Loading...
1/*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include <linux/log2.h>
19
20#include "xfs.h"
21#include "xfs_fs.h"
22#include "xfs_types.h"
23#include "xfs_bit.h"
24#include "xfs_log.h"
25#include "xfs_inum.h"
26#include "xfs_trans.h"
27#include "xfs_trans_priv.h"
28#include "xfs_sb.h"
29#include "xfs_ag.h"
30#include "xfs_mount.h"
31#include "xfs_bmap_btree.h"
32#include "xfs_alloc_btree.h"
33#include "xfs_ialloc_btree.h"
34#include "xfs_attr_sf.h"
35#include "xfs_dinode.h"
36#include "xfs_inode.h"
37#include "xfs_buf_item.h"
38#include "xfs_inode_item.h"
39#include "xfs_btree.h"
40#include "xfs_alloc.h"
41#include "xfs_ialloc.h"
42#include "xfs_bmap.h"
43#include "xfs_error.h"
44#include "xfs_utils.h"
45#include "xfs_quota.h"
46#include "xfs_filestream.h"
47#include "xfs_vnodeops.h"
48#include "xfs_trace.h"
49
50kmem_zone_t *xfs_ifork_zone;
51kmem_zone_t *xfs_inode_zone;
52
53/*
54 * Used in xfs_itruncate_extents(). This is the maximum number of extents
55 * freed from a file in a single transaction.
56 */
57#define XFS_ITRUNC_MAX_EXTENTS 2
58
59STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
60STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
61STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
62STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
63
64#ifdef DEBUG
65/*
66 * Make sure that the extents in the given memory buffer
67 * are valid.
68 */
69STATIC void
70xfs_validate_extents(
71 xfs_ifork_t *ifp,
72 int nrecs,
73 xfs_exntfmt_t fmt)
74{
75 xfs_bmbt_irec_t irec;
76 xfs_bmbt_rec_host_t rec;
77 int i;
78
79 for (i = 0; i < nrecs; i++) {
80 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
81 rec.l0 = get_unaligned(&ep->l0);
82 rec.l1 = get_unaligned(&ep->l1);
83 xfs_bmbt_get_all(&rec, &irec);
84 if (fmt == XFS_EXTFMT_NOSTATE)
85 ASSERT(irec.br_state == XFS_EXT_NORM);
86 }
87}
88#else /* DEBUG */
89#define xfs_validate_extents(ifp, nrecs, fmt)
90#endif /* DEBUG */
91
92/*
93 * Check that none of the inode's in the buffer have a next
94 * unlinked field of 0.
95 */
96#if defined(DEBUG)
97void
98xfs_inobp_check(
99 xfs_mount_t *mp,
100 xfs_buf_t *bp)
101{
102 int i;
103 int j;
104 xfs_dinode_t *dip;
105
106 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog;
107
108 for (i = 0; i < j; i++) {
109 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
110 i * mp->m_sb.sb_inodesize);
111 if (!dip->di_next_unlinked) {
112 xfs_alert(mp,
113 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.",
114 bp);
115 ASSERT(dip->di_next_unlinked);
116 }
117 }
118}
119#endif
120
121/*
122 * Find the buffer associated with the given inode map
123 * We do basic validation checks on the buffer once it has been
124 * retrieved from disk.
125 */
126STATIC int
127xfs_imap_to_bp(
128 xfs_mount_t *mp,
129 xfs_trans_t *tp,
130 struct xfs_imap *imap,
131 xfs_buf_t **bpp,
132 uint buf_flags,
133 uint iget_flags)
134{
135 int error;
136 int i;
137 int ni;
138 xfs_buf_t *bp;
139
140 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
141 (int)imap->im_len, buf_flags, &bp);
142 if (error) {
143 if (error != EAGAIN) {
144 xfs_warn(mp,
145 "%s: xfs_trans_read_buf() returned error %d.",
146 __func__, error);
147 } else {
148 ASSERT(buf_flags & XBF_TRYLOCK);
149 }
150 return error;
151 }
152
153 /*
154 * Validate the magic number and version of every inode in the buffer
155 * (if DEBUG kernel) or the first inode in the buffer, otherwise.
156 */
157#ifdef DEBUG
158 ni = BBTOB(imap->im_len) >> mp->m_sb.sb_inodelog;
159#else /* usual case */
160 ni = 1;
161#endif
162
163 for (i = 0; i < ni; i++) {
164 int di_ok;
165 xfs_dinode_t *dip;
166
167 dip = (xfs_dinode_t *)xfs_buf_offset(bp,
168 (i << mp->m_sb.sb_inodelog));
169 di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) &&
170 XFS_DINODE_GOOD_VERSION(dip->di_version);
171 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
172 XFS_ERRTAG_ITOBP_INOTOBP,
173 XFS_RANDOM_ITOBP_INOTOBP))) {
174 if (iget_flags & XFS_IGET_UNTRUSTED) {
175 xfs_trans_brelse(tp, bp);
176 return XFS_ERROR(EINVAL);
177 }
178 XFS_CORRUPTION_ERROR("xfs_imap_to_bp",
179 XFS_ERRLEVEL_HIGH, mp, dip);
180#ifdef DEBUG
181 xfs_emerg(mp,
182 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
183 (unsigned long long)imap->im_blkno, i,
184 be16_to_cpu(dip->di_magic));
185 ASSERT(0);
186#endif
187 xfs_trans_brelse(tp, bp);
188 return XFS_ERROR(EFSCORRUPTED);
189 }
190 }
191
192 xfs_inobp_check(mp, bp);
193
194 /*
195 * Mark the buffer as an inode buffer now that it looks good
196 */
197 XFS_BUF_SET_VTYPE(bp, B_FS_INO);
198
199 *bpp = bp;
200 return 0;
201}
202
203/*
204 * This routine is called to map an inode number within a file
205 * system to the buffer containing the on-disk version of the
206 * inode. It returns a pointer to the buffer containing the
207 * on-disk inode in the bpp parameter, and in the dip parameter
208 * it returns a pointer to the on-disk inode within that buffer.
209 *
210 * If a non-zero error is returned, then the contents of bpp and
211 * dipp are undefined.
212 *
213 * Use xfs_imap() to determine the size and location of the
214 * buffer to read from disk.
215 */
216int
217xfs_inotobp(
218 xfs_mount_t *mp,
219 xfs_trans_t *tp,
220 xfs_ino_t ino,
221 xfs_dinode_t **dipp,
222 xfs_buf_t **bpp,
223 int *offset,
224 uint imap_flags)
225{
226 struct xfs_imap imap;
227 xfs_buf_t *bp;
228 int error;
229
230 imap.im_blkno = 0;
231 error = xfs_imap(mp, tp, ino, &imap, imap_flags);
232 if (error)
233 return error;
234
235 error = xfs_imap_to_bp(mp, tp, &imap, &bp, XBF_LOCK, imap_flags);
236 if (error)
237 return error;
238
239 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
240 *bpp = bp;
241 *offset = imap.im_boffset;
242 return 0;
243}
244
245
246/*
247 * This routine is called to map an inode to the buffer containing
248 * the on-disk version of the inode. It returns a pointer to the
249 * buffer containing the on-disk inode in the bpp parameter, and in
250 * the dip parameter it returns a pointer to the on-disk inode within
251 * that buffer.
252 *
253 * If a non-zero error is returned, then the contents of bpp and
254 * dipp are undefined.
255 *
256 * The inode is expected to already been mapped to its buffer and read
257 * in once, thus we can use the mapping information stored in the inode
258 * rather than calling xfs_imap(). This allows us to avoid the overhead
259 * of looking at the inode btree for small block file systems
260 * (see xfs_imap()).
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 uint buf_flags)
270{
271 xfs_buf_t *bp;
272 int error;
273
274 ASSERT(ip->i_imap.im_blkno != 0);
275
276 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
277 if (error)
278 return error;
279
280 if (!bp) {
281 ASSERT(buf_flags & XBF_TRYLOCK);
282 ASSERT(tp == NULL);
283 *bpp = NULL;
284 return EAGAIN;
285 }
286
287 *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
288 *bpp = bp;
289 return 0;
290}
291
292/*
293 * Move inode type and inode format specific information from the
294 * on-disk inode to the in-core inode. For fifos, devs, and sockets
295 * this means set if_rdev to the proper value. For files, directories,
296 * and symlinks this means to bring in the in-line data or extent
297 * pointers. For a file in B-tree format, only the root is immediately
298 * brought in-core. The rest will be in-lined in if_extents when it
299 * is first referenced (see xfs_iread_extents()).
300 */
301STATIC int
302xfs_iformat(
303 xfs_inode_t *ip,
304 xfs_dinode_t *dip)
305{
306 xfs_attr_shortform_t *atp;
307 int size;
308 int error;
309 xfs_fsize_t di_size;
310 ip->i_df.if_ext_max =
311 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
312 error = 0;
313
314 if (unlikely(be32_to_cpu(dip->di_nextents) +
315 be16_to_cpu(dip->di_anextents) >
316 be64_to_cpu(dip->di_nblocks))) {
317 xfs_warn(ip->i_mount,
318 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
319 (unsigned long long)ip->i_ino,
320 (int)(be32_to_cpu(dip->di_nextents) +
321 be16_to_cpu(dip->di_anextents)),
322 (unsigned long long)
323 be64_to_cpu(dip->di_nblocks));
324 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
325 ip->i_mount, dip);
326 return XFS_ERROR(EFSCORRUPTED);
327 }
328
329 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
330 xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
331 (unsigned long long)ip->i_ino,
332 dip->di_forkoff);
333 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
334 ip->i_mount, dip);
335 return XFS_ERROR(EFSCORRUPTED);
336 }
337
338 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
339 !ip->i_mount->m_rtdev_targp)) {
340 xfs_warn(ip->i_mount,
341 "corrupt dinode %Lu, has realtime flag set.",
342 ip->i_ino);
343 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
344 XFS_ERRLEVEL_LOW, ip->i_mount, dip);
345 return XFS_ERROR(EFSCORRUPTED);
346 }
347
348 switch (ip->i_d.di_mode & S_IFMT) {
349 case S_IFIFO:
350 case S_IFCHR:
351 case S_IFBLK:
352 case S_IFSOCK:
353 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
354 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
355 ip->i_mount, dip);
356 return XFS_ERROR(EFSCORRUPTED);
357 }
358 ip->i_d.di_size = 0;
359 ip->i_size = 0;
360 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
361 break;
362
363 case S_IFREG:
364 case S_IFLNK:
365 case S_IFDIR:
366 switch (dip->di_format) {
367 case XFS_DINODE_FMT_LOCAL:
368 /*
369 * no local regular files yet
370 */
371 if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) {
372 xfs_warn(ip->i_mount,
373 "corrupt inode %Lu (local format for regular file).",
374 (unsigned long long) ip->i_ino);
375 XFS_CORRUPTION_ERROR("xfs_iformat(4)",
376 XFS_ERRLEVEL_LOW,
377 ip->i_mount, dip);
378 return XFS_ERROR(EFSCORRUPTED);
379 }
380
381 di_size = be64_to_cpu(dip->di_size);
382 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
383 xfs_warn(ip->i_mount,
384 "corrupt inode %Lu (bad size %Ld for local inode).",
385 (unsigned long long) ip->i_ino,
386 (long long) di_size);
387 XFS_CORRUPTION_ERROR("xfs_iformat(5)",
388 XFS_ERRLEVEL_LOW,
389 ip->i_mount, dip);
390 return XFS_ERROR(EFSCORRUPTED);
391 }
392
393 size = (int)di_size;
394 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
395 break;
396 case XFS_DINODE_FMT_EXTENTS:
397 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
398 break;
399 case XFS_DINODE_FMT_BTREE:
400 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
401 break;
402 default:
403 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
404 ip->i_mount);
405 return XFS_ERROR(EFSCORRUPTED);
406 }
407 break;
408
409 default:
410 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
411 return XFS_ERROR(EFSCORRUPTED);
412 }
413 if (error) {
414 return error;
415 }
416 if (!XFS_DFORK_Q(dip))
417 return 0;
418 ASSERT(ip->i_afp == NULL);
419 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
420 ip->i_afp->if_ext_max =
421 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
422 switch (dip->di_aformat) {
423 case XFS_DINODE_FMT_LOCAL:
424 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
425 size = be16_to_cpu(atp->hdr.totsize);
426
427 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
428 xfs_warn(ip->i_mount,
429 "corrupt inode %Lu (bad attr fork size %Ld).",
430 (unsigned long long) ip->i_ino,
431 (long long) size);
432 XFS_CORRUPTION_ERROR("xfs_iformat(8)",
433 XFS_ERRLEVEL_LOW,
434 ip->i_mount, dip);
435 return XFS_ERROR(EFSCORRUPTED);
436 }
437
438 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
439 break;
440 case XFS_DINODE_FMT_EXTENTS:
441 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
442 break;
443 case XFS_DINODE_FMT_BTREE:
444 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
445 break;
446 default:
447 error = XFS_ERROR(EFSCORRUPTED);
448 break;
449 }
450 if (error) {
451 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
452 ip->i_afp = NULL;
453 xfs_idestroy_fork(ip, XFS_DATA_FORK);
454 }
455 return error;
456}
457
458/*
459 * The file is in-lined in the on-disk inode.
460 * If it fits into if_inline_data, then copy
461 * it there, otherwise allocate a buffer for it
462 * and copy the data there. Either way, set
463 * if_data to point at the data.
464 * If we allocate a buffer for the data, make
465 * sure that its size is a multiple of 4 and
466 * record the real size in i_real_bytes.
467 */
468STATIC int
469xfs_iformat_local(
470 xfs_inode_t *ip,
471 xfs_dinode_t *dip,
472 int whichfork,
473 int size)
474{
475 xfs_ifork_t *ifp;
476 int real_size;
477
478 /*
479 * If the size is unreasonable, then something
480 * is wrong and we just bail out rather than crash in
481 * kmem_alloc() or memcpy() below.
482 */
483 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
484 xfs_warn(ip->i_mount,
485 "corrupt inode %Lu (bad size %d for local fork, size = %d).",
486 (unsigned long long) ip->i_ino, size,
487 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
488 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
489 ip->i_mount, dip);
490 return XFS_ERROR(EFSCORRUPTED);
491 }
492 ifp = XFS_IFORK_PTR(ip, whichfork);
493 real_size = 0;
494 if (size == 0)
495 ifp->if_u1.if_data = NULL;
496 else if (size <= sizeof(ifp->if_u2.if_inline_data))
497 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
498 else {
499 real_size = roundup(size, 4);
500 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
501 }
502 ifp->if_bytes = size;
503 ifp->if_real_bytes = real_size;
504 if (size)
505 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
506 ifp->if_flags &= ~XFS_IFEXTENTS;
507 ifp->if_flags |= XFS_IFINLINE;
508 return 0;
509}
510
511/*
512 * The file consists of a set of extents all
513 * of which fit into the on-disk inode.
514 * If there are few enough extents to fit into
515 * the if_inline_ext, then copy them there.
516 * Otherwise allocate a buffer for them and copy
517 * them into it. Either way, set if_extents
518 * to point at the extents.
519 */
520STATIC int
521xfs_iformat_extents(
522 xfs_inode_t *ip,
523 xfs_dinode_t *dip,
524 int whichfork)
525{
526 xfs_bmbt_rec_t *dp;
527 xfs_ifork_t *ifp;
528 int nex;
529 int size;
530 int i;
531
532 ifp = XFS_IFORK_PTR(ip, whichfork);
533 nex = XFS_DFORK_NEXTENTS(dip, whichfork);
534 size = nex * (uint)sizeof(xfs_bmbt_rec_t);
535
536 /*
537 * If the number of extents is unreasonable, then something
538 * is wrong and we just bail out rather than crash in
539 * kmem_alloc() or memcpy() below.
540 */
541 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
542 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
543 (unsigned long long) ip->i_ino, nex);
544 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
545 ip->i_mount, dip);
546 return XFS_ERROR(EFSCORRUPTED);
547 }
548
549 ifp->if_real_bytes = 0;
550 if (nex == 0)
551 ifp->if_u1.if_extents = NULL;
552 else if (nex <= XFS_INLINE_EXTS)
553 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
554 else
555 xfs_iext_add(ifp, 0, nex);
556
557 ifp->if_bytes = size;
558 if (size) {
559 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
560 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
561 for (i = 0; i < nex; i++, dp++) {
562 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
563 ep->l0 = get_unaligned_be64(&dp->l0);
564 ep->l1 = get_unaligned_be64(&dp->l1);
565 }
566 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
567 if (whichfork != XFS_DATA_FORK ||
568 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
569 if (unlikely(xfs_check_nostate_extents(
570 ifp, 0, nex))) {
571 XFS_ERROR_REPORT("xfs_iformat_extents(2)",
572 XFS_ERRLEVEL_LOW,
573 ip->i_mount);
574 return XFS_ERROR(EFSCORRUPTED);
575 }
576 }
577 ifp->if_flags |= XFS_IFEXTENTS;
578 return 0;
579}
580
581/*
582 * The file has too many extents to fit into
583 * the inode, so they are in B-tree format.
584 * Allocate a buffer for the root of the B-tree
585 * and copy the root into it. The i_extents
586 * field will remain NULL until all of the
587 * extents are read in (when they are needed).
588 */
589STATIC int
590xfs_iformat_btree(
591 xfs_inode_t *ip,
592 xfs_dinode_t *dip,
593 int whichfork)
594{
595 xfs_bmdr_block_t *dfp;
596 xfs_ifork_t *ifp;
597 /* REFERENCED */
598 int nrecs;
599 int size;
600
601 ifp = XFS_IFORK_PTR(ip, whichfork);
602 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
603 size = XFS_BMAP_BROOT_SPACE(dfp);
604 nrecs = be16_to_cpu(dfp->bb_numrecs);
605
606 /*
607 * blow out if -- fork has less extents than can fit in
608 * fork (fork shouldn't be a btree format), root btree
609 * block has more records than can fit into the fork,
610 * or the number of extents is greater than the number of
611 * blocks.
612 */
613 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max
614 || XFS_BMDR_SPACE_CALC(nrecs) >
615 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)
616 || XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
617 xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).",
618 (unsigned long long) ip->i_ino);
619 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
620 ip->i_mount, dip);
621 return XFS_ERROR(EFSCORRUPTED);
622 }
623
624 ifp->if_broot_bytes = size;
625 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
626 ASSERT(ifp->if_broot != NULL);
627 /*
628 * Copy and convert from the on-disk structure
629 * to the in-memory structure.
630 */
631 xfs_bmdr_to_bmbt(ip->i_mount, dfp,
632 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
633 ifp->if_broot, size);
634 ifp->if_flags &= ~XFS_IFEXTENTS;
635 ifp->if_flags |= XFS_IFBROOT;
636
637 return 0;
638}
639
640STATIC void
641xfs_dinode_from_disk(
642 xfs_icdinode_t *to,
643 xfs_dinode_t *from)
644{
645 to->di_magic = be16_to_cpu(from->di_magic);
646 to->di_mode = be16_to_cpu(from->di_mode);
647 to->di_version = from ->di_version;
648 to->di_format = from->di_format;
649 to->di_onlink = be16_to_cpu(from->di_onlink);
650 to->di_uid = be32_to_cpu(from->di_uid);
651 to->di_gid = be32_to_cpu(from->di_gid);
652 to->di_nlink = be32_to_cpu(from->di_nlink);
653 to->di_projid_lo = be16_to_cpu(from->di_projid_lo);
654 to->di_projid_hi = be16_to_cpu(from->di_projid_hi);
655 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
656 to->di_flushiter = be16_to_cpu(from->di_flushiter);
657 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec);
658 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec);
659 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec);
660 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec);
661 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec);
662 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec);
663 to->di_size = be64_to_cpu(from->di_size);
664 to->di_nblocks = be64_to_cpu(from->di_nblocks);
665 to->di_extsize = be32_to_cpu(from->di_extsize);
666 to->di_nextents = be32_to_cpu(from->di_nextents);
667 to->di_anextents = be16_to_cpu(from->di_anextents);
668 to->di_forkoff = from->di_forkoff;
669 to->di_aformat = from->di_aformat;
670 to->di_dmevmask = be32_to_cpu(from->di_dmevmask);
671 to->di_dmstate = be16_to_cpu(from->di_dmstate);
672 to->di_flags = be16_to_cpu(from->di_flags);
673 to->di_gen = be32_to_cpu(from->di_gen);
674}
675
676void
677xfs_dinode_to_disk(
678 xfs_dinode_t *to,
679 xfs_icdinode_t *from)
680{
681 to->di_magic = cpu_to_be16(from->di_magic);
682 to->di_mode = cpu_to_be16(from->di_mode);
683 to->di_version = from ->di_version;
684 to->di_format = from->di_format;
685 to->di_onlink = cpu_to_be16(from->di_onlink);
686 to->di_uid = cpu_to_be32(from->di_uid);
687 to->di_gid = cpu_to_be32(from->di_gid);
688 to->di_nlink = cpu_to_be32(from->di_nlink);
689 to->di_projid_lo = cpu_to_be16(from->di_projid_lo);
690 to->di_projid_hi = cpu_to_be16(from->di_projid_hi);
691 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad));
692 to->di_flushiter = cpu_to_be16(from->di_flushiter);
693 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec);
694 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec);
695 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec);
696 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec);
697 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec);
698 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec);
699 to->di_size = cpu_to_be64(from->di_size);
700 to->di_nblocks = cpu_to_be64(from->di_nblocks);
701 to->di_extsize = cpu_to_be32(from->di_extsize);
702 to->di_nextents = cpu_to_be32(from->di_nextents);
703 to->di_anextents = cpu_to_be16(from->di_anextents);
704 to->di_forkoff = from->di_forkoff;
705 to->di_aformat = from->di_aformat;
706 to->di_dmevmask = cpu_to_be32(from->di_dmevmask);
707 to->di_dmstate = cpu_to_be16(from->di_dmstate);
708 to->di_flags = cpu_to_be16(from->di_flags);
709 to->di_gen = cpu_to_be32(from->di_gen);
710}
711
712STATIC uint
713_xfs_dic2xflags(
714 __uint16_t di_flags)
715{
716 uint flags = 0;
717
718 if (di_flags & XFS_DIFLAG_ANY) {
719 if (di_flags & XFS_DIFLAG_REALTIME)
720 flags |= XFS_XFLAG_REALTIME;
721 if (di_flags & XFS_DIFLAG_PREALLOC)
722 flags |= XFS_XFLAG_PREALLOC;
723 if (di_flags & XFS_DIFLAG_IMMUTABLE)
724 flags |= XFS_XFLAG_IMMUTABLE;
725 if (di_flags & XFS_DIFLAG_APPEND)
726 flags |= XFS_XFLAG_APPEND;
727 if (di_flags & XFS_DIFLAG_SYNC)
728 flags |= XFS_XFLAG_SYNC;
729 if (di_flags & XFS_DIFLAG_NOATIME)
730 flags |= XFS_XFLAG_NOATIME;
731 if (di_flags & XFS_DIFLAG_NODUMP)
732 flags |= XFS_XFLAG_NODUMP;
733 if (di_flags & XFS_DIFLAG_RTINHERIT)
734 flags |= XFS_XFLAG_RTINHERIT;
735 if (di_flags & XFS_DIFLAG_PROJINHERIT)
736 flags |= XFS_XFLAG_PROJINHERIT;
737 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
738 flags |= XFS_XFLAG_NOSYMLINKS;
739 if (di_flags & XFS_DIFLAG_EXTSIZE)
740 flags |= XFS_XFLAG_EXTSIZE;
741 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
742 flags |= XFS_XFLAG_EXTSZINHERIT;
743 if (di_flags & XFS_DIFLAG_NODEFRAG)
744 flags |= XFS_XFLAG_NODEFRAG;
745 if (di_flags & XFS_DIFLAG_FILESTREAM)
746 flags |= XFS_XFLAG_FILESTREAM;
747 }
748
749 return flags;
750}
751
752uint
753xfs_ip2xflags(
754 xfs_inode_t *ip)
755{
756 xfs_icdinode_t *dic = &ip->i_d;
757
758 return _xfs_dic2xflags(dic->di_flags) |
759 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
760}
761
762uint
763xfs_dic2xflags(
764 xfs_dinode_t *dip)
765{
766 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
767 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
768}
769
770/*
771 * Read the disk inode attributes into the in-core inode structure.
772 */
773int
774xfs_iread(
775 xfs_mount_t *mp,
776 xfs_trans_t *tp,
777 xfs_inode_t *ip,
778 uint iget_flags)
779{
780 xfs_buf_t *bp;
781 xfs_dinode_t *dip;
782 int error;
783
784 /*
785 * Fill in the location information in the in-core inode.
786 */
787 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
788 if (error)
789 return error;
790
791 /*
792 * Get pointers to the on-disk inode and the buffer containing it.
793 */
794 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
795 XBF_LOCK, iget_flags);
796 if (error)
797 return error;
798 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
799
800 /*
801 * If we got something that isn't an inode it means someone
802 * (nfs or dmi) has a stale handle.
803 */
804 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) {
805#ifdef DEBUG
806 xfs_alert(mp,
807 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)",
808 __func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC);
809#endif /* DEBUG */
810 error = XFS_ERROR(EINVAL);
811 goto out_brelse;
812 }
813
814 /*
815 * If the on-disk inode is already linked to a directory
816 * entry, copy all of the inode into the in-core inode.
817 * xfs_iformat() handles copying in the inode format
818 * specific information.
819 * Otherwise, just get the truly permanent information.
820 */
821 if (dip->di_mode) {
822 xfs_dinode_from_disk(&ip->i_d, dip);
823 error = xfs_iformat(ip, dip);
824 if (error) {
825#ifdef DEBUG
826 xfs_alert(mp, "%s: xfs_iformat() returned error %d",
827 __func__, error);
828#endif /* DEBUG */
829 goto out_brelse;
830 }
831 } else {
832 ip->i_d.di_magic = be16_to_cpu(dip->di_magic);
833 ip->i_d.di_version = dip->di_version;
834 ip->i_d.di_gen = be32_to_cpu(dip->di_gen);
835 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter);
836 /*
837 * Make sure to pull in the mode here as well in
838 * case the inode is released without being used.
839 * This ensures that xfs_inactive() will see that
840 * the inode is already free and not try to mess
841 * with the uninitialized part of it.
842 */
843 ip->i_d.di_mode = 0;
844 /*
845 * Initialize the per-fork minima and maxima for a new
846 * inode here. xfs_iformat will do it for old inodes.
847 */
848 ip->i_df.if_ext_max =
849 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
850 }
851
852 /*
853 * The inode format changed when we moved the link count and
854 * made it 32 bits long. If this is an old format inode,
855 * convert it in memory to look like a new one. If it gets
856 * flushed to disk we will convert back before flushing or
857 * logging it. We zero out the new projid field and the old link
858 * count field. We'll handle clearing the pad field (the remains
859 * of the old uuid field) when we actually convert the inode to
860 * the new format. We don't change the version number so that we
861 * can distinguish this from a real new format inode.
862 */
863 if (ip->i_d.di_version == 1) {
864 ip->i_d.di_nlink = ip->i_d.di_onlink;
865 ip->i_d.di_onlink = 0;
866 xfs_set_projid(ip, 0);
867 }
868
869 ip->i_delayed_blks = 0;
870 ip->i_size = ip->i_d.di_size;
871
872 /*
873 * Mark the buffer containing the inode as something to keep
874 * around for a while. This helps to keep recently accessed
875 * meta-data in-core longer.
876 */
877 xfs_buf_set_ref(bp, XFS_INO_REF);
878
879 /*
880 * Use xfs_trans_brelse() to release the buffer containing the
881 * on-disk inode, because it was acquired with xfs_trans_read_buf()
882 * in xfs_itobp() above. If tp is NULL, this is just a normal
883 * brelse(). If we're within a transaction, then xfs_trans_brelse()
884 * will only release the buffer if it is not dirty within the
885 * transaction. It will be OK to release the buffer in this case,
886 * because inodes on disk are never destroyed and we will be
887 * locking the new in-core inode before putting it in the hash
888 * table where other processes can find it. Thus we don't have
889 * to worry about the inode being changed just because we released
890 * the buffer.
891 */
892 out_brelse:
893 xfs_trans_brelse(tp, bp);
894 return error;
895}
896
897/*
898 * Read in extents from a btree-format inode.
899 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c.
900 */
901int
902xfs_iread_extents(
903 xfs_trans_t *tp,
904 xfs_inode_t *ip,
905 int whichfork)
906{
907 int error;
908 xfs_ifork_t *ifp;
909 xfs_extnum_t nextents;
910
911 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
912 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
913 ip->i_mount);
914 return XFS_ERROR(EFSCORRUPTED);
915 }
916 nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
917 ifp = XFS_IFORK_PTR(ip, whichfork);
918
919 /*
920 * We know that the size is valid (it's checked in iformat_btree)
921 */
922 ifp->if_bytes = ifp->if_real_bytes = 0;
923 ifp->if_flags |= XFS_IFEXTENTS;
924 xfs_iext_add(ifp, 0, nextents);
925 error = xfs_bmap_read_extents(tp, ip, whichfork);
926 if (error) {
927 xfs_iext_destroy(ifp);
928 ifp->if_flags &= ~XFS_IFEXTENTS;
929 return error;
930 }
931 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
932 return 0;
933}
934
935/*
936 * Allocate an inode on disk and return a copy of its in-core version.
937 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
938 * appropriately within the inode. The uid and gid for the inode are
939 * set according to the contents of the given cred structure.
940 *
941 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
942 * has a free inode available, call xfs_iget()
943 * to obtain the in-core version of the allocated inode. Finally,
944 * fill in the inode and log its initial contents. In this case,
945 * ialloc_context would be set to NULL and call_again set to false.
946 *
947 * If xfs_dialloc() does not have an available inode,
948 * it will replenish its supply by doing an allocation. Since we can
949 * only do one allocation within a transaction without deadlocks, we
950 * must commit the current transaction before returning the inode itself.
951 * In this case, therefore, we will set call_again to true and return.
952 * The caller should then commit the current transaction, start a new
953 * transaction, and call xfs_ialloc() again to actually get the inode.
954 *
955 * To ensure that some other process does not grab the inode that
956 * was allocated during the first call to xfs_ialloc(), this routine
957 * also returns the [locked] bp pointing to the head of the freelist
958 * as ialloc_context. The caller should hold this buffer across
959 * the commit and pass it back into this routine on the second call.
960 *
961 * If we are allocating quota inodes, we do not have a parent inode
962 * to attach to or associate with (i.e. pip == NULL) because they
963 * are not linked into the directory structure - they are attached
964 * directly to the superblock - and so have no parent.
965 */
966int
967xfs_ialloc(
968 xfs_trans_t *tp,
969 xfs_inode_t *pip,
970 mode_t mode,
971 xfs_nlink_t nlink,
972 xfs_dev_t rdev,
973 prid_t prid,
974 int okalloc,
975 xfs_buf_t **ialloc_context,
976 boolean_t *call_again,
977 xfs_inode_t **ipp)
978{
979 xfs_ino_t ino;
980 xfs_inode_t *ip;
981 uint flags;
982 int error;
983 timespec_t tv;
984 int filestreams = 0;
985
986 /*
987 * Call the space management code to pick
988 * the on-disk inode to be allocated.
989 */
990 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
991 ialloc_context, call_again, &ino);
992 if (error)
993 return error;
994 if (*call_again || ino == NULLFSINO) {
995 *ipp = NULL;
996 return 0;
997 }
998 ASSERT(*ialloc_context == NULL);
999
1000 /*
1001 * Get the in-core inode with the lock held exclusively.
1002 * This is because we're setting fields here we need
1003 * to prevent others from looking at until we're done.
1004 */
1005 error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE,
1006 XFS_ILOCK_EXCL, &ip);
1007 if (error)
1008 return error;
1009 ASSERT(ip != NULL);
1010
1011 ip->i_d.di_mode = (__uint16_t)mode;
1012 ip->i_d.di_onlink = 0;
1013 ip->i_d.di_nlink = nlink;
1014 ASSERT(ip->i_d.di_nlink == nlink);
1015 ip->i_d.di_uid = current_fsuid();
1016 ip->i_d.di_gid = current_fsgid();
1017 xfs_set_projid(ip, prid);
1018 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1019
1020 /*
1021 * If the superblock version is up to where we support new format
1022 * inodes and this is currently an old format inode, then change
1023 * the inode version number now. This way we only do the conversion
1024 * here rather than here and in the flush/logging code.
1025 */
1026 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) &&
1027 ip->i_d.di_version == 1) {
1028 ip->i_d.di_version = 2;
1029 /*
1030 * We've already zeroed the old link count, the projid field,
1031 * and the pad field.
1032 */
1033 }
1034
1035 /*
1036 * Project ids won't be stored on disk if we are using a version 1 inode.
1037 */
1038 if ((prid != 0) && (ip->i_d.di_version == 1))
1039 xfs_bump_ino_vers2(tp, ip);
1040
1041 if (pip && XFS_INHERIT_GID(pip)) {
1042 ip->i_d.di_gid = pip->i_d.di_gid;
1043 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
1044 ip->i_d.di_mode |= S_ISGID;
1045 }
1046 }
1047
1048 /*
1049 * If the group ID of the new file does not match the effective group
1050 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
1051 * (and only if the irix_sgid_inherit compatibility variable is set).
1052 */
1053 if ((irix_sgid_inherit) &&
1054 (ip->i_d.di_mode & S_ISGID) &&
1055 (!in_group_p((gid_t)ip->i_d.di_gid))) {
1056 ip->i_d.di_mode &= ~S_ISGID;
1057 }
1058
1059 ip->i_d.di_size = 0;
1060 ip->i_size = 0;
1061 ip->i_d.di_nextents = 0;
1062 ASSERT(ip->i_d.di_nblocks == 0);
1063
1064 nanotime(&tv);
1065 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
1066 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
1067 ip->i_d.di_atime = ip->i_d.di_mtime;
1068 ip->i_d.di_ctime = ip->i_d.di_mtime;
1069
1070 /*
1071 * di_gen will have been taken care of in xfs_iread.
1072 */
1073 ip->i_d.di_extsize = 0;
1074 ip->i_d.di_dmevmask = 0;
1075 ip->i_d.di_dmstate = 0;
1076 ip->i_d.di_flags = 0;
1077 flags = XFS_ILOG_CORE;
1078 switch (mode & S_IFMT) {
1079 case S_IFIFO:
1080 case S_IFCHR:
1081 case S_IFBLK:
1082 case S_IFSOCK:
1083 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
1084 ip->i_df.if_u2.if_rdev = rdev;
1085 ip->i_df.if_flags = 0;
1086 flags |= XFS_ILOG_DEV;
1087 break;
1088 case S_IFREG:
1089 /*
1090 * we can't set up filestreams until after the VFS inode
1091 * is set up properly.
1092 */
1093 if (pip && xfs_inode_is_filestream(pip))
1094 filestreams = 1;
1095 /* fall through */
1096 case S_IFDIR:
1097 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
1098 uint di_flags = 0;
1099
1100 if (S_ISDIR(mode)) {
1101 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1102 di_flags |= XFS_DIFLAG_RTINHERIT;
1103 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1104 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
1105 ip->i_d.di_extsize = pip->i_d.di_extsize;
1106 }
1107 } else if (S_ISREG(mode)) {
1108 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
1109 di_flags |= XFS_DIFLAG_REALTIME;
1110 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
1111 di_flags |= XFS_DIFLAG_EXTSIZE;
1112 ip->i_d.di_extsize = pip->i_d.di_extsize;
1113 }
1114 }
1115 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
1116 xfs_inherit_noatime)
1117 di_flags |= XFS_DIFLAG_NOATIME;
1118 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
1119 xfs_inherit_nodump)
1120 di_flags |= XFS_DIFLAG_NODUMP;
1121 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
1122 xfs_inherit_sync)
1123 di_flags |= XFS_DIFLAG_SYNC;
1124 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
1125 xfs_inherit_nosymlinks)
1126 di_flags |= XFS_DIFLAG_NOSYMLINKS;
1127 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1128 di_flags |= XFS_DIFLAG_PROJINHERIT;
1129 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
1130 xfs_inherit_nodefrag)
1131 di_flags |= XFS_DIFLAG_NODEFRAG;
1132 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
1133 di_flags |= XFS_DIFLAG_FILESTREAM;
1134 ip->i_d.di_flags |= di_flags;
1135 }
1136 /* FALLTHROUGH */
1137 case S_IFLNK:
1138 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1139 ip->i_df.if_flags = XFS_IFEXTENTS;
1140 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
1141 ip->i_df.if_u1.if_extents = NULL;
1142 break;
1143 default:
1144 ASSERT(0);
1145 }
1146 /*
1147 * Attribute fork settings for new inode.
1148 */
1149 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1150 ip->i_d.di_anextents = 0;
1151
1152 /*
1153 * Log the new values stuffed into the inode.
1154 */
1155 xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
1156 xfs_trans_log_inode(tp, ip, flags);
1157
1158 /* now that we have an i_mode we can setup inode ops and unlock */
1159 xfs_setup_inode(ip);
1160
1161 /* now we have set up the vfs inode we can associate the filestream */
1162 if (filestreams) {
1163 error = xfs_filestream_associate(pip, ip);
1164 if (error < 0)
1165 return -error;
1166 if (!error)
1167 xfs_iflags_set(ip, XFS_IFILESTREAM);
1168 }
1169
1170 *ipp = ip;
1171 return 0;
1172}
1173
1174/*
1175 * Check to make sure that there are no blocks allocated to the
1176 * file beyond the size of the file. We don't check this for
1177 * files with fixed size extents or real time extents, but we
1178 * at least do it for regular files.
1179 */
1180#ifdef DEBUG
1181STATIC void
1182xfs_isize_check(
1183 struct xfs_inode *ip,
1184 xfs_fsize_t isize)
1185{
1186 struct xfs_mount *mp = ip->i_mount;
1187 xfs_fileoff_t map_first;
1188 int nimaps;
1189 xfs_bmbt_irec_t imaps[2];
1190
1191 if (!S_ISREG(ip->i_d.di_mode))
1192 return;
1193
1194 if (XFS_IS_REALTIME_INODE(ip))
1195 return;
1196
1197 if (ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE)
1198 return;
1199
1200 nimaps = 2;
1201 map_first = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
1202 /*
1203 * The filesystem could be shutting down, so bmapi may return
1204 * an error.
1205 */
1206 if (xfs_bmapi(NULL, ip, map_first,
1207 (XFS_B_TO_FSB(mp,
1208 (xfs_ufsize_t)XFS_MAXIOFFSET(mp)) -
1209 map_first),
1210 XFS_BMAPI_ENTIRE, NULL, 0, imaps, &nimaps,
1211 NULL))
1212 return;
1213 ASSERT(nimaps == 1);
1214 ASSERT(imaps[0].br_startblock == HOLESTARTBLOCK);
1215}
1216#else /* DEBUG */
1217#define xfs_isize_check(ip, isize)
1218#endif /* DEBUG */
1219
1220/*
1221 * Free up the underlying blocks past new_size. The new size must be smaller
1222 * than the current size. This routine can be used both for the attribute and
1223 * data fork, and does not modify the inode size, which is left to the caller.
1224 *
1225 * The transaction passed to this routine must have made a permanent log
1226 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1227 * given transaction and start new ones, so make sure everything involved in
1228 * the transaction is tidy before calling here. Some transaction will be
1229 * returned to the caller to be committed. The incoming transaction must
1230 * already include the inode, and both inode locks must be held exclusively.
1231 * The inode must also be "held" within the transaction. On return the inode
1232 * will be "held" within the returned transaction. This routine does NOT
1233 * require any disk space to be reserved for it within the transaction.
1234 *
1235 * If we get an error, we must return with the inode locked and linked into the
1236 * current transaction. This keeps things simple for the higher level code,
1237 * because it always knows that the inode is locked and held in the transaction
1238 * that returns to it whether errors occur or not. We don't mark the inode
1239 * dirty on error so that transactions can be easily aborted if possible.
1240 */
1241int
1242xfs_itruncate_extents(
1243 struct xfs_trans **tpp,
1244 struct xfs_inode *ip,
1245 int whichfork,
1246 xfs_fsize_t new_size)
1247{
1248 struct xfs_mount *mp = ip->i_mount;
1249 struct xfs_trans *tp = *tpp;
1250 struct xfs_trans *ntp;
1251 xfs_bmap_free_t free_list;
1252 xfs_fsblock_t first_block;
1253 xfs_fileoff_t first_unmap_block;
1254 xfs_fileoff_t last_block;
1255 xfs_filblks_t unmap_len;
1256 int committed;
1257 int error = 0;
1258 int done = 0;
1259
1260 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
1261 ASSERT(new_size <= ip->i_size);
1262 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1263 ASSERT(ip->i_itemp != NULL);
1264 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1265 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1266
1267 /*
1268 * Since it is possible for space to become allocated beyond
1269 * the end of the file (in a crash where the space is allocated
1270 * but the inode size is not yet updated), simply remove any
1271 * blocks which show up between the new EOF and the maximum
1272 * possible file size. If the first block to be removed is
1273 * beyond the maximum file size (ie it is the same as last_block),
1274 * then there is nothing to do.
1275 */
1276 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1277 last_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1278 if (first_unmap_block == last_block)
1279 return 0;
1280
1281 ASSERT(first_unmap_block < last_block);
1282 unmap_len = last_block - first_unmap_block + 1;
1283 while (!done) {
1284 xfs_bmap_init(&free_list, &first_block);
1285 error = xfs_bunmapi(tp, ip,
1286 first_unmap_block, unmap_len,
1287 xfs_bmapi_aflag(whichfork),
1288 XFS_ITRUNC_MAX_EXTENTS,
1289 &first_block, &free_list,
1290 &done);
1291 if (error)
1292 goto out_bmap_cancel;
1293
1294 /*
1295 * Duplicate the transaction that has the permanent
1296 * reservation and commit the old transaction.
1297 */
1298 error = xfs_bmap_finish(&tp, &free_list, &committed);
1299 if (committed)
1300 xfs_trans_ijoin(tp, ip);
1301 if (error)
1302 goto out_bmap_cancel;
1303
1304 if (committed) {
1305 /*
1306 * Mark the inode dirty so it will be logged and
1307 * moved forward in the log as part of every commit.
1308 */
1309 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1310 }
1311
1312 ntp = xfs_trans_dup(tp);
1313 error = xfs_trans_commit(tp, 0);
1314 tp = ntp;
1315
1316 xfs_trans_ijoin(tp, ip);
1317
1318 if (error)
1319 goto out;
1320
1321 /*
1322 * Transaction commit worked ok so we can drop the extra ticket
1323 * reference that we gained in xfs_trans_dup()
1324 */
1325 xfs_log_ticket_put(tp->t_ticket);
1326 error = xfs_trans_reserve(tp, 0,
1327 XFS_ITRUNCATE_LOG_RES(mp), 0,
1328 XFS_TRANS_PERM_LOG_RES,
1329 XFS_ITRUNCATE_LOG_COUNT);
1330 if (error)
1331 goto out;
1332 }
1333
1334out:
1335 *tpp = tp;
1336 return error;
1337out_bmap_cancel:
1338 /*
1339 * If the bunmapi call encounters an error, return to the caller where
1340 * the transaction can be properly aborted. We just need to make sure
1341 * we're not holding any resources that we were not when we came in.
1342 */
1343 xfs_bmap_cancel(&free_list);
1344 goto out;
1345}
1346
1347int
1348xfs_itruncate_data(
1349 struct xfs_trans **tpp,
1350 struct xfs_inode *ip,
1351 xfs_fsize_t new_size)
1352{
1353 int error;
1354
1355 trace_xfs_itruncate_data_start(ip, new_size);
1356
1357 /*
1358 * The first thing we do is set the size to new_size permanently on
1359 * disk. This way we don't have to worry about anyone ever being able
1360 * to look at the data being freed even in the face of a crash.
1361 * What we're getting around here is the case where we free a block, it
1362 * is allocated to another file, it is written to, and then we crash.
1363 * If the new data gets written to the file but the log buffers
1364 * containing the free and reallocation don't, then we'd end up with
1365 * garbage in the blocks being freed. As long as we make the new_size
1366 * permanent before actually freeing any blocks it doesn't matter if
1367 * they get written to.
1368 */
1369 if (ip->i_d.di_nextents > 0) {
1370 /*
1371 * If we are not changing the file size then do not update
1372 * the on-disk file size - we may be called from
1373 * xfs_inactive_free_eofblocks(). If we update the on-disk
1374 * file size and then the system crashes before the contents
1375 * of the file are flushed to disk then the files may be
1376 * full of holes (ie NULL files bug).
1377 */
1378 if (ip->i_size != new_size) {
1379 ip->i_d.di_size = new_size;
1380 ip->i_size = new_size;
1381 xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1382 }
1383 }
1384
1385 error = xfs_itruncate_extents(tpp, ip, XFS_DATA_FORK, new_size);
1386 if (error)
1387 return error;
1388
1389 /*
1390 * If we are not changing the file size then do not update the on-disk
1391 * file size - we may be called from xfs_inactive_free_eofblocks().
1392 * If we update the on-disk file size and then the system crashes
1393 * before the contents of the file are flushed to disk then the files
1394 * may be full of holes (ie NULL files bug).
1395 */
1396 xfs_isize_check(ip, new_size);
1397 if (ip->i_size != new_size) {
1398 ip->i_d.di_size = new_size;
1399 ip->i_size = new_size;
1400 }
1401
1402 ASSERT(new_size != 0 || ip->i_delayed_blks == 0);
1403 ASSERT(new_size != 0 || ip->i_d.di_nextents == 0);
1404
1405 /*
1406 * Always re-log the inode so that our permanent transaction can keep
1407 * on rolling it forward in the log.
1408 */
1409 xfs_trans_log_inode(*tpp, ip, XFS_ILOG_CORE);
1410
1411 trace_xfs_itruncate_data_end(ip, new_size);
1412 return 0;
1413}
1414
1415/*
1416 * This is called when the inode's link count goes to 0.
1417 * We place the on-disk inode on a list in the AGI. It
1418 * will be pulled from this list when the inode is freed.
1419 */
1420int
1421xfs_iunlink(
1422 xfs_trans_t *tp,
1423 xfs_inode_t *ip)
1424{
1425 xfs_mount_t *mp;
1426 xfs_agi_t *agi;
1427 xfs_dinode_t *dip;
1428 xfs_buf_t *agibp;
1429 xfs_buf_t *ibp;
1430 xfs_agino_t agino;
1431 short bucket_index;
1432 int offset;
1433 int error;
1434
1435 ASSERT(ip->i_d.di_nlink == 0);
1436 ASSERT(ip->i_d.di_mode != 0);
1437
1438 mp = tp->t_mountp;
1439
1440 /*
1441 * Get the agi buffer first. It ensures lock ordering
1442 * on the list.
1443 */
1444 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1445 if (error)
1446 return error;
1447 agi = XFS_BUF_TO_AGI(agibp);
1448
1449 /*
1450 * Get the index into the agi hash table for the
1451 * list this inode will go on.
1452 */
1453 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1454 ASSERT(agino != 0);
1455 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1456 ASSERT(agi->agi_unlinked[bucket_index]);
1457 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1458
1459 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1460 /*
1461 * There is already another inode in the bucket we need
1462 * to add ourselves to. Add us at the front of the list.
1463 * Here we put the head pointer into our next pointer,
1464 * and then we fall through to point the head at us.
1465 */
1466 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1467 if (error)
1468 return error;
1469
1470 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1471 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1472 offset = ip->i_imap.im_boffset +
1473 offsetof(xfs_dinode_t, di_next_unlinked);
1474 xfs_trans_inode_buf(tp, ibp);
1475 xfs_trans_log_buf(tp, ibp, offset,
1476 (offset + sizeof(xfs_agino_t) - 1));
1477 xfs_inobp_check(mp, ibp);
1478 }
1479
1480 /*
1481 * Point the bucket head pointer at the inode being inserted.
1482 */
1483 ASSERT(agino != 0);
1484 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1485 offset = offsetof(xfs_agi_t, agi_unlinked) +
1486 (sizeof(xfs_agino_t) * bucket_index);
1487 xfs_trans_log_buf(tp, agibp, offset,
1488 (offset + sizeof(xfs_agino_t) - 1));
1489 return 0;
1490}
1491
1492/*
1493 * Pull the on-disk inode from the AGI unlinked list.
1494 */
1495STATIC int
1496xfs_iunlink_remove(
1497 xfs_trans_t *tp,
1498 xfs_inode_t *ip)
1499{
1500 xfs_ino_t next_ino;
1501 xfs_mount_t *mp;
1502 xfs_agi_t *agi;
1503 xfs_dinode_t *dip;
1504 xfs_buf_t *agibp;
1505 xfs_buf_t *ibp;
1506 xfs_agnumber_t agno;
1507 xfs_agino_t agino;
1508 xfs_agino_t next_agino;
1509 xfs_buf_t *last_ibp;
1510 xfs_dinode_t *last_dip = NULL;
1511 short bucket_index;
1512 int offset, last_offset = 0;
1513 int error;
1514
1515 mp = tp->t_mountp;
1516 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1517
1518 /*
1519 * Get the agi buffer first. It ensures lock ordering
1520 * on the list.
1521 */
1522 error = xfs_read_agi(mp, tp, agno, &agibp);
1523 if (error)
1524 return error;
1525
1526 agi = XFS_BUF_TO_AGI(agibp);
1527
1528 /*
1529 * Get the index into the agi hash table for the
1530 * list this inode will go on.
1531 */
1532 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1533 ASSERT(agino != 0);
1534 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1535 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
1536 ASSERT(agi->agi_unlinked[bucket_index]);
1537
1538 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
1539 /*
1540 * We're at the head of the list. Get the inode's
1541 * on-disk buffer to see if there is anyone after us
1542 * on the list. Only modify our next pointer if it
1543 * is not already NULLAGINO. This saves us the overhead
1544 * of dealing with the buffer when there is no need to
1545 * change it.
1546 */
1547 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1548 if (error) {
1549 xfs_warn(mp, "%s: xfs_itobp() returned error %d.",
1550 __func__, error);
1551 return error;
1552 }
1553 next_agino = be32_to_cpu(dip->di_next_unlinked);
1554 ASSERT(next_agino != 0);
1555 if (next_agino != NULLAGINO) {
1556 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1557 offset = ip->i_imap.im_boffset +
1558 offsetof(xfs_dinode_t, di_next_unlinked);
1559 xfs_trans_inode_buf(tp, ibp);
1560 xfs_trans_log_buf(tp, ibp, offset,
1561 (offset + sizeof(xfs_agino_t) - 1));
1562 xfs_inobp_check(mp, ibp);
1563 } else {
1564 xfs_trans_brelse(tp, ibp);
1565 }
1566 /*
1567 * Point the bucket head pointer at the next inode.
1568 */
1569 ASSERT(next_agino != 0);
1570 ASSERT(next_agino != agino);
1571 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
1572 offset = offsetof(xfs_agi_t, agi_unlinked) +
1573 (sizeof(xfs_agino_t) * bucket_index);
1574 xfs_trans_log_buf(tp, agibp, offset,
1575 (offset + sizeof(xfs_agino_t) - 1));
1576 } else {
1577 /*
1578 * We need to search the list for the inode being freed.
1579 */
1580 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1581 last_ibp = NULL;
1582 while (next_agino != agino) {
1583 /*
1584 * If the last inode wasn't the one pointing to
1585 * us, then release its buffer since we're not
1586 * going to do anything with it.
1587 */
1588 if (last_ibp != NULL) {
1589 xfs_trans_brelse(tp, last_ibp);
1590 }
1591 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
1592 error = xfs_inotobp(mp, tp, next_ino, &last_dip,
1593 &last_ibp, &last_offset, 0);
1594 if (error) {
1595 xfs_warn(mp,
1596 "%s: xfs_inotobp() returned error %d.",
1597 __func__, error);
1598 return error;
1599 }
1600 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
1601 ASSERT(next_agino != NULLAGINO);
1602 ASSERT(next_agino != 0);
1603 }
1604 /*
1605 * Now last_ibp points to the buffer previous to us on
1606 * the unlinked list. Pull us from the list.
1607 */
1608 error = xfs_itobp(mp, tp, ip, &dip, &ibp, XBF_LOCK);
1609 if (error) {
1610 xfs_warn(mp, "%s: xfs_itobp(2) returned error %d.",
1611 __func__, error);
1612 return error;
1613 }
1614 next_agino = be32_to_cpu(dip->di_next_unlinked);
1615 ASSERT(next_agino != 0);
1616 ASSERT(next_agino != agino);
1617 if (next_agino != NULLAGINO) {
1618 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
1619 offset = ip->i_imap.im_boffset +
1620 offsetof(xfs_dinode_t, di_next_unlinked);
1621 xfs_trans_inode_buf(tp, ibp);
1622 xfs_trans_log_buf(tp, ibp, offset,
1623 (offset + sizeof(xfs_agino_t) - 1));
1624 xfs_inobp_check(mp, ibp);
1625 } else {
1626 xfs_trans_brelse(tp, ibp);
1627 }
1628 /*
1629 * Point the previous inode on the list to the next inode.
1630 */
1631 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
1632 ASSERT(next_agino != 0);
1633 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
1634 xfs_trans_inode_buf(tp, last_ibp);
1635 xfs_trans_log_buf(tp, last_ibp, offset,
1636 (offset + sizeof(xfs_agino_t) - 1));
1637 xfs_inobp_check(mp, last_ibp);
1638 }
1639 return 0;
1640}
1641
1642/*
1643 * A big issue when freeing the inode cluster is is that we _cannot_ skip any
1644 * inodes that are in memory - they all must be marked stale and attached to
1645 * the cluster buffer.
1646 */
1647STATIC void
1648xfs_ifree_cluster(
1649 xfs_inode_t *free_ip,
1650 xfs_trans_t *tp,
1651 xfs_ino_t inum)
1652{
1653 xfs_mount_t *mp = free_ip->i_mount;
1654 int blks_per_cluster;
1655 int nbufs;
1656 int ninodes;
1657 int i, j;
1658 xfs_daddr_t blkno;
1659 xfs_buf_t *bp;
1660 xfs_inode_t *ip;
1661 xfs_inode_log_item_t *iip;
1662 xfs_log_item_t *lip;
1663 struct xfs_perag *pag;
1664
1665 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
1666 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
1667 blks_per_cluster = 1;
1668 ninodes = mp->m_sb.sb_inopblock;
1669 nbufs = XFS_IALLOC_BLOCKS(mp);
1670 } else {
1671 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
1672 mp->m_sb.sb_blocksize;
1673 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
1674 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
1675 }
1676
1677 for (j = 0; j < nbufs; j++, inum += ninodes) {
1678 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
1679 XFS_INO_TO_AGBNO(mp, inum));
1680
1681 /*
1682 * We obtain and lock the backing buffer first in the process
1683 * here, as we have to ensure that any dirty inode that we
1684 * can't get the flush lock on is attached to the buffer.
1685 * If we scan the in-memory inodes first, then buffer IO can
1686 * complete before we get a lock on it, and hence we may fail
1687 * to mark all the active inodes on the buffer stale.
1688 */
1689 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
1690 mp->m_bsize * blks_per_cluster,
1691 XBF_LOCK);
1692
1693 /*
1694 * Walk the inodes already attached to the buffer and mark them
1695 * stale. These will all have the flush locks held, so an
1696 * in-memory inode walk can't lock them. By marking them all
1697 * stale first, we will not attempt to lock them in the loop
1698 * below as the XFS_ISTALE flag will be set.
1699 */
1700 lip = bp->b_fspriv;
1701 while (lip) {
1702 if (lip->li_type == XFS_LI_INODE) {
1703 iip = (xfs_inode_log_item_t *)lip;
1704 ASSERT(iip->ili_logged == 1);
1705 lip->li_cb = xfs_istale_done;
1706 xfs_trans_ail_copy_lsn(mp->m_ail,
1707 &iip->ili_flush_lsn,
1708 &iip->ili_item.li_lsn);
1709 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
1710 }
1711 lip = lip->li_bio_list;
1712 }
1713
1714
1715 /*
1716 * For each inode in memory attempt to add it to the inode
1717 * buffer and set it up for being staled on buffer IO
1718 * completion. This is safe as we've locked out tail pushing
1719 * and flushing by locking the buffer.
1720 *
1721 * We have already marked every inode that was part of a
1722 * transaction stale above, which means there is no point in
1723 * even trying to lock them.
1724 */
1725 for (i = 0; i < ninodes; i++) {
1726retry:
1727 rcu_read_lock();
1728 ip = radix_tree_lookup(&pag->pag_ici_root,
1729 XFS_INO_TO_AGINO(mp, (inum + i)));
1730
1731 /* Inode not in memory, nothing to do */
1732 if (!ip) {
1733 rcu_read_unlock();
1734 continue;
1735 }
1736
1737 /*
1738 * because this is an RCU protected lookup, we could
1739 * find a recently freed or even reallocated inode
1740 * during the lookup. We need to check under the
1741 * i_flags_lock for a valid inode here. Skip it if it
1742 * is not valid, the wrong inode or stale.
1743 */
1744 spin_lock(&ip->i_flags_lock);
1745 if (ip->i_ino != inum + i ||
1746 __xfs_iflags_test(ip, XFS_ISTALE)) {
1747 spin_unlock(&ip->i_flags_lock);
1748 rcu_read_unlock();
1749 continue;
1750 }
1751 spin_unlock(&ip->i_flags_lock);
1752
1753 /*
1754 * Don't try to lock/unlock the current inode, but we
1755 * _cannot_ skip the other inodes that we did not find
1756 * in the list attached to the buffer and are not
1757 * already marked stale. If we can't lock it, back off
1758 * and retry.
1759 */
1760 if (ip != free_ip &&
1761 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
1762 rcu_read_unlock();
1763 delay(1);
1764 goto retry;
1765 }
1766 rcu_read_unlock();
1767
1768 xfs_iflock(ip);
1769 xfs_iflags_set(ip, XFS_ISTALE);
1770
1771 /*
1772 * we don't need to attach clean inodes or those only
1773 * with unlogged changes (which we throw away, anyway).
1774 */
1775 iip = ip->i_itemp;
1776 if (!iip || xfs_inode_clean(ip)) {
1777 ASSERT(ip != free_ip);
1778 ip->i_update_core = 0;
1779 xfs_ifunlock(ip);
1780 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1781 continue;
1782 }
1783
1784 iip->ili_last_fields = iip->ili_format.ilf_fields;
1785 iip->ili_format.ilf_fields = 0;
1786 iip->ili_logged = 1;
1787 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
1788 &iip->ili_item.li_lsn);
1789
1790 xfs_buf_attach_iodone(bp, xfs_istale_done,
1791 &iip->ili_item);
1792
1793 if (ip != free_ip)
1794 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1795 }
1796
1797 xfs_trans_stale_inode_buf(tp, bp);
1798 xfs_trans_binval(tp, bp);
1799 }
1800
1801 xfs_perag_put(pag);
1802}
1803
1804/*
1805 * This is called to return an inode to the inode free list.
1806 * The inode should already be truncated to 0 length and have
1807 * no pages associated with it. This routine also assumes that
1808 * the inode is already a part of the transaction.
1809 *
1810 * The on-disk copy of the inode will have been added to the list
1811 * of unlinked inodes in the AGI. We need to remove the inode from
1812 * that list atomically with respect to freeing it here.
1813 */
1814int
1815xfs_ifree(
1816 xfs_trans_t *tp,
1817 xfs_inode_t *ip,
1818 xfs_bmap_free_t *flist)
1819{
1820 int error;
1821 int delete;
1822 xfs_ino_t first_ino;
1823 xfs_dinode_t *dip;
1824 xfs_buf_t *ibp;
1825
1826 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1827 ASSERT(ip->i_d.di_nlink == 0);
1828 ASSERT(ip->i_d.di_nextents == 0);
1829 ASSERT(ip->i_d.di_anextents == 0);
1830 ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
1831 (!S_ISREG(ip->i_d.di_mode)));
1832 ASSERT(ip->i_d.di_nblocks == 0);
1833
1834 /*
1835 * Pull the on-disk inode from the AGI unlinked list.
1836 */
1837 error = xfs_iunlink_remove(tp, ip);
1838 if (error != 0) {
1839 return error;
1840 }
1841
1842 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
1843 if (error != 0) {
1844 return error;
1845 }
1846 ip->i_d.di_mode = 0; /* mark incore inode as free */
1847 ip->i_d.di_flags = 0;
1848 ip->i_d.di_dmevmask = 0;
1849 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
1850 ip->i_df.if_ext_max =
1851 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
1852 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
1853 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1854 /*
1855 * Bump the generation count so no one will be confused
1856 * by reincarnations of this inode.
1857 */
1858 ip->i_d.di_gen++;
1859
1860 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1861
1862 error = xfs_itobp(ip->i_mount, tp, ip, &dip, &ibp, XBF_LOCK);
1863 if (error)
1864 return error;
1865
1866 /*
1867 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat
1868 * from picking up this inode when it is reclaimed (its incore state
1869 * initialzed but not flushed to disk yet). The in-core di_mode is
1870 * already cleared and a corresponding transaction logged.
1871 * The hack here just synchronizes the in-core to on-disk
1872 * di_mode value in advance before the actual inode sync to disk.
1873 * This is OK because the inode is already unlinked and would never
1874 * change its di_mode again for this inode generation.
1875 * This is a temporary hack that would require a proper fix
1876 * in the future.
1877 */
1878 dip->di_mode = 0;
1879
1880 if (delete) {
1881 xfs_ifree_cluster(ip, tp, first_ino);
1882 }
1883
1884 return 0;
1885}
1886
1887/*
1888 * Reallocate the space for if_broot based on the number of records
1889 * being added or deleted as indicated in rec_diff. Move the records
1890 * and pointers in if_broot to fit the new size. When shrinking this
1891 * will eliminate holes between the records and pointers created by
1892 * the caller. When growing this will create holes to be filled in
1893 * by the caller.
1894 *
1895 * The caller must not request to add more records than would fit in
1896 * the on-disk inode root. If the if_broot is currently NULL, then
1897 * if we adding records one will be allocated. The caller must also
1898 * not request that the number of records go below zero, although
1899 * it can go to zero.
1900 *
1901 * ip -- the inode whose if_broot area is changing
1902 * ext_diff -- the change in the number of records, positive or negative,
1903 * requested for the if_broot array.
1904 */
1905void
1906xfs_iroot_realloc(
1907 xfs_inode_t *ip,
1908 int rec_diff,
1909 int whichfork)
1910{
1911 struct xfs_mount *mp = ip->i_mount;
1912 int cur_max;
1913 xfs_ifork_t *ifp;
1914 struct xfs_btree_block *new_broot;
1915 int new_max;
1916 size_t new_size;
1917 char *np;
1918 char *op;
1919
1920 /*
1921 * Handle the degenerate case quietly.
1922 */
1923 if (rec_diff == 0) {
1924 return;
1925 }
1926
1927 ifp = XFS_IFORK_PTR(ip, whichfork);
1928 if (rec_diff > 0) {
1929 /*
1930 * If there wasn't any memory allocated before, just
1931 * allocate it now and get out.
1932 */
1933 if (ifp->if_broot_bytes == 0) {
1934 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff);
1935 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1936 ifp->if_broot_bytes = (int)new_size;
1937 return;
1938 }
1939
1940 /*
1941 * If there is already an existing if_broot, then we need
1942 * to realloc() it and shift the pointers to their new
1943 * location. The records don't change location because
1944 * they are kept butted up against the btree block header.
1945 */
1946 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1947 new_max = cur_max + rec_diff;
1948 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1949 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
1950 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */
1951 KM_SLEEP | KM_NOFS);
1952 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1953 ifp->if_broot_bytes);
1954 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
1955 (int)new_size);
1956 ifp->if_broot_bytes = (int)new_size;
1957 ASSERT(ifp->if_broot_bytes <=
1958 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
1959 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t));
1960 return;
1961 }
1962
1963 /*
1964 * rec_diff is less than 0. In this case, we are shrinking the
1965 * if_broot buffer. It must already exist. If we go to zero
1966 * records, just get rid of the root and clear the status bit.
1967 */
1968 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
1969 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
1970 new_max = cur_max + rec_diff;
1971 ASSERT(new_max >= 0);
1972 if (new_max > 0)
1973 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max);
1974 else
1975 new_size = 0;
1976 if (new_size > 0) {
1977 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
1978 /*
1979 * First copy over the btree block header.
1980 */
1981 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN);
1982 } else {
1983 new_broot = NULL;
1984 ifp->if_flags &= ~XFS_IFBROOT;
1985 }
1986
1987 /*
1988 * Only copy the records and pointers if there are any.
1989 */
1990 if (new_max > 0) {
1991 /*
1992 * First copy the records.
1993 */
1994 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
1995 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
1996 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
1997
1998 /*
1999 * Then copy the pointers.
2000 */
2001 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
2002 ifp->if_broot_bytes);
2003 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
2004 (int)new_size);
2005 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t));
2006 }
2007 kmem_free(ifp->if_broot);
2008 ifp->if_broot = new_broot;
2009 ifp->if_broot_bytes = (int)new_size;
2010 ASSERT(ifp->if_broot_bytes <=
2011 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ);
2012 return;
2013}
2014
2015
2016/*
2017 * This is called when the amount of space needed for if_data
2018 * is increased or decreased. The change in size is indicated by
2019 * the number of bytes that need to be added or deleted in the
2020 * byte_diff parameter.
2021 *
2022 * If the amount of space needed has decreased below the size of the
2023 * inline buffer, then switch to using the inline buffer. Otherwise,
2024 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
2025 * to what is needed.
2026 *
2027 * ip -- the inode whose if_data area is changing
2028 * byte_diff -- the change in the number of bytes, positive or negative,
2029 * requested for the if_data array.
2030 */
2031void
2032xfs_idata_realloc(
2033 xfs_inode_t *ip,
2034 int byte_diff,
2035 int whichfork)
2036{
2037 xfs_ifork_t *ifp;
2038 int new_size;
2039 int real_size;
2040
2041 if (byte_diff == 0) {
2042 return;
2043 }
2044
2045 ifp = XFS_IFORK_PTR(ip, whichfork);
2046 new_size = (int)ifp->if_bytes + byte_diff;
2047 ASSERT(new_size >= 0);
2048
2049 if (new_size == 0) {
2050 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2051 kmem_free(ifp->if_u1.if_data);
2052 }
2053 ifp->if_u1.if_data = NULL;
2054 real_size = 0;
2055 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
2056 /*
2057 * If the valid extents/data can fit in if_inline_ext/data,
2058 * copy them from the malloc'd vector and free it.
2059 */
2060 if (ifp->if_u1.if_data == NULL) {
2061 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2062 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2063 ASSERT(ifp->if_real_bytes != 0);
2064 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
2065 new_size);
2066 kmem_free(ifp->if_u1.if_data);
2067 ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
2068 }
2069 real_size = 0;
2070 } else {
2071 /*
2072 * Stuck with malloc/realloc.
2073 * For inline data, the underlying buffer must be
2074 * a multiple of 4 bytes in size so that it can be
2075 * logged and stay on word boundaries. We enforce
2076 * that here.
2077 */
2078 real_size = roundup(new_size, 4);
2079 if (ifp->if_u1.if_data == NULL) {
2080 ASSERT(ifp->if_real_bytes == 0);
2081 ifp->if_u1.if_data = kmem_alloc(real_size,
2082 KM_SLEEP | KM_NOFS);
2083 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
2084 /*
2085 * Only do the realloc if the underlying size
2086 * is really changing.
2087 */
2088 if (ifp->if_real_bytes != real_size) {
2089 ifp->if_u1.if_data =
2090 kmem_realloc(ifp->if_u1.if_data,
2091 real_size,
2092 ifp->if_real_bytes,
2093 KM_SLEEP | KM_NOFS);
2094 }
2095 } else {
2096 ASSERT(ifp->if_real_bytes == 0);
2097 ifp->if_u1.if_data = kmem_alloc(real_size,
2098 KM_SLEEP | KM_NOFS);
2099 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
2100 ifp->if_bytes);
2101 }
2102 }
2103 ifp->if_real_bytes = real_size;
2104 ifp->if_bytes = new_size;
2105 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2106}
2107
2108void
2109xfs_idestroy_fork(
2110 xfs_inode_t *ip,
2111 int whichfork)
2112{
2113 xfs_ifork_t *ifp;
2114
2115 ifp = XFS_IFORK_PTR(ip, whichfork);
2116 if (ifp->if_broot != NULL) {
2117 kmem_free(ifp->if_broot);
2118 ifp->if_broot = NULL;
2119 }
2120
2121 /*
2122 * If the format is local, then we can't have an extents
2123 * array so just look for an inline data array. If we're
2124 * not local then we may or may not have an extents list,
2125 * so check and free it up if we do.
2126 */
2127 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
2128 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
2129 (ifp->if_u1.if_data != NULL)) {
2130 ASSERT(ifp->if_real_bytes != 0);
2131 kmem_free(ifp->if_u1.if_data);
2132 ifp->if_u1.if_data = NULL;
2133 ifp->if_real_bytes = 0;
2134 }
2135 } else if ((ifp->if_flags & XFS_IFEXTENTS) &&
2136 ((ifp->if_flags & XFS_IFEXTIREC) ||
2137 ((ifp->if_u1.if_extents != NULL) &&
2138 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
2139 ASSERT(ifp->if_real_bytes != 0);
2140 xfs_iext_destroy(ifp);
2141 }
2142 ASSERT(ifp->if_u1.if_extents == NULL ||
2143 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
2144 ASSERT(ifp->if_real_bytes == 0);
2145 if (whichfork == XFS_ATTR_FORK) {
2146 kmem_zone_free(xfs_ifork_zone, ip->i_afp);
2147 ip->i_afp = NULL;
2148 }
2149}
2150
2151/*
2152 * This is called to unpin an inode. The caller must have the inode locked
2153 * in at least shared mode so that the buffer cannot be subsequently pinned
2154 * once someone is waiting for it to be unpinned.
2155 */
2156static void
2157xfs_iunpin_nowait(
2158 struct xfs_inode *ip)
2159{
2160 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2161
2162 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2163
2164 /* Give the log a push to start the unpinning I/O */
2165 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2166
2167}
2168
2169void
2170xfs_iunpin_wait(
2171 struct xfs_inode *ip)
2172{
2173 if (xfs_ipincount(ip)) {
2174 xfs_iunpin_nowait(ip);
2175 wait_event(ip->i_ipin_wait, (xfs_ipincount(ip) == 0));
2176 }
2177}
2178
2179/*
2180 * xfs_iextents_copy()
2181 *
2182 * This is called to copy the REAL extents (as opposed to the delayed
2183 * allocation extents) from the inode into the given buffer. It
2184 * returns the number of bytes copied into the buffer.
2185 *
2186 * If there are no delayed allocation extents, then we can just
2187 * memcpy() the extents into the buffer. Otherwise, we need to
2188 * examine each extent in turn and skip those which are delayed.
2189 */
2190int
2191xfs_iextents_copy(
2192 xfs_inode_t *ip,
2193 xfs_bmbt_rec_t *dp,
2194 int whichfork)
2195{
2196 int copied;
2197 int i;
2198 xfs_ifork_t *ifp;
2199 int nrecs;
2200 xfs_fsblock_t start_block;
2201
2202 ifp = XFS_IFORK_PTR(ip, whichfork);
2203 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2204 ASSERT(ifp->if_bytes > 0);
2205
2206 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2207 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
2208 ASSERT(nrecs > 0);
2209
2210 /*
2211 * There are some delayed allocation extents in the
2212 * inode, so copy the extents one at a time and skip
2213 * the delayed ones. There must be at least one
2214 * non-delayed extent.
2215 */
2216 copied = 0;
2217 for (i = 0; i < nrecs; i++) {
2218 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
2219 start_block = xfs_bmbt_get_startblock(ep);
2220 if (isnullstartblock(start_block)) {
2221 /*
2222 * It's a delayed allocation extent, so skip it.
2223 */
2224 continue;
2225 }
2226
2227 /* Translate to on disk format */
2228 put_unaligned(cpu_to_be64(ep->l0), &dp->l0);
2229 put_unaligned(cpu_to_be64(ep->l1), &dp->l1);
2230 dp++;
2231 copied++;
2232 }
2233 ASSERT(copied != 0);
2234 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
2235
2236 return (copied * (uint)sizeof(xfs_bmbt_rec_t));
2237}
2238
2239/*
2240 * Each of the following cases stores data into the same region
2241 * of the on-disk inode, so only one of them can be valid at
2242 * any given time. While it is possible to have conflicting formats
2243 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
2244 * in EXTENTS format, this can only happen when the fork has
2245 * changed formats after being modified but before being flushed.
2246 * In these cases, the format always takes precedence, because the
2247 * format indicates the current state of the fork.
2248 */
2249/*ARGSUSED*/
2250STATIC void
2251xfs_iflush_fork(
2252 xfs_inode_t *ip,
2253 xfs_dinode_t *dip,
2254 xfs_inode_log_item_t *iip,
2255 int whichfork,
2256 xfs_buf_t *bp)
2257{
2258 char *cp;
2259 xfs_ifork_t *ifp;
2260 xfs_mount_t *mp;
2261#ifdef XFS_TRANS_DEBUG
2262 int first;
2263#endif
2264 static const short brootflag[2] =
2265 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
2266 static const short dataflag[2] =
2267 { XFS_ILOG_DDATA, XFS_ILOG_ADATA };
2268 static const short extflag[2] =
2269 { XFS_ILOG_DEXT, XFS_ILOG_AEXT };
2270
2271 if (!iip)
2272 return;
2273 ifp = XFS_IFORK_PTR(ip, whichfork);
2274 /*
2275 * This can happen if we gave up in iformat in an error path,
2276 * for the attribute fork.
2277 */
2278 if (!ifp) {
2279 ASSERT(whichfork == XFS_ATTR_FORK);
2280 return;
2281 }
2282 cp = XFS_DFORK_PTR(dip, whichfork);
2283 mp = ip->i_mount;
2284 switch (XFS_IFORK_FORMAT(ip, whichfork)) {
2285 case XFS_DINODE_FMT_LOCAL:
2286 if ((iip->ili_format.ilf_fields & dataflag[whichfork]) &&
2287 (ifp->if_bytes > 0)) {
2288 ASSERT(ifp->if_u1.if_data != NULL);
2289 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
2290 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
2291 }
2292 break;
2293
2294 case XFS_DINODE_FMT_EXTENTS:
2295 ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
2296 !(iip->ili_format.ilf_fields & extflag[whichfork]));
2297 if ((iip->ili_format.ilf_fields & extflag[whichfork]) &&
2298 (ifp->if_bytes > 0)) {
2299 ASSERT(xfs_iext_get_ext(ifp, 0));
2300 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
2301 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
2302 whichfork);
2303 }
2304 break;
2305
2306 case XFS_DINODE_FMT_BTREE:
2307 if ((iip->ili_format.ilf_fields & brootflag[whichfork]) &&
2308 (ifp->if_broot_bytes > 0)) {
2309 ASSERT(ifp->if_broot != NULL);
2310 ASSERT(ifp->if_broot_bytes <=
2311 (XFS_IFORK_SIZE(ip, whichfork) +
2312 XFS_BROOT_SIZE_ADJ));
2313 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
2314 (xfs_bmdr_block_t *)cp,
2315 XFS_DFORK_SIZE(dip, mp, whichfork));
2316 }
2317 break;
2318
2319 case XFS_DINODE_FMT_DEV:
2320 if (iip->ili_format.ilf_fields & XFS_ILOG_DEV) {
2321 ASSERT(whichfork == XFS_DATA_FORK);
2322 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
2323 }
2324 break;
2325
2326 case XFS_DINODE_FMT_UUID:
2327 if (iip->ili_format.ilf_fields & XFS_ILOG_UUID) {
2328 ASSERT(whichfork == XFS_DATA_FORK);
2329 memcpy(XFS_DFORK_DPTR(dip),
2330 &ip->i_df.if_u2.if_uuid,
2331 sizeof(uuid_t));
2332 }
2333 break;
2334
2335 default:
2336 ASSERT(0);
2337 break;
2338 }
2339}
2340
2341STATIC int
2342xfs_iflush_cluster(
2343 xfs_inode_t *ip,
2344 xfs_buf_t *bp)
2345{
2346 xfs_mount_t *mp = ip->i_mount;
2347 struct xfs_perag *pag;
2348 unsigned long first_index, mask;
2349 unsigned long inodes_per_cluster;
2350 int ilist_size;
2351 xfs_inode_t **ilist;
2352 xfs_inode_t *iq;
2353 int nr_found;
2354 int clcount = 0;
2355 int bufwasdelwri;
2356 int i;
2357
2358 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2359
2360 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2361 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2362 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2363 if (!ilist)
2364 goto out_put;
2365
2366 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2367 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2368 rcu_read_lock();
2369 /* really need a gang lookup range call here */
2370 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2371 first_index, inodes_per_cluster);
2372 if (nr_found == 0)
2373 goto out_free;
2374
2375 for (i = 0; i < nr_found; i++) {
2376 iq = ilist[i];
2377 if (iq == ip)
2378 continue;
2379
2380 /*
2381 * because this is an RCU protected lookup, we could find a
2382 * recently freed or even reallocated inode during the lookup.
2383 * We need to check under the i_flags_lock for a valid inode
2384 * here. Skip it if it is not valid or the wrong inode.
2385 */
2386 spin_lock(&ip->i_flags_lock);
2387 if (!ip->i_ino ||
2388 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2389 spin_unlock(&ip->i_flags_lock);
2390 continue;
2391 }
2392 spin_unlock(&ip->i_flags_lock);
2393
2394 /*
2395 * Do an un-protected check to see if the inode is dirty and
2396 * is a candidate for flushing. These checks will be repeated
2397 * later after the appropriate locks are acquired.
2398 */
2399 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2400 continue;
2401
2402 /*
2403 * Try to get locks. If any are unavailable or it is pinned,
2404 * then this inode cannot be flushed and is skipped.
2405 */
2406
2407 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2408 continue;
2409 if (!xfs_iflock_nowait(iq)) {
2410 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2411 continue;
2412 }
2413 if (xfs_ipincount(iq)) {
2414 xfs_ifunlock(iq);
2415 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2416 continue;
2417 }
2418
2419 /*
2420 * arriving here means that this inode can be flushed. First
2421 * re-check that it's dirty before flushing.
2422 */
2423 if (!xfs_inode_clean(iq)) {
2424 int error;
2425 error = xfs_iflush_int(iq, bp);
2426 if (error) {
2427 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2428 goto cluster_corrupt_out;
2429 }
2430 clcount++;
2431 } else {
2432 xfs_ifunlock(iq);
2433 }
2434 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2435 }
2436
2437 if (clcount) {
2438 XFS_STATS_INC(xs_icluster_flushcnt);
2439 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2440 }
2441
2442out_free:
2443 rcu_read_unlock();
2444 kmem_free(ilist);
2445out_put:
2446 xfs_perag_put(pag);
2447 return 0;
2448
2449
2450cluster_corrupt_out:
2451 /*
2452 * Corruption detected in the clustering loop. Invalidate the
2453 * inode buffer and shut down the filesystem.
2454 */
2455 rcu_read_unlock();
2456 /*
2457 * Clean up the buffer. If it was B_DELWRI, just release it --
2458 * brelse can handle it with no problems. If not, shut down the
2459 * filesystem before releasing the buffer.
2460 */
2461 bufwasdelwri = XFS_BUF_ISDELAYWRITE(bp);
2462 if (bufwasdelwri)
2463 xfs_buf_relse(bp);
2464
2465 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2466
2467 if (!bufwasdelwri) {
2468 /*
2469 * Just like incore_relse: if we have b_iodone functions,
2470 * mark the buffer as an error and call them. Otherwise
2471 * mark it as stale and brelse.
2472 */
2473 if (bp->b_iodone) {
2474 XFS_BUF_UNDONE(bp);
2475 XFS_BUF_STALE(bp);
2476 xfs_buf_ioerror(bp, EIO);
2477 xfs_buf_ioend(bp, 0);
2478 } else {
2479 XFS_BUF_STALE(bp);
2480 xfs_buf_relse(bp);
2481 }
2482 }
2483
2484 /*
2485 * Unlocks the flush lock
2486 */
2487 xfs_iflush_abort(iq);
2488 kmem_free(ilist);
2489 xfs_perag_put(pag);
2490 return XFS_ERROR(EFSCORRUPTED);
2491}
2492
2493/*
2494 * xfs_iflush() will write a modified inode's changes out to the
2495 * inode's on disk home. The caller must have the inode lock held
2496 * in at least shared mode and the inode flush completion must be
2497 * active as well. The inode lock will still be held upon return from
2498 * the call and the caller is free to unlock it.
2499 * The inode flush will be completed when the inode reaches the disk.
2500 * The flags indicate how the inode's buffer should be written out.
2501 */
2502int
2503xfs_iflush(
2504 xfs_inode_t *ip,
2505 uint flags)
2506{
2507 xfs_inode_log_item_t *iip;
2508 xfs_buf_t *bp;
2509 xfs_dinode_t *dip;
2510 xfs_mount_t *mp;
2511 int error;
2512
2513 XFS_STATS_INC(xs_iflush_count);
2514
2515 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2516 ASSERT(!completion_done(&ip->i_flush));
2517 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2518 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2519
2520 iip = ip->i_itemp;
2521 mp = ip->i_mount;
2522
2523 /*
2524 * We can't flush the inode until it is unpinned, so wait for it if we
2525 * are allowed to block. We know no one new can pin it, because we are
2526 * holding the inode lock shared and you need to hold it exclusively to
2527 * pin the inode.
2528 *
2529 * If we are not allowed to block, force the log out asynchronously so
2530 * that when we come back the inode will be unpinned. If other inodes
2531 * in the same cluster are dirty, they will probably write the inode
2532 * out for us if they occur after the log force completes.
2533 */
2534 if (!(flags & SYNC_WAIT) && xfs_ipincount(ip)) {
2535 xfs_iunpin_nowait(ip);
2536 xfs_ifunlock(ip);
2537 return EAGAIN;
2538 }
2539 xfs_iunpin_wait(ip);
2540
2541 /*
2542 * For stale inodes we cannot rely on the backing buffer remaining
2543 * stale in cache for the remaining life of the stale inode and so
2544 * xfs_itobp() below may give us a buffer that no longer contains
2545 * inodes below. We have to check this after ensuring the inode is
2546 * unpinned so that it is safe to reclaim the stale inode after the
2547 * flush call.
2548 */
2549 if (xfs_iflags_test(ip, XFS_ISTALE)) {
2550 xfs_ifunlock(ip);
2551 return 0;
2552 }
2553
2554 /*
2555 * This may have been unpinned because the filesystem is shutting
2556 * down forcibly. If that's the case we must not write this inode
2557 * to disk, because the log record didn't make it to disk!
2558 */
2559 if (XFS_FORCED_SHUTDOWN(mp)) {
2560 ip->i_update_core = 0;
2561 if (iip)
2562 iip->ili_format.ilf_fields = 0;
2563 xfs_ifunlock(ip);
2564 return XFS_ERROR(EIO);
2565 }
2566
2567 /*
2568 * Get the buffer containing the on-disk inode.
2569 */
2570 error = xfs_itobp(mp, NULL, ip, &dip, &bp,
2571 (flags & SYNC_TRYLOCK) ? XBF_TRYLOCK : XBF_LOCK);
2572 if (error || !bp) {
2573 xfs_ifunlock(ip);
2574 return error;
2575 }
2576
2577 /*
2578 * First flush out the inode that xfs_iflush was called with.
2579 */
2580 error = xfs_iflush_int(ip, bp);
2581 if (error)
2582 goto corrupt_out;
2583
2584 /*
2585 * If the buffer is pinned then push on the log now so we won't
2586 * get stuck waiting in the write for too long.
2587 */
2588 if (xfs_buf_ispinned(bp))
2589 xfs_log_force(mp, 0);
2590
2591 /*
2592 * inode clustering:
2593 * see if other inodes can be gathered into this write
2594 */
2595 error = xfs_iflush_cluster(ip, bp);
2596 if (error)
2597 goto cluster_corrupt_out;
2598
2599 if (flags & SYNC_WAIT)
2600 error = xfs_bwrite(mp, bp);
2601 else
2602 xfs_bdwrite(mp, bp);
2603 return error;
2604
2605corrupt_out:
2606 xfs_buf_relse(bp);
2607 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
2608cluster_corrupt_out:
2609 /*
2610 * Unlocks the flush lock
2611 */
2612 xfs_iflush_abort(ip);
2613 return XFS_ERROR(EFSCORRUPTED);
2614}
2615
2616
2617STATIC int
2618xfs_iflush_int(
2619 xfs_inode_t *ip,
2620 xfs_buf_t *bp)
2621{
2622 xfs_inode_log_item_t *iip;
2623 xfs_dinode_t *dip;
2624 xfs_mount_t *mp;
2625#ifdef XFS_TRANS_DEBUG
2626 int first;
2627#endif
2628
2629 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2630 ASSERT(!completion_done(&ip->i_flush));
2631 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
2632 ip->i_d.di_nextents > ip->i_df.if_ext_max);
2633
2634 iip = ip->i_itemp;
2635 mp = ip->i_mount;
2636
2637 /* set *dip = inode's place in the buffer */
2638 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
2639
2640 /*
2641 * Clear i_update_core before copying out the data.
2642 * This is for coordination with our timestamp updates
2643 * that don't hold the inode lock. They will always
2644 * update the timestamps BEFORE setting i_update_core,
2645 * so if we clear i_update_core after they set it we
2646 * are guaranteed to see their updates to the timestamps.
2647 * I believe that this depends on strongly ordered memory
2648 * semantics, but we have that. We use the SYNCHRONIZE
2649 * macro to make sure that the compiler does not reorder
2650 * the i_update_core access below the data copy below.
2651 */
2652 ip->i_update_core = 0;
2653 SYNCHRONIZE();
2654
2655 /*
2656 * Make sure to get the latest timestamps from the Linux inode.
2657 */
2658 xfs_synchronize_times(ip);
2659
2660 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
2661 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
2662 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2663 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
2664 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
2665 goto corrupt_out;
2666 }
2667 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
2668 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
2669 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2670 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
2671 __func__, ip->i_ino, ip, ip->i_d.di_magic);
2672 goto corrupt_out;
2673 }
2674 if (S_ISREG(ip->i_d.di_mode)) {
2675 if (XFS_TEST_ERROR(
2676 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2677 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
2678 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
2679 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2680 "%s: Bad regular inode %Lu, ptr 0x%p",
2681 __func__, ip->i_ino, ip);
2682 goto corrupt_out;
2683 }
2684 } else if (S_ISDIR(ip->i_d.di_mode)) {
2685 if (XFS_TEST_ERROR(
2686 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
2687 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
2688 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
2689 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
2690 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2691 "%s: Bad directory inode %Lu, ptr 0x%p",
2692 __func__, ip->i_ino, ip);
2693 goto corrupt_out;
2694 }
2695 }
2696 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
2697 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
2698 XFS_RANDOM_IFLUSH_5)) {
2699 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2700 "%s: detected corrupt incore inode %Lu, "
2701 "total extents = %d, nblocks = %Ld, ptr 0x%p",
2702 __func__, ip->i_ino,
2703 ip->i_d.di_nextents + ip->i_d.di_anextents,
2704 ip->i_d.di_nblocks, ip);
2705 goto corrupt_out;
2706 }
2707 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
2708 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
2709 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
2710 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
2711 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
2712 goto corrupt_out;
2713 }
2714 /*
2715 * bump the flush iteration count, used to detect flushes which
2716 * postdate a log record during recovery.
2717 */
2718
2719 ip->i_d.di_flushiter++;
2720
2721 /*
2722 * Copy the dirty parts of the inode into the on-disk
2723 * inode. We always copy out the core of the inode,
2724 * because if the inode is dirty at all the core must
2725 * be.
2726 */
2727 xfs_dinode_to_disk(dip, &ip->i_d);
2728
2729 /* Wrap, we never let the log put out DI_MAX_FLUSH */
2730 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
2731 ip->i_d.di_flushiter = 0;
2732
2733 /*
2734 * If this is really an old format inode and the superblock version
2735 * has not been updated to support only new format inodes, then
2736 * convert back to the old inode format. If the superblock version
2737 * has been updated, then make the conversion permanent.
2738 */
2739 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
2740 if (ip->i_d.di_version == 1) {
2741 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
2742 /*
2743 * Convert it back.
2744 */
2745 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
2746 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
2747 } else {
2748 /*
2749 * The superblock version has already been bumped,
2750 * so just make the conversion to the new inode
2751 * format permanent.
2752 */
2753 ip->i_d.di_version = 2;
2754 dip->di_version = 2;
2755 ip->i_d.di_onlink = 0;
2756 dip->di_onlink = 0;
2757 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
2758 memset(&(dip->di_pad[0]), 0,
2759 sizeof(dip->di_pad));
2760 ASSERT(xfs_get_projid(ip) == 0);
2761 }
2762 }
2763
2764 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
2765 if (XFS_IFORK_Q(ip))
2766 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
2767 xfs_inobp_check(mp, bp);
2768
2769 /*
2770 * We've recorded everything logged in the inode, so we'd
2771 * like to clear the ilf_fields bits so we don't log and
2772 * flush things unnecessarily. However, we can't stop
2773 * logging all this information until the data we've copied
2774 * into the disk buffer is written to disk. If we did we might
2775 * overwrite the copy of the inode in the log with all the
2776 * data after re-logging only part of it, and in the face of
2777 * a crash we wouldn't have all the data we need to recover.
2778 *
2779 * What we do is move the bits to the ili_last_fields field.
2780 * When logging the inode, these bits are moved back to the
2781 * ilf_fields field. In the xfs_iflush_done() routine we
2782 * clear ili_last_fields, since we know that the information
2783 * those bits represent is permanently on disk. As long as
2784 * the flush completes before the inode is logged again, then
2785 * both ilf_fields and ili_last_fields will be cleared.
2786 *
2787 * We can play with the ilf_fields bits here, because the inode
2788 * lock must be held exclusively in order to set bits there
2789 * and the flush lock protects the ili_last_fields bits.
2790 * Set ili_logged so the flush done
2791 * routine can tell whether or not to look in the AIL.
2792 * Also, store the current LSN of the inode so that we can tell
2793 * whether the item has moved in the AIL from xfs_iflush_done().
2794 * In order to read the lsn we need the AIL lock, because
2795 * it is a 64 bit value that cannot be read atomically.
2796 */
2797 if (iip != NULL && iip->ili_format.ilf_fields != 0) {
2798 iip->ili_last_fields = iip->ili_format.ilf_fields;
2799 iip->ili_format.ilf_fields = 0;
2800 iip->ili_logged = 1;
2801
2802 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2803 &iip->ili_item.li_lsn);
2804
2805 /*
2806 * Attach the function xfs_iflush_done to the inode's
2807 * buffer. This will remove the inode from the AIL
2808 * and unlock the inode's flush lock when the inode is
2809 * completely written to disk.
2810 */
2811 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
2812
2813 ASSERT(bp->b_fspriv != NULL);
2814 ASSERT(bp->b_iodone != NULL);
2815 } else {
2816 /*
2817 * We're flushing an inode which is not in the AIL and has
2818 * not been logged but has i_update_core set. For this
2819 * case we can use a B_DELWRI flush and immediately drop
2820 * the inode flush lock because we can avoid the whole
2821 * AIL state thing. It's OK to drop the flush lock now,
2822 * because we've already locked the buffer and to do anything
2823 * you really need both.
2824 */
2825 if (iip != NULL) {
2826 ASSERT(iip->ili_logged == 0);
2827 ASSERT(iip->ili_last_fields == 0);
2828 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0);
2829 }
2830 xfs_ifunlock(ip);
2831 }
2832
2833 return 0;
2834
2835corrupt_out:
2836 return XFS_ERROR(EFSCORRUPTED);
2837}
2838
2839/*
2840 * Return a pointer to the extent record at file index idx.
2841 */
2842xfs_bmbt_rec_host_t *
2843xfs_iext_get_ext(
2844 xfs_ifork_t *ifp, /* inode fork pointer */
2845 xfs_extnum_t idx) /* index of target extent */
2846{
2847 ASSERT(idx >= 0);
2848 ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
2849
2850 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
2851 return ifp->if_u1.if_ext_irec->er_extbuf;
2852 } else if (ifp->if_flags & XFS_IFEXTIREC) {
2853 xfs_ext_irec_t *erp; /* irec pointer */
2854 int erp_idx = 0; /* irec index */
2855 xfs_extnum_t page_idx = idx; /* ext index in target list */
2856
2857 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
2858 return &erp->er_extbuf[page_idx];
2859 } else if (ifp->if_bytes) {
2860 return &ifp->if_u1.if_extents[idx];
2861 } else {
2862 return NULL;
2863 }
2864}
2865
2866/*
2867 * Insert new item(s) into the extent records for incore inode
2868 * fork 'ifp'. 'count' new items are inserted at index 'idx'.
2869 */
2870void
2871xfs_iext_insert(
2872 xfs_inode_t *ip, /* incore inode pointer */
2873 xfs_extnum_t idx, /* starting index of new items */
2874 xfs_extnum_t count, /* number of inserted items */
2875 xfs_bmbt_irec_t *new, /* items to insert */
2876 int state) /* type of extent conversion */
2877{
2878 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
2879 xfs_extnum_t i; /* extent record index */
2880
2881 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
2882
2883 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2884 xfs_iext_add(ifp, idx, count);
2885 for (i = idx; i < idx + count; i++, new++)
2886 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
2887}
2888
2889/*
2890 * This is called when the amount of space required for incore file
2891 * extents needs to be increased. The ext_diff parameter stores the
2892 * number of new extents being added and the idx parameter contains
2893 * the extent index where the new extents will be added. If the new
2894 * extents are being appended, then we just need to (re)allocate and
2895 * initialize the space. Otherwise, if the new extents are being
2896 * inserted into the middle of the existing entries, a bit more work
2897 * is required to make room for the new extents to be inserted. The
2898 * caller is responsible for filling in the new extent entries upon
2899 * return.
2900 */
2901void
2902xfs_iext_add(
2903 xfs_ifork_t *ifp, /* inode fork pointer */
2904 xfs_extnum_t idx, /* index to begin adding exts */
2905 int ext_diff) /* number of extents to add */
2906{
2907 int byte_diff; /* new bytes being added */
2908 int new_size; /* size of extents after adding */
2909 xfs_extnum_t nextents; /* number of extents in file */
2910
2911 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2912 ASSERT((idx >= 0) && (idx <= nextents));
2913 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
2914 new_size = ifp->if_bytes + byte_diff;
2915 /*
2916 * If the new number of extents (nextents + ext_diff)
2917 * fits inside the inode, then continue to use the inline
2918 * extent buffer.
2919 */
2920 if (nextents + ext_diff <= XFS_INLINE_EXTS) {
2921 if (idx < nextents) {
2922 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
2923 &ifp->if_u2.if_inline_ext[idx],
2924 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2925 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
2926 }
2927 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
2928 ifp->if_real_bytes = 0;
2929 }
2930 /*
2931 * Otherwise use a linear (direct) extent list.
2932 * If the extents are currently inside the inode,
2933 * xfs_iext_realloc_direct will switch us from
2934 * inline to direct extent allocation mode.
2935 */
2936 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
2937 xfs_iext_realloc_direct(ifp, new_size);
2938 if (idx < nextents) {
2939 memmove(&ifp->if_u1.if_extents[idx + ext_diff],
2940 &ifp->if_u1.if_extents[idx],
2941 (nextents - idx) * sizeof(xfs_bmbt_rec_t));
2942 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
2943 }
2944 }
2945 /* Indirection array */
2946 else {
2947 xfs_ext_irec_t *erp;
2948 int erp_idx = 0;
2949 int page_idx = idx;
2950
2951 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
2952 if (ifp->if_flags & XFS_IFEXTIREC) {
2953 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
2954 } else {
2955 xfs_iext_irec_init(ifp);
2956 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
2957 erp = ifp->if_u1.if_ext_irec;
2958 }
2959 /* Extents fit in target extent page */
2960 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
2961 if (page_idx < erp->er_extcount) {
2962 memmove(&erp->er_extbuf[page_idx + ext_diff],
2963 &erp->er_extbuf[page_idx],
2964 (erp->er_extcount - page_idx) *
2965 sizeof(xfs_bmbt_rec_t));
2966 memset(&erp->er_extbuf[page_idx], 0, byte_diff);
2967 }
2968 erp->er_extcount += ext_diff;
2969 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
2970 }
2971 /* Insert a new extent page */
2972 else if (erp) {
2973 xfs_iext_add_indirect_multi(ifp,
2974 erp_idx, page_idx, ext_diff);
2975 }
2976 /*
2977 * If extent(s) are being appended to the last page in
2978 * the indirection array and the new extent(s) don't fit
2979 * in the page, then erp is NULL and erp_idx is set to
2980 * the next index needed in the indirection array.
2981 */
2982 else {
2983 int count = ext_diff;
2984
2985 while (count) {
2986 erp = xfs_iext_irec_new(ifp, erp_idx);
2987 erp->er_extcount = count;
2988 count -= MIN(count, (int)XFS_LINEAR_EXTS);
2989 if (count) {
2990 erp_idx++;
2991 }
2992 }
2993 }
2994 }
2995 ifp->if_bytes = new_size;
2996}
2997
2998/*
2999 * This is called when incore extents are being added to the indirection
3000 * array and the new extents do not fit in the target extent list. The
3001 * erp_idx parameter contains the irec index for the target extent list
3002 * in the indirection array, and the idx parameter contains the extent
3003 * index within the list. The number of extents being added is stored
3004 * in the count parameter.
3005 *
3006 * |-------| |-------|
3007 * | | | | idx - number of extents before idx
3008 * | idx | | count |
3009 * | | | | count - number of extents being inserted at idx
3010 * |-------| |-------|
3011 * | count | | nex2 | nex2 - number of extents after idx + count
3012 * |-------| |-------|
3013 */
3014void
3015xfs_iext_add_indirect_multi(
3016 xfs_ifork_t *ifp, /* inode fork pointer */
3017 int erp_idx, /* target extent irec index */
3018 xfs_extnum_t idx, /* index within target list */
3019 int count) /* new extents being added */
3020{
3021 int byte_diff; /* new bytes being added */
3022 xfs_ext_irec_t *erp; /* pointer to irec entry */
3023 xfs_extnum_t ext_diff; /* number of extents to add */
3024 xfs_extnum_t ext_cnt; /* new extents still needed */
3025 xfs_extnum_t nex2; /* extents after idx + count */
3026 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */
3027 int nlists; /* number of irec's (lists) */
3028
3029 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3030 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3031 nex2 = erp->er_extcount - idx;
3032 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3033
3034 /*
3035 * Save second part of target extent list
3036 * (all extents past */
3037 if (nex2) {
3038 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3039 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
3040 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
3041 erp->er_extcount -= nex2;
3042 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
3043 memset(&erp->er_extbuf[idx], 0, byte_diff);
3044 }
3045
3046 /*
3047 * Add the new extents to the end of the target
3048 * list, then allocate new irec record(s) and
3049 * extent buffer(s) as needed to store the rest
3050 * of the new extents.
3051 */
3052 ext_cnt = count;
3053 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
3054 if (ext_diff) {
3055 erp->er_extcount += ext_diff;
3056 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3057 ext_cnt -= ext_diff;
3058 }
3059 while (ext_cnt) {
3060 erp_idx++;
3061 erp = xfs_iext_irec_new(ifp, erp_idx);
3062 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
3063 erp->er_extcount = ext_diff;
3064 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
3065 ext_cnt -= ext_diff;
3066 }
3067
3068 /* Add nex2 extents back to indirection array */
3069 if (nex2) {
3070 xfs_extnum_t ext_avail;
3071 int i;
3072
3073 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
3074 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
3075 i = 0;
3076 /*
3077 * If nex2 extents fit in the current page, append
3078 * nex2_ep after the new extents.
3079 */
3080 if (nex2 <= ext_avail) {
3081 i = erp->er_extcount;
3082 }
3083 /*
3084 * Otherwise, check if space is available in the
3085 * next page.
3086 */
3087 else if ((erp_idx < nlists - 1) &&
3088 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
3089 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
3090 erp_idx++;
3091 erp++;
3092 /* Create a hole for nex2 extents */
3093 memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
3094 erp->er_extcount * sizeof(xfs_bmbt_rec_t));
3095 }
3096 /*
3097 * Final choice, create a new extent page for
3098 * nex2 extents.
3099 */
3100 else {
3101 erp_idx++;
3102 erp = xfs_iext_irec_new(ifp, erp_idx);
3103 }
3104 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
3105 kmem_free(nex2_ep);
3106 erp->er_extcount += nex2;
3107 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
3108 }
3109}
3110
3111/*
3112 * This is called when the amount of space required for incore file
3113 * extents needs to be decreased. The ext_diff parameter stores the
3114 * number of extents to be removed and the idx parameter contains
3115 * the extent index where the extents will be removed from.
3116 *
3117 * If the amount of space needed has decreased below the linear
3118 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
3119 * extent array. Otherwise, use kmem_realloc() to adjust the
3120 * size to what is needed.
3121 */
3122void
3123xfs_iext_remove(
3124 xfs_inode_t *ip, /* incore inode pointer */
3125 xfs_extnum_t idx, /* index to begin removing exts */
3126 int ext_diff, /* number of extents to remove */
3127 int state) /* type of extent conversion */
3128{
3129 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
3130 xfs_extnum_t nextents; /* number of extents in file */
3131 int new_size; /* size of extents after removal */
3132
3133 trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
3134
3135 ASSERT(ext_diff > 0);
3136 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3137 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
3138
3139 if (new_size == 0) {
3140 xfs_iext_destroy(ifp);
3141 } else if (ifp->if_flags & XFS_IFEXTIREC) {
3142 xfs_iext_remove_indirect(ifp, idx, ext_diff);
3143 } else if (ifp->if_real_bytes) {
3144 xfs_iext_remove_direct(ifp, idx, ext_diff);
3145 } else {
3146 xfs_iext_remove_inline(ifp, idx, ext_diff);
3147 }
3148 ifp->if_bytes = new_size;
3149}
3150
3151/*
3152 * This removes ext_diff extents from the inline buffer, beginning
3153 * at extent index idx.
3154 */
3155void
3156xfs_iext_remove_inline(
3157 xfs_ifork_t *ifp, /* inode fork pointer */
3158 xfs_extnum_t idx, /* index to begin removing exts */
3159 int ext_diff) /* number of extents to remove */
3160{
3161 int nextents; /* number of extents in file */
3162
3163 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3164 ASSERT(idx < XFS_INLINE_EXTS);
3165 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3166 ASSERT(((nextents - ext_diff) > 0) &&
3167 (nextents - ext_diff) < XFS_INLINE_EXTS);
3168
3169 if (idx + ext_diff < nextents) {
3170 memmove(&ifp->if_u2.if_inline_ext[idx],
3171 &ifp->if_u2.if_inline_ext[idx + ext_diff],
3172 (nextents - (idx + ext_diff)) *
3173 sizeof(xfs_bmbt_rec_t));
3174 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
3175 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3176 } else {
3177 memset(&ifp->if_u2.if_inline_ext[idx], 0,
3178 ext_diff * sizeof(xfs_bmbt_rec_t));
3179 }
3180}
3181
3182/*
3183 * This removes ext_diff extents from a linear (direct) extent list,
3184 * beginning at extent index idx. If the extents are being removed
3185 * from the end of the list (ie. truncate) then we just need to re-
3186 * allocate the list to remove the extra space. Otherwise, if the
3187 * extents are being removed from the middle of the existing extent
3188 * entries, then we first need to move the extent records beginning
3189 * at idx + ext_diff up in the list to overwrite the records being
3190 * removed, then remove the extra space via kmem_realloc.
3191 */
3192void
3193xfs_iext_remove_direct(
3194 xfs_ifork_t *ifp, /* inode fork pointer */
3195 xfs_extnum_t idx, /* index to begin removing exts */
3196 int ext_diff) /* number of extents to remove */
3197{
3198 xfs_extnum_t nextents; /* number of extents in file */
3199 int new_size; /* size of extents after removal */
3200
3201 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3202 new_size = ifp->if_bytes -
3203 (ext_diff * sizeof(xfs_bmbt_rec_t));
3204 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3205
3206 if (new_size == 0) {
3207 xfs_iext_destroy(ifp);
3208 return;
3209 }
3210 /* Move extents up in the list (if needed) */
3211 if (idx + ext_diff < nextents) {
3212 memmove(&ifp->if_u1.if_extents[idx],
3213 &ifp->if_u1.if_extents[idx + ext_diff],
3214 (nextents - (idx + ext_diff)) *
3215 sizeof(xfs_bmbt_rec_t));
3216 }
3217 memset(&ifp->if_u1.if_extents[nextents - ext_diff],
3218 0, ext_diff * sizeof(xfs_bmbt_rec_t));
3219 /*
3220 * Reallocate the direct extent list. If the extents
3221 * will fit inside the inode then xfs_iext_realloc_direct
3222 * will switch from direct to inline extent allocation
3223 * mode for us.
3224 */
3225 xfs_iext_realloc_direct(ifp, new_size);
3226 ifp->if_bytes = new_size;
3227}
3228
3229/*
3230 * This is called when incore extents are being removed from the
3231 * indirection array and the extents being removed span multiple extent
3232 * buffers. The idx parameter contains the file extent index where we
3233 * want to begin removing extents, and the count parameter contains
3234 * how many extents need to be removed.
3235 *
3236 * |-------| |-------|
3237 * | nex1 | | | nex1 - number of extents before idx
3238 * |-------| | count |
3239 * | | | | count - number of extents being removed at idx
3240 * | count | |-------|
3241 * | | | nex2 | nex2 - number of extents after idx + count
3242 * |-------| |-------|
3243 */
3244void
3245xfs_iext_remove_indirect(
3246 xfs_ifork_t *ifp, /* inode fork pointer */
3247 xfs_extnum_t idx, /* index to begin removing extents */
3248 int count) /* number of extents to remove */
3249{
3250 xfs_ext_irec_t *erp; /* indirection array pointer */
3251 int erp_idx = 0; /* indirection array index */
3252 xfs_extnum_t ext_cnt; /* extents left to remove */
3253 xfs_extnum_t ext_diff; /* extents to remove in current list */
3254 xfs_extnum_t nex1; /* number of extents before idx */
3255 xfs_extnum_t nex2; /* extents after idx + count */
3256 int page_idx = idx; /* index in target extent list */
3257
3258 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3259 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
3260 ASSERT(erp != NULL);
3261 nex1 = page_idx;
3262 ext_cnt = count;
3263 while (ext_cnt) {
3264 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
3265 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
3266 /*
3267 * Check for deletion of entire list;
3268 * xfs_iext_irec_remove() updates extent offsets.
3269 */
3270 if (ext_diff == erp->er_extcount) {
3271 xfs_iext_irec_remove(ifp, erp_idx);
3272 ext_cnt -= ext_diff;
3273 nex1 = 0;
3274 if (ext_cnt) {
3275 ASSERT(erp_idx < ifp->if_real_bytes /
3276 XFS_IEXT_BUFSZ);
3277 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3278 nex1 = 0;
3279 continue;
3280 } else {
3281 break;
3282 }
3283 }
3284 /* Move extents up (if needed) */
3285 if (nex2) {
3286 memmove(&erp->er_extbuf[nex1],
3287 &erp->er_extbuf[nex1 + ext_diff],
3288 nex2 * sizeof(xfs_bmbt_rec_t));
3289 }
3290 /* Zero out rest of page */
3291 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
3292 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
3293 /* Update remaining counters */
3294 erp->er_extcount -= ext_diff;
3295 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
3296 ext_cnt -= ext_diff;
3297 nex1 = 0;
3298 erp_idx++;
3299 erp++;
3300 }
3301 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
3302 xfs_iext_irec_compact(ifp);
3303}
3304
3305/*
3306 * Create, destroy, or resize a linear (direct) block of extents.
3307 */
3308void
3309xfs_iext_realloc_direct(
3310 xfs_ifork_t *ifp, /* inode fork pointer */
3311 int new_size) /* new size of extents */
3312{
3313 int rnew_size; /* real new size of extents */
3314
3315 rnew_size = new_size;
3316
3317 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
3318 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
3319 (new_size != ifp->if_real_bytes)));
3320
3321 /* Free extent records */
3322 if (new_size == 0) {
3323 xfs_iext_destroy(ifp);
3324 }
3325 /* Resize direct extent list and zero any new bytes */
3326 else if (ifp->if_real_bytes) {
3327 /* Check if extents will fit inside the inode */
3328 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
3329 xfs_iext_direct_to_inline(ifp, new_size /
3330 (uint)sizeof(xfs_bmbt_rec_t));
3331 ifp->if_bytes = new_size;
3332 return;
3333 }
3334 if (!is_power_of_2(new_size)){
3335 rnew_size = roundup_pow_of_two(new_size);
3336 }
3337 if (rnew_size != ifp->if_real_bytes) {
3338 ifp->if_u1.if_extents =
3339 kmem_realloc(ifp->if_u1.if_extents,
3340 rnew_size,
3341 ifp->if_real_bytes, KM_NOFS);
3342 }
3343 if (rnew_size > ifp->if_real_bytes) {
3344 memset(&ifp->if_u1.if_extents[ifp->if_bytes /
3345 (uint)sizeof(xfs_bmbt_rec_t)], 0,
3346 rnew_size - ifp->if_real_bytes);
3347 }
3348 }
3349 /*
3350 * Switch from the inline extent buffer to a direct
3351 * extent list. Be sure to include the inline extent
3352 * bytes in new_size.
3353 */
3354 else {
3355 new_size += ifp->if_bytes;
3356 if (!is_power_of_2(new_size)) {
3357 rnew_size = roundup_pow_of_two(new_size);
3358 }
3359 xfs_iext_inline_to_direct(ifp, rnew_size);
3360 }
3361 ifp->if_real_bytes = rnew_size;
3362 ifp->if_bytes = new_size;
3363}
3364
3365/*
3366 * Switch from linear (direct) extent records to inline buffer.
3367 */
3368void
3369xfs_iext_direct_to_inline(
3370 xfs_ifork_t *ifp, /* inode fork pointer */
3371 xfs_extnum_t nextents) /* number of extents in file */
3372{
3373 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3374 ASSERT(nextents <= XFS_INLINE_EXTS);
3375 /*
3376 * The inline buffer was zeroed when we switched
3377 * from inline to direct extent allocation mode,
3378 * so we don't need to clear it here.
3379 */
3380 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
3381 nextents * sizeof(xfs_bmbt_rec_t));
3382 kmem_free(ifp->if_u1.if_extents);
3383 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
3384 ifp->if_real_bytes = 0;
3385}
3386
3387/*
3388 * Switch from inline buffer to linear (direct) extent records.
3389 * new_size should already be rounded up to the next power of 2
3390 * by the caller (when appropriate), so use new_size as it is.
3391 * However, since new_size may be rounded up, we can't update
3392 * if_bytes here. It is the caller's responsibility to update
3393 * if_bytes upon return.
3394 */
3395void
3396xfs_iext_inline_to_direct(
3397 xfs_ifork_t *ifp, /* inode fork pointer */
3398 int new_size) /* number of extents in file */
3399{
3400 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
3401 memset(ifp->if_u1.if_extents, 0, new_size);
3402 if (ifp->if_bytes) {
3403 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
3404 ifp->if_bytes);
3405 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3406 sizeof(xfs_bmbt_rec_t));
3407 }
3408 ifp->if_real_bytes = new_size;
3409}
3410
3411/*
3412 * Resize an extent indirection array to new_size bytes.
3413 */
3414STATIC void
3415xfs_iext_realloc_indirect(
3416 xfs_ifork_t *ifp, /* inode fork pointer */
3417 int new_size) /* new indirection array size */
3418{
3419 int nlists; /* number of irec's (ex lists) */
3420 int size; /* current indirection array size */
3421
3422 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3423 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3424 size = nlists * sizeof(xfs_ext_irec_t);
3425 ASSERT(ifp->if_real_bytes);
3426 ASSERT((new_size >= 0) && (new_size != size));
3427 if (new_size == 0) {
3428 xfs_iext_destroy(ifp);
3429 } else {
3430 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
3431 kmem_realloc(ifp->if_u1.if_ext_irec,
3432 new_size, size, KM_NOFS);
3433 }
3434}
3435
3436/*
3437 * Switch from indirection array to linear (direct) extent allocations.
3438 */
3439STATIC void
3440xfs_iext_indirect_to_direct(
3441 xfs_ifork_t *ifp) /* inode fork pointer */
3442{
3443 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
3444 xfs_extnum_t nextents; /* number of extents in file */
3445 int size; /* size of file extents */
3446
3447 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3448 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3449 ASSERT(nextents <= XFS_LINEAR_EXTS);
3450 size = nextents * sizeof(xfs_bmbt_rec_t);
3451
3452 xfs_iext_irec_compact_pages(ifp);
3453 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
3454
3455 ep = ifp->if_u1.if_ext_irec->er_extbuf;
3456 kmem_free(ifp->if_u1.if_ext_irec);
3457 ifp->if_flags &= ~XFS_IFEXTIREC;
3458 ifp->if_u1.if_extents = ep;
3459 ifp->if_bytes = size;
3460 if (nextents < XFS_LINEAR_EXTS) {
3461 xfs_iext_realloc_direct(ifp, size);
3462 }
3463}
3464
3465/*
3466 * Free incore file extents.
3467 */
3468void
3469xfs_iext_destroy(
3470 xfs_ifork_t *ifp) /* inode fork pointer */
3471{
3472 if (ifp->if_flags & XFS_IFEXTIREC) {
3473 int erp_idx;
3474 int nlists;
3475
3476 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3477 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
3478 xfs_iext_irec_remove(ifp, erp_idx);
3479 }
3480 ifp->if_flags &= ~XFS_IFEXTIREC;
3481 } else if (ifp->if_real_bytes) {
3482 kmem_free(ifp->if_u1.if_extents);
3483 } else if (ifp->if_bytes) {
3484 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
3485 sizeof(xfs_bmbt_rec_t));
3486 }
3487 ifp->if_u1.if_extents = NULL;
3488 ifp->if_real_bytes = 0;
3489 ifp->if_bytes = 0;
3490}
3491
3492/*
3493 * Return a pointer to the extent record for file system block bno.
3494 */
3495xfs_bmbt_rec_host_t * /* pointer to found extent record */
3496xfs_iext_bno_to_ext(
3497 xfs_ifork_t *ifp, /* inode fork pointer */
3498 xfs_fileoff_t bno, /* block number to search for */
3499 xfs_extnum_t *idxp) /* index of target extent */
3500{
3501 xfs_bmbt_rec_host_t *base; /* pointer to first extent */
3502 xfs_filblks_t blockcount = 0; /* number of blocks in extent */
3503 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */
3504 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3505 int high; /* upper boundary in search */
3506 xfs_extnum_t idx = 0; /* index of target extent */
3507 int low; /* lower boundary in search */
3508 xfs_extnum_t nextents; /* number of file extents */
3509 xfs_fileoff_t startoff = 0; /* start offset of extent */
3510
3511 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3512 if (nextents == 0) {
3513 *idxp = 0;
3514 return NULL;
3515 }
3516 low = 0;
3517 if (ifp->if_flags & XFS_IFEXTIREC) {
3518 /* Find target extent list */
3519 int erp_idx = 0;
3520 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
3521 base = erp->er_extbuf;
3522 high = erp->er_extcount - 1;
3523 } else {
3524 base = ifp->if_u1.if_extents;
3525 high = nextents - 1;
3526 }
3527 /* Binary search extent records */
3528 while (low <= high) {
3529 idx = (low + high) >> 1;
3530 ep = base + idx;
3531 startoff = xfs_bmbt_get_startoff(ep);
3532 blockcount = xfs_bmbt_get_blockcount(ep);
3533 if (bno < startoff) {
3534 high = idx - 1;
3535 } else if (bno >= startoff + blockcount) {
3536 low = idx + 1;
3537 } else {
3538 /* Convert back to file-based extent index */
3539 if (ifp->if_flags & XFS_IFEXTIREC) {
3540 idx += erp->er_extoff;
3541 }
3542 *idxp = idx;
3543 return ep;
3544 }
3545 }
3546 /* Convert back to file-based extent index */
3547 if (ifp->if_flags & XFS_IFEXTIREC) {
3548 idx += erp->er_extoff;
3549 }
3550 if (bno >= startoff + blockcount) {
3551 if (++idx == nextents) {
3552 ep = NULL;
3553 } else {
3554 ep = xfs_iext_get_ext(ifp, idx);
3555 }
3556 }
3557 *idxp = idx;
3558 return ep;
3559}
3560
3561/*
3562 * Return a pointer to the indirection array entry containing the
3563 * extent record for filesystem block bno. Store the index of the
3564 * target irec in *erp_idxp.
3565 */
3566xfs_ext_irec_t * /* pointer to found extent record */
3567xfs_iext_bno_to_irec(
3568 xfs_ifork_t *ifp, /* inode fork pointer */
3569 xfs_fileoff_t bno, /* block number to search for */
3570 int *erp_idxp) /* irec index of target ext list */
3571{
3572 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */
3573 xfs_ext_irec_t *erp_next; /* next indirection array entry */
3574 int erp_idx; /* indirection array index */
3575 int nlists; /* number of extent irec's (lists) */
3576 int high; /* binary search upper limit */
3577 int low; /* binary search lower limit */
3578
3579 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3580 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3581 erp_idx = 0;
3582 low = 0;
3583 high = nlists - 1;
3584 while (low <= high) {
3585 erp_idx = (low + high) >> 1;
3586 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3587 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
3588 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
3589 high = erp_idx - 1;
3590 } else if (erp_next && bno >=
3591 xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
3592 low = erp_idx + 1;
3593 } else {
3594 break;
3595 }
3596 }
3597 *erp_idxp = erp_idx;
3598 return erp;
3599}
3600
3601/*
3602 * Return a pointer to the indirection array entry containing the
3603 * extent record at file extent index *idxp. Store the index of the
3604 * target irec in *erp_idxp and store the page index of the target
3605 * extent record in *idxp.
3606 */
3607xfs_ext_irec_t *
3608xfs_iext_idx_to_irec(
3609 xfs_ifork_t *ifp, /* inode fork pointer */
3610 xfs_extnum_t *idxp, /* extent index (file -> page) */
3611 int *erp_idxp, /* pointer to target irec */
3612 int realloc) /* new bytes were just added */
3613{
3614 xfs_ext_irec_t *prev; /* pointer to previous irec */
3615 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */
3616 int erp_idx; /* indirection array index */
3617 int nlists; /* number of irec's (ex lists) */
3618 int high; /* binary search upper limit */
3619 int low; /* binary search lower limit */
3620 xfs_extnum_t page_idx = *idxp; /* extent index in target list */
3621
3622 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3623 ASSERT(page_idx >= 0);
3624 ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
3625 ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
3626
3627 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3628 erp_idx = 0;
3629 low = 0;
3630 high = nlists - 1;
3631
3632 /* Binary search extent irec's */
3633 while (low <= high) {
3634 erp_idx = (low + high) >> 1;
3635 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3636 prev = erp_idx > 0 ? erp - 1 : NULL;
3637 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
3638 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
3639 high = erp_idx - 1;
3640 } else if (page_idx > erp->er_extoff + erp->er_extcount ||
3641 (page_idx == erp->er_extoff + erp->er_extcount &&
3642 !realloc)) {
3643 low = erp_idx + 1;
3644 } else if (page_idx == erp->er_extoff + erp->er_extcount &&
3645 erp->er_extcount == XFS_LINEAR_EXTS) {
3646 ASSERT(realloc);
3647 page_idx = 0;
3648 erp_idx++;
3649 erp = erp_idx < nlists ? erp + 1 : NULL;
3650 break;
3651 } else {
3652 page_idx -= erp->er_extoff;
3653 break;
3654 }
3655 }
3656 *idxp = page_idx;
3657 *erp_idxp = erp_idx;
3658 return(erp);
3659}
3660
3661/*
3662 * Allocate and initialize an indirection array once the space needed
3663 * for incore extents increases above XFS_IEXT_BUFSZ.
3664 */
3665void
3666xfs_iext_irec_init(
3667 xfs_ifork_t *ifp) /* inode fork pointer */
3668{
3669 xfs_ext_irec_t *erp; /* indirection array pointer */
3670 xfs_extnum_t nextents; /* number of extents in file */
3671
3672 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
3673 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3674 ASSERT(nextents <= XFS_LINEAR_EXTS);
3675
3676 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
3677
3678 if (nextents == 0) {
3679 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3680 } else if (!ifp->if_real_bytes) {
3681 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
3682 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
3683 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
3684 }
3685 erp->er_extbuf = ifp->if_u1.if_extents;
3686 erp->er_extcount = nextents;
3687 erp->er_extoff = 0;
3688
3689 ifp->if_flags |= XFS_IFEXTIREC;
3690 ifp->if_real_bytes = XFS_IEXT_BUFSZ;
3691 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
3692 ifp->if_u1.if_ext_irec = erp;
3693
3694 return;
3695}
3696
3697/*
3698 * Allocate and initialize a new entry in the indirection array.
3699 */
3700xfs_ext_irec_t *
3701xfs_iext_irec_new(
3702 xfs_ifork_t *ifp, /* inode fork pointer */
3703 int erp_idx) /* index for new irec */
3704{
3705 xfs_ext_irec_t *erp; /* indirection array pointer */
3706 int i; /* loop counter */
3707 int nlists; /* number of irec's (ex lists) */
3708
3709 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3710 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3711
3712 /* Resize indirection array */
3713 xfs_iext_realloc_indirect(ifp, ++nlists *
3714 sizeof(xfs_ext_irec_t));
3715 /*
3716 * Move records down in the array so the
3717 * new page can use erp_idx.
3718 */
3719 erp = ifp->if_u1.if_ext_irec;
3720 for (i = nlists - 1; i > erp_idx; i--) {
3721 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
3722 }
3723 ASSERT(i == erp_idx);
3724
3725 /* Initialize new extent record */
3726 erp = ifp->if_u1.if_ext_irec;
3727 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
3728 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3729 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
3730 erp[erp_idx].er_extcount = 0;
3731 erp[erp_idx].er_extoff = erp_idx > 0 ?
3732 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
3733 return (&erp[erp_idx]);
3734}
3735
3736/*
3737 * Remove a record from the indirection array.
3738 */
3739void
3740xfs_iext_irec_remove(
3741 xfs_ifork_t *ifp, /* inode fork pointer */
3742 int erp_idx) /* irec index to remove */
3743{
3744 xfs_ext_irec_t *erp; /* indirection array pointer */
3745 int i; /* loop counter */
3746 int nlists; /* number of irec's (ex lists) */
3747
3748 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3749 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3750 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3751 if (erp->er_extbuf) {
3752 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
3753 -erp->er_extcount);
3754 kmem_free(erp->er_extbuf);
3755 }
3756 /* Compact extent records */
3757 erp = ifp->if_u1.if_ext_irec;
3758 for (i = erp_idx; i < nlists - 1; i++) {
3759 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
3760 }
3761 /*
3762 * Manually free the last extent record from the indirection
3763 * array. A call to xfs_iext_realloc_indirect() with a size
3764 * of zero would result in a call to xfs_iext_destroy() which
3765 * would in turn call this function again, creating a nasty
3766 * infinite loop.
3767 */
3768 if (--nlists) {
3769 xfs_iext_realloc_indirect(ifp,
3770 nlists * sizeof(xfs_ext_irec_t));
3771 } else {
3772 kmem_free(ifp->if_u1.if_ext_irec);
3773 }
3774 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
3775}
3776
3777/*
3778 * This is called to clean up large amounts of unused memory allocated
3779 * by the indirection array. Before compacting anything though, verify
3780 * that the indirection array is still needed and switch back to the
3781 * linear extent list (or even the inline buffer) if possible. The
3782 * compaction policy is as follows:
3783 *
3784 * Full Compaction: Extents fit into a single page (or inline buffer)
3785 * Partial Compaction: Extents occupy less than 50% of allocated space
3786 * No Compaction: Extents occupy at least 50% of allocated space
3787 */
3788void
3789xfs_iext_irec_compact(
3790 xfs_ifork_t *ifp) /* inode fork pointer */
3791{
3792 xfs_extnum_t nextents; /* number of extents in file */
3793 int nlists; /* number of irec's (ex lists) */
3794
3795 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3796 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3797 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3798
3799 if (nextents == 0) {
3800 xfs_iext_destroy(ifp);
3801 } else if (nextents <= XFS_INLINE_EXTS) {
3802 xfs_iext_indirect_to_direct(ifp);
3803 xfs_iext_direct_to_inline(ifp, nextents);
3804 } else if (nextents <= XFS_LINEAR_EXTS) {
3805 xfs_iext_indirect_to_direct(ifp);
3806 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
3807 xfs_iext_irec_compact_pages(ifp);
3808 }
3809}
3810
3811/*
3812 * Combine extents from neighboring extent pages.
3813 */
3814void
3815xfs_iext_irec_compact_pages(
3816 xfs_ifork_t *ifp) /* inode fork pointer */
3817{
3818 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */
3819 int erp_idx = 0; /* indirection array index */
3820 int nlists; /* number of irec's (ex lists) */
3821
3822 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3823 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3824 while (erp_idx < nlists - 1) {
3825 erp = &ifp->if_u1.if_ext_irec[erp_idx];
3826 erp_next = erp + 1;
3827 if (erp_next->er_extcount <=
3828 (XFS_LINEAR_EXTS - erp->er_extcount)) {
3829 memcpy(&erp->er_extbuf[erp->er_extcount],
3830 erp_next->er_extbuf, erp_next->er_extcount *
3831 sizeof(xfs_bmbt_rec_t));
3832 erp->er_extcount += erp_next->er_extcount;
3833 /*
3834 * Free page before removing extent record
3835 * so er_extoffs don't get modified in
3836 * xfs_iext_irec_remove.
3837 */
3838 kmem_free(erp_next->er_extbuf);
3839 erp_next->er_extbuf = NULL;
3840 xfs_iext_irec_remove(ifp, erp_idx + 1);
3841 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3842 } else {
3843 erp_idx++;
3844 }
3845 }
3846}
3847
3848/*
3849 * This is called to update the er_extoff field in the indirection
3850 * array when extents have been added or removed from one of the
3851 * extent lists. erp_idx contains the irec index to begin updating
3852 * at and ext_diff contains the number of extents that were added
3853 * or removed.
3854 */
3855void
3856xfs_iext_irec_update_extoffs(
3857 xfs_ifork_t *ifp, /* inode fork pointer */
3858 int erp_idx, /* irec index to update */
3859 int ext_diff) /* number of new extents */
3860{
3861 int i; /* loop counter */
3862 int nlists; /* number of irec's (ex lists */
3863
3864 ASSERT(ifp->if_flags & XFS_IFEXTIREC);
3865 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
3866 for (i = erp_idx; i < nlists; i++) {
3867 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
3868 }
3869}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include <linux/iversion.h>
7
8#include "xfs.h"
9#include "xfs_fs.h"
10#include "xfs_shared.h"
11#include "xfs_format.h"
12#include "xfs_log_format.h"
13#include "xfs_trans_resv.h"
14#include "xfs_mount.h"
15#include "xfs_defer.h"
16#include "xfs_inode.h"
17#include "xfs_dir2.h"
18#include "xfs_attr.h"
19#include "xfs_trans_space.h"
20#include "xfs_trans.h"
21#include "xfs_buf_item.h"
22#include "xfs_inode_item.h"
23#include "xfs_iunlink_item.h"
24#include "xfs_ialloc.h"
25#include "xfs_bmap.h"
26#include "xfs_bmap_util.h"
27#include "xfs_errortag.h"
28#include "xfs_error.h"
29#include "xfs_quota.h"
30#include "xfs_filestream.h"
31#include "xfs_trace.h"
32#include "xfs_icache.h"
33#include "xfs_symlink.h"
34#include "xfs_trans_priv.h"
35#include "xfs_log.h"
36#include "xfs_bmap_btree.h"
37#include "xfs_reflink.h"
38#include "xfs_ag.h"
39#include "xfs_log_priv.h"
40#include "xfs_health.h"
41
42struct kmem_cache *xfs_inode_cache;
43
44STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
45STATIC int xfs_iunlink_remove(struct xfs_trans *tp, struct xfs_perag *pag,
46 struct xfs_inode *);
47
48/*
49 * helper function to extract extent size hint from inode
50 */
51xfs_extlen_t
52xfs_get_extsz_hint(
53 struct xfs_inode *ip)
54{
55 /*
56 * No point in aligning allocations if we need to COW to actually
57 * write to them.
58 */
59 if (xfs_is_always_cow_inode(ip))
60 return 0;
61 if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
62 return ip->i_extsize;
63 if (XFS_IS_REALTIME_INODE(ip))
64 return ip->i_mount->m_sb.sb_rextsize;
65 return 0;
66}
67
68/*
69 * Helper function to extract CoW extent size hint from inode.
70 * Between the extent size hint and the CoW extent size hint, we
71 * return the greater of the two. If the value is zero (automatic),
72 * use the default size.
73 */
74xfs_extlen_t
75xfs_get_cowextsz_hint(
76 struct xfs_inode *ip)
77{
78 xfs_extlen_t a, b;
79
80 a = 0;
81 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
82 a = ip->i_cowextsize;
83 b = xfs_get_extsz_hint(ip);
84
85 a = max(a, b);
86 if (a == 0)
87 return XFS_DEFAULT_COWEXTSZ_HINT;
88 return a;
89}
90
91/*
92 * These two are wrapper routines around the xfs_ilock() routine used to
93 * centralize some grungy code. They are used in places that wish to lock the
94 * inode solely for reading the extents. The reason these places can't just
95 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
96 * bringing in of the extents from disk for a file in b-tree format. If the
97 * inode is in b-tree format, then we need to lock the inode exclusively until
98 * the extents are read in. Locking it exclusively all the time would limit
99 * our parallelism unnecessarily, though. What we do instead is check to see
100 * if the extents have been read in yet, and only lock the inode exclusively
101 * if they have not.
102 *
103 * The functions return a value which should be given to the corresponding
104 * xfs_iunlock() call.
105 */
106uint
107xfs_ilock_data_map_shared(
108 struct xfs_inode *ip)
109{
110 uint lock_mode = XFS_ILOCK_SHARED;
111
112 if (xfs_need_iread_extents(&ip->i_df))
113 lock_mode = XFS_ILOCK_EXCL;
114 xfs_ilock(ip, lock_mode);
115 return lock_mode;
116}
117
118uint
119xfs_ilock_attr_map_shared(
120 struct xfs_inode *ip)
121{
122 uint lock_mode = XFS_ILOCK_SHARED;
123
124 if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af))
125 lock_mode = XFS_ILOCK_EXCL;
126 xfs_ilock(ip, lock_mode);
127 return lock_mode;
128}
129
130/*
131 * You can't set both SHARED and EXCL for the same lock,
132 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED,
133 * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values
134 * to set in lock_flags.
135 */
136static inline void
137xfs_lock_flags_assert(
138 uint lock_flags)
139{
140 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
141 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
142 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
143 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
144 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
145 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
146 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
147 ASSERT(lock_flags != 0);
148}
149
150/*
151 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
152 * multi-reader locks: invalidate_lock and the i_lock. This routine allows
153 * various combinations of the locks to be obtained.
154 *
155 * The 3 locks should always be ordered so that the IO lock is obtained first,
156 * the mmap lock second and the ilock last in order to prevent deadlock.
157 *
158 * Basic locking order:
159 *
160 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock
161 *
162 * mmap_lock locking order:
163 *
164 * i_rwsem -> page lock -> mmap_lock
165 * mmap_lock -> invalidate_lock -> page_lock
166 *
167 * The difference in mmap_lock locking order mean that we cannot hold the
168 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths
169 * can fault in pages during copy in/out (for buffered IO) or require the
170 * mmap_lock in get_user_pages() to map the user pages into the kernel address
171 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page
172 * fault because page faults already hold the mmap_lock.
173 *
174 * Hence to serialise fully against both syscall and mmap based IO, we need to
175 * take both the i_rwsem and the invalidate_lock. These locks should *only* be
176 * both taken in places where we need to invalidate the page cache in a race
177 * free manner (e.g. truncate, hole punch and other extent manipulation
178 * functions).
179 */
180void
181xfs_ilock(
182 xfs_inode_t *ip,
183 uint lock_flags)
184{
185 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
186
187 xfs_lock_flags_assert(lock_flags);
188
189 if (lock_flags & XFS_IOLOCK_EXCL) {
190 down_write_nested(&VFS_I(ip)->i_rwsem,
191 XFS_IOLOCK_DEP(lock_flags));
192 } else if (lock_flags & XFS_IOLOCK_SHARED) {
193 down_read_nested(&VFS_I(ip)->i_rwsem,
194 XFS_IOLOCK_DEP(lock_flags));
195 }
196
197 if (lock_flags & XFS_MMAPLOCK_EXCL) {
198 down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
199 XFS_MMAPLOCK_DEP(lock_flags));
200 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
201 down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
202 XFS_MMAPLOCK_DEP(lock_flags));
203 }
204
205 if (lock_flags & XFS_ILOCK_EXCL)
206 down_write_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
207 else if (lock_flags & XFS_ILOCK_SHARED)
208 down_read_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
209}
210
211/*
212 * This is just like xfs_ilock(), except that the caller
213 * is guaranteed not to sleep. It returns 1 if it gets
214 * the requested locks and 0 otherwise. If the IO lock is
215 * obtained but the inode lock cannot be, then the IO lock
216 * is dropped before returning.
217 *
218 * ip -- the inode being locked
219 * lock_flags -- this parameter indicates the inode's locks to be
220 * to be locked. See the comment for xfs_ilock() for a list
221 * of valid values.
222 */
223int
224xfs_ilock_nowait(
225 xfs_inode_t *ip,
226 uint lock_flags)
227{
228 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
229
230 xfs_lock_flags_assert(lock_flags);
231
232 if (lock_flags & XFS_IOLOCK_EXCL) {
233 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
234 goto out;
235 } else if (lock_flags & XFS_IOLOCK_SHARED) {
236 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
237 goto out;
238 }
239
240 if (lock_flags & XFS_MMAPLOCK_EXCL) {
241 if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
242 goto out_undo_iolock;
243 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
244 if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
245 goto out_undo_iolock;
246 }
247
248 if (lock_flags & XFS_ILOCK_EXCL) {
249 if (!down_write_trylock(&ip->i_lock))
250 goto out_undo_mmaplock;
251 } else if (lock_flags & XFS_ILOCK_SHARED) {
252 if (!down_read_trylock(&ip->i_lock))
253 goto out_undo_mmaplock;
254 }
255 return 1;
256
257out_undo_mmaplock:
258 if (lock_flags & XFS_MMAPLOCK_EXCL)
259 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
260 else if (lock_flags & XFS_MMAPLOCK_SHARED)
261 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
262out_undo_iolock:
263 if (lock_flags & XFS_IOLOCK_EXCL)
264 up_write(&VFS_I(ip)->i_rwsem);
265 else if (lock_flags & XFS_IOLOCK_SHARED)
266 up_read(&VFS_I(ip)->i_rwsem);
267out:
268 return 0;
269}
270
271/*
272 * xfs_iunlock() is used to drop the inode locks acquired with
273 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
274 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
275 * that we know which locks to drop.
276 *
277 * ip -- the inode being unlocked
278 * lock_flags -- this parameter indicates the inode's locks to be
279 * to be unlocked. See the comment for xfs_ilock() for a list
280 * of valid values for this parameter.
281 *
282 */
283void
284xfs_iunlock(
285 xfs_inode_t *ip,
286 uint lock_flags)
287{
288 xfs_lock_flags_assert(lock_flags);
289
290 if (lock_flags & XFS_IOLOCK_EXCL)
291 up_write(&VFS_I(ip)->i_rwsem);
292 else if (lock_flags & XFS_IOLOCK_SHARED)
293 up_read(&VFS_I(ip)->i_rwsem);
294
295 if (lock_flags & XFS_MMAPLOCK_EXCL)
296 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
297 else if (lock_flags & XFS_MMAPLOCK_SHARED)
298 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
299
300 if (lock_flags & XFS_ILOCK_EXCL)
301 up_write(&ip->i_lock);
302 else if (lock_flags & XFS_ILOCK_SHARED)
303 up_read(&ip->i_lock);
304
305 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
306}
307
308/*
309 * give up write locks. the i/o lock cannot be held nested
310 * if it is being demoted.
311 */
312void
313xfs_ilock_demote(
314 xfs_inode_t *ip,
315 uint lock_flags)
316{
317 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
318 ASSERT((lock_flags &
319 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
320
321 if (lock_flags & XFS_ILOCK_EXCL)
322 downgrade_write(&ip->i_lock);
323 if (lock_flags & XFS_MMAPLOCK_EXCL)
324 downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);
325 if (lock_flags & XFS_IOLOCK_EXCL)
326 downgrade_write(&VFS_I(ip)->i_rwsem);
327
328 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
329}
330
331void
332xfs_assert_ilocked(
333 struct xfs_inode *ip,
334 uint lock_flags)
335{
336 /*
337 * Sometimes we assert the ILOCK is held exclusively, but we're in
338 * a workqueue, so lockdep doesn't know we're the owner.
339 */
340 if (lock_flags & XFS_ILOCK_SHARED)
341 rwsem_assert_held(&ip->i_lock);
342 else if (lock_flags & XFS_ILOCK_EXCL)
343 rwsem_assert_held_write_nolockdep(&ip->i_lock);
344
345 if (lock_flags & XFS_MMAPLOCK_SHARED)
346 rwsem_assert_held(&VFS_I(ip)->i_mapping->invalidate_lock);
347 else if (lock_flags & XFS_MMAPLOCK_EXCL)
348 rwsem_assert_held_write(&VFS_I(ip)->i_mapping->invalidate_lock);
349
350 if (lock_flags & XFS_IOLOCK_SHARED)
351 rwsem_assert_held(&VFS_I(ip)->i_rwsem);
352 else if (lock_flags & XFS_IOLOCK_EXCL)
353 rwsem_assert_held_write(&VFS_I(ip)->i_rwsem);
354}
355
356/*
357 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
358 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
359 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
360 * errors and warnings.
361 */
362#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
363static bool
364xfs_lockdep_subclass_ok(
365 int subclass)
366{
367 return subclass < MAX_LOCKDEP_SUBCLASSES;
368}
369#else
370#define xfs_lockdep_subclass_ok(subclass) (true)
371#endif
372
373/*
374 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
375 * value. This can be called for any type of inode lock combination, including
376 * parent locking. Care must be taken to ensure we don't overrun the subclass
377 * storage fields in the class mask we build.
378 */
379static inline uint
380xfs_lock_inumorder(
381 uint lock_mode,
382 uint subclass)
383{
384 uint class = 0;
385
386 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
387 XFS_ILOCK_RTSUM)));
388 ASSERT(xfs_lockdep_subclass_ok(subclass));
389
390 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
391 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
392 class += subclass << XFS_IOLOCK_SHIFT;
393 }
394
395 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
396 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
397 class += subclass << XFS_MMAPLOCK_SHIFT;
398 }
399
400 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
401 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
402 class += subclass << XFS_ILOCK_SHIFT;
403 }
404
405 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
406}
407
408/*
409 * The following routine will lock n inodes in exclusive mode. We assume the
410 * caller calls us with the inodes in i_ino order.
411 *
412 * We need to detect deadlock where an inode that we lock is in the AIL and we
413 * start waiting for another inode that is locked by a thread in a long running
414 * transaction (such as truncate). This can result in deadlock since the long
415 * running trans might need to wait for the inode we just locked in order to
416 * push the tail and free space in the log.
417 *
418 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
419 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
420 * lock more than one at a time, lockdep will report false positives saying we
421 * have violated locking orders.
422 */
423static void
424xfs_lock_inodes(
425 struct xfs_inode **ips,
426 int inodes,
427 uint lock_mode)
428{
429 int attempts = 0;
430 uint i;
431 int j;
432 bool try_lock;
433 struct xfs_log_item *lp;
434
435 /*
436 * Currently supports between 2 and 5 inodes with exclusive locking. We
437 * support an arbitrary depth of locking here, but absolute limits on
438 * inodes depend on the type of locking and the limits placed by
439 * lockdep annotations in xfs_lock_inumorder. These are all checked by
440 * the asserts.
441 */
442 ASSERT(ips && inodes >= 2 && inodes <= 5);
443 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
444 XFS_ILOCK_EXCL));
445 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
446 XFS_ILOCK_SHARED)));
447 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
448 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
449 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
450 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
451
452 if (lock_mode & XFS_IOLOCK_EXCL) {
453 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
454 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
455 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
456
457again:
458 try_lock = false;
459 i = 0;
460 for (; i < inodes; i++) {
461 ASSERT(ips[i]);
462
463 if (i && (ips[i] == ips[i - 1])) /* Already locked */
464 continue;
465
466 /*
467 * If try_lock is not set yet, make sure all locked inodes are
468 * not in the AIL. If any are, set try_lock to be used later.
469 */
470 if (!try_lock) {
471 for (j = (i - 1); j >= 0 && !try_lock; j--) {
472 lp = &ips[j]->i_itemp->ili_item;
473 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
474 try_lock = true;
475 }
476 }
477
478 /*
479 * If any of the previous locks we have locked is in the AIL,
480 * we must TRY to get the second and subsequent locks. If
481 * we can't get any, we must release all we have
482 * and try again.
483 */
484 if (!try_lock) {
485 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
486 continue;
487 }
488
489 /* try_lock means we have an inode locked that is in the AIL. */
490 ASSERT(i != 0);
491 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
492 continue;
493
494 /*
495 * Unlock all previous guys and try again. xfs_iunlock will try
496 * to push the tail if the inode is in the AIL.
497 */
498 attempts++;
499 for (j = i - 1; j >= 0; j--) {
500 /*
501 * Check to see if we've already unlocked this one. Not
502 * the first one going back, and the inode ptr is the
503 * same.
504 */
505 if (j != (i - 1) && ips[j] == ips[j + 1])
506 continue;
507
508 xfs_iunlock(ips[j], lock_mode);
509 }
510
511 if ((attempts % 5) == 0) {
512 delay(1); /* Don't just spin the CPU */
513 }
514 goto again;
515 }
516}
517
518/*
519 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
520 * mmaplock must be double-locked separately since we use i_rwsem and
521 * invalidate_lock for that. We now support taking one lock EXCL and the
522 * other SHARED.
523 */
524void
525xfs_lock_two_inodes(
526 struct xfs_inode *ip0,
527 uint ip0_mode,
528 struct xfs_inode *ip1,
529 uint ip1_mode)
530{
531 int attempts = 0;
532 struct xfs_log_item *lp;
533
534 ASSERT(hweight32(ip0_mode) == 1);
535 ASSERT(hweight32(ip1_mode) == 1);
536 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
537 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
538 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
539 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
540 ASSERT(ip0->i_ino != ip1->i_ino);
541
542 if (ip0->i_ino > ip1->i_ino) {
543 swap(ip0, ip1);
544 swap(ip0_mode, ip1_mode);
545 }
546
547 again:
548 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
549
550 /*
551 * If the first lock we have locked is in the AIL, we must TRY to get
552 * the second lock. If we can't get it, we must release the first one
553 * and try again.
554 */
555 lp = &ip0->i_itemp->ili_item;
556 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
557 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
558 xfs_iunlock(ip0, ip0_mode);
559 if ((++attempts % 5) == 0)
560 delay(1); /* Don't just spin the CPU */
561 goto again;
562 }
563 } else {
564 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
565 }
566}
567
568uint
569xfs_ip2xflags(
570 struct xfs_inode *ip)
571{
572 uint flags = 0;
573
574 if (ip->i_diflags & XFS_DIFLAG_ANY) {
575 if (ip->i_diflags & XFS_DIFLAG_REALTIME)
576 flags |= FS_XFLAG_REALTIME;
577 if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
578 flags |= FS_XFLAG_PREALLOC;
579 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
580 flags |= FS_XFLAG_IMMUTABLE;
581 if (ip->i_diflags & XFS_DIFLAG_APPEND)
582 flags |= FS_XFLAG_APPEND;
583 if (ip->i_diflags & XFS_DIFLAG_SYNC)
584 flags |= FS_XFLAG_SYNC;
585 if (ip->i_diflags & XFS_DIFLAG_NOATIME)
586 flags |= FS_XFLAG_NOATIME;
587 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
588 flags |= FS_XFLAG_NODUMP;
589 if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
590 flags |= FS_XFLAG_RTINHERIT;
591 if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
592 flags |= FS_XFLAG_PROJINHERIT;
593 if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
594 flags |= FS_XFLAG_NOSYMLINKS;
595 if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
596 flags |= FS_XFLAG_EXTSIZE;
597 if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
598 flags |= FS_XFLAG_EXTSZINHERIT;
599 if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
600 flags |= FS_XFLAG_NODEFRAG;
601 if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
602 flags |= FS_XFLAG_FILESTREAM;
603 }
604
605 if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
606 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
607 flags |= FS_XFLAG_DAX;
608 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
609 flags |= FS_XFLAG_COWEXTSIZE;
610 }
611
612 if (xfs_inode_has_attr_fork(ip))
613 flags |= FS_XFLAG_HASATTR;
614 return flags;
615}
616
617/*
618 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
619 * is allowed, otherwise it has to be an exact match. If a CI match is found,
620 * ci_name->name will point to a the actual name (caller must free) or
621 * will be set to NULL if an exact match is found.
622 */
623int
624xfs_lookup(
625 struct xfs_inode *dp,
626 const struct xfs_name *name,
627 struct xfs_inode **ipp,
628 struct xfs_name *ci_name)
629{
630 xfs_ino_t inum;
631 int error;
632
633 trace_xfs_lookup(dp, name);
634
635 if (xfs_is_shutdown(dp->i_mount))
636 return -EIO;
637 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
638 return -EIO;
639
640 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
641 if (error)
642 goto out_unlock;
643
644 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
645 if (error)
646 goto out_free_name;
647
648 return 0;
649
650out_free_name:
651 if (ci_name)
652 kfree(ci_name->name);
653out_unlock:
654 *ipp = NULL;
655 return error;
656}
657
658/* Propagate di_flags from a parent inode to a child inode. */
659static void
660xfs_inode_inherit_flags(
661 struct xfs_inode *ip,
662 const struct xfs_inode *pip)
663{
664 unsigned int di_flags = 0;
665 xfs_failaddr_t failaddr;
666 umode_t mode = VFS_I(ip)->i_mode;
667
668 if (S_ISDIR(mode)) {
669 if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
670 di_flags |= XFS_DIFLAG_RTINHERIT;
671 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
672 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
673 ip->i_extsize = pip->i_extsize;
674 }
675 if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
676 di_flags |= XFS_DIFLAG_PROJINHERIT;
677 } else if (S_ISREG(mode)) {
678 if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
679 xfs_has_realtime(ip->i_mount))
680 di_flags |= XFS_DIFLAG_REALTIME;
681 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
682 di_flags |= XFS_DIFLAG_EXTSIZE;
683 ip->i_extsize = pip->i_extsize;
684 }
685 }
686 if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
687 xfs_inherit_noatime)
688 di_flags |= XFS_DIFLAG_NOATIME;
689 if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
690 xfs_inherit_nodump)
691 di_flags |= XFS_DIFLAG_NODUMP;
692 if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
693 xfs_inherit_sync)
694 di_flags |= XFS_DIFLAG_SYNC;
695 if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
696 xfs_inherit_nosymlinks)
697 di_flags |= XFS_DIFLAG_NOSYMLINKS;
698 if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
699 xfs_inherit_nodefrag)
700 di_flags |= XFS_DIFLAG_NODEFRAG;
701 if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
702 di_flags |= XFS_DIFLAG_FILESTREAM;
703
704 ip->i_diflags |= di_flags;
705
706 /*
707 * Inode verifiers on older kernels only check that the extent size
708 * hint is an integer multiple of the rt extent size on realtime files.
709 * They did not check the hint alignment on a directory with both
710 * rtinherit and extszinherit flags set. If the misaligned hint is
711 * propagated from a directory into a new realtime file, new file
712 * allocations will fail due to math errors in the rt allocator and/or
713 * trip the verifiers. Validate the hint settings in the new file so
714 * that we don't let broken hints propagate.
715 */
716 failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
717 VFS_I(ip)->i_mode, ip->i_diflags);
718 if (failaddr) {
719 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
720 XFS_DIFLAG_EXTSZINHERIT);
721 ip->i_extsize = 0;
722 }
723}
724
725/* Propagate di_flags2 from a parent inode to a child inode. */
726static void
727xfs_inode_inherit_flags2(
728 struct xfs_inode *ip,
729 const struct xfs_inode *pip)
730{
731 xfs_failaddr_t failaddr;
732
733 if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
734 ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
735 ip->i_cowextsize = pip->i_cowextsize;
736 }
737 if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
738 ip->i_diflags2 |= XFS_DIFLAG2_DAX;
739
740 /* Don't let invalid cowextsize hints propagate. */
741 failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
742 VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
743 if (failaddr) {
744 ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
745 ip->i_cowextsize = 0;
746 }
747}
748
749/*
750 * Initialise a newly allocated inode and return the in-core inode to the
751 * caller locked exclusively.
752 */
753int
754xfs_init_new_inode(
755 struct mnt_idmap *idmap,
756 struct xfs_trans *tp,
757 struct xfs_inode *pip,
758 xfs_ino_t ino,
759 umode_t mode,
760 xfs_nlink_t nlink,
761 dev_t rdev,
762 prid_t prid,
763 bool init_xattrs,
764 struct xfs_inode **ipp)
765{
766 struct inode *dir = pip ? VFS_I(pip) : NULL;
767 struct xfs_mount *mp = tp->t_mountp;
768 struct xfs_inode *ip;
769 unsigned int flags;
770 int error;
771 struct timespec64 tv;
772 struct inode *inode;
773
774 /*
775 * Protect against obviously corrupt allocation btree records. Later
776 * xfs_iget checks will catch re-allocation of other active in-memory
777 * and on-disk inodes. If we don't catch reallocating the parent inode
778 * here we will deadlock in xfs_iget() so we have to do these checks
779 * first.
780 */
781 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
782 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
783 xfs_agno_mark_sick(mp, XFS_INO_TO_AGNO(mp, ino),
784 XFS_SICK_AG_INOBT);
785 return -EFSCORRUPTED;
786 }
787
788 /*
789 * Get the in-core inode with the lock held exclusively to prevent
790 * others from looking at until we're done.
791 */
792 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
793 if (error)
794 return error;
795
796 ASSERT(ip != NULL);
797 inode = VFS_I(ip);
798 set_nlink(inode, nlink);
799 inode->i_rdev = rdev;
800 ip->i_projid = prid;
801
802 if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) {
803 inode_fsuid_set(inode, idmap);
804 inode->i_gid = dir->i_gid;
805 inode->i_mode = mode;
806 } else {
807 inode_init_owner(idmap, inode, dir, mode);
808 }
809
810 /*
811 * If the group ID of the new file does not match the effective group
812 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
813 * (and only if the irix_sgid_inherit compatibility variable is set).
814 */
815 if (irix_sgid_inherit && (inode->i_mode & S_ISGID) &&
816 !vfsgid_in_group_p(i_gid_into_vfsgid(idmap, inode)))
817 inode->i_mode &= ~S_ISGID;
818
819 ip->i_disk_size = 0;
820 ip->i_df.if_nextents = 0;
821 ASSERT(ip->i_nblocks == 0);
822
823 tv = inode_set_ctime_current(inode);
824 inode_set_mtime_to_ts(inode, tv);
825 inode_set_atime_to_ts(inode, tv);
826
827 ip->i_extsize = 0;
828 ip->i_diflags = 0;
829
830 if (xfs_has_v3inodes(mp)) {
831 inode_set_iversion(inode, 1);
832 ip->i_cowextsize = 0;
833 ip->i_crtime = tv;
834 }
835
836 flags = XFS_ILOG_CORE;
837 switch (mode & S_IFMT) {
838 case S_IFIFO:
839 case S_IFCHR:
840 case S_IFBLK:
841 case S_IFSOCK:
842 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
843 flags |= XFS_ILOG_DEV;
844 break;
845 case S_IFREG:
846 case S_IFDIR:
847 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
848 xfs_inode_inherit_flags(ip, pip);
849 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
850 xfs_inode_inherit_flags2(ip, pip);
851 fallthrough;
852 case S_IFLNK:
853 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
854 ip->i_df.if_bytes = 0;
855 ip->i_df.if_data = NULL;
856 break;
857 default:
858 ASSERT(0);
859 }
860
861 /*
862 * If we need to create attributes immediately after allocating the
863 * inode, initialise an empty attribute fork right now. We use the
864 * default fork offset for attributes here as we don't know exactly what
865 * size or how many attributes we might be adding. We can do this
866 * safely here because we know the data fork is completely empty and
867 * this saves us from needing to run a separate transaction to set the
868 * fork offset in the immediate future.
869 */
870 if (init_xattrs && xfs_has_attr(mp)) {
871 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
872 xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
873 }
874
875 /*
876 * Log the new values stuffed into the inode.
877 */
878 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
879 xfs_trans_log_inode(tp, ip, flags);
880
881 /* now that we have an i_mode we can setup the inode structure */
882 xfs_setup_inode(ip);
883
884 *ipp = ip;
885 return 0;
886}
887
888/*
889 * Decrement the link count on an inode & log the change. If this causes the
890 * link count to go to zero, move the inode to AGI unlinked list so that it can
891 * be freed when the last active reference goes away via xfs_inactive().
892 */
893static int /* error */
894xfs_droplink(
895 xfs_trans_t *tp,
896 xfs_inode_t *ip)
897{
898 if (VFS_I(ip)->i_nlink == 0) {
899 xfs_alert(ip->i_mount,
900 "%s: Attempt to drop inode (%llu) with nlink zero.",
901 __func__, ip->i_ino);
902 return -EFSCORRUPTED;
903 }
904
905 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
906
907 drop_nlink(VFS_I(ip));
908 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
909
910 if (VFS_I(ip)->i_nlink)
911 return 0;
912
913 return xfs_iunlink(tp, ip);
914}
915
916/*
917 * Increment the link count on an inode & log the change.
918 */
919static void
920xfs_bumplink(
921 xfs_trans_t *tp,
922 xfs_inode_t *ip)
923{
924 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
925
926 inc_nlink(VFS_I(ip));
927 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
928}
929
930#ifdef CONFIG_XFS_LIVE_HOOKS
931/*
932 * Use a static key here to reduce the overhead of directory live update hooks.
933 * If the compiler supports jump labels, the static branch will be replaced by
934 * a nop sled when there are no hook users. Online fsck is currently the only
935 * caller, so this is a reasonable tradeoff.
936 *
937 * Note: Patching the kernel code requires taking the cpu hotplug lock. Other
938 * parts of the kernel allocate memory with that lock held, which means that
939 * XFS callers cannot hold any locks that might be used by memory reclaim or
940 * writeback when calling the static_branch_{inc,dec} functions.
941 */
942DEFINE_STATIC_XFS_HOOK_SWITCH(xfs_dir_hooks_switch);
943
944void
945xfs_dir_hook_disable(void)
946{
947 xfs_hooks_switch_off(&xfs_dir_hooks_switch);
948}
949
950void
951xfs_dir_hook_enable(void)
952{
953 xfs_hooks_switch_on(&xfs_dir_hooks_switch);
954}
955
956/* Call hooks for a directory update relating to a child dirent update. */
957inline void
958xfs_dir_update_hook(
959 struct xfs_inode *dp,
960 struct xfs_inode *ip,
961 int delta,
962 const struct xfs_name *name)
963{
964 if (xfs_hooks_switched_on(&xfs_dir_hooks_switch)) {
965 struct xfs_dir_update_params p = {
966 .dp = dp,
967 .ip = ip,
968 .delta = delta,
969 .name = name,
970 };
971 struct xfs_mount *mp = ip->i_mount;
972
973 xfs_hooks_call(&mp->m_dir_update_hooks, 0, &p);
974 }
975}
976
977/* Call the specified function during a directory update. */
978int
979xfs_dir_hook_add(
980 struct xfs_mount *mp,
981 struct xfs_dir_hook *hook)
982{
983 return xfs_hooks_add(&mp->m_dir_update_hooks, &hook->dirent_hook);
984}
985
986/* Stop calling the specified function during a directory update. */
987void
988xfs_dir_hook_del(
989 struct xfs_mount *mp,
990 struct xfs_dir_hook *hook)
991{
992 xfs_hooks_del(&mp->m_dir_update_hooks, &hook->dirent_hook);
993}
994
995/* Configure directory update hook functions. */
996void
997xfs_dir_hook_setup(
998 struct xfs_dir_hook *hook,
999 notifier_fn_t mod_fn)
1000{
1001 xfs_hook_setup(&hook->dirent_hook, mod_fn);
1002}
1003#endif /* CONFIG_XFS_LIVE_HOOKS */
1004
1005int
1006xfs_create(
1007 struct mnt_idmap *idmap,
1008 xfs_inode_t *dp,
1009 struct xfs_name *name,
1010 umode_t mode,
1011 dev_t rdev,
1012 bool init_xattrs,
1013 xfs_inode_t **ipp)
1014{
1015 int is_dir = S_ISDIR(mode);
1016 struct xfs_mount *mp = dp->i_mount;
1017 struct xfs_inode *ip = NULL;
1018 struct xfs_trans *tp = NULL;
1019 int error;
1020 bool unlock_dp_on_error = false;
1021 prid_t prid;
1022 struct xfs_dquot *udqp = NULL;
1023 struct xfs_dquot *gdqp = NULL;
1024 struct xfs_dquot *pdqp = NULL;
1025 struct xfs_trans_res *tres;
1026 uint resblks;
1027 xfs_ino_t ino;
1028
1029 trace_xfs_create(dp, name);
1030
1031 if (xfs_is_shutdown(mp))
1032 return -EIO;
1033 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
1034 return -EIO;
1035
1036 prid = xfs_get_initial_prid(dp);
1037
1038 /*
1039 * Make sure that we have allocated dquot(s) on disk.
1040 */
1041 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns),
1042 mapped_fsgid(idmap, &init_user_ns), prid,
1043 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1044 &udqp, &gdqp, &pdqp);
1045 if (error)
1046 return error;
1047
1048 if (is_dir) {
1049 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1050 tres = &M_RES(mp)->tr_mkdir;
1051 } else {
1052 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1053 tres = &M_RES(mp)->tr_create;
1054 }
1055
1056 /*
1057 * Initially assume that the file does not exist and
1058 * reserve the resources for that case. If that is not
1059 * the case we'll drop the one we have and get a more
1060 * appropriate transaction later.
1061 */
1062 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1063 &tp);
1064 if (error == -ENOSPC) {
1065 /* flush outstanding delalloc blocks and retry */
1066 xfs_flush_inodes(mp);
1067 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1068 resblks, &tp);
1069 }
1070 if (error)
1071 goto out_release_dquots;
1072
1073 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1074 unlock_dp_on_error = true;
1075
1076 /*
1077 * A newly created regular or special file just has one directory
1078 * entry pointing to them, but a directory also the "." entry
1079 * pointing to itself.
1080 */
1081 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1082 if (!error)
1083 error = xfs_init_new_inode(idmap, tp, dp, ino, mode,
1084 is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
1085 if (error)
1086 goto out_trans_cancel;
1087
1088 /*
1089 * Now we join the directory inode to the transaction. We do not do it
1090 * earlier because xfs_dialloc might commit the previous transaction
1091 * (and release all the locks). An error from here on will result in
1092 * the transaction cancel unlocking dp so don't do it explicitly in the
1093 * error path.
1094 */
1095 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1096 unlock_dp_on_error = false;
1097
1098 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1099 resblks - XFS_IALLOC_SPACE_RES(mp));
1100 if (error) {
1101 ASSERT(error != -ENOSPC);
1102 goto out_trans_cancel;
1103 }
1104 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1105 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1106
1107 if (is_dir) {
1108 error = xfs_dir_init(tp, ip, dp);
1109 if (error)
1110 goto out_trans_cancel;
1111
1112 xfs_bumplink(tp, dp);
1113 }
1114
1115 /*
1116 * Create ip with a reference from dp, and add '.' and '..' references
1117 * if it's a directory.
1118 */
1119 xfs_dir_update_hook(dp, ip, 1, name);
1120
1121 /*
1122 * If this is a synchronous mount, make sure that the
1123 * create transaction goes to disk before returning to
1124 * the user.
1125 */
1126 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1127 xfs_trans_set_sync(tp);
1128
1129 /*
1130 * Attach the dquot(s) to the inodes and modify them incore.
1131 * These ids of the inode couldn't have changed since the new
1132 * inode has been locked ever since it was created.
1133 */
1134 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1135
1136 error = xfs_trans_commit(tp);
1137 if (error)
1138 goto out_release_inode;
1139
1140 xfs_qm_dqrele(udqp);
1141 xfs_qm_dqrele(gdqp);
1142 xfs_qm_dqrele(pdqp);
1143
1144 *ipp = ip;
1145 return 0;
1146
1147 out_trans_cancel:
1148 xfs_trans_cancel(tp);
1149 out_release_inode:
1150 /*
1151 * Wait until after the current transaction is aborted to finish the
1152 * setup of the inode and release the inode. This prevents recursive
1153 * transactions and deadlocks from xfs_inactive.
1154 */
1155 if (ip) {
1156 xfs_finish_inode_setup(ip);
1157 xfs_irele(ip);
1158 }
1159 out_release_dquots:
1160 xfs_qm_dqrele(udqp);
1161 xfs_qm_dqrele(gdqp);
1162 xfs_qm_dqrele(pdqp);
1163
1164 if (unlock_dp_on_error)
1165 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1166 return error;
1167}
1168
1169int
1170xfs_create_tmpfile(
1171 struct mnt_idmap *idmap,
1172 struct xfs_inode *dp,
1173 umode_t mode,
1174 struct xfs_inode **ipp)
1175{
1176 struct xfs_mount *mp = dp->i_mount;
1177 struct xfs_inode *ip = NULL;
1178 struct xfs_trans *tp = NULL;
1179 int error;
1180 prid_t prid;
1181 struct xfs_dquot *udqp = NULL;
1182 struct xfs_dquot *gdqp = NULL;
1183 struct xfs_dquot *pdqp = NULL;
1184 struct xfs_trans_res *tres;
1185 uint resblks;
1186 xfs_ino_t ino;
1187
1188 if (xfs_is_shutdown(mp))
1189 return -EIO;
1190
1191 prid = xfs_get_initial_prid(dp);
1192
1193 /*
1194 * Make sure that we have allocated dquot(s) on disk.
1195 */
1196 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(idmap, &init_user_ns),
1197 mapped_fsgid(idmap, &init_user_ns), prid,
1198 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1199 &udqp, &gdqp, &pdqp);
1200 if (error)
1201 return error;
1202
1203 resblks = XFS_IALLOC_SPACE_RES(mp);
1204 tres = &M_RES(mp)->tr_create_tmpfile;
1205
1206 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1207 &tp);
1208 if (error)
1209 goto out_release_dquots;
1210
1211 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1212 if (!error)
1213 error = xfs_init_new_inode(idmap, tp, dp, ino, mode,
1214 0, 0, prid, false, &ip);
1215 if (error)
1216 goto out_trans_cancel;
1217
1218 if (xfs_has_wsync(mp))
1219 xfs_trans_set_sync(tp);
1220
1221 /*
1222 * Attach the dquot(s) to the inodes and modify them incore.
1223 * These ids of the inode couldn't have changed since the new
1224 * inode has been locked ever since it was created.
1225 */
1226 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1227
1228 error = xfs_iunlink(tp, ip);
1229 if (error)
1230 goto out_trans_cancel;
1231
1232 error = xfs_trans_commit(tp);
1233 if (error)
1234 goto out_release_inode;
1235
1236 xfs_qm_dqrele(udqp);
1237 xfs_qm_dqrele(gdqp);
1238 xfs_qm_dqrele(pdqp);
1239
1240 *ipp = ip;
1241 return 0;
1242
1243 out_trans_cancel:
1244 xfs_trans_cancel(tp);
1245 out_release_inode:
1246 /*
1247 * Wait until after the current transaction is aborted to finish the
1248 * setup of the inode and release the inode. This prevents recursive
1249 * transactions and deadlocks from xfs_inactive.
1250 */
1251 if (ip) {
1252 xfs_finish_inode_setup(ip);
1253 xfs_irele(ip);
1254 }
1255 out_release_dquots:
1256 xfs_qm_dqrele(udqp);
1257 xfs_qm_dqrele(gdqp);
1258 xfs_qm_dqrele(pdqp);
1259
1260 return error;
1261}
1262
1263int
1264xfs_link(
1265 xfs_inode_t *tdp,
1266 xfs_inode_t *sip,
1267 struct xfs_name *target_name)
1268{
1269 xfs_mount_t *mp = tdp->i_mount;
1270 xfs_trans_t *tp;
1271 int error, nospace_error = 0;
1272 int resblks;
1273
1274 trace_xfs_link(tdp, target_name);
1275
1276 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1277
1278 if (xfs_is_shutdown(mp))
1279 return -EIO;
1280 if (xfs_ifork_zapped(tdp, XFS_DATA_FORK))
1281 return -EIO;
1282
1283 error = xfs_qm_dqattach(sip);
1284 if (error)
1285 goto std_return;
1286
1287 error = xfs_qm_dqattach(tdp);
1288 if (error)
1289 goto std_return;
1290
1291 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1292 error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,
1293 &tp, &nospace_error);
1294 if (error)
1295 goto std_return;
1296
1297 /*
1298 * If we are using project inheritance, we only allow hard link
1299 * creation in our tree when the project IDs are the same; else
1300 * the tree quota mechanism could be circumvented.
1301 */
1302 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1303 tdp->i_projid != sip->i_projid)) {
1304 /*
1305 * Project quota setup skips special files which can
1306 * leave inodes in a PROJINHERIT directory without a
1307 * project ID set. We need to allow links to be made
1308 * to these "project-less" inodes because userspace
1309 * expects them to succeed after project ID setup,
1310 * but everything else should be rejected.
1311 */
1312 if (!special_file(VFS_I(sip)->i_mode) ||
1313 sip->i_projid != 0) {
1314 error = -EXDEV;
1315 goto error_return;
1316 }
1317 }
1318
1319 if (!resblks) {
1320 error = xfs_dir_canenter(tp, tdp, target_name);
1321 if (error)
1322 goto error_return;
1323 }
1324
1325 /*
1326 * Handle initial link state of O_TMPFILE inode
1327 */
1328 if (VFS_I(sip)->i_nlink == 0) {
1329 struct xfs_perag *pag;
1330
1331 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
1332 error = xfs_iunlink_remove(tp, pag, sip);
1333 xfs_perag_put(pag);
1334 if (error)
1335 goto error_return;
1336 }
1337
1338 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1339 resblks);
1340 if (error)
1341 goto error_return;
1342 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1343 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1344
1345 xfs_bumplink(tp, sip);
1346 xfs_dir_update_hook(tdp, sip, 1, target_name);
1347
1348 /*
1349 * If this is a synchronous mount, make sure that the
1350 * link transaction goes to disk before returning to
1351 * the user.
1352 */
1353 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1354 xfs_trans_set_sync(tp);
1355
1356 return xfs_trans_commit(tp);
1357
1358 error_return:
1359 xfs_trans_cancel(tp);
1360 std_return:
1361 if (error == -ENOSPC && nospace_error)
1362 error = nospace_error;
1363 return error;
1364}
1365
1366/* Clear the reflink flag and the cowblocks tag if possible. */
1367static void
1368xfs_itruncate_clear_reflink_flags(
1369 struct xfs_inode *ip)
1370{
1371 struct xfs_ifork *dfork;
1372 struct xfs_ifork *cfork;
1373
1374 if (!xfs_is_reflink_inode(ip))
1375 return;
1376 dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);
1377 cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);
1378 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1379 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1380 if (cfork->if_bytes == 0)
1381 xfs_inode_clear_cowblocks_tag(ip);
1382}
1383
1384/*
1385 * Free up the underlying blocks past new_size. The new size must be smaller
1386 * than the current size. This routine can be used both for the attribute and
1387 * data fork, and does not modify the inode size, which is left to the caller.
1388 *
1389 * The transaction passed to this routine must have made a permanent log
1390 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1391 * given transaction and start new ones, so make sure everything involved in
1392 * the transaction is tidy before calling here. Some transaction will be
1393 * returned to the caller to be committed. The incoming transaction must
1394 * already include the inode, and both inode locks must be held exclusively.
1395 * The inode must also be "held" within the transaction. On return the inode
1396 * will be "held" within the returned transaction. This routine does NOT
1397 * require any disk space to be reserved for it within the transaction.
1398 *
1399 * If we get an error, we must return with the inode locked and linked into the
1400 * current transaction. This keeps things simple for the higher level code,
1401 * because it always knows that the inode is locked and held in the transaction
1402 * that returns to it whether errors occur or not. We don't mark the inode
1403 * dirty on error so that transactions can be easily aborted if possible.
1404 */
1405int
1406xfs_itruncate_extents_flags(
1407 struct xfs_trans **tpp,
1408 struct xfs_inode *ip,
1409 int whichfork,
1410 xfs_fsize_t new_size,
1411 int flags)
1412{
1413 struct xfs_mount *mp = ip->i_mount;
1414 struct xfs_trans *tp = *tpp;
1415 xfs_fileoff_t first_unmap_block;
1416 int error = 0;
1417
1418 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1419 if (atomic_read(&VFS_I(ip)->i_count))
1420 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);
1421 ASSERT(new_size <= XFS_ISIZE(ip));
1422 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1423 ASSERT(ip->i_itemp != NULL);
1424 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1425 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1426
1427 trace_xfs_itruncate_extents_start(ip, new_size);
1428
1429 flags |= xfs_bmapi_aflag(whichfork);
1430
1431 /*
1432 * Since it is possible for space to become allocated beyond
1433 * the end of the file (in a crash where the space is allocated
1434 * but the inode size is not yet updated), simply remove any
1435 * blocks which show up between the new EOF and the maximum
1436 * possible file size.
1437 *
1438 * We have to free all the blocks to the bmbt maximum offset, even if
1439 * the page cache can't scale that far.
1440 */
1441 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1442 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1443 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1444 return 0;
1445 }
1446
1447 error = xfs_bunmapi_range(&tp, ip, flags, first_unmap_block,
1448 XFS_MAX_FILEOFF);
1449 if (error)
1450 goto out;
1451
1452 if (whichfork == XFS_DATA_FORK) {
1453 /* Remove all pending CoW reservations. */
1454 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1455 first_unmap_block, XFS_MAX_FILEOFF, true);
1456 if (error)
1457 goto out;
1458
1459 xfs_itruncate_clear_reflink_flags(ip);
1460 }
1461
1462 /*
1463 * Always re-log the inode so that our permanent transaction can keep
1464 * on rolling it forward in the log.
1465 */
1466 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1467
1468 trace_xfs_itruncate_extents_end(ip, new_size);
1469
1470out:
1471 *tpp = tp;
1472 return error;
1473}
1474
1475int
1476xfs_release(
1477 xfs_inode_t *ip)
1478{
1479 xfs_mount_t *mp = ip->i_mount;
1480 int error = 0;
1481
1482 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1483 return 0;
1484
1485 /* If this is a read-only mount, don't do this (would generate I/O) */
1486 if (xfs_is_readonly(mp))
1487 return 0;
1488
1489 if (!xfs_is_shutdown(mp)) {
1490 int truncated;
1491
1492 /*
1493 * If we previously truncated this file and removed old data
1494 * in the process, we want to initiate "early" writeout on
1495 * the last close. This is an attempt to combat the notorious
1496 * NULL files problem which is particularly noticeable from a
1497 * truncate down, buffered (re-)write (delalloc), followed by
1498 * a crash. What we are effectively doing here is
1499 * significantly reducing the time window where we'd otherwise
1500 * be exposed to that problem.
1501 */
1502 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1503 if (truncated) {
1504 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1505 if (ip->i_delayed_blks > 0) {
1506 error = filemap_flush(VFS_I(ip)->i_mapping);
1507 if (error)
1508 return error;
1509 }
1510 }
1511 }
1512
1513 if (VFS_I(ip)->i_nlink == 0)
1514 return 0;
1515
1516 /*
1517 * If we can't get the iolock just skip truncating the blocks past EOF
1518 * because we could deadlock with the mmap_lock otherwise. We'll get
1519 * another chance to drop them once the last reference to the inode is
1520 * dropped, so we'll never leak blocks permanently.
1521 */
1522 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
1523 return 0;
1524
1525 if (xfs_can_free_eofblocks(ip, false)) {
1526 /*
1527 * Check if the inode is being opened, written and closed
1528 * frequently and we have delayed allocation blocks outstanding
1529 * (e.g. streaming writes from the NFS server), truncating the
1530 * blocks past EOF will cause fragmentation to occur.
1531 *
1532 * In this case don't do the truncation, but we have to be
1533 * careful how we detect this case. Blocks beyond EOF show up as
1534 * i_delayed_blks even when the inode is clean, so we need to
1535 * truncate them away first before checking for a dirty release.
1536 * Hence on the first dirty close we will still remove the
1537 * speculative allocation, but after that we will leave it in
1538 * place.
1539 */
1540 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1541 goto out_unlock;
1542
1543 error = xfs_free_eofblocks(ip);
1544 if (error)
1545 goto out_unlock;
1546
1547 /* delalloc blocks after truncation means it really is dirty */
1548 if (ip->i_delayed_blks)
1549 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1550 }
1551
1552out_unlock:
1553 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1554 return error;
1555}
1556
1557/*
1558 * xfs_inactive_truncate
1559 *
1560 * Called to perform a truncate when an inode becomes unlinked.
1561 */
1562STATIC int
1563xfs_inactive_truncate(
1564 struct xfs_inode *ip)
1565{
1566 struct xfs_mount *mp = ip->i_mount;
1567 struct xfs_trans *tp;
1568 int error;
1569
1570 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1571 if (error) {
1572 ASSERT(xfs_is_shutdown(mp));
1573 return error;
1574 }
1575 xfs_ilock(ip, XFS_ILOCK_EXCL);
1576 xfs_trans_ijoin(tp, ip, 0);
1577
1578 /*
1579 * Log the inode size first to prevent stale data exposure in the event
1580 * of a system crash before the truncate completes. See the related
1581 * comment in xfs_vn_setattr_size() for details.
1582 */
1583 ip->i_disk_size = 0;
1584 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1585
1586 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1587 if (error)
1588 goto error_trans_cancel;
1589
1590 ASSERT(ip->i_df.if_nextents == 0);
1591
1592 error = xfs_trans_commit(tp);
1593 if (error)
1594 goto error_unlock;
1595
1596 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1597 return 0;
1598
1599error_trans_cancel:
1600 xfs_trans_cancel(tp);
1601error_unlock:
1602 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1603 return error;
1604}
1605
1606/*
1607 * xfs_inactive_ifree()
1608 *
1609 * Perform the inode free when an inode is unlinked.
1610 */
1611STATIC int
1612xfs_inactive_ifree(
1613 struct xfs_inode *ip)
1614{
1615 struct xfs_mount *mp = ip->i_mount;
1616 struct xfs_trans *tp;
1617 int error;
1618
1619 /*
1620 * We try to use a per-AG reservation for any block needed by the finobt
1621 * tree, but as the finobt feature predates the per-AG reservation
1622 * support a degraded file system might not have enough space for the
1623 * reservation at mount time. In that case try to dip into the reserved
1624 * pool and pray.
1625 *
1626 * Send a warning if the reservation does happen to fail, as the inode
1627 * now remains allocated and sits on the unlinked list until the fs is
1628 * repaired.
1629 */
1630 if (unlikely(mp->m_finobt_nores)) {
1631 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1632 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1633 &tp);
1634 } else {
1635 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1636 }
1637 if (error) {
1638 if (error == -ENOSPC) {
1639 xfs_warn_ratelimited(mp,
1640 "Failed to remove inode(s) from unlinked list. "
1641 "Please free space, unmount and run xfs_repair.");
1642 } else {
1643 ASSERT(xfs_is_shutdown(mp));
1644 }
1645 return error;
1646 }
1647
1648 /*
1649 * We do not hold the inode locked across the entire rolling transaction
1650 * here. We only need to hold it for the first transaction that
1651 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1652 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1653 * here breaks the relationship between cluster buffer invalidation and
1654 * stale inode invalidation on cluster buffer item journal commit
1655 * completion, and can result in leaving dirty stale inodes hanging
1656 * around in memory.
1657 *
1658 * We have no need for serialising this inode operation against other
1659 * operations - we freed the inode and hence reallocation is required
1660 * and that will serialise on reallocating the space the deferops need
1661 * to free. Hence we can unlock the inode on the first commit of
1662 * the transaction rather than roll it right through the deferops. This
1663 * avoids relogging the XFS_ISTALE inode.
1664 *
1665 * We check that xfs_ifree() hasn't grown an internal transaction roll
1666 * by asserting that the inode is still locked when it returns.
1667 */
1668 xfs_ilock(ip, XFS_ILOCK_EXCL);
1669 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1670
1671 error = xfs_ifree(tp, ip);
1672 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
1673 if (error) {
1674 /*
1675 * If we fail to free the inode, shut down. The cancel
1676 * might do that, we need to make sure. Otherwise the
1677 * inode might be lost for a long time or forever.
1678 */
1679 if (!xfs_is_shutdown(mp)) {
1680 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1681 __func__, error);
1682 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1683 }
1684 xfs_trans_cancel(tp);
1685 return error;
1686 }
1687
1688 /*
1689 * Credit the quota account(s). The inode is gone.
1690 */
1691 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1692
1693 return xfs_trans_commit(tp);
1694}
1695
1696/*
1697 * Returns true if we need to update the on-disk metadata before we can free
1698 * the memory used by this inode. Updates include freeing post-eof
1699 * preallocations; freeing COW staging extents; and marking the inode free in
1700 * the inobt if it is on the unlinked list.
1701 */
1702bool
1703xfs_inode_needs_inactive(
1704 struct xfs_inode *ip)
1705{
1706 struct xfs_mount *mp = ip->i_mount;
1707 struct xfs_ifork *cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
1708
1709 /*
1710 * If the inode is already free, then there can be nothing
1711 * to clean up here.
1712 */
1713 if (VFS_I(ip)->i_mode == 0)
1714 return false;
1715
1716 /*
1717 * If this is a read-only mount, don't do this (would generate I/O)
1718 * unless we're in log recovery and cleaning the iunlinked list.
1719 */
1720 if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1721 return false;
1722
1723 /* If the log isn't running, push inodes straight to reclaim. */
1724 if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1725 return false;
1726
1727 /* Metadata inodes require explicit resource cleanup. */
1728 if (xfs_is_metadata_inode(ip))
1729 return false;
1730
1731 /* Want to clean out the cow blocks if there are any. */
1732 if (cow_ifp && cow_ifp->if_bytes > 0)
1733 return true;
1734
1735 /* Unlinked files must be freed. */
1736 if (VFS_I(ip)->i_nlink == 0)
1737 return true;
1738
1739 /*
1740 * This file isn't being freed, so check if there are post-eof blocks
1741 * to free. @force is true because we are evicting an inode from the
1742 * cache. Post-eof blocks must be freed, lest we end up with broken
1743 * free space accounting.
1744 *
1745 * Note: don't bother with iolock here since lockdep complains about
1746 * acquiring it in reclaim context. We have the only reference to the
1747 * inode at this point anyways.
1748 */
1749 return xfs_can_free_eofblocks(ip, true);
1750}
1751
1752/*
1753 * Save health status somewhere, if we're dumping an inode with uncorrected
1754 * errors and online repair isn't running.
1755 */
1756static inline void
1757xfs_inactive_health(
1758 struct xfs_inode *ip)
1759{
1760 struct xfs_mount *mp = ip->i_mount;
1761 struct xfs_perag *pag;
1762 unsigned int sick;
1763 unsigned int checked;
1764
1765 xfs_inode_measure_sickness(ip, &sick, &checked);
1766 if (!sick)
1767 return;
1768
1769 trace_xfs_inode_unfixed_corruption(ip, sick);
1770
1771 if (sick & XFS_SICK_INO_FORGET)
1772 return;
1773
1774 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1775 if (!pag) {
1776 /* There had better still be a perag structure! */
1777 ASSERT(0);
1778 return;
1779 }
1780
1781 xfs_ag_mark_sick(pag, XFS_SICK_AG_INODES);
1782 xfs_perag_put(pag);
1783}
1784
1785/*
1786 * xfs_inactive
1787 *
1788 * This is called when the vnode reference count for the vnode
1789 * goes to zero. If the file has been unlinked, then it must
1790 * now be truncated. Also, we clear all of the read-ahead state
1791 * kept for the inode here since the file is now closed.
1792 */
1793int
1794xfs_inactive(
1795 xfs_inode_t *ip)
1796{
1797 struct xfs_mount *mp;
1798 int error = 0;
1799 int truncate = 0;
1800
1801 /*
1802 * If the inode is already free, then there can be nothing
1803 * to clean up here.
1804 */
1805 if (VFS_I(ip)->i_mode == 0) {
1806 ASSERT(ip->i_df.if_broot_bytes == 0);
1807 goto out;
1808 }
1809
1810 mp = ip->i_mount;
1811 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1812
1813 xfs_inactive_health(ip);
1814
1815 /*
1816 * If this is a read-only mount, don't do this (would generate I/O)
1817 * unless we're in log recovery and cleaning the iunlinked list.
1818 */
1819 if (xfs_is_readonly(mp) && !xlog_recovery_needed(mp->m_log))
1820 goto out;
1821
1822 /* Metadata inodes require explicit resource cleanup. */
1823 if (xfs_is_metadata_inode(ip))
1824 goto out;
1825
1826 /* Try to clean out the cow blocks if there are any. */
1827 if (xfs_inode_has_cow_data(ip))
1828 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1829
1830 if (VFS_I(ip)->i_nlink != 0) {
1831 /*
1832 * force is true because we are evicting an inode from the
1833 * cache. Post-eof blocks must be freed, lest we end up with
1834 * broken free space accounting.
1835 *
1836 * Note: don't bother with iolock here since lockdep complains
1837 * about acquiring it in reclaim context. We have the only
1838 * reference to the inode at this point anyways.
1839 */
1840 if (xfs_can_free_eofblocks(ip, true))
1841 error = xfs_free_eofblocks(ip);
1842
1843 goto out;
1844 }
1845
1846 if (S_ISREG(VFS_I(ip)->i_mode) &&
1847 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1848 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1849 truncate = 1;
1850
1851 if (xfs_iflags_test(ip, XFS_IQUOTAUNCHECKED)) {
1852 /*
1853 * If this inode is being inactivated during a quotacheck and
1854 * has not yet been scanned by quotacheck, we /must/ remove
1855 * the dquots from the inode before inactivation changes the
1856 * block and inode counts. Most probably this is a result of
1857 * reloading the incore iunlinked list to purge unrecovered
1858 * unlinked inodes.
1859 */
1860 xfs_qm_dqdetach(ip);
1861 } else {
1862 error = xfs_qm_dqattach(ip);
1863 if (error)
1864 goto out;
1865 }
1866
1867 if (S_ISLNK(VFS_I(ip)->i_mode))
1868 error = xfs_inactive_symlink(ip);
1869 else if (truncate)
1870 error = xfs_inactive_truncate(ip);
1871 if (error)
1872 goto out;
1873
1874 /*
1875 * If there are attributes associated with the file then blow them away
1876 * now. The code calls a routine that recursively deconstructs the
1877 * attribute fork. If also blows away the in-core attribute fork.
1878 */
1879 if (xfs_inode_has_attr_fork(ip)) {
1880 error = xfs_attr_inactive(ip);
1881 if (error)
1882 goto out;
1883 }
1884
1885 ASSERT(ip->i_forkoff == 0);
1886
1887 /*
1888 * Free the inode.
1889 */
1890 error = xfs_inactive_ifree(ip);
1891
1892out:
1893 /*
1894 * We're done making metadata updates for this inode, so we can release
1895 * the attached dquots.
1896 */
1897 xfs_qm_dqdetach(ip);
1898 return error;
1899}
1900
1901/*
1902 * In-Core Unlinked List Lookups
1903 * =============================
1904 *
1905 * Every inode is supposed to be reachable from some other piece of metadata
1906 * with the exception of the root directory. Inodes with a connection to a
1907 * file descriptor but not linked from anywhere in the on-disk directory tree
1908 * are collectively known as unlinked inodes, though the filesystem itself
1909 * maintains links to these inodes so that on-disk metadata are consistent.
1910 *
1911 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1912 * header contains a number of buckets that point to an inode, and each inode
1913 * record has a pointer to the next inode in the hash chain. This
1914 * singly-linked list causes scaling problems in the iunlink remove function
1915 * because we must walk that list to find the inode that points to the inode
1916 * being removed from the unlinked hash bucket list.
1917 *
1918 * Hence we keep an in-memory double linked list to link each inode on an
1919 * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer
1920 * based lists would require having 64 list heads in the perag, one for each
1921 * list. This is expensive in terms of memory (think millions of AGs) and cache
1922 * misses on lookups. Instead, use the fact that inodes on the unlinked list
1923 * must be referenced at the VFS level to keep them on the list and hence we
1924 * have an existence guarantee for inodes on the unlinked list.
1925 *
1926 * Given we have an existence guarantee, we can use lockless inode cache lookups
1927 * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode
1928 * for the double linked unlinked list, and we don't need any extra locking to
1929 * keep the list safe as all manipulations are done under the AGI buffer lock.
1930 * Keeping the list up to date does not require memory allocation, just finding
1931 * the XFS inode and updating the next/prev unlinked list aginos.
1932 */
1933
1934/*
1935 * Find an inode on the unlinked list. This does not take references to the
1936 * inode as we have existence guarantees by holding the AGI buffer lock and that
1937 * only unlinked, referenced inodes can be on the unlinked inode list. If we
1938 * don't find the inode in cache, then let the caller handle the situation.
1939 */
1940static struct xfs_inode *
1941xfs_iunlink_lookup(
1942 struct xfs_perag *pag,
1943 xfs_agino_t agino)
1944{
1945 struct xfs_inode *ip;
1946
1947 rcu_read_lock();
1948 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
1949 if (!ip) {
1950 /* Caller can handle inode not being in memory. */
1951 rcu_read_unlock();
1952 return NULL;
1953 }
1954
1955 /*
1956 * Inode in RCU freeing limbo should not happen. Warn about this and
1957 * let the caller handle the failure.
1958 */
1959 if (WARN_ON_ONCE(!ip->i_ino)) {
1960 rcu_read_unlock();
1961 return NULL;
1962 }
1963 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));
1964 rcu_read_unlock();
1965 return ip;
1966}
1967
1968/*
1969 * Update the prev pointer of the next agino. Returns -ENOLINK if the inode
1970 * is not in cache.
1971 */
1972static int
1973xfs_iunlink_update_backref(
1974 struct xfs_perag *pag,
1975 xfs_agino_t prev_agino,
1976 xfs_agino_t next_agino)
1977{
1978 struct xfs_inode *ip;
1979
1980 /* No update necessary if we are at the end of the list. */
1981 if (next_agino == NULLAGINO)
1982 return 0;
1983
1984 ip = xfs_iunlink_lookup(pag, next_agino);
1985 if (!ip)
1986 return -ENOLINK;
1987
1988 ip->i_prev_unlinked = prev_agino;
1989 return 0;
1990}
1991
1992/*
1993 * Point the AGI unlinked bucket at an inode and log the results. The caller
1994 * is responsible for validating the old value.
1995 */
1996STATIC int
1997xfs_iunlink_update_bucket(
1998 struct xfs_trans *tp,
1999 struct xfs_perag *pag,
2000 struct xfs_buf *agibp,
2001 unsigned int bucket_index,
2002 xfs_agino_t new_agino)
2003{
2004 struct xfs_agi *agi = agibp->b_addr;
2005 xfs_agino_t old_value;
2006 int offset;
2007
2008 ASSERT(xfs_verify_agino_or_null(pag, new_agino));
2009
2010 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2011 trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
2012 old_value, new_agino);
2013
2014 /*
2015 * We should never find the head of the list already set to the value
2016 * passed in because either we're adding or removing ourselves from the
2017 * head of the list.
2018 */
2019 if (old_value == new_agino) {
2020 xfs_buf_mark_corrupt(agibp);
2021 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
2022 return -EFSCORRUPTED;
2023 }
2024
2025 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
2026 offset = offsetof(struct xfs_agi, agi_unlinked) +
2027 (sizeof(xfs_agino_t) * bucket_index);
2028 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
2029 return 0;
2030}
2031
2032/*
2033 * Load the inode @next_agino into the cache and set its prev_unlinked pointer
2034 * to @prev_agino. Caller must hold the AGI to synchronize with other changes
2035 * to the unlinked list.
2036 */
2037STATIC int
2038xfs_iunlink_reload_next(
2039 struct xfs_trans *tp,
2040 struct xfs_buf *agibp,
2041 xfs_agino_t prev_agino,
2042 xfs_agino_t next_agino)
2043{
2044 struct xfs_perag *pag = agibp->b_pag;
2045 struct xfs_mount *mp = pag->pag_mount;
2046 struct xfs_inode *next_ip = NULL;
2047 xfs_ino_t ino;
2048 int error;
2049
2050 ASSERT(next_agino != NULLAGINO);
2051
2052#ifdef DEBUG
2053 rcu_read_lock();
2054 next_ip = radix_tree_lookup(&pag->pag_ici_root, next_agino);
2055 ASSERT(next_ip == NULL);
2056 rcu_read_unlock();
2057#endif
2058
2059 xfs_info_ratelimited(mp,
2060 "Found unrecovered unlinked inode 0x%x in AG 0x%x. Initiating recovery.",
2061 next_agino, pag->pag_agno);
2062
2063 /*
2064 * Use an untrusted lookup just to be cautious in case the AGI has been
2065 * corrupted and now points at a free inode. That shouldn't happen,
2066 * but we'd rather shut down now since we're already running in a weird
2067 * situation.
2068 */
2069 ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, next_agino);
2070 error = xfs_iget(mp, tp, ino, XFS_IGET_UNTRUSTED, 0, &next_ip);
2071 if (error) {
2072 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
2073 return error;
2074 }
2075
2076 /* If this is not an unlinked inode, something is very wrong. */
2077 if (VFS_I(next_ip)->i_nlink != 0) {
2078 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
2079 error = -EFSCORRUPTED;
2080 goto rele;
2081 }
2082
2083 next_ip->i_prev_unlinked = prev_agino;
2084 trace_xfs_iunlink_reload_next(next_ip);
2085rele:
2086 ASSERT(!(VFS_I(next_ip)->i_state & I_DONTCACHE));
2087 if (xfs_is_quotacheck_running(mp) && next_ip)
2088 xfs_iflags_set(next_ip, XFS_IQUOTAUNCHECKED);
2089 xfs_irele(next_ip);
2090 return error;
2091}
2092
2093static int
2094xfs_iunlink_insert_inode(
2095 struct xfs_trans *tp,
2096 struct xfs_perag *pag,
2097 struct xfs_buf *agibp,
2098 struct xfs_inode *ip)
2099{
2100 struct xfs_mount *mp = tp->t_mountp;
2101 struct xfs_agi *agi = agibp->b_addr;
2102 xfs_agino_t next_agino;
2103 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2104 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2105 int error;
2106
2107 /*
2108 * Get the index into the agi hash table for the list this inode will
2109 * go on. Make sure the pointer isn't garbage and that this inode
2110 * isn't already on the list.
2111 */
2112 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2113 if (next_agino == agino ||
2114 !xfs_verify_agino_or_null(pag, next_agino)) {
2115 xfs_buf_mark_corrupt(agibp);
2116 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
2117 return -EFSCORRUPTED;
2118 }
2119
2120 /*
2121 * Update the prev pointer in the next inode to point back to this
2122 * inode.
2123 */
2124 error = xfs_iunlink_update_backref(pag, agino, next_agino);
2125 if (error == -ENOLINK)
2126 error = xfs_iunlink_reload_next(tp, agibp, agino, next_agino);
2127 if (error)
2128 return error;
2129
2130 if (next_agino != NULLAGINO) {
2131 /*
2132 * There is already another inode in the bucket, so point this
2133 * inode to the current head of the list.
2134 */
2135 error = xfs_iunlink_log_inode(tp, ip, pag, next_agino);
2136 if (error)
2137 return error;
2138 ip->i_next_unlinked = next_agino;
2139 }
2140
2141 /* Point the head of the list to point to this inode. */
2142 ip->i_prev_unlinked = NULLAGINO;
2143 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
2144}
2145
2146/*
2147 * This is called when the inode's link count has gone to 0 or we are creating
2148 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
2149 *
2150 * We place the on-disk inode on a list in the AGI. It will be pulled from this
2151 * list when the inode is freed.
2152 */
2153STATIC int
2154xfs_iunlink(
2155 struct xfs_trans *tp,
2156 struct xfs_inode *ip)
2157{
2158 struct xfs_mount *mp = tp->t_mountp;
2159 struct xfs_perag *pag;
2160 struct xfs_buf *agibp;
2161 int error;
2162
2163 ASSERT(VFS_I(ip)->i_nlink == 0);
2164 ASSERT(VFS_I(ip)->i_mode != 0);
2165 trace_xfs_iunlink(ip);
2166
2167 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2168
2169 /* Get the agi buffer first. It ensures lock ordering on the list. */
2170 error = xfs_read_agi(pag, tp, &agibp);
2171 if (error)
2172 goto out;
2173
2174 error = xfs_iunlink_insert_inode(tp, pag, agibp, ip);
2175out:
2176 xfs_perag_put(pag);
2177 return error;
2178}
2179
2180static int
2181xfs_iunlink_remove_inode(
2182 struct xfs_trans *tp,
2183 struct xfs_perag *pag,
2184 struct xfs_buf *agibp,
2185 struct xfs_inode *ip)
2186{
2187 struct xfs_mount *mp = tp->t_mountp;
2188 struct xfs_agi *agi = agibp->b_addr;
2189 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2190 xfs_agino_t head_agino;
2191 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2192 int error;
2193
2194 trace_xfs_iunlink_remove(ip);
2195
2196 /*
2197 * Get the index into the agi hash table for the list this inode will
2198 * go on. Make sure the head pointer isn't garbage.
2199 */
2200 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2201 if (!xfs_verify_agino(pag, head_agino)) {
2202 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2203 agi, sizeof(*agi));
2204 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
2205 return -EFSCORRUPTED;
2206 }
2207
2208 /*
2209 * Set our inode's next_unlinked pointer to NULL and then return
2210 * the old pointer value so that we can update whatever was previous
2211 * to us in the list to point to whatever was next in the list.
2212 */
2213 error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO);
2214 if (error)
2215 return error;
2216
2217 /*
2218 * Update the prev pointer in the next inode to point back to previous
2219 * inode in the chain.
2220 */
2221 error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked,
2222 ip->i_next_unlinked);
2223 if (error == -ENOLINK)
2224 error = xfs_iunlink_reload_next(tp, agibp, ip->i_prev_unlinked,
2225 ip->i_next_unlinked);
2226 if (error)
2227 return error;
2228
2229 if (head_agino != agino) {
2230 struct xfs_inode *prev_ip;
2231
2232 prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked);
2233 if (!prev_ip) {
2234 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
2235 return -EFSCORRUPTED;
2236 }
2237
2238 error = xfs_iunlink_log_inode(tp, prev_ip, pag,
2239 ip->i_next_unlinked);
2240 prev_ip->i_next_unlinked = ip->i_next_unlinked;
2241 } else {
2242 /* Point the head of the list to the next unlinked inode. */
2243 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
2244 ip->i_next_unlinked);
2245 }
2246
2247 ip->i_next_unlinked = NULLAGINO;
2248 ip->i_prev_unlinked = 0;
2249 return error;
2250}
2251
2252/*
2253 * Pull the on-disk inode from the AGI unlinked list.
2254 */
2255STATIC int
2256xfs_iunlink_remove(
2257 struct xfs_trans *tp,
2258 struct xfs_perag *pag,
2259 struct xfs_inode *ip)
2260{
2261 struct xfs_buf *agibp;
2262 int error;
2263
2264 trace_xfs_iunlink_remove(ip);
2265
2266 /* Get the agi buffer first. It ensures lock ordering on the list. */
2267 error = xfs_read_agi(pag, tp, &agibp);
2268 if (error)
2269 return error;
2270
2271 return xfs_iunlink_remove_inode(tp, pag, agibp, ip);
2272}
2273
2274/*
2275 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2276 * mark it stale. We should only find clean inodes in this lookup that aren't
2277 * already stale.
2278 */
2279static void
2280xfs_ifree_mark_inode_stale(
2281 struct xfs_perag *pag,
2282 struct xfs_inode *free_ip,
2283 xfs_ino_t inum)
2284{
2285 struct xfs_mount *mp = pag->pag_mount;
2286 struct xfs_inode_log_item *iip;
2287 struct xfs_inode *ip;
2288
2289retry:
2290 rcu_read_lock();
2291 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2292
2293 /* Inode not in memory, nothing to do */
2294 if (!ip) {
2295 rcu_read_unlock();
2296 return;
2297 }
2298
2299 /*
2300 * because this is an RCU protected lookup, we could find a recently
2301 * freed or even reallocated inode during the lookup. We need to check
2302 * under the i_flags_lock for a valid inode here. Skip it if it is not
2303 * valid, the wrong inode or stale.
2304 */
2305 spin_lock(&ip->i_flags_lock);
2306 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2307 goto out_iflags_unlock;
2308
2309 /*
2310 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2311 * other inodes that we did not find in the list attached to the buffer
2312 * and are not already marked stale. If we can't lock it, back off and
2313 * retry.
2314 */
2315 if (ip != free_ip) {
2316 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2317 spin_unlock(&ip->i_flags_lock);
2318 rcu_read_unlock();
2319 delay(1);
2320 goto retry;
2321 }
2322 }
2323 ip->i_flags |= XFS_ISTALE;
2324
2325 /*
2326 * If the inode is flushing, it is already attached to the buffer. All
2327 * we needed to do here is mark the inode stale so buffer IO completion
2328 * will remove it from the AIL.
2329 */
2330 iip = ip->i_itemp;
2331 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2332 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2333 ASSERT(iip->ili_last_fields);
2334 goto out_iunlock;
2335 }
2336
2337 /*
2338 * Inodes not attached to the buffer can be released immediately.
2339 * Everything else has to go through xfs_iflush_abort() on journal
2340 * commit as the flock synchronises removal of the inode from the
2341 * cluster buffer against inode reclaim.
2342 */
2343 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2344 goto out_iunlock;
2345
2346 __xfs_iflags_set(ip, XFS_IFLUSHING);
2347 spin_unlock(&ip->i_flags_lock);
2348 rcu_read_unlock();
2349
2350 /* we have a dirty inode in memory that has not yet been flushed. */
2351 spin_lock(&iip->ili_lock);
2352 iip->ili_last_fields = iip->ili_fields;
2353 iip->ili_fields = 0;
2354 iip->ili_fsync_fields = 0;
2355 spin_unlock(&iip->ili_lock);
2356 ASSERT(iip->ili_last_fields);
2357
2358 if (ip != free_ip)
2359 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2360 return;
2361
2362out_iunlock:
2363 if (ip != free_ip)
2364 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2365out_iflags_unlock:
2366 spin_unlock(&ip->i_flags_lock);
2367 rcu_read_unlock();
2368}
2369
2370/*
2371 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2372 * inodes that are in memory - they all must be marked stale and attached to
2373 * the cluster buffer.
2374 */
2375static int
2376xfs_ifree_cluster(
2377 struct xfs_trans *tp,
2378 struct xfs_perag *pag,
2379 struct xfs_inode *free_ip,
2380 struct xfs_icluster *xic)
2381{
2382 struct xfs_mount *mp = free_ip->i_mount;
2383 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2384 struct xfs_buf *bp;
2385 xfs_daddr_t blkno;
2386 xfs_ino_t inum = xic->first_ino;
2387 int nbufs;
2388 int i, j;
2389 int ioffset;
2390 int error;
2391
2392 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2393
2394 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2395 /*
2396 * The allocation bitmap tells us which inodes of the chunk were
2397 * physically allocated. Skip the cluster if an inode falls into
2398 * a sparse region.
2399 */
2400 ioffset = inum - xic->first_ino;
2401 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2402 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2403 continue;
2404 }
2405
2406 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2407 XFS_INO_TO_AGBNO(mp, inum));
2408
2409 /*
2410 * We obtain and lock the backing buffer first in the process
2411 * here to ensure dirty inodes attached to the buffer remain in
2412 * the flushing state while we mark them stale.
2413 *
2414 * If we scan the in-memory inodes first, then buffer IO can
2415 * complete before we get a lock on it, and hence we may fail
2416 * to mark all the active inodes on the buffer stale.
2417 */
2418 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2419 mp->m_bsize * igeo->blocks_per_cluster,
2420 XBF_UNMAPPED, &bp);
2421 if (error)
2422 return error;
2423
2424 /*
2425 * This buffer may not have been correctly initialised as we
2426 * didn't read it from disk. That's not important because we are
2427 * only using to mark the buffer as stale in the log, and to
2428 * attach stale cached inodes on it. That means it will never be
2429 * dispatched for IO. If it is, we want to know about it, and we
2430 * want it to fail. We can acheive this by adding a write
2431 * verifier to the buffer.
2432 */
2433 bp->b_ops = &xfs_inode_buf_ops;
2434
2435 /*
2436 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2437 * too. This requires lookups, and will skip inodes that we've
2438 * already marked XFS_ISTALE.
2439 */
2440 for (i = 0; i < igeo->inodes_per_cluster; i++)
2441 xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
2442
2443 xfs_trans_stale_inode_buf(tp, bp);
2444 xfs_trans_binval(tp, bp);
2445 }
2446 return 0;
2447}
2448
2449/*
2450 * This is called to return an inode to the inode free list. The inode should
2451 * already be truncated to 0 length and have no pages associated with it. This
2452 * routine also assumes that the inode is already a part of the transaction.
2453 *
2454 * The on-disk copy of the inode will have been added to the list of unlinked
2455 * inodes in the AGI. We need to remove the inode from that list atomically with
2456 * respect to freeing it here.
2457 */
2458int
2459xfs_ifree(
2460 struct xfs_trans *tp,
2461 struct xfs_inode *ip)
2462{
2463 struct xfs_mount *mp = ip->i_mount;
2464 struct xfs_perag *pag;
2465 struct xfs_icluster xic = { 0 };
2466 struct xfs_inode_log_item *iip = ip->i_itemp;
2467 int error;
2468
2469 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);
2470 ASSERT(VFS_I(ip)->i_nlink == 0);
2471 ASSERT(ip->i_df.if_nextents == 0);
2472 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2473 ASSERT(ip->i_nblocks == 0);
2474
2475 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2476
2477 /*
2478 * Free the inode first so that we guarantee that the AGI lock is going
2479 * to be taken before we remove the inode from the unlinked list. This
2480 * makes the AGI lock -> unlinked list modification order the same as
2481 * used in O_TMPFILE creation.
2482 */
2483 error = xfs_difree(tp, pag, ip->i_ino, &xic);
2484 if (error)
2485 goto out;
2486
2487 error = xfs_iunlink_remove(tp, pag, ip);
2488 if (error)
2489 goto out;
2490
2491 /*
2492 * Free any local-format data sitting around before we reset the
2493 * data fork to extents format. Note that the attr fork data has
2494 * already been freed by xfs_attr_inactive.
2495 */
2496 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2497 kfree(ip->i_df.if_data);
2498 ip->i_df.if_data = NULL;
2499 ip->i_df.if_bytes = 0;
2500 }
2501
2502 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2503 ip->i_diflags = 0;
2504 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
2505 ip->i_forkoff = 0; /* mark the attr fork not in use */
2506 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2507 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2508 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2509
2510 /* Don't attempt to replay owner changes for a deleted inode */
2511 spin_lock(&iip->ili_lock);
2512 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2513 spin_unlock(&iip->ili_lock);
2514
2515 /*
2516 * Bump the generation count so no one will be confused
2517 * by reincarnations of this inode.
2518 */
2519 VFS_I(ip)->i_generation++;
2520 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2521
2522 if (xic.deleted)
2523 error = xfs_ifree_cluster(tp, pag, ip, &xic);
2524out:
2525 xfs_perag_put(pag);
2526 return error;
2527}
2528
2529/*
2530 * This is called to unpin an inode. The caller must have the inode locked
2531 * in at least shared mode so that the buffer cannot be subsequently pinned
2532 * once someone is waiting for it to be unpinned.
2533 */
2534static void
2535xfs_iunpin(
2536 struct xfs_inode *ip)
2537{
2538 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
2539
2540 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2541
2542 /* Give the log a push to start the unpinning I/O */
2543 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2544
2545}
2546
2547static void
2548__xfs_iunpin_wait(
2549 struct xfs_inode *ip)
2550{
2551 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2552 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2553
2554 xfs_iunpin(ip);
2555
2556 do {
2557 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2558 if (xfs_ipincount(ip))
2559 io_schedule();
2560 } while (xfs_ipincount(ip));
2561 finish_wait(wq, &wait.wq_entry);
2562}
2563
2564void
2565xfs_iunpin_wait(
2566 struct xfs_inode *ip)
2567{
2568 if (xfs_ipincount(ip))
2569 __xfs_iunpin_wait(ip);
2570}
2571
2572/*
2573 * Removing an inode from the namespace involves removing the directory entry
2574 * and dropping the link count on the inode. Removing the directory entry can
2575 * result in locking an AGF (directory blocks were freed) and removing a link
2576 * count can result in placing the inode on an unlinked list which results in
2577 * locking an AGI.
2578 *
2579 * The big problem here is that we have an ordering constraint on AGF and AGI
2580 * locking - inode allocation locks the AGI, then can allocate a new extent for
2581 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2582 * removes the inode from the unlinked list, requiring that we lock the AGI
2583 * first, and then freeing the inode can result in an inode chunk being freed
2584 * and hence freeing disk space requiring that we lock an AGF.
2585 *
2586 * Hence the ordering that is imposed by other parts of the code is AGI before
2587 * AGF. This means we cannot remove the directory entry before we drop the inode
2588 * reference count and put it on the unlinked list as this results in a lock
2589 * order of AGF then AGI, and this can deadlock against inode allocation and
2590 * freeing. Therefore we must drop the link counts before we remove the
2591 * directory entry.
2592 *
2593 * This is still safe from a transactional point of view - it is not until we
2594 * get to xfs_defer_finish() that we have the possibility of multiple
2595 * transactions in this operation. Hence as long as we remove the directory
2596 * entry and drop the link count in the first transaction of the remove
2597 * operation, there are no transactional constraints on the ordering here.
2598 */
2599int
2600xfs_remove(
2601 xfs_inode_t *dp,
2602 struct xfs_name *name,
2603 xfs_inode_t *ip)
2604{
2605 xfs_mount_t *mp = dp->i_mount;
2606 xfs_trans_t *tp = NULL;
2607 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2608 int dontcare;
2609 int error = 0;
2610 uint resblks;
2611
2612 trace_xfs_remove(dp, name);
2613
2614 if (xfs_is_shutdown(mp))
2615 return -EIO;
2616 if (xfs_ifork_zapped(dp, XFS_DATA_FORK))
2617 return -EIO;
2618
2619 error = xfs_qm_dqattach(dp);
2620 if (error)
2621 goto std_return;
2622
2623 error = xfs_qm_dqattach(ip);
2624 if (error)
2625 goto std_return;
2626
2627 /*
2628 * We try to get the real space reservation first, allowing for
2629 * directory btree deletion(s) implying possible bmap insert(s). If we
2630 * can't get the space reservation then we use 0 instead, and avoid the
2631 * bmap btree insert(s) in the directory code by, if the bmap insert
2632 * tries to happen, instead trimming the LAST block from the directory.
2633 *
2634 * Ignore EDQUOT and ENOSPC being returned via nospace_error because
2635 * the directory code can handle a reservationless update and we don't
2636 * want to prevent a user from trying to free space by deleting things.
2637 */
2638 resblks = XFS_REMOVE_SPACE_RES(mp);
2639 error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,
2640 &tp, &dontcare);
2641 if (error) {
2642 ASSERT(error != -ENOSPC);
2643 goto std_return;
2644 }
2645
2646 /*
2647 * If we're removing a directory perform some additional validation.
2648 */
2649 if (is_dir) {
2650 ASSERT(VFS_I(ip)->i_nlink >= 2);
2651 if (VFS_I(ip)->i_nlink != 2) {
2652 error = -ENOTEMPTY;
2653 goto out_trans_cancel;
2654 }
2655 if (!xfs_dir_isempty(ip)) {
2656 error = -ENOTEMPTY;
2657 goto out_trans_cancel;
2658 }
2659
2660 /* Drop the link from ip's "..". */
2661 error = xfs_droplink(tp, dp);
2662 if (error)
2663 goto out_trans_cancel;
2664
2665 /* Drop the "." link from ip to self. */
2666 error = xfs_droplink(tp, ip);
2667 if (error)
2668 goto out_trans_cancel;
2669
2670 /*
2671 * Point the unlinked child directory's ".." entry to the root
2672 * directory to eliminate back-references to inodes that may
2673 * get freed before the child directory is closed. If the fs
2674 * gets shrunk, this can lead to dirent inode validation errors.
2675 */
2676 if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
2677 error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
2678 tp->t_mountp->m_sb.sb_rootino, 0);
2679 if (error)
2680 goto out_trans_cancel;
2681 }
2682 } else {
2683 /*
2684 * When removing a non-directory we need to log the parent
2685 * inode here. For a directory this is done implicitly
2686 * by the xfs_droplink call for the ".." entry.
2687 */
2688 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2689 }
2690 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2691
2692 /* Drop the link from dp to ip. */
2693 error = xfs_droplink(tp, ip);
2694 if (error)
2695 goto out_trans_cancel;
2696
2697 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2698 if (error) {
2699 ASSERT(error != -ENOENT);
2700 goto out_trans_cancel;
2701 }
2702
2703 /*
2704 * Drop the link from dp to ip, and if ip was a directory, remove the
2705 * '.' and '..' references since we freed the directory.
2706 */
2707 xfs_dir_update_hook(dp, ip, -1, name);
2708
2709 /*
2710 * If this is a synchronous mount, make sure that the
2711 * remove transaction goes to disk before returning to
2712 * the user.
2713 */
2714 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
2715 xfs_trans_set_sync(tp);
2716
2717 error = xfs_trans_commit(tp);
2718 if (error)
2719 goto std_return;
2720
2721 if (is_dir && xfs_inode_is_filestream(ip))
2722 xfs_filestream_deassociate(ip);
2723
2724 return 0;
2725
2726 out_trans_cancel:
2727 xfs_trans_cancel(tp);
2728 std_return:
2729 return error;
2730}
2731
2732/*
2733 * Enter all inodes for a rename transaction into a sorted array.
2734 */
2735#define __XFS_SORT_INODES 5
2736STATIC void
2737xfs_sort_for_rename(
2738 struct xfs_inode *dp1, /* in: old (source) directory inode */
2739 struct xfs_inode *dp2, /* in: new (target) directory inode */
2740 struct xfs_inode *ip1, /* in: inode of old entry */
2741 struct xfs_inode *ip2, /* in: inode of new entry */
2742 struct xfs_inode *wip, /* in: whiteout inode */
2743 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2744 int *num_inodes) /* in/out: inodes in array */
2745{
2746 int i, j;
2747
2748 ASSERT(*num_inodes == __XFS_SORT_INODES);
2749 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2750
2751 /*
2752 * i_tab contains a list of pointers to inodes. We initialize
2753 * the table here & we'll sort it. We will then use it to
2754 * order the acquisition of the inode locks.
2755 *
2756 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2757 */
2758 i = 0;
2759 i_tab[i++] = dp1;
2760 i_tab[i++] = dp2;
2761 i_tab[i++] = ip1;
2762 if (ip2)
2763 i_tab[i++] = ip2;
2764 if (wip)
2765 i_tab[i++] = wip;
2766 *num_inodes = i;
2767
2768 /*
2769 * Sort the elements via bubble sort. (Remember, there are at
2770 * most 5 elements to sort, so this is adequate.)
2771 */
2772 for (i = 0; i < *num_inodes; i++) {
2773 for (j = 1; j < *num_inodes; j++) {
2774 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2775 struct xfs_inode *temp = i_tab[j];
2776 i_tab[j] = i_tab[j-1];
2777 i_tab[j-1] = temp;
2778 }
2779 }
2780 }
2781}
2782
2783static int
2784xfs_finish_rename(
2785 struct xfs_trans *tp)
2786{
2787 /*
2788 * If this is a synchronous mount, make sure that the rename transaction
2789 * goes to disk before returning to the user.
2790 */
2791 if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2792 xfs_trans_set_sync(tp);
2793
2794 return xfs_trans_commit(tp);
2795}
2796
2797/*
2798 * xfs_cross_rename()
2799 *
2800 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2801 */
2802STATIC int
2803xfs_cross_rename(
2804 struct xfs_trans *tp,
2805 struct xfs_inode *dp1,
2806 struct xfs_name *name1,
2807 struct xfs_inode *ip1,
2808 struct xfs_inode *dp2,
2809 struct xfs_name *name2,
2810 struct xfs_inode *ip2,
2811 int spaceres)
2812{
2813 int error = 0;
2814 int ip1_flags = 0;
2815 int ip2_flags = 0;
2816 int dp2_flags = 0;
2817
2818 /* Swap inode number for dirent in first parent */
2819 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2820 if (error)
2821 goto out_trans_abort;
2822
2823 /* Swap inode number for dirent in second parent */
2824 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2825 if (error)
2826 goto out_trans_abort;
2827
2828 /*
2829 * If we're renaming one or more directories across different parents,
2830 * update the respective ".." entries (and link counts) to match the new
2831 * parents.
2832 */
2833 if (dp1 != dp2) {
2834 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2835
2836 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2837 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2838 dp1->i_ino, spaceres);
2839 if (error)
2840 goto out_trans_abort;
2841
2842 /* transfer ip2 ".." reference to dp1 */
2843 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2844 error = xfs_droplink(tp, dp2);
2845 if (error)
2846 goto out_trans_abort;
2847 xfs_bumplink(tp, dp1);
2848 }
2849
2850 /*
2851 * Although ip1 isn't changed here, userspace needs
2852 * to be warned about the change, so that applications
2853 * relying on it (like backup ones), will properly
2854 * notify the change
2855 */
2856 ip1_flags |= XFS_ICHGTIME_CHG;
2857 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2858 }
2859
2860 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2861 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2862 dp2->i_ino, spaceres);
2863 if (error)
2864 goto out_trans_abort;
2865
2866 /* transfer ip1 ".." reference to dp2 */
2867 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2868 error = xfs_droplink(tp, dp1);
2869 if (error)
2870 goto out_trans_abort;
2871 xfs_bumplink(tp, dp2);
2872 }
2873
2874 /*
2875 * Although ip2 isn't changed here, userspace needs
2876 * to be warned about the change, so that applications
2877 * relying on it (like backup ones), will properly
2878 * notify the change
2879 */
2880 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2881 ip2_flags |= XFS_ICHGTIME_CHG;
2882 }
2883 }
2884
2885 if (ip1_flags) {
2886 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2887 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2888 }
2889 if (ip2_flags) {
2890 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2891 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2892 }
2893 if (dp2_flags) {
2894 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2895 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2896 }
2897 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2898 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2899
2900 /*
2901 * Inform our hook clients that we've finished an exchange operation as
2902 * follows: removed the source and target files from their directories;
2903 * added the target to the source directory; and added the source to
2904 * the target directory. All inodes are locked, so it's ok to model a
2905 * rename this way so long as we say we deleted entries before we add
2906 * new ones.
2907 */
2908 xfs_dir_update_hook(dp1, ip1, -1, name1);
2909 xfs_dir_update_hook(dp2, ip2, -1, name2);
2910 xfs_dir_update_hook(dp1, ip2, 1, name1);
2911 xfs_dir_update_hook(dp2, ip1, 1, name2);
2912
2913 return xfs_finish_rename(tp);
2914
2915out_trans_abort:
2916 xfs_trans_cancel(tp);
2917 return error;
2918}
2919
2920/*
2921 * xfs_rename_alloc_whiteout()
2922 *
2923 * Return a referenced, unlinked, unlocked inode that can be used as a
2924 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2925 * crash between allocating the inode and linking it into the rename transaction
2926 * recovery will free the inode and we won't leak it.
2927 */
2928static int
2929xfs_rename_alloc_whiteout(
2930 struct mnt_idmap *idmap,
2931 struct xfs_name *src_name,
2932 struct xfs_inode *dp,
2933 struct xfs_inode **wip)
2934{
2935 struct xfs_inode *tmpfile;
2936 struct qstr name;
2937 int error;
2938
2939 error = xfs_create_tmpfile(idmap, dp, S_IFCHR | WHITEOUT_MODE,
2940 &tmpfile);
2941 if (error)
2942 return error;
2943
2944 name.name = src_name->name;
2945 name.len = src_name->len;
2946 error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);
2947 if (error) {
2948 xfs_finish_inode_setup(tmpfile);
2949 xfs_irele(tmpfile);
2950 return error;
2951 }
2952
2953 /*
2954 * Prepare the tmpfile inode as if it were created through the VFS.
2955 * Complete the inode setup and flag it as linkable. nlink is already
2956 * zero, so we can skip the drop_nlink.
2957 */
2958 xfs_setup_iops(tmpfile);
2959 xfs_finish_inode_setup(tmpfile);
2960 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2961
2962 *wip = tmpfile;
2963 return 0;
2964}
2965
2966/*
2967 * xfs_rename
2968 */
2969int
2970xfs_rename(
2971 struct mnt_idmap *idmap,
2972 struct xfs_inode *src_dp,
2973 struct xfs_name *src_name,
2974 struct xfs_inode *src_ip,
2975 struct xfs_inode *target_dp,
2976 struct xfs_name *target_name,
2977 struct xfs_inode *target_ip,
2978 unsigned int flags)
2979{
2980 struct xfs_mount *mp = src_dp->i_mount;
2981 struct xfs_trans *tp;
2982 struct xfs_inode *wip = NULL; /* whiteout inode */
2983 struct xfs_inode *inodes[__XFS_SORT_INODES];
2984 int i;
2985 int num_inodes = __XFS_SORT_INODES;
2986 bool new_parent = (src_dp != target_dp);
2987 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2988 int spaceres;
2989 bool retried = false;
2990 int error, nospace_error = 0;
2991
2992 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2993
2994 if ((flags & RENAME_EXCHANGE) && !target_ip)
2995 return -EINVAL;
2996
2997 /*
2998 * If we are doing a whiteout operation, allocate the whiteout inode
2999 * we will be placing at the target and ensure the type is set
3000 * appropriately.
3001 */
3002 if (flags & RENAME_WHITEOUT) {
3003 error = xfs_rename_alloc_whiteout(idmap, src_name,
3004 target_dp, &wip);
3005 if (error)
3006 return error;
3007
3008 /* setup target dirent info as whiteout */
3009 src_name->type = XFS_DIR3_FT_CHRDEV;
3010 }
3011
3012 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
3013 inodes, &num_inodes);
3014
3015retry:
3016 nospace_error = 0;
3017 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
3018 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
3019 if (error == -ENOSPC) {
3020 nospace_error = error;
3021 spaceres = 0;
3022 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
3023 &tp);
3024 }
3025 if (error)
3026 goto out_release_wip;
3027
3028 /*
3029 * Attach the dquots to the inodes
3030 */
3031 error = xfs_qm_vop_rename_dqattach(inodes);
3032 if (error)
3033 goto out_trans_cancel;
3034
3035 /*
3036 * Lock all the participating inodes. Depending upon whether
3037 * the target_name exists in the target directory, and
3038 * whether the target directory is the same as the source
3039 * directory, we can lock from 2 to 5 inodes.
3040 */
3041 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
3042
3043 /*
3044 * Join all the inodes to the transaction. From this point on,
3045 * we can rely on either trans_commit or trans_cancel to unlock
3046 * them.
3047 */
3048 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
3049 if (new_parent)
3050 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
3051 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
3052 if (target_ip)
3053 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
3054 if (wip)
3055 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
3056
3057 /*
3058 * If we are using project inheritance, we only allow renames
3059 * into our tree when the project IDs are the same; else the
3060 * tree quota mechanism would be circumvented.
3061 */
3062 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
3063 target_dp->i_projid != src_ip->i_projid)) {
3064 error = -EXDEV;
3065 goto out_trans_cancel;
3066 }
3067
3068 /* RENAME_EXCHANGE is unique from here on. */
3069 if (flags & RENAME_EXCHANGE)
3070 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
3071 target_dp, target_name, target_ip,
3072 spaceres);
3073
3074 /*
3075 * Try to reserve quota to handle an expansion of the target directory.
3076 * We'll allow the rename to continue in reservationless mode if we hit
3077 * a space usage constraint. If we trigger reservationless mode, save
3078 * the errno if there isn't any free space in the target directory.
3079 */
3080 if (spaceres != 0) {
3081 error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,
3082 0, false);
3083 if (error == -EDQUOT || error == -ENOSPC) {
3084 if (!retried) {
3085 xfs_trans_cancel(tp);
3086 xfs_blockgc_free_quota(target_dp, 0);
3087 retried = true;
3088 goto retry;
3089 }
3090
3091 nospace_error = error;
3092 spaceres = 0;
3093 error = 0;
3094 }
3095 if (error)
3096 goto out_trans_cancel;
3097 }
3098
3099 /*
3100 * Check for expected errors before we dirty the transaction
3101 * so we can return an error without a transaction abort.
3102 */
3103 if (target_ip == NULL) {
3104 /*
3105 * If there's no space reservation, check the entry will
3106 * fit before actually inserting it.
3107 */
3108 if (!spaceres) {
3109 error = xfs_dir_canenter(tp, target_dp, target_name);
3110 if (error)
3111 goto out_trans_cancel;
3112 }
3113 } else {
3114 /*
3115 * If target exists and it's a directory, check that whether
3116 * it can be destroyed.
3117 */
3118 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
3119 (!xfs_dir_isempty(target_ip) ||
3120 (VFS_I(target_ip)->i_nlink > 2))) {
3121 error = -EEXIST;
3122 goto out_trans_cancel;
3123 }
3124 }
3125
3126 /*
3127 * Lock the AGI buffers we need to handle bumping the nlink of the
3128 * whiteout inode off the unlinked list and to handle dropping the
3129 * nlink of the target inode. Per locking order rules, do this in
3130 * increasing AG order and before directory block allocation tries to
3131 * grab AGFs because we grab AGIs before AGFs.
3132 *
3133 * The (vfs) caller must ensure that if src is a directory then
3134 * target_ip is either null or an empty directory.
3135 */
3136 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
3137 if (inodes[i] == wip ||
3138 (inodes[i] == target_ip &&
3139 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
3140 struct xfs_perag *pag;
3141 struct xfs_buf *bp;
3142
3143 pag = xfs_perag_get(mp,
3144 XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));
3145 error = xfs_read_agi(pag, tp, &bp);
3146 xfs_perag_put(pag);
3147 if (error)
3148 goto out_trans_cancel;
3149 }
3150 }
3151
3152 /*
3153 * Directory entry creation below may acquire the AGF. Remove
3154 * the whiteout from the unlinked list first to preserve correct
3155 * AGI/AGF locking order. This dirties the transaction so failures
3156 * after this point will abort and log recovery will clean up the
3157 * mess.
3158 *
3159 * For whiteouts, we need to bump the link count on the whiteout
3160 * inode. After this point, we have a real link, clear the tmpfile
3161 * state flag from the inode so it doesn't accidentally get misused
3162 * in future.
3163 */
3164 if (wip) {
3165 struct xfs_perag *pag;
3166
3167 ASSERT(VFS_I(wip)->i_nlink == 0);
3168
3169 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
3170 error = xfs_iunlink_remove(tp, pag, wip);
3171 xfs_perag_put(pag);
3172 if (error)
3173 goto out_trans_cancel;
3174
3175 xfs_bumplink(tp, wip);
3176 VFS_I(wip)->i_state &= ~I_LINKABLE;
3177 }
3178
3179 /*
3180 * Set up the target.
3181 */
3182 if (target_ip == NULL) {
3183 /*
3184 * If target does not exist and the rename crosses
3185 * directories, adjust the target directory link count
3186 * to account for the ".." reference from the new entry.
3187 */
3188 error = xfs_dir_createname(tp, target_dp, target_name,
3189 src_ip->i_ino, spaceres);
3190 if (error)
3191 goto out_trans_cancel;
3192
3193 xfs_trans_ichgtime(tp, target_dp,
3194 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3195
3196 if (new_parent && src_is_directory) {
3197 xfs_bumplink(tp, target_dp);
3198 }
3199 } else { /* target_ip != NULL */
3200 /*
3201 * Link the source inode under the target name.
3202 * If the source inode is a directory and we are moving
3203 * it across directories, its ".." entry will be
3204 * inconsistent until we replace that down below.
3205 *
3206 * In case there is already an entry with the same
3207 * name at the destination directory, remove it first.
3208 */
3209 error = xfs_dir_replace(tp, target_dp, target_name,
3210 src_ip->i_ino, spaceres);
3211 if (error)
3212 goto out_trans_cancel;
3213
3214 xfs_trans_ichgtime(tp, target_dp,
3215 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3216
3217 /*
3218 * Decrement the link count on the target since the target
3219 * dir no longer points to it.
3220 */
3221 error = xfs_droplink(tp, target_ip);
3222 if (error)
3223 goto out_trans_cancel;
3224
3225 if (src_is_directory) {
3226 /*
3227 * Drop the link from the old "." entry.
3228 */
3229 error = xfs_droplink(tp, target_ip);
3230 if (error)
3231 goto out_trans_cancel;
3232 }
3233 } /* target_ip != NULL */
3234
3235 /*
3236 * Remove the source.
3237 */
3238 if (new_parent && src_is_directory) {
3239 /*
3240 * Rewrite the ".." entry to point to the new
3241 * directory.
3242 */
3243 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3244 target_dp->i_ino, spaceres);
3245 ASSERT(error != -EEXIST);
3246 if (error)
3247 goto out_trans_cancel;
3248 }
3249
3250 /*
3251 * We always want to hit the ctime on the source inode.
3252 *
3253 * This isn't strictly required by the standards since the source
3254 * inode isn't really being changed, but old unix file systems did
3255 * it and some incremental backup programs won't work without it.
3256 */
3257 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3258 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3259
3260 /*
3261 * Adjust the link count on src_dp. This is necessary when
3262 * renaming a directory, either within one parent when
3263 * the target existed, or across two parent directories.
3264 */
3265 if (src_is_directory && (new_parent || target_ip != NULL)) {
3266
3267 /*
3268 * Decrement link count on src_directory since the
3269 * entry that's moved no longer points to it.
3270 */
3271 error = xfs_droplink(tp, src_dp);
3272 if (error)
3273 goto out_trans_cancel;
3274 }
3275
3276 /*
3277 * For whiteouts, we only need to update the source dirent with the
3278 * inode number of the whiteout inode rather than removing it
3279 * altogether.
3280 */
3281 if (wip)
3282 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3283 spaceres);
3284 else
3285 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3286 spaceres);
3287
3288 if (error)
3289 goto out_trans_cancel;
3290
3291 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3292 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3293 if (new_parent)
3294 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3295
3296 /*
3297 * Inform our hook clients that we've finished a rename operation as
3298 * follows: removed the source and target files from their directories;
3299 * that we've added the source to the target directory; and finally
3300 * that we've added the whiteout, if there was one. All inodes are
3301 * locked, so it's ok to model a rename this way so long as we say we
3302 * deleted entries before we add new ones.
3303 */
3304 if (target_ip)
3305 xfs_dir_update_hook(target_dp, target_ip, -1, target_name);
3306 xfs_dir_update_hook(src_dp, src_ip, -1, src_name);
3307 xfs_dir_update_hook(target_dp, src_ip, 1, target_name);
3308 if (wip)
3309 xfs_dir_update_hook(src_dp, wip, 1, src_name);
3310
3311 error = xfs_finish_rename(tp);
3312 if (wip)
3313 xfs_irele(wip);
3314 return error;
3315
3316out_trans_cancel:
3317 xfs_trans_cancel(tp);
3318out_release_wip:
3319 if (wip)
3320 xfs_irele(wip);
3321 if (error == -ENOSPC && nospace_error)
3322 error = nospace_error;
3323 return error;
3324}
3325
3326static int
3327xfs_iflush(
3328 struct xfs_inode *ip,
3329 struct xfs_buf *bp)
3330{
3331 struct xfs_inode_log_item *iip = ip->i_itemp;
3332 struct xfs_dinode *dip;
3333 struct xfs_mount *mp = ip->i_mount;
3334 int error;
3335
3336 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED);
3337 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3338 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3339 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3340 ASSERT(iip->ili_item.li_buf == bp);
3341
3342 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3343
3344 /*
3345 * We don't flush the inode if any of the following checks fail, but we
3346 * do still update the log item and attach to the backing buffer as if
3347 * the flush happened. This is a formality to facilitate predictable
3348 * error handling as the caller will shutdown and fail the buffer.
3349 */
3350 error = -EFSCORRUPTED;
3351 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3352 mp, XFS_ERRTAG_IFLUSH_1)) {
3353 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3354 "%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,
3355 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3356 goto flush_out;
3357 }
3358 if (S_ISREG(VFS_I(ip)->i_mode)) {
3359 if (XFS_TEST_ERROR(
3360 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3361 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3362 mp, XFS_ERRTAG_IFLUSH_3)) {
3363 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3364 "%s: Bad regular inode %llu, ptr "PTR_FMT,
3365 __func__, ip->i_ino, ip);
3366 goto flush_out;
3367 }
3368 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3369 if (XFS_TEST_ERROR(
3370 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3371 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3372 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3373 mp, XFS_ERRTAG_IFLUSH_4)) {
3374 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3375 "%s: Bad directory inode %llu, ptr "PTR_FMT,
3376 __func__, ip->i_ino, ip);
3377 goto flush_out;
3378 }
3379 }
3380 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >
3381 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3382 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3383 "%s: detected corrupt incore inode %llu, "
3384 "total extents = %llu nblocks = %lld, ptr "PTR_FMT,
3385 __func__, ip->i_ino,
3386 ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),
3387 ip->i_nblocks, ip);
3388 goto flush_out;
3389 }
3390 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3391 mp, XFS_ERRTAG_IFLUSH_6)) {
3392 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3393 "%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,
3394 __func__, ip->i_ino, ip->i_forkoff, ip);
3395 goto flush_out;
3396 }
3397
3398 /*
3399 * Inode item log recovery for v2 inodes are dependent on the flushiter
3400 * count for correct sequencing. We bump the flush iteration count so
3401 * we can detect flushes which postdate a log record during recovery.
3402 * This is redundant as we now log every change and hence this can't
3403 * happen but we need to still do it to ensure backwards compatibility
3404 * with old kernels that predate logging all inode changes.
3405 */
3406 if (!xfs_has_v3inodes(mp))
3407 ip->i_flushiter++;
3408
3409 /*
3410 * If there are inline format data / attr forks attached to this inode,
3411 * make sure they are not corrupt.
3412 */
3413 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3414 xfs_ifork_verify_local_data(ip))
3415 goto flush_out;
3416 if (xfs_inode_has_attr_fork(ip) &&
3417 ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&
3418 xfs_ifork_verify_local_attr(ip))
3419 goto flush_out;
3420
3421 /*
3422 * Copy the dirty parts of the inode into the on-disk inode. We always
3423 * copy out the core of the inode, because if the inode is dirty at all
3424 * the core must be.
3425 */
3426 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3427
3428 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3429 if (!xfs_has_v3inodes(mp)) {
3430 if (ip->i_flushiter == DI_MAX_FLUSH)
3431 ip->i_flushiter = 0;
3432 }
3433
3434 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3435 if (xfs_inode_has_attr_fork(ip))
3436 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3437
3438 /*
3439 * We've recorded everything logged in the inode, so we'd like to clear
3440 * the ili_fields bits so we don't log and flush things unnecessarily.
3441 * However, we can't stop logging all this information until the data
3442 * we've copied into the disk buffer is written to disk. If we did we
3443 * might overwrite the copy of the inode in the log with all the data
3444 * after re-logging only part of it, and in the face of a crash we
3445 * wouldn't have all the data we need to recover.
3446 *
3447 * What we do is move the bits to the ili_last_fields field. When
3448 * logging the inode, these bits are moved back to the ili_fields field.
3449 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3450 * we know that the information those bits represent is permanently on
3451 * disk. As long as the flush completes before the inode is logged
3452 * again, then both ili_fields and ili_last_fields will be cleared.
3453 */
3454 error = 0;
3455flush_out:
3456 spin_lock(&iip->ili_lock);
3457 iip->ili_last_fields = iip->ili_fields;
3458 iip->ili_fields = 0;
3459 iip->ili_fsync_fields = 0;
3460 spin_unlock(&iip->ili_lock);
3461
3462 /*
3463 * Store the current LSN of the inode so that we can tell whether the
3464 * item has moved in the AIL from xfs_buf_inode_iodone().
3465 */
3466 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3467 &iip->ili_item.li_lsn);
3468
3469 /* generate the checksum. */
3470 xfs_dinode_calc_crc(mp, dip);
3471 if (error)
3472 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
3473 return error;
3474}
3475
3476/*
3477 * Non-blocking flush of dirty inode metadata into the backing buffer.
3478 *
3479 * The caller must have a reference to the inode and hold the cluster buffer
3480 * locked. The function will walk across all the inodes on the cluster buffer it
3481 * can find and lock without blocking, and flush them to the cluster buffer.
3482 *
3483 * On successful flushing of at least one inode, the caller must write out the
3484 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3485 * the caller needs to release the buffer. On failure, the filesystem will be
3486 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3487 * will be returned.
3488 */
3489int
3490xfs_iflush_cluster(
3491 struct xfs_buf *bp)
3492{
3493 struct xfs_mount *mp = bp->b_mount;
3494 struct xfs_log_item *lip, *n;
3495 struct xfs_inode *ip;
3496 struct xfs_inode_log_item *iip;
3497 int clcount = 0;
3498 int error = 0;
3499
3500 /*
3501 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3502 * will remove itself from the list.
3503 */
3504 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3505 iip = (struct xfs_inode_log_item *)lip;
3506 ip = iip->ili_inode;
3507
3508 /*
3509 * Quick and dirty check to avoid locks if possible.
3510 */
3511 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3512 continue;
3513 if (xfs_ipincount(ip))
3514 continue;
3515
3516 /*
3517 * The inode is still attached to the buffer, which means it is
3518 * dirty but reclaim might try to grab it. Check carefully for
3519 * that, and grab the ilock while still holding the i_flags_lock
3520 * to guarantee reclaim will not be able to reclaim this inode
3521 * once we drop the i_flags_lock.
3522 */
3523 spin_lock(&ip->i_flags_lock);
3524 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3525 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3526 spin_unlock(&ip->i_flags_lock);
3527 continue;
3528 }
3529
3530 /*
3531 * ILOCK will pin the inode against reclaim and prevent
3532 * concurrent transactions modifying the inode while we are
3533 * flushing the inode. If we get the lock, set the flushing
3534 * state before we drop the i_flags_lock.
3535 */
3536 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3537 spin_unlock(&ip->i_flags_lock);
3538 continue;
3539 }
3540 __xfs_iflags_set(ip, XFS_IFLUSHING);
3541 spin_unlock(&ip->i_flags_lock);
3542
3543 /*
3544 * Abort flushing this inode if we are shut down because the
3545 * inode may not currently be in the AIL. This can occur when
3546 * log I/O failure unpins the inode without inserting into the
3547 * AIL, leaving a dirty/unpinned inode attached to the buffer
3548 * that otherwise looks like it should be flushed.
3549 */
3550 if (xlog_is_shutdown(mp->m_log)) {
3551 xfs_iunpin_wait(ip);
3552 xfs_iflush_abort(ip);
3553 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3554 error = -EIO;
3555 continue;
3556 }
3557
3558 /* don't block waiting on a log force to unpin dirty inodes */
3559 if (xfs_ipincount(ip)) {
3560 xfs_iflags_clear(ip, XFS_IFLUSHING);
3561 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3562 continue;
3563 }
3564
3565 if (!xfs_inode_clean(ip))
3566 error = xfs_iflush(ip, bp);
3567 else
3568 xfs_iflags_clear(ip, XFS_IFLUSHING);
3569 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3570 if (error)
3571 break;
3572 clcount++;
3573 }
3574
3575 if (error) {
3576 /*
3577 * Shutdown first so we kill the log before we release this
3578 * buffer. If it is an INODE_ALLOC buffer and pins the tail
3579 * of the log, failing it before the _log_ is shut down can
3580 * result in the log tail being moved forward in the journal
3581 * on disk because log writes can still be taking place. Hence
3582 * unpinning the tail will allow the ICREATE intent to be
3583 * removed from the log an recovery will fail with uninitialised
3584 * inode cluster buffers.
3585 */
3586 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3587 bp->b_flags |= XBF_ASYNC;
3588 xfs_buf_ioend_fail(bp);
3589 return error;
3590 }
3591
3592 if (!clcount)
3593 return -EAGAIN;
3594
3595 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3596 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3597 return 0;
3598
3599}
3600
3601/* Release an inode. */
3602void
3603xfs_irele(
3604 struct xfs_inode *ip)
3605{
3606 trace_xfs_irele(ip, _RET_IP_);
3607 iput(VFS_I(ip));
3608}
3609
3610/*
3611 * Ensure all commited transactions touching the inode are written to the log.
3612 */
3613int
3614xfs_log_force_inode(
3615 struct xfs_inode *ip)
3616{
3617 xfs_csn_t seq = 0;
3618
3619 xfs_ilock(ip, XFS_ILOCK_SHARED);
3620 if (xfs_ipincount(ip))
3621 seq = ip->i_itemp->ili_commit_seq;
3622 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3623
3624 if (!seq)
3625 return 0;
3626 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3627}
3628
3629/*
3630 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3631 * abide vfs locking order (lowest pointer value goes first) and breaking the
3632 * layout leases before proceeding. The loop is needed because we cannot call
3633 * the blocking break_layout() with the iolocks held, and therefore have to
3634 * back out both locks.
3635 */
3636static int
3637xfs_iolock_two_inodes_and_break_layout(
3638 struct inode *src,
3639 struct inode *dest)
3640{
3641 int error;
3642
3643 if (src > dest)
3644 swap(src, dest);
3645
3646retry:
3647 /* Wait to break both inodes' layouts before we start locking. */
3648 error = break_layout(src, true);
3649 if (error)
3650 return error;
3651 if (src != dest) {
3652 error = break_layout(dest, true);
3653 if (error)
3654 return error;
3655 }
3656
3657 /* Lock one inode and make sure nobody got in and leased it. */
3658 inode_lock(src);
3659 error = break_layout(src, false);
3660 if (error) {
3661 inode_unlock(src);
3662 if (error == -EWOULDBLOCK)
3663 goto retry;
3664 return error;
3665 }
3666
3667 if (src == dest)
3668 return 0;
3669
3670 /* Lock the other inode and make sure nobody got in and leased it. */
3671 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3672 error = break_layout(dest, false);
3673 if (error) {
3674 inode_unlock(src);
3675 inode_unlock(dest);
3676 if (error == -EWOULDBLOCK)
3677 goto retry;
3678 return error;
3679 }
3680
3681 return 0;
3682}
3683
3684static int
3685xfs_mmaplock_two_inodes_and_break_dax_layout(
3686 struct xfs_inode *ip1,
3687 struct xfs_inode *ip2)
3688{
3689 int error;
3690 bool retry;
3691 struct page *page;
3692
3693 if (ip1->i_ino > ip2->i_ino)
3694 swap(ip1, ip2);
3695
3696again:
3697 retry = false;
3698 /* Lock the first inode */
3699 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3700 error = xfs_break_dax_layouts(VFS_I(ip1), &retry);
3701 if (error || retry) {
3702 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3703 if (error == 0 && retry)
3704 goto again;
3705 return error;
3706 }
3707
3708 if (ip1 == ip2)
3709 return 0;
3710
3711 /* Nested lock the second inode */
3712 xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));
3713 /*
3714 * We cannot use xfs_break_dax_layouts() directly here because it may
3715 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable
3716 * for this nested lock case.
3717 */
3718 page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);
3719 if (page && page_ref_count(page) != 1) {
3720 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3721 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3722 goto again;
3723 }
3724
3725 return 0;
3726}
3727
3728/*
3729 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3730 * mmap activity.
3731 */
3732int
3733xfs_ilock2_io_mmap(
3734 struct xfs_inode *ip1,
3735 struct xfs_inode *ip2)
3736{
3737 int ret;
3738
3739 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3740 if (ret)
3741 return ret;
3742
3743 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3744 ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);
3745 if (ret) {
3746 inode_unlock(VFS_I(ip2));
3747 if (ip1 != ip2)
3748 inode_unlock(VFS_I(ip1));
3749 return ret;
3750 }
3751 } else
3752 filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
3753 VFS_I(ip2)->i_mapping);
3754
3755 return 0;
3756}
3757
3758/* Unlock both inodes to allow IO and mmap activity. */
3759void
3760xfs_iunlock2_io_mmap(
3761 struct xfs_inode *ip1,
3762 struct xfs_inode *ip2)
3763{
3764 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3765 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3766 if (ip1 != ip2)
3767 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3768 } else
3769 filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
3770 VFS_I(ip2)->i_mapping);
3771
3772 inode_unlock(VFS_I(ip2));
3773 if (ip1 != ip2)
3774 inode_unlock(VFS_I(ip1));
3775}
3776
3777/* Drop the MMAPLOCK and the IOLOCK after a remap completes. */
3778void
3779xfs_iunlock2_remapping(
3780 struct xfs_inode *ip1,
3781 struct xfs_inode *ip2)
3782{
3783 xfs_iflags_clear(ip1, XFS_IREMAPPING);
3784
3785 if (ip1 != ip2)
3786 xfs_iunlock(ip1, XFS_MMAPLOCK_SHARED);
3787 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3788
3789 if (ip1 != ip2)
3790 inode_unlock_shared(VFS_I(ip1));
3791 inode_unlock(VFS_I(ip2));
3792}
3793
3794/*
3795 * Reload the incore inode list for this inode. Caller should ensure that
3796 * the link count cannot change, either by taking ILOCK_SHARED or otherwise
3797 * preventing other threads from executing.
3798 */
3799int
3800xfs_inode_reload_unlinked_bucket(
3801 struct xfs_trans *tp,
3802 struct xfs_inode *ip)
3803{
3804 struct xfs_mount *mp = tp->t_mountp;
3805 struct xfs_buf *agibp;
3806 struct xfs_agi *agi;
3807 struct xfs_perag *pag;
3808 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
3809 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
3810 xfs_agino_t prev_agino, next_agino;
3811 unsigned int bucket;
3812 bool foundit = false;
3813 int error;
3814
3815 /* Grab the first inode in the list */
3816 pag = xfs_perag_get(mp, agno);
3817 error = xfs_ialloc_read_agi(pag, tp, &agibp);
3818 xfs_perag_put(pag);
3819 if (error)
3820 return error;
3821
3822 /*
3823 * We've taken ILOCK_SHARED and the AGI buffer lock to stabilize the
3824 * incore unlinked list pointers for this inode. Check once more to
3825 * see if we raced with anyone else to reload the unlinked list.
3826 */
3827 if (!xfs_inode_unlinked_incomplete(ip)) {
3828 foundit = true;
3829 goto out_agibp;
3830 }
3831
3832 bucket = agino % XFS_AGI_UNLINKED_BUCKETS;
3833 agi = agibp->b_addr;
3834
3835 trace_xfs_inode_reload_unlinked_bucket(ip);
3836
3837 xfs_info_ratelimited(mp,
3838 "Found unrecovered unlinked inode 0x%x in AG 0x%x. Initiating list recovery.",
3839 agino, agno);
3840
3841 prev_agino = NULLAGINO;
3842 next_agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3843 while (next_agino != NULLAGINO) {
3844 struct xfs_inode *next_ip = NULL;
3845
3846 /* Found this caller's inode, set its backlink. */
3847 if (next_agino == agino) {
3848 next_ip = ip;
3849 next_ip->i_prev_unlinked = prev_agino;
3850 foundit = true;
3851 goto next_inode;
3852 }
3853
3854 /* Try in-memory lookup first. */
3855 next_ip = xfs_iunlink_lookup(pag, next_agino);
3856 if (next_ip)
3857 goto next_inode;
3858
3859 /* Inode not in memory, try reloading it. */
3860 error = xfs_iunlink_reload_next(tp, agibp, prev_agino,
3861 next_agino);
3862 if (error)
3863 break;
3864
3865 /* Grab the reloaded inode. */
3866 next_ip = xfs_iunlink_lookup(pag, next_agino);
3867 if (!next_ip) {
3868 /* No incore inode at all? We reloaded it... */
3869 ASSERT(next_ip != NULL);
3870 error = -EFSCORRUPTED;
3871 break;
3872 }
3873
3874next_inode:
3875 prev_agino = next_agino;
3876 next_agino = next_ip->i_next_unlinked;
3877 }
3878
3879out_agibp:
3880 xfs_trans_brelse(tp, agibp);
3881 /* Should have found this inode somewhere in the iunlinked bucket. */
3882 if (!error && !foundit)
3883 error = -EFSCORRUPTED;
3884 return error;
3885}
3886
3887/* Decide if this inode is missing its unlinked list and reload it. */
3888int
3889xfs_inode_reload_unlinked(
3890 struct xfs_inode *ip)
3891{
3892 struct xfs_trans *tp;
3893 int error;
3894
3895 error = xfs_trans_alloc_empty(ip->i_mount, &tp);
3896 if (error)
3897 return error;
3898
3899 xfs_ilock(ip, XFS_ILOCK_SHARED);
3900 if (xfs_inode_unlinked_incomplete(ip))
3901 error = xfs_inode_reload_unlinked_bucket(tp, ip);
3902 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3903 xfs_trans_cancel(tp);
3904
3905 return error;
3906}
3907
3908/* Has this inode fork been zapped by repair? */
3909bool
3910xfs_ifork_zapped(
3911 const struct xfs_inode *ip,
3912 int whichfork)
3913{
3914 unsigned int datamask = 0;
3915
3916 switch (whichfork) {
3917 case XFS_DATA_FORK:
3918 switch (ip->i_vnode.i_mode & S_IFMT) {
3919 case S_IFDIR:
3920 datamask = XFS_SICK_INO_DIR_ZAPPED;
3921 break;
3922 case S_IFLNK:
3923 datamask = XFS_SICK_INO_SYMLINK_ZAPPED;
3924 break;
3925 }
3926 return ip->i_sick & (XFS_SICK_INO_BMBTD_ZAPPED | datamask);
3927 case XFS_ATTR_FORK:
3928 return ip->i_sick & XFS_SICK_INO_BMBTA_ZAPPED;
3929 default:
3930 return false;
3931 }
3932}
3933
3934/* Compute the number of data and realtime blocks used by a file. */
3935void
3936xfs_inode_count_blocks(
3937 struct xfs_trans *tp,
3938 struct xfs_inode *ip,
3939 xfs_filblks_t *dblocks,
3940 xfs_filblks_t *rblocks)
3941{
3942 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
3943
3944 *rblocks = 0;
3945 if (XFS_IS_REALTIME_INODE(ip))
3946 xfs_bmap_count_leaves(ifp, rblocks);
3947 *dblocks = ip->i_nblocks - *rblocks;
3948}