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_attr.h"
17#include "xfs_trans.h"
18#include "xfs_trace.h"
19#include "xfs_icache.h"
20#include "xfs_symlink.h"
21#include "xfs_dir2.h"
22#include "xfs_iomap.h"
23#include "xfs_error.h"
24
25#include <linux/posix_acl.h>
26#include <linux/security.h>
27#include <linux/iversion.h>
28#include <linux/fiemap.h>
29
30/*
31 * Directories have different lock order w.r.t. mmap_lock compared to regular
32 * files. This is due to readdir potentially triggering page faults on a user
33 * buffer inside filldir(), and this happens with the ilock on the directory
34 * held. For regular files, the lock order is the other way around - the
35 * mmap_lock is taken during the page fault, and then we lock the ilock to do
36 * block mapping. Hence we need a different class for the directory ilock so
37 * that lockdep can tell them apart.
38 */
39static struct lock_class_key xfs_nondir_ilock_class;
40static struct lock_class_key xfs_dir_ilock_class;
41
42static int
43xfs_initxattrs(
44 struct inode *inode,
45 const struct xattr *xattr_array,
46 void *fs_info)
47{
48 const struct xattr *xattr;
49 struct xfs_inode *ip = XFS_I(inode);
50 int error = 0;
51
52 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
53 struct xfs_da_args args = {
54 .dp = ip,
55 .attr_filter = XFS_ATTR_SECURE,
56 .name = xattr->name,
57 .namelen = strlen(xattr->name),
58 .value = xattr->value,
59 .valuelen = xattr->value_len,
60 };
61 error = xfs_attr_set(&args);
62 if (error < 0)
63 break;
64 }
65 return error;
66}
67
68/*
69 * Hook in SELinux. This is not quite correct yet, what we really need
70 * here (as we do for default ACLs) is a mechanism by which creation of
71 * these attrs can be journalled at inode creation time (along with the
72 * inode, of course, such that log replay can't cause these to be lost).
73 */
74
75STATIC int
76xfs_init_security(
77 struct inode *inode,
78 struct inode *dir,
79 const struct qstr *qstr)
80{
81 return security_inode_init_security(inode, dir, qstr,
82 &xfs_initxattrs, NULL);
83}
84
85static void
86xfs_dentry_to_name(
87 struct xfs_name *namep,
88 struct dentry *dentry)
89{
90 namep->name = dentry->d_name.name;
91 namep->len = dentry->d_name.len;
92 namep->type = XFS_DIR3_FT_UNKNOWN;
93}
94
95static int
96xfs_dentry_mode_to_name(
97 struct xfs_name *namep,
98 struct dentry *dentry,
99 int mode)
100{
101 namep->name = dentry->d_name.name;
102 namep->len = dentry->d_name.len;
103 namep->type = xfs_mode_to_ftype(mode);
104
105 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
106 return -EFSCORRUPTED;
107
108 return 0;
109}
110
111STATIC void
112xfs_cleanup_inode(
113 struct inode *dir,
114 struct inode *inode,
115 struct dentry *dentry)
116{
117 struct xfs_name teardown;
118
119 /* Oh, the horror.
120 * If we can't add the ACL or we fail in
121 * xfs_init_security we must back out.
122 * ENOSPC can hit here, among other things.
123 */
124 xfs_dentry_to_name(&teardown, dentry);
125
126 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
127}
128
129STATIC int
130xfs_generic_create(
131 struct inode *dir,
132 struct dentry *dentry,
133 umode_t mode,
134 dev_t rdev,
135 bool tmpfile) /* unnamed file */
136{
137 struct inode *inode;
138 struct xfs_inode *ip = NULL;
139 struct posix_acl *default_acl, *acl;
140 struct xfs_name name;
141 int error;
142
143 /*
144 * Irix uses Missed'em'V split, but doesn't want to see
145 * the upper 5 bits of (14bit) major.
146 */
147 if (S_ISCHR(mode) || S_ISBLK(mode)) {
148 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
149 return -EINVAL;
150 } else {
151 rdev = 0;
152 }
153
154 error = posix_acl_create(dir, &mode, &default_acl, &acl);
155 if (error)
156 return error;
157
158 /* Verify mode is valid also for tmpfile case */
159 error = xfs_dentry_mode_to_name(&name, dentry, mode);
160 if (unlikely(error))
161 goto out_free_acl;
162
163 if (!tmpfile) {
164 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
165 } else {
166 error = xfs_create_tmpfile(XFS_I(dir), mode, &ip);
167 }
168 if (unlikely(error))
169 goto out_free_acl;
170
171 inode = VFS_I(ip);
172
173 error = xfs_init_security(inode, dir, &dentry->d_name);
174 if (unlikely(error))
175 goto out_cleanup_inode;
176
177#ifdef CONFIG_XFS_POSIX_ACL
178 if (default_acl) {
179 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
180 if (error)
181 goto out_cleanup_inode;
182 }
183 if (acl) {
184 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
185 if (error)
186 goto out_cleanup_inode;
187 }
188#endif
189
190 xfs_setup_iops(ip);
191
192 if (tmpfile) {
193 /*
194 * The VFS requires that any inode fed to d_tmpfile must have
195 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
196 * However, we created the temp file with nlink == 0 because
197 * we're not allowed to put an inode with nlink > 0 on the
198 * unlinked list. Therefore we have to set nlink to 1 so that
199 * d_tmpfile can immediately set it back to zero.
200 */
201 set_nlink(inode, 1);
202 d_tmpfile(dentry, inode);
203 } else
204 d_instantiate(dentry, inode);
205
206 xfs_finish_inode_setup(ip);
207
208 out_free_acl:
209 if (default_acl)
210 posix_acl_release(default_acl);
211 if (acl)
212 posix_acl_release(acl);
213 return error;
214
215 out_cleanup_inode:
216 xfs_finish_inode_setup(ip);
217 if (!tmpfile)
218 xfs_cleanup_inode(dir, inode, dentry);
219 xfs_irele(ip);
220 goto out_free_acl;
221}
222
223STATIC int
224xfs_vn_mknod(
225 struct inode *dir,
226 struct dentry *dentry,
227 umode_t mode,
228 dev_t rdev)
229{
230 return xfs_generic_create(dir, dentry, mode, rdev, false);
231}
232
233STATIC int
234xfs_vn_create(
235 struct inode *dir,
236 struct dentry *dentry,
237 umode_t mode,
238 bool flags)
239{
240 return xfs_vn_mknod(dir, dentry, mode, 0);
241}
242
243STATIC int
244xfs_vn_mkdir(
245 struct inode *dir,
246 struct dentry *dentry,
247 umode_t mode)
248{
249 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
250}
251
252STATIC struct dentry *
253xfs_vn_lookup(
254 struct inode *dir,
255 struct dentry *dentry,
256 unsigned int flags)
257{
258 struct inode *inode;
259 struct xfs_inode *cip;
260 struct xfs_name name;
261 int error;
262
263 if (dentry->d_name.len >= MAXNAMELEN)
264 return ERR_PTR(-ENAMETOOLONG);
265
266 xfs_dentry_to_name(&name, dentry);
267 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
268 if (likely(!error))
269 inode = VFS_I(cip);
270 else if (likely(error == -ENOENT))
271 inode = NULL;
272 else
273 inode = ERR_PTR(error);
274 return d_splice_alias(inode, dentry);
275}
276
277STATIC struct dentry *
278xfs_vn_ci_lookup(
279 struct inode *dir,
280 struct dentry *dentry,
281 unsigned int flags)
282{
283 struct xfs_inode *ip;
284 struct xfs_name xname;
285 struct xfs_name ci_name;
286 struct qstr dname;
287 int error;
288
289 if (dentry->d_name.len >= MAXNAMELEN)
290 return ERR_PTR(-ENAMETOOLONG);
291
292 xfs_dentry_to_name(&xname, dentry);
293 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
294 if (unlikely(error)) {
295 if (unlikely(error != -ENOENT))
296 return ERR_PTR(error);
297 /*
298 * call d_add(dentry, NULL) here when d_drop_negative_children
299 * is called in xfs_vn_mknod (ie. allow negative dentries
300 * with CI filesystems).
301 */
302 return NULL;
303 }
304
305 /* if exact match, just splice and exit */
306 if (!ci_name.name)
307 return d_splice_alias(VFS_I(ip), dentry);
308
309 /* else case-insensitive match... */
310 dname.name = ci_name.name;
311 dname.len = ci_name.len;
312 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
313 kmem_free(ci_name.name);
314 return dentry;
315}
316
317STATIC int
318xfs_vn_link(
319 struct dentry *old_dentry,
320 struct inode *dir,
321 struct dentry *dentry)
322{
323 struct inode *inode = d_inode(old_dentry);
324 struct xfs_name name;
325 int error;
326
327 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
328 if (unlikely(error))
329 return error;
330
331 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
332 if (unlikely(error))
333 return error;
334
335 ihold(inode);
336 d_instantiate(dentry, inode);
337 return 0;
338}
339
340STATIC int
341xfs_vn_unlink(
342 struct inode *dir,
343 struct dentry *dentry)
344{
345 struct xfs_name name;
346 int error;
347
348 xfs_dentry_to_name(&name, dentry);
349
350 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
351 if (error)
352 return error;
353
354 /*
355 * With unlink, the VFS makes the dentry "negative": no inode,
356 * but still hashed. This is incompatible with case-insensitive
357 * mode, so invalidate (unhash) the dentry in CI-mode.
358 */
359 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
360 d_invalidate(dentry);
361 return 0;
362}
363
364STATIC int
365xfs_vn_symlink(
366 struct inode *dir,
367 struct dentry *dentry,
368 const char *symname)
369{
370 struct inode *inode;
371 struct xfs_inode *cip = NULL;
372 struct xfs_name name;
373 int error;
374 umode_t mode;
375
376 mode = S_IFLNK |
377 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
378 error = xfs_dentry_mode_to_name(&name, dentry, mode);
379 if (unlikely(error))
380 goto out;
381
382 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
383 if (unlikely(error))
384 goto out;
385
386 inode = VFS_I(cip);
387
388 error = xfs_init_security(inode, dir, &dentry->d_name);
389 if (unlikely(error))
390 goto out_cleanup_inode;
391
392 xfs_setup_iops(cip);
393
394 d_instantiate(dentry, inode);
395 xfs_finish_inode_setup(cip);
396 return 0;
397
398 out_cleanup_inode:
399 xfs_finish_inode_setup(cip);
400 xfs_cleanup_inode(dir, inode, dentry);
401 xfs_irele(cip);
402 out:
403 return error;
404}
405
406STATIC int
407xfs_vn_rename(
408 struct inode *odir,
409 struct dentry *odentry,
410 struct inode *ndir,
411 struct dentry *ndentry,
412 unsigned int flags)
413{
414 struct inode *new_inode = d_inode(ndentry);
415 int omode = 0;
416 int error;
417 struct xfs_name oname;
418 struct xfs_name nname;
419
420 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
421 return -EINVAL;
422
423 /* if we are exchanging files, we need to set i_mode of both files */
424 if (flags & RENAME_EXCHANGE)
425 omode = d_inode(ndentry)->i_mode;
426
427 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
428 if (omode && unlikely(error))
429 return error;
430
431 error = xfs_dentry_mode_to_name(&nname, ndentry,
432 d_inode(odentry)->i_mode);
433 if (unlikely(error))
434 return error;
435
436 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
437 XFS_I(ndir), &nname,
438 new_inode ? XFS_I(new_inode) : NULL, flags);
439}
440
441/*
442 * careful here - this function can get called recursively, so
443 * we need to be very careful about how much stack we use.
444 * uio is kmalloced for this reason...
445 */
446STATIC const char *
447xfs_vn_get_link(
448 struct dentry *dentry,
449 struct inode *inode,
450 struct delayed_call *done)
451{
452 char *link;
453 int error = -ENOMEM;
454
455 if (!dentry)
456 return ERR_PTR(-ECHILD);
457
458 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
459 if (!link)
460 goto out_err;
461
462 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
463 if (unlikely(error))
464 goto out_kfree;
465
466 set_delayed_call(done, kfree_link, link);
467 return link;
468
469 out_kfree:
470 kfree(link);
471 out_err:
472 return ERR_PTR(error);
473}
474
475STATIC const char *
476xfs_vn_get_link_inline(
477 struct dentry *dentry,
478 struct inode *inode,
479 struct delayed_call *done)
480{
481 struct xfs_inode *ip = XFS_I(inode);
482 char *link;
483
484 ASSERT(ip->i_df.if_flags & XFS_IFINLINE);
485
486 /*
487 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if
488 * if_data is junk.
489 */
490 link = ip->i_df.if_u1.if_data;
491 if (XFS_IS_CORRUPT(ip->i_mount, !link))
492 return ERR_PTR(-EFSCORRUPTED);
493 return link;
494}
495
496static uint32_t
497xfs_stat_blksize(
498 struct xfs_inode *ip)
499{
500 struct xfs_mount *mp = ip->i_mount;
501
502 /*
503 * If the file blocks are being allocated from a realtime volume, then
504 * always return the realtime extent size.
505 */
506 if (XFS_IS_REALTIME_INODE(ip))
507 return xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
508
509 /*
510 * Allow large block sizes to be reported to userspace programs if the
511 * "largeio" mount option is used.
512 *
513 * If compatibility mode is specified, simply return the basic unit of
514 * caching so that we don't get inefficient read/modify/write I/O from
515 * user apps. Otherwise....
516 *
517 * If the underlying volume is a stripe, then return the stripe width in
518 * bytes as the recommended I/O size. It is not a stripe and we've set a
519 * default buffered I/O size, return that, otherwise return the compat
520 * default.
521 */
522 if (mp->m_flags & XFS_MOUNT_LARGEIO) {
523 if (mp->m_swidth)
524 return mp->m_swidth << mp->m_sb.sb_blocklog;
525 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
526 return 1U << mp->m_allocsize_log;
527 }
528
529 return PAGE_SIZE;
530}
531
532STATIC int
533xfs_vn_getattr(
534 const struct path *path,
535 struct kstat *stat,
536 u32 request_mask,
537 unsigned int query_flags)
538{
539 struct inode *inode = d_inode(path->dentry);
540 struct xfs_inode *ip = XFS_I(inode);
541 struct xfs_mount *mp = ip->i_mount;
542
543 trace_xfs_getattr(ip);
544
545 if (XFS_FORCED_SHUTDOWN(mp))
546 return -EIO;
547
548 stat->size = XFS_ISIZE(ip);
549 stat->dev = inode->i_sb->s_dev;
550 stat->mode = inode->i_mode;
551 stat->nlink = inode->i_nlink;
552 stat->uid = inode->i_uid;
553 stat->gid = inode->i_gid;
554 stat->ino = ip->i_ino;
555 stat->atime = inode->i_atime;
556 stat->mtime = inode->i_mtime;
557 stat->ctime = inode->i_ctime;
558 stat->blocks =
559 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
560
561 if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
562 if (request_mask & STATX_BTIME) {
563 stat->result_mask |= STATX_BTIME;
564 stat->btime = ip->i_d.di_crtime;
565 }
566 }
567
568 /*
569 * Note: If you add another clause to set an attribute flag, please
570 * update attributes_mask below.
571 */
572 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
573 stat->attributes |= STATX_ATTR_IMMUTABLE;
574 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
575 stat->attributes |= STATX_ATTR_APPEND;
576 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
577 stat->attributes |= STATX_ATTR_NODUMP;
578
579 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
580 STATX_ATTR_APPEND |
581 STATX_ATTR_NODUMP);
582
583 switch (inode->i_mode & S_IFMT) {
584 case S_IFBLK:
585 case S_IFCHR:
586 stat->blksize = BLKDEV_IOSIZE;
587 stat->rdev = inode->i_rdev;
588 break;
589 default:
590 stat->blksize = xfs_stat_blksize(ip);
591 stat->rdev = 0;
592 break;
593 }
594
595 return 0;
596}
597
598static void
599xfs_setattr_mode(
600 struct xfs_inode *ip,
601 struct iattr *iattr)
602{
603 struct inode *inode = VFS_I(ip);
604 umode_t mode = iattr->ia_mode;
605
606 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
607
608 inode->i_mode &= S_IFMT;
609 inode->i_mode |= mode & ~S_IFMT;
610}
611
612void
613xfs_setattr_time(
614 struct xfs_inode *ip,
615 struct iattr *iattr)
616{
617 struct inode *inode = VFS_I(ip);
618
619 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
620
621 if (iattr->ia_valid & ATTR_ATIME)
622 inode->i_atime = iattr->ia_atime;
623 if (iattr->ia_valid & ATTR_CTIME)
624 inode->i_ctime = iattr->ia_ctime;
625 if (iattr->ia_valid & ATTR_MTIME)
626 inode->i_mtime = iattr->ia_mtime;
627}
628
629static int
630xfs_vn_change_ok(
631 struct dentry *dentry,
632 struct iattr *iattr)
633{
634 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
635
636 if (mp->m_flags & XFS_MOUNT_RDONLY)
637 return -EROFS;
638
639 if (XFS_FORCED_SHUTDOWN(mp))
640 return -EIO;
641
642 return setattr_prepare(dentry, iattr);
643}
644
645/*
646 * Set non-size attributes of an inode.
647 *
648 * Caution: The caller of this function is responsible for calling
649 * setattr_prepare() or otherwise verifying the change is fine.
650 */
651int
652xfs_setattr_nonsize(
653 struct xfs_inode *ip,
654 struct iattr *iattr,
655 int flags)
656{
657 xfs_mount_t *mp = ip->i_mount;
658 struct inode *inode = VFS_I(ip);
659 int mask = iattr->ia_valid;
660 xfs_trans_t *tp;
661 int error;
662 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
663 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
664 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
665 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
666
667 ASSERT((mask & ATTR_SIZE) == 0);
668
669 /*
670 * If disk quotas is on, we make sure that the dquots do exist on disk,
671 * before we start any other transactions. Trying to do this later
672 * is messy. We don't care to take a readlock to look at the ids
673 * in inode here, because we can't hold it across the trans_reserve.
674 * If the IDs do change before we take the ilock, we're covered
675 * because the i_*dquot fields will get updated anyway.
676 */
677 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
678 uint qflags = 0;
679
680 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
681 uid = iattr->ia_uid;
682 qflags |= XFS_QMOPT_UQUOTA;
683 } else {
684 uid = inode->i_uid;
685 }
686 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
687 gid = iattr->ia_gid;
688 qflags |= XFS_QMOPT_GQUOTA;
689 } else {
690 gid = inode->i_gid;
691 }
692
693 /*
694 * We take a reference when we initialize udqp and gdqp,
695 * so it is important that we never blindly double trip on
696 * the same variable. See xfs_create() for an example.
697 */
698 ASSERT(udqp == NULL);
699 ASSERT(gdqp == NULL);
700 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_d.di_projid,
701 qflags, &udqp, &gdqp, NULL);
702 if (error)
703 return error;
704 }
705
706 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
707 if (error)
708 goto out_dqrele;
709
710 xfs_ilock(ip, XFS_ILOCK_EXCL);
711 xfs_trans_ijoin(tp, ip, 0);
712
713 /*
714 * Change file ownership. Must be the owner or privileged.
715 */
716 if (mask & (ATTR_UID|ATTR_GID)) {
717 /*
718 * These IDs could have changed since we last looked at them.
719 * But, we're assured that if the ownership did change
720 * while we didn't have the inode locked, inode's dquot(s)
721 * would have changed also.
722 */
723 iuid = inode->i_uid;
724 igid = inode->i_gid;
725 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
726 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
727
728 /*
729 * Do a quota reservation only if uid/gid is actually
730 * going to change.
731 */
732 if (XFS_IS_QUOTA_RUNNING(mp) &&
733 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
734 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
735 ASSERT(tp);
736 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
737 NULL, capable(CAP_FOWNER) ?
738 XFS_QMOPT_FORCE_RES : 0);
739 if (error) /* out of quota */
740 goto out_cancel;
741 }
742
743 /*
744 * CAP_FSETID overrides the following restrictions:
745 *
746 * The set-user-ID and set-group-ID bits of a file will be
747 * cleared upon successful return from chown()
748 */
749 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
750 !capable(CAP_FSETID))
751 inode->i_mode &= ~(S_ISUID|S_ISGID);
752
753 /*
754 * Change the ownerships and register quota modifications
755 * in the transaction.
756 */
757 if (!uid_eq(iuid, uid)) {
758 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
759 ASSERT(mask & ATTR_UID);
760 ASSERT(udqp);
761 olddquot1 = xfs_qm_vop_chown(tp, ip,
762 &ip->i_udquot, udqp);
763 }
764 inode->i_uid = uid;
765 }
766 if (!gid_eq(igid, gid)) {
767 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
768 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
769 !XFS_IS_PQUOTA_ON(mp));
770 ASSERT(mask & ATTR_GID);
771 ASSERT(gdqp);
772 olddquot2 = xfs_qm_vop_chown(tp, ip,
773 &ip->i_gdquot, gdqp);
774 }
775 inode->i_gid = gid;
776 }
777 }
778
779 if (mask & ATTR_MODE)
780 xfs_setattr_mode(ip, iattr);
781 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
782 xfs_setattr_time(ip, iattr);
783
784 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
785
786 XFS_STATS_INC(mp, xs_ig_attrchg);
787
788 if (mp->m_flags & XFS_MOUNT_WSYNC)
789 xfs_trans_set_sync(tp);
790 error = xfs_trans_commit(tp);
791
792 xfs_iunlock(ip, XFS_ILOCK_EXCL);
793
794 /*
795 * Release any dquot(s) the inode had kept before chown.
796 */
797 xfs_qm_dqrele(olddquot1);
798 xfs_qm_dqrele(olddquot2);
799 xfs_qm_dqrele(udqp);
800 xfs_qm_dqrele(gdqp);
801
802 if (error)
803 return error;
804
805 /*
806 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
807 * update. We could avoid this with linked transactions
808 * and passing down the transaction pointer all the way
809 * to attr_set. No previous user of the generic
810 * Posix ACL code seems to care about this issue either.
811 */
812 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
813 error = posix_acl_chmod(inode, inode->i_mode);
814 if (error)
815 return error;
816 }
817
818 return 0;
819
820out_cancel:
821 xfs_trans_cancel(tp);
822 xfs_iunlock(ip, XFS_ILOCK_EXCL);
823out_dqrele:
824 xfs_qm_dqrele(udqp);
825 xfs_qm_dqrele(gdqp);
826 return error;
827}
828
829int
830xfs_vn_setattr_nonsize(
831 struct dentry *dentry,
832 struct iattr *iattr)
833{
834 struct xfs_inode *ip = XFS_I(d_inode(dentry));
835 int error;
836
837 trace_xfs_setattr(ip);
838
839 error = xfs_vn_change_ok(dentry, iattr);
840 if (error)
841 return error;
842 return xfs_setattr_nonsize(ip, iattr, 0);
843}
844
845/*
846 * Truncate file. Must have write permission and not be a directory.
847 *
848 * Caution: The caller of this function is responsible for calling
849 * setattr_prepare() or otherwise verifying the change is fine.
850 */
851STATIC int
852xfs_setattr_size(
853 struct xfs_inode *ip,
854 struct iattr *iattr)
855{
856 struct xfs_mount *mp = ip->i_mount;
857 struct inode *inode = VFS_I(ip);
858 xfs_off_t oldsize, newsize;
859 struct xfs_trans *tp;
860 int error;
861 uint lock_flags = 0;
862 bool did_zeroing = false;
863
864 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
865 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
866 ASSERT(S_ISREG(inode->i_mode));
867 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
868 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
869
870 oldsize = inode->i_size;
871 newsize = iattr->ia_size;
872
873 /*
874 * Short circuit the truncate case for zero length files.
875 */
876 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
877 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
878 return 0;
879
880 /*
881 * Use the regular setattr path to update the timestamps.
882 */
883 iattr->ia_valid &= ~ATTR_SIZE;
884 return xfs_setattr_nonsize(ip, iattr, 0);
885 }
886
887 /*
888 * Make sure that the dquots are attached to the inode.
889 */
890 error = xfs_qm_dqattach(ip);
891 if (error)
892 return error;
893
894 /*
895 * Wait for all direct I/O to complete.
896 */
897 inode_dio_wait(inode);
898
899 /*
900 * File data changes must be complete before we start the transaction to
901 * modify the inode. This needs to be done before joining the inode to
902 * the transaction because the inode cannot be unlocked once it is a
903 * part of the transaction.
904 *
905 * Start with zeroing any data beyond EOF that we may expose on file
906 * extension, or zeroing out the rest of the block on a downward
907 * truncate.
908 */
909 if (newsize > oldsize) {
910 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
911 error = iomap_zero_range(inode, oldsize, newsize - oldsize,
912 &did_zeroing, &xfs_buffered_write_iomap_ops);
913 } else {
914 error = iomap_truncate_page(inode, newsize, &did_zeroing,
915 &xfs_buffered_write_iomap_ops);
916 }
917
918 if (error)
919 return error;
920
921 /*
922 * We've already locked out new page faults, so now we can safely remove
923 * pages from the page cache knowing they won't get refaulted until we
924 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
925 * complete. The truncate_setsize() call also cleans partial EOF page
926 * PTEs on extending truncates and hence ensures sub-page block size
927 * filesystems are correctly handled, too.
928 *
929 * We have to do all the page cache truncate work outside the
930 * transaction context as the "lock" order is page lock->log space
931 * reservation as defined by extent allocation in the writeback path.
932 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
933 * having already truncated the in-memory version of the file (i.e. made
934 * user visible changes). There's not much we can do about this, except
935 * to hope that the caller sees ENOMEM and retries the truncate
936 * operation.
937 *
938 * And we update in-core i_size and truncate page cache beyond newsize
939 * before writeback the [di_size, newsize] range, so we're guaranteed
940 * not to write stale data past the new EOF on truncate down.
941 */
942 truncate_setsize(inode, newsize);
943
944 /*
945 * We are going to log the inode size change in this transaction so
946 * any previous writes that are beyond the on disk EOF and the new
947 * EOF that have not been written out need to be written here. If we
948 * do not write the data out, we expose ourselves to the null files
949 * problem. Note that this includes any block zeroing we did above;
950 * otherwise those blocks may not be zeroed after a crash.
951 */
952 if (did_zeroing ||
953 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
954 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
955 ip->i_d.di_size, newsize - 1);
956 if (error)
957 return error;
958 }
959
960 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
961 if (error)
962 return error;
963
964 lock_flags |= XFS_ILOCK_EXCL;
965 xfs_ilock(ip, XFS_ILOCK_EXCL);
966 xfs_trans_ijoin(tp, ip, 0);
967
968 /*
969 * Only change the c/mtime if we are changing the size or we are
970 * explicitly asked to change it. This handles the semantic difference
971 * between truncate() and ftruncate() as implemented in the VFS.
972 *
973 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
974 * special case where we need to update the times despite not having
975 * these flags set. For all other operations the VFS set these flags
976 * explicitly if it wants a timestamp update.
977 */
978 if (newsize != oldsize &&
979 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
980 iattr->ia_ctime = iattr->ia_mtime =
981 current_time(inode);
982 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
983 }
984
985 /*
986 * The first thing we do is set the size to new_size permanently on
987 * disk. This way we don't have to worry about anyone ever being able
988 * to look at the data being freed even in the face of a crash.
989 * What we're getting around here is the case where we free a block, it
990 * is allocated to another file, it is written to, and then we crash.
991 * If the new data gets written to the file but the log buffers
992 * containing the free and reallocation don't, then we'd end up with
993 * garbage in the blocks being freed. As long as we make the new size
994 * permanent before actually freeing any blocks it doesn't matter if
995 * they get written to.
996 */
997 ip->i_d.di_size = newsize;
998 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
999
1000 if (newsize <= oldsize) {
1001 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
1002 if (error)
1003 goto out_trans_cancel;
1004
1005 /*
1006 * Truncated "down", so we're removing references to old data
1007 * here - if we delay flushing for a long time, we expose
1008 * ourselves unduly to the notorious NULL files problem. So,
1009 * we mark this inode and flush it when the file is closed,
1010 * and do not wait the usual (long) time for writeout.
1011 */
1012 xfs_iflags_set(ip, XFS_ITRUNCATED);
1013
1014 /* A truncate down always removes post-EOF blocks. */
1015 xfs_inode_clear_eofblocks_tag(ip);
1016 }
1017
1018 if (iattr->ia_valid & ATTR_MODE)
1019 xfs_setattr_mode(ip, iattr);
1020 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
1021 xfs_setattr_time(ip, iattr);
1022
1023 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1024
1025 XFS_STATS_INC(mp, xs_ig_attrchg);
1026
1027 if (mp->m_flags & XFS_MOUNT_WSYNC)
1028 xfs_trans_set_sync(tp);
1029
1030 error = xfs_trans_commit(tp);
1031out_unlock:
1032 if (lock_flags)
1033 xfs_iunlock(ip, lock_flags);
1034 return error;
1035
1036out_trans_cancel:
1037 xfs_trans_cancel(tp);
1038 goto out_unlock;
1039}
1040
1041int
1042xfs_vn_setattr_size(
1043 struct dentry *dentry,
1044 struct iattr *iattr)
1045{
1046 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1047 int error;
1048
1049 trace_xfs_setattr(ip);
1050
1051 error = xfs_vn_change_ok(dentry, iattr);
1052 if (error)
1053 return error;
1054 return xfs_setattr_size(ip, iattr);
1055}
1056
1057STATIC int
1058xfs_vn_setattr(
1059 struct dentry *dentry,
1060 struct iattr *iattr)
1061{
1062 int error;
1063
1064 if (iattr->ia_valid & ATTR_SIZE) {
1065 struct inode *inode = d_inode(dentry);
1066 struct xfs_inode *ip = XFS_I(inode);
1067 uint iolock;
1068
1069 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1070 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1071
1072 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1073 if (error) {
1074 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1075 return error;
1076 }
1077
1078 error = xfs_vn_setattr_size(dentry, iattr);
1079 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1080 } else {
1081 error = xfs_vn_setattr_nonsize(dentry, iattr);
1082 }
1083
1084 return error;
1085}
1086
1087STATIC int
1088xfs_vn_update_time(
1089 struct inode *inode,
1090 struct timespec64 *now,
1091 int flags)
1092{
1093 struct xfs_inode *ip = XFS_I(inode);
1094 struct xfs_mount *mp = ip->i_mount;
1095 int log_flags = XFS_ILOG_TIMESTAMP;
1096 struct xfs_trans *tp;
1097 int error;
1098
1099 trace_xfs_update_time(ip);
1100
1101 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1102 if (!((flags & S_VERSION) &&
1103 inode_maybe_inc_iversion(inode, false)))
1104 return generic_update_time(inode, now, flags);
1105
1106 /* Capture the iversion update that just occurred */
1107 log_flags |= XFS_ILOG_CORE;
1108 }
1109
1110 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1111 if (error)
1112 return error;
1113
1114 xfs_ilock(ip, XFS_ILOCK_EXCL);
1115 if (flags & S_CTIME)
1116 inode->i_ctime = *now;
1117 if (flags & S_MTIME)
1118 inode->i_mtime = *now;
1119 if (flags & S_ATIME)
1120 inode->i_atime = *now;
1121
1122 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1123 xfs_trans_log_inode(tp, ip, log_flags);
1124 return xfs_trans_commit(tp);
1125}
1126
1127STATIC int
1128xfs_vn_fiemap(
1129 struct inode *inode,
1130 struct fiemap_extent_info *fieinfo,
1131 u64 start,
1132 u64 length)
1133{
1134 int error;
1135
1136 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1137 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1138 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1139 error = iomap_fiemap(inode, fieinfo, start, length,
1140 &xfs_xattr_iomap_ops);
1141 } else {
1142 error = iomap_fiemap(inode, fieinfo, start, length,
1143 &xfs_read_iomap_ops);
1144 }
1145 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1146
1147 return error;
1148}
1149
1150STATIC int
1151xfs_vn_tmpfile(
1152 struct inode *dir,
1153 struct dentry *dentry,
1154 umode_t mode)
1155{
1156 return xfs_generic_create(dir, dentry, mode, 0, true);
1157}
1158
1159static const struct inode_operations xfs_inode_operations = {
1160 .get_acl = xfs_get_acl,
1161 .set_acl = xfs_set_acl,
1162 .getattr = xfs_vn_getattr,
1163 .setattr = xfs_vn_setattr,
1164 .listxattr = xfs_vn_listxattr,
1165 .fiemap = xfs_vn_fiemap,
1166 .update_time = xfs_vn_update_time,
1167};
1168
1169static const struct inode_operations xfs_dir_inode_operations = {
1170 .create = xfs_vn_create,
1171 .lookup = xfs_vn_lookup,
1172 .link = xfs_vn_link,
1173 .unlink = xfs_vn_unlink,
1174 .symlink = xfs_vn_symlink,
1175 .mkdir = xfs_vn_mkdir,
1176 /*
1177 * Yes, XFS uses the same method for rmdir and unlink.
1178 *
1179 * There are some subtile differences deeper in the code,
1180 * but we use S_ISDIR to check for those.
1181 */
1182 .rmdir = xfs_vn_unlink,
1183 .mknod = xfs_vn_mknod,
1184 .rename = xfs_vn_rename,
1185 .get_acl = xfs_get_acl,
1186 .set_acl = xfs_set_acl,
1187 .getattr = xfs_vn_getattr,
1188 .setattr = xfs_vn_setattr,
1189 .listxattr = xfs_vn_listxattr,
1190 .update_time = xfs_vn_update_time,
1191 .tmpfile = xfs_vn_tmpfile,
1192};
1193
1194static const struct inode_operations xfs_dir_ci_inode_operations = {
1195 .create = xfs_vn_create,
1196 .lookup = xfs_vn_ci_lookup,
1197 .link = xfs_vn_link,
1198 .unlink = xfs_vn_unlink,
1199 .symlink = xfs_vn_symlink,
1200 .mkdir = xfs_vn_mkdir,
1201 /*
1202 * Yes, XFS uses the same method for rmdir and unlink.
1203 *
1204 * There are some subtile differences deeper in the code,
1205 * but we use S_ISDIR to check for those.
1206 */
1207 .rmdir = xfs_vn_unlink,
1208 .mknod = xfs_vn_mknod,
1209 .rename = xfs_vn_rename,
1210 .get_acl = xfs_get_acl,
1211 .set_acl = xfs_set_acl,
1212 .getattr = xfs_vn_getattr,
1213 .setattr = xfs_vn_setattr,
1214 .listxattr = xfs_vn_listxattr,
1215 .update_time = xfs_vn_update_time,
1216 .tmpfile = xfs_vn_tmpfile,
1217};
1218
1219static const struct inode_operations xfs_symlink_inode_operations = {
1220 .get_link = xfs_vn_get_link,
1221 .getattr = xfs_vn_getattr,
1222 .setattr = xfs_vn_setattr,
1223 .listxattr = xfs_vn_listxattr,
1224 .update_time = xfs_vn_update_time,
1225};
1226
1227static const struct inode_operations xfs_inline_symlink_inode_operations = {
1228 .get_link = xfs_vn_get_link_inline,
1229 .getattr = xfs_vn_getattr,
1230 .setattr = xfs_vn_setattr,
1231 .listxattr = xfs_vn_listxattr,
1232 .update_time = xfs_vn_update_time,
1233};
1234
1235/* Figure out if this file actually supports DAX. */
1236static bool
1237xfs_inode_supports_dax(
1238 struct xfs_inode *ip)
1239{
1240 struct xfs_mount *mp = ip->i_mount;
1241
1242 /* Only supported on regular files. */
1243 if (!S_ISREG(VFS_I(ip)->i_mode))
1244 return false;
1245
1246 /* Only supported on non-reflinked files. */
1247 if (xfs_is_reflink_inode(ip))
1248 return false;
1249
1250 /* Block size must match page size */
1251 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1252 return false;
1253
1254 /* Device has to support DAX too. */
1255 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1256}
1257
1258static bool
1259xfs_inode_should_enable_dax(
1260 struct xfs_inode *ip)
1261{
1262 if (!IS_ENABLED(CONFIG_FS_DAX))
1263 return false;
1264 if (ip->i_mount->m_flags & XFS_MOUNT_DAX_NEVER)
1265 return false;
1266 if (!xfs_inode_supports_dax(ip))
1267 return false;
1268 if (ip->i_mount->m_flags & XFS_MOUNT_DAX_ALWAYS)
1269 return true;
1270 if (ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)
1271 return true;
1272 return false;
1273}
1274
1275void
1276xfs_diflags_to_iflags(
1277 struct xfs_inode *ip,
1278 bool init)
1279{
1280 struct inode *inode = VFS_I(ip);
1281 unsigned int xflags = xfs_ip2xflags(ip);
1282 unsigned int flags = 0;
1283
1284 ASSERT(!(IS_DAX(inode) && init));
1285
1286 if (xflags & FS_XFLAG_IMMUTABLE)
1287 flags |= S_IMMUTABLE;
1288 if (xflags & FS_XFLAG_APPEND)
1289 flags |= S_APPEND;
1290 if (xflags & FS_XFLAG_SYNC)
1291 flags |= S_SYNC;
1292 if (xflags & FS_XFLAG_NOATIME)
1293 flags |= S_NOATIME;
1294 if (init && xfs_inode_should_enable_dax(ip))
1295 flags |= S_DAX;
1296
1297 /*
1298 * S_DAX can only be set during inode initialization and is never set by
1299 * the VFS, so we cannot mask off S_DAX in i_flags.
1300 */
1301 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1302 inode->i_flags |= flags;
1303}
1304
1305/*
1306 * Initialize the Linux inode.
1307 *
1308 * When reading existing inodes from disk this is called directly from xfs_iget,
1309 * when creating a new inode it is called from xfs_ialloc after setting up the
1310 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1311 * it up to the caller to deal with unlocking the inode appropriately.
1312 */
1313void
1314xfs_setup_inode(
1315 struct xfs_inode *ip)
1316{
1317 struct inode *inode = &ip->i_vnode;
1318 gfp_t gfp_mask;
1319
1320 inode->i_ino = ip->i_ino;
1321 inode->i_state = I_NEW;
1322
1323 inode_sb_list_add(inode);
1324 /* make the inode look hashed for the writeback code */
1325 inode_fake_hash(inode);
1326
1327 i_size_write(inode, ip->i_d.di_size);
1328 xfs_diflags_to_iflags(ip, true);
1329
1330 if (S_ISDIR(inode->i_mode)) {
1331 /*
1332 * We set the i_rwsem class here to avoid potential races with
1333 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1334 * after a filehandle lookup has already found the inode in
1335 * cache before it has been unlocked via unlock_new_inode().
1336 */
1337 lockdep_set_class(&inode->i_rwsem,
1338 &inode->i_sb->s_type->i_mutex_dir_key);
1339 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1340 } else {
1341 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1342 }
1343
1344 /*
1345 * Ensure all page cache allocations are done from GFP_NOFS context to
1346 * prevent direct reclaim recursion back into the filesystem and blowing
1347 * stacks or deadlocking.
1348 */
1349 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1350 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1351
1352 /*
1353 * If there is no attribute fork no ACL can exist on this inode,
1354 * and it can't have any file capabilities attached to it either.
1355 */
1356 if (!XFS_IFORK_Q(ip)) {
1357 inode_has_no_xattr(inode);
1358 cache_no_acl(inode);
1359 }
1360}
1361
1362void
1363xfs_setup_iops(
1364 struct xfs_inode *ip)
1365{
1366 struct inode *inode = &ip->i_vnode;
1367
1368 switch (inode->i_mode & S_IFMT) {
1369 case S_IFREG:
1370 inode->i_op = &xfs_inode_operations;
1371 inode->i_fop = &xfs_file_operations;
1372 if (IS_DAX(inode))
1373 inode->i_mapping->a_ops = &xfs_dax_aops;
1374 else
1375 inode->i_mapping->a_ops = &xfs_address_space_operations;
1376 break;
1377 case S_IFDIR:
1378 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1379 inode->i_op = &xfs_dir_ci_inode_operations;
1380 else
1381 inode->i_op = &xfs_dir_inode_operations;
1382 inode->i_fop = &xfs_dir_file_operations;
1383 break;
1384 case S_IFLNK:
1385 if (ip->i_df.if_flags & XFS_IFINLINE)
1386 inode->i_op = &xfs_inline_symlink_inode_operations;
1387 else
1388 inode->i_op = &xfs_symlink_inode_operations;
1389 break;
1390 default:
1391 inode->i_op = &xfs_inode_operations;
1392 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1393 break;
1394 }
1395}