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
xfs_flags2diflags(struct xfs_inode * ip,unsigned int xflags)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
xfs_flags2diflags2(struct xfs_inode * ip,unsigned int xflags)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
xfs_ip2xflags(struct xfs_inode * ip)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
xfs_get_initial_prid(struct xfs_inode * dp)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
xfs_inode_inherit_flags(struct xfs_inode * ip,const struct xfs_inode * pip)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
xfs_inode_inherit_flags2(struct xfs_inode * ip,const struct xfs_inode * pip)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 if (xfs_is_metadir_inode(pip))
228 ip->i_diflags2 |= XFS_DIFLAG2_METADATA;
229
230 /* Don't let invalid cowextsize hints propagate. */
231 failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
232 VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
233 if (failaddr) {
234 ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
235 ip->i_cowextsize = 0;
236 }
237 }
238
239 /*
240 * If we need to create attributes immediately after allocating the inode,
241 * initialise an empty attribute fork right now. We use the default fork offset
242 * for attributes here as we don't know exactly what size or how many
243 * attributes we might be adding. We can do this safely here because we know
244 * the data fork is completely empty and this saves us from needing to run a
245 * separate transaction to set the fork offset in the immediate future.
246 *
247 * If we have parent pointers and the caller hasn't told us that the file will
248 * never be linked into a directory tree, we /must/ create the attr fork.
249 */
250 static inline bool
xfs_icreate_want_attrfork(struct xfs_mount * mp,const struct xfs_icreate_args * args)251 xfs_icreate_want_attrfork(
252 struct xfs_mount *mp,
253 const struct xfs_icreate_args *args)
254 {
255 if (args->flags & XFS_ICREATE_INIT_XATTRS)
256 return true;
257
258 if (!(args->flags & XFS_ICREATE_UNLINKABLE) && xfs_has_parent(mp))
259 return true;
260
261 return false;
262 }
263
264 /* Initialise an inode's attributes. */
265 void
xfs_inode_init(struct xfs_trans * tp,const struct xfs_icreate_args * args,struct xfs_inode * ip)266 xfs_inode_init(
267 struct xfs_trans *tp,
268 const struct xfs_icreate_args *args,
269 struct xfs_inode *ip)
270 {
271 struct xfs_inode *pip = args->pip;
272 struct inode *dir = pip ? VFS_I(pip) : NULL;
273 struct xfs_mount *mp = tp->t_mountp;
274 struct inode *inode = VFS_I(ip);
275 unsigned int flags;
276 int times = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG |
277 XFS_ICHGTIME_ACCESS;
278
279 if (args->flags & XFS_ICREATE_TMPFILE)
280 set_nlink(inode, 0);
281 else if (S_ISDIR(args->mode))
282 set_nlink(inode, 2);
283 else
284 set_nlink(inode, 1);
285 inode->i_rdev = args->rdev;
286
287 if (!args->idmap || pip == NULL) {
288 /* creating a tree root, sb rooted, or detached file */
289 inode->i_uid = GLOBAL_ROOT_UID;
290 inode->i_gid = GLOBAL_ROOT_GID;
291 ip->i_projid = 0;
292 inode->i_mode = args->mode;
293 } else {
294 /* creating a child in the directory tree */
295 if (dir && !(dir->i_mode & S_ISGID) && xfs_has_grpid(mp)) {
296 inode_fsuid_set(inode, args->idmap);
297 inode->i_gid = dir->i_gid;
298 inode->i_mode = args->mode;
299 } else {
300 inode_init_owner(args->idmap, inode, dir, args->mode);
301 }
302 ip->i_projid = xfs_get_initial_prid(pip);
303 }
304
305 ip->i_disk_size = 0;
306 ip->i_df.if_nextents = 0;
307 ASSERT(ip->i_nblocks == 0);
308
309 ip->i_extsize = 0;
310 ip->i_diflags = 0;
311
312 if (xfs_has_v3inodes(mp)) {
313 inode_set_iversion(inode, 1);
314 /* also covers the di_used_blocks union arm: */
315 ip->i_cowextsize = 0;
316 times |= XFS_ICHGTIME_CREATE;
317 }
318
319 xfs_trans_ichgtime(tp, ip, times);
320
321 flags = XFS_ILOG_CORE;
322 switch (args->mode & S_IFMT) {
323 case S_IFIFO:
324 case S_IFCHR:
325 case S_IFBLK:
326 case S_IFSOCK:
327 ip->i_df.if_format = XFS_DINODE_FMT_DEV;
328 flags |= XFS_ILOG_DEV;
329 break;
330 case S_IFREG:
331 case S_IFDIR:
332 if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
333 xfs_inode_inherit_flags(ip, pip);
334 if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
335 xfs_inode_inherit_flags2(ip, pip);
336 fallthrough;
337 case S_IFLNK:
338 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
339 ip->i_df.if_bytes = 0;
340 ip->i_df.if_data = NULL;
341 break;
342 default:
343 ASSERT(0);
344 }
345
346 if (xfs_icreate_want_attrfork(mp, args)) {
347 ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
348 xfs_ifork_init_attr(ip, XFS_DINODE_FMT_EXTENTS, 0);
349
350 if (!xfs_has_attr(mp)) {
351 spin_lock(&mp->m_sb_lock);
352 xfs_add_attr(mp);
353 spin_unlock(&mp->m_sb_lock);
354 xfs_log_sb(tp);
355 }
356 }
357
358 xfs_trans_log_inode(tp, ip, flags);
359 }
360
361 /*
362 * In-Core Unlinked List Lookups
363 * =============================
364 *
365 * Every inode is supposed to be reachable from some other piece of metadata
366 * with the exception of the root directory. Inodes with a connection to a
367 * file descriptor but not linked from anywhere in the on-disk directory tree
368 * are collectively known as unlinked inodes, though the filesystem itself
369 * maintains links to these inodes so that on-disk metadata are consistent.
370 *
371 * XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
372 * header contains a number of buckets that point to an inode, and each inode
373 * record has a pointer to the next inode in the hash chain. This
374 * singly-linked list causes scaling problems in the iunlink remove function
375 * because we must walk that list to find the inode that points to the inode
376 * being removed from the unlinked hash bucket list.
377 *
378 * Hence we keep an in-memory double linked list to link each inode on an
379 * unlinked list. Because there are 64 unlinked lists per AGI, keeping pointer
380 * based lists would require having 64 list heads in the perag, one for each
381 * list. This is expensive in terms of memory (think millions of AGs) and cache
382 * misses on lookups. Instead, use the fact that inodes on the unlinked list
383 * must be referenced at the VFS level to keep them on the list and hence we
384 * have an existence guarantee for inodes on the unlinked list.
385 *
386 * Given we have an existence guarantee, we can use lockless inode cache lookups
387 * to resolve aginos to xfs inodes. This means we only need 8 bytes per inode
388 * for the double linked unlinked list, and we don't need any extra locking to
389 * keep the list safe as all manipulations are done under the AGI buffer lock.
390 * Keeping the list up to date does not require memory allocation, just finding
391 * the XFS inode and updating the next/prev unlinked list aginos.
392 */
393
394 /*
395 * Update the prev pointer of the next agino. Returns -ENOLINK if the inode
396 * is not in cache.
397 */
398 static int
xfs_iunlink_update_backref(struct xfs_perag * pag,xfs_agino_t prev_agino,xfs_agino_t next_agino)399 xfs_iunlink_update_backref(
400 struct xfs_perag *pag,
401 xfs_agino_t prev_agino,
402 xfs_agino_t next_agino)
403 {
404 struct xfs_inode *ip;
405
406 /* No update necessary if we are at the end of the list. */
407 if (next_agino == NULLAGINO)
408 return 0;
409
410 ip = xfs_iunlink_lookup(pag, next_agino);
411 if (!ip)
412 return -ENOLINK;
413
414 ip->i_prev_unlinked = prev_agino;
415 return 0;
416 }
417
418 /*
419 * Point the AGI unlinked bucket at an inode and log the results. The caller
420 * is responsible for validating the old value.
421 */
422 STATIC int
xfs_iunlink_update_bucket(struct xfs_trans * tp,struct xfs_perag * pag,struct xfs_buf * agibp,unsigned int bucket_index,xfs_agino_t new_agino)423 xfs_iunlink_update_bucket(
424 struct xfs_trans *tp,
425 struct xfs_perag *pag,
426 struct xfs_buf *agibp,
427 unsigned int bucket_index,
428 xfs_agino_t new_agino)
429 {
430 struct xfs_agi *agi = agibp->b_addr;
431 xfs_agino_t old_value;
432 int offset;
433
434 ASSERT(xfs_verify_agino_or_null(pag, new_agino));
435
436 old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
437 trace_xfs_iunlink_update_bucket(pag, bucket_index, old_value,
438 new_agino);
439
440 /*
441 * We should never find the head of the list already set to the value
442 * passed in because either we're adding or removing ourselves from the
443 * head of the list.
444 */
445 if (old_value == new_agino) {
446 xfs_buf_mark_corrupt(agibp);
447 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
448 return -EFSCORRUPTED;
449 }
450
451 agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
452 offset = offsetof(struct xfs_agi, agi_unlinked) +
453 (sizeof(xfs_agino_t) * bucket_index);
454 xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
455 return 0;
456 }
457
458 static int
xfs_iunlink_insert_inode(struct xfs_trans * tp,struct xfs_perag * pag,struct xfs_buf * agibp,struct xfs_inode * ip)459 xfs_iunlink_insert_inode(
460 struct xfs_trans *tp,
461 struct xfs_perag *pag,
462 struct xfs_buf *agibp,
463 struct xfs_inode *ip)
464 {
465 struct xfs_mount *mp = tp->t_mountp;
466 struct xfs_agi *agi = agibp->b_addr;
467 xfs_agino_t next_agino;
468 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
469 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
470 int error;
471
472 /*
473 * Get the index into the agi hash table for the list this inode will
474 * go on. Make sure the pointer isn't garbage and that this inode
475 * isn't already on the list.
476 */
477 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
478 if (next_agino == agino ||
479 !xfs_verify_agino_or_null(pag, next_agino)) {
480 xfs_buf_mark_corrupt(agibp);
481 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
482 return -EFSCORRUPTED;
483 }
484
485 /*
486 * Update the prev pointer in the next inode to point back to this
487 * inode.
488 */
489 error = xfs_iunlink_update_backref(pag, agino, next_agino);
490 if (error == -ENOLINK)
491 error = xfs_iunlink_reload_next(tp, agibp, agino, next_agino);
492 if (error)
493 return error;
494
495 if (next_agino != NULLAGINO) {
496 /*
497 * There is already another inode in the bucket, so point this
498 * inode to the current head of the list.
499 */
500 error = xfs_iunlink_log_inode(tp, ip, pag, next_agino);
501 if (error)
502 return error;
503 ip->i_next_unlinked = next_agino;
504 }
505
506 /* Point the head of the list to point to this inode. */
507 ip->i_prev_unlinked = NULLAGINO;
508 return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
509 }
510
511 /*
512 * This is called when the inode's link count has gone to 0 or we are creating
513 * a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
514 *
515 * We place the on-disk inode on a list in the AGI. It will be pulled from this
516 * list when the inode is freed.
517 */
518 int
xfs_iunlink(struct xfs_trans * tp,struct xfs_inode * ip)519 xfs_iunlink(
520 struct xfs_trans *tp,
521 struct xfs_inode *ip)
522 {
523 struct xfs_mount *mp = tp->t_mountp;
524 struct xfs_perag *pag;
525 struct xfs_buf *agibp;
526 int error;
527
528 ASSERT(VFS_I(ip)->i_nlink == 0);
529 ASSERT(VFS_I(ip)->i_mode != 0);
530 trace_xfs_iunlink(ip);
531
532 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
533
534 /* Get the agi buffer first. It ensures lock ordering on the list. */
535 error = xfs_read_agi(pag, tp, 0, &agibp);
536 if (error)
537 goto out;
538
539 error = xfs_iunlink_insert_inode(tp, pag, agibp, ip);
540 out:
541 xfs_perag_put(pag);
542 return error;
543 }
544
545 static int
xfs_iunlink_remove_inode(struct xfs_trans * tp,struct xfs_perag * pag,struct xfs_buf * agibp,struct xfs_inode * ip)546 xfs_iunlink_remove_inode(
547 struct xfs_trans *tp,
548 struct xfs_perag *pag,
549 struct xfs_buf *agibp,
550 struct xfs_inode *ip)
551 {
552 struct xfs_mount *mp = tp->t_mountp;
553 struct xfs_agi *agi = agibp->b_addr;
554 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
555 xfs_agino_t head_agino;
556 short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
557 int error;
558
559 trace_xfs_iunlink_remove(ip);
560
561 /*
562 * Get the index into the agi hash table for the list this inode will
563 * go on. Make sure the head pointer isn't garbage.
564 */
565 head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
566 if (!xfs_verify_agino(pag, head_agino)) {
567 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
568 agi, sizeof(*agi));
569 xfs_ag_mark_sick(pag, XFS_SICK_AG_AGI);
570 return -EFSCORRUPTED;
571 }
572
573 /*
574 * Set our inode's next_unlinked pointer to NULL and then return
575 * the old pointer value so that we can update whatever was previous
576 * to us in the list to point to whatever was next in the list.
577 */
578 error = xfs_iunlink_log_inode(tp, ip, pag, NULLAGINO);
579 if (error)
580 return error;
581
582 /*
583 * Update the prev pointer in the next inode to point back to previous
584 * inode in the chain.
585 */
586 error = xfs_iunlink_update_backref(pag, ip->i_prev_unlinked,
587 ip->i_next_unlinked);
588 if (error == -ENOLINK)
589 error = xfs_iunlink_reload_next(tp, agibp, ip->i_prev_unlinked,
590 ip->i_next_unlinked);
591 if (error)
592 return error;
593
594 if (head_agino != agino) {
595 struct xfs_inode *prev_ip;
596
597 prev_ip = xfs_iunlink_lookup(pag, ip->i_prev_unlinked);
598 if (!prev_ip) {
599 xfs_inode_mark_sick(ip, XFS_SICK_INO_CORE);
600 return -EFSCORRUPTED;
601 }
602
603 error = xfs_iunlink_log_inode(tp, prev_ip, pag,
604 ip->i_next_unlinked);
605 prev_ip->i_next_unlinked = ip->i_next_unlinked;
606 } else {
607 /* Point the head of the list to the next unlinked inode. */
608 error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
609 ip->i_next_unlinked);
610 }
611
612 ip->i_next_unlinked = NULLAGINO;
613 ip->i_prev_unlinked = 0;
614 return error;
615 }
616
617 /*
618 * Pull the on-disk inode from the AGI unlinked list.
619 */
620 int
xfs_iunlink_remove(struct xfs_trans * tp,struct xfs_perag * pag,struct xfs_inode * ip)621 xfs_iunlink_remove(
622 struct xfs_trans *tp,
623 struct xfs_perag *pag,
624 struct xfs_inode *ip)
625 {
626 struct xfs_buf *agibp;
627 int error;
628
629 trace_xfs_iunlink_remove(ip);
630
631 /* Get the agi buffer first. It ensures lock ordering on the list. */
632 error = xfs_read_agi(pag, tp, 0, &agibp);
633 if (error)
634 return error;
635
636 return xfs_iunlink_remove_inode(tp, pag, agibp, ip);
637 }
638
639 /*
640 * Decrement the link count on an inode & log the change. If this causes the
641 * link count to go to zero, move the inode to AGI unlinked list so that it can
642 * be freed when the last active reference goes away via xfs_inactive().
643 */
644 int
xfs_droplink(struct xfs_trans * tp,struct xfs_inode * ip)645 xfs_droplink(
646 struct xfs_trans *tp,
647 struct xfs_inode *ip)
648 {
649 struct inode *inode = VFS_I(ip);
650
651 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
652
653 if (inode->i_nlink == 0) {
654 xfs_info_ratelimited(tp->t_mountp,
655 "Inode 0x%llx link count dropped below zero. Pinning link count.",
656 ip->i_ino);
657 set_nlink(inode, XFS_NLINK_PINNED);
658 }
659 if (inode->i_nlink != XFS_NLINK_PINNED)
660 drop_nlink(inode);
661
662 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
663
664 if (inode->i_nlink)
665 return 0;
666
667 return xfs_iunlink(tp, ip);
668 }
669
670 /*
671 * Increment the link count on an inode & log the change.
672 */
673 void
xfs_bumplink(struct xfs_trans * tp,struct xfs_inode * ip)674 xfs_bumplink(
675 struct xfs_trans *tp,
676 struct xfs_inode *ip)
677 {
678 struct inode *inode = VFS_I(ip);
679
680 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
681
682 if (inode->i_nlink == XFS_NLINK_PINNED - 1)
683 xfs_info_ratelimited(tp->t_mountp,
684 "Inode 0x%llx link count exceeded maximum. Pinning link count.",
685 ip->i_ino);
686 if (inode->i_nlink != XFS_NLINK_PINNED)
687 inc_nlink(inode);
688
689 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
690 }
691
692 /* Free an inode in the ondisk index and zero it out. */
693 int
xfs_inode_uninit(struct xfs_trans * tp,struct xfs_perag * pag,struct xfs_inode * ip,struct xfs_icluster * xic)694 xfs_inode_uninit(
695 struct xfs_trans *tp,
696 struct xfs_perag *pag,
697 struct xfs_inode *ip,
698 struct xfs_icluster *xic)
699 {
700 struct xfs_mount *mp = ip->i_mount;
701 int error;
702
703 /*
704 * Free the inode first so that we guarantee that the AGI lock is going
705 * to be taken before we remove the inode from the unlinked list. This
706 * makes the AGI lock -> unlinked list modification order the same as
707 * used in O_TMPFILE creation.
708 */
709 error = xfs_difree(tp, pag, ip->i_ino, xic);
710 if (error)
711 return error;
712
713 error = xfs_iunlink_remove(tp, pag, ip);
714 if (error)
715 return error;
716
717 /*
718 * Free any local-format data sitting around before we reset the
719 * data fork to extents format. Note that the attr fork data has
720 * already been freed by xfs_attr_inactive.
721 */
722 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
723 kfree(ip->i_df.if_data);
724 ip->i_df.if_data = NULL;
725 ip->i_df.if_bytes = 0;
726 }
727
728 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
729 ip->i_diflags = 0;
730 ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
731 ip->i_forkoff = 0; /* mark the attr fork not in use */
732 ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
733
734 /*
735 * Bump the generation count so no one will be confused
736 * by reincarnations of this inode.
737 */
738 VFS_I(ip)->i_generation++;
739 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
740 return 0;
741 }
742