Loading...
1/*
2 * Copyright (c) 2000-2005 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 "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_acl.h"
21#include "xfs_bit.h"
22#include "xfs_log.h"
23#include "xfs_inum.h"
24#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
27#include "xfs_alloc.h"
28#include "xfs_quota.h"
29#include "xfs_mount.h"
30#include "xfs_bmap_btree.h"
31#include "xfs_dinode.h"
32#include "xfs_inode.h"
33#include "xfs_bmap.h"
34#include "xfs_rtalloc.h"
35#include "xfs_error.h"
36#include "xfs_itable.h"
37#include "xfs_rw.h"
38#include "xfs_attr.h"
39#include "xfs_buf_item.h"
40#include "xfs_utils.h"
41#include "xfs_vnodeops.h"
42#include "xfs_inode_item.h"
43#include "xfs_trace.h"
44
45#include <linux/capability.h>
46#include <linux/xattr.h>
47#include <linux/namei.h>
48#include <linux/posix_acl.h>
49#include <linux/security.h>
50#include <linux/fiemap.h>
51#include <linux/slab.h>
52
53/*
54 * Bring the timestamps in the XFS inode uptodate.
55 *
56 * Used before writing the inode to disk.
57 */
58void
59xfs_synchronize_times(
60 xfs_inode_t *ip)
61{
62 struct inode *inode = VFS_I(ip);
63
64 ip->i_d.di_atime.t_sec = (__int32_t)inode->i_atime.tv_sec;
65 ip->i_d.di_atime.t_nsec = (__int32_t)inode->i_atime.tv_nsec;
66 ip->i_d.di_ctime.t_sec = (__int32_t)inode->i_ctime.tv_sec;
67 ip->i_d.di_ctime.t_nsec = (__int32_t)inode->i_ctime.tv_nsec;
68 ip->i_d.di_mtime.t_sec = (__int32_t)inode->i_mtime.tv_sec;
69 ip->i_d.di_mtime.t_nsec = (__int32_t)inode->i_mtime.tv_nsec;
70}
71
72/*
73 * If the linux inode is valid, mark it dirty, else mark the dirty state
74 * in the XFS inode to make sure we pick it up when reclaiming the inode.
75 */
76void
77xfs_mark_inode_dirty_sync(
78 xfs_inode_t *ip)
79{
80 struct inode *inode = VFS_I(ip);
81
82 if (!(inode->i_state & (I_WILL_FREE|I_FREEING)))
83 mark_inode_dirty_sync(inode);
84 else {
85 barrier();
86 ip->i_update_core = 1;
87 }
88}
89
90void
91xfs_mark_inode_dirty(
92 xfs_inode_t *ip)
93{
94 struct inode *inode = VFS_I(ip);
95
96 if (!(inode->i_state & (I_WILL_FREE|I_FREEING)))
97 mark_inode_dirty(inode);
98 else {
99 barrier();
100 ip->i_update_core = 1;
101 }
102
103}
104
105/*
106 * Hook in SELinux. This is not quite correct yet, what we really need
107 * here (as we do for default ACLs) is a mechanism by which creation of
108 * these attrs can be journalled at inode creation time (along with the
109 * inode, of course, such that log replay can't cause these to be lost).
110 */
111STATIC int
112xfs_init_security(
113 struct inode *inode,
114 struct inode *dir,
115 const struct qstr *qstr)
116{
117 struct xfs_inode *ip = XFS_I(inode);
118 size_t length;
119 void *value;
120 unsigned char *name;
121 int error;
122
123 error = security_inode_init_security(inode, dir, qstr, (char **)&name,
124 &value, &length);
125 if (error) {
126 if (error == -EOPNOTSUPP)
127 return 0;
128 return -error;
129 }
130
131 error = xfs_attr_set(ip, name, value, length, ATTR_SECURE);
132
133 kfree(name);
134 kfree(value);
135 return error;
136}
137
138static void
139xfs_dentry_to_name(
140 struct xfs_name *namep,
141 struct dentry *dentry)
142{
143 namep->name = dentry->d_name.name;
144 namep->len = dentry->d_name.len;
145}
146
147STATIC void
148xfs_cleanup_inode(
149 struct inode *dir,
150 struct inode *inode,
151 struct dentry *dentry)
152{
153 struct xfs_name teardown;
154
155 /* Oh, the horror.
156 * If we can't add the ACL or we fail in
157 * xfs_init_security we must back out.
158 * ENOSPC can hit here, among other things.
159 */
160 xfs_dentry_to_name(&teardown, dentry);
161
162 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
163 iput(inode);
164}
165
166STATIC int
167xfs_vn_mknod(
168 struct inode *dir,
169 struct dentry *dentry,
170 int mode,
171 dev_t rdev)
172{
173 struct inode *inode;
174 struct xfs_inode *ip = NULL;
175 struct posix_acl *default_acl = NULL;
176 struct xfs_name name;
177 int error;
178
179 /*
180 * Irix uses Missed'em'V split, but doesn't want to see
181 * the upper 5 bits of (14bit) major.
182 */
183 if (S_ISCHR(mode) || S_ISBLK(mode)) {
184 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
185 return -EINVAL;
186 rdev = sysv_encode_dev(rdev);
187 } else {
188 rdev = 0;
189 }
190
191 if (IS_POSIXACL(dir)) {
192 default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
193 if (IS_ERR(default_acl))
194 return PTR_ERR(default_acl);
195
196 if (!default_acl)
197 mode &= ~current_umask();
198 }
199
200 xfs_dentry_to_name(&name, dentry);
201 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
202 if (unlikely(error))
203 goto out_free_acl;
204
205 inode = VFS_I(ip);
206
207 error = xfs_init_security(inode, dir, &dentry->d_name);
208 if (unlikely(error))
209 goto out_cleanup_inode;
210
211 if (default_acl) {
212 error = -xfs_inherit_acl(inode, default_acl);
213 default_acl = NULL;
214 if (unlikely(error))
215 goto out_cleanup_inode;
216 }
217
218
219 d_instantiate(dentry, inode);
220 return -error;
221
222 out_cleanup_inode:
223 xfs_cleanup_inode(dir, inode, dentry);
224 out_free_acl:
225 posix_acl_release(default_acl);
226 return -error;
227}
228
229STATIC int
230xfs_vn_create(
231 struct inode *dir,
232 struct dentry *dentry,
233 int mode,
234 struct nameidata *nd)
235{
236 return xfs_vn_mknod(dir, dentry, mode, 0);
237}
238
239STATIC int
240xfs_vn_mkdir(
241 struct inode *dir,
242 struct dentry *dentry,
243 int mode)
244{
245 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
246}
247
248STATIC struct dentry *
249xfs_vn_lookup(
250 struct inode *dir,
251 struct dentry *dentry,
252 struct nameidata *nd)
253{
254 struct xfs_inode *cip;
255 struct xfs_name name;
256 int error;
257
258 if (dentry->d_name.len >= MAXNAMELEN)
259 return ERR_PTR(-ENAMETOOLONG);
260
261 xfs_dentry_to_name(&name, dentry);
262 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
263 if (unlikely(error)) {
264 if (unlikely(error != ENOENT))
265 return ERR_PTR(-error);
266 d_add(dentry, NULL);
267 return NULL;
268 }
269
270 return d_splice_alias(VFS_I(cip), dentry);
271}
272
273STATIC struct dentry *
274xfs_vn_ci_lookup(
275 struct inode *dir,
276 struct dentry *dentry,
277 struct nameidata *nd)
278{
279 struct xfs_inode *ip;
280 struct xfs_name xname;
281 struct xfs_name ci_name;
282 struct qstr dname;
283 int error;
284
285 if (dentry->d_name.len >= MAXNAMELEN)
286 return ERR_PTR(-ENAMETOOLONG);
287
288 xfs_dentry_to_name(&xname, dentry);
289 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
290 if (unlikely(error)) {
291 if (unlikely(error != ENOENT))
292 return ERR_PTR(-error);
293 /*
294 * call d_add(dentry, NULL) here when d_drop_negative_children
295 * is called in xfs_vn_mknod (ie. allow negative dentries
296 * with CI filesystems).
297 */
298 return NULL;
299 }
300
301 /* if exact match, just splice and exit */
302 if (!ci_name.name)
303 return d_splice_alias(VFS_I(ip), dentry);
304
305 /* else case-insensitive match... */
306 dname.name = ci_name.name;
307 dname.len = ci_name.len;
308 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
309 kmem_free(ci_name.name);
310 return dentry;
311}
312
313STATIC int
314xfs_vn_link(
315 struct dentry *old_dentry,
316 struct inode *dir,
317 struct dentry *dentry)
318{
319 struct inode *inode = old_dentry->d_inode;
320 struct xfs_name name;
321 int error;
322
323 xfs_dentry_to_name(&name, dentry);
324
325 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
326 if (unlikely(error))
327 return -error;
328
329 ihold(inode);
330 d_instantiate(dentry, inode);
331 return 0;
332}
333
334STATIC int
335xfs_vn_unlink(
336 struct inode *dir,
337 struct dentry *dentry)
338{
339 struct xfs_name name;
340 int error;
341
342 xfs_dentry_to_name(&name, dentry);
343
344 error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
345 if (error)
346 return error;
347
348 /*
349 * With unlink, the VFS makes the dentry "negative": no inode,
350 * but still hashed. This is incompatible with case-insensitive
351 * mode, so invalidate (unhash) the dentry in CI-mode.
352 */
353 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
354 d_invalidate(dentry);
355 return 0;
356}
357
358STATIC int
359xfs_vn_symlink(
360 struct inode *dir,
361 struct dentry *dentry,
362 const char *symname)
363{
364 struct inode *inode;
365 struct xfs_inode *cip = NULL;
366 struct xfs_name name;
367 int error;
368 mode_t mode;
369
370 mode = S_IFLNK |
371 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
372 xfs_dentry_to_name(&name, dentry);
373
374 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
375 if (unlikely(error))
376 goto out;
377
378 inode = VFS_I(cip);
379
380 error = xfs_init_security(inode, dir, &dentry->d_name);
381 if (unlikely(error))
382 goto out_cleanup_inode;
383
384 d_instantiate(dentry, inode);
385 return 0;
386
387 out_cleanup_inode:
388 xfs_cleanup_inode(dir, inode, dentry);
389 out:
390 return -error;
391}
392
393STATIC int
394xfs_vn_rename(
395 struct inode *odir,
396 struct dentry *odentry,
397 struct inode *ndir,
398 struct dentry *ndentry)
399{
400 struct inode *new_inode = ndentry->d_inode;
401 struct xfs_name oname;
402 struct xfs_name nname;
403
404 xfs_dentry_to_name(&oname, odentry);
405 xfs_dentry_to_name(&nname, ndentry);
406
407 return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
408 XFS_I(ndir), &nname, new_inode ?
409 XFS_I(new_inode) : NULL);
410}
411
412/*
413 * careful here - this function can get called recursively, so
414 * we need to be very careful about how much stack we use.
415 * uio is kmalloced for this reason...
416 */
417STATIC void *
418xfs_vn_follow_link(
419 struct dentry *dentry,
420 struct nameidata *nd)
421{
422 char *link;
423 int error = -ENOMEM;
424
425 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
426 if (!link)
427 goto out_err;
428
429 error = -xfs_readlink(XFS_I(dentry->d_inode), link);
430 if (unlikely(error))
431 goto out_kfree;
432
433 nd_set_link(nd, link);
434 return NULL;
435
436 out_kfree:
437 kfree(link);
438 out_err:
439 nd_set_link(nd, ERR_PTR(error));
440 return NULL;
441}
442
443STATIC void
444xfs_vn_put_link(
445 struct dentry *dentry,
446 struct nameidata *nd,
447 void *p)
448{
449 char *s = nd_get_link(nd);
450
451 if (!IS_ERR(s))
452 kfree(s);
453}
454
455STATIC int
456xfs_vn_getattr(
457 struct vfsmount *mnt,
458 struct dentry *dentry,
459 struct kstat *stat)
460{
461 struct inode *inode = dentry->d_inode;
462 struct xfs_inode *ip = XFS_I(inode);
463 struct xfs_mount *mp = ip->i_mount;
464
465 trace_xfs_getattr(ip);
466
467 if (XFS_FORCED_SHUTDOWN(mp))
468 return XFS_ERROR(EIO);
469
470 stat->size = XFS_ISIZE(ip);
471 stat->dev = inode->i_sb->s_dev;
472 stat->mode = ip->i_d.di_mode;
473 stat->nlink = ip->i_d.di_nlink;
474 stat->uid = ip->i_d.di_uid;
475 stat->gid = ip->i_d.di_gid;
476 stat->ino = ip->i_ino;
477 stat->atime = inode->i_atime;
478 stat->mtime = inode->i_mtime;
479 stat->ctime = inode->i_ctime;
480 stat->blocks =
481 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
482
483
484 switch (inode->i_mode & S_IFMT) {
485 case S_IFBLK:
486 case S_IFCHR:
487 stat->blksize = BLKDEV_IOSIZE;
488 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
489 sysv_minor(ip->i_df.if_u2.if_rdev));
490 break;
491 default:
492 if (XFS_IS_REALTIME_INODE(ip)) {
493 /*
494 * If the file blocks are being allocated from a
495 * realtime volume, then return the inode's realtime
496 * extent size or the realtime volume's extent size.
497 */
498 stat->blksize =
499 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
500 } else
501 stat->blksize = xfs_preferred_iosize(mp);
502 stat->rdev = 0;
503 break;
504 }
505
506 return 0;
507}
508
509int
510xfs_setattr_nonsize(
511 struct xfs_inode *ip,
512 struct iattr *iattr,
513 int flags)
514{
515 xfs_mount_t *mp = ip->i_mount;
516 struct inode *inode = VFS_I(ip);
517 int mask = iattr->ia_valid;
518 xfs_trans_t *tp;
519 int error;
520 uid_t uid = 0, iuid = 0;
521 gid_t gid = 0, igid = 0;
522 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
523 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
524
525 trace_xfs_setattr(ip);
526
527 if (mp->m_flags & XFS_MOUNT_RDONLY)
528 return XFS_ERROR(EROFS);
529
530 if (XFS_FORCED_SHUTDOWN(mp))
531 return XFS_ERROR(EIO);
532
533 error = -inode_change_ok(inode, iattr);
534 if (error)
535 return XFS_ERROR(error);
536
537 ASSERT((mask & ATTR_SIZE) == 0);
538
539 /*
540 * If disk quotas is on, we make sure that the dquots do exist on disk,
541 * before we start any other transactions. Trying to do this later
542 * is messy. We don't care to take a readlock to look at the ids
543 * in inode here, because we can't hold it across the trans_reserve.
544 * If the IDs do change before we take the ilock, we're covered
545 * because the i_*dquot fields will get updated anyway.
546 */
547 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
548 uint qflags = 0;
549
550 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
551 uid = iattr->ia_uid;
552 qflags |= XFS_QMOPT_UQUOTA;
553 } else {
554 uid = ip->i_d.di_uid;
555 }
556 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
557 gid = iattr->ia_gid;
558 qflags |= XFS_QMOPT_GQUOTA;
559 } else {
560 gid = ip->i_d.di_gid;
561 }
562
563 /*
564 * We take a reference when we initialize udqp and gdqp,
565 * so it is important that we never blindly double trip on
566 * the same variable. See xfs_create() for an example.
567 */
568 ASSERT(udqp == NULL);
569 ASSERT(gdqp == NULL);
570 error = xfs_qm_vop_dqalloc(ip, uid, gid, xfs_get_projid(ip),
571 qflags, &udqp, &gdqp);
572 if (error)
573 return error;
574 }
575
576 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
577 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0);
578 if (error)
579 goto out_dqrele;
580
581 xfs_ilock(ip, XFS_ILOCK_EXCL);
582
583 /*
584 * Change file ownership. Must be the owner or privileged.
585 */
586 if (mask & (ATTR_UID|ATTR_GID)) {
587 /*
588 * These IDs could have changed since we last looked at them.
589 * But, we're assured that if the ownership did change
590 * while we didn't have the inode locked, inode's dquot(s)
591 * would have changed also.
592 */
593 iuid = ip->i_d.di_uid;
594 igid = ip->i_d.di_gid;
595 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
596 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
597
598 /*
599 * Do a quota reservation only if uid/gid is actually
600 * going to change.
601 */
602 if (XFS_IS_QUOTA_RUNNING(mp) &&
603 ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
604 (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
605 ASSERT(tp);
606 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
607 capable(CAP_FOWNER) ?
608 XFS_QMOPT_FORCE_RES : 0);
609 if (error) /* out of quota */
610 goto out_trans_cancel;
611 }
612 }
613
614 xfs_trans_ijoin(tp, ip);
615
616 /*
617 * Change file ownership. Must be the owner or privileged.
618 */
619 if (mask & (ATTR_UID|ATTR_GID)) {
620 /*
621 * CAP_FSETID overrides the following restrictions:
622 *
623 * The set-user-ID and set-group-ID bits of a file will be
624 * cleared upon successful return from chown()
625 */
626 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
627 !capable(CAP_FSETID))
628 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
629
630 /*
631 * Change the ownerships and register quota modifications
632 * in the transaction.
633 */
634 if (iuid != uid) {
635 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
636 ASSERT(mask & ATTR_UID);
637 ASSERT(udqp);
638 olddquot1 = xfs_qm_vop_chown(tp, ip,
639 &ip->i_udquot, udqp);
640 }
641 ip->i_d.di_uid = uid;
642 inode->i_uid = uid;
643 }
644 if (igid != gid) {
645 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
646 ASSERT(!XFS_IS_PQUOTA_ON(mp));
647 ASSERT(mask & ATTR_GID);
648 ASSERT(gdqp);
649 olddquot2 = xfs_qm_vop_chown(tp, ip,
650 &ip->i_gdquot, gdqp);
651 }
652 ip->i_d.di_gid = gid;
653 inode->i_gid = gid;
654 }
655 }
656
657 /*
658 * Change file access modes.
659 */
660 if (mask & ATTR_MODE) {
661 umode_t mode = iattr->ia_mode;
662
663 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
664 mode &= ~S_ISGID;
665
666 ip->i_d.di_mode &= S_IFMT;
667 ip->i_d.di_mode |= mode & ~S_IFMT;
668
669 inode->i_mode &= S_IFMT;
670 inode->i_mode |= mode & ~S_IFMT;
671 }
672
673 /*
674 * Change file access or modified times.
675 */
676 if (mask & ATTR_ATIME) {
677 inode->i_atime = iattr->ia_atime;
678 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
679 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
680 ip->i_update_core = 1;
681 }
682 if (mask & ATTR_CTIME) {
683 inode->i_ctime = iattr->ia_ctime;
684 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
685 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
686 ip->i_update_core = 1;
687 }
688 if (mask & ATTR_MTIME) {
689 inode->i_mtime = iattr->ia_mtime;
690 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
691 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
692 ip->i_update_core = 1;
693 }
694
695 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
696
697 XFS_STATS_INC(xs_ig_attrchg);
698
699 if (mp->m_flags & XFS_MOUNT_WSYNC)
700 xfs_trans_set_sync(tp);
701 error = xfs_trans_commit(tp, 0);
702
703 xfs_iunlock(ip, XFS_ILOCK_EXCL);
704
705 /*
706 * Release any dquot(s) the inode had kept before chown.
707 */
708 xfs_qm_dqrele(olddquot1);
709 xfs_qm_dqrele(olddquot2);
710 xfs_qm_dqrele(udqp);
711 xfs_qm_dqrele(gdqp);
712
713 if (error)
714 return XFS_ERROR(error);
715
716 /*
717 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
718 * update. We could avoid this with linked transactions
719 * and passing down the transaction pointer all the way
720 * to attr_set. No previous user of the generic
721 * Posix ACL code seems to care about this issue either.
722 */
723 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
724 error = -xfs_acl_chmod(inode);
725 if (error)
726 return XFS_ERROR(error);
727 }
728
729 return 0;
730
731out_trans_cancel:
732 xfs_trans_cancel(tp, 0);
733 xfs_iunlock(ip, XFS_ILOCK_EXCL);
734out_dqrele:
735 xfs_qm_dqrele(udqp);
736 xfs_qm_dqrele(gdqp);
737 return error;
738}
739
740/*
741 * Truncate file. Must have write permission and not be a directory.
742 */
743int
744xfs_setattr_size(
745 struct xfs_inode *ip,
746 struct iattr *iattr,
747 int flags)
748{
749 struct xfs_mount *mp = ip->i_mount;
750 struct inode *inode = VFS_I(ip);
751 int mask = iattr->ia_valid;
752 struct xfs_trans *tp;
753 int error;
754 uint lock_flags;
755 uint commit_flags = 0;
756
757 trace_xfs_setattr(ip);
758
759 if (mp->m_flags & XFS_MOUNT_RDONLY)
760 return XFS_ERROR(EROFS);
761
762 if (XFS_FORCED_SHUTDOWN(mp))
763 return XFS_ERROR(EIO);
764
765 error = -inode_change_ok(inode, iattr);
766 if (error)
767 return XFS_ERROR(error);
768
769 ASSERT(S_ISREG(ip->i_d.di_mode));
770 ASSERT((mask & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
771 ATTR_MTIME_SET|ATTR_KILL_SUID|ATTR_KILL_SGID|
772 ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
773
774 lock_flags = XFS_ILOCK_EXCL;
775 if (!(flags & XFS_ATTR_NOLOCK))
776 lock_flags |= XFS_IOLOCK_EXCL;
777 xfs_ilock(ip, lock_flags);
778
779 /*
780 * Short circuit the truncate case for zero length files.
781 */
782 if (iattr->ia_size == 0 &&
783 ip->i_size == 0 && ip->i_d.di_nextents == 0) {
784 if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
785 goto out_unlock;
786
787 /*
788 * Use the regular setattr path to update the timestamps.
789 */
790 xfs_iunlock(ip, lock_flags);
791 iattr->ia_valid &= ~ATTR_SIZE;
792 return xfs_setattr_nonsize(ip, iattr, 0);
793 }
794
795 /*
796 * Make sure that the dquots are attached to the inode.
797 */
798 error = xfs_qm_dqattach_locked(ip, 0);
799 if (error)
800 goto out_unlock;
801
802 /*
803 * Now we can make the changes. Before we join the inode to the
804 * transaction, take care of the part of the truncation that must be
805 * done without the inode lock. This needs to be done before joining
806 * the inode to the transaction, because the inode cannot be unlocked
807 * once it is a part of the transaction.
808 */
809 if (iattr->ia_size > ip->i_size) {
810 /*
811 * Do the first part of growing a file: zero any data in the
812 * last block that is beyond the old EOF. We need to do this
813 * before the inode is joined to the transaction to modify
814 * i_size.
815 */
816 error = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
817 if (error)
818 goto out_unlock;
819 }
820 xfs_iunlock(ip, XFS_ILOCK_EXCL);
821 lock_flags &= ~XFS_ILOCK_EXCL;
822
823 /*
824 * We are going to log the inode size change in this transaction so
825 * any previous writes that are beyond the on disk EOF and the new
826 * EOF that have not been written out need to be written here. If we
827 * do not write the data out, we expose ourselves to the null files
828 * problem.
829 *
830 * Only flush from the on disk size to the smaller of the in memory
831 * file size or the new size as that's the range we really care about
832 * here and prevents waiting for other data not within the range we
833 * care about here.
834 */
835 if (ip->i_size != ip->i_d.di_size && iattr->ia_size > ip->i_d.di_size) {
836 error = xfs_flush_pages(ip, ip->i_d.di_size, iattr->ia_size,
837 XBF_ASYNC, FI_NONE);
838 if (error)
839 goto out_unlock;
840 }
841
842 /*
843 * Wait for all I/O to complete.
844 */
845 xfs_ioend_wait(ip);
846
847 error = -block_truncate_page(inode->i_mapping, iattr->ia_size,
848 xfs_get_blocks);
849 if (error)
850 goto out_unlock;
851
852 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
853 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
854 XFS_TRANS_PERM_LOG_RES,
855 XFS_ITRUNCATE_LOG_COUNT);
856 if (error)
857 goto out_trans_cancel;
858
859 truncate_setsize(inode, iattr->ia_size);
860
861 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
862 lock_flags |= XFS_ILOCK_EXCL;
863
864 xfs_ilock(ip, XFS_ILOCK_EXCL);
865
866 xfs_trans_ijoin(tp, ip);
867
868 /*
869 * Only change the c/mtime if we are changing the size or we are
870 * explicitly asked to change it. This handles the semantic difference
871 * between truncate() and ftruncate() as implemented in the VFS.
872 *
873 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
874 * special case where we need to update the times despite not having
875 * these flags set. For all other operations the VFS set these flags
876 * explicitly if it wants a timestamp update.
877 */
878 if (iattr->ia_size != ip->i_size &&
879 (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
880 iattr->ia_ctime = iattr->ia_mtime =
881 current_fs_time(inode->i_sb);
882 mask |= ATTR_CTIME | ATTR_MTIME;
883 }
884
885 if (iattr->ia_size > ip->i_size) {
886 ip->i_d.di_size = iattr->ia_size;
887 ip->i_size = iattr->ia_size;
888 } else if (iattr->ia_size <= ip->i_size ||
889 (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
890 error = xfs_itruncate_data(&tp, ip, iattr->ia_size);
891 if (error)
892 goto out_trans_abort;
893
894 /*
895 * Truncated "down", so we're removing references to old data
896 * here - if we delay flushing for a long time, we expose
897 * ourselves unduly to the notorious NULL files problem. So,
898 * we mark this inode and flush it when the file is closed,
899 * and do not wait the usual (long) time for writeout.
900 */
901 xfs_iflags_set(ip, XFS_ITRUNCATED);
902 }
903
904 if (mask & ATTR_CTIME) {
905 inode->i_ctime = iattr->ia_ctime;
906 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
907 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
908 ip->i_update_core = 1;
909 }
910 if (mask & ATTR_MTIME) {
911 inode->i_mtime = iattr->ia_mtime;
912 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
913 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
914 ip->i_update_core = 1;
915 }
916
917 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
918
919 XFS_STATS_INC(xs_ig_attrchg);
920
921 if (mp->m_flags & XFS_MOUNT_WSYNC)
922 xfs_trans_set_sync(tp);
923
924 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
925out_unlock:
926 if (lock_flags)
927 xfs_iunlock(ip, lock_flags);
928 return error;
929
930out_trans_abort:
931 commit_flags |= XFS_TRANS_ABORT;
932out_trans_cancel:
933 xfs_trans_cancel(tp, commit_flags);
934 goto out_unlock;
935}
936
937STATIC int
938xfs_vn_setattr(
939 struct dentry *dentry,
940 struct iattr *iattr)
941{
942 if (iattr->ia_valid & ATTR_SIZE)
943 return -xfs_setattr_size(XFS_I(dentry->d_inode), iattr, 0);
944 return -xfs_setattr_nonsize(XFS_I(dentry->d_inode), iattr, 0);
945}
946
947#define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
948
949/*
950 * Call fiemap helper to fill in user data.
951 * Returns positive errors to xfs_getbmap.
952 */
953STATIC int
954xfs_fiemap_format(
955 void **arg,
956 struct getbmapx *bmv,
957 int *full)
958{
959 int error;
960 struct fiemap_extent_info *fieinfo = *arg;
961 u32 fiemap_flags = 0;
962 u64 logical, physical, length;
963
964 /* Do nothing for a hole */
965 if (bmv->bmv_block == -1LL)
966 return 0;
967
968 logical = BBTOB(bmv->bmv_offset);
969 physical = BBTOB(bmv->bmv_block);
970 length = BBTOB(bmv->bmv_length);
971
972 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
973 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
974 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
975 fiemap_flags |= FIEMAP_EXTENT_DELALLOC;
976 physical = 0; /* no block yet */
977 }
978 if (bmv->bmv_oflags & BMV_OF_LAST)
979 fiemap_flags |= FIEMAP_EXTENT_LAST;
980
981 error = fiemap_fill_next_extent(fieinfo, logical, physical,
982 length, fiemap_flags);
983 if (error > 0) {
984 error = 0;
985 *full = 1; /* user array now full */
986 }
987
988 return -error;
989}
990
991STATIC int
992xfs_vn_fiemap(
993 struct inode *inode,
994 struct fiemap_extent_info *fieinfo,
995 u64 start,
996 u64 length)
997{
998 xfs_inode_t *ip = XFS_I(inode);
999 struct getbmapx bm;
1000 int error;
1001
1002 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1003 if (error)
1004 return error;
1005
1006 /* Set up bmap header for xfs internal routine */
1007 bm.bmv_offset = BTOBB(start);
1008 /* Special case for whole file */
1009 if (length == FIEMAP_MAX_OFFSET)
1010 bm.bmv_length = -1LL;
1011 else
1012 bm.bmv_length = BTOBB(length);
1013
1014 /* We add one because in getbmap world count includes the header */
1015 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1016 fieinfo->fi_extents_max + 1;
1017 bm.bmv_count = min_t(__s32, bm.bmv_count,
1018 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1019 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1020 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1021 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1022 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1023 bm.bmv_iflags |= BMV_IF_DELALLOC;
1024
1025 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1026 if (error)
1027 return -error;
1028
1029 return 0;
1030}
1031
1032static const struct inode_operations xfs_inode_operations = {
1033 .get_acl = xfs_get_acl,
1034 .getattr = xfs_vn_getattr,
1035 .setattr = xfs_vn_setattr,
1036 .setxattr = generic_setxattr,
1037 .getxattr = generic_getxattr,
1038 .removexattr = generic_removexattr,
1039 .listxattr = xfs_vn_listxattr,
1040 .fiemap = xfs_vn_fiemap,
1041};
1042
1043static const struct inode_operations xfs_dir_inode_operations = {
1044 .create = xfs_vn_create,
1045 .lookup = xfs_vn_lookup,
1046 .link = xfs_vn_link,
1047 .unlink = xfs_vn_unlink,
1048 .symlink = xfs_vn_symlink,
1049 .mkdir = xfs_vn_mkdir,
1050 /*
1051 * Yes, XFS uses the same method for rmdir and unlink.
1052 *
1053 * There are some subtile differences deeper in the code,
1054 * but we use S_ISDIR to check for those.
1055 */
1056 .rmdir = xfs_vn_unlink,
1057 .mknod = xfs_vn_mknod,
1058 .rename = xfs_vn_rename,
1059 .get_acl = xfs_get_acl,
1060 .getattr = xfs_vn_getattr,
1061 .setattr = xfs_vn_setattr,
1062 .setxattr = generic_setxattr,
1063 .getxattr = generic_getxattr,
1064 .removexattr = generic_removexattr,
1065 .listxattr = xfs_vn_listxattr,
1066};
1067
1068static const struct inode_operations xfs_dir_ci_inode_operations = {
1069 .create = xfs_vn_create,
1070 .lookup = xfs_vn_ci_lookup,
1071 .link = xfs_vn_link,
1072 .unlink = xfs_vn_unlink,
1073 .symlink = xfs_vn_symlink,
1074 .mkdir = xfs_vn_mkdir,
1075 /*
1076 * Yes, XFS uses the same method for rmdir and unlink.
1077 *
1078 * There are some subtile differences deeper in the code,
1079 * but we use S_ISDIR to check for those.
1080 */
1081 .rmdir = xfs_vn_unlink,
1082 .mknod = xfs_vn_mknod,
1083 .rename = xfs_vn_rename,
1084 .get_acl = xfs_get_acl,
1085 .getattr = xfs_vn_getattr,
1086 .setattr = xfs_vn_setattr,
1087 .setxattr = generic_setxattr,
1088 .getxattr = generic_getxattr,
1089 .removexattr = generic_removexattr,
1090 .listxattr = xfs_vn_listxattr,
1091};
1092
1093static const struct inode_operations xfs_symlink_inode_operations = {
1094 .readlink = generic_readlink,
1095 .follow_link = xfs_vn_follow_link,
1096 .put_link = xfs_vn_put_link,
1097 .get_acl = xfs_get_acl,
1098 .getattr = xfs_vn_getattr,
1099 .setattr = xfs_vn_setattr,
1100 .setxattr = generic_setxattr,
1101 .getxattr = generic_getxattr,
1102 .removexattr = generic_removexattr,
1103 .listxattr = xfs_vn_listxattr,
1104};
1105
1106STATIC void
1107xfs_diflags_to_iflags(
1108 struct inode *inode,
1109 struct xfs_inode *ip)
1110{
1111 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1112 inode->i_flags |= S_IMMUTABLE;
1113 else
1114 inode->i_flags &= ~S_IMMUTABLE;
1115 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1116 inode->i_flags |= S_APPEND;
1117 else
1118 inode->i_flags &= ~S_APPEND;
1119 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1120 inode->i_flags |= S_SYNC;
1121 else
1122 inode->i_flags &= ~S_SYNC;
1123 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1124 inode->i_flags |= S_NOATIME;
1125 else
1126 inode->i_flags &= ~S_NOATIME;
1127}
1128
1129/*
1130 * Initialize the Linux inode, set up the operation vectors and
1131 * unlock the inode.
1132 *
1133 * When reading existing inodes from disk this is called directly
1134 * from xfs_iget, when creating a new inode it is called from
1135 * xfs_ialloc after setting up the inode.
1136 *
1137 * We are always called with an uninitialised linux inode here.
1138 * We need to initialise the necessary fields and take a reference
1139 * on it.
1140 */
1141void
1142xfs_setup_inode(
1143 struct xfs_inode *ip)
1144{
1145 struct inode *inode = &ip->i_vnode;
1146
1147 inode->i_ino = ip->i_ino;
1148 inode->i_state = I_NEW;
1149
1150 inode_sb_list_add(inode);
1151 /* make the inode look hashed for the writeback code */
1152 hlist_add_fake(&inode->i_hash);
1153
1154 inode->i_mode = ip->i_d.di_mode;
1155 inode->i_nlink = ip->i_d.di_nlink;
1156 inode->i_uid = ip->i_d.di_uid;
1157 inode->i_gid = ip->i_d.di_gid;
1158
1159 switch (inode->i_mode & S_IFMT) {
1160 case S_IFBLK:
1161 case S_IFCHR:
1162 inode->i_rdev =
1163 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1164 sysv_minor(ip->i_df.if_u2.if_rdev));
1165 break;
1166 default:
1167 inode->i_rdev = 0;
1168 break;
1169 }
1170
1171 inode->i_generation = ip->i_d.di_gen;
1172 i_size_write(inode, ip->i_d.di_size);
1173 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
1174 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
1175 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
1176 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
1177 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
1178 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
1179 xfs_diflags_to_iflags(inode, ip);
1180
1181 switch (inode->i_mode & S_IFMT) {
1182 case S_IFREG:
1183 inode->i_op = &xfs_inode_operations;
1184 inode->i_fop = &xfs_file_operations;
1185 inode->i_mapping->a_ops = &xfs_address_space_operations;
1186 break;
1187 case S_IFDIR:
1188 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1189 inode->i_op = &xfs_dir_ci_inode_operations;
1190 else
1191 inode->i_op = &xfs_dir_inode_operations;
1192 inode->i_fop = &xfs_dir_file_operations;
1193 break;
1194 case S_IFLNK:
1195 inode->i_op = &xfs_symlink_inode_operations;
1196 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1197 inode->i_mapping->a_ops = &xfs_address_space_operations;
1198 break;
1199 default:
1200 inode->i_op = &xfs_inode_operations;
1201 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1202 break;
1203 }
1204
1205 /*
1206 * If there is no attribute fork no ACL can exist on this inode,
1207 * and it can't have any file capabilities attached to it either.
1208 */
1209 if (!XFS_IFORK_Q(ip)) {
1210 inode_has_no_xattr(inode);
1211 cache_no_acl(inode);
1212 }
1213
1214 xfs_iflags_clear(ip, XFS_INEW);
1215 barrier();
1216
1217 unlock_new_inode(inode);
1218}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_inode.h"
14#include "xfs_acl.h"
15#include "xfs_quota.h"
16#include "xfs_da_format.h"
17#include "xfs_da_btree.h"
18#include "xfs_attr.h"
19#include "xfs_trans.h"
20#include "xfs_trans_space.h"
21#include "xfs_bmap_btree.h"
22#include "xfs_trace.h"
23#include "xfs_icache.h"
24#include "xfs_symlink.h"
25#include "xfs_dir2.h"
26#include "xfs_iomap.h"
27#include "xfs_error.h"
28#include "xfs_ioctl.h"
29#include "xfs_xattr.h"
30#include "xfs_file.h"
31#include "xfs_bmap.h"
32
33#include <linux/posix_acl.h>
34#include <linux/security.h>
35#include <linux/iversion.h>
36#include <linux/fiemap.h>
37
38/*
39 * Directories have different lock order w.r.t. mmap_lock compared to regular
40 * files. This is due to readdir potentially triggering page faults on a user
41 * buffer inside filldir(), and this happens with the ilock on the directory
42 * held. For regular files, the lock order is the other way around - the
43 * mmap_lock is taken during the page fault, and then we lock the ilock to do
44 * block mapping. Hence we need a different class for the directory ilock so
45 * that lockdep can tell them apart. Directories in the metadata directory
46 * tree get a separate class so that lockdep reports will warn us if someone
47 * ever tries to lock regular directories after locking metadata directories.
48 */
49static struct lock_class_key xfs_nondir_ilock_class;
50static struct lock_class_key xfs_dir_ilock_class;
51
52static int
53xfs_initxattrs(
54 struct inode *inode,
55 const struct xattr *xattr_array,
56 void *fs_info)
57{
58 const struct xattr *xattr;
59 struct xfs_inode *ip = XFS_I(inode);
60 int error = 0;
61
62 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
63 struct xfs_da_args args = {
64 .dp = ip,
65 .attr_filter = XFS_ATTR_SECURE,
66 .name = xattr->name,
67 .namelen = strlen(xattr->name),
68 .value = xattr->value,
69 .valuelen = xattr->value_len,
70 };
71 error = xfs_attr_change(&args, XFS_ATTRUPDATE_UPSERT);
72 if (error < 0)
73 break;
74 }
75 return error;
76}
77
78/*
79 * Hook in SELinux. This is not quite correct yet, what we really need
80 * here (as we do for default ACLs) is a mechanism by which creation of
81 * these attrs can be journalled at inode creation time (along with the
82 * inode, of course, such that log replay can't cause these to be lost).
83 */
84int
85xfs_inode_init_security(
86 struct inode *inode,
87 struct inode *dir,
88 const struct qstr *qstr)
89{
90 return security_inode_init_security(inode, dir, qstr,
91 &xfs_initxattrs, NULL);
92}
93
94static void
95xfs_dentry_to_name(
96 struct xfs_name *namep,
97 struct dentry *dentry)
98{
99 namep->name = dentry->d_name.name;
100 namep->len = dentry->d_name.len;
101 namep->type = XFS_DIR3_FT_UNKNOWN;
102}
103
104static int
105xfs_dentry_mode_to_name(
106 struct xfs_name *namep,
107 struct dentry *dentry,
108 int mode)
109{
110 namep->name = dentry->d_name.name;
111 namep->len = dentry->d_name.len;
112 namep->type = xfs_mode_to_ftype(mode);
113
114 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
115 return -EFSCORRUPTED;
116
117 return 0;
118}
119
120STATIC void
121xfs_cleanup_inode(
122 struct inode *dir,
123 struct inode *inode,
124 struct dentry *dentry)
125{
126 struct xfs_name teardown;
127
128 /* Oh, the horror.
129 * If we can't add the ACL or we fail in
130 * xfs_inode_init_security we must back out.
131 * ENOSPC can hit here, among other things.
132 */
133 xfs_dentry_to_name(&teardown, dentry);
134
135 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
136}
137
138/*
139 * Check to see if we are likely to need an extended attribute to be added to
140 * the inode we are about to allocate. This allows the attribute fork to be
141 * created during the inode allocation, reducing the number of transactions we
142 * need to do in this fast path.
143 *
144 * The security checks are optimistic, but not guaranteed. The two LSMs that
145 * require xattrs to be added here (selinux and smack) are also the only two
146 * LSMs that add a sb->s_security structure to the superblock. Hence if security
147 * is enabled and sb->s_security is set, we have a pretty good idea that we are
148 * going to be asked to add a security xattr immediately after allocating the
149 * xfs inode and instantiating the VFS inode.
150 */
151static inline bool
152xfs_create_need_xattr(
153 struct inode *dir,
154 struct posix_acl *default_acl,
155 struct posix_acl *acl)
156{
157 if (acl)
158 return true;
159 if (default_acl)
160 return true;
161#if IS_ENABLED(CONFIG_SECURITY)
162 if (dir->i_sb->s_security)
163 return true;
164#endif
165 return false;
166}
167
168
169STATIC int
170xfs_generic_create(
171 struct mnt_idmap *idmap,
172 struct inode *dir,
173 struct dentry *dentry,
174 umode_t mode,
175 dev_t rdev,
176 struct file *tmpfile) /* unnamed file */
177{
178 struct xfs_icreate_args args = {
179 .idmap = idmap,
180 .pip = XFS_I(dir),
181 .rdev = rdev,
182 .mode = mode,
183 };
184 struct inode *inode;
185 struct xfs_inode *ip = NULL;
186 struct posix_acl *default_acl, *acl;
187 struct xfs_name name;
188 int error;
189
190 /*
191 * Irix uses Missed'em'V split, but doesn't want to see
192 * the upper 5 bits of (14bit) major.
193 */
194 if (S_ISCHR(args.mode) || S_ISBLK(args.mode)) {
195 if (unlikely(!sysv_valid_dev(args.rdev) ||
196 MAJOR(args.rdev) & ~0x1ff))
197 return -EINVAL;
198 } else {
199 args.rdev = 0;
200 }
201
202 error = posix_acl_create(dir, &args.mode, &default_acl, &acl);
203 if (error)
204 return error;
205
206 /* Verify mode is valid also for tmpfile case */
207 error = xfs_dentry_mode_to_name(&name, dentry, args.mode);
208 if (unlikely(error))
209 goto out_free_acl;
210
211 if (!tmpfile) {
212 if (xfs_create_need_xattr(dir, default_acl, acl))
213 args.flags |= XFS_ICREATE_INIT_XATTRS;
214
215 error = xfs_create(&args, &name, &ip);
216 } else {
217 args.flags |= XFS_ICREATE_TMPFILE;
218
219 /*
220 * If this temporary file will not be linkable, don't bother
221 * creating an attr fork to receive a parent pointer.
222 */
223 if (tmpfile->f_flags & O_EXCL)
224 args.flags |= XFS_ICREATE_UNLINKABLE;
225
226 error = xfs_create_tmpfile(&args, &ip);
227 }
228 if (unlikely(error))
229 goto out_free_acl;
230
231 inode = VFS_I(ip);
232
233 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
234 if (unlikely(error))
235 goto out_cleanup_inode;
236
237 if (default_acl) {
238 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
239 if (error)
240 goto out_cleanup_inode;
241 }
242 if (acl) {
243 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
244 if (error)
245 goto out_cleanup_inode;
246 }
247
248 xfs_setup_iops(ip);
249
250 if (tmpfile) {
251 /*
252 * The VFS requires that any inode fed to d_tmpfile must have
253 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
254 * However, we created the temp file with nlink == 0 because
255 * we're not allowed to put an inode with nlink > 0 on the
256 * unlinked list. Therefore we have to set nlink to 1 so that
257 * d_tmpfile can immediately set it back to zero.
258 */
259 set_nlink(inode, 1);
260 d_tmpfile(tmpfile, inode);
261 } else
262 d_instantiate(dentry, inode);
263
264 xfs_finish_inode_setup(ip);
265
266 out_free_acl:
267 posix_acl_release(default_acl);
268 posix_acl_release(acl);
269 return error;
270
271 out_cleanup_inode:
272 xfs_finish_inode_setup(ip);
273 if (!tmpfile)
274 xfs_cleanup_inode(dir, inode, dentry);
275 xfs_irele(ip);
276 goto out_free_acl;
277}
278
279STATIC int
280xfs_vn_mknod(
281 struct mnt_idmap *idmap,
282 struct inode *dir,
283 struct dentry *dentry,
284 umode_t mode,
285 dev_t rdev)
286{
287 return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL);
288}
289
290STATIC int
291xfs_vn_create(
292 struct mnt_idmap *idmap,
293 struct inode *dir,
294 struct dentry *dentry,
295 umode_t mode,
296 bool flags)
297{
298 return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL);
299}
300
301STATIC int
302xfs_vn_mkdir(
303 struct mnt_idmap *idmap,
304 struct inode *dir,
305 struct dentry *dentry,
306 umode_t mode)
307{
308 return xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL);
309}
310
311STATIC struct dentry *
312xfs_vn_lookup(
313 struct inode *dir,
314 struct dentry *dentry,
315 unsigned int flags)
316{
317 struct inode *inode;
318 struct xfs_inode *cip;
319 struct xfs_name name;
320 int error;
321
322 if (dentry->d_name.len >= MAXNAMELEN)
323 return ERR_PTR(-ENAMETOOLONG);
324
325 xfs_dentry_to_name(&name, dentry);
326 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
327 if (likely(!error))
328 inode = VFS_I(cip);
329 else if (likely(error == -ENOENT))
330 inode = NULL;
331 else
332 inode = ERR_PTR(error);
333 return d_splice_alias(inode, dentry);
334}
335
336STATIC struct dentry *
337xfs_vn_ci_lookup(
338 struct inode *dir,
339 struct dentry *dentry,
340 unsigned int flags)
341{
342 struct xfs_inode *ip;
343 struct xfs_name xname;
344 struct xfs_name ci_name;
345 struct qstr dname;
346 int error;
347
348 if (dentry->d_name.len >= MAXNAMELEN)
349 return ERR_PTR(-ENAMETOOLONG);
350
351 xfs_dentry_to_name(&xname, dentry);
352 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
353 if (unlikely(error)) {
354 if (unlikely(error != -ENOENT))
355 return ERR_PTR(error);
356 /*
357 * call d_add(dentry, NULL) here when d_drop_negative_children
358 * is called in xfs_vn_mknod (ie. allow negative dentries
359 * with CI filesystems).
360 */
361 return NULL;
362 }
363
364 /* if exact match, just splice and exit */
365 if (!ci_name.name)
366 return d_splice_alias(VFS_I(ip), dentry);
367
368 /* else case-insensitive match... */
369 dname.name = ci_name.name;
370 dname.len = ci_name.len;
371 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
372 kfree(ci_name.name);
373 return dentry;
374}
375
376STATIC int
377xfs_vn_link(
378 struct dentry *old_dentry,
379 struct inode *dir,
380 struct dentry *dentry)
381{
382 struct inode *inode = d_inode(old_dentry);
383 struct xfs_name name;
384 int error;
385
386 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
387 if (unlikely(error))
388 return error;
389
390 if (IS_PRIVATE(inode))
391 return -EPERM;
392
393 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
394 if (unlikely(error))
395 return error;
396
397 ihold(inode);
398 d_instantiate(dentry, inode);
399 return 0;
400}
401
402STATIC int
403xfs_vn_unlink(
404 struct inode *dir,
405 struct dentry *dentry)
406{
407 struct xfs_name name;
408 int error;
409
410 xfs_dentry_to_name(&name, dentry);
411
412 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
413 if (error)
414 return error;
415
416 /*
417 * With unlink, the VFS makes the dentry "negative": no inode,
418 * but still hashed. This is incompatible with case-insensitive
419 * mode, so invalidate (unhash) the dentry in CI-mode.
420 */
421 if (xfs_has_asciici(XFS_M(dir->i_sb)))
422 d_invalidate(dentry);
423 return 0;
424}
425
426STATIC int
427xfs_vn_symlink(
428 struct mnt_idmap *idmap,
429 struct inode *dir,
430 struct dentry *dentry,
431 const char *symname)
432{
433 struct inode *inode;
434 struct xfs_inode *cip = NULL;
435 struct xfs_name name;
436 int error;
437 umode_t mode;
438
439 mode = S_IFLNK |
440 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
441 error = xfs_dentry_mode_to_name(&name, dentry, mode);
442 if (unlikely(error))
443 goto out;
444
445 error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip);
446 if (unlikely(error))
447 goto out;
448
449 inode = VFS_I(cip);
450
451 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
452 if (unlikely(error))
453 goto out_cleanup_inode;
454
455 xfs_setup_iops(cip);
456
457 d_instantiate(dentry, inode);
458 xfs_finish_inode_setup(cip);
459 return 0;
460
461 out_cleanup_inode:
462 xfs_finish_inode_setup(cip);
463 xfs_cleanup_inode(dir, inode, dentry);
464 xfs_irele(cip);
465 out:
466 return error;
467}
468
469STATIC int
470xfs_vn_rename(
471 struct mnt_idmap *idmap,
472 struct inode *odir,
473 struct dentry *odentry,
474 struct inode *ndir,
475 struct dentry *ndentry,
476 unsigned int flags)
477{
478 struct inode *new_inode = d_inode(ndentry);
479 int omode = 0;
480 int error;
481 struct xfs_name oname;
482 struct xfs_name nname;
483
484 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
485 return -EINVAL;
486
487 /* if we are exchanging files, we need to set i_mode of both files */
488 if (flags & RENAME_EXCHANGE)
489 omode = d_inode(ndentry)->i_mode;
490
491 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
492 if (omode && unlikely(error))
493 return error;
494
495 error = xfs_dentry_mode_to_name(&nname, ndentry,
496 d_inode(odentry)->i_mode);
497 if (unlikely(error))
498 return error;
499
500 return xfs_rename(idmap, XFS_I(odir), &oname,
501 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
502 new_inode ? XFS_I(new_inode) : NULL, flags);
503}
504
505/*
506 * careful here - this function can get called recursively, so
507 * we need to be very careful about how much stack we use.
508 * uio is kmalloced for this reason...
509 */
510STATIC const char *
511xfs_vn_get_link(
512 struct dentry *dentry,
513 struct inode *inode,
514 struct delayed_call *done)
515{
516 char *link;
517 int error = -ENOMEM;
518
519 if (!dentry)
520 return ERR_PTR(-ECHILD);
521
522 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
523 if (!link)
524 goto out_err;
525
526 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
527 if (unlikely(error))
528 goto out_kfree;
529
530 set_delayed_call(done, kfree_link, link);
531 return link;
532
533 out_kfree:
534 kfree(link);
535 out_err:
536 return ERR_PTR(error);
537}
538
539static uint32_t
540xfs_stat_blksize(
541 struct xfs_inode *ip)
542{
543 struct xfs_mount *mp = ip->i_mount;
544
545 /*
546 * If the file blocks are being allocated from a realtime volume, then
547 * always return the realtime extent size.
548 */
549 if (XFS_IS_REALTIME_INODE(ip))
550 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip) ? : 1);
551
552 /*
553 * Allow large block sizes to be reported to userspace programs if the
554 * "largeio" mount option is used.
555 *
556 * If compatibility mode is specified, simply return the basic unit of
557 * caching so that we don't get inefficient read/modify/write I/O from
558 * user apps. Otherwise....
559 *
560 * If the underlying volume is a stripe, then return the stripe width in
561 * bytes as the recommended I/O size. It is not a stripe and we've set a
562 * default buffered I/O size, return that, otherwise return the compat
563 * default.
564 */
565 if (xfs_has_large_iosize(mp)) {
566 if (mp->m_swidth)
567 return XFS_FSB_TO_B(mp, mp->m_swidth);
568 if (xfs_has_allocsize(mp))
569 return 1U << mp->m_allocsize_log;
570 }
571
572 return max_t(uint32_t, PAGE_SIZE, mp->m_sb.sb_blocksize);
573}
574
575static void
576xfs_get_atomic_write_attr(
577 struct xfs_inode *ip,
578 unsigned int *unit_min,
579 unsigned int *unit_max)
580{
581 if (!xfs_inode_can_atomicwrite(ip)) {
582 *unit_min = *unit_max = 0;
583 return;
584 }
585
586 *unit_min = *unit_max = ip->i_mount->m_sb.sb_blocksize;
587}
588
589STATIC int
590xfs_vn_getattr(
591 struct mnt_idmap *idmap,
592 const struct path *path,
593 struct kstat *stat,
594 u32 request_mask,
595 unsigned int query_flags)
596{
597 struct inode *inode = d_inode(path->dentry);
598 struct xfs_inode *ip = XFS_I(inode);
599 struct xfs_mount *mp = ip->i_mount;
600 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
601 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
602
603 trace_xfs_getattr(ip);
604
605 if (xfs_is_shutdown(mp))
606 return -EIO;
607
608 stat->size = XFS_ISIZE(ip);
609 stat->dev = inode->i_sb->s_dev;
610 stat->mode = inode->i_mode;
611 stat->nlink = inode->i_nlink;
612 stat->uid = vfsuid_into_kuid(vfsuid);
613 stat->gid = vfsgid_into_kgid(vfsgid);
614 stat->ino = ip->i_ino;
615 stat->atime = inode_get_atime(inode);
616
617 fill_mg_cmtime(stat, request_mask, inode);
618
619 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
620
621 if (xfs_has_v3inodes(mp)) {
622 if (request_mask & STATX_BTIME) {
623 stat->result_mask |= STATX_BTIME;
624 stat->btime = ip->i_crtime;
625 }
626 }
627
628 /*
629 * Note: If you add another clause to set an attribute flag, please
630 * update attributes_mask below.
631 */
632 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
633 stat->attributes |= STATX_ATTR_IMMUTABLE;
634 if (ip->i_diflags & XFS_DIFLAG_APPEND)
635 stat->attributes |= STATX_ATTR_APPEND;
636 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
637 stat->attributes |= STATX_ATTR_NODUMP;
638
639 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
640 STATX_ATTR_APPEND |
641 STATX_ATTR_NODUMP);
642
643 switch (inode->i_mode & S_IFMT) {
644 case S_IFBLK:
645 case S_IFCHR:
646 stat->blksize = BLKDEV_IOSIZE;
647 stat->rdev = inode->i_rdev;
648 break;
649 case S_IFREG:
650 if (request_mask & STATX_DIOALIGN) {
651 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
652 struct block_device *bdev = target->bt_bdev;
653
654 stat->result_mask |= STATX_DIOALIGN;
655 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
656 stat->dio_offset_align = bdev_logical_block_size(bdev);
657 }
658 if (request_mask & STATX_WRITE_ATOMIC) {
659 unsigned int unit_min, unit_max;
660
661 xfs_get_atomic_write_attr(ip, &unit_min,
662 &unit_max);
663 generic_fill_statx_atomic_writes(stat,
664 unit_min, unit_max);
665 }
666 fallthrough;
667 default:
668 stat->blksize = xfs_stat_blksize(ip);
669 stat->rdev = 0;
670 break;
671 }
672
673 return 0;
674}
675
676static int
677xfs_vn_change_ok(
678 struct mnt_idmap *idmap,
679 struct dentry *dentry,
680 struct iattr *iattr)
681{
682 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
683
684 if (xfs_is_readonly(mp))
685 return -EROFS;
686
687 if (xfs_is_shutdown(mp))
688 return -EIO;
689
690 return setattr_prepare(idmap, dentry, iattr);
691}
692
693/*
694 * Set non-size attributes of an inode.
695 *
696 * Caution: The caller of this function is responsible for calling
697 * setattr_prepare() or otherwise verifying the change is fine.
698 */
699static int
700xfs_setattr_nonsize(
701 struct mnt_idmap *idmap,
702 struct dentry *dentry,
703 struct xfs_inode *ip,
704 struct iattr *iattr)
705{
706 xfs_mount_t *mp = ip->i_mount;
707 struct inode *inode = VFS_I(ip);
708 int mask = iattr->ia_valid;
709 xfs_trans_t *tp;
710 int error;
711 kuid_t uid = GLOBAL_ROOT_UID;
712 kgid_t gid = GLOBAL_ROOT_GID;
713 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
714 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL;
715
716 ASSERT((mask & ATTR_SIZE) == 0);
717
718 /*
719 * If disk quotas is on, we make sure that the dquots do exist on disk,
720 * before we start any other transactions. Trying to do this later
721 * is messy. We don't care to take a readlock to look at the ids
722 * in inode here, because we can't hold it across the trans_reserve.
723 * If the IDs do change before we take the ilock, we're covered
724 * because the i_*dquot fields will get updated anyway.
725 */
726 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
727 uint qflags = 0;
728
729 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
730 uid = from_vfsuid(idmap, i_user_ns(inode),
731 iattr->ia_vfsuid);
732 qflags |= XFS_QMOPT_UQUOTA;
733 } else {
734 uid = inode->i_uid;
735 }
736 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
737 gid = from_vfsgid(idmap, i_user_ns(inode),
738 iattr->ia_vfsgid);
739 qflags |= XFS_QMOPT_GQUOTA;
740 } else {
741 gid = inode->i_gid;
742 }
743
744 /*
745 * We take a reference when we initialize udqp and gdqp,
746 * so it is important that we never blindly double trip on
747 * the same variable. See xfs_create() for an example.
748 */
749 ASSERT(udqp == NULL);
750 ASSERT(gdqp == NULL);
751 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
752 qflags, &udqp, &gdqp, NULL);
753 if (error)
754 return error;
755 }
756
757 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
758 has_capability_noaudit(current, CAP_FOWNER), &tp);
759 if (error)
760 goto out_dqrele;
761
762 /*
763 * Register quota modifications in the transaction. Must be the owner
764 * or privileged. These IDs could have changed since we last looked at
765 * them. But, we're assured that if the ownership did change while we
766 * didn't have the inode locked, inode's dquot(s) would have changed
767 * also.
768 */
769 if (XFS_IS_UQUOTA_ON(mp) &&
770 i_uid_needs_update(idmap, iattr, inode)) {
771 ASSERT(udqp);
772 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp);
773 }
774 if (XFS_IS_GQUOTA_ON(mp) &&
775 i_gid_needs_update(idmap, iattr, inode)) {
776 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp));
777 ASSERT(gdqp);
778 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp);
779 }
780
781 setattr_copy(idmap, inode, iattr);
782 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
783
784 XFS_STATS_INC(mp, xs_ig_attrchg);
785
786 if (xfs_has_wsync(mp))
787 xfs_trans_set_sync(tp);
788 error = xfs_trans_commit(tp);
789
790 /*
791 * Release any dquot(s) the inode had kept before chown.
792 */
793 xfs_qm_dqrele(old_udqp);
794 xfs_qm_dqrele(old_gdqp);
795 xfs_qm_dqrele(udqp);
796 xfs_qm_dqrele(gdqp);
797
798 if (error)
799 return error;
800
801 /*
802 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
803 * update. We could avoid this with linked transactions
804 * and passing down the transaction pointer all the way
805 * to attr_set. No previous user of the generic
806 * Posix ACL code seems to care about this issue either.
807 */
808 if (mask & ATTR_MODE) {
809 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
810 if (error)
811 return error;
812 }
813
814 return 0;
815
816out_dqrele:
817 xfs_qm_dqrele(udqp);
818 xfs_qm_dqrele(gdqp);
819 return error;
820}
821
822/*
823 * Truncate file. Must have write permission and not be a directory.
824 *
825 * Caution: The caller of this function is responsible for calling
826 * setattr_prepare() or otherwise verifying the change is fine.
827 */
828STATIC int
829xfs_setattr_size(
830 struct mnt_idmap *idmap,
831 struct dentry *dentry,
832 struct xfs_inode *ip,
833 struct iattr *iattr)
834{
835 struct xfs_mount *mp = ip->i_mount;
836 struct inode *inode = VFS_I(ip);
837 xfs_off_t oldsize, newsize;
838 struct xfs_trans *tp;
839 int error;
840 uint lock_flags = 0;
841 uint resblks = 0;
842 bool did_zeroing = false;
843
844 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
845 ASSERT(S_ISREG(inode->i_mode));
846 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
847 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
848
849 oldsize = inode->i_size;
850 newsize = iattr->ia_size;
851
852 /*
853 * Short circuit the truncate case for zero length files.
854 */
855 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
856 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
857 return 0;
858
859 /*
860 * Use the regular setattr path to update the timestamps.
861 */
862 iattr->ia_valid &= ~ATTR_SIZE;
863 return xfs_setattr_nonsize(idmap, dentry, ip, iattr);
864 }
865
866 /*
867 * Make sure that the dquots are attached to the inode.
868 */
869 error = xfs_qm_dqattach(ip);
870 if (error)
871 return error;
872
873 /*
874 * Wait for all direct I/O to complete.
875 */
876 inode_dio_wait(inode);
877
878 /*
879 * File data changes must be complete before we start the transaction to
880 * modify the inode. This needs to be done before joining the inode to
881 * the transaction because the inode cannot be unlocked once it is a
882 * part of the transaction.
883 *
884 * Start with zeroing any data beyond EOF that we may expose on file
885 * extension, or zeroing out the rest of the block on a downward
886 * truncate.
887 */
888 if (newsize > oldsize) {
889 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
890 error = xfs_zero_range(ip, oldsize, newsize - oldsize,
891 &did_zeroing);
892 } else {
893 error = xfs_truncate_page(ip, newsize, &did_zeroing);
894 }
895
896 if (error)
897 return error;
898
899 /*
900 * We've already locked out new page faults, so now we can safely remove
901 * pages from the page cache knowing they won't get refaulted until we
902 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
903 * complete. The truncate_setsize() call also cleans partial EOF page
904 * PTEs on extending truncates and hence ensures sub-page block size
905 * filesystems are correctly handled, too.
906 *
907 * We have to do all the page cache truncate work outside the
908 * transaction context as the "lock" order is page lock->log space
909 * reservation as defined by extent allocation in the writeback path.
910 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
911 * having already truncated the in-memory version of the file (i.e. made
912 * user visible changes). There's not much we can do about this, except
913 * to hope that the caller sees ENOMEM and retries the truncate
914 * operation.
915 *
916 * And we update in-core i_size and truncate page cache beyond newsize
917 * before writeback the [i_disk_size, newsize] range, so we're
918 * guaranteed not to write stale data past the new EOF on truncate down.
919 */
920 truncate_setsize(inode, newsize);
921
922 /*
923 * We are going to log the inode size change in this transaction so
924 * any previous writes that are beyond the on disk EOF and the new
925 * EOF that have not been written out need to be written here. If we
926 * do not write the data out, we expose ourselves to the null files
927 * problem. Note that this includes any block zeroing we did above;
928 * otherwise those blocks may not be zeroed after a crash.
929 */
930 if (did_zeroing ||
931 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
932 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
933 ip->i_disk_size, newsize - 1);
934 if (error)
935 return error;
936 }
937
938 /*
939 * For realtime inode with more than one block rtextsize, we need the
940 * block reservation for bmap btree block allocations/splits that can
941 * happen since it could split the tail written extent and convert the
942 * right beyond EOF one to unwritten.
943 */
944 if (xfs_inode_has_bigrtalloc(ip))
945 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
946
947 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks,
948 0, 0, &tp);
949 if (error)
950 return error;
951
952 lock_flags |= XFS_ILOCK_EXCL;
953 xfs_ilock(ip, XFS_ILOCK_EXCL);
954 xfs_trans_ijoin(tp, ip, 0);
955
956 /*
957 * Only change the c/mtime if we are changing the size or we are
958 * explicitly asked to change it. This handles the semantic difference
959 * between truncate() and ftruncate() as implemented in the VFS.
960 *
961 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
962 * special case where we need to update the times despite not having
963 * these flags set. For all other operations the VFS set these flags
964 * explicitly if it wants a timestamp update.
965 */
966 if (newsize != oldsize &&
967 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
968 iattr->ia_ctime = iattr->ia_mtime =
969 current_time(inode);
970 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
971 }
972
973 /*
974 * The first thing we do is set the size to new_size permanently on
975 * disk. This way we don't have to worry about anyone ever being able
976 * to look at the data being freed even in the face of a crash.
977 * What we're getting around here is the case where we free a block, it
978 * is allocated to another file, it is written to, and then we crash.
979 * If the new data gets written to the file but the log buffers
980 * containing the free and reallocation don't, then we'd end up with
981 * garbage in the blocks being freed. As long as we make the new size
982 * permanent before actually freeing any blocks it doesn't matter if
983 * they get written to.
984 */
985 ip->i_disk_size = newsize;
986 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
987
988 if (newsize <= oldsize) {
989 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
990 if (error)
991 goto out_trans_cancel;
992
993 /*
994 * Truncated "down", so we're removing references to old data
995 * here - if we delay flushing for a long time, we expose
996 * ourselves unduly to the notorious NULL files problem. So,
997 * we mark this inode and flush it when the file is closed,
998 * and do not wait the usual (long) time for writeout.
999 */
1000 xfs_iflags_set(ip, XFS_ITRUNCATED);
1001
1002 /* A truncate down always removes post-EOF blocks. */
1003 xfs_inode_clear_eofblocks_tag(ip);
1004 }
1005
1006 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
1007 setattr_copy(idmap, inode, iattr);
1008 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1009
1010 XFS_STATS_INC(mp, xs_ig_attrchg);
1011
1012 if (xfs_has_wsync(mp))
1013 xfs_trans_set_sync(tp);
1014
1015 error = xfs_trans_commit(tp);
1016out_unlock:
1017 if (lock_flags)
1018 xfs_iunlock(ip, lock_flags);
1019 return error;
1020
1021out_trans_cancel:
1022 xfs_trans_cancel(tp);
1023 goto out_unlock;
1024}
1025
1026int
1027xfs_vn_setattr_size(
1028 struct mnt_idmap *idmap,
1029 struct dentry *dentry,
1030 struct iattr *iattr)
1031{
1032 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1033 int error;
1034
1035 trace_xfs_setattr(ip);
1036
1037 error = xfs_vn_change_ok(idmap, dentry, iattr);
1038 if (error)
1039 return error;
1040 return xfs_setattr_size(idmap, dentry, ip, iattr);
1041}
1042
1043STATIC int
1044xfs_vn_setattr(
1045 struct mnt_idmap *idmap,
1046 struct dentry *dentry,
1047 struct iattr *iattr)
1048{
1049 struct inode *inode = d_inode(dentry);
1050 struct xfs_inode *ip = XFS_I(inode);
1051 int error;
1052
1053 if (iattr->ia_valid & ATTR_SIZE) {
1054 uint iolock;
1055
1056 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1057 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1058
1059 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1060 if (error) {
1061 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1062 return error;
1063 }
1064
1065 error = xfs_vn_setattr_size(idmap, dentry, iattr);
1066 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1067 } else {
1068 trace_xfs_setattr(ip);
1069
1070 error = xfs_vn_change_ok(idmap, dentry, iattr);
1071 if (!error)
1072 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr);
1073 }
1074
1075 return error;
1076}
1077
1078STATIC int
1079xfs_vn_update_time(
1080 struct inode *inode,
1081 int flags)
1082{
1083 struct xfs_inode *ip = XFS_I(inode);
1084 struct xfs_mount *mp = ip->i_mount;
1085 int log_flags = XFS_ILOG_TIMESTAMP;
1086 struct xfs_trans *tp;
1087 int error;
1088 struct timespec64 now;
1089
1090 trace_xfs_update_time(ip);
1091
1092 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1093 if (!((flags & S_VERSION) &&
1094 inode_maybe_inc_iversion(inode, false))) {
1095 generic_update_time(inode, flags);
1096 return 0;
1097 }
1098
1099 /* Capture the iversion update that just occurred */
1100 log_flags |= XFS_ILOG_CORE;
1101 }
1102
1103 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1104 if (error)
1105 return error;
1106
1107 xfs_ilock(ip, XFS_ILOCK_EXCL);
1108 if (flags & (S_CTIME|S_MTIME))
1109 now = inode_set_ctime_current(inode);
1110 else
1111 now = current_time(inode);
1112
1113 if (flags & S_MTIME)
1114 inode_set_mtime_to_ts(inode, now);
1115 if (flags & S_ATIME)
1116 inode_set_atime_to_ts(inode, now);
1117
1118 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1119 xfs_trans_log_inode(tp, ip, log_flags);
1120 return xfs_trans_commit(tp);
1121}
1122
1123STATIC int
1124xfs_vn_fiemap(
1125 struct inode *inode,
1126 struct fiemap_extent_info *fieinfo,
1127 u64 start,
1128 u64 length)
1129{
1130 int error;
1131
1132 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1133 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1134 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1135 error = iomap_fiemap(inode, fieinfo, start, length,
1136 &xfs_xattr_iomap_ops);
1137 } else {
1138 error = iomap_fiemap(inode, fieinfo, start, length,
1139 &xfs_read_iomap_ops);
1140 }
1141 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1142
1143 return error;
1144}
1145
1146STATIC int
1147xfs_vn_tmpfile(
1148 struct mnt_idmap *idmap,
1149 struct inode *dir,
1150 struct file *file,
1151 umode_t mode)
1152{
1153 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file);
1154
1155 return finish_open_simple(file, err);
1156}
1157
1158static const struct inode_operations xfs_inode_operations = {
1159 .get_inode_acl = xfs_get_acl,
1160 .set_acl = xfs_set_acl,
1161 .getattr = xfs_vn_getattr,
1162 .setattr = xfs_vn_setattr,
1163 .listxattr = xfs_vn_listxattr,
1164 .fiemap = xfs_vn_fiemap,
1165 .update_time = xfs_vn_update_time,
1166 .fileattr_get = xfs_fileattr_get,
1167 .fileattr_set = xfs_fileattr_set,
1168};
1169
1170static const struct inode_operations xfs_dir_inode_operations = {
1171 .create = xfs_vn_create,
1172 .lookup = xfs_vn_lookup,
1173 .link = xfs_vn_link,
1174 .unlink = xfs_vn_unlink,
1175 .symlink = xfs_vn_symlink,
1176 .mkdir = xfs_vn_mkdir,
1177 /*
1178 * Yes, XFS uses the same method for rmdir and unlink.
1179 *
1180 * There are some subtile differences deeper in the code,
1181 * but we use S_ISDIR to check for those.
1182 */
1183 .rmdir = xfs_vn_unlink,
1184 .mknod = xfs_vn_mknod,
1185 .rename = xfs_vn_rename,
1186 .get_inode_acl = xfs_get_acl,
1187 .set_acl = xfs_set_acl,
1188 .getattr = xfs_vn_getattr,
1189 .setattr = xfs_vn_setattr,
1190 .listxattr = xfs_vn_listxattr,
1191 .update_time = xfs_vn_update_time,
1192 .tmpfile = xfs_vn_tmpfile,
1193 .fileattr_get = xfs_fileattr_get,
1194 .fileattr_set = xfs_fileattr_set,
1195};
1196
1197static const struct inode_operations xfs_dir_ci_inode_operations = {
1198 .create = xfs_vn_create,
1199 .lookup = xfs_vn_ci_lookup,
1200 .link = xfs_vn_link,
1201 .unlink = xfs_vn_unlink,
1202 .symlink = xfs_vn_symlink,
1203 .mkdir = xfs_vn_mkdir,
1204 /*
1205 * Yes, XFS uses the same method for rmdir and unlink.
1206 *
1207 * There are some subtile differences deeper in the code,
1208 * but we use S_ISDIR to check for those.
1209 */
1210 .rmdir = xfs_vn_unlink,
1211 .mknod = xfs_vn_mknod,
1212 .rename = xfs_vn_rename,
1213 .get_inode_acl = xfs_get_acl,
1214 .set_acl = xfs_set_acl,
1215 .getattr = xfs_vn_getattr,
1216 .setattr = xfs_vn_setattr,
1217 .listxattr = xfs_vn_listxattr,
1218 .update_time = xfs_vn_update_time,
1219 .tmpfile = xfs_vn_tmpfile,
1220 .fileattr_get = xfs_fileattr_get,
1221 .fileattr_set = xfs_fileattr_set,
1222};
1223
1224static const struct inode_operations xfs_symlink_inode_operations = {
1225 .get_link = xfs_vn_get_link,
1226 .getattr = xfs_vn_getattr,
1227 .setattr = xfs_vn_setattr,
1228 .listxattr = xfs_vn_listxattr,
1229 .update_time = xfs_vn_update_time,
1230};
1231
1232/* Figure out if this file actually supports DAX. */
1233static bool
1234xfs_inode_supports_dax(
1235 struct xfs_inode *ip)
1236{
1237 struct xfs_mount *mp = ip->i_mount;
1238
1239 /* Only supported on regular files. */
1240 if (!S_ISREG(VFS_I(ip)->i_mode))
1241 return false;
1242
1243 /* Block size must match page size */
1244 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1245 return false;
1246
1247 /* Device has to support DAX too. */
1248 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1249}
1250
1251static bool
1252xfs_inode_should_enable_dax(
1253 struct xfs_inode *ip)
1254{
1255 if (!IS_ENABLED(CONFIG_FS_DAX))
1256 return false;
1257 if (xfs_has_dax_never(ip->i_mount))
1258 return false;
1259 if (!xfs_inode_supports_dax(ip))
1260 return false;
1261 if (xfs_has_dax_always(ip->i_mount))
1262 return true;
1263 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1264 return true;
1265 return false;
1266}
1267
1268void
1269xfs_diflags_to_iflags(
1270 struct xfs_inode *ip,
1271 bool init)
1272{
1273 struct inode *inode = VFS_I(ip);
1274 unsigned int xflags = xfs_ip2xflags(ip);
1275 unsigned int flags = 0;
1276
1277 ASSERT(!(IS_DAX(inode) && init));
1278
1279 if (xflags & FS_XFLAG_IMMUTABLE)
1280 flags |= S_IMMUTABLE;
1281 if (xflags & FS_XFLAG_APPEND)
1282 flags |= S_APPEND;
1283 if (xflags & FS_XFLAG_SYNC)
1284 flags |= S_SYNC;
1285 if (xflags & FS_XFLAG_NOATIME)
1286 flags |= S_NOATIME;
1287 if (init && xfs_inode_should_enable_dax(ip))
1288 flags |= S_DAX;
1289
1290 /*
1291 * S_DAX can only be set during inode initialization and is never set by
1292 * the VFS, so we cannot mask off S_DAX in i_flags.
1293 */
1294 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1295 inode->i_flags |= flags;
1296}
1297
1298/*
1299 * Initialize the Linux inode.
1300 *
1301 * When reading existing inodes from disk this is called directly from xfs_iget,
1302 * when creating a new inode it is called from xfs_init_new_inode after setting
1303 * up the inode. These callers have different criteria for clearing XFS_INEW, so
1304 * leave it up to the caller to deal with unlocking the inode appropriately.
1305 */
1306void
1307xfs_setup_inode(
1308 struct xfs_inode *ip)
1309{
1310 struct inode *inode = &ip->i_vnode;
1311 gfp_t gfp_mask;
1312 bool is_meta = xfs_is_internal_inode(ip);
1313
1314 inode->i_ino = ip->i_ino;
1315 inode->i_state |= I_NEW;
1316
1317 inode_sb_list_add(inode);
1318 /* make the inode look hashed for the writeback code */
1319 inode_fake_hash(inode);
1320
1321 i_size_write(inode, ip->i_disk_size);
1322 xfs_diflags_to_iflags(ip, true);
1323
1324 /*
1325 * Mark our metadata files as private so that LSMs and the ACL code
1326 * don't try to add their own metadata or reason about these files,
1327 * and users cannot ever obtain file handles to them.
1328 */
1329 if (is_meta) {
1330 inode->i_flags |= S_PRIVATE;
1331 inode->i_opflags &= ~IOP_XATTR;
1332 }
1333
1334 if (S_ISDIR(inode->i_mode)) {
1335 /*
1336 * We set the i_rwsem class here to avoid potential races with
1337 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1338 * after a filehandle lookup has already found the inode in
1339 * cache before it has been unlocked via unlock_new_inode().
1340 */
1341 lockdep_set_class(&inode->i_rwsem,
1342 &inode->i_sb->s_type->i_mutex_dir_key);
1343 lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class);
1344 } else {
1345 lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class);
1346 }
1347
1348 /*
1349 * Ensure all page cache allocations are done from GFP_NOFS context to
1350 * prevent direct reclaim recursion back into the filesystem and blowing
1351 * stacks or deadlocking.
1352 */
1353 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1354 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1355
1356 /*
1357 * For real-time inodes update the stable write flags to that of the RT
1358 * device instead of the data device.
1359 */
1360 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip))
1361 xfs_update_stable_writes(ip);
1362
1363 /*
1364 * If there is no attribute fork no ACL can exist on this inode,
1365 * and it can't have any file capabilities attached to it either.
1366 */
1367 if (!xfs_inode_has_attr_fork(ip)) {
1368 inode_has_no_xattr(inode);
1369 cache_no_acl(inode);
1370 }
1371}
1372
1373void
1374xfs_setup_iops(
1375 struct xfs_inode *ip)
1376{
1377 struct inode *inode = &ip->i_vnode;
1378
1379 switch (inode->i_mode & S_IFMT) {
1380 case S_IFREG:
1381 inode->i_op = &xfs_inode_operations;
1382 inode->i_fop = &xfs_file_operations;
1383 if (IS_DAX(inode))
1384 inode->i_mapping->a_ops = &xfs_dax_aops;
1385 else
1386 inode->i_mapping->a_ops = &xfs_address_space_operations;
1387 break;
1388 case S_IFDIR:
1389 if (xfs_has_asciici(XFS_M(inode->i_sb)))
1390 inode->i_op = &xfs_dir_ci_inode_operations;
1391 else
1392 inode->i_op = &xfs_dir_inode_operations;
1393 inode->i_fop = &xfs_dir_file_operations;
1394 break;
1395 case S_IFLNK:
1396 inode->i_op = &xfs_symlink_inode_operations;
1397 break;
1398 default:
1399 inode->i_op = &xfs_inode_operations;
1400 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1401 break;
1402 }
1403}