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
41struct kmem_cache *xfs_inode_cache;
42
43/*
44 * Used in xfs_itruncate_extents(). This is the maximum number of extents
45 * freed from a file in a single transaction.
46 */
47#define XFS_ITRUNC_MAX_EXTENTS 2
48
49STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
50STATIC int xfs_iunlink_remove(struct xfs_trans *tp, struct xfs_perag *pag,
51 struct xfs_inode *);
52
53/*
54 * helper function to extract extent size hint from inode
55 */
56xfs_extlen_t
57xfs_get_extsz_hint(
58 struct xfs_inode *ip)
59{
60 /*
61 * No point in aligning allocations if we need to COW to actually
62 * write to them.
63 */
64 if (xfs_is_always_cow_inode(ip))
65 return 0;
66 if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
67 return ip->i_extsize;
68 if (XFS_IS_REALTIME_INODE(ip))
69 return ip->i_mount->m_sb.sb_rextsize;
70 return 0;
71}
72
73/*
74 * Helper function to extract CoW extent size hint from inode.
75 * Between the extent size hint and the CoW extent size hint, we
76 * return the greater of the two. If the value is zero (automatic),
77 * use the default size.
78 */
79xfs_extlen_t
80xfs_get_cowextsz_hint(
81 struct xfs_inode *ip)
82{
83 xfs_extlen_t a, b;
84
85 a = 0;
86 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
87 a = ip->i_cowextsize;
88 b = xfs_get_extsz_hint(ip);
89
90 a = max(a, b);
91 if (a == 0)
92 return XFS_DEFAULT_COWEXTSZ_HINT;
93 return a;
94}
95
96/*
97 * These two are wrapper routines around the xfs_ilock() routine used to
98 * centralize some grungy code. They are used in places that wish to lock the
99 * inode solely for reading the extents. The reason these places can't just
100 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
101 * bringing in of the extents from disk for a file in b-tree format. If the
102 * inode is in b-tree format, then we need to lock the inode exclusively until
103 * the extents are read in. Locking it exclusively all the time would limit
104 * our parallelism unnecessarily, though. What we do instead is check to see
105 * if the extents have been read in yet, and only lock the inode exclusively
106 * if they have not.
107 *
108 * The functions return a value which should be given to the corresponding
109 * xfs_iunlock() call.
110 */
111uint
112xfs_ilock_data_map_shared(
113 struct xfs_inode *ip)
114{
115 uint lock_mode = XFS_ILOCK_SHARED;
116
117 if (xfs_need_iread_extents(&ip->i_df))
118 lock_mode = XFS_ILOCK_EXCL;
119 xfs_ilock(ip, lock_mode);
120 return lock_mode;
121}
122
123uint
124xfs_ilock_attr_map_shared(
125 struct xfs_inode *ip)
126{
127 uint lock_mode = XFS_ILOCK_SHARED;
128
129 if (xfs_inode_has_attr_fork(ip) && xfs_need_iread_extents(&ip->i_af))
130 lock_mode = XFS_ILOCK_EXCL;
131 xfs_ilock(ip, lock_mode);
132 return lock_mode;
133}
134
135/*
136 * You can't set both SHARED and EXCL for the same lock,
137 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_MMAPLOCK_SHARED,
138 * XFS_MMAPLOCK_EXCL, XFS_ILOCK_SHARED, XFS_ILOCK_EXCL are valid values
139 * to set in lock_flags.
140 */
141static inline void
142xfs_lock_flags_assert(
143 uint lock_flags)
144{
145 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
146 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
147 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
148 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
149 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
150 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
151 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
152 ASSERT(lock_flags != 0);
153}
154
155/*
156 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2
157 * multi-reader locks: invalidate_lock and the i_lock. This routine allows
158 * various combinations of the locks to be obtained.
159 *
160 * The 3 locks should always be ordered so that the IO lock is obtained first,
161 * the mmap lock second and the ilock last in order to prevent deadlock.
162 *
163 * Basic locking order:
164 *
165 * i_rwsem -> invalidate_lock -> page_lock -> i_ilock
166 *
167 * mmap_lock locking order:
168 *
169 * i_rwsem -> page lock -> mmap_lock
170 * mmap_lock -> invalidate_lock -> page_lock
171 *
172 * The difference in mmap_lock locking order mean that we cannot hold the
173 * invalidate_lock over syscall based read(2)/write(2) based IO. These IO paths
174 * can fault in pages during copy in/out (for buffered IO) or require the
175 * mmap_lock in get_user_pages() to map the user pages into the kernel address
176 * space for direct IO. Similarly the i_rwsem cannot be taken inside a page
177 * fault because page faults already hold the mmap_lock.
178 *
179 * Hence to serialise fully against both syscall and mmap based IO, we need to
180 * take both the i_rwsem and the invalidate_lock. These locks should *only* be
181 * both taken in places where we need to invalidate the page cache in a race
182 * free manner (e.g. truncate, hole punch and other extent manipulation
183 * functions).
184 */
185void
186xfs_ilock(
187 xfs_inode_t *ip,
188 uint lock_flags)
189{
190 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
191
192 xfs_lock_flags_assert(lock_flags);
193
194 if (lock_flags & XFS_IOLOCK_EXCL) {
195 down_write_nested(&VFS_I(ip)->i_rwsem,
196 XFS_IOLOCK_DEP(lock_flags));
197 } else if (lock_flags & XFS_IOLOCK_SHARED) {
198 down_read_nested(&VFS_I(ip)->i_rwsem,
199 XFS_IOLOCK_DEP(lock_flags));
200 }
201
202 if (lock_flags & XFS_MMAPLOCK_EXCL) {
203 down_write_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
204 XFS_MMAPLOCK_DEP(lock_flags));
205 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
206 down_read_nested(&VFS_I(ip)->i_mapping->invalidate_lock,
207 XFS_MMAPLOCK_DEP(lock_flags));
208 }
209
210 if (lock_flags & XFS_ILOCK_EXCL)
211 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
212 else if (lock_flags & XFS_ILOCK_SHARED)
213 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
214}
215
216/*
217 * This is just like xfs_ilock(), except that the caller
218 * is guaranteed not to sleep. It returns 1 if it gets
219 * the requested locks and 0 otherwise. If the IO lock is
220 * obtained but the inode lock cannot be, then the IO lock
221 * is dropped before returning.
222 *
223 * ip -- the inode being locked
224 * lock_flags -- this parameter indicates the inode's locks to be
225 * to be locked. See the comment for xfs_ilock() for a list
226 * of valid values.
227 */
228int
229xfs_ilock_nowait(
230 xfs_inode_t *ip,
231 uint lock_flags)
232{
233 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
234
235 xfs_lock_flags_assert(lock_flags);
236
237 if (lock_flags & XFS_IOLOCK_EXCL) {
238 if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
239 goto out;
240 } else if (lock_flags & XFS_IOLOCK_SHARED) {
241 if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
242 goto out;
243 }
244
245 if (lock_flags & XFS_MMAPLOCK_EXCL) {
246 if (!down_write_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
247 goto out_undo_iolock;
248 } else if (lock_flags & XFS_MMAPLOCK_SHARED) {
249 if (!down_read_trylock(&VFS_I(ip)->i_mapping->invalidate_lock))
250 goto out_undo_iolock;
251 }
252
253 if (lock_flags & XFS_ILOCK_EXCL) {
254 if (!mrtryupdate(&ip->i_lock))
255 goto out_undo_mmaplock;
256 } else if (lock_flags & XFS_ILOCK_SHARED) {
257 if (!mrtryaccess(&ip->i_lock))
258 goto out_undo_mmaplock;
259 }
260 return 1;
261
262out_undo_mmaplock:
263 if (lock_flags & XFS_MMAPLOCK_EXCL)
264 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
265 else if (lock_flags & XFS_MMAPLOCK_SHARED)
266 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
267out_undo_iolock:
268 if (lock_flags & XFS_IOLOCK_EXCL)
269 up_write(&VFS_I(ip)->i_rwsem);
270 else if (lock_flags & XFS_IOLOCK_SHARED)
271 up_read(&VFS_I(ip)->i_rwsem);
272out:
273 return 0;
274}
275
276/*
277 * xfs_iunlock() is used to drop the inode locks acquired with
278 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
279 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
280 * that we know which locks to drop.
281 *
282 * ip -- the inode being unlocked
283 * lock_flags -- this parameter indicates the inode's locks to be
284 * to be unlocked. See the comment for xfs_ilock() for a list
285 * of valid values for this parameter.
286 *
287 */
288void
289xfs_iunlock(
290 xfs_inode_t *ip,
291 uint lock_flags)
292{
293 xfs_lock_flags_assert(lock_flags);
294
295 if (lock_flags & XFS_IOLOCK_EXCL)
296 up_write(&VFS_I(ip)->i_rwsem);
297 else if (lock_flags & XFS_IOLOCK_SHARED)
298 up_read(&VFS_I(ip)->i_rwsem);
299
300 if (lock_flags & XFS_MMAPLOCK_EXCL)
301 up_write(&VFS_I(ip)->i_mapping->invalidate_lock);
302 else if (lock_flags & XFS_MMAPLOCK_SHARED)
303 up_read(&VFS_I(ip)->i_mapping->invalidate_lock);
304
305 if (lock_flags & XFS_ILOCK_EXCL)
306 mrunlock_excl(&ip->i_lock);
307 else if (lock_flags & XFS_ILOCK_SHARED)
308 mrunlock_shared(&ip->i_lock);
309
310 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
311}
312
313/*
314 * give up write locks. the i/o lock cannot be held nested
315 * if it is being demoted.
316 */
317void
318xfs_ilock_demote(
319 xfs_inode_t *ip,
320 uint lock_flags)
321{
322 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
323 ASSERT((lock_flags &
324 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
325
326 if (lock_flags & XFS_ILOCK_EXCL)
327 mrdemote(&ip->i_lock);
328 if (lock_flags & XFS_MMAPLOCK_EXCL)
329 downgrade_write(&VFS_I(ip)->i_mapping->invalidate_lock);
330 if (lock_flags & XFS_IOLOCK_EXCL)
331 downgrade_write(&VFS_I(ip)->i_rwsem);
332
333 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
334}
335
336#if defined(DEBUG) || defined(XFS_WARN)
337static inline bool
338__xfs_rwsem_islocked(
339 struct rw_semaphore *rwsem,
340 bool shared)
341{
342 if (!debug_locks)
343 return rwsem_is_locked(rwsem);
344
345 if (!shared)
346 return lockdep_is_held_type(rwsem, 0);
347
348 /*
349 * We are checking that the lock is held at least in shared
350 * mode but don't care that it might be held exclusively
351 * (i.e. shared | excl). Hence we check if the lock is held
352 * in any mode rather than an explicit shared mode.
353 */
354 return lockdep_is_held_type(rwsem, -1);
355}
356
357bool
358xfs_isilocked(
359 struct xfs_inode *ip,
360 uint lock_flags)
361{
362 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
363 if (!(lock_flags & XFS_ILOCK_SHARED))
364 return !!ip->i_lock.mr_writer;
365 return rwsem_is_locked(&ip->i_lock.mr_lock);
366 }
367
368 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
369 return __xfs_rwsem_islocked(&VFS_I(ip)->i_mapping->invalidate_lock,
370 (lock_flags & XFS_MMAPLOCK_SHARED));
371 }
372
373 if (lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) {
374 return __xfs_rwsem_islocked(&VFS_I(ip)->i_rwsem,
375 (lock_flags & XFS_IOLOCK_SHARED));
376 }
377
378 ASSERT(0);
379 return false;
380}
381#endif
382
383/*
384 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
385 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
386 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
387 * errors and warnings.
388 */
389#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
390static bool
391xfs_lockdep_subclass_ok(
392 int subclass)
393{
394 return subclass < MAX_LOCKDEP_SUBCLASSES;
395}
396#else
397#define xfs_lockdep_subclass_ok(subclass) (true)
398#endif
399
400/*
401 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different
402 * value. This can be called for any type of inode lock combination, including
403 * parent locking. Care must be taken to ensure we don't overrun the subclass
404 * storage fields in the class mask we build.
405 */
406static inline uint
407xfs_lock_inumorder(
408 uint lock_mode,
409 uint subclass)
410{
411 uint class = 0;
412
413 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
414 XFS_ILOCK_RTSUM)));
415 ASSERT(xfs_lockdep_subclass_ok(subclass));
416
417 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
418 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
419 class += subclass << XFS_IOLOCK_SHIFT;
420 }
421
422 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
423 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
424 class += subclass << XFS_MMAPLOCK_SHIFT;
425 }
426
427 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
428 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
429 class += subclass << XFS_ILOCK_SHIFT;
430 }
431
432 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
433}
434
435/*
436 * The following routine will lock n inodes in exclusive mode. We assume the
437 * caller calls us with the inodes in i_ino order.
438 *
439 * We need to detect deadlock where an inode that we lock is in the AIL and we
440 * start waiting for another inode that is locked by a thread in a long running
441 * transaction (such as truncate). This can result in deadlock since the long
442 * running trans might need to wait for the inode we just locked in order to
443 * push the tail and free space in the log.
444 *
445 * xfs_lock_inodes() can only be used to lock one type of lock at a time -
446 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we
447 * lock more than one at a time, lockdep will report false positives saying we
448 * have violated locking orders.
449 */
450static void
451xfs_lock_inodes(
452 struct xfs_inode **ips,
453 int inodes,
454 uint lock_mode)
455{
456 int attempts = 0;
457 uint i;
458 int j;
459 bool try_lock;
460 struct xfs_log_item *lp;
461
462 /*
463 * Currently supports between 2 and 5 inodes with exclusive locking. We
464 * support an arbitrary depth of locking here, but absolute limits on
465 * inodes depend on the type of locking and the limits placed by
466 * lockdep annotations in xfs_lock_inumorder. These are all checked by
467 * the asserts.
468 */
469 ASSERT(ips && inodes >= 2 && inodes <= 5);
470 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
471 XFS_ILOCK_EXCL));
472 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
473 XFS_ILOCK_SHARED)));
474 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
475 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
476 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
477 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
478
479 if (lock_mode & XFS_IOLOCK_EXCL) {
480 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
481 } else if (lock_mode & XFS_MMAPLOCK_EXCL)
482 ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
483
484again:
485 try_lock = false;
486 i = 0;
487 for (; i < inodes; i++) {
488 ASSERT(ips[i]);
489
490 if (i && (ips[i] == ips[i - 1])) /* Already locked */
491 continue;
492
493 /*
494 * If try_lock is not set yet, make sure all locked inodes are
495 * not in the AIL. If any are, set try_lock to be used later.
496 */
497 if (!try_lock) {
498 for (j = (i - 1); j >= 0 && !try_lock; j--) {
499 lp = &ips[j]->i_itemp->ili_item;
500 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
501 try_lock = true;
502 }
503 }
504
505 /*
506 * If any of the previous locks we have locked is in the AIL,
507 * we must TRY to get the second and subsequent locks. If
508 * we can't get any, we must release all we have
509 * and try again.
510 */
511 if (!try_lock) {
512 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
513 continue;
514 }
515
516 /* try_lock means we have an inode locked that is in the AIL. */
517 ASSERT(i != 0);
518 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
519 continue;
520
521 /*
522 * Unlock all previous guys and try again. xfs_iunlock will try
523 * to push the tail if the inode is in the AIL.
524 */
525 attempts++;
526 for (j = i - 1; j >= 0; j--) {
527 /*
528 * Check to see if we've already unlocked this one. Not
529 * the first one going back, and the inode ptr is the
530 * same.
531 */
532 if (j != (i - 1) && ips[j] == ips[j + 1])
533 continue;
534
535 xfs_iunlock(ips[j], lock_mode);
536 }
537
538 if ((attempts % 5) == 0) {
539 delay(1); /* Don't just spin the CPU */
540 }
541 goto again;
542 }
543}
544
545/*
546 * xfs_lock_two_inodes() can only be used to lock ilock. The iolock and
547 * mmaplock must be double-locked separately since we use i_rwsem and
548 * invalidate_lock for that. We now support taking one lock EXCL and the
549 * other SHARED.
550 */
551void
552xfs_lock_two_inodes(
553 struct xfs_inode *ip0,
554 uint ip0_mode,
555 struct xfs_inode *ip1,
556 uint ip1_mode)
557{
558 int attempts = 0;
559 struct xfs_log_item *lp;
560
561 ASSERT(hweight32(ip0_mode) == 1);
562 ASSERT(hweight32(ip1_mode) == 1);
563 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
564 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
565 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
566 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)));
567 ASSERT(ip0->i_ino != ip1->i_ino);
568
569 if (ip0->i_ino > ip1->i_ino) {
570 swap(ip0, ip1);
571 swap(ip0_mode, ip1_mode);
572 }
573
574 again:
575 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
576
577 /*
578 * If the first lock we have locked is in the AIL, we must TRY to get
579 * the second lock. If we can't get it, we must release the first one
580 * and try again.
581 */
582 lp = &ip0->i_itemp->ili_item;
583 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
584 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
585 xfs_iunlock(ip0, ip0_mode);
586 if ((++attempts % 5) == 0)
587 delay(1); /* Don't just spin the CPU */
588 goto again;
589 }
590 } else {
591 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
592 }
593}
594
595uint
596xfs_ip2xflags(
597 struct xfs_inode *ip)
598{
599 uint flags = 0;
600
601 if (ip->i_diflags & XFS_DIFLAG_ANY) {
602 if (ip->i_diflags & XFS_DIFLAG_REALTIME)
603 flags |= FS_XFLAG_REALTIME;
604 if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
605 flags |= FS_XFLAG_PREALLOC;
606 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
607 flags |= FS_XFLAG_IMMUTABLE;
608 if (ip->i_diflags & XFS_DIFLAG_APPEND)
609 flags |= FS_XFLAG_APPEND;
610 if (ip->i_diflags & XFS_DIFLAG_SYNC)
611 flags |= FS_XFLAG_SYNC;
612 if (ip->i_diflags & XFS_DIFLAG_NOATIME)
613 flags |= FS_XFLAG_NOATIME;
614 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
615 flags |= FS_XFLAG_NODUMP;
616 if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
617 flags |= FS_XFLAG_RTINHERIT;
618 if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
619 flags |= FS_XFLAG_PROJINHERIT;
620 if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
621 flags |= FS_XFLAG_NOSYMLINKS;
622 if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
623 flags |= FS_XFLAG_EXTSIZE;
624 if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
625 flags |= FS_XFLAG_EXTSZINHERIT;
626 if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
627 flags |= FS_XFLAG_NODEFRAG;
628 if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
629 flags |= FS_XFLAG_FILESTREAM;
630 }
631
632 if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
633 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
634 flags |= FS_XFLAG_DAX;
635 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
636 flags |= FS_XFLAG_COWEXTSIZE;
637 }
638
639 if (xfs_inode_has_attr_fork(ip))
640 flags |= FS_XFLAG_HASATTR;
641 return flags;
642}
643
644/*
645 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
646 * is allowed, otherwise it has to be an exact match. If a CI match is found,
647 * ci_name->name will point to a the actual name (caller must free) or
648 * will be set to NULL if an exact match is found.
649 */
650int
651xfs_lookup(
652 struct xfs_inode *dp,
653 const struct xfs_name *name,
654 struct xfs_inode **ipp,
655 struct xfs_name *ci_name)
656{
657 xfs_ino_t inum;
658 int error;
659
660 trace_xfs_lookup(dp, name);
661
662 if (xfs_is_shutdown(dp->i_mount))
663 return -EIO;
664
665 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
666 if (error)
667 goto out_unlock;
668
669 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
670 if (error)
671 goto out_free_name;
672
673 return 0;
674
675out_free_name:
676 if (ci_name)
677 kmem_free(ci_name->name);
678out_unlock:
679 *ipp = NULL;
680 return error;
681}
682
683/* Propagate di_flags from a parent inode to a child inode. */
684static void
685xfs_inode_inherit_flags(
686 struct xfs_inode *ip,
687 const struct xfs_inode *pip)
688{
689 unsigned int di_flags = 0;
690 xfs_failaddr_t failaddr;
691 umode_t mode = VFS_I(ip)->i_mode;
692
693 if (S_ISDIR(mode)) {
694 if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
695 di_flags |= XFS_DIFLAG_RTINHERIT;
696 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
697 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
698 ip->i_extsize = pip->i_extsize;
699 }
700 if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
701 di_flags |= XFS_DIFLAG_PROJINHERIT;
702 } else if (S_ISREG(mode)) {
703 if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
704 xfs_has_realtime(ip->i_mount))
705 di_flags |= XFS_DIFLAG_REALTIME;
706 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
707 di_flags |= XFS_DIFLAG_EXTSIZE;
708 ip->i_extsize = pip->i_extsize;
709 }
710 }
711 if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
712 xfs_inherit_noatime)
713 di_flags |= XFS_DIFLAG_NOATIME;
714 if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
715 xfs_inherit_nodump)
716 di_flags |= XFS_DIFLAG_NODUMP;
717 if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
718 xfs_inherit_sync)
719 di_flags |= XFS_DIFLAG_SYNC;
720 if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
721 xfs_inherit_nosymlinks)
722 di_flags |= XFS_DIFLAG_NOSYMLINKS;
723 if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
724 xfs_inherit_nodefrag)
725 di_flags |= XFS_DIFLAG_NODEFRAG;
726 if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
727 di_flags |= XFS_DIFLAG_FILESTREAM;
728
729 ip->i_diflags |= di_flags;
730
731 /*
732 * Inode verifiers on older kernels only check that the extent size
733 * hint is an integer multiple of the rt extent size on realtime files.
734 * They did not check the hint alignment on a directory with both
735 * rtinherit and extszinherit flags set. If the misaligned hint is
736 * propagated from a directory into a new realtime file, new file
737 * allocations will fail due to math errors in the rt allocator and/or
738 * trip the verifiers. Validate the hint settings in the new file so
739 * that we don't let broken hints propagate.
740 */
741 failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
742 VFS_I(ip)->i_mode, ip->i_diflags);
743 if (failaddr) {
744 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
745 XFS_DIFLAG_EXTSZINHERIT);
746 ip->i_extsize = 0;
747 }
748}
749
750/* Propagate di_flags2 from a parent inode to a child inode. */
751static void
752xfs_inode_inherit_flags2(
753 struct xfs_inode *ip,
754 const struct xfs_inode *pip)
755{
756 xfs_failaddr_t failaddr;
757
758 if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
759 ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
760 ip->i_cowextsize = pip->i_cowextsize;
761 }
762 if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
763 ip->i_diflags2 |= XFS_DIFLAG2_DAX;
764
765 /* Don't let invalid cowextsize hints propagate. */
766 failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
767 VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
768 if (failaddr) {
769 ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
770 ip->i_cowextsize = 0;
771 }
772}
773
774/*
775 * Initialise a newly allocated inode and return the in-core inode to the
776 * caller locked exclusively.
777 */
778int
779xfs_init_new_inode(
780 struct user_namespace *mnt_userns,
781 struct xfs_trans *tp,
782 struct xfs_inode *pip,
783 xfs_ino_t ino,
784 umode_t mode,
785 xfs_nlink_t nlink,
786 dev_t rdev,
787 prid_t prid,
788 bool init_xattrs,
789 struct xfs_inode **ipp)
790{
791 struct inode *dir = pip ? VFS_I(pip) : NULL;
792 struct xfs_mount *mp = tp->t_mountp;
793 struct xfs_inode *ip;
794 unsigned int flags;
795 int error;
796 struct timespec64 tv;
797 struct inode *inode;
798
799 /*
800 * Protect against obviously corrupt allocation btree records. Later
801 * xfs_iget checks will catch re-allocation of other active in-memory
802 * and on-disk inodes. If we don't catch reallocating the parent inode
803 * here we will deadlock in xfs_iget() so we have to do these checks
804 * first.
805 */
806 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
807 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
808 return -EFSCORRUPTED;
809 }
810
811 /*
812 * Get the in-core inode with the lock held exclusively to prevent
813 * others from looking at until we're done.
814 */
815 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
816 if (error)
817 return error;
818
819 ASSERT(ip != NULL);
820 inode = VFS_I(ip);
821 set_nlink(inode, nlink);
822 inode->i_rdev = rdev;
823 ip->i_projid = prid;
824
825 if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) {
826 inode_fsuid_set(inode, mnt_userns);
827 inode->i_gid = dir->i_gid;
828 inode->i_mode = mode;
829 } else {
830 inode_init_owner(mnt_userns, inode, dir, mode);
831 }
832
833 /*
834 * If the group ID of the new file does not match the effective group
835 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
836 * (and only if the irix_sgid_inherit compatibility variable is set).
837 */
838 if (irix_sgid_inherit && (inode->i_mode & S_ISGID) &&
839 !vfsgid_in_group_p(i_gid_into_vfsgid(mnt_userns, inode)))
840 inode->i_mode &= ~S_ISGID;
841
842 ip->i_disk_size = 0;
843 ip->i_df.if_nextents = 0;
844 ASSERT(ip->i_nblocks == 0);
845
846 tv = current_time(inode);
847 inode->i_mtime = tv;
848 inode->i_atime = tv;
849 inode->i_ctime = tv;
850
851 ip->i_extsize = 0;
852 ip->i_diflags = 0;
853
854 if (xfs_has_v3inodes(mp)) {
855 inode_set_iversion(inode, 1);
856 ip->i_cowextsize = 0;
857 ip->i_crtime = tv;
858 }
859
860 flags = XFS_ILOG_CORE;
861 switch (mode & S_IFMT) {
862 case S_IFIFO:
863 case S_IFCHR:
864 case S_IFBLK:
865 case S_IFSOCK:
866 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
867 flags |= XFS_ILOG_DEV;
868 break;
869 case S_IFREG:
870 case S_IFDIR:
871 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
872 xfs_inode_inherit_flags(ip, pip);
873 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
874 xfs_inode_inherit_flags2(ip, pip);
875 fallthrough;
876 case S_IFLNK:
877 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
878 ip->i_df.if_bytes = 0;
879 ip->i_df.if_u1.if_root = NULL;
880 break;
881 default:
882 ASSERT(0);
883 }
884
885 /*
886 * If we need to create attributes immediately after allocating the
887 * inode, initialise an empty attribute fork right now. We use the
888 * default fork offset for attributes here as we don't know exactly what
889 * size or how many attributes we might be adding. We can do this
890 * safely here because we know the data fork is completely empty and
891 * this saves us from needing to run a separate transaction to set the
892 * fork offset in the immediate future.
893 */
894 if (init_xattrs && xfs_has_attr(mp)) {
895 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
896 xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
897 }
898
899 /*
900 * Log the new values stuffed into the inode.
901 */
902 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
903 xfs_trans_log_inode(tp, ip, flags);
904
905 /* now that we have an i_mode we can setup the inode structure */
906 xfs_setup_inode(ip);
907
908 *ipp = ip;
909 return 0;
910}
911
912/*
913 * Decrement the link count on an inode & log the change. If this causes the
914 * link count to go to zero, move the inode to AGI unlinked list so that it can
915 * be freed when the last active reference goes away via xfs_inactive().
916 */
917static int /* error */
918xfs_droplink(
919 xfs_trans_t *tp,
920 xfs_inode_t *ip)
921{
922 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
923
924 drop_nlink(VFS_I(ip));
925 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
926
927 if (VFS_I(ip)->i_nlink)
928 return 0;
929
930 return xfs_iunlink(tp, ip);
931}
932
933/*
934 * Increment the link count on an inode & log the change.
935 */
936static void
937xfs_bumplink(
938 xfs_trans_t *tp,
939 xfs_inode_t *ip)
940{
941 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
942
943 inc_nlink(VFS_I(ip));
944 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
945}
946
947int
948xfs_create(
949 struct user_namespace *mnt_userns,
950 xfs_inode_t *dp,
951 struct xfs_name *name,
952 umode_t mode,
953 dev_t rdev,
954 bool init_xattrs,
955 xfs_inode_t **ipp)
956{
957 int is_dir = S_ISDIR(mode);
958 struct xfs_mount *mp = dp->i_mount;
959 struct xfs_inode *ip = NULL;
960 struct xfs_trans *tp = NULL;
961 int error;
962 bool unlock_dp_on_error = false;
963 prid_t prid;
964 struct xfs_dquot *udqp = NULL;
965 struct xfs_dquot *gdqp = NULL;
966 struct xfs_dquot *pdqp = NULL;
967 struct xfs_trans_res *tres;
968 uint resblks;
969 xfs_ino_t ino;
970
971 trace_xfs_create(dp, name);
972
973 if (xfs_is_shutdown(mp))
974 return -EIO;
975
976 prid = xfs_get_initial_prid(dp);
977
978 /*
979 * Make sure that we have allocated dquot(s) on disk.
980 */
981 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns, &init_user_ns),
982 mapped_fsgid(mnt_userns, &init_user_ns), prid,
983 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
984 &udqp, &gdqp, &pdqp);
985 if (error)
986 return error;
987
988 if (is_dir) {
989 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
990 tres = &M_RES(mp)->tr_mkdir;
991 } else {
992 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
993 tres = &M_RES(mp)->tr_create;
994 }
995
996 /*
997 * Initially assume that the file does not exist and
998 * reserve the resources for that case. If that is not
999 * the case we'll drop the one we have and get a more
1000 * appropriate transaction later.
1001 */
1002 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1003 &tp);
1004 if (error == -ENOSPC) {
1005 /* flush outstanding delalloc blocks and retry */
1006 xfs_flush_inodes(mp);
1007 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
1008 resblks, &tp);
1009 }
1010 if (error)
1011 goto out_release_dquots;
1012
1013 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1014 unlock_dp_on_error = true;
1015
1016 /*
1017 * A newly created regular or special file just has one directory
1018 * entry pointing to them, but a directory also the "." entry
1019 * pointing to itself.
1020 */
1021 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1022 if (!error)
1023 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1024 is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
1025 if (error)
1026 goto out_trans_cancel;
1027
1028 /*
1029 * Now we join the directory inode to the transaction. We do not do it
1030 * earlier because xfs_dialloc might commit the previous transaction
1031 * (and release all the locks). An error from here on will result in
1032 * the transaction cancel unlocking dp so don't do it explicitly in the
1033 * error path.
1034 */
1035 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1036 unlock_dp_on_error = false;
1037
1038 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1039 resblks - XFS_IALLOC_SPACE_RES(mp));
1040 if (error) {
1041 ASSERT(error != -ENOSPC);
1042 goto out_trans_cancel;
1043 }
1044 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1045 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1046
1047 if (is_dir) {
1048 error = xfs_dir_init(tp, ip, dp);
1049 if (error)
1050 goto out_trans_cancel;
1051
1052 xfs_bumplink(tp, dp);
1053 }
1054
1055 /*
1056 * If this is a synchronous mount, make sure that the
1057 * create transaction goes to disk before returning to
1058 * the user.
1059 */
1060 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1061 xfs_trans_set_sync(tp);
1062
1063 /*
1064 * Attach the dquot(s) to the inodes and modify them incore.
1065 * These ids of the inode couldn't have changed since the new
1066 * inode has been locked ever since it was created.
1067 */
1068 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1069
1070 error = xfs_trans_commit(tp);
1071 if (error)
1072 goto out_release_inode;
1073
1074 xfs_qm_dqrele(udqp);
1075 xfs_qm_dqrele(gdqp);
1076 xfs_qm_dqrele(pdqp);
1077
1078 *ipp = ip;
1079 return 0;
1080
1081 out_trans_cancel:
1082 xfs_trans_cancel(tp);
1083 out_release_inode:
1084 /*
1085 * Wait until after the current transaction is aborted to finish the
1086 * setup of the inode and release the inode. This prevents recursive
1087 * transactions and deadlocks from xfs_inactive.
1088 */
1089 if (ip) {
1090 xfs_finish_inode_setup(ip);
1091 xfs_irele(ip);
1092 }
1093 out_release_dquots:
1094 xfs_qm_dqrele(udqp);
1095 xfs_qm_dqrele(gdqp);
1096 xfs_qm_dqrele(pdqp);
1097
1098 if (unlock_dp_on_error)
1099 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1100 return error;
1101}
1102
1103int
1104xfs_create_tmpfile(
1105 struct user_namespace *mnt_userns,
1106 struct xfs_inode *dp,
1107 umode_t mode,
1108 struct xfs_inode **ipp)
1109{
1110 struct xfs_mount *mp = dp->i_mount;
1111 struct xfs_inode *ip = NULL;
1112 struct xfs_trans *tp = NULL;
1113 int error;
1114 prid_t prid;
1115 struct xfs_dquot *udqp = NULL;
1116 struct xfs_dquot *gdqp = NULL;
1117 struct xfs_dquot *pdqp = NULL;
1118 struct xfs_trans_res *tres;
1119 uint resblks;
1120 xfs_ino_t ino;
1121
1122 if (xfs_is_shutdown(mp))
1123 return -EIO;
1124
1125 prid = xfs_get_initial_prid(dp);
1126
1127 /*
1128 * Make sure that we have allocated dquot(s) on disk.
1129 */
1130 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns, &init_user_ns),
1131 mapped_fsgid(mnt_userns, &init_user_ns), prid,
1132 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1133 &udqp, &gdqp, &pdqp);
1134 if (error)
1135 return error;
1136
1137 resblks = XFS_IALLOC_SPACE_RES(mp);
1138 tres = &M_RES(mp)->tr_create_tmpfile;
1139
1140 error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
1141 &tp);
1142 if (error)
1143 goto out_release_dquots;
1144
1145 error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
1146 if (!error)
1147 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
1148 0, 0, prid, false, &ip);
1149 if (error)
1150 goto out_trans_cancel;
1151
1152 if (xfs_has_wsync(mp))
1153 xfs_trans_set_sync(tp);
1154
1155 /*
1156 * Attach the dquot(s) to the inodes and modify them incore.
1157 * These ids of the inode couldn't have changed since the new
1158 * inode has been locked ever since it was created.
1159 */
1160 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1161
1162 error = xfs_iunlink(tp, ip);
1163 if (error)
1164 goto out_trans_cancel;
1165
1166 error = xfs_trans_commit(tp);
1167 if (error)
1168 goto out_release_inode;
1169
1170 xfs_qm_dqrele(udqp);
1171 xfs_qm_dqrele(gdqp);
1172 xfs_qm_dqrele(pdqp);
1173
1174 *ipp = ip;
1175 return 0;
1176
1177 out_trans_cancel:
1178 xfs_trans_cancel(tp);
1179 out_release_inode:
1180 /*
1181 * Wait until after the current transaction is aborted to finish the
1182 * setup of the inode and release the inode. This prevents recursive
1183 * transactions and deadlocks from xfs_inactive.
1184 */
1185 if (ip) {
1186 xfs_finish_inode_setup(ip);
1187 xfs_irele(ip);
1188 }
1189 out_release_dquots:
1190 xfs_qm_dqrele(udqp);
1191 xfs_qm_dqrele(gdqp);
1192 xfs_qm_dqrele(pdqp);
1193
1194 return error;
1195}
1196
1197int
1198xfs_link(
1199 xfs_inode_t *tdp,
1200 xfs_inode_t *sip,
1201 struct xfs_name *target_name)
1202{
1203 xfs_mount_t *mp = tdp->i_mount;
1204 xfs_trans_t *tp;
1205 int error, nospace_error = 0;
1206 int resblks;
1207
1208 trace_xfs_link(tdp, target_name);
1209
1210 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
1211
1212 if (xfs_is_shutdown(mp))
1213 return -EIO;
1214
1215 error = xfs_qm_dqattach(sip);
1216 if (error)
1217 goto std_return;
1218
1219 error = xfs_qm_dqattach(tdp);
1220 if (error)
1221 goto std_return;
1222
1223 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1224 error = xfs_trans_alloc_dir(tdp, &M_RES(mp)->tr_link, sip, &resblks,
1225 &tp, &nospace_error);
1226 if (error)
1227 goto std_return;
1228
1229 /*
1230 * If we are using project inheritance, we only allow hard link
1231 * creation in our tree when the project IDs are the same; else
1232 * the tree quota mechanism could be circumvented.
1233 */
1234 if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
1235 tdp->i_projid != sip->i_projid)) {
1236 error = -EXDEV;
1237 goto error_return;
1238 }
1239
1240 if (!resblks) {
1241 error = xfs_dir_canenter(tp, tdp, target_name);
1242 if (error)
1243 goto error_return;
1244 }
1245
1246 /*
1247 * Handle initial link state of O_TMPFILE inode
1248 */
1249 if (VFS_I(sip)->i_nlink == 0) {
1250 struct xfs_perag *pag;
1251
1252 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
1253 error = xfs_iunlink_remove(tp, pag, sip);
1254 xfs_perag_put(pag);
1255 if (error)
1256 goto error_return;
1257 }
1258
1259 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1260 resblks);
1261 if (error)
1262 goto error_return;
1263 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1264 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1265
1266 xfs_bumplink(tp, sip);
1267
1268 /*
1269 * If this is a synchronous mount, make sure that the
1270 * link transaction goes to disk before returning to
1271 * the user.
1272 */
1273 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
1274 xfs_trans_set_sync(tp);
1275
1276 return xfs_trans_commit(tp);
1277
1278 error_return:
1279 xfs_trans_cancel(tp);
1280 std_return:
1281 if (error == -ENOSPC && nospace_error)
1282 error = nospace_error;
1283 return error;
1284}
1285
1286/* Clear the reflink flag and the cowblocks tag if possible. */
1287static void
1288xfs_itruncate_clear_reflink_flags(
1289 struct xfs_inode *ip)
1290{
1291 struct xfs_ifork *dfork;
1292 struct xfs_ifork *cfork;
1293
1294 if (!xfs_is_reflink_inode(ip))
1295 return;
1296 dfork = xfs_ifork_ptr(ip, XFS_DATA_FORK);
1297 cfork = xfs_ifork_ptr(ip, XFS_COW_FORK);
1298 if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
1299 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1300 if (cfork->if_bytes == 0)
1301 xfs_inode_clear_cowblocks_tag(ip);
1302}
1303
1304/*
1305 * Free up the underlying blocks past new_size. The new size must be smaller
1306 * than the current size. This routine can be used both for the attribute and
1307 * data fork, and does not modify the inode size, which is left to the caller.
1308 *
1309 * The transaction passed to this routine must have made a permanent log
1310 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1311 * given transaction and start new ones, so make sure everything involved in
1312 * the transaction is tidy before calling here. Some transaction will be
1313 * returned to the caller to be committed. The incoming transaction must
1314 * already include the inode, and both inode locks must be held exclusively.
1315 * The inode must also be "held" within the transaction. On return the inode
1316 * will be "held" within the returned transaction. This routine does NOT
1317 * require any disk space to be reserved for it within the transaction.
1318 *
1319 * If we get an error, we must return with the inode locked and linked into the
1320 * current transaction. This keeps things simple for the higher level code,
1321 * because it always knows that the inode is locked and held in the transaction
1322 * that returns to it whether errors occur or not. We don't mark the inode
1323 * dirty on error so that transactions can be easily aborted if possible.
1324 */
1325int
1326xfs_itruncate_extents_flags(
1327 struct xfs_trans **tpp,
1328 struct xfs_inode *ip,
1329 int whichfork,
1330 xfs_fsize_t new_size,
1331 int flags)
1332{
1333 struct xfs_mount *mp = ip->i_mount;
1334 struct xfs_trans *tp = *tpp;
1335 xfs_fileoff_t first_unmap_block;
1336 xfs_filblks_t unmap_len;
1337 int error = 0;
1338
1339 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1340 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1341 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1342 ASSERT(new_size <= XFS_ISIZE(ip));
1343 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1344 ASSERT(ip->i_itemp != NULL);
1345 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1346 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1347
1348 trace_xfs_itruncate_extents_start(ip, new_size);
1349
1350 flags |= xfs_bmapi_aflag(whichfork);
1351
1352 /*
1353 * Since it is possible for space to become allocated beyond
1354 * the end of the file (in a crash where the space is allocated
1355 * but the inode size is not yet updated), simply remove any
1356 * blocks which show up between the new EOF and the maximum
1357 * possible file size.
1358 *
1359 * We have to free all the blocks to the bmbt maximum offset, even if
1360 * the page cache can't scale that far.
1361 */
1362 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1363 if (!xfs_verify_fileoff(mp, first_unmap_block)) {
1364 WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
1365 return 0;
1366 }
1367
1368 unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
1369 while (unmap_len > 0) {
1370 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1371 error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
1372 flags, XFS_ITRUNC_MAX_EXTENTS);
1373 if (error)
1374 goto out;
1375
1376 /* free the just unmapped extents */
1377 error = xfs_defer_finish(&tp);
1378 if (error)
1379 goto out;
1380 }
1381
1382 if (whichfork == XFS_DATA_FORK) {
1383 /* Remove all pending CoW reservations. */
1384 error = xfs_reflink_cancel_cow_blocks(ip, &tp,
1385 first_unmap_block, XFS_MAX_FILEOFF, true);
1386 if (error)
1387 goto out;
1388
1389 xfs_itruncate_clear_reflink_flags(ip);
1390 }
1391
1392 /*
1393 * Always re-log the inode so that our permanent transaction can keep
1394 * on rolling it forward in the log.
1395 */
1396 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1397
1398 trace_xfs_itruncate_extents_end(ip, new_size);
1399
1400out:
1401 *tpp = tp;
1402 return error;
1403}
1404
1405int
1406xfs_release(
1407 xfs_inode_t *ip)
1408{
1409 xfs_mount_t *mp = ip->i_mount;
1410 int error = 0;
1411
1412 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
1413 return 0;
1414
1415 /* If this is a read-only mount, don't do this (would generate I/O) */
1416 if (xfs_is_readonly(mp))
1417 return 0;
1418
1419 if (!xfs_is_shutdown(mp)) {
1420 int truncated;
1421
1422 /*
1423 * If we previously truncated this file and removed old data
1424 * in the process, we want to initiate "early" writeout on
1425 * the last close. This is an attempt to combat the notorious
1426 * NULL files problem which is particularly noticeable from a
1427 * truncate down, buffered (re-)write (delalloc), followed by
1428 * a crash. What we are effectively doing here is
1429 * significantly reducing the time window where we'd otherwise
1430 * be exposed to that problem.
1431 */
1432 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1433 if (truncated) {
1434 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1435 if (ip->i_delayed_blks > 0) {
1436 error = filemap_flush(VFS_I(ip)->i_mapping);
1437 if (error)
1438 return error;
1439 }
1440 }
1441 }
1442
1443 if (VFS_I(ip)->i_nlink == 0)
1444 return 0;
1445
1446 /*
1447 * If we can't get the iolock just skip truncating the blocks past EOF
1448 * because we could deadlock with the mmap_lock otherwise. We'll get
1449 * another chance to drop them once the last reference to the inode is
1450 * dropped, so we'll never leak blocks permanently.
1451 */
1452 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
1453 return 0;
1454
1455 if (xfs_can_free_eofblocks(ip, false)) {
1456 /*
1457 * Check if the inode is being opened, written and closed
1458 * frequently and we have delayed allocation blocks outstanding
1459 * (e.g. streaming writes from the NFS server), truncating the
1460 * blocks past EOF will cause fragmentation to occur.
1461 *
1462 * In this case don't do the truncation, but we have to be
1463 * careful how we detect this case. Blocks beyond EOF show up as
1464 * i_delayed_blks even when the inode is clean, so we need to
1465 * truncate them away first before checking for a dirty release.
1466 * Hence on the first dirty close we will still remove the
1467 * speculative allocation, but after that we will leave it in
1468 * place.
1469 */
1470 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1471 goto out_unlock;
1472
1473 error = xfs_free_eofblocks(ip);
1474 if (error)
1475 goto out_unlock;
1476
1477 /* delalloc blocks after truncation means it really is dirty */
1478 if (ip->i_delayed_blks)
1479 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1480 }
1481
1482out_unlock:
1483 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1484 return error;
1485}
1486
1487/*
1488 * xfs_inactive_truncate
1489 *
1490 * Called to perform a truncate when an inode becomes unlinked.
1491 */
1492STATIC int
1493xfs_inactive_truncate(
1494 struct xfs_inode *ip)
1495{
1496 struct xfs_mount *mp = ip->i_mount;
1497 struct xfs_trans *tp;
1498 int error;
1499
1500 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
1501 if (error) {
1502 ASSERT(xfs_is_shutdown(mp));
1503 return error;
1504 }
1505 xfs_ilock(ip, XFS_ILOCK_EXCL);
1506 xfs_trans_ijoin(tp, ip, 0);
1507
1508 /*
1509 * Log the inode size first to prevent stale data exposure in the event
1510 * of a system crash before the truncate completes. See the related
1511 * comment in xfs_vn_setattr_size() for details.
1512 */
1513 ip->i_disk_size = 0;
1514 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1515
1516 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1517 if (error)
1518 goto error_trans_cancel;
1519
1520 ASSERT(ip->i_df.if_nextents == 0);
1521
1522 error = xfs_trans_commit(tp);
1523 if (error)
1524 goto error_unlock;
1525
1526 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1527 return 0;
1528
1529error_trans_cancel:
1530 xfs_trans_cancel(tp);
1531error_unlock:
1532 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1533 return error;
1534}
1535
1536/*
1537 * xfs_inactive_ifree()
1538 *
1539 * Perform the inode free when an inode is unlinked.
1540 */
1541STATIC int
1542xfs_inactive_ifree(
1543 struct xfs_inode *ip)
1544{
1545 struct xfs_mount *mp = ip->i_mount;
1546 struct xfs_trans *tp;
1547 int error;
1548
1549 /*
1550 * We try to use a per-AG reservation for any block needed by the finobt
1551 * tree, but as the finobt feature predates the per-AG reservation
1552 * support a degraded file system might not have enough space for the
1553 * reservation at mount time. In that case try to dip into the reserved
1554 * pool and pray.
1555 *
1556 * Send a warning if the reservation does happen to fail, as the inode
1557 * now remains allocated and sits on the unlinked list until the fs is
1558 * repaired.
1559 */
1560 if (unlikely(mp->m_finobt_nores)) {
1561 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
1562 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
1563 &tp);
1564 } else {
1565 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
1566 }
1567 if (error) {
1568 if (error == -ENOSPC) {
1569 xfs_warn_ratelimited(mp,
1570 "Failed to remove inode(s) from unlinked list. "
1571 "Please free space, unmount and run xfs_repair.");
1572 } else {
1573 ASSERT(xfs_is_shutdown(mp));
1574 }
1575 return error;
1576 }
1577
1578 /*
1579 * We do not hold the inode locked across the entire rolling transaction
1580 * here. We only need to hold it for the first transaction that
1581 * xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
1582 * underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
1583 * here breaks the relationship between cluster buffer invalidation and
1584 * stale inode invalidation on cluster buffer item journal commit
1585 * completion, and can result in leaving dirty stale inodes hanging
1586 * around in memory.
1587 *
1588 * We have no need for serialising this inode operation against other
1589 * operations - we freed the inode and hence reallocation is required
1590 * and that will serialise on reallocating the space the deferops need
1591 * to free. Hence we can unlock the inode on the first commit of
1592 * the transaction rather than roll it right through the deferops. This
1593 * avoids relogging the XFS_ISTALE inode.
1594 *
1595 * We check that xfs_ifree() hasn't grown an internal transaction roll
1596 * by asserting that the inode is still locked when it returns.
1597 */
1598 xfs_ilock(ip, XFS_ILOCK_EXCL);
1599 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1600
1601 error = xfs_ifree(tp, ip);
1602 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1603 if (error) {
1604 /*
1605 * If we fail to free the inode, shut down. The cancel
1606 * might do that, we need to make sure. Otherwise the
1607 * inode might be lost for a long time or forever.
1608 */
1609 if (!xfs_is_shutdown(mp)) {
1610 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1611 __func__, error);
1612 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1613 }
1614 xfs_trans_cancel(tp);
1615 return error;
1616 }
1617
1618 /*
1619 * Credit the quota account(s). The inode is gone.
1620 */
1621 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1622
1623 /*
1624 * Just ignore errors at this point. There is nothing we can do except
1625 * to try to keep going. Make sure it's not a silent error.
1626 */
1627 error = xfs_trans_commit(tp);
1628 if (error)
1629 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1630 __func__, error);
1631
1632 return 0;
1633}
1634
1635/*
1636 * Returns true if we need to update the on-disk metadata before we can free
1637 * the memory used by this inode. Updates include freeing post-eof
1638 * preallocations; freeing COW staging extents; and marking the inode free in
1639 * the inobt if it is on the unlinked list.
1640 */
1641bool
1642xfs_inode_needs_inactive(
1643 struct xfs_inode *ip)
1644{
1645 struct xfs_mount *mp = ip->i_mount;
1646 struct xfs_ifork *cow_ifp = xfs_ifork_ptr(ip, XFS_COW_FORK);
1647
1648 /*
1649 * If the inode is already free, then there can be nothing
1650 * to clean up here.
1651 */
1652 if (VFS_I(ip)->i_mode == 0)
1653 return false;
1654
1655 /* If this is a read-only mount, don't do this (would generate I/O) */
1656 if (xfs_is_readonly(mp))
1657 return false;
1658
1659 /* If the log isn't running, push inodes straight to reclaim. */
1660 if (xfs_is_shutdown(mp) || xfs_has_norecovery(mp))
1661 return false;
1662
1663 /* Metadata inodes require explicit resource cleanup. */
1664 if (xfs_is_metadata_inode(ip))
1665 return false;
1666
1667 /* Want to clean out the cow blocks if there are any. */
1668 if (cow_ifp && cow_ifp->if_bytes > 0)
1669 return true;
1670
1671 /* Unlinked files must be freed. */
1672 if (VFS_I(ip)->i_nlink == 0)
1673 return true;
1674
1675 /*
1676 * This file isn't being freed, so check if there are post-eof blocks
1677 * to free. @force is true because we are evicting an inode from the
1678 * cache. Post-eof blocks must be freed, lest we end up with broken
1679 * free space accounting.
1680 *
1681 * Note: don't bother with iolock here since lockdep complains about
1682 * acquiring it in reclaim context. We have the only reference to the
1683 * inode at this point anyways.
1684 */
1685 return xfs_can_free_eofblocks(ip, true);
1686}
1687
1688/*
1689 * xfs_inactive
1690 *
1691 * This is called when the vnode reference count for the vnode
1692 * goes to zero. If the file has been unlinked, then it must
1693 * now be truncated. Also, we clear all of the read-ahead state
1694 * kept for the inode here since the file is now closed.
1695 */
1696void
1697xfs_inactive(
1698 xfs_inode_t *ip)
1699{
1700 struct xfs_mount *mp;
1701 int error;
1702 int truncate = 0;
1703
1704 /*
1705 * If the inode is already free, then there can be nothing
1706 * to clean up here.
1707 */
1708 if (VFS_I(ip)->i_mode == 0) {
1709 ASSERT(ip->i_df.if_broot_bytes == 0);
1710 goto out;
1711 }
1712
1713 mp = ip->i_mount;
1714 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
1715
1716 /* If this is a read-only mount, don't do this (would generate I/O) */
1717 if (xfs_is_readonly(mp))
1718 goto out;
1719
1720 /* Metadata inodes require explicit resource cleanup. */
1721 if (xfs_is_metadata_inode(ip))
1722 goto out;
1723
1724 /* Try to clean out the cow blocks if there are any. */
1725 if (xfs_inode_has_cow_data(ip))
1726 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
1727
1728 if (VFS_I(ip)->i_nlink != 0) {
1729 /*
1730 * force is true because we are evicting an inode from the
1731 * cache. Post-eof blocks must be freed, lest we end up with
1732 * broken free space accounting.
1733 *
1734 * Note: don't bother with iolock here since lockdep complains
1735 * about acquiring it in reclaim context. We have the only
1736 * reference to the inode at this point anyways.
1737 */
1738 if (xfs_can_free_eofblocks(ip, true))
1739 xfs_free_eofblocks(ip);
1740
1741 goto out;
1742 }
1743
1744 if (S_ISREG(VFS_I(ip)->i_mode) &&
1745 (ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
1746 ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
1747 truncate = 1;
1748
1749 error = xfs_qm_dqattach(ip);
1750 if (error)
1751 goto out;
1752
1753 if (S_ISLNK(VFS_I(ip)->i_mode))
1754 error = xfs_inactive_symlink(ip);
1755 else if (truncate)
1756 error = xfs_inactive_truncate(ip);
1757 if (error)
1758 goto out;
1759
1760 /*
1761 * If there are attributes associated with the file then blow them away
1762 * now. The code calls a routine that recursively deconstructs the
1763 * attribute fork. If also blows away the in-core attribute fork.
1764 */
1765 if (xfs_inode_has_attr_fork(ip)) {
1766 error = xfs_attr_inactive(ip);
1767 if (error)
1768 goto out;
1769 }
1770
1771 ASSERT(ip->i_forkoff == 0);
1772
1773 /*
1774 * Free the inode.
1775 */
1776 xfs_inactive_ifree(ip);
1777
1778out:
1779 /*
1780 * We're done making metadata updates for this inode, so we can release
1781 * the attached dquots.
1782 */
1783 xfs_qm_dqdetach(ip);
1784}
1785
1786/*
1787 * In-Core Unlinked List Lookups
1788 * =============================
1789 *
1790 * Every inode is supposed to be reachable from some other piece of metadata
1791 * with the exception of the root directory. Inodes with a connection to a
1792 * file descriptor but not linked from anywhere in the on-disk directory tree
1793 * are collectively known as unlinked inodes, though the filesystem itself
1794 * maintains links to these inodes so that on-disk metadata are consistent.
1795 *
1796 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
1797 * header contains a number of buckets that point to an inode, and each inode
1798 * record has a pointer to the next inode in the hash chain. This
1799 * singly-linked list causes scaling problems in the iunlink remove function
1800 * because we must walk that list to find the inode that points to the inode
1801 * being removed from the unlinked hash bucket list.
1802 *
1803 * Hence we keep an in-memory double linked list to link each inode on an
1804 * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer
1805 * based lists would require having 64 list heads in the perag, one for each
1806 * list. This is expensive in terms of memory (think millions of AGs) and cache
1807 * misses on lookups. Instead, use the fact that inodes on the unlinked list
1808 * must be referenced at the VFS level to keep them on the list and hence we
1809 * have an existence guarantee for inodes on the unlinked list.
1810 *
1811 * Given we have an existence guarantee, we can use lockless inode cache lookups
1812 * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode
1813 * for the double linked unlinked list, and we don't need any extra locking to
1814 * keep the list safe as all manipulations are done under the AGI buffer lock.
1815 * Keeping the list up to date does not require memory allocation, just finding
1816 * the XFS inode and updating the next/prev unlinked list aginos.
1817 */
1818
1819/*
1820 * Find an inode on the unlinked list. This does not take references to the
1821 * inode as we have existence guarantees by holding the AGI buffer lock and that
1822 * only unlinked, referenced inodes can be on the unlinked inode list. If we
1823 * don't find the inode in cache, then let the caller handle the situation.
1824 */
1825static struct xfs_inode *
1826xfs_iunlink_lookup(
1827 struct xfs_perag *pag,
1828 xfs_agino_t agino)
1829{
1830 struct xfs_inode *ip;
1831
1832 rcu_read_lock();
1833 ip = radix_tree_lookup(&pag->pag_ici_root, agino);
1834
1835 /*
1836 * Inode not in memory or in RCU freeing limbo should not happen.
1837 * Warn about this and let the caller handle the failure.
1838 */
1839 if (WARN_ON_ONCE(!ip || !ip->i_ino)) {
1840 rcu_read_unlock();
1841 return NULL;
1842 }
1843 ASSERT(!xfs_iflags_test(ip, XFS_IRECLAIMABLE | XFS_IRECLAIM));
1844 rcu_read_unlock();
1845 return ip;
1846}
1847
1848/* Update the prev pointer of the next agino. */
1849static int
1850xfs_iunlink_update_backref(
1851 struct xfs_perag *pag,
1852 xfs_agino_t prev_agino,
1853 xfs_agino_t next_agino)
1854{
1855 struct xfs_inode *ip;
1856
1857 /* No update necessary if we are at the end of the list. */
1858 if (next_agino == NULLAGINO)
1859 return 0;
1860
1861 ip = xfs_iunlink_lookup(pag, next_agino);
1862 if (!ip)
1863 return -EFSCORRUPTED;
1864 ip->i_prev_unlinked = prev_agino;
1865 return 0;
1866}
1867
1868/*
1869 * Point the AGI unlinked bucket at an inode and log the results. The caller
1870 * is responsible for validating the old value.
1871 */
1872STATIC int
1873xfs_iunlink_update_bucket(
1874 struct xfs_trans *tp,
1875 struct xfs_perag *pag,
1876 struct xfs_buf *agibp,
1877 unsigned int bucket_index,
1878 xfs_agino_t new_agino)
1879{
1880 struct xfs_agi *agi = agibp->b_addr;
1881 xfs_agino_t old_value;
1882 int offset;
1883
1884 ASSERT(xfs_verify_agino_or_null(pag, new_agino));
1885
1886 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1887 trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
1888 old_value, new_agino);
1889
1890 /*
1891 * We should never find the head of the list already set to the value
1892 * passed in because either we're adding or removing ourselves from the
1893 * head of the list.
1894 */
1895 if (old_value == new_agino) {
1896 xfs_buf_mark_corrupt(agibp);
1897 return -EFSCORRUPTED;
1898 }
1899
1900 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
1901 offset = offsetof(struct xfs_agi, agi_unlinked) +
1902 (sizeof(xfs_agino_t) * bucket_index);
1903 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
1904 return 0;
1905}
1906
1907static int
1908xfs_iunlink_insert_inode(
1909 struct xfs_trans *tp,
1910 struct xfs_perag *pag,
1911 struct xfs_buf *agibp,
1912 struct xfs_inode *ip)
1913{
1914 struct xfs_mount *mp = tp->t_mountp;
1915 struct xfs_agi *agi = agibp->b_addr;
1916 xfs_agino_t next_agino;
1917 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1918 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1919 int error;
1920
1921 /*
1922 * Get the index into the agi hash table for the list this inode will
1923 * go on. Make sure the pointer isn't garbage and that this inode
1924 * isn't already on the list.
1925 */
1926 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
1927 if (next_agino == agino ||
1928 !xfs_verify_agino_or_null(pag, next_agino)) {
1929 xfs_buf_mark_corrupt(agibp);
1930 return -EFSCORRUPTED;
1931 }
1932
1933 /*
1934 * Update the prev pointer in the next inode to point back to this
1935 * inode.
1936 */
1937 error = xfs_iunlink_update_backref(pag, agino, next_agino);
1938 if (error)
1939 return error;
1940
1941 if (next_agino != NULLAGINO) {
1942 /*
1943 * There is already another inode in the bucket, so point this
1944 * inode to the current head of the list.
1945 */
1946 error = xfs_iunlink_log_inode(tp, ip, pag, next_agino);
1947 if (error)
1948 return error;
1949 ip->i_next_unlinked = next_agino;
1950 }
1951
1952 /* Point the head of the list to point to this inode. */
1953 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
1954}
1955
1956/*
1957 * This is called when the inode's link count has gone to 0 or we are creating
1958 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
1959 *
1960 * We place the on-disk inode on a list in the AGI. It will be pulled from this
1961 * list when the inode is freed.
1962 */
1963STATIC int
1964xfs_iunlink(
1965 struct xfs_trans *tp,
1966 struct xfs_inode *ip)
1967{
1968 struct xfs_mount *mp = tp->t_mountp;
1969 struct xfs_perag *pag;
1970 struct xfs_buf *agibp;
1971 int error;
1972
1973 ASSERT(VFS_I(ip)->i_nlink == 0);
1974 ASSERT(VFS_I(ip)->i_mode != 0);
1975 trace_xfs_iunlink(ip);
1976
1977 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1978
1979 /* Get the agi buffer first. It ensures lock ordering on the list. */
1980 error = xfs_read_agi(pag, tp, &agibp);
1981 if (error)
1982 goto out;
1983
1984 error = xfs_iunlink_insert_inode(tp, pag, agibp, ip);
1985out:
1986 xfs_perag_put(pag);
1987 return error;
1988}
1989
1990static int
1991xfs_iunlink_remove_inode(
1992 struct xfs_trans *tp,
1993 struct xfs_perag *pag,
1994 struct xfs_buf *agibp,
1995 struct xfs_inode *ip)
1996{
1997 struct xfs_mount *mp = tp->t_mountp;
1998 struct xfs_agi *agi = agibp->b_addr;
1999 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2000 xfs_agino_t head_agino;
2001 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2002 int error;
2003
2004 trace_xfs_iunlink_remove(ip);
2005
2006 /*
2007 * Get the index into the agi hash table for the list this inode will
2008 * go on. Make sure the head pointer isn't garbage.
2009 */
2010 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2011 if (!xfs_verify_agino(pag, head_agino)) {
2012 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
2013 agi, sizeof(*agi));
2014 return -EFSCORRUPTED;
2015 }
2016
2017 /*
2018 * Set our inode's next_unlinked pointer to NULL and then return
2019 * the old pointer value so that we can update whatever was previous
2020 * to us in the list to point to whatever was next in the list.
2021 */
2022 error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO);
2023 if (error)
2024 return error;
2025
2026 /*
2027 * Update the prev pointer in the next inode to point back to previous
2028 * inode in the chain.
2029 */
2030 error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked,
2031 ip->i_next_unlinked);
2032 if (error)
2033 return error;
2034
2035 if (head_agino != agino) {
2036 struct xfs_inode *prev_ip;
2037
2038 prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked);
2039 if (!prev_ip)
2040 return -EFSCORRUPTED;
2041
2042 error = xfs_iunlink_log_inode(tp, prev_ip, pag,
2043 ip->i_next_unlinked);
2044 prev_ip->i_next_unlinked = ip->i_next_unlinked;
2045 } else {
2046 /* Point the head of the list to the next unlinked inode. */
2047 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
2048 ip->i_next_unlinked);
2049 }
2050
2051 ip->i_next_unlinked = NULLAGINO;
2052 ip->i_prev_unlinked = NULLAGINO;
2053 return error;
2054}
2055
2056/*
2057 * Pull the on-disk inode from the AGI unlinked list.
2058 */
2059STATIC int
2060xfs_iunlink_remove(
2061 struct xfs_trans *tp,
2062 struct xfs_perag *pag,
2063 struct xfs_inode *ip)
2064{
2065 struct xfs_buf *agibp;
2066 int error;
2067
2068 trace_xfs_iunlink_remove(ip);
2069
2070 /* Get the agi buffer first. It ensures lock ordering on the list. */
2071 error = xfs_read_agi(pag, tp, &agibp);
2072 if (error)
2073 return error;
2074
2075 return xfs_iunlink_remove_inode(tp, pag, agibp, ip);
2076}
2077
2078/*
2079 * Look up the inode number specified and if it is not already marked XFS_ISTALE
2080 * mark it stale. We should only find clean inodes in this lookup that aren't
2081 * already stale.
2082 */
2083static void
2084xfs_ifree_mark_inode_stale(
2085 struct xfs_perag *pag,
2086 struct xfs_inode *free_ip,
2087 xfs_ino_t inum)
2088{
2089 struct xfs_mount *mp = pag->pag_mount;
2090 struct xfs_inode_log_item *iip;
2091 struct xfs_inode *ip;
2092
2093retry:
2094 rcu_read_lock();
2095 ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
2096
2097 /* Inode not in memory, nothing to do */
2098 if (!ip) {
2099 rcu_read_unlock();
2100 return;
2101 }
2102
2103 /*
2104 * because this is an RCU protected lookup, we could find a recently
2105 * freed or even reallocated inode during the lookup. We need to check
2106 * under the i_flags_lock for a valid inode here. Skip it if it is not
2107 * valid, the wrong inode or stale.
2108 */
2109 spin_lock(&ip->i_flags_lock);
2110 if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
2111 goto out_iflags_unlock;
2112
2113 /*
2114 * Don't try to lock/unlock the current inode, but we _cannot_ skip the
2115 * other inodes that we did not find in the list attached to the buffer
2116 * and are not already marked stale. If we can't lock it, back off and
2117 * retry.
2118 */
2119 if (ip != free_ip) {
2120 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2121 spin_unlock(&ip->i_flags_lock);
2122 rcu_read_unlock();
2123 delay(1);
2124 goto retry;
2125 }
2126 }
2127 ip->i_flags |= XFS_ISTALE;
2128
2129 /*
2130 * If the inode is flushing, it is already attached to the buffer. All
2131 * we needed to do here is mark the inode stale so buffer IO completion
2132 * will remove it from the AIL.
2133 */
2134 iip = ip->i_itemp;
2135 if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
2136 ASSERT(!list_empty(&iip->ili_item.li_bio_list));
2137 ASSERT(iip->ili_last_fields);
2138 goto out_iunlock;
2139 }
2140
2141 /*
2142 * Inodes not attached to the buffer can be released immediately.
2143 * Everything else has to go through xfs_iflush_abort() on journal
2144 * commit as the flock synchronises removal of the inode from the
2145 * cluster buffer against inode reclaim.
2146 */
2147 if (!iip || list_empty(&iip->ili_item.li_bio_list))
2148 goto out_iunlock;
2149
2150 __xfs_iflags_set(ip, XFS_IFLUSHING);
2151 spin_unlock(&ip->i_flags_lock);
2152 rcu_read_unlock();
2153
2154 /* we have a dirty inode in memory that has not yet been flushed. */
2155 spin_lock(&iip->ili_lock);
2156 iip->ili_last_fields = iip->ili_fields;
2157 iip->ili_fields = 0;
2158 iip->ili_fsync_fields = 0;
2159 spin_unlock(&iip->ili_lock);
2160 ASSERT(iip->ili_last_fields);
2161
2162 if (ip != free_ip)
2163 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2164 return;
2165
2166out_iunlock:
2167 if (ip != free_ip)
2168 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2169out_iflags_unlock:
2170 spin_unlock(&ip->i_flags_lock);
2171 rcu_read_unlock();
2172}
2173
2174/*
2175 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2176 * inodes that are in memory - they all must be marked stale and attached to
2177 * the cluster buffer.
2178 */
2179static int
2180xfs_ifree_cluster(
2181 struct xfs_trans *tp,
2182 struct xfs_perag *pag,
2183 struct xfs_inode *free_ip,
2184 struct xfs_icluster *xic)
2185{
2186 struct xfs_mount *mp = free_ip->i_mount;
2187 struct xfs_ino_geometry *igeo = M_IGEO(mp);
2188 struct xfs_buf *bp;
2189 xfs_daddr_t blkno;
2190 xfs_ino_t inum = xic->first_ino;
2191 int nbufs;
2192 int i, j;
2193 int ioffset;
2194 int error;
2195
2196 nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
2197
2198 for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
2199 /*
2200 * The allocation bitmap tells us which inodes of the chunk were
2201 * physically allocated. Skip the cluster if an inode falls into
2202 * a sparse region.
2203 */
2204 ioffset = inum - xic->first_ino;
2205 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
2206 ASSERT(ioffset % igeo->inodes_per_cluster == 0);
2207 continue;
2208 }
2209
2210 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2211 XFS_INO_TO_AGBNO(mp, inum));
2212
2213 /*
2214 * We obtain and lock the backing buffer first in the process
2215 * here to ensure dirty inodes attached to the buffer remain in
2216 * the flushing state while we mark them stale.
2217 *
2218 * If we scan the in-memory inodes first, then buffer IO can
2219 * complete before we get a lock on it, and hence we may fail
2220 * to mark all the active inodes on the buffer stale.
2221 */
2222 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2223 mp->m_bsize * igeo->blocks_per_cluster,
2224 XBF_UNMAPPED, &bp);
2225 if (error)
2226 return error;
2227
2228 /*
2229 * This buffer may not have been correctly initialised as we
2230 * didn't read it from disk. That's not important because we are
2231 * only using to mark the buffer as stale in the log, and to
2232 * attach stale cached inodes on it. That means it will never be
2233 * dispatched for IO. If it is, we want to know about it, and we
2234 * want it to fail. We can acheive this by adding a write
2235 * verifier to the buffer.
2236 */
2237 bp->b_ops = &xfs_inode_buf_ops;
2238
2239 /*
2240 * Now we need to set all the cached clean inodes as XFS_ISTALE,
2241 * too. This requires lookups, and will skip inodes that we've
2242 * already marked XFS_ISTALE.
2243 */
2244 for (i = 0; i < igeo->inodes_per_cluster; i++)
2245 xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
2246
2247 xfs_trans_stale_inode_buf(tp, bp);
2248 xfs_trans_binval(tp, bp);
2249 }
2250 return 0;
2251}
2252
2253/*
2254 * This is called to return an inode to the inode free list. The inode should
2255 * already be truncated to 0 length and have no pages associated with it. This
2256 * routine also assumes that the inode is already a part of the transaction.
2257 *
2258 * The on-disk copy of the inode will have been added to the list of unlinked
2259 * inodes in the AGI. We need to remove the inode from that list atomically with
2260 * respect to freeing it here.
2261 */
2262int
2263xfs_ifree(
2264 struct xfs_trans *tp,
2265 struct xfs_inode *ip)
2266{
2267 struct xfs_mount *mp = ip->i_mount;
2268 struct xfs_perag *pag;
2269 struct xfs_icluster xic = { 0 };
2270 struct xfs_inode_log_item *iip = ip->i_itemp;
2271 int error;
2272
2273 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2274 ASSERT(VFS_I(ip)->i_nlink == 0);
2275 ASSERT(ip->i_df.if_nextents == 0);
2276 ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
2277 ASSERT(ip->i_nblocks == 0);
2278
2279 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2280
2281 /*
2282 * Free the inode first so that we guarantee that the AGI lock is going
2283 * to be taken before we remove the inode from the unlinked list. This
2284 * makes the AGI lock -> unlinked list modification order the same as
2285 * used in O_TMPFILE creation.
2286 */
2287 error = xfs_difree(tp, pag, ip->i_ino, &xic);
2288 if (error)
2289 goto out;
2290
2291 error = xfs_iunlink_remove(tp, pag, ip);
2292 if (error)
2293 goto out;
2294
2295 /*
2296 * Free any local-format data sitting around before we reset the
2297 * data fork to extents format. Note that the attr fork data has
2298 * already been freed by xfs_attr_inactive.
2299 */
2300 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
2301 kmem_free(ip->i_df.if_u1.if_data);
2302 ip->i_df.if_u1.if_data = NULL;
2303 ip->i_df.if_bytes = 0;
2304 }
2305
2306 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
2307 ip->i_diflags = 0;
2308 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
2309 ip->i_forkoff = 0; /* mark the attr fork not in use */
2310 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
2311 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
2312 xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
2313
2314 /* Don't attempt to replay owner changes for a deleted inode */
2315 spin_lock(&iip->ili_lock);
2316 iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
2317 spin_unlock(&iip->ili_lock);
2318
2319 /*
2320 * Bump the generation count so no one will be confused
2321 * by reincarnations of this inode.
2322 */
2323 VFS_I(ip)->i_generation++;
2324 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2325
2326 if (xic.deleted)
2327 error = xfs_ifree_cluster(tp, pag, ip, &xic);
2328out:
2329 xfs_perag_put(pag);
2330 return error;
2331}
2332
2333/*
2334 * This is called to unpin an inode. The caller must have the inode locked
2335 * in at least shared mode so that the buffer cannot be subsequently pinned
2336 * once someone is waiting for it to be unpinned.
2337 */
2338static void
2339xfs_iunpin(
2340 struct xfs_inode *ip)
2341{
2342 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2343
2344 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2345
2346 /* Give the log a push to start the unpinning I/O */
2347 xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
2348
2349}
2350
2351static void
2352__xfs_iunpin_wait(
2353 struct xfs_inode *ip)
2354{
2355 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2356 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2357
2358 xfs_iunpin(ip);
2359
2360 do {
2361 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2362 if (xfs_ipincount(ip))
2363 io_schedule();
2364 } while (xfs_ipincount(ip));
2365 finish_wait(wq, &wait.wq_entry);
2366}
2367
2368void
2369xfs_iunpin_wait(
2370 struct xfs_inode *ip)
2371{
2372 if (xfs_ipincount(ip))
2373 __xfs_iunpin_wait(ip);
2374}
2375
2376/*
2377 * Removing an inode from the namespace involves removing the directory entry
2378 * and dropping the link count on the inode. Removing the directory entry can
2379 * result in locking an AGF (directory blocks were freed) and removing a link
2380 * count can result in placing the inode on an unlinked list which results in
2381 * locking an AGI.
2382 *
2383 * The big problem here is that we have an ordering constraint on AGF and AGI
2384 * locking - inode allocation locks the AGI, then can allocate a new extent for
2385 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode
2386 * removes the inode from the unlinked list, requiring that we lock the AGI
2387 * first, and then freeing the inode can result in an inode chunk being freed
2388 * and hence freeing disk space requiring that we lock an AGF.
2389 *
2390 * Hence the ordering that is imposed by other parts of the code is AGI before
2391 * AGF. This means we cannot remove the directory entry before we drop the inode
2392 * reference count and put it on the unlinked list as this results in a lock
2393 * order of AGF then AGI, and this can deadlock against inode allocation and
2394 * freeing. Therefore we must drop the link counts before we remove the
2395 * directory entry.
2396 *
2397 * This is still safe from a transactional point of view - it is not until we
2398 * get to xfs_defer_finish() that we have the possibility of multiple
2399 * transactions in this operation. Hence as long as we remove the directory
2400 * entry and drop the link count in the first transaction of the remove
2401 * operation, there are no transactional constraints on the ordering here.
2402 */
2403int
2404xfs_remove(
2405 xfs_inode_t *dp,
2406 struct xfs_name *name,
2407 xfs_inode_t *ip)
2408{
2409 xfs_mount_t *mp = dp->i_mount;
2410 xfs_trans_t *tp = NULL;
2411 int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
2412 int dontcare;
2413 int error = 0;
2414 uint resblks;
2415
2416 trace_xfs_remove(dp, name);
2417
2418 if (xfs_is_shutdown(mp))
2419 return -EIO;
2420
2421 error = xfs_qm_dqattach(dp);
2422 if (error)
2423 goto std_return;
2424
2425 error = xfs_qm_dqattach(ip);
2426 if (error)
2427 goto std_return;
2428
2429 /*
2430 * We try to get the real space reservation first, allowing for
2431 * directory btree deletion(s) implying possible bmap insert(s). If we
2432 * can't get the space reservation then we use 0 instead, and avoid the
2433 * bmap btree insert(s) in the directory code by, if the bmap insert
2434 * tries to happen, instead trimming the LAST block from the directory.
2435 *
2436 * Ignore EDQUOT and ENOSPC being returned via nospace_error because
2437 * the directory code can handle a reservationless update and we don't
2438 * want to prevent a user from trying to free space by deleting things.
2439 */
2440 resblks = XFS_REMOVE_SPACE_RES(mp);
2441 error = xfs_trans_alloc_dir(dp, &M_RES(mp)->tr_remove, ip, &resblks,
2442 &tp, &dontcare);
2443 if (error) {
2444 ASSERT(error != -ENOSPC);
2445 goto std_return;
2446 }
2447
2448 /*
2449 * If we're removing a directory perform some additional validation.
2450 */
2451 if (is_dir) {
2452 ASSERT(VFS_I(ip)->i_nlink >= 2);
2453 if (VFS_I(ip)->i_nlink != 2) {
2454 error = -ENOTEMPTY;
2455 goto out_trans_cancel;
2456 }
2457 if (!xfs_dir_isempty(ip)) {
2458 error = -ENOTEMPTY;
2459 goto out_trans_cancel;
2460 }
2461
2462 /* Drop the link from ip's "..". */
2463 error = xfs_droplink(tp, dp);
2464 if (error)
2465 goto out_trans_cancel;
2466
2467 /* Drop the "." link from ip to self. */
2468 error = xfs_droplink(tp, ip);
2469 if (error)
2470 goto out_trans_cancel;
2471
2472 /*
2473 * Point the unlinked child directory's ".." entry to the root
2474 * directory to eliminate back-references to inodes that may
2475 * get freed before the child directory is closed. If the fs
2476 * gets shrunk, this can lead to dirent inode validation errors.
2477 */
2478 if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
2479 error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
2480 tp->t_mountp->m_sb.sb_rootino, 0);
2481 if (error)
2482 goto out_trans_cancel;
2483 }
2484 } else {
2485 /*
2486 * When removing a non-directory we need to log the parent
2487 * inode here. For a directory this is done implicitly
2488 * by the xfs_droplink call for the ".." entry.
2489 */
2490 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2491 }
2492 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2493
2494 /* Drop the link from dp to ip. */
2495 error = xfs_droplink(tp, ip);
2496 if (error)
2497 goto out_trans_cancel;
2498
2499 error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
2500 if (error) {
2501 ASSERT(error != -ENOENT);
2502 goto out_trans_cancel;
2503 }
2504
2505 /*
2506 * If this is a synchronous mount, make sure that the
2507 * remove transaction goes to disk before returning to
2508 * the user.
2509 */
2510 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp))
2511 xfs_trans_set_sync(tp);
2512
2513 error = xfs_trans_commit(tp);
2514 if (error)
2515 goto std_return;
2516
2517 if (is_dir && xfs_inode_is_filestream(ip))
2518 xfs_filestream_deassociate(ip);
2519
2520 return 0;
2521
2522 out_trans_cancel:
2523 xfs_trans_cancel(tp);
2524 std_return:
2525 return error;
2526}
2527
2528/*
2529 * Enter all inodes for a rename transaction into a sorted array.
2530 */
2531#define __XFS_SORT_INODES 5
2532STATIC void
2533xfs_sort_for_rename(
2534 struct xfs_inode *dp1, /* in: old (source) directory inode */
2535 struct xfs_inode *dp2, /* in: new (target) directory inode */
2536 struct xfs_inode *ip1, /* in: inode of old entry */
2537 struct xfs_inode *ip2, /* in: inode of new entry */
2538 struct xfs_inode *wip, /* in: whiteout inode */
2539 struct xfs_inode **i_tab,/* out: sorted array of inodes */
2540 int *num_inodes) /* in/out: inodes in array */
2541{
2542 int i, j;
2543
2544 ASSERT(*num_inodes == __XFS_SORT_INODES);
2545 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
2546
2547 /*
2548 * i_tab contains a list of pointers to inodes. We initialize
2549 * the table here & we'll sort it. We will then use it to
2550 * order the acquisition of the inode locks.
2551 *
2552 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2553 */
2554 i = 0;
2555 i_tab[i++] = dp1;
2556 i_tab[i++] = dp2;
2557 i_tab[i++] = ip1;
2558 if (ip2)
2559 i_tab[i++] = ip2;
2560 if (wip)
2561 i_tab[i++] = wip;
2562 *num_inodes = i;
2563
2564 /*
2565 * Sort the elements via bubble sort. (Remember, there are at
2566 * most 5 elements to sort, so this is adequate.)
2567 */
2568 for (i = 0; i < *num_inodes; i++) {
2569 for (j = 1; j < *num_inodes; j++) {
2570 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2571 struct xfs_inode *temp = i_tab[j];
2572 i_tab[j] = i_tab[j-1];
2573 i_tab[j-1] = temp;
2574 }
2575 }
2576 }
2577}
2578
2579static int
2580xfs_finish_rename(
2581 struct xfs_trans *tp)
2582{
2583 /*
2584 * If this is a synchronous mount, make sure that the rename transaction
2585 * goes to disk before returning to the user.
2586 */
2587 if (xfs_has_wsync(tp->t_mountp) || xfs_has_dirsync(tp->t_mountp))
2588 xfs_trans_set_sync(tp);
2589
2590 return xfs_trans_commit(tp);
2591}
2592
2593/*
2594 * xfs_cross_rename()
2595 *
2596 * responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
2597 */
2598STATIC int
2599xfs_cross_rename(
2600 struct xfs_trans *tp,
2601 struct xfs_inode *dp1,
2602 struct xfs_name *name1,
2603 struct xfs_inode *ip1,
2604 struct xfs_inode *dp2,
2605 struct xfs_name *name2,
2606 struct xfs_inode *ip2,
2607 int spaceres)
2608{
2609 int error = 0;
2610 int ip1_flags = 0;
2611 int ip2_flags = 0;
2612 int dp2_flags = 0;
2613
2614 /* Swap inode number for dirent in first parent */
2615 error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
2616 if (error)
2617 goto out_trans_abort;
2618
2619 /* Swap inode number for dirent in second parent */
2620 error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
2621 if (error)
2622 goto out_trans_abort;
2623
2624 /*
2625 * If we're renaming one or more directories across different parents,
2626 * update the respective ".." entries (and link counts) to match the new
2627 * parents.
2628 */
2629 if (dp1 != dp2) {
2630 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2631
2632 if (S_ISDIR(VFS_I(ip2)->i_mode)) {
2633 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
2634 dp1->i_ino, spaceres);
2635 if (error)
2636 goto out_trans_abort;
2637
2638 /* transfer ip2 ".." reference to dp1 */
2639 if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
2640 error = xfs_droplink(tp, dp2);
2641 if (error)
2642 goto out_trans_abort;
2643 xfs_bumplink(tp, dp1);
2644 }
2645
2646 /*
2647 * Although ip1 isn't changed here, userspace needs
2648 * to be warned about the change, so that applications
2649 * relying on it (like backup ones), will properly
2650 * notify the change
2651 */
2652 ip1_flags |= XFS_ICHGTIME_CHG;
2653 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2654 }
2655
2656 if (S_ISDIR(VFS_I(ip1)->i_mode)) {
2657 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
2658 dp2->i_ino, spaceres);
2659 if (error)
2660 goto out_trans_abort;
2661
2662 /* transfer ip1 ".." reference to dp2 */
2663 if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
2664 error = xfs_droplink(tp, dp1);
2665 if (error)
2666 goto out_trans_abort;
2667 xfs_bumplink(tp, dp2);
2668 }
2669
2670 /*
2671 * Although ip2 isn't changed here, userspace needs
2672 * to be warned about the change, so that applications
2673 * relying on it (like backup ones), will properly
2674 * notify the change
2675 */
2676 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
2677 ip2_flags |= XFS_ICHGTIME_CHG;
2678 }
2679 }
2680
2681 if (ip1_flags) {
2682 xfs_trans_ichgtime(tp, ip1, ip1_flags);
2683 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
2684 }
2685 if (ip2_flags) {
2686 xfs_trans_ichgtime(tp, ip2, ip2_flags);
2687 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
2688 }
2689 if (dp2_flags) {
2690 xfs_trans_ichgtime(tp, dp2, dp2_flags);
2691 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
2692 }
2693 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2694 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
2695 return xfs_finish_rename(tp);
2696
2697out_trans_abort:
2698 xfs_trans_cancel(tp);
2699 return error;
2700}
2701
2702/*
2703 * xfs_rename_alloc_whiteout()
2704 *
2705 * Return a referenced, unlinked, unlocked inode that can be used as a
2706 * whiteout in a rename transaction. We use a tmpfile inode here so that if we
2707 * crash between allocating the inode and linking it into the rename transaction
2708 * recovery will free the inode and we won't leak it.
2709 */
2710static int
2711xfs_rename_alloc_whiteout(
2712 struct user_namespace *mnt_userns,
2713 struct xfs_name *src_name,
2714 struct xfs_inode *dp,
2715 struct xfs_inode **wip)
2716{
2717 struct xfs_inode *tmpfile;
2718 struct qstr name;
2719 int error;
2720
2721 error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
2722 &tmpfile);
2723 if (error)
2724 return error;
2725
2726 name.name = src_name->name;
2727 name.len = src_name->len;
2728 error = xfs_inode_init_security(VFS_I(tmpfile), VFS_I(dp), &name);
2729 if (error) {
2730 xfs_finish_inode_setup(tmpfile);
2731 xfs_irele(tmpfile);
2732 return error;
2733 }
2734
2735 /*
2736 * Prepare the tmpfile inode as if it were created through the VFS.
2737 * Complete the inode setup and flag it as linkable. nlink is already
2738 * zero, so we can skip the drop_nlink.
2739 */
2740 xfs_setup_iops(tmpfile);
2741 xfs_finish_inode_setup(tmpfile);
2742 VFS_I(tmpfile)->i_state |= I_LINKABLE;
2743
2744 *wip = tmpfile;
2745 return 0;
2746}
2747
2748/*
2749 * xfs_rename
2750 */
2751int
2752xfs_rename(
2753 struct user_namespace *mnt_userns,
2754 struct xfs_inode *src_dp,
2755 struct xfs_name *src_name,
2756 struct xfs_inode *src_ip,
2757 struct xfs_inode *target_dp,
2758 struct xfs_name *target_name,
2759 struct xfs_inode *target_ip,
2760 unsigned int flags)
2761{
2762 struct xfs_mount *mp = src_dp->i_mount;
2763 struct xfs_trans *tp;
2764 struct xfs_inode *wip = NULL; /* whiteout inode */
2765 struct xfs_inode *inodes[__XFS_SORT_INODES];
2766 int i;
2767 int num_inodes = __XFS_SORT_INODES;
2768 bool new_parent = (src_dp != target_dp);
2769 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
2770 int spaceres;
2771 bool retried = false;
2772 int error, nospace_error = 0;
2773
2774 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2775
2776 if ((flags & RENAME_EXCHANGE) && !target_ip)
2777 return -EINVAL;
2778
2779 /*
2780 * If we are doing a whiteout operation, allocate the whiteout inode
2781 * we will be placing at the target and ensure the type is set
2782 * appropriately.
2783 */
2784 if (flags & RENAME_WHITEOUT) {
2785 error = xfs_rename_alloc_whiteout(mnt_userns, src_name,
2786 target_dp, &wip);
2787 if (error)
2788 return error;
2789
2790 /* setup target dirent info as whiteout */
2791 src_name->type = XFS_DIR3_FT_CHRDEV;
2792 }
2793
2794 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
2795 inodes, &num_inodes);
2796
2797retry:
2798 nospace_error = 0;
2799 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2800 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
2801 if (error == -ENOSPC) {
2802 nospace_error = error;
2803 spaceres = 0;
2804 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
2805 &tp);
2806 }
2807 if (error)
2808 goto out_release_wip;
2809
2810 /*
2811 * Attach the dquots to the inodes
2812 */
2813 error = xfs_qm_vop_rename_dqattach(inodes);
2814 if (error)
2815 goto out_trans_cancel;
2816
2817 /*
2818 * Lock all the participating inodes. Depending upon whether
2819 * the target_name exists in the target directory, and
2820 * whether the target directory is the same as the source
2821 * directory, we can lock from 2 to 5 inodes.
2822 */
2823 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2824
2825 /*
2826 * Join all the inodes to the transaction. From this point on,
2827 * we can rely on either trans_commit or trans_cancel to unlock
2828 * them.
2829 */
2830 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2831 if (new_parent)
2832 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2833 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2834 if (target_ip)
2835 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2836 if (wip)
2837 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
2838
2839 /*
2840 * If we are using project inheritance, we only allow renames
2841 * into our tree when the project IDs are the same; else the
2842 * tree quota mechanism would be circumvented.
2843 */
2844 if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
2845 target_dp->i_projid != src_ip->i_projid)) {
2846 error = -EXDEV;
2847 goto out_trans_cancel;
2848 }
2849
2850 /* RENAME_EXCHANGE is unique from here on. */
2851 if (flags & RENAME_EXCHANGE)
2852 return xfs_cross_rename(tp, src_dp, src_name, src_ip,
2853 target_dp, target_name, target_ip,
2854 spaceres);
2855
2856 /*
2857 * Try to reserve quota to handle an expansion of the target directory.
2858 * We'll allow the rename to continue in reservationless mode if we hit
2859 * a space usage constraint. If we trigger reservationless mode, save
2860 * the errno if there isn't any free space in the target directory.
2861 */
2862 if (spaceres != 0) {
2863 error = xfs_trans_reserve_quota_nblks(tp, target_dp, spaceres,
2864 0, false);
2865 if (error == -EDQUOT || error == -ENOSPC) {
2866 if (!retried) {
2867 xfs_trans_cancel(tp);
2868 xfs_blockgc_free_quota(target_dp, 0);
2869 retried = true;
2870 goto retry;
2871 }
2872
2873 nospace_error = error;
2874 spaceres = 0;
2875 error = 0;
2876 }
2877 if (error)
2878 goto out_trans_cancel;
2879 }
2880
2881 /*
2882 * Check for expected errors before we dirty the transaction
2883 * so we can return an error without a transaction abort.
2884 */
2885 if (target_ip == NULL) {
2886 /*
2887 * If there's no space reservation, check the entry will
2888 * fit before actually inserting it.
2889 */
2890 if (!spaceres) {
2891 error = xfs_dir_canenter(tp, target_dp, target_name);
2892 if (error)
2893 goto out_trans_cancel;
2894 }
2895 } else {
2896 /*
2897 * If target exists and it's a directory, check that whether
2898 * it can be destroyed.
2899 */
2900 if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
2901 (!xfs_dir_isempty(target_ip) ||
2902 (VFS_I(target_ip)->i_nlink > 2))) {
2903 error = -EEXIST;
2904 goto out_trans_cancel;
2905 }
2906 }
2907
2908 /*
2909 * Lock the AGI buffers we need to handle bumping the nlink of the
2910 * whiteout inode off the unlinked list and to handle dropping the
2911 * nlink of the target inode. Per locking order rules, do this in
2912 * increasing AG order and before directory block allocation tries to
2913 * grab AGFs because we grab AGIs before AGFs.
2914 *
2915 * The (vfs) caller must ensure that if src is a directory then
2916 * target_ip is either null or an empty directory.
2917 */
2918 for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
2919 if (inodes[i] == wip ||
2920 (inodes[i] == target_ip &&
2921 (VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
2922 struct xfs_perag *pag;
2923 struct xfs_buf *bp;
2924
2925 pag = xfs_perag_get(mp,
2926 XFS_INO_TO_AGNO(mp, inodes[i]->i_ino));
2927 error = xfs_read_agi(pag, tp, &bp);
2928 xfs_perag_put(pag);
2929 if (error)
2930 goto out_trans_cancel;
2931 }
2932 }
2933
2934 /*
2935 * Directory entry creation below may acquire the AGF. Remove
2936 * the whiteout from the unlinked list first to preserve correct
2937 * AGI/AGF locking order. This dirties the transaction so failures
2938 * after this point will abort and log recovery will clean up the
2939 * mess.
2940 *
2941 * For whiteouts, we need to bump the link count on the whiteout
2942 * inode. After this point, we have a real link, clear the tmpfile
2943 * state flag from the inode so it doesn't accidentally get misused
2944 * in future.
2945 */
2946 if (wip) {
2947 struct xfs_perag *pag;
2948
2949 ASSERT(VFS_I(wip)->i_nlink == 0);
2950
2951 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
2952 error = xfs_iunlink_remove(tp, pag, wip);
2953 xfs_perag_put(pag);
2954 if (error)
2955 goto out_trans_cancel;
2956
2957 xfs_bumplink(tp, wip);
2958 VFS_I(wip)->i_state &= ~I_LINKABLE;
2959 }
2960
2961 /*
2962 * Set up the target.
2963 */
2964 if (target_ip == NULL) {
2965 /*
2966 * If target does not exist and the rename crosses
2967 * directories, adjust the target directory link count
2968 * to account for the ".." reference from the new entry.
2969 */
2970 error = xfs_dir_createname(tp, target_dp, target_name,
2971 src_ip->i_ino, spaceres);
2972 if (error)
2973 goto out_trans_cancel;
2974
2975 xfs_trans_ichgtime(tp, target_dp,
2976 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2977
2978 if (new_parent && src_is_directory) {
2979 xfs_bumplink(tp, target_dp);
2980 }
2981 } else { /* target_ip != NULL */
2982 /*
2983 * Link the source inode under the target name.
2984 * If the source inode is a directory and we are moving
2985 * it across directories, its ".." entry will be
2986 * inconsistent until we replace that down below.
2987 *
2988 * In case there is already an entry with the same
2989 * name at the destination directory, remove it first.
2990 */
2991 error = xfs_dir_replace(tp, target_dp, target_name,
2992 src_ip->i_ino, spaceres);
2993 if (error)
2994 goto out_trans_cancel;
2995
2996 xfs_trans_ichgtime(tp, target_dp,
2997 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2998
2999 /*
3000 * Decrement the link count on the target since the target
3001 * dir no longer points to it.
3002 */
3003 error = xfs_droplink(tp, target_ip);
3004 if (error)
3005 goto out_trans_cancel;
3006
3007 if (src_is_directory) {
3008 /*
3009 * Drop the link from the old "." entry.
3010 */
3011 error = xfs_droplink(tp, target_ip);
3012 if (error)
3013 goto out_trans_cancel;
3014 }
3015 } /* target_ip != NULL */
3016
3017 /*
3018 * Remove the source.
3019 */
3020 if (new_parent && src_is_directory) {
3021 /*
3022 * Rewrite the ".." entry to point to the new
3023 * directory.
3024 */
3025 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
3026 target_dp->i_ino, spaceres);
3027 ASSERT(error != -EEXIST);
3028 if (error)
3029 goto out_trans_cancel;
3030 }
3031
3032 /*
3033 * We always want to hit the ctime on the source inode.
3034 *
3035 * This isn't strictly required by the standards since the source
3036 * inode isn't really being changed, but old unix file systems did
3037 * it and some incremental backup programs won't work without it.
3038 */
3039 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
3040 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
3041
3042 /*
3043 * Adjust the link count on src_dp. This is necessary when
3044 * renaming a directory, either within one parent when
3045 * the target existed, or across two parent directories.
3046 */
3047 if (src_is_directory && (new_parent || target_ip != NULL)) {
3048
3049 /*
3050 * Decrement link count on src_directory since the
3051 * entry that's moved no longer points to it.
3052 */
3053 error = xfs_droplink(tp, src_dp);
3054 if (error)
3055 goto out_trans_cancel;
3056 }
3057
3058 /*
3059 * For whiteouts, we only need to update the source dirent with the
3060 * inode number of the whiteout inode rather than removing it
3061 * altogether.
3062 */
3063 if (wip)
3064 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
3065 spaceres);
3066 else
3067 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
3068 spaceres);
3069
3070 if (error)
3071 goto out_trans_cancel;
3072
3073 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3074 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
3075 if (new_parent)
3076 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
3077
3078 error = xfs_finish_rename(tp);
3079 if (wip)
3080 xfs_irele(wip);
3081 return error;
3082
3083out_trans_cancel:
3084 xfs_trans_cancel(tp);
3085out_release_wip:
3086 if (wip)
3087 xfs_irele(wip);
3088 if (error == -ENOSPC && nospace_error)
3089 error = nospace_error;
3090 return error;
3091}
3092
3093static int
3094xfs_iflush(
3095 struct xfs_inode *ip,
3096 struct xfs_buf *bp)
3097{
3098 struct xfs_inode_log_item *iip = ip->i_itemp;
3099 struct xfs_dinode *dip;
3100 struct xfs_mount *mp = ip->i_mount;
3101 int error;
3102
3103 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3104 ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
3105 ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
3106 ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3107 ASSERT(iip->ili_item.li_buf == bp);
3108
3109 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
3110
3111 /*
3112 * We don't flush the inode if any of the following checks fail, but we
3113 * do still update the log item and attach to the backing buffer as if
3114 * the flush happened. This is a formality to facilitate predictable
3115 * error handling as the caller will shutdown and fail the buffer.
3116 */
3117 error = -EFSCORRUPTED;
3118 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3119 mp, XFS_ERRTAG_IFLUSH_1)) {
3120 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3121 "%s: Bad inode %llu magic number 0x%x, ptr "PTR_FMT,
3122 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3123 goto flush_out;
3124 }
3125 if (S_ISREG(VFS_I(ip)->i_mode)) {
3126 if (XFS_TEST_ERROR(
3127 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3128 ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
3129 mp, XFS_ERRTAG_IFLUSH_3)) {
3130 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3131 "%s: Bad regular inode %llu, ptr "PTR_FMT,
3132 __func__, ip->i_ino, ip);
3133 goto flush_out;
3134 }
3135 } else if (S_ISDIR(VFS_I(ip)->i_mode)) {
3136 if (XFS_TEST_ERROR(
3137 ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
3138 ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
3139 ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
3140 mp, XFS_ERRTAG_IFLUSH_4)) {
3141 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3142 "%s: Bad directory inode %llu, ptr "PTR_FMT,
3143 __func__, ip->i_ino, ip);
3144 goto flush_out;
3145 }
3146 }
3147 if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af) >
3148 ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
3149 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3150 "%s: detected corrupt incore inode %llu, "
3151 "total extents = %llu nblocks = %lld, ptr "PTR_FMT,
3152 __func__, ip->i_ino,
3153 ip->i_df.if_nextents + xfs_ifork_nextents(&ip->i_af),
3154 ip->i_nblocks, ip);
3155 goto flush_out;
3156 }
3157 if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
3158 mp, XFS_ERRTAG_IFLUSH_6)) {
3159 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3160 "%s: bad inode %llu, forkoff 0x%x, ptr "PTR_FMT,
3161 __func__, ip->i_ino, ip->i_forkoff, ip);
3162 goto flush_out;
3163 }
3164
3165 /*
3166 * Inode item log recovery for v2 inodes are dependent on the flushiter
3167 * count for correct sequencing. We bump the flush iteration count so
3168 * we can detect flushes which postdate a log record during recovery.
3169 * This is redundant as we now log every change and hence this can't
3170 * happen but we need to still do it to ensure backwards compatibility
3171 * with old kernels that predate logging all inode changes.
3172 */
3173 if (!xfs_has_v3inodes(mp))
3174 ip->i_flushiter++;
3175
3176 /*
3177 * If there are inline format data / attr forks attached to this inode,
3178 * make sure they are not corrupt.
3179 */
3180 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
3181 xfs_ifork_verify_local_data(ip))
3182 goto flush_out;
3183 if (xfs_inode_has_attr_fork(ip) &&
3184 ip->i_af.if_format == XFS_DINODE_FMT_LOCAL &&
3185 xfs_ifork_verify_local_attr(ip))
3186 goto flush_out;
3187
3188 /*
3189 * Copy the dirty parts of the inode into the on-disk inode. We always
3190 * copy out the core of the inode, because if the inode is dirty at all
3191 * the core must be.
3192 */
3193 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
3194
3195 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3196 if (!xfs_has_v3inodes(mp)) {
3197 if (ip->i_flushiter == DI_MAX_FLUSH)
3198 ip->i_flushiter = 0;
3199 }
3200
3201 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
3202 if (xfs_inode_has_attr_fork(ip))
3203 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
3204
3205 /*
3206 * We've recorded everything logged in the inode, so we'd like to clear
3207 * the ili_fields bits so we don't log and flush things unnecessarily.
3208 * However, we can't stop logging all this information until the data
3209 * we've copied into the disk buffer is written to disk. If we did we
3210 * might overwrite the copy of the inode in the log with all the data
3211 * after re-logging only part of it, and in the face of a crash we
3212 * wouldn't have all the data we need to recover.
3213 *
3214 * What we do is move the bits to the ili_last_fields field. When
3215 * logging the inode, these bits are moved back to the ili_fields field.
3216 * In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
3217 * we know that the information those bits represent is permanently on
3218 * disk. As long as the flush completes before the inode is logged
3219 * again, then both ili_fields and ili_last_fields will be cleared.
3220 */
3221 error = 0;
3222flush_out:
3223 spin_lock(&iip->ili_lock);
3224 iip->ili_last_fields = iip->ili_fields;
3225 iip->ili_fields = 0;
3226 iip->ili_fsync_fields = 0;
3227 spin_unlock(&iip->ili_lock);
3228
3229 /*
3230 * Store the current LSN of the inode so that we can tell whether the
3231 * item has moved in the AIL from xfs_buf_inode_iodone().
3232 */
3233 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3234 &iip->ili_item.li_lsn);
3235
3236 /* generate the checksum. */
3237 xfs_dinode_calc_crc(mp, dip);
3238 return error;
3239}
3240
3241/*
3242 * Non-blocking flush of dirty inode metadata into the backing buffer.
3243 *
3244 * The caller must have a reference to the inode and hold the cluster buffer
3245 * locked. The function will walk across all the inodes on the cluster buffer it
3246 * can find and lock without blocking, and flush them to the cluster buffer.
3247 *
3248 * On successful flushing of at least one inode, the caller must write out the
3249 * buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
3250 * the caller needs to release the buffer. On failure, the filesystem will be
3251 * shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
3252 * will be returned.
3253 */
3254int
3255xfs_iflush_cluster(
3256 struct xfs_buf *bp)
3257{
3258 struct xfs_mount *mp = bp->b_mount;
3259 struct xfs_log_item *lip, *n;
3260 struct xfs_inode *ip;
3261 struct xfs_inode_log_item *iip;
3262 int clcount = 0;
3263 int error = 0;
3264
3265 /*
3266 * We must use the safe variant here as on shutdown xfs_iflush_abort()
3267 * will remove itself from the list.
3268 */
3269 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
3270 iip = (struct xfs_inode_log_item *)lip;
3271 ip = iip->ili_inode;
3272
3273 /*
3274 * Quick and dirty check to avoid locks if possible.
3275 */
3276 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
3277 continue;
3278 if (xfs_ipincount(ip))
3279 continue;
3280
3281 /*
3282 * The inode is still attached to the buffer, which means it is
3283 * dirty but reclaim might try to grab it. Check carefully for
3284 * that, and grab the ilock while still holding the i_flags_lock
3285 * to guarantee reclaim will not be able to reclaim this inode
3286 * once we drop the i_flags_lock.
3287 */
3288 spin_lock(&ip->i_flags_lock);
3289 ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
3290 if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
3291 spin_unlock(&ip->i_flags_lock);
3292 continue;
3293 }
3294
3295 /*
3296 * ILOCK will pin the inode against reclaim and prevent
3297 * concurrent transactions modifying the inode while we are
3298 * flushing the inode. If we get the lock, set the flushing
3299 * state before we drop the i_flags_lock.
3300 */
3301 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3302 spin_unlock(&ip->i_flags_lock);
3303 continue;
3304 }
3305 __xfs_iflags_set(ip, XFS_IFLUSHING);
3306 spin_unlock(&ip->i_flags_lock);
3307
3308 /*
3309 * Abort flushing this inode if we are shut down because the
3310 * inode may not currently be in the AIL. This can occur when
3311 * log I/O failure unpins the inode without inserting into the
3312 * AIL, leaving a dirty/unpinned inode attached to the buffer
3313 * that otherwise looks like it should be flushed.
3314 */
3315 if (xlog_is_shutdown(mp->m_log)) {
3316 xfs_iunpin_wait(ip);
3317 xfs_iflush_abort(ip);
3318 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3319 error = -EIO;
3320 continue;
3321 }
3322
3323 /* don't block waiting on a log force to unpin dirty inodes */
3324 if (xfs_ipincount(ip)) {
3325 xfs_iflags_clear(ip, XFS_IFLUSHING);
3326 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3327 continue;
3328 }
3329
3330 if (!xfs_inode_clean(ip))
3331 error = xfs_iflush(ip, bp);
3332 else
3333 xfs_iflags_clear(ip, XFS_IFLUSHING);
3334 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3335 if (error)
3336 break;
3337 clcount++;
3338 }
3339
3340 if (error) {
3341 /*
3342 * Shutdown first so we kill the log before we release this
3343 * buffer. If it is an INODE_ALLOC buffer and pins the tail
3344 * of the log, failing it before the _log_ is shut down can
3345 * result in the log tail being moved forward in the journal
3346 * on disk because log writes can still be taking place. Hence
3347 * unpinning the tail will allow the ICREATE intent to be
3348 * removed from the log an recovery will fail with uninitialised
3349 * inode cluster buffers.
3350 */
3351 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3352 bp->b_flags |= XBF_ASYNC;
3353 xfs_buf_ioend_fail(bp);
3354 return error;
3355 }
3356
3357 if (!clcount)
3358 return -EAGAIN;
3359
3360 XFS_STATS_INC(mp, xs_icluster_flushcnt);
3361 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
3362 return 0;
3363
3364}
3365
3366/* Release an inode. */
3367void
3368xfs_irele(
3369 struct xfs_inode *ip)
3370{
3371 trace_xfs_irele(ip, _RET_IP_);
3372 iput(VFS_I(ip));
3373}
3374
3375/*
3376 * Ensure all commited transactions touching the inode are written to the log.
3377 */
3378int
3379xfs_log_force_inode(
3380 struct xfs_inode *ip)
3381{
3382 xfs_csn_t seq = 0;
3383
3384 xfs_ilock(ip, XFS_ILOCK_SHARED);
3385 if (xfs_ipincount(ip))
3386 seq = ip->i_itemp->ili_commit_seq;
3387 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3388
3389 if (!seq)
3390 return 0;
3391 return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
3392}
3393
3394/*
3395 * Grab the exclusive iolock for a data copy from src to dest, making sure to
3396 * abide vfs locking order (lowest pointer value goes first) and breaking the
3397 * layout leases before proceeding. The loop is needed because we cannot call
3398 * the blocking break_layout() with the iolocks held, and therefore have to
3399 * back out both locks.
3400 */
3401static int
3402xfs_iolock_two_inodes_and_break_layout(
3403 struct inode *src,
3404 struct inode *dest)
3405{
3406 int error;
3407
3408 if (src > dest)
3409 swap(src, dest);
3410
3411retry:
3412 /* Wait to break both inodes' layouts before we start locking. */
3413 error = break_layout(src, true);
3414 if (error)
3415 return error;
3416 if (src != dest) {
3417 error = break_layout(dest, true);
3418 if (error)
3419 return error;
3420 }
3421
3422 /* Lock one inode and make sure nobody got in and leased it. */
3423 inode_lock(src);
3424 error = break_layout(src, false);
3425 if (error) {
3426 inode_unlock(src);
3427 if (error == -EWOULDBLOCK)
3428 goto retry;
3429 return error;
3430 }
3431
3432 if (src == dest)
3433 return 0;
3434
3435 /* Lock the other inode and make sure nobody got in and leased it. */
3436 inode_lock_nested(dest, I_MUTEX_NONDIR2);
3437 error = break_layout(dest, false);
3438 if (error) {
3439 inode_unlock(src);
3440 inode_unlock(dest);
3441 if (error == -EWOULDBLOCK)
3442 goto retry;
3443 return error;
3444 }
3445
3446 return 0;
3447}
3448
3449static int
3450xfs_mmaplock_two_inodes_and_break_dax_layout(
3451 struct xfs_inode *ip1,
3452 struct xfs_inode *ip2)
3453{
3454 int error;
3455 bool retry;
3456 struct page *page;
3457
3458 if (ip1->i_ino > ip2->i_ino)
3459 swap(ip1, ip2);
3460
3461again:
3462 retry = false;
3463 /* Lock the first inode */
3464 xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
3465 error = xfs_break_dax_layouts(VFS_I(ip1), &retry);
3466 if (error || retry) {
3467 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3468 if (error == 0 && retry)
3469 goto again;
3470 return error;
3471 }
3472
3473 if (ip1 == ip2)
3474 return 0;
3475
3476 /* Nested lock the second inode */
3477 xfs_ilock(ip2, xfs_lock_inumorder(XFS_MMAPLOCK_EXCL, 1));
3478 /*
3479 * We cannot use xfs_break_dax_layouts() directly here because it may
3480 * need to unlock & lock the XFS_MMAPLOCK_EXCL which is not suitable
3481 * for this nested lock case.
3482 */
3483 page = dax_layout_busy_page(VFS_I(ip2)->i_mapping);
3484 if (page && page_ref_count(page) != 1) {
3485 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3486 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3487 goto again;
3488 }
3489
3490 return 0;
3491}
3492
3493/*
3494 * Lock two inodes so that userspace cannot initiate I/O via file syscalls or
3495 * mmap activity.
3496 */
3497int
3498xfs_ilock2_io_mmap(
3499 struct xfs_inode *ip1,
3500 struct xfs_inode *ip2)
3501{
3502 int ret;
3503
3504 ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
3505 if (ret)
3506 return ret;
3507
3508 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3509 ret = xfs_mmaplock_two_inodes_and_break_dax_layout(ip1, ip2);
3510 if (ret) {
3511 inode_unlock(VFS_I(ip2));
3512 if (ip1 != ip2)
3513 inode_unlock(VFS_I(ip1));
3514 return ret;
3515 }
3516 } else
3517 filemap_invalidate_lock_two(VFS_I(ip1)->i_mapping,
3518 VFS_I(ip2)->i_mapping);
3519
3520 return 0;
3521}
3522
3523/* Unlock both inodes to allow IO and mmap activity. */
3524void
3525xfs_iunlock2_io_mmap(
3526 struct xfs_inode *ip1,
3527 struct xfs_inode *ip2)
3528{
3529 if (IS_DAX(VFS_I(ip1)) && IS_DAX(VFS_I(ip2))) {
3530 xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
3531 if (ip1 != ip2)
3532 xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
3533 } else
3534 filemap_invalidate_unlock_two(VFS_I(ip1)->i_mapping,
3535 VFS_I(ip2)->i_mapping);
3536
3537 inode_unlock(VFS_I(ip2));
3538 if (ip1 != ip2)
3539 inode_unlock(VFS_I(ip1));
3540}