the core of the inode,
3431 * because if the inode is dirty at all the core must
3432 * be.
3433 */
3434 xfs_xlate_dinode_core((xfs_caddr_t)&(dip->di_core), &(ip->i_d), -1);
3435
3436 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3437 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3438 ip->i_d.di_flushiter = 0;
3439
3440 /*
3441 * If this is really an old format inode and the superblock version
3442 * has not been updated to support only new format inodes, then
3443 * convert back to the old inode format. If the superblock version
3444 * has been updated, then make the conversion permanent.
3445 */
3446 ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1 ||
3447 XFS_SB_VERSION_HASNLINK(&mp->m_sb));
3448 if (ip->i_d.di_version == XFS_DINODE_VERSION_1) {
3449 if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) {
3450 /*
3451 * Convert it back.
3452 */
3453 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3454 INT_SET(dip->di_core.di_onlink, ARCH_CONVERT, ip->i_d.di_nlink);
3455 } else {
3456 /*
3457 * The superblock version has already been bumped,
3458 * so just make the conversion to the new inode
3459 * format permanent.
3460 */
3461 ip->i_d.di_version = XFS_DINODE_VERSION_2;
3462 INT_SET(dip->di_core.di_version, ARCH_CONVERT, XFS_DINODE_VERSION_2);
3463 ip->i_d.di_onlink = 0;
3464 dip->di_core.di_onlink = 0;
3465 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3466 memset(&(dip->di_core.di_pad[0]), 0,
3467 sizeof(dip->di_core.di_pad));
3468 ASSERT(ip->i_d.di_projid == 0);
3469 }
3470 }
3471
3472 if (xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp) == EFSCORRUPTED) {
3473 goto corrupt_out;
3474 }
3475
3476 if (XFS_IFORK_Q(ip)) {
3477 /*
3478 * The only error from xfs_iflush_fork is on the data fork.
3479 */
3480 (void) xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3481 }
3482 xfs_inobp_check(mp, bp);
3483
3484 /*
3485 * We've recorded everything logged in the inode, so we'd
3486 * like to clear the ilf_fields bits so we don't log and
3487 * flush things unnecessarily. However, we can't stop
3488 * logging all this information until the data we've copied
3489 * into the disk buffer is written to disk. If we did we might
3490 * overwrite the copy of the inode in the log with all the
3491 * data after re-logging only part of it, and in the face of
3492 * a crash we wouldn't have all the data we need to recover.
3493 *
3494 * What we do is move the bits to the ili_last_fields field.
3495 * When logging the inode, these bits are moved back to the
3496 * ilf_fields field. In the xfs_iflush_done() routine we
3497 * clear ili_last_fields, since we know that the information
3498 * those bits represent is permanently on disk. As long as
3499 * the flush completes before the inode is logged again, then
3500 * both ilf_fields and ili_last_fields will be cleared.
3501 *
3502 * We can play with the ilf_fields bits here, because the inode
3503 * lock must be held exclusively in order to set bits there
3504 * and the flush lock protects the ili_last_fields bits.
3505 * Set ili_logged so the flush done
3506 * routine can tell whether or not to look in the AIL.
3507 * Also, store the current LSN of the inode so that we can tell
3508 * whether the item has moved in the AIL from xfs_iflush_done().
3509 * In order to read the lsn we need the AIL lock, because
3510 * it is a 64 bit value that cannot be read atomically.
3511 */
3512 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
3513 iip->ili_last_fields = iip->ili_format.ilf_fields;
3514 iip->ili_format.ilf_fields = 0;
3515 iip->ili_logged = 1;
3516
3517 ASSERT(sizeof(xfs_lsn_t) == 8); /* don't lock if it shrinks */
3518 AIL_LOCK(mp,s);
3519 iip->ili_flush_lsn = iip->ili_item.li_lsn;
3520 AIL_UNLOCK(mp, s);
3521
3522 /*
3523 * Attach the function xfs_iflush_done to the inode's
3524 * buffer. This will remove the inode from the AIL
3525 * and unlock the inode's flush lock when the inode is
3526 * completely written to disk.
3527 */
3528 xfs_buf_attach_iodone(bp, (void(*)(xfs_buf_t*,xfs_log_item_t*))
3529 xfs_iflush_done, (xfs_log_item_t *)iip);
3530
3531 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
3532 ASSERT(XFS_BUF_IODONE_FUNC(bp) != NULL);
3533 } else {
3534 /*
3535 * We're flushing an inode which is not in the AIL and has
3536 * not been logged but has i_update_core set. For this
3537 * case we can use a B_DELWRI flush and immediately drop
3538 * the inode flush lock because we can avoid the whole
3539 * AIL state thing. It's OK to drop the flush lock now,
3540 * because we've already locked the buffer and to do anything
3541 * you really need both.
3542 */
3543 if (iip != NULL) {
3544 ASSERT(iip->ili_logged == 0);
3545 ASSERT(iip->ili_last_fields == 0);
3546 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
3547 }
3548 xfs_ifunlock(ip);
3549 }
3550
3551 return 0;
3552
3553corrupt_out:
3554 return XFS_ERROR(EFSCORRUPTED);
3555}
3556
3557
3558/*
3559 * Flush all inactive inodes in mp. Return true if no user references
3560 * were found, false otherwise.
3561 */
3562int
3563xfs_iflush_all(
3564 xfs_mount_t *mp,
3565 int flag)
3566{
3567 int busy;
3568 int done;
3569 int purged;
3570 xfs_inode_t *ip;
3571 vmap_t vmap;
3572 vnode_t *vp;
3573
3574 busy = done = 0;
3575 while (!done) {
3576 purged = 0;
3577 XFS_MOUNT_ILOCK(mp);
3578 ip = mp->m_inodes;
3579 if (ip == NULL) {
3580 break;
3581 }
3582 do {
3583 /* Make sure we skip markers inserted by sync */
3584 if (ip->i_mount == NULL) {
3585 ip = ip->i_mnext;
3586 continue;
3587 }
3588
3589 /*
3590 * It's up to our caller to purge the root
3591 * and quota vnodes later.
3592 */
3593 vp = XFS_ITOV_NULL(ip);
3594
3595 if (!vp) {
3596 XFS_MOUNT_IUNLOCK(mp);
3597 xfs_finish_reclaim(ip, 0, XFS_IFLUSH_ASYNC);
3598 purged = 1;
3599 break;
3600 }
3601
3602 if (vn_count(vp) != 0) {
3603 if (vn_count(vp) == 1 &&
3604 (ip == mp->m_rootip ||
3605 (mp->m_quotainfo &&
3606 (ip->i_ino == mp->m_sb.sb_uquotino ||
3607 ip->i_ino == mp->m_sb.sb_gquotino)))) {
3608
3609 ip = ip->i_mnext;
3610 continue;
3611 }
3612 if (!(flag & XFS_FLUSH_ALL)) {
3613 busy = 1;
3614 done = 1;
3615 break;
3616 }
3617 /*
3618 * Ignore busy inodes but continue flushing
3619 * others.
3620 */
3621 ip = ip->i_mnext;
3622 continue;
3623 }
3624 /*
3625 * Sample vp mapping while holding mp locked on MP
3626 * systems, so we don't purge a reclaimed or
3627 * nonexistent vnode. We break from the loop
3628 * since we know that we modify
3629 * it by pulling ourselves from it in xfs_reclaim()
3630 * called via vn_purge() below. Set ip to the next
3631 * entry in the list anyway so we'll know below
3632 * whether we reached the end or not.
3633 */
3634 VMAP(vp, vmap);
3635 XFS_MOUNT_IUNLOCK(mp);
3636
3637 vn_purge(vp, &vmap);
3638
3639 purged = 1;
3640 break;
3641 } while (ip != mp->m_inodes);
3642 /*
3643 * We need to distinguish between when we exit the loop
3644 * after a purge and when we simply hit the end of the
3645 * list. We can't use the (ip == mp->m_inodes) test,
3646 * because when we purge an inode at the start of the list
3647 * the next inode on the list becomes mp->m_inodes. That
3648 * would cause such a test to bail out early. The purged
3649 * variable tells us how we got out of the loop.
3650 */
3651 if (!purged) {
3652 done = 1;
3653 }
3654 }
3655 XFS_MOUNT_IUNLOCK(mp);
3656 return !busy;
3657}
3658
3659
3660/*
3661 * xfs_iaccess: check accessibility of inode for mode.
3662 */
3663int
3664xfs_iaccess(
3665 xfs_inode_t *ip,
3666 mode_t mode,
3667 cred_t *cr)
3668{
3669 int error;
3670 mode_t orgmode = mode;
3671 struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip));
3672
3673 if (mode & S_IWUSR) {
3674 umode_t imode = inode->i_mode;
3675
3676 if (IS_RDONLY(inode) &&
3677 (S_ISREG(imode) || S_ISDIR(imode) || S_ISLNK(imode)))
3678 return XFS_ERROR(EROFS);
3679
3680 if (IS_IMMUTABLE(inode))
3681 return XFS_ERROR(EACCES);
3682 }
3683
3684 /*
3685 * If there's an Access Control List it's used instead of
3686 * the mode bits.
3687 */
3688 if ((error = _ACL_XFS_IACCESS(ip, mode, cr)) != -1)
3689 return error ? XFS_ERROR(error) : 0;
3690
3691 if (current_fsuid(cr) != ip->i_d.di_uid) {
3692 mode >>= 3;
3693 if (!in_group_p((gid_t)ip->i_d.di_gid))
3694 mode >>= 3;
3695 }
3696
3697 /*
3698 * If the DACs are ok we don't need any capability check.
3699 */
3700 if ((ip->i_d.di_mode & mode) == mode)
3701 return 0;
3702 /*
3703 * Read/write DACs are always overridable.
3704 * Executable DACs are overridable if at least one exec bit is set.
3705 */
3706 if (!(orgmode & S_IXUSR) ||
3707 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
3708 if (capable_cred(cr, CAP_DAC_OVERRIDE))
3709 return 0;
3710
3711 if ((orgmode == S_IRUSR) ||
3712 (S_ISDIR(inode->i_mode) && (!(orgmode & S_IWUSR)))) {
3713 if (capable_cred(cr, CAP_DAC_READ_SEARCH))
3714 return 0;
3715#ifdef NOISE
3716 cmn_err(CE_NOTE, "Ick: mode=%o, orgmode=%o", mode, orgmode);
3717#endif /* NOISE */
3718 return XFS_ERROR(EACCES);
3719 }
3720 return XFS_ERROR(EACCES);
3721}
3722
3723/*
3724 * xfs_iroundup: round up argument to next power of two
3725 */
3726uint
3727xfs_iroundup(
3728 uint v)
3729{
3730 int i;
3731 uint m;
3732
3733 if ((v & (v - 1)) == 0)
3734 return v;
3735 ASSERT((v & 0x80000000) == 0);
3736 if ((v & (v + 1)) == 0)
3737 return v + 1;
3738 for (i = 0, m = 1; i < 31; i++, m <<= 1) {
3739 if (v & m)
3740 continue;
3741 v |= m;
3742 if ((v & (v + 1)) == 0)
3743 return v + 1;
3744 }
3745 ASSERT(0);
3746 return( 0 );
3747}
3748
3749/*
3750 * Change the requested timestamp in the given inode.
3751 * We don't lock across timestamp updates, and we don't log them but
3752 * we do record the fact that there is dirty information in core.
3753 *
3754 * NOTE -- callers MUST combine XFS_ICHGTIME_MOD or XFS_ICHGTIME_CHG
3755 * with XFS_ICHGTIME_ACC to be sure that access time
3756 * update will take. Calling first with XFS_ICHGTIME_ACC
3757 * and then XFS_ICHGTIME_MOD may fail to modify the access
3758 * timestamp if the filesystem is mounted noacctm.
3759 */
3760void
3761xfs_ichgtime(xfs_inode_t *ip,
3762 int flags)
3763{
3764 timespec_t tv;
3765 vnode_t *vp = XFS_ITOV(ip);
3766 struct inode *inode = LINVFS_GET_IP(vp);
3767
3768 /*
3769 * We're not supposed to change timestamps in readonly-mounted
3770 * filesystems. Throw it away if anyone asks us.
3771 */
3772 if (unlikely(vp->v_vfsp->vfs_flag & VFS_RDONLY))
3773 return;
3774
3775 /*
3776 * Don't update access timestamps on reads if mounted "noatime"
3777 * Throw it away if anyone asks us.
3778 */
3779 if ((ip->i_mount->m_flags & XFS_MOUNT_NOATIME || IS_NOATIME(inode)) &&
3780 ((flags & (XFS_ICHGTIME_ACC|XFS_ICHGTIME_MOD|XFS_ICHGTIME_CHG))
3781 == XFS_ICHGTIME_ACC))
3782 return;
3783
3784 nanotime(&tv);
3785 if (flags & XFS_ICHGTIME_MOD) {
3786 VN_MTIMESET(vp, &tv);
3787 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
3788 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
3789 }
3790 if (flags & XFS_ICHGTIME_ACC) {
3791 VN_ATIMESET(vp, &tv);
3792 ip->i_d.di_atime.t_sec = (__int32_t)tv.tv_sec;
3793 ip->i_d.di_atime.t_nsec = (__int32_t)tv.tv_nsec;
3794 }
3795 if (flags & XFS_ICHGTIME_CHG) {
3796 VN_CTIMESET(vp, &tv);
3797 ip->i_d.di_ctime.t_sec = (__int32_t)tv.tv_sec;
3798 ip->i_d.di_ctime.t_nsec = (__int32_t)tv.tv_nsec;
3799 }
3800
3801 /*
3802 * We update the i_update_core field _after_ changing
3803 * the timestamps in order to coordinate properly with
3804 * xfs_iflush() so that we don't lose timestamp updates.
3805 * This keeps us from having to hold the inode lock
3806 * while doing this. We use the SYNCHRONIZE macro to
3807 * ensure that the compiler does not reorder the update
3808 * of i_update_core above the timestamp updates above.
3809 */
3810 SYNCHRONIZE();
3811 ip->i_update_core = 1;
3812 if (!(inode->i_state & I_LOCK))
3813 mark_inode_dirty_sync(inode);
3814}
3815
3816#ifdef XFS_ILOCK_TRACE
3817ktrace_t *xfs_ilock_trace_buf;
3818
3819void
3820xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
3821{
3822 ktrace_enter(ip->i_lock_trace,
3823 (void *)ip,
3824 (void *)(unsigned long)lock, /* 1 = LOCK, 3=UNLOCK, etc */
3825 (void *)(unsigned long)lockflags, /* XFS_ILOCK_EXCL etc */
3826 (void *)ra, /* caller of ilock */
3827 (void *)(unsigned long)current_cpu(),
3828 (void *)(unsigned long)current_pid(),
3829 NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL);
3830}
3831#endif