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_shared.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_mount.h"
25#include "xfs_da_format.h"
26#include "xfs_inode.h"
27#include "xfs_bmap.h"
28#include "xfs_bmap_util.h"
29#include "xfs_acl.h"
30#include "xfs_quota.h"
31#include "xfs_error.h"
32#include "xfs_attr.h"
33#include "xfs_trans.h"
34#include "xfs_trace.h"
35#include "xfs_icache.h"
36#include "xfs_symlink.h"
37#include "xfs_da_btree.h"
38#include "xfs_dir2.h"
39#include "xfs_trans_space.h"
40#include "xfs_pnfs.h"
41#include "xfs_iomap.h"
42
43#include <linux/capability.h>
44#include <linux/xattr.h>
45#include <linux/posix_acl.h>
46#include <linux/security.h>
47#include <linux/iomap.h>
48#include <linux/slab.h>
49
50/*
51 * Directories have different lock order w.r.t. mmap_sem compared to regular
52 * files. This is due to readdir potentially triggering page faults on a user
53 * buffer inside filldir(), and this happens with the ilock on the directory
54 * held. For regular files, the lock order is the other way around - the
55 * mmap_sem is taken during the page fault, and then we lock the ilock to do
56 * block mapping. Hence we need a different class for the directory ilock so
57 * that lockdep can tell them apart.
58 */
59static struct lock_class_key xfs_nondir_ilock_class;
60static struct lock_class_key xfs_dir_ilock_class;
61
62static int
63xfs_initxattrs(
64 struct inode *inode,
65 const struct xattr *xattr_array,
66 void *fs_info)
67{
68 const struct xattr *xattr;
69 struct xfs_inode *ip = XFS_I(inode);
70 int error = 0;
71
72 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
73 error = xfs_attr_set(ip, xattr->name, xattr->value,
74 xattr->value_len, ATTR_SECURE);
75 if (error < 0)
76 break;
77 }
78 return error;
79}
80
81/*
82 * Hook in SELinux. This is not quite correct yet, what we really need
83 * here (as we do for default ACLs) is a mechanism by which creation of
84 * these attrs can be journalled at inode creation time (along with the
85 * inode, of course, such that log replay can't cause these to be lost).
86 */
87
88STATIC int
89xfs_init_security(
90 struct inode *inode,
91 struct inode *dir,
92 const struct qstr *qstr)
93{
94 return security_inode_init_security(inode, dir, qstr,
95 &xfs_initxattrs, NULL);
96}
97
98static void
99xfs_dentry_to_name(
100 struct xfs_name *namep,
101 struct dentry *dentry)
102{
103 namep->name = dentry->d_name.name;
104 namep->len = dentry->d_name.len;
105 namep->type = XFS_DIR3_FT_UNKNOWN;
106}
107
108static int
109xfs_dentry_mode_to_name(
110 struct xfs_name *namep,
111 struct dentry *dentry,
112 int mode)
113{
114 namep->name = dentry->d_name.name;
115 namep->len = dentry->d_name.len;
116 namep->type = xfs_mode_to_ftype(mode);
117
118 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
119 return -EFSCORRUPTED;
120
121 return 0;
122}
123
124STATIC void
125xfs_cleanup_inode(
126 struct inode *dir,
127 struct inode *inode,
128 struct dentry *dentry)
129{
130 struct xfs_name teardown;
131
132 /* Oh, the horror.
133 * If we can't add the ACL or we fail in
134 * xfs_init_security we must back out.
135 * ENOSPC can hit here, among other things.
136 */
137 xfs_dentry_to_name(&teardown, dentry);
138
139 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
140}
141
142STATIC int
143xfs_generic_create(
144 struct inode *dir,
145 struct dentry *dentry,
146 umode_t mode,
147 dev_t rdev,
148 bool tmpfile) /* unnamed file */
149{
150 struct inode *inode;
151 struct xfs_inode *ip = NULL;
152 struct posix_acl *default_acl, *acl;
153 struct xfs_name name;
154 int error;
155
156 /*
157 * Irix uses Missed'em'V split, but doesn't want to see
158 * the upper 5 bits of (14bit) major.
159 */
160 if (S_ISCHR(mode) || S_ISBLK(mode)) {
161 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
162 return -EINVAL;
163 rdev = sysv_encode_dev(rdev);
164 } else {
165 rdev = 0;
166 }
167
168 error = posix_acl_create(dir, &mode, &default_acl, &acl);
169 if (error)
170 return error;
171
172 /* Verify mode is valid also for tmpfile case */
173 error = xfs_dentry_mode_to_name(&name, dentry, mode);
174 if (unlikely(error))
175 goto out_free_acl;
176
177 if (!tmpfile) {
178 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
179 } else {
180 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
181 }
182 if (unlikely(error))
183 goto out_free_acl;
184
185 inode = VFS_I(ip);
186
187 error = xfs_init_security(inode, dir, &dentry->d_name);
188 if (unlikely(error))
189 goto out_cleanup_inode;
190
191#ifdef CONFIG_XFS_POSIX_ACL
192 if (default_acl) {
193 error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
194 if (error)
195 goto out_cleanup_inode;
196 }
197 if (acl) {
198 error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
199 if (error)
200 goto out_cleanup_inode;
201 }
202#endif
203
204 xfs_setup_iops(ip);
205
206 if (tmpfile)
207 d_tmpfile(dentry, inode);
208 else
209 d_instantiate(dentry, inode);
210
211 xfs_finish_inode_setup(ip);
212
213 out_free_acl:
214 if (default_acl)
215 posix_acl_release(default_acl);
216 if (acl)
217 posix_acl_release(acl);
218 return error;
219
220 out_cleanup_inode:
221 xfs_finish_inode_setup(ip);
222 if (!tmpfile)
223 xfs_cleanup_inode(dir, inode, dentry);
224 iput(inode);
225 goto out_free_acl;
226}
227
228STATIC int
229xfs_vn_mknod(
230 struct inode *dir,
231 struct dentry *dentry,
232 umode_t mode,
233 dev_t rdev)
234{
235 return xfs_generic_create(dir, dentry, mode, rdev, false);
236}
237
238STATIC int
239xfs_vn_create(
240 struct inode *dir,
241 struct dentry *dentry,
242 umode_t mode,
243 bool flags)
244{
245 return xfs_vn_mknod(dir, dentry, mode, 0);
246}
247
248STATIC int
249xfs_vn_mkdir(
250 struct inode *dir,
251 struct dentry *dentry,
252 umode_t mode)
253{
254 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
255}
256
257STATIC struct dentry *
258xfs_vn_lookup(
259 struct inode *dir,
260 struct dentry *dentry,
261 unsigned int flags)
262{
263 struct xfs_inode *cip;
264 struct xfs_name name;
265 int error;
266
267 if (dentry->d_name.len >= MAXNAMELEN)
268 return ERR_PTR(-ENAMETOOLONG);
269
270 xfs_dentry_to_name(&name, dentry);
271 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
272 if (unlikely(error)) {
273 if (unlikely(error != -ENOENT))
274 return ERR_PTR(error);
275 d_add(dentry, NULL);
276 return NULL;
277 }
278
279 return d_splice_alias(VFS_I(cip), dentry);
280}
281
282STATIC struct dentry *
283xfs_vn_ci_lookup(
284 struct inode *dir,
285 struct dentry *dentry,
286 unsigned int flags)
287{
288 struct xfs_inode *ip;
289 struct xfs_name xname;
290 struct xfs_name ci_name;
291 struct qstr dname;
292 int error;
293
294 if (dentry->d_name.len >= MAXNAMELEN)
295 return ERR_PTR(-ENAMETOOLONG);
296
297 xfs_dentry_to_name(&xname, dentry);
298 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
299 if (unlikely(error)) {
300 if (unlikely(error != -ENOENT))
301 return ERR_PTR(error);
302 /*
303 * call d_add(dentry, NULL) here when d_drop_negative_children
304 * is called in xfs_vn_mknod (ie. allow negative dentries
305 * with CI filesystems).
306 */
307 return NULL;
308 }
309
310 /* if exact match, just splice and exit */
311 if (!ci_name.name)
312 return d_splice_alias(VFS_I(ip), dentry);
313
314 /* else case-insensitive match... */
315 dname.name = ci_name.name;
316 dname.len = ci_name.len;
317 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
318 kmem_free(ci_name.name);
319 return dentry;
320}
321
322STATIC int
323xfs_vn_link(
324 struct dentry *old_dentry,
325 struct inode *dir,
326 struct dentry *dentry)
327{
328 struct inode *inode = d_inode(old_dentry);
329 struct xfs_name name;
330 int error;
331
332 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
333 if (unlikely(error))
334 return error;
335
336 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
337 if (unlikely(error))
338 return error;
339
340 ihold(inode);
341 d_instantiate(dentry, inode);
342 return 0;
343}
344
345STATIC int
346xfs_vn_unlink(
347 struct inode *dir,
348 struct dentry *dentry)
349{
350 struct xfs_name name;
351 int error;
352
353 xfs_dentry_to_name(&name, dentry);
354
355 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
356 if (error)
357 return error;
358
359 /*
360 * With unlink, the VFS makes the dentry "negative": no inode,
361 * but still hashed. This is incompatible with case-insensitive
362 * mode, so invalidate (unhash) the dentry in CI-mode.
363 */
364 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
365 d_invalidate(dentry);
366 return 0;
367}
368
369STATIC int
370xfs_vn_symlink(
371 struct inode *dir,
372 struct dentry *dentry,
373 const char *symname)
374{
375 struct inode *inode;
376 struct xfs_inode *cip = NULL;
377 struct xfs_name name;
378 int error;
379 umode_t mode;
380
381 mode = S_IFLNK |
382 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
383 error = xfs_dentry_mode_to_name(&name, dentry, mode);
384 if (unlikely(error))
385 goto out;
386
387 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
388 if (unlikely(error))
389 goto out;
390
391 inode = VFS_I(cip);
392
393 error = xfs_init_security(inode, dir, &dentry->d_name);
394 if (unlikely(error))
395 goto out_cleanup_inode;
396
397 xfs_setup_iops(cip);
398
399 d_instantiate(dentry, inode);
400 xfs_finish_inode_setup(cip);
401 return 0;
402
403 out_cleanup_inode:
404 xfs_finish_inode_setup(cip);
405 xfs_cleanup_inode(dir, inode, dentry);
406 iput(inode);
407 out:
408 return error;
409}
410
411STATIC int
412xfs_vn_rename(
413 struct inode *odir,
414 struct dentry *odentry,
415 struct inode *ndir,
416 struct dentry *ndentry,
417 unsigned int flags)
418{
419 struct inode *new_inode = d_inode(ndentry);
420 int omode = 0;
421 int error;
422 struct xfs_name oname;
423 struct xfs_name nname;
424
425 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
426 return -EINVAL;
427
428 /* if we are exchanging files, we need to set i_mode of both files */
429 if (flags & RENAME_EXCHANGE)
430 omode = d_inode(ndentry)->i_mode;
431
432 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
433 if (omode && unlikely(error))
434 return error;
435
436 error = xfs_dentry_mode_to_name(&nname, ndentry,
437 d_inode(odentry)->i_mode);
438 if (unlikely(error))
439 return error;
440
441 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
442 XFS_I(ndir), &nname,
443 new_inode ? XFS_I(new_inode) : NULL, flags);
444}
445
446/*
447 * careful here - this function can get called recursively, so
448 * we need to be very careful about how much stack we use.
449 * uio is kmalloced for this reason...
450 */
451STATIC const char *
452xfs_vn_get_link(
453 struct dentry *dentry,
454 struct inode *inode,
455 struct delayed_call *done)
456{
457 char *link;
458 int error = -ENOMEM;
459
460 if (!dentry)
461 return ERR_PTR(-ECHILD);
462
463 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
464 if (!link)
465 goto out_err;
466
467 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
468 if (unlikely(error))
469 goto out_kfree;
470
471 set_delayed_call(done, kfree_link, link);
472 return link;
473
474 out_kfree:
475 kfree(link);
476 out_err:
477 return ERR_PTR(error);
478}
479
480STATIC const char *
481xfs_vn_get_link_inline(
482 struct dentry *dentry,
483 struct inode *inode,
484 struct delayed_call *done)
485{
486 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
487 return XFS_I(inode)->i_df.if_u1.if_data;
488}
489
490STATIC int
491xfs_vn_getattr(
492 struct vfsmount *mnt,
493 struct dentry *dentry,
494 struct kstat *stat)
495{
496 struct inode *inode = d_inode(dentry);
497 struct xfs_inode *ip = XFS_I(inode);
498 struct xfs_mount *mp = ip->i_mount;
499
500 trace_xfs_getattr(ip);
501
502 if (XFS_FORCED_SHUTDOWN(mp))
503 return -EIO;
504
505 stat->size = XFS_ISIZE(ip);
506 stat->dev = inode->i_sb->s_dev;
507 stat->mode = inode->i_mode;
508 stat->nlink = inode->i_nlink;
509 stat->uid = inode->i_uid;
510 stat->gid = inode->i_gid;
511 stat->ino = ip->i_ino;
512 stat->atime = inode->i_atime;
513 stat->mtime = inode->i_mtime;
514 stat->ctime = inode->i_ctime;
515 stat->blocks =
516 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
517
518
519 switch (inode->i_mode & S_IFMT) {
520 case S_IFBLK:
521 case S_IFCHR:
522 stat->blksize = BLKDEV_IOSIZE;
523 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
524 sysv_minor(ip->i_df.if_u2.if_rdev));
525 break;
526 default:
527 if (XFS_IS_REALTIME_INODE(ip)) {
528 /*
529 * If the file blocks are being allocated from a
530 * realtime volume, then return the inode's realtime
531 * extent size or the realtime volume's extent size.
532 */
533 stat->blksize =
534 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
535 } else
536 stat->blksize = xfs_preferred_iosize(mp);
537 stat->rdev = 0;
538 break;
539 }
540
541 return 0;
542}
543
544static void
545xfs_setattr_mode(
546 struct xfs_inode *ip,
547 struct iattr *iattr)
548{
549 struct inode *inode = VFS_I(ip);
550 umode_t mode = iattr->ia_mode;
551
552 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
553
554 inode->i_mode &= S_IFMT;
555 inode->i_mode |= mode & ~S_IFMT;
556}
557
558void
559xfs_setattr_time(
560 struct xfs_inode *ip,
561 struct iattr *iattr)
562{
563 struct inode *inode = VFS_I(ip);
564
565 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
566
567 if (iattr->ia_valid & ATTR_ATIME)
568 inode->i_atime = iattr->ia_atime;
569 if (iattr->ia_valid & ATTR_CTIME)
570 inode->i_ctime = iattr->ia_ctime;
571 if (iattr->ia_valid & ATTR_MTIME)
572 inode->i_mtime = iattr->ia_mtime;
573}
574
575static int
576xfs_vn_change_ok(
577 struct dentry *dentry,
578 struct iattr *iattr)
579{
580 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
581
582 if (mp->m_flags & XFS_MOUNT_RDONLY)
583 return -EROFS;
584
585 if (XFS_FORCED_SHUTDOWN(mp))
586 return -EIO;
587
588 return setattr_prepare(dentry, iattr);
589}
590
591/*
592 * Set non-size attributes of an inode.
593 *
594 * Caution: The caller of this function is responsible for calling
595 * setattr_prepare() or otherwise verifying the change is fine.
596 */
597int
598xfs_setattr_nonsize(
599 struct xfs_inode *ip,
600 struct iattr *iattr,
601 int flags)
602{
603 xfs_mount_t *mp = ip->i_mount;
604 struct inode *inode = VFS_I(ip);
605 int mask = iattr->ia_valid;
606 xfs_trans_t *tp;
607 int error;
608 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
609 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
610 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
611 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
612
613 ASSERT((mask & ATTR_SIZE) == 0);
614
615 /*
616 * If disk quotas is on, we make sure that the dquots do exist on disk,
617 * before we start any other transactions. Trying to do this later
618 * is messy. We don't care to take a readlock to look at the ids
619 * in inode here, because we can't hold it across the trans_reserve.
620 * If the IDs do change before we take the ilock, we're covered
621 * because the i_*dquot fields will get updated anyway.
622 */
623 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
624 uint qflags = 0;
625
626 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
627 uid = iattr->ia_uid;
628 qflags |= XFS_QMOPT_UQUOTA;
629 } else {
630 uid = inode->i_uid;
631 }
632 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
633 gid = iattr->ia_gid;
634 qflags |= XFS_QMOPT_GQUOTA;
635 } else {
636 gid = inode->i_gid;
637 }
638
639 /*
640 * We take a reference when we initialize udqp and gdqp,
641 * so it is important that we never blindly double trip on
642 * the same variable. See xfs_create() for an example.
643 */
644 ASSERT(udqp == NULL);
645 ASSERT(gdqp == NULL);
646 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
647 xfs_kgid_to_gid(gid),
648 xfs_get_projid(ip),
649 qflags, &udqp, &gdqp, NULL);
650 if (error)
651 return error;
652 }
653
654 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
655 if (error)
656 goto out_dqrele;
657
658 xfs_ilock(ip, XFS_ILOCK_EXCL);
659 xfs_trans_ijoin(tp, ip, 0);
660
661 /*
662 * Change file ownership. Must be the owner or privileged.
663 */
664 if (mask & (ATTR_UID|ATTR_GID)) {
665 /*
666 * These IDs could have changed since we last looked at them.
667 * But, we're assured that if the ownership did change
668 * while we didn't have the inode locked, inode's dquot(s)
669 * would have changed also.
670 */
671 iuid = inode->i_uid;
672 igid = inode->i_gid;
673 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
674 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
675
676 /*
677 * Do a quota reservation only if uid/gid is actually
678 * going to change.
679 */
680 if (XFS_IS_QUOTA_RUNNING(mp) &&
681 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
682 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
683 ASSERT(tp);
684 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
685 NULL, capable(CAP_FOWNER) ?
686 XFS_QMOPT_FORCE_RES : 0);
687 if (error) /* out of quota */
688 goto out_cancel;
689 }
690 }
691
692 /*
693 * Change file ownership. Must be the owner or privileged.
694 */
695 if (mask & (ATTR_UID|ATTR_GID)) {
696 /*
697 * CAP_FSETID overrides the following restrictions:
698 *
699 * The set-user-ID and set-group-ID bits of a file will be
700 * cleared upon successful return from chown()
701 */
702 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
703 !capable(CAP_FSETID))
704 inode->i_mode &= ~(S_ISUID|S_ISGID);
705
706 /*
707 * Change the ownerships and register quota modifications
708 * in the transaction.
709 */
710 if (!uid_eq(iuid, uid)) {
711 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
712 ASSERT(mask & ATTR_UID);
713 ASSERT(udqp);
714 olddquot1 = xfs_qm_vop_chown(tp, ip,
715 &ip->i_udquot, udqp);
716 }
717 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
718 inode->i_uid = uid;
719 }
720 if (!gid_eq(igid, gid)) {
721 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
722 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
723 !XFS_IS_PQUOTA_ON(mp));
724 ASSERT(mask & ATTR_GID);
725 ASSERT(gdqp);
726 olddquot2 = xfs_qm_vop_chown(tp, ip,
727 &ip->i_gdquot, gdqp);
728 }
729 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
730 inode->i_gid = gid;
731 }
732 }
733
734 if (mask & ATTR_MODE)
735 xfs_setattr_mode(ip, iattr);
736 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
737 xfs_setattr_time(ip, iattr);
738
739 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
740
741 XFS_STATS_INC(mp, xs_ig_attrchg);
742
743 if (mp->m_flags & XFS_MOUNT_WSYNC)
744 xfs_trans_set_sync(tp);
745 error = xfs_trans_commit(tp);
746
747 xfs_iunlock(ip, XFS_ILOCK_EXCL);
748
749 /*
750 * Release any dquot(s) the inode had kept before chown.
751 */
752 xfs_qm_dqrele(olddquot1);
753 xfs_qm_dqrele(olddquot2);
754 xfs_qm_dqrele(udqp);
755 xfs_qm_dqrele(gdqp);
756
757 if (error)
758 return error;
759
760 /*
761 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
762 * update. We could avoid this with linked transactions
763 * and passing down the transaction pointer all the way
764 * to attr_set. No previous user of the generic
765 * Posix ACL code seems to care about this issue either.
766 */
767 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
768 error = posix_acl_chmod(inode, inode->i_mode);
769 if (error)
770 return error;
771 }
772
773 return 0;
774
775out_cancel:
776 xfs_trans_cancel(tp);
777out_dqrele:
778 xfs_qm_dqrele(udqp);
779 xfs_qm_dqrele(gdqp);
780 return error;
781}
782
783int
784xfs_vn_setattr_nonsize(
785 struct dentry *dentry,
786 struct iattr *iattr)
787{
788 struct xfs_inode *ip = XFS_I(d_inode(dentry));
789 int error;
790
791 trace_xfs_setattr(ip);
792
793 error = xfs_vn_change_ok(dentry, iattr);
794 if (error)
795 return error;
796 return xfs_setattr_nonsize(ip, iattr, 0);
797}
798
799/*
800 * Truncate file. Must have write permission and not be a directory.
801 *
802 * Caution: The caller of this function is responsible for calling
803 * setattr_prepare() or otherwise verifying the change is fine.
804 */
805int
806xfs_setattr_size(
807 struct xfs_inode *ip,
808 struct iattr *iattr)
809{
810 struct xfs_mount *mp = ip->i_mount;
811 struct inode *inode = VFS_I(ip);
812 xfs_off_t oldsize, newsize;
813 struct xfs_trans *tp;
814 int error;
815 uint lock_flags = 0;
816 bool did_zeroing = false;
817
818 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
819 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
820 ASSERT(S_ISREG(inode->i_mode));
821 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
822 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
823
824 oldsize = inode->i_size;
825 newsize = iattr->ia_size;
826
827 /*
828 * Short circuit the truncate case for zero length files.
829 */
830 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
831 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
832 return 0;
833
834 /*
835 * Use the regular setattr path to update the timestamps.
836 */
837 iattr->ia_valid &= ~ATTR_SIZE;
838 return xfs_setattr_nonsize(ip, iattr, 0);
839 }
840
841 /*
842 * Make sure that the dquots are attached to the inode.
843 */
844 error = xfs_qm_dqattach(ip, 0);
845 if (error)
846 return error;
847
848 /*
849 * Wait for all direct I/O to complete.
850 */
851 inode_dio_wait(inode);
852
853 /*
854 * File data changes must be complete before we start the transaction to
855 * modify the inode. This needs to be done before joining the inode to
856 * the transaction because the inode cannot be unlocked once it is a
857 * part of the transaction.
858 *
859 * Start with zeroing any data beyond EOF that we may expose on file
860 * extension, or zeroing out the rest of the block on a downward
861 * truncate.
862 */
863 if (newsize > oldsize) {
864 error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
865 } else {
866 error = iomap_truncate_page(inode, newsize, &did_zeroing,
867 &xfs_iomap_ops);
868 }
869
870 if (error)
871 return error;
872
873 /*
874 * We are going to log the inode size change in this transaction so
875 * any previous writes that are beyond the on disk EOF and the new
876 * EOF that have not been written out need to be written here. If we
877 * do not write the data out, we expose ourselves to the null files
878 * problem. Note that this includes any block zeroing we did above;
879 * otherwise those blocks may not be zeroed after a crash.
880 */
881 if (did_zeroing ||
882 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
883 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
884 ip->i_d.di_size, newsize);
885 if (error)
886 return error;
887 }
888
889 /*
890 * We've already locked out new page faults, so now we can safely remove
891 * pages from the page cache knowing they won't get refaulted until we
892 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
893 * complete. The truncate_setsize() call also cleans partial EOF page
894 * PTEs on extending truncates and hence ensures sub-page block size
895 * filesystems are correctly handled, too.
896 *
897 * We have to do all the page cache truncate work outside the
898 * transaction context as the "lock" order is page lock->log space
899 * reservation as defined by extent allocation in the writeback path.
900 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
901 * having already truncated the in-memory version of the file (i.e. made
902 * user visible changes). There's not much we can do about this, except
903 * to hope that the caller sees ENOMEM and retries the truncate
904 * operation.
905 */
906 truncate_setsize(inode, newsize);
907
908 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
909 if (error)
910 return error;
911
912 lock_flags |= XFS_ILOCK_EXCL;
913 xfs_ilock(ip, XFS_ILOCK_EXCL);
914 xfs_trans_ijoin(tp, ip, 0);
915
916 /*
917 * Only change the c/mtime if we are changing the size or we are
918 * explicitly asked to change it. This handles the semantic difference
919 * between truncate() and ftruncate() as implemented in the VFS.
920 *
921 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
922 * special case where we need to update the times despite not having
923 * these flags set. For all other operations the VFS set these flags
924 * explicitly if it wants a timestamp update.
925 */
926 if (newsize != oldsize &&
927 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
928 iattr->ia_ctime = iattr->ia_mtime =
929 current_time(inode);
930 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
931 }
932
933 /*
934 * The first thing we do is set the size to new_size permanently on
935 * disk. This way we don't have to worry about anyone ever being able
936 * to look at the data being freed even in the face of a crash.
937 * What we're getting around here is the case where we free a block, it
938 * is allocated to another file, it is written to, and then we crash.
939 * If the new data gets written to the file but the log buffers
940 * containing the free and reallocation don't, then we'd end up with
941 * garbage in the blocks being freed. As long as we make the new size
942 * permanent before actually freeing any blocks it doesn't matter if
943 * they get written to.
944 */
945 ip->i_d.di_size = newsize;
946 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
947
948 if (newsize <= oldsize) {
949 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
950 if (error)
951 goto out_trans_cancel;
952
953 /*
954 * Truncated "down", so we're removing references to old data
955 * here - if we delay flushing for a long time, we expose
956 * ourselves unduly to the notorious NULL files problem. So,
957 * we mark this inode and flush it when the file is closed,
958 * and do not wait the usual (long) time for writeout.
959 */
960 xfs_iflags_set(ip, XFS_ITRUNCATED);
961
962 /* A truncate down always removes post-EOF blocks. */
963 xfs_inode_clear_eofblocks_tag(ip);
964 }
965
966 if (iattr->ia_valid & ATTR_MODE)
967 xfs_setattr_mode(ip, iattr);
968 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
969 xfs_setattr_time(ip, iattr);
970
971 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
972
973 XFS_STATS_INC(mp, xs_ig_attrchg);
974
975 if (mp->m_flags & XFS_MOUNT_WSYNC)
976 xfs_trans_set_sync(tp);
977
978 error = xfs_trans_commit(tp);
979out_unlock:
980 if (lock_flags)
981 xfs_iunlock(ip, lock_flags);
982 return error;
983
984out_trans_cancel:
985 xfs_trans_cancel(tp);
986 goto out_unlock;
987}
988
989int
990xfs_vn_setattr_size(
991 struct dentry *dentry,
992 struct iattr *iattr)
993{
994 struct xfs_inode *ip = XFS_I(d_inode(dentry));
995 int error;
996
997 trace_xfs_setattr(ip);
998
999 error = xfs_vn_change_ok(dentry, iattr);
1000 if (error)
1001 return error;
1002 return xfs_setattr_size(ip, iattr);
1003}
1004
1005STATIC int
1006xfs_vn_setattr(
1007 struct dentry *dentry,
1008 struct iattr *iattr)
1009{
1010 int error;
1011
1012 if (iattr->ia_valid & ATTR_SIZE) {
1013 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1014 uint iolock = XFS_IOLOCK_EXCL;
1015
1016 error = xfs_break_layouts(d_inode(dentry), &iolock);
1017 if (error)
1018 return error;
1019
1020 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1021 error = xfs_vn_setattr_size(dentry, iattr);
1022 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1023 } else {
1024 error = xfs_vn_setattr_nonsize(dentry, iattr);
1025 }
1026
1027 return error;
1028}
1029
1030STATIC int
1031xfs_vn_update_time(
1032 struct inode *inode,
1033 struct timespec *now,
1034 int flags)
1035{
1036 struct xfs_inode *ip = XFS_I(inode);
1037 struct xfs_mount *mp = ip->i_mount;
1038 struct xfs_trans *tp;
1039 int error;
1040
1041 trace_xfs_update_time(ip);
1042
1043 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1044 if (error)
1045 return error;
1046
1047 xfs_ilock(ip, XFS_ILOCK_EXCL);
1048 if (flags & S_CTIME)
1049 inode->i_ctime = *now;
1050 if (flags & S_MTIME)
1051 inode->i_mtime = *now;
1052 if (flags & S_ATIME)
1053 inode->i_atime = *now;
1054
1055 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1056 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1057 return xfs_trans_commit(tp);
1058}
1059
1060STATIC int
1061xfs_vn_fiemap(
1062 struct inode *inode,
1063 struct fiemap_extent_info *fieinfo,
1064 u64 start,
1065 u64 length)
1066{
1067 int error;
1068
1069 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1070 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1071 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1072 error = iomap_fiemap(inode, fieinfo, start, length,
1073 &xfs_xattr_iomap_ops);
1074 } else {
1075 error = iomap_fiemap(inode, fieinfo, start, length,
1076 &xfs_iomap_ops);
1077 }
1078 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1079
1080 return error;
1081}
1082
1083STATIC int
1084xfs_vn_tmpfile(
1085 struct inode *dir,
1086 struct dentry *dentry,
1087 umode_t mode)
1088{
1089 return xfs_generic_create(dir, dentry, mode, 0, true);
1090}
1091
1092static const struct inode_operations xfs_inode_operations = {
1093 .get_acl = xfs_get_acl,
1094 .set_acl = xfs_set_acl,
1095 .getattr = xfs_vn_getattr,
1096 .setattr = xfs_vn_setattr,
1097 .listxattr = xfs_vn_listxattr,
1098 .fiemap = xfs_vn_fiemap,
1099 .update_time = xfs_vn_update_time,
1100};
1101
1102static const struct inode_operations xfs_dir_inode_operations = {
1103 .create = xfs_vn_create,
1104 .lookup = xfs_vn_lookup,
1105 .link = xfs_vn_link,
1106 .unlink = xfs_vn_unlink,
1107 .symlink = xfs_vn_symlink,
1108 .mkdir = xfs_vn_mkdir,
1109 /*
1110 * Yes, XFS uses the same method for rmdir and unlink.
1111 *
1112 * There are some subtile differences deeper in the code,
1113 * but we use S_ISDIR to check for those.
1114 */
1115 .rmdir = xfs_vn_unlink,
1116 .mknod = xfs_vn_mknod,
1117 .rename = xfs_vn_rename,
1118 .get_acl = xfs_get_acl,
1119 .set_acl = xfs_set_acl,
1120 .getattr = xfs_vn_getattr,
1121 .setattr = xfs_vn_setattr,
1122 .listxattr = xfs_vn_listxattr,
1123 .update_time = xfs_vn_update_time,
1124 .tmpfile = xfs_vn_tmpfile,
1125};
1126
1127static const struct inode_operations xfs_dir_ci_inode_operations = {
1128 .create = xfs_vn_create,
1129 .lookup = xfs_vn_ci_lookup,
1130 .link = xfs_vn_link,
1131 .unlink = xfs_vn_unlink,
1132 .symlink = xfs_vn_symlink,
1133 .mkdir = xfs_vn_mkdir,
1134 /*
1135 * Yes, XFS uses the same method for rmdir and unlink.
1136 *
1137 * There are some subtile differences deeper in the code,
1138 * but we use S_ISDIR to check for those.
1139 */
1140 .rmdir = xfs_vn_unlink,
1141 .mknod = xfs_vn_mknod,
1142 .rename = xfs_vn_rename,
1143 .get_acl = xfs_get_acl,
1144 .set_acl = xfs_set_acl,
1145 .getattr = xfs_vn_getattr,
1146 .setattr = xfs_vn_setattr,
1147 .listxattr = xfs_vn_listxattr,
1148 .update_time = xfs_vn_update_time,
1149 .tmpfile = xfs_vn_tmpfile,
1150};
1151
1152static const struct inode_operations xfs_symlink_inode_operations = {
1153 .get_link = xfs_vn_get_link,
1154 .getattr = xfs_vn_getattr,
1155 .setattr = xfs_vn_setattr,
1156 .listxattr = xfs_vn_listxattr,
1157 .update_time = xfs_vn_update_time,
1158};
1159
1160static const struct inode_operations xfs_inline_symlink_inode_operations = {
1161 .get_link = xfs_vn_get_link_inline,
1162 .getattr = xfs_vn_getattr,
1163 .setattr = xfs_vn_setattr,
1164 .listxattr = xfs_vn_listxattr,
1165 .update_time = xfs_vn_update_time,
1166};
1167
1168STATIC void
1169xfs_diflags_to_iflags(
1170 struct inode *inode,
1171 struct xfs_inode *ip)
1172{
1173 uint16_t flags = ip->i_d.di_flags;
1174
1175 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1176 S_NOATIME | S_DAX);
1177
1178 if (flags & XFS_DIFLAG_IMMUTABLE)
1179 inode->i_flags |= S_IMMUTABLE;
1180 if (flags & XFS_DIFLAG_APPEND)
1181 inode->i_flags |= S_APPEND;
1182 if (flags & XFS_DIFLAG_SYNC)
1183 inode->i_flags |= S_SYNC;
1184 if (flags & XFS_DIFLAG_NOATIME)
1185 inode->i_flags |= S_NOATIME;
1186 if (S_ISREG(inode->i_mode) &&
1187 ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
1188 !xfs_is_reflink_inode(ip) &&
1189 (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
1190 ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1191 inode->i_flags |= S_DAX;
1192}
1193
1194/*
1195 * Initialize the Linux inode.
1196 *
1197 * When reading existing inodes from disk this is called directly from xfs_iget,
1198 * when creating a new inode it is called from xfs_ialloc after setting up the
1199 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1200 * it up to the caller to deal with unlocking the inode appropriately.
1201 */
1202void
1203xfs_setup_inode(
1204 struct xfs_inode *ip)
1205{
1206 struct inode *inode = &ip->i_vnode;
1207 gfp_t gfp_mask;
1208
1209 inode->i_ino = ip->i_ino;
1210 inode->i_state = I_NEW;
1211
1212 inode_sb_list_add(inode);
1213 /* make the inode look hashed for the writeback code */
1214 hlist_add_fake(&inode->i_hash);
1215
1216 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1217 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1218
1219 switch (inode->i_mode & S_IFMT) {
1220 case S_IFBLK:
1221 case S_IFCHR:
1222 inode->i_rdev =
1223 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1224 sysv_minor(ip->i_df.if_u2.if_rdev));
1225 break;
1226 default:
1227 inode->i_rdev = 0;
1228 break;
1229 }
1230
1231 i_size_write(inode, ip->i_d.di_size);
1232 xfs_diflags_to_iflags(inode, ip);
1233
1234 if (S_ISDIR(inode->i_mode)) {
1235 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1236 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1237 } else {
1238 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1239 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1240 }
1241
1242 /*
1243 * Ensure all page cache allocations are done from GFP_NOFS context to
1244 * prevent direct reclaim recursion back into the filesystem and blowing
1245 * stacks or deadlocking.
1246 */
1247 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1248 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1249
1250 /*
1251 * If there is no attribute fork no ACL can exist on this inode,
1252 * and it can't have any file capabilities attached to it either.
1253 */
1254 if (!XFS_IFORK_Q(ip)) {
1255 inode_has_no_xattr(inode);
1256 cache_no_acl(inode);
1257 }
1258}
1259
1260void
1261xfs_setup_iops(
1262 struct xfs_inode *ip)
1263{
1264 struct inode *inode = &ip->i_vnode;
1265
1266 switch (inode->i_mode & S_IFMT) {
1267 case S_IFREG:
1268 inode->i_op = &xfs_inode_operations;
1269 inode->i_fop = &xfs_file_operations;
1270 inode->i_mapping->a_ops = &xfs_address_space_operations;
1271 break;
1272 case S_IFDIR:
1273 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1274 inode->i_op = &xfs_dir_ci_inode_operations;
1275 else
1276 inode->i_op = &xfs_dir_inode_operations;
1277 inode->i_fop = &xfs_dir_file_operations;
1278 break;
1279 case S_IFLNK:
1280 if (ip->i_df.if_flags & XFS_IFINLINE)
1281 inode->i_op = &xfs_inline_symlink_inode_operations;
1282 else
1283 inode->i_op = &xfs_symlink_inode_operations;
1284 break;
1285 default:
1286 inode->i_op = &xfs_inode_operations;
1287 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1288 break;
1289 }
1290}
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_trace.h"
21#include "xfs_icache.h"
22#include "xfs_symlink.h"
23#include "xfs_dir2.h"
24#include "xfs_iomap.h"
25#include "xfs_error.h"
26#include "xfs_ioctl.h"
27#include "xfs_xattr.h"
28
29#include <linux/posix_acl.h>
30#include <linux/security.h>
31#include <linux/iversion.h>
32#include <linux/fiemap.h>
33
34/*
35 * Directories have different lock order w.r.t. mmap_lock compared to regular
36 * files. This is due to readdir potentially triggering page faults on a user
37 * buffer inside filldir(), and this happens with the ilock on the directory
38 * held. For regular files, the lock order is the other way around - the
39 * mmap_lock is taken during the page fault, and then we lock the ilock to do
40 * block mapping. Hence we need a different class for the directory ilock so
41 * that lockdep can tell them apart.
42 */
43static struct lock_class_key xfs_nondir_ilock_class;
44static struct lock_class_key xfs_dir_ilock_class;
45
46static int
47xfs_initxattrs(
48 struct inode *inode,
49 const struct xattr *xattr_array,
50 void *fs_info)
51{
52 const struct xattr *xattr;
53 struct xfs_inode *ip = XFS_I(inode);
54 int error = 0;
55
56 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
57 struct xfs_da_args args = {
58 .dp = ip,
59 .attr_filter = XFS_ATTR_SECURE,
60 .name = xattr->name,
61 .namelen = strlen(xattr->name),
62 .value = xattr->value,
63 .valuelen = xattr->value_len,
64 };
65 error = xfs_attr_change(&args);
66 if (error < 0)
67 break;
68 }
69 return error;
70}
71
72/*
73 * Hook in SELinux. This is not quite correct yet, what we really need
74 * here (as we do for default ACLs) is a mechanism by which creation of
75 * these attrs can be journalled at inode creation time (along with the
76 * inode, of course, such that log replay can't cause these to be lost).
77 */
78int
79xfs_inode_init_security(
80 struct inode *inode,
81 struct inode *dir,
82 const struct qstr *qstr)
83{
84 return security_inode_init_security(inode, dir, qstr,
85 &xfs_initxattrs, NULL);
86}
87
88static void
89xfs_dentry_to_name(
90 struct xfs_name *namep,
91 struct dentry *dentry)
92{
93 namep->name = dentry->d_name.name;
94 namep->len = dentry->d_name.len;
95 namep->type = XFS_DIR3_FT_UNKNOWN;
96}
97
98static int
99xfs_dentry_mode_to_name(
100 struct xfs_name *namep,
101 struct dentry *dentry,
102 int mode)
103{
104 namep->name = dentry->d_name.name;
105 namep->len = dentry->d_name.len;
106 namep->type = xfs_mode_to_ftype(mode);
107
108 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
109 return -EFSCORRUPTED;
110
111 return 0;
112}
113
114STATIC void
115xfs_cleanup_inode(
116 struct inode *dir,
117 struct inode *inode,
118 struct dentry *dentry)
119{
120 struct xfs_name teardown;
121
122 /* Oh, the horror.
123 * If we can't add the ACL or we fail in
124 * xfs_inode_init_security we must back out.
125 * ENOSPC can hit here, among other things.
126 */
127 xfs_dentry_to_name(&teardown, dentry);
128
129 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
130}
131
132/*
133 * Check to see if we are likely to need an extended attribute to be added to
134 * the inode we are about to allocate. This allows the attribute fork to be
135 * created during the inode allocation, reducing the number of transactions we
136 * need to do in this fast path.
137 *
138 * The security checks are optimistic, but not guaranteed. The two LSMs that
139 * require xattrs to be added here (selinux and smack) are also the only two
140 * LSMs that add a sb->s_security structure to the superblock. Hence if security
141 * is enabled and sb->s_security is set, we have a pretty good idea that we are
142 * going to be asked to add a security xattr immediately after allocating the
143 * xfs inode and instantiating the VFS inode.
144 */
145static inline bool
146xfs_create_need_xattr(
147 struct inode *dir,
148 struct posix_acl *default_acl,
149 struct posix_acl *acl)
150{
151 if (acl)
152 return true;
153 if (default_acl)
154 return true;
155#if IS_ENABLED(CONFIG_SECURITY)
156 if (dir->i_sb->s_security)
157 return true;
158#endif
159 return false;
160}
161
162
163STATIC int
164xfs_generic_create(
165 struct user_namespace *mnt_userns,
166 struct inode *dir,
167 struct dentry *dentry,
168 umode_t mode,
169 dev_t rdev,
170 struct file *tmpfile) /* unnamed file */
171{
172 struct inode *inode;
173 struct xfs_inode *ip = NULL;
174 struct posix_acl *default_acl, *acl;
175 struct xfs_name name;
176 int error;
177
178 /*
179 * Irix uses Missed'em'V split, but doesn't want to see
180 * the upper 5 bits of (14bit) major.
181 */
182 if (S_ISCHR(mode) || S_ISBLK(mode)) {
183 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
184 return -EINVAL;
185 } else {
186 rdev = 0;
187 }
188
189 error = posix_acl_create(dir, &mode, &default_acl, &acl);
190 if (error)
191 return error;
192
193 /* Verify mode is valid also for tmpfile case */
194 error = xfs_dentry_mode_to_name(&name, dentry, mode);
195 if (unlikely(error))
196 goto out_free_acl;
197
198 if (!tmpfile) {
199 error = xfs_create(mnt_userns, XFS_I(dir), &name, mode, rdev,
200 xfs_create_need_xattr(dir, default_acl, acl),
201 &ip);
202 } else {
203 error = xfs_create_tmpfile(mnt_userns, XFS_I(dir), mode, &ip);
204 }
205 if (unlikely(error))
206 goto out_free_acl;
207
208 inode = VFS_I(ip);
209
210 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
211 if (unlikely(error))
212 goto out_cleanup_inode;
213
214 if (default_acl) {
215 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
216 if (error)
217 goto out_cleanup_inode;
218 }
219 if (acl) {
220 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
221 if (error)
222 goto out_cleanup_inode;
223 }
224
225 xfs_setup_iops(ip);
226
227 if (tmpfile) {
228 /*
229 * The VFS requires that any inode fed to d_tmpfile must have
230 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
231 * However, we created the temp file with nlink == 0 because
232 * we're not allowed to put an inode with nlink > 0 on the
233 * unlinked list. Therefore we have to set nlink to 1 so that
234 * d_tmpfile can immediately set it back to zero.
235 */
236 set_nlink(inode, 1);
237 d_tmpfile(tmpfile, inode);
238 } else
239 d_instantiate(dentry, inode);
240
241 xfs_finish_inode_setup(ip);
242
243 out_free_acl:
244 posix_acl_release(default_acl);
245 posix_acl_release(acl);
246 return error;
247
248 out_cleanup_inode:
249 xfs_finish_inode_setup(ip);
250 if (!tmpfile)
251 xfs_cleanup_inode(dir, inode, dentry);
252 xfs_irele(ip);
253 goto out_free_acl;
254}
255
256STATIC int
257xfs_vn_mknod(
258 struct user_namespace *mnt_userns,
259 struct inode *dir,
260 struct dentry *dentry,
261 umode_t mode,
262 dev_t rdev)
263{
264 return xfs_generic_create(mnt_userns, dir, dentry, mode, rdev, NULL);
265}
266
267STATIC int
268xfs_vn_create(
269 struct user_namespace *mnt_userns,
270 struct inode *dir,
271 struct dentry *dentry,
272 umode_t mode,
273 bool flags)
274{
275 return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, NULL);
276}
277
278STATIC int
279xfs_vn_mkdir(
280 struct user_namespace *mnt_userns,
281 struct inode *dir,
282 struct dentry *dentry,
283 umode_t mode)
284{
285 return xfs_generic_create(mnt_userns, dir, dentry, mode | S_IFDIR, 0,
286 NULL);
287}
288
289STATIC struct dentry *
290xfs_vn_lookup(
291 struct inode *dir,
292 struct dentry *dentry,
293 unsigned int flags)
294{
295 struct inode *inode;
296 struct xfs_inode *cip;
297 struct xfs_name name;
298 int error;
299
300 if (dentry->d_name.len >= MAXNAMELEN)
301 return ERR_PTR(-ENAMETOOLONG);
302
303 xfs_dentry_to_name(&name, dentry);
304 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
305 if (likely(!error))
306 inode = VFS_I(cip);
307 else if (likely(error == -ENOENT))
308 inode = NULL;
309 else
310 inode = ERR_PTR(error);
311 return d_splice_alias(inode, dentry);
312}
313
314STATIC struct dentry *
315xfs_vn_ci_lookup(
316 struct inode *dir,
317 struct dentry *dentry,
318 unsigned int flags)
319{
320 struct xfs_inode *ip;
321 struct xfs_name xname;
322 struct xfs_name ci_name;
323 struct qstr dname;
324 int error;
325
326 if (dentry->d_name.len >= MAXNAMELEN)
327 return ERR_PTR(-ENAMETOOLONG);
328
329 xfs_dentry_to_name(&xname, dentry);
330 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
331 if (unlikely(error)) {
332 if (unlikely(error != -ENOENT))
333 return ERR_PTR(error);
334 /*
335 * call d_add(dentry, NULL) here when d_drop_negative_children
336 * is called in xfs_vn_mknod (ie. allow negative dentries
337 * with CI filesystems).
338 */
339 return NULL;
340 }
341
342 /* if exact match, just splice and exit */
343 if (!ci_name.name)
344 return d_splice_alias(VFS_I(ip), dentry);
345
346 /* else case-insensitive match... */
347 dname.name = ci_name.name;
348 dname.len = ci_name.len;
349 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
350 kmem_free(ci_name.name);
351 return dentry;
352}
353
354STATIC int
355xfs_vn_link(
356 struct dentry *old_dentry,
357 struct inode *dir,
358 struct dentry *dentry)
359{
360 struct inode *inode = d_inode(old_dentry);
361 struct xfs_name name;
362 int error;
363
364 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
365 if (unlikely(error))
366 return error;
367
368 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
369 if (unlikely(error))
370 return error;
371
372 ihold(inode);
373 d_instantiate(dentry, inode);
374 return 0;
375}
376
377STATIC int
378xfs_vn_unlink(
379 struct inode *dir,
380 struct dentry *dentry)
381{
382 struct xfs_name name;
383 int error;
384
385 xfs_dentry_to_name(&name, dentry);
386
387 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
388 if (error)
389 return error;
390
391 /*
392 * With unlink, the VFS makes the dentry "negative": no inode,
393 * but still hashed. This is incompatible with case-insensitive
394 * mode, so invalidate (unhash) the dentry in CI-mode.
395 */
396 if (xfs_has_asciici(XFS_M(dir->i_sb)))
397 d_invalidate(dentry);
398 return 0;
399}
400
401STATIC int
402xfs_vn_symlink(
403 struct user_namespace *mnt_userns,
404 struct inode *dir,
405 struct dentry *dentry,
406 const char *symname)
407{
408 struct inode *inode;
409 struct xfs_inode *cip = NULL;
410 struct xfs_name name;
411 int error;
412 umode_t mode;
413
414 mode = S_IFLNK |
415 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
416 error = xfs_dentry_mode_to_name(&name, dentry, mode);
417 if (unlikely(error))
418 goto out;
419
420 error = xfs_symlink(mnt_userns, XFS_I(dir), &name, symname, mode, &cip);
421 if (unlikely(error))
422 goto out;
423
424 inode = VFS_I(cip);
425
426 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
427 if (unlikely(error))
428 goto out_cleanup_inode;
429
430 xfs_setup_iops(cip);
431
432 d_instantiate(dentry, inode);
433 xfs_finish_inode_setup(cip);
434 return 0;
435
436 out_cleanup_inode:
437 xfs_finish_inode_setup(cip);
438 xfs_cleanup_inode(dir, inode, dentry);
439 xfs_irele(cip);
440 out:
441 return error;
442}
443
444STATIC int
445xfs_vn_rename(
446 struct user_namespace *mnt_userns,
447 struct inode *odir,
448 struct dentry *odentry,
449 struct inode *ndir,
450 struct dentry *ndentry,
451 unsigned int flags)
452{
453 struct inode *new_inode = d_inode(ndentry);
454 int omode = 0;
455 int error;
456 struct xfs_name oname;
457 struct xfs_name nname;
458
459 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
460 return -EINVAL;
461
462 /* if we are exchanging files, we need to set i_mode of both files */
463 if (flags & RENAME_EXCHANGE)
464 omode = d_inode(ndentry)->i_mode;
465
466 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
467 if (omode && unlikely(error))
468 return error;
469
470 error = xfs_dentry_mode_to_name(&nname, ndentry,
471 d_inode(odentry)->i_mode);
472 if (unlikely(error))
473 return error;
474
475 return xfs_rename(mnt_userns, XFS_I(odir), &oname,
476 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
477 new_inode ? XFS_I(new_inode) : NULL, flags);
478}
479
480/*
481 * careful here - this function can get called recursively, so
482 * we need to be very careful about how much stack we use.
483 * uio is kmalloced for this reason...
484 */
485STATIC const char *
486xfs_vn_get_link(
487 struct dentry *dentry,
488 struct inode *inode,
489 struct delayed_call *done)
490{
491 char *link;
492 int error = -ENOMEM;
493
494 if (!dentry)
495 return ERR_PTR(-ECHILD);
496
497 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
498 if (!link)
499 goto out_err;
500
501 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
502 if (unlikely(error))
503 goto out_kfree;
504
505 set_delayed_call(done, kfree_link, link);
506 return link;
507
508 out_kfree:
509 kfree(link);
510 out_err:
511 return ERR_PTR(error);
512}
513
514static uint32_t
515xfs_stat_blksize(
516 struct xfs_inode *ip)
517{
518 struct xfs_mount *mp = ip->i_mount;
519
520 /*
521 * If the file blocks are being allocated from a realtime volume, then
522 * always return the realtime extent size.
523 */
524 if (XFS_IS_REALTIME_INODE(ip))
525 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip));
526
527 /*
528 * Allow large block sizes to be reported to userspace programs if the
529 * "largeio" mount option is used.
530 *
531 * If compatibility mode is specified, simply return the basic unit of
532 * caching so that we don't get inefficient read/modify/write I/O from
533 * user apps. Otherwise....
534 *
535 * If the underlying volume is a stripe, then return the stripe width in
536 * bytes as the recommended I/O size. It is not a stripe and we've set a
537 * default buffered I/O size, return that, otherwise return the compat
538 * default.
539 */
540 if (xfs_has_large_iosize(mp)) {
541 if (mp->m_swidth)
542 return XFS_FSB_TO_B(mp, mp->m_swidth);
543 if (xfs_has_allocsize(mp))
544 return 1U << mp->m_allocsize_log;
545 }
546
547 return PAGE_SIZE;
548}
549
550STATIC int
551xfs_vn_getattr(
552 struct user_namespace *mnt_userns,
553 const struct path *path,
554 struct kstat *stat,
555 u32 request_mask,
556 unsigned int query_flags)
557{
558 struct inode *inode = d_inode(path->dentry);
559 struct xfs_inode *ip = XFS_I(inode);
560 struct xfs_mount *mp = ip->i_mount;
561 vfsuid_t vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
562 vfsgid_t vfsgid = i_gid_into_vfsgid(mnt_userns, inode);
563
564 trace_xfs_getattr(ip);
565
566 if (xfs_is_shutdown(mp))
567 return -EIO;
568
569 stat->size = XFS_ISIZE(ip);
570 stat->dev = inode->i_sb->s_dev;
571 stat->mode = inode->i_mode;
572 stat->nlink = inode->i_nlink;
573 stat->uid = vfsuid_into_kuid(vfsuid);
574 stat->gid = vfsgid_into_kgid(vfsgid);
575 stat->ino = ip->i_ino;
576 stat->atime = inode->i_atime;
577 stat->mtime = inode->i_mtime;
578 stat->ctime = inode->i_ctime;
579 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
580
581 if (xfs_has_v3inodes(mp)) {
582 if (request_mask & STATX_BTIME) {
583 stat->result_mask |= STATX_BTIME;
584 stat->btime = ip->i_crtime;
585 }
586 }
587
588 /*
589 * Note: If you add another clause to set an attribute flag, please
590 * update attributes_mask below.
591 */
592 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
593 stat->attributes |= STATX_ATTR_IMMUTABLE;
594 if (ip->i_diflags & XFS_DIFLAG_APPEND)
595 stat->attributes |= STATX_ATTR_APPEND;
596 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
597 stat->attributes |= STATX_ATTR_NODUMP;
598
599 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
600 STATX_ATTR_APPEND |
601 STATX_ATTR_NODUMP);
602
603 switch (inode->i_mode & S_IFMT) {
604 case S_IFBLK:
605 case S_IFCHR:
606 stat->blksize = BLKDEV_IOSIZE;
607 stat->rdev = inode->i_rdev;
608 break;
609 case S_IFREG:
610 if (request_mask & STATX_DIOALIGN) {
611 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
612 struct block_device *bdev = target->bt_bdev;
613
614 stat->result_mask |= STATX_DIOALIGN;
615 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
616 stat->dio_offset_align = bdev_logical_block_size(bdev);
617 }
618 fallthrough;
619 default:
620 stat->blksize = xfs_stat_blksize(ip);
621 stat->rdev = 0;
622 break;
623 }
624
625 return 0;
626}
627
628static int
629xfs_vn_change_ok(
630 struct user_namespace *mnt_userns,
631 struct dentry *dentry,
632 struct iattr *iattr)
633{
634 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
635
636 if (xfs_is_readonly(mp))
637 return -EROFS;
638
639 if (xfs_is_shutdown(mp))
640 return -EIO;
641
642 return setattr_prepare(mnt_userns, 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 */
651static int
652xfs_setattr_nonsize(
653 struct user_namespace *mnt_userns,
654 struct dentry *dentry,
655 struct xfs_inode *ip,
656 struct iattr *iattr)
657{
658 xfs_mount_t *mp = ip->i_mount;
659 struct inode *inode = VFS_I(ip);
660 int mask = iattr->ia_valid;
661 xfs_trans_t *tp;
662 int error;
663 kuid_t uid = GLOBAL_ROOT_UID;
664 kgid_t gid = GLOBAL_ROOT_GID;
665 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
666 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL;
667
668 ASSERT((mask & ATTR_SIZE) == 0);
669
670 /*
671 * If disk quotas is on, we make sure that the dquots do exist on disk,
672 * before we start any other transactions. Trying to do this later
673 * is messy. We don't care to take a readlock to look at the ids
674 * in inode here, because we can't hold it across the trans_reserve.
675 * If the IDs do change before we take the ilock, we're covered
676 * because the i_*dquot fields will get updated anyway.
677 */
678 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
679 uint qflags = 0;
680
681 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
682 uid = from_vfsuid(mnt_userns, i_user_ns(inode),
683 iattr->ia_vfsuid);
684 qflags |= XFS_QMOPT_UQUOTA;
685 } else {
686 uid = inode->i_uid;
687 }
688 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
689 gid = from_vfsgid(mnt_userns, i_user_ns(inode),
690 iattr->ia_vfsgid);
691 qflags |= XFS_QMOPT_GQUOTA;
692 } else {
693 gid = inode->i_gid;
694 }
695
696 /*
697 * We take a reference when we initialize udqp and gdqp,
698 * so it is important that we never blindly double trip on
699 * the same variable. See xfs_create() for an example.
700 */
701 ASSERT(udqp == NULL);
702 ASSERT(gdqp == NULL);
703 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
704 qflags, &udqp, &gdqp, NULL);
705 if (error)
706 return error;
707 }
708
709 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
710 has_capability_noaudit(current, CAP_FOWNER), &tp);
711 if (error)
712 goto out_dqrele;
713
714 /*
715 * Register quota modifications in the transaction. Must be the owner
716 * or privileged. These IDs could have changed since we last looked at
717 * them. But, we're assured that if the ownership did change while we
718 * didn't have the inode locked, inode's dquot(s) would have changed
719 * also.
720 */
721 if (XFS_IS_UQUOTA_ON(mp) &&
722 i_uid_needs_update(mnt_userns, iattr, inode)) {
723 ASSERT(udqp);
724 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp);
725 }
726 if (XFS_IS_GQUOTA_ON(mp) &&
727 i_gid_needs_update(mnt_userns, iattr, inode)) {
728 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp));
729 ASSERT(gdqp);
730 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp);
731 }
732
733 setattr_copy(mnt_userns, inode, iattr);
734 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
735
736 XFS_STATS_INC(mp, xs_ig_attrchg);
737
738 if (xfs_has_wsync(mp))
739 xfs_trans_set_sync(tp);
740 error = xfs_trans_commit(tp);
741
742 /*
743 * Release any dquot(s) the inode had kept before chown.
744 */
745 xfs_qm_dqrele(old_udqp);
746 xfs_qm_dqrele(old_gdqp);
747 xfs_qm_dqrele(udqp);
748 xfs_qm_dqrele(gdqp);
749
750 if (error)
751 return error;
752
753 /*
754 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
755 * update. We could avoid this with linked transactions
756 * and passing down the transaction pointer all the way
757 * to attr_set. No previous user of the generic
758 * Posix ACL code seems to care about this issue either.
759 */
760 if (mask & ATTR_MODE) {
761 error = posix_acl_chmod(mnt_userns, dentry, inode->i_mode);
762 if (error)
763 return error;
764 }
765
766 return 0;
767
768out_dqrele:
769 xfs_qm_dqrele(udqp);
770 xfs_qm_dqrele(gdqp);
771 return error;
772}
773
774/*
775 * Truncate file. Must have write permission and not be a directory.
776 *
777 * Caution: The caller of this function is responsible for calling
778 * setattr_prepare() or otherwise verifying the change is fine.
779 */
780STATIC int
781xfs_setattr_size(
782 struct user_namespace *mnt_userns,
783 struct dentry *dentry,
784 struct xfs_inode *ip,
785 struct iattr *iattr)
786{
787 struct xfs_mount *mp = ip->i_mount;
788 struct inode *inode = VFS_I(ip);
789 xfs_off_t oldsize, newsize;
790 struct xfs_trans *tp;
791 int error;
792 uint lock_flags = 0;
793 bool did_zeroing = false;
794
795 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
796 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
797 ASSERT(S_ISREG(inode->i_mode));
798 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
799 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
800
801 oldsize = inode->i_size;
802 newsize = iattr->ia_size;
803
804 /*
805 * Short circuit the truncate case for zero length files.
806 */
807 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
808 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
809 return 0;
810
811 /*
812 * Use the regular setattr path to update the timestamps.
813 */
814 iattr->ia_valid &= ~ATTR_SIZE;
815 return xfs_setattr_nonsize(mnt_userns, dentry, ip, iattr);
816 }
817
818 /*
819 * Make sure that the dquots are attached to the inode.
820 */
821 error = xfs_qm_dqattach(ip);
822 if (error)
823 return error;
824
825 /*
826 * Wait for all direct I/O to complete.
827 */
828 inode_dio_wait(inode);
829
830 /*
831 * File data changes must be complete before we start the transaction to
832 * modify the inode. This needs to be done before joining the inode to
833 * the transaction because the inode cannot be unlocked once it is a
834 * part of the transaction.
835 *
836 * Start with zeroing any data beyond EOF that we may expose on file
837 * extension, or zeroing out the rest of the block on a downward
838 * truncate.
839 */
840 if (newsize > oldsize) {
841 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
842 error = xfs_zero_range(ip, oldsize, newsize - oldsize,
843 &did_zeroing);
844 } else {
845 /*
846 * iomap won't detect a dirty page over an unwritten block (or a
847 * cow block over a hole) and subsequently skips zeroing the
848 * newly post-EOF portion of the page. Flush the new EOF to
849 * convert the block before the pagecache truncate.
850 */
851 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
852 newsize);
853 if (error)
854 return error;
855 error = xfs_truncate_page(ip, newsize, &did_zeroing);
856 }
857
858 if (error)
859 return error;
860
861 /*
862 * We've already locked out new page faults, so now we can safely remove
863 * pages from the page cache knowing they won't get refaulted until we
864 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
865 * complete. The truncate_setsize() call also cleans partial EOF page
866 * PTEs on extending truncates and hence ensures sub-page block size
867 * filesystems are correctly handled, too.
868 *
869 * We have to do all the page cache truncate work outside the
870 * transaction context as the "lock" order is page lock->log space
871 * reservation as defined by extent allocation in the writeback path.
872 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
873 * having already truncated the in-memory version of the file (i.e. made
874 * user visible changes). There's not much we can do about this, except
875 * to hope that the caller sees ENOMEM and retries the truncate
876 * operation.
877 *
878 * And we update in-core i_size and truncate page cache beyond newsize
879 * before writeback the [i_disk_size, newsize] range, so we're
880 * guaranteed not to write stale data past the new EOF on truncate down.
881 */
882 truncate_setsize(inode, newsize);
883
884 /*
885 * We are going to log the inode size change in this transaction so
886 * any previous writes that are beyond the on disk EOF and the new
887 * EOF that have not been written out need to be written here. If we
888 * do not write the data out, we expose ourselves to the null files
889 * problem. Note that this includes any block zeroing we did above;
890 * otherwise those blocks may not be zeroed after a crash.
891 */
892 if (did_zeroing ||
893 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
894 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
895 ip->i_disk_size, newsize - 1);
896 if (error)
897 return error;
898 }
899
900 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
901 if (error)
902 return error;
903
904 lock_flags |= XFS_ILOCK_EXCL;
905 xfs_ilock(ip, XFS_ILOCK_EXCL);
906 xfs_trans_ijoin(tp, ip, 0);
907
908 /*
909 * Only change the c/mtime if we are changing the size or we are
910 * explicitly asked to change it. This handles the semantic difference
911 * between truncate() and ftruncate() as implemented in the VFS.
912 *
913 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
914 * special case where we need to update the times despite not having
915 * these flags set. For all other operations the VFS set these flags
916 * explicitly if it wants a timestamp update.
917 */
918 if (newsize != oldsize &&
919 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
920 iattr->ia_ctime = iattr->ia_mtime =
921 current_time(inode);
922 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
923 }
924
925 /*
926 * The first thing we do is set the size to new_size permanently on
927 * disk. This way we don't have to worry about anyone ever being able
928 * to look at the data being freed even in the face of a crash.
929 * What we're getting around here is the case where we free a block, it
930 * is allocated to another file, it is written to, and then we crash.
931 * If the new data gets written to the file but the log buffers
932 * containing the free and reallocation don't, then we'd end up with
933 * garbage in the blocks being freed. As long as we make the new size
934 * permanent before actually freeing any blocks it doesn't matter if
935 * they get written to.
936 */
937 ip->i_disk_size = newsize;
938 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
939
940 if (newsize <= oldsize) {
941 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
942 if (error)
943 goto out_trans_cancel;
944
945 /*
946 * Truncated "down", so we're removing references to old data
947 * here - if we delay flushing for a long time, we expose
948 * ourselves unduly to the notorious NULL files problem. So,
949 * we mark this inode and flush it when the file is closed,
950 * and do not wait the usual (long) time for writeout.
951 */
952 xfs_iflags_set(ip, XFS_ITRUNCATED);
953
954 /* A truncate down always removes post-EOF blocks. */
955 xfs_inode_clear_eofblocks_tag(ip);
956 }
957
958 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
959 setattr_copy(mnt_userns, inode, iattr);
960 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
961
962 XFS_STATS_INC(mp, xs_ig_attrchg);
963
964 if (xfs_has_wsync(mp))
965 xfs_trans_set_sync(tp);
966
967 error = xfs_trans_commit(tp);
968out_unlock:
969 if (lock_flags)
970 xfs_iunlock(ip, lock_flags);
971 return error;
972
973out_trans_cancel:
974 xfs_trans_cancel(tp);
975 goto out_unlock;
976}
977
978int
979xfs_vn_setattr_size(
980 struct user_namespace *mnt_userns,
981 struct dentry *dentry,
982 struct iattr *iattr)
983{
984 struct xfs_inode *ip = XFS_I(d_inode(dentry));
985 int error;
986
987 trace_xfs_setattr(ip);
988
989 error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
990 if (error)
991 return error;
992 return xfs_setattr_size(mnt_userns, dentry, ip, iattr);
993}
994
995STATIC int
996xfs_vn_setattr(
997 struct user_namespace *mnt_userns,
998 struct dentry *dentry,
999 struct iattr *iattr)
1000{
1001 struct inode *inode = d_inode(dentry);
1002 struct xfs_inode *ip = XFS_I(inode);
1003 int error;
1004
1005 if (iattr->ia_valid & ATTR_SIZE) {
1006 uint iolock;
1007
1008 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1009 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1010
1011 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1012 if (error) {
1013 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1014 return error;
1015 }
1016
1017 error = xfs_vn_setattr_size(mnt_userns, dentry, iattr);
1018 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1019 } else {
1020 trace_xfs_setattr(ip);
1021
1022 error = xfs_vn_change_ok(mnt_userns, dentry, iattr);
1023 if (!error)
1024 error = xfs_setattr_nonsize(mnt_userns, dentry, ip, iattr);
1025 }
1026
1027 return error;
1028}
1029
1030STATIC int
1031xfs_vn_update_time(
1032 struct inode *inode,
1033 struct timespec64 *now,
1034 int flags)
1035{
1036 struct xfs_inode *ip = XFS_I(inode);
1037 struct xfs_mount *mp = ip->i_mount;
1038 int log_flags = XFS_ILOG_TIMESTAMP;
1039 struct xfs_trans *tp;
1040 int error;
1041
1042 trace_xfs_update_time(ip);
1043
1044 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1045 if (!((flags & S_VERSION) &&
1046 inode_maybe_inc_iversion(inode, false)))
1047 return generic_update_time(inode, now, flags);
1048
1049 /* Capture the iversion update that just occurred */
1050 log_flags |= XFS_ILOG_CORE;
1051 }
1052
1053 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1054 if (error)
1055 return error;
1056
1057 xfs_ilock(ip, XFS_ILOCK_EXCL);
1058 if (flags & S_CTIME)
1059 inode->i_ctime = *now;
1060 if (flags & S_MTIME)
1061 inode->i_mtime = *now;
1062 if (flags & S_ATIME)
1063 inode->i_atime = *now;
1064
1065 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1066 xfs_trans_log_inode(tp, ip, log_flags);
1067 return xfs_trans_commit(tp);
1068}
1069
1070STATIC int
1071xfs_vn_fiemap(
1072 struct inode *inode,
1073 struct fiemap_extent_info *fieinfo,
1074 u64 start,
1075 u64 length)
1076{
1077 int error;
1078
1079 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1080 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1081 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1082 error = iomap_fiemap(inode, fieinfo, start, length,
1083 &xfs_xattr_iomap_ops);
1084 } else {
1085 error = iomap_fiemap(inode, fieinfo, start, length,
1086 &xfs_read_iomap_ops);
1087 }
1088 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1089
1090 return error;
1091}
1092
1093STATIC int
1094xfs_vn_tmpfile(
1095 struct user_namespace *mnt_userns,
1096 struct inode *dir,
1097 struct file *file,
1098 umode_t mode)
1099{
1100 int err = xfs_generic_create(mnt_userns, dir, file->f_path.dentry, mode, 0, file);
1101
1102 return finish_open_simple(file, err);
1103}
1104
1105static const struct inode_operations xfs_inode_operations = {
1106 .get_inode_acl = xfs_get_acl,
1107 .set_acl = xfs_set_acl,
1108 .getattr = xfs_vn_getattr,
1109 .setattr = xfs_vn_setattr,
1110 .listxattr = xfs_vn_listxattr,
1111 .fiemap = xfs_vn_fiemap,
1112 .update_time = xfs_vn_update_time,
1113 .fileattr_get = xfs_fileattr_get,
1114 .fileattr_set = xfs_fileattr_set,
1115};
1116
1117static const struct inode_operations xfs_dir_inode_operations = {
1118 .create = xfs_vn_create,
1119 .lookup = xfs_vn_lookup,
1120 .link = xfs_vn_link,
1121 .unlink = xfs_vn_unlink,
1122 .symlink = xfs_vn_symlink,
1123 .mkdir = xfs_vn_mkdir,
1124 /*
1125 * Yes, XFS uses the same method for rmdir and unlink.
1126 *
1127 * There are some subtile differences deeper in the code,
1128 * but we use S_ISDIR to check for those.
1129 */
1130 .rmdir = xfs_vn_unlink,
1131 .mknod = xfs_vn_mknod,
1132 .rename = xfs_vn_rename,
1133 .get_inode_acl = xfs_get_acl,
1134 .set_acl = xfs_set_acl,
1135 .getattr = xfs_vn_getattr,
1136 .setattr = xfs_vn_setattr,
1137 .listxattr = xfs_vn_listxattr,
1138 .update_time = xfs_vn_update_time,
1139 .tmpfile = xfs_vn_tmpfile,
1140 .fileattr_get = xfs_fileattr_get,
1141 .fileattr_set = xfs_fileattr_set,
1142};
1143
1144static const struct inode_operations xfs_dir_ci_inode_operations = {
1145 .create = xfs_vn_create,
1146 .lookup = xfs_vn_ci_lookup,
1147 .link = xfs_vn_link,
1148 .unlink = xfs_vn_unlink,
1149 .symlink = xfs_vn_symlink,
1150 .mkdir = xfs_vn_mkdir,
1151 /*
1152 * Yes, XFS uses the same method for rmdir and unlink.
1153 *
1154 * There are some subtile differences deeper in the code,
1155 * but we use S_ISDIR to check for those.
1156 */
1157 .rmdir = xfs_vn_unlink,
1158 .mknod = xfs_vn_mknod,
1159 .rename = xfs_vn_rename,
1160 .get_inode_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 .update_time = xfs_vn_update_time,
1166 .tmpfile = xfs_vn_tmpfile,
1167 .fileattr_get = xfs_fileattr_get,
1168 .fileattr_set = xfs_fileattr_set,
1169};
1170
1171static const struct inode_operations xfs_symlink_inode_operations = {
1172 .get_link = xfs_vn_get_link,
1173 .getattr = xfs_vn_getattr,
1174 .setattr = xfs_vn_setattr,
1175 .listxattr = xfs_vn_listxattr,
1176 .update_time = xfs_vn_update_time,
1177};
1178
1179/* Figure out if this file actually supports DAX. */
1180static bool
1181xfs_inode_supports_dax(
1182 struct xfs_inode *ip)
1183{
1184 struct xfs_mount *mp = ip->i_mount;
1185
1186 /* Only supported on regular files. */
1187 if (!S_ISREG(VFS_I(ip)->i_mode))
1188 return false;
1189
1190 /* Block size must match page size */
1191 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1192 return false;
1193
1194 /* Device has to support DAX too. */
1195 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1196}
1197
1198static bool
1199xfs_inode_should_enable_dax(
1200 struct xfs_inode *ip)
1201{
1202 if (!IS_ENABLED(CONFIG_FS_DAX))
1203 return false;
1204 if (xfs_has_dax_never(ip->i_mount))
1205 return false;
1206 if (!xfs_inode_supports_dax(ip))
1207 return false;
1208 if (xfs_has_dax_always(ip->i_mount))
1209 return true;
1210 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1211 return true;
1212 return false;
1213}
1214
1215void
1216xfs_diflags_to_iflags(
1217 struct xfs_inode *ip,
1218 bool init)
1219{
1220 struct inode *inode = VFS_I(ip);
1221 unsigned int xflags = xfs_ip2xflags(ip);
1222 unsigned int flags = 0;
1223
1224 ASSERT(!(IS_DAX(inode) && init));
1225
1226 if (xflags & FS_XFLAG_IMMUTABLE)
1227 flags |= S_IMMUTABLE;
1228 if (xflags & FS_XFLAG_APPEND)
1229 flags |= S_APPEND;
1230 if (xflags & FS_XFLAG_SYNC)
1231 flags |= S_SYNC;
1232 if (xflags & FS_XFLAG_NOATIME)
1233 flags |= S_NOATIME;
1234 if (init && xfs_inode_should_enable_dax(ip))
1235 flags |= S_DAX;
1236
1237 /*
1238 * S_DAX can only be set during inode initialization and is never set by
1239 * the VFS, so we cannot mask off S_DAX in i_flags.
1240 */
1241 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1242 inode->i_flags |= flags;
1243}
1244
1245/*
1246 * Initialize the Linux inode.
1247 *
1248 * When reading existing inodes from disk this is called directly from xfs_iget,
1249 * when creating a new inode it is called from xfs_init_new_inode after setting
1250 * up the inode. These callers have different criteria for clearing XFS_INEW, so
1251 * leave it up to the caller to deal with unlocking the inode appropriately.
1252 */
1253void
1254xfs_setup_inode(
1255 struct xfs_inode *ip)
1256{
1257 struct inode *inode = &ip->i_vnode;
1258 gfp_t gfp_mask;
1259
1260 inode->i_ino = ip->i_ino;
1261 inode->i_state |= I_NEW;
1262
1263 inode_sb_list_add(inode);
1264 /* make the inode look hashed for the writeback code */
1265 inode_fake_hash(inode);
1266
1267 i_size_write(inode, ip->i_disk_size);
1268 xfs_diflags_to_iflags(ip, true);
1269
1270 if (S_ISDIR(inode->i_mode)) {
1271 /*
1272 * We set the i_rwsem class here to avoid potential races with
1273 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1274 * after a filehandle lookup has already found the inode in
1275 * cache before it has been unlocked via unlock_new_inode().
1276 */
1277 lockdep_set_class(&inode->i_rwsem,
1278 &inode->i_sb->s_type->i_mutex_dir_key);
1279 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1280 } else {
1281 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1282 }
1283
1284 /*
1285 * Ensure all page cache allocations are done from GFP_NOFS context to
1286 * prevent direct reclaim recursion back into the filesystem and blowing
1287 * stacks or deadlocking.
1288 */
1289 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1290 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1291
1292 /*
1293 * If there is no attribute fork no ACL can exist on this inode,
1294 * and it can't have any file capabilities attached to it either.
1295 */
1296 if (!xfs_inode_has_attr_fork(ip)) {
1297 inode_has_no_xattr(inode);
1298 cache_no_acl(inode);
1299 }
1300}
1301
1302void
1303xfs_setup_iops(
1304 struct xfs_inode *ip)
1305{
1306 struct inode *inode = &ip->i_vnode;
1307
1308 switch (inode->i_mode & S_IFMT) {
1309 case S_IFREG:
1310 inode->i_op = &xfs_inode_operations;
1311 inode->i_fop = &xfs_file_operations;
1312 if (IS_DAX(inode))
1313 inode->i_mapping->a_ops = &xfs_dax_aops;
1314 else
1315 inode->i_mapping->a_ops = &xfs_address_space_operations;
1316 break;
1317 case S_IFDIR:
1318 if (xfs_has_asciici(XFS_M(inode->i_sb)))
1319 inode->i_op = &xfs_dir_ci_inode_operations;
1320 else
1321 inode->i_op = &xfs_dir_inode_operations;
1322 inode->i_fop = &xfs_dir_file_operations;
1323 break;
1324 case S_IFLNK:
1325 inode->i_op = &xfs_symlink_inode_operations;
1326 break;
1327 default:
1328 inode->i_op = &xfs_inode_operations;
1329 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1330 break;
1331 }
1332}