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 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_inode.h"
16 #include "xfs_inode_util.h"
17 #include "xfs_trans.h"
18 #include "xfs_ialloc.h"
19 #include "xfs_health.h"
20 #include "xfs_bmap.h"
21 #include "xfs_error.h"
22 #include "xfs_trace.h"
23 #include "xfs_ag.h"
24 #include "xfs_iunlink_item.h"
25 #include "xfs_inode_item.h"
26
27 uint16_t
28 xfs_flags2diflags(
29 struct xfs_inode *ip,
30 unsigned int xflags)
31 {
32 /* can't set PREALLOC this way, just preserve it */
33 uint16_t di_flags =
34 (ip->i_diflags & XFS_DIFLAG_PREALLOC);
35
36 if (xflags & FS_XFLAG_IMMUTABLE)
37 di_flags |= XFS_DIFLAG_IMMUTABLE;
38 if (xflags & FS_XFLAG_APPEND)
39 di_flags |= XFS_DIFLAG_APPEND;
40 if (xflags & FS_XFLAG_SYNC)
41 di_flags |= XFS_DIFLAG_SYNC;
42 if (xflags & FS_XFLAG_NOATIME)
43 di_flags |= XFS_DIFLAG_NOATIME;
44 if (xflags & FS_XFLAG_NODUMP)
45 di_flags |= XFS_DIFLAG_NODUMP;
46 if (xflags & FS_XFLAG_NODEFRAG)
47 di_flags |= XFS_DIFLAG_NODEFRAG;
48 if (xflags & FS_XFLAG_FILESTREAM)
49 di_flags |= XFS_DIFLAG_FILESTREAM;
50 if (S_ISDIR(VFS_I(ip)->i_mode)) {
51 if (xflags & FS_XFLAG_RTINHERIT)
52 di_flags |= XFS_DIFLAG_RTINHERIT;
53 if (xflags & FS_XFLAG_NOSYMLINKS)
54 di_flags |= XFS_DIFLAG_NOSYMLINKS;
55 if (xflags & FS_XFLAG_EXTSZINHERIT)
56 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
57 if (xflags & FS_XFLAG_PROJINHERIT)
58 di_flags |= XFS_DIFLAG_PROJINHERIT;
59 } else if (S_ISREG(VFS_I(ip)->i_mode)) {
60 if (xflags & FS_XFLAG_REALTIME)
61 di_flags |= XFS_DIFLAG_REALTIME;
62 if (xflags & FS_XFLAG_EXTSIZE)
63 di_flags |= XFS_DIFLAG_EXTSIZE;
64 }
65
66 return di_flags;
67 }
68
69 uint64_t
70 xfs_flags2diflags2(
71 struct xfs_inode *ip,
72 unsigned int xflags)
73 {
74 uint64_t di_flags2 =
75 (ip->i_diflags2 & (XFS_DIFLAG2_REFLINK |
76 XFS_DIFLAG2_BIGTIME |
77 XFS_DIFLAG2_NREXT64));
78
79 if (xflags & FS_XFLAG_DAX)
80 di_flags2 |= XFS_DIFLAG2_DAX;
81 if (xflags & FS_XFLAG_COWEXTSIZE)
82 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE;
83
84 return di_flags2;
85 }
86
87 uint32_t
88 xfs_ip2xflags(
89 struct xfs_inode *ip)
90 {
91 uint32_t flags = 0;
92
93 if (ip->i_diflags & XFS_DIFLAG_ANY) {
94 if (ip->i_diflags & XFS_DIFLAG_REALTIME)
95 flags |= FS_XFLAG_REALTIME;
96 if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
97 flags |= FS_XFLAG_PREALLOC;
98 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
99 flags |= FS_XFLAG_IMMUTABLE;
100 if (ip->i_diflags & XFS_DIFLAG_APPEND)
101 flags |= FS_XFLAG_APPEND;
102 if (ip->i_diflags & XFS_DIFLAG_SYNC)
103 flags |= FS_XFLAG_SYNC;
104 if (ip->i_diflags & XFS_DIFLAG_NOATIME)
105 flags |= FS_XFLAG_NOATIME;
106 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
107 flags |= FS_XFLAG_NODUMP;
108 if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
109 flags |= FS_XFLAG_RTINHERIT;
110 if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
111 flags |= FS_XFLAG_PROJINHERIT;
112 if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
113 flags |= FS_XFLAG_NOSYMLINKS;
114 if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
115 flags |= FS_XFLAG_EXTSIZE;
116 if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
117 flags |= FS_XFLAG_EXTSZINHERIT;
118 if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
119 flags |= FS_XFLAG_NODEFRAG;
120 if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
121 flags |= FS_XFLAG_FILESTREAM;
122 }
123
124 if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
125 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
126 flags |= FS_XFLAG_DAX;
127 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
128 flags |= FS_XFLAG_COWEXTSIZE;
129 }
130
131 if (xfs_inode_has_attr_fork(ip))
132 flags |= FS_XFLAG_HASATTR;
133 return flags;
134 }
135
136 prid_t
137 xfs_get_initial_prid(struct xfs_inode *dp)
138 {
139 if (dp->i_diflags & XFS_DIFLAG_PROJINHERIT)
140 return dp->i_projid;
141
142 /* Assign to the root project by default. */
143 return 0;
144 }
145
146 /* Propagate di_flags from a parent inode to a child inode. */
147 static inline void
148 xfs_inode_inherit_flags(
149 struct xfs_inode *ip,
150 const struct xfs_inode *pip)
151 {
152 unsigned int di_flags = 0;
153 xfs_failaddr_t failaddr;
154 umode_t mode = VFS_I(ip)->i_mode;
155
156 if (S_ISDIR(mode)) {
157 if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
158 di_flags |= XFS_DIFLAG_RTINHERIT;
159 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
160 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
161 ip->i_extsize = pip->i_extsize;
162 }
163 if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
164 di_flags |= XFS_DIFLAG_PROJINHERIT;
165 } else if (S_ISREG(mode)) {
166 if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
167 xfs_has_realtime(ip->i_mount))
168 di_flags |= XFS_DIFLAG_REALTIME;
169 if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
170 di_flags |= XFS_DIFLAG_EXTSIZE;
171 ip->i_extsize = pip->i_extsize;
172 }
173 }
174 if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
175 xfs_inherit_noatime)
176 di_flags |= XFS_DIFLAG_NOATIME;
177 if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
178 xfs_inherit_nodump)
179 di_flags |= XFS_DIFLAG_NODUMP;
180 if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
181 xfs_inherit_sync)
182 di_flags |= XFS_DIFLAG_SYNC;
183 if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
184 xfs_inherit_nosymlinks)
185 di_flags |= XFS_DIFLAG_NOSYMLINKS;
186 if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
187 xfs_inherit_nodefrag)
188 di_flags |= XFS_DIFLAG_NODEFRAG;
189 if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
190 di_flags |= XFS_DIFLAG_FILESTREAM;
191
192 ip->i_diflags |= di_flags;
193
194 /*
195 * Inode verifiers on older kernels only check that the extent size
196 * hint is an integer multiple of the rt extent size on realtime files.
197 * They did not check the hint alignment on a directory with both
198 * rtinherit and extszinherit flags set. If the misaligned hint is
199 * propagated from a directory into a new realtime file, new file
200 * allocations will fail due to math errors in the rt allocator and/or
201 * trip the verifiers. Validate the hint settings in the new file so
202 * that we don't let broken hints propagate.
203 */
204 failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
205 VFS_I(ip)->i_mode, ip->i_diflags);
206 if (failaddr) {
207 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
208 XFS_DIFLAG_EXTSZINHERIT);
209 ip->i_extsize = 0;
210 }
211 }
212
213 /* Propagate di_flags2 from a parent inode to a child inode. */
214 static inline void
215 xfs_inode_inherit_flags2(
216 struct xfs_inode *ip,
217 const struct xfs_inode *pip)
218 {
219 xfs_failaddr_t failaddr;
220
221 if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
222 ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
223 ip->i_cowextsize = pip->i_cowextsize;
224 }
225 if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
226 ip->i_diflags2 |= XFS_DIFLAG2_DAX;
227
228 /* Don't let invalid cowextsize hints propagate. */
229 failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
230 VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
231 if (failaddr) {
232 ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
233 ip->i_cowextsize = 0;
234 }
235 }
236
237 /*
238 * If we need to create attributes immediately after allocating the inode,
239 * initialise an empty attribute fork right now. We use the default fork offset
240 * for attributes here as we don't know exactly what size or how many
241 * attributes we might be adding. We can do this safely here because we know
242 * the data fork is completely empty and this saves us from needing to run a
243 * separate transaction to set the fork offset in the immediate future.
244 *
245 * If we have parent pointers and the caller hasn't told us that the file will
246 * never be linked into a directory tree, we /must/ create the attr fork.
247 */
248 static inline bool
249 xfs_icreate_want_attrfork(
250 struct xfs_mount *mp,
251 const struct xfs_icreate_args *args)
252 {
253 if (args->flags & XFS_ICREATE_INIT_XATTRS)
254 return true;
255
256 if (!(args->flags & XFS_ICREATE_UNLINKABLE) && xfs_has_parent(mp))
257 return true;
258
259 return false;
260 }
261
262 /* Initialise an inode's attributes. */
263 void
264 xfs_inode_init(
265 struct xfs_trans *tp,
266 const struct xfs_icreate_args *args,
267 struct xfs_inode *ip)
268 {
269 struct xfs_inode *pip = args->pip;
270 struct inode *dir = pip ? VFS_I(pip) : NULL;
271 struct xfs_mount *mp = tp->t_mountp;
272 struct inode *inode = VFS_I(ip);
273 unsigned int flags;
274 int times = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG |
275 XFS_ICHGTIME_ACCESS;
276
277 if (args->flags & XFS_ICREATE_TMPFILE)
278 set_nlink(inode, 0);
279 else if (S_ISDIR(args->mode))
280 set_nlink(inode, 2);
281 else
282 set_nlink(inode, 1);
283 inode->i_rdev = args->rdev;
284
285 if (!args->idmap || pip == NULL) {
286 /* creating a tree root, sb rooted, or detached file */
287 inode->i_uid = GLOBAL_ROOT_UID;
288 inode->i_gid = GLOBAL_ROOT_GID;
289 ip->i_projid = 0;
290 inode->i_mode = args->mode;
291 } else {
292 /* creating a child in the directory tree */
293 if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) {
294 inode_fsuid_set(inode, args->idmap);
295 inode->i_gid = dir->i_gid;
296 inode->i_mode = args->mode;
297 } else {
298 inode_init_owner(args->idmap, inode, dir, args->mode);
299 }
300
301 /*
302 * If the group ID of the new file does not match the effective
303 * group ID or one of the supplementary group IDs, the S_ISGID
304 * bit is cleared (and only if the irix_sgid_inherit
305 * compatibility variable is set).
306 */
307 if (irix_sgid_inherit && (inode->i_mode & S_ISGID) &&
308 !vfsgid_in_group_p(i_gid_into_vfsgid(args->idmap, inode)))
309 inode->i_mode &= ~S_ISGID;
310
311 ip->i_projid = pip ? xfs_get_initial_prid(pip) : 0;
312 }
313
314 ip->i_disk_size = 0;
315 ip->i_df.if_nextents = 0;
316 ASSERT(ip->i_nblocks == 0);
317
318 ip->i_extsize = 0;
319 ip->i_diflags = 0;
320
321 if (xfs_has_v3inodes(mp)) {
322 inode_set_iversion(inode, 1);
323 ip->i_cowextsize = 0;
324 times |= XFS_ICHGTIME_CREATE;
325 }
326
327 xfs_trans_ichgtime(tp, ip, times);
328
329 flags = XFS_ILOG_CORE;
330 switch (args->mode & S_IFMT) {
331 case S_IFIFO:
332 case S_IFCHR:
333 case S_IFBLK:
334 case S_IFSOCK:
335 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
336 flags |= XFS_ILOG_DEV;
337 break;
338 case S_IFREG:
339 case S_IFDIR:
340 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
341 xfs_inode_inherit_flags(ip, pip);
342 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
343 xfs_inode_inherit_flags2(ip, pip);
344 fallthrough;
345 case S_IFLNK:
346 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
347 ip->i_df.if_bytes = 0;
348 ip->i_df.if_data = NULL;
349 break;
350 default:
351 ASSERT(0);
352 }
353
354 if (xfs_icreate_want_attrfork(mp, args)) {
355 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
356 xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
357
358 if (!xfs_has_attr(mp)) {
359 spin_lock(&mp->m_sb_lock);
360 xfs_add_attr(mp);
361 spin_unlock(&mp->m_sb_lock);
362 xfs_log_sb(tp);
363 }
364 }
365
366 xfs_trans_log_inode(tp, ip, flags);
367 }
368
369 /*
370 * In-Core Unlinked List Lookups
371 * =============================
372 *
373 * Every inode is supposed to be reachable from some other piece of metadata
374 * with the exception of the root directory. Inodes with a connection to a
375 * file descriptor but not linked from anywhere in the on-disk directory tree
376 * are collectively known as unlinked inodes, though the filesystem itself
377 * maintains links to these inodes so that on-disk metadata are consistent.
378 *
379 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
380 * header contains a number of buckets that point to an inode, and each inode
381 * record has a pointer to the next inode in the hash chain. This
382 * singly-linked list causes scaling problems in the iunlink remove function
383 * because we must walk that list to find the inode that points to the inode
384 * being removed from the unlinked hash bucket list.
385 *
386 * Hence we keep an in-memory double linked list to link each inode on an
387 * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer
388 * based lists would require having 64 list heads in the perag, one for each
389 * list. This is expensive in terms of memory (think millions of AGs) and cache
390 * misses on lookups. Instead, use the fact that inodes on the unlinked list
391 * must be referenced at the VFS level to keep them on the list and hence we
392 * have an existence guarantee for inodes on the unlinked list.
393 *
394 * Given we have an existence guarantee, we can use lockless inode cache lookups
395 * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode
396 * for the double linked unlinked list, and we don't need any extra locking to
397 * keep the list safe as all manipulations are done under the AGI buffer lock.
398 * Keeping the list up to date does not require memory allocation, just finding
399 * the XFS inode and updating the next/prev unlinked list aginos.
400 */
401
402 /*
403 * Update the prev pointer of the next agino. Returns -ENOLINK if the inode
404 * is not in cache.
405 */
406 static int
407 xfs_iunlink_update_backref(
408 struct xfs_perag *pag,
409 xfs_agino_t prev_agino,
410 xfs_agino_t next_agino)
411 {
412 struct xfs_inode *ip;
413
414 /* No update necessary if we are at the end of the list. */
415 if (next_agino == NULLAGINO)
416 return 0;
417
418 ip = xfs_iunlink_lookup(pag, next_agino);
419 if (!ip)
420 return -ENOLINK;
421
422 ip->i_prev_unlinked = prev_agino;
423 return 0;
424 }
425
426 /*
427 * Point the AGI unlinked bucket at an inode and log the results. The caller
428 * is responsible for validating the old value.
429 */
430 STATIC int
431 xfs_iunlink_update_bucket(
432 struct xfs_trans *tp,
433 struct xfs_perag *pag,
434 struct xfs_buf *agibp,
435 unsigned int bucket_index,
436 xfs_agino_t new_agino)
437 {
438 struct xfs_agi *agi = agibp->b_addr;
439 xfs_agino_t old_value;
440 int offset;
441
442 ASSERT(xfs_verify_agino_or_null(pag, new_agino));
443
444 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
445 trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
446 old_value, new_agino);
447
448 /*
449 * We should never find the head of the list already set to the value
450 * passed in because either we're adding or removing ourselves from the
451 * head of the list.
452 */
453 if (old_value == new_agino) {
454 xfs_buf_mark_corrupt(agibp);
455 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
456 return -EFSCORRUPTED;
457 }
458
459 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
460 offset = offsetof(struct xfs_agi, agi_unlinked) +
461 (sizeof(xfs_agino_t) * bucket_index);
462 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
463 return 0;
464 }
465
466 static int
467 xfs_iunlink_insert_inode(
468 struct xfs_trans *tp,
469 struct xfs_perag *pag,
470 struct xfs_buf *agibp,
471 struct xfs_inode *ip)
472 {
473 struct xfs_mount *mp = tp->t_mountp;
474 struct xfs_agi *agi = agibp->b_addr;
475 xfs_agino_t next_agino;
476 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
477 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
478 int error;
479
480 /*
481 * Get the index into the agi hash table for the list this inode will
482 * go on. Make sure the pointer isn't garbage and that this inode
483 * isn't already on the list.
484 */
485 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
486 if (next_agino == agino ||
487 !xfs_verify_agino_or_null(pag, next_agino)) {
488 xfs_buf_mark_corrupt(agibp);
489 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
490 return -EFSCORRUPTED;
491 }
492
493 /*
494 * Update the prev pointer in the next inode to point back to this
495 * inode.
496 */
497 error = xfs_iunlink_update_backref(pag, agino, next_agino);
498 if (error == -ENOLINK)
499 error = xfs_iunlink_reload_next(tp, agibp, agino, next_agino);
500 if (error)
501 return error;
502
503 if (next_agino != NULLAGINO) {
504 /*
505 * There is already another inode in the bucket, so point this
506 * inode to the current head of the list.
507 */
508 error = xfs_iunlink_log_inode(tp, ip, pag, next_agino);
509 if (error)
510 return error;
511 ip->i_next_unlinked = next_agino;
512 }
513
514 /* Point the head of the list to point to this inode. */
515 ip->i_prev_unlinked = NULLAGINO;
516 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
517 }
518
519 /*
520 * This is called when the inode's link count has gone to 0 or we are creating
521 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
522 *
523 * We place the on-disk inode on a list in the AGI. It will be pulled from this
524 * list when the inode is freed.
525 */
526 int
527 xfs_iunlink(
528 struct xfs_trans *tp,
529 struct xfs_inode *ip)
530 {
531 struct xfs_mount *mp = tp->t_mountp;
532 struct xfs_perag *pag;
533 struct xfs_buf *agibp;
534 int error;
535
536 ASSERT(VFS_I(ip)->i_nlink == 0);
537 ASSERT(VFS_I(ip)->i_mode != 0);
538 trace_xfs_iunlink(ip);
539
540 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
541
542 /* Get the agi buffer first. It ensures lock ordering on the list. */
543 error = xfs_read_agi(pag, tp, 0, &agibp);
544 if (error)
545 goto out;
546
547 error = xfs_iunlink_insert_inode(tp, pag, agibp, ip);
548 out:
549 xfs_perag_put(pag);
550 return error;
551 }
552
553 static int
554 xfs_iunlink_remove_inode(
555 struct xfs_trans *tp,
556 struct xfs_perag *pag,
557 struct xfs_buf *agibp,
558 struct xfs_inode *ip)
559 {
560 struct xfs_mount *mp = tp->t_mountp;
561 struct xfs_agi *agi = agibp->b_addr;
562 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
563 xfs_agino_t head_agino;
564 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
565 int error;
566
567 trace_xfs_iunlink_remove(ip);
568
569 /*
570 * Get the index into the agi hash table for the list this inode will
571 * go on. Make sure the head pointer isn't garbage.
572 */
573 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
574 if (!xfs_verify_agino(pag, head_agino)) {
575 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
576 agi, sizeof(*agi));
577 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
578 return -EFSCORRUPTED;
579 }
580
581 /*
582 * Set our inode's next_unlinked pointer to NULL and then return
583 * the old pointer value so that we can update whatever was previous
584 * to us in the list to point to whatever was next in the list.
585 */
586 error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO);
587 if (error)
588 return error;
589
590 /*
591 * Update the prev pointer in the next inode to point back to previous
592 * inode in the chain.
593 */
594 error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked,
595 ip->i_next_unlinked);
596 if (error == -ENOLINK)
597 error = xfs_iunlink_reload_next(tp, agibp, ip->i_prev_unlinked,
598 ip->i_next_unlinked);
599 if (error)
600 return error;
601
602 if (head_agino != agino) {
603 struct xfs_inode *prev_ip;
604
605 prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked);
606 if (!prev_ip) {
607 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
608 return -EFSCORRUPTED;
609 }
610
611 error = xfs_iunlink_log_inode(tp, prev_ip, pag,
612 ip->i_next_unlinked);
613 prev_ip->i_next_unlinked = ip->i_next_unlinked;
614 } else {
615 /* Point the head of the list to the next unlinked inode. */
616 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
617 ip->i_next_unlinked);
618 }
619
620 ip->i_next_unlinked = NULLAGINO;
621 ip->i_prev_unlinked = 0;
622 return error;
623 }
624
625 /*
626 * Pull the on-disk inode from the AGI unlinked list.
627 */
628 int
629 xfs_iunlink_remove(
630 struct xfs_trans *tp,
631 struct xfs_perag *pag,
632 struct xfs_inode *ip)
633 {
634 struct xfs_buf *agibp;
635 int error;
636
637 trace_xfs_iunlink_remove(ip);
638
639 /* Get the agi buffer first. It ensures lock ordering on the list. */
640 error = xfs_read_agi(pag, tp, 0, &agibp);
641 if (error)
642 return error;
643
644 return xfs_iunlink_remove_inode(tp, pag, agibp, ip);
645 }
646
647 /*
648 * Decrement the link count on an inode & log the change. If this causes the
649 * link count to go to zero, move the inode to AGI unlinked list so that it can
650 * be freed when the last active reference goes away via xfs_inactive().
651 */
652 int
653 xfs_droplink(
654 struct xfs_trans *tp,
655 struct xfs_inode *ip)
656 {
657 struct inode *inode = VFS_I(ip);
658
659 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
660
661 if (inode->i_nlink == 0) {
662 xfs_info_ratelimited(tp->t_mountp,
663 "Inode 0x%llx link count dropped below zero. Pinning link count.",
664 ip->i_ino);
665 set_nlink(inode, XFS_NLINK_PINNED);
666 }
667 if (inode->i_nlink != XFS_NLINK_PINNED)
668 drop_nlink(inode);
669
670 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
671
672 if (inode->i_nlink)
673 return 0;
674
675 return xfs_iunlink(tp, ip);
676 }
677
678 /*
679 * Increment the link count on an inode & log the change.
680 */
681 void
682 xfs_bumplink(
683 struct xfs_trans *tp,
684 struct xfs_inode *ip)
685 {
686 struct inode *inode = VFS_I(ip);
687
688 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
689
690 if (inode->i_nlink == XFS_NLINK_PINNED - 1)
691 xfs_info_ratelimited(tp->t_mountp,
692 "Inode 0x%llx link count exceeded maximum. Pinning link count.",
693 ip->i_ino);
694 if (inode->i_nlink != XFS_NLINK_PINNED)
695 inc_nlink(inode);
696
697 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
698 }
699
700 /* Free an inode in the ondisk index and zero it out. */
701 int
702 xfs_inode_uninit(
703 struct xfs_trans *tp,
704 struct xfs_perag *pag,
705 struct xfs_inode *ip,
706 struct xfs_icluster *xic)
707 {
708 struct xfs_mount *mp = ip->i_mount;
709 int error;
710
711 /*
712 * Free the inode first so that we guarantee that the AGI lock is going
713 * to be taken before we remove the inode from the unlinked list. This
714 * makes the AGI lock -> unlinked list modification order the same as
715 * used in O_TMPFILE creation.
716 */
717 error = xfs_difree(tp, pag, ip->i_ino, xic);
718 if (error)
719 return error;
720
721 error = xfs_iunlink_remove(tp, pag, ip);
722 if (error)
723 return error;
724
725 /*
726 * Free any local-format data sitting around before we reset the
727 * data fork to extents format. Note that the attr fork data has
728 * already been freed by xfs_attr_inactive.
729 */
730 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
731 kfree(ip->i_df.if_data);
732 ip->i_df.if_data = NULL;
733 ip->i_df.if_bytes = 0;
734 }
735
736 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
737 ip->i_diflags = 0;
738 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
739 ip->i_forkoff = 0; /* mark the attr fork not in use */
740 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
741
742 /*
743 * Bump the generation count so no one will be confused
744 * by reincarnations of this inode.
745 */
746 VFS_I(ip)->i_generation++;
747 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
748 return 0;
749 }
750
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.