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
42#include <linux/capability.h>
43#include <linux/xattr.h>
44#include <linux/posix_acl.h>
45#include <linux/security.h>
46#include <linux/fiemap.h>
47#include <linux/slab.h>
48
49/*
50 * Directories have different lock order w.r.t. mmap_sem compared to regular
51 * files. This is due to readdir potentially triggering page faults on a user
52 * buffer inside filldir(), and this happens with the ilock on the directory
53 * held. For regular files, the lock order is the other way around - the
54 * mmap_sem is taken during the page fault, and then we lock the ilock to do
55 * block mapping. Hence we need a different class for the directory ilock so
56 * that lockdep can tell them apart.
57 */
58static struct lock_class_key xfs_nondir_ilock_class;
59static struct lock_class_key xfs_dir_ilock_class;
60
61static int
62xfs_initxattrs(
63 struct inode *inode,
64 const struct xattr *xattr_array,
65 void *fs_info)
66{
67 const struct xattr *xattr;
68 struct xfs_inode *ip = XFS_I(inode);
69 int error = 0;
70
71 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
72 error = xfs_attr_set(ip, xattr->name, xattr->value,
73 xattr->value_len, ATTR_SECURE);
74 if (error < 0)
75 break;
76 }
77 return error;
78}
79
80/*
81 * Hook in SELinux. This is not quite correct yet, what we really need
82 * here (as we do for default ACLs) is a mechanism by which creation of
83 * these attrs can be journalled at inode creation time (along with the
84 * inode, of course, such that log replay can't cause these to be lost).
85 */
86
87STATIC int
88xfs_init_security(
89 struct inode *inode,
90 struct inode *dir,
91 const struct qstr *qstr)
92{
93 return security_inode_init_security(inode, dir, qstr,
94 &xfs_initxattrs, NULL);
95}
96
97static void
98xfs_dentry_to_name(
99 struct xfs_name *namep,
100 struct dentry *dentry,
101 int mode)
102{
103 namep->name = dentry->d_name.name;
104 namep->len = dentry->d_name.len;
105 namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
106}
107
108STATIC void
109xfs_cleanup_inode(
110 struct inode *dir,
111 struct inode *inode,
112 struct dentry *dentry)
113{
114 struct xfs_name teardown;
115
116 /* Oh, the horror.
117 * If we can't add the ACL or we fail in
118 * xfs_init_security we must back out.
119 * ENOSPC can hit here, among other things.
120 */
121 xfs_dentry_to_name(&teardown, dentry, 0);
122
123 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
124}
125
126STATIC int
127xfs_generic_create(
128 struct inode *dir,
129 struct dentry *dentry,
130 umode_t mode,
131 dev_t rdev,
132 bool tmpfile) /* unnamed file */
133{
134 struct inode *inode;
135 struct xfs_inode *ip = NULL;
136 struct posix_acl *default_acl, *acl;
137 struct xfs_name name;
138 int error;
139
140 /*
141 * Irix uses Missed'em'V split, but doesn't want to see
142 * the upper 5 bits of (14bit) major.
143 */
144 if (S_ISCHR(mode) || S_ISBLK(mode)) {
145 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
146 return -EINVAL;
147 rdev = sysv_encode_dev(rdev);
148 } else {
149 rdev = 0;
150 }
151
152 error = posix_acl_create(dir, &mode, &default_acl, &acl);
153 if (error)
154 return error;
155
156 if (!tmpfile) {
157 xfs_dentry_to_name(&name, dentry, mode);
158 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
159 } else {
160 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
161 }
162 if (unlikely(error))
163 goto out_free_acl;
164
165 inode = VFS_I(ip);
166
167 error = xfs_init_security(inode, dir, &dentry->d_name);
168 if (unlikely(error))
169 goto out_cleanup_inode;
170
171#ifdef CONFIG_XFS_POSIX_ACL
172 if (default_acl) {
173 error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
174 if (error)
175 goto out_cleanup_inode;
176 }
177 if (acl) {
178 error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
179 if (error)
180 goto out_cleanup_inode;
181 }
182#endif
183
184 if (tmpfile)
185 d_tmpfile(dentry, inode);
186 else
187 d_instantiate(dentry, inode);
188
189 xfs_finish_inode_setup(ip);
190
191 out_free_acl:
192 if (default_acl)
193 posix_acl_release(default_acl);
194 if (acl)
195 posix_acl_release(acl);
196 return error;
197
198 out_cleanup_inode:
199 xfs_finish_inode_setup(ip);
200 if (!tmpfile)
201 xfs_cleanup_inode(dir, inode, dentry);
202 iput(inode);
203 goto out_free_acl;
204}
205
206STATIC int
207xfs_vn_mknod(
208 struct inode *dir,
209 struct dentry *dentry,
210 umode_t mode,
211 dev_t rdev)
212{
213 return xfs_generic_create(dir, dentry, mode, rdev, false);
214}
215
216STATIC int
217xfs_vn_create(
218 struct inode *dir,
219 struct dentry *dentry,
220 umode_t mode,
221 bool flags)
222{
223 return xfs_vn_mknod(dir, dentry, mode, 0);
224}
225
226STATIC int
227xfs_vn_mkdir(
228 struct inode *dir,
229 struct dentry *dentry,
230 umode_t mode)
231{
232 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
233}
234
235STATIC struct dentry *
236xfs_vn_lookup(
237 struct inode *dir,
238 struct dentry *dentry,
239 unsigned int flags)
240{
241 struct xfs_inode *cip;
242 struct xfs_name name;
243 int error;
244
245 if (dentry->d_name.len >= MAXNAMELEN)
246 return ERR_PTR(-ENAMETOOLONG);
247
248 xfs_dentry_to_name(&name, dentry, 0);
249 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
250 if (unlikely(error)) {
251 if (unlikely(error != -ENOENT))
252 return ERR_PTR(error);
253 d_add(dentry, NULL);
254 return NULL;
255 }
256
257 return d_splice_alias(VFS_I(cip), dentry);
258}
259
260STATIC struct dentry *
261xfs_vn_ci_lookup(
262 struct inode *dir,
263 struct dentry *dentry,
264 unsigned int flags)
265{
266 struct xfs_inode *ip;
267 struct xfs_name xname;
268 struct xfs_name ci_name;
269 struct qstr dname;
270 int error;
271
272 if (dentry->d_name.len >= MAXNAMELEN)
273 return ERR_PTR(-ENAMETOOLONG);
274
275 xfs_dentry_to_name(&xname, dentry, 0);
276 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
277 if (unlikely(error)) {
278 if (unlikely(error != -ENOENT))
279 return ERR_PTR(error);
280 /*
281 * call d_add(dentry, NULL) here when d_drop_negative_children
282 * is called in xfs_vn_mknod (ie. allow negative dentries
283 * with CI filesystems).
284 */
285 return NULL;
286 }
287
288 /* if exact match, just splice and exit */
289 if (!ci_name.name)
290 return d_splice_alias(VFS_I(ip), dentry);
291
292 /* else case-insensitive match... */
293 dname.name = ci_name.name;
294 dname.len = ci_name.len;
295 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
296 kmem_free(ci_name.name);
297 return dentry;
298}
299
300STATIC int
301xfs_vn_link(
302 struct dentry *old_dentry,
303 struct inode *dir,
304 struct dentry *dentry)
305{
306 struct inode *inode = d_inode(old_dentry);
307 struct xfs_name name;
308 int error;
309
310 xfs_dentry_to_name(&name, dentry, inode->i_mode);
311
312 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
313 if (unlikely(error))
314 return error;
315
316 ihold(inode);
317 d_instantiate(dentry, inode);
318 return 0;
319}
320
321STATIC int
322xfs_vn_unlink(
323 struct inode *dir,
324 struct dentry *dentry)
325{
326 struct xfs_name name;
327 int error;
328
329 xfs_dentry_to_name(&name, dentry, 0);
330
331 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
332 if (error)
333 return error;
334
335 /*
336 * With unlink, the VFS makes the dentry "negative": no inode,
337 * but still hashed. This is incompatible with case-insensitive
338 * mode, so invalidate (unhash) the dentry in CI-mode.
339 */
340 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
341 d_invalidate(dentry);
342 return 0;
343}
344
345STATIC int
346xfs_vn_symlink(
347 struct inode *dir,
348 struct dentry *dentry,
349 const char *symname)
350{
351 struct inode *inode;
352 struct xfs_inode *cip = NULL;
353 struct xfs_name name;
354 int error;
355 umode_t mode;
356
357 mode = S_IFLNK |
358 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
359 xfs_dentry_to_name(&name, dentry, mode);
360
361 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
362 if (unlikely(error))
363 goto out;
364
365 inode = VFS_I(cip);
366
367 error = xfs_init_security(inode, dir, &dentry->d_name);
368 if (unlikely(error))
369 goto out_cleanup_inode;
370
371 d_instantiate(dentry, inode);
372 xfs_finish_inode_setup(cip);
373 return 0;
374
375 out_cleanup_inode:
376 xfs_finish_inode_setup(cip);
377 xfs_cleanup_inode(dir, inode, dentry);
378 iput(inode);
379 out:
380 return error;
381}
382
383STATIC int
384xfs_vn_rename(
385 struct inode *odir,
386 struct dentry *odentry,
387 struct inode *ndir,
388 struct dentry *ndentry,
389 unsigned int flags)
390{
391 struct inode *new_inode = d_inode(ndentry);
392 int omode = 0;
393 struct xfs_name oname;
394 struct xfs_name nname;
395
396 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
397 return -EINVAL;
398
399 /* if we are exchanging files, we need to set i_mode of both files */
400 if (flags & RENAME_EXCHANGE)
401 omode = d_inode(ndentry)->i_mode;
402
403 xfs_dentry_to_name(&oname, odentry, omode);
404 xfs_dentry_to_name(&nname, ndentry, d_inode(odentry)->i_mode);
405
406 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
407 XFS_I(ndir), &nname,
408 new_inode ? XFS_I(new_inode) : NULL, flags);
409}
410
411/*
412 * careful here - this function can get called recursively, so
413 * we need to be very careful about how much stack we use.
414 * uio is kmalloced for this reason...
415 */
416STATIC const char *
417xfs_vn_get_link(
418 struct dentry *dentry,
419 struct inode *inode,
420 struct delayed_call *done)
421{
422 char *link;
423 int error = -ENOMEM;
424
425 if (!dentry)
426 return ERR_PTR(-ECHILD);
427
428 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
429 if (!link)
430 goto out_err;
431
432 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
433 if (unlikely(error))
434 goto out_kfree;
435
436 set_delayed_call(done, kfree_link, link);
437 return link;
438
439 out_kfree:
440 kfree(link);
441 out_err:
442 return ERR_PTR(error);
443}
444
445STATIC int
446xfs_vn_getattr(
447 struct vfsmount *mnt,
448 struct dentry *dentry,
449 struct kstat *stat)
450{
451 struct inode *inode = d_inode(dentry);
452 struct xfs_inode *ip = XFS_I(inode);
453 struct xfs_mount *mp = ip->i_mount;
454
455 trace_xfs_getattr(ip);
456
457 if (XFS_FORCED_SHUTDOWN(mp))
458 return -EIO;
459
460 stat->size = XFS_ISIZE(ip);
461 stat->dev = inode->i_sb->s_dev;
462 stat->mode = inode->i_mode;
463 stat->nlink = inode->i_nlink;
464 stat->uid = inode->i_uid;
465 stat->gid = inode->i_gid;
466 stat->ino = ip->i_ino;
467 stat->atime = inode->i_atime;
468 stat->mtime = inode->i_mtime;
469 stat->ctime = inode->i_ctime;
470 stat->blocks =
471 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
472
473
474 switch (inode->i_mode & S_IFMT) {
475 case S_IFBLK:
476 case S_IFCHR:
477 stat->blksize = BLKDEV_IOSIZE;
478 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
479 sysv_minor(ip->i_df.if_u2.if_rdev));
480 break;
481 default:
482 if (XFS_IS_REALTIME_INODE(ip)) {
483 /*
484 * If the file blocks are being allocated from a
485 * realtime volume, then return the inode's realtime
486 * extent size or the realtime volume's extent size.
487 */
488 stat->blksize =
489 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
490 } else
491 stat->blksize = xfs_preferred_iosize(mp);
492 stat->rdev = 0;
493 break;
494 }
495
496 return 0;
497}
498
499static void
500xfs_setattr_mode(
501 struct xfs_inode *ip,
502 struct iattr *iattr)
503{
504 struct inode *inode = VFS_I(ip);
505 umode_t mode = iattr->ia_mode;
506
507 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
508
509 inode->i_mode &= S_IFMT;
510 inode->i_mode |= mode & ~S_IFMT;
511}
512
513void
514xfs_setattr_time(
515 struct xfs_inode *ip,
516 struct iattr *iattr)
517{
518 struct inode *inode = VFS_I(ip);
519
520 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
521
522 if (iattr->ia_valid & ATTR_ATIME)
523 inode->i_atime = iattr->ia_atime;
524 if (iattr->ia_valid & ATTR_CTIME)
525 inode->i_ctime = iattr->ia_ctime;
526 if (iattr->ia_valid & ATTR_MTIME)
527 inode->i_mtime = iattr->ia_mtime;
528}
529
530int
531xfs_setattr_nonsize(
532 struct xfs_inode *ip,
533 struct iattr *iattr,
534 int flags)
535{
536 xfs_mount_t *mp = ip->i_mount;
537 struct inode *inode = VFS_I(ip);
538 int mask = iattr->ia_valid;
539 xfs_trans_t *tp;
540 int error;
541 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
542 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
543 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
544 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
545
546 trace_xfs_setattr(ip);
547
548 /* If acls are being inherited, we already have this checked */
549 if (!(flags & XFS_ATTR_NOACL)) {
550 if (mp->m_flags & XFS_MOUNT_RDONLY)
551 return -EROFS;
552
553 if (XFS_FORCED_SHUTDOWN(mp))
554 return -EIO;
555
556 error = inode_change_ok(inode, iattr);
557 if (error)
558 return error;
559 }
560
561 ASSERT((mask & ATTR_SIZE) == 0);
562
563 /*
564 * If disk quotas is on, we make sure that the dquots do exist on disk,
565 * before we start any other transactions. Trying to do this later
566 * is messy. We don't care to take a readlock to look at the ids
567 * in inode here, because we can't hold it across the trans_reserve.
568 * If the IDs do change before we take the ilock, we're covered
569 * because the i_*dquot fields will get updated anyway.
570 */
571 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
572 uint qflags = 0;
573
574 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
575 uid = iattr->ia_uid;
576 qflags |= XFS_QMOPT_UQUOTA;
577 } else {
578 uid = inode->i_uid;
579 }
580 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
581 gid = iattr->ia_gid;
582 qflags |= XFS_QMOPT_GQUOTA;
583 } else {
584 gid = inode->i_gid;
585 }
586
587 /*
588 * We take a reference when we initialize udqp and gdqp,
589 * so it is important that we never blindly double trip on
590 * the same variable. See xfs_create() for an example.
591 */
592 ASSERT(udqp == NULL);
593 ASSERT(gdqp == NULL);
594 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
595 xfs_kgid_to_gid(gid),
596 xfs_get_projid(ip),
597 qflags, &udqp, &gdqp, NULL);
598 if (error)
599 return error;
600 }
601
602 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
603 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
604 if (error)
605 goto out_trans_cancel;
606
607 xfs_ilock(ip, XFS_ILOCK_EXCL);
608
609 /*
610 * Change file ownership. Must be the owner or privileged.
611 */
612 if (mask & (ATTR_UID|ATTR_GID)) {
613 /*
614 * These IDs could have changed since we last looked at them.
615 * But, we're assured that if the ownership did change
616 * while we didn't have the inode locked, inode's dquot(s)
617 * would have changed also.
618 */
619 iuid = inode->i_uid;
620 igid = inode->i_gid;
621 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
622 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
623
624 /*
625 * Do a quota reservation only if uid/gid is actually
626 * going to change.
627 */
628 if (XFS_IS_QUOTA_RUNNING(mp) &&
629 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
630 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
631 ASSERT(tp);
632 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
633 NULL, capable(CAP_FOWNER) ?
634 XFS_QMOPT_FORCE_RES : 0);
635 if (error) /* out of quota */
636 goto out_unlock;
637 }
638 }
639
640 xfs_trans_ijoin(tp, ip, 0);
641
642 /*
643 * Change file ownership. Must be the owner or privileged.
644 */
645 if (mask & (ATTR_UID|ATTR_GID)) {
646 /*
647 * CAP_FSETID overrides the following restrictions:
648 *
649 * The set-user-ID and set-group-ID bits of a file will be
650 * cleared upon successful return from chown()
651 */
652 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
653 !capable(CAP_FSETID))
654 inode->i_mode &= ~(S_ISUID|S_ISGID);
655
656 /*
657 * Change the ownerships and register quota modifications
658 * in the transaction.
659 */
660 if (!uid_eq(iuid, uid)) {
661 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
662 ASSERT(mask & ATTR_UID);
663 ASSERT(udqp);
664 olddquot1 = xfs_qm_vop_chown(tp, ip,
665 &ip->i_udquot, udqp);
666 }
667 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
668 inode->i_uid = uid;
669 }
670 if (!gid_eq(igid, gid)) {
671 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
672 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
673 !XFS_IS_PQUOTA_ON(mp));
674 ASSERT(mask & ATTR_GID);
675 ASSERT(gdqp);
676 olddquot2 = xfs_qm_vop_chown(tp, ip,
677 &ip->i_gdquot, gdqp);
678 }
679 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
680 inode->i_gid = gid;
681 }
682 }
683
684 if (mask & ATTR_MODE)
685 xfs_setattr_mode(ip, iattr);
686 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
687 xfs_setattr_time(ip, iattr);
688
689 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
690
691 XFS_STATS_INC(mp, xs_ig_attrchg);
692
693 if (mp->m_flags & XFS_MOUNT_WSYNC)
694 xfs_trans_set_sync(tp);
695 error = xfs_trans_commit(tp);
696
697 xfs_iunlock(ip, XFS_ILOCK_EXCL);
698
699 /*
700 * Release any dquot(s) the inode had kept before chown.
701 */
702 xfs_qm_dqrele(olddquot1);
703 xfs_qm_dqrele(olddquot2);
704 xfs_qm_dqrele(udqp);
705 xfs_qm_dqrele(gdqp);
706
707 if (error)
708 return error;
709
710 /*
711 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
712 * update. We could avoid this with linked transactions
713 * and passing down the transaction pointer all the way
714 * to attr_set. No previous user of the generic
715 * Posix ACL code seems to care about this issue either.
716 */
717 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
718 error = posix_acl_chmod(inode, inode->i_mode);
719 if (error)
720 return error;
721 }
722
723 return 0;
724
725out_unlock:
726 xfs_iunlock(ip, XFS_ILOCK_EXCL);
727out_trans_cancel:
728 xfs_trans_cancel(tp);
729 xfs_qm_dqrele(udqp);
730 xfs_qm_dqrele(gdqp);
731 return error;
732}
733
734/*
735 * Truncate file. Must have write permission and not be a directory.
736 */
737int
738xfs_setattr_size(
739 struct xfs_inode *ip,
740 struct iattr *iattr)
741{
742 struct xfs_mount *mp = ip->i_mount;
743 struct inode *inode = VFS_I(ip);
744 xfs_off_t oldsize, newsize;
745 struct xfs_trans *tp;
746 int error;
747 uint lock_flags = 0;
748 bool did_zeroing = false;
749
750 trace_xfs_setattr(ip);
751
752 if (mp->m_flags & XFS_MOUNT_RDONLY)
753 return -EROFS;
754
755 if (XFS_FORCED_SHUTDOWN(mp))
756 return -EIO;
757
758 error = inode_change_ok(inode, iattr);
759 if (error)
760 return error;
761
762 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
763 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
764 ASSERT(S_ISREG(inode->i_mode));
765 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
766 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
767
768 oldsize = inode->i_size;
769 newsize = iattr->ia_size;
770
771 /*
772 * Short circuit the truncate case for zero length files.
773 */
774 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
775 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
776 return 0;
777
778 /*
779 * Use the regular setattr path to update the timestamps.
780 */
781 iattr->ia_valid &= ~ATTR_SIZE;
782 return xfs_setattr_nonsize(ip, iattr, 0);
783 }
784
785 /*
786 * Make sure that the dquots are attached to the inode.
787 */
788 error = xfs_qm_dqattach(ip, 0);
789 if (error)
790 return error;
791
792 /*
793 * File data changes must be complete before we start the transaction to
794 * modify the inode. This needs to be done before joining the inode to
795 * the transaction because the inode cannot be unlocked once it is a
796 * part of the transaction.
797 *
798 * Start with zeroing any data block beyond EOF that we may expose on
799 * file extension.
800 */
801 if (newsize > oldsize) {
802 error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
803 if (error)
804 return error;
805 }
806
807 /*
808 * We are going to log the inode size change in this transaction so
809 * any previous writes that are beyond the on disk EOF and the new
810 * EOF that have not been written out need to be written here. If we
811 * do not write the data out, we expose ourselves to the null files
812 * problem. Note that this includes any block zeroing we did above;
813 * otherwise those blocks may not be zeroed after a crash.
814 */
815 if (newsize > ip->i_d.di_size &&
816 (oldsize != ip->i_d.di_size || did_zeroing)) {
817 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
818 ip->i_d.di_size, newsize);
819 if (error)
820 return error;
821 }
822
823 /* Now wait for all direct I/O to complete. */
824 inode_dio_wait(inode);
825
826 /*
827 * We've already locked out new page faults, so now we can safely remove
828 * pages from the page cache knowing they won't get refaulted until we
829 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
830 * complete. The truncate_setsize() call also cleans partial EOF page
831 * PTEs on extending truncates and hence ensures sub-page block size
832 * filesystems are correctly handled, too.
833 *
834 * We have to do all the page cache truncate work outside the
835 * transaction context as the "lock" order is page lock->log space
836 * reservation as defined by extent allocation in the writeback path.
837 * Hence a truncate can fail with ENOMEM from xfs_trans_reserve(), but
838 * having already truncated the in-memory version of the file (i.e. made
839 * user visible changes). There's not much we can do about this, except
840 * to hope that the caller sees ENOMEM and retries the truncate
841 * operation.
842 */
843 if (IS_DAX(inode))
844 error = dax_truncate_page(inode, newsize, xfs_get_blocks_direct);
845 else
846 error = block_truncate_page(inode->i_mapping, newsize,
847 xfs_get_blocks);
848 if (error)
849 return error;
850 truncate_setsize(inode, newsize);
851
852 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
853 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
854 if (error)
855 goto out_trans_cancel;
856
857 lock_flags |= XFS_ILOCK_EXCL;
858 xfs_ilock(ip, XFS_ILOCK_EXCL);
859 xfs_trans_ijoin(tp, ip, 0);
860
861 /*
862 * Only change the c/mtime if we are changing the size or we are
863 * explicitly asked to change it. This handles the semantic difference
864 * between truncate() and ftruncate() as implemented in the VFS.
865 *
866 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
867 * special case where we need to update the times despite not having
868 * these flags set. For all other operations the VFS set these flags
869 * explicitly if it wants a timestamp update.
870 */
871 if (newsize != oldsize &&
872 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
873 iattr->ia_ctime = iattr->ia_mtime =
874 current_fs_time(inode->i_sb);
875 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
876 }
877
878 /*
879 * The first thing we do is set the size to new_size permanently on
880 * disk. This way we don't have to worry about anyone ever being able
881 * to look at the data being freed even in the face of a crash.
882 * What we're getting around here is the case where we free a block, it
883 * is allocated to another file, it is written to, and then we crash.
884 * If the new data gets written to the file but the log buffers
885 * containing the free and reallocation don't, then we'd end up with
886 * garbage in the blocks being freed. As long as we make the new size
887 * permanent before actually freeing any blocks it doesn't matter if
888 * they get written to.
889 */
890 ip->i_d.di_size = newsize;
891 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
892
893 if (newsize <= oldsize) {
894 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
895 if (error)
896 goto out_trans_cancel;
897
898 /*
899 * Truncated "down", so we're removing references to old data
900 * here - if we delay flushing for a long time, we expose
901 * ourselves unduly to the notorious NULL files problem. So,
902 * we mark this inode and flush it when the file is closed,
903 * and do not wait the usual (long) time for writeout.
904 */
905 xfs_iflags_set(ip, XFS_ITRUNCATED);
906
907 /* A truncate down always removes post-EOF blocks. */
908 xfs_inode_clear_eofblocks_tag(ip);
909 }
910
911 if (iattr->ia_valid & ATTR_MODE)
912 xfs_setattr_mode(ip, iattr);
913 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
914 xfs_setattr_time(ip, iattr);
915
916 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
917
918 XFS_STATS_INC(mp, xs_ig_attrchg);
919
920 if (mp->m_flags & XFS_MOUNT_WSYNC)
921 xfs_trans_set_sync(tp);
922
923 error = xfs_trans_commit(tp);
924out_unlock:
925 if (lock_flags)
926 xfs_iunlock(ip, lock_flags);
927 return error;
928
929out_trans_cancel:
930 xfs_trans_cancel(tp);
931 goto out_unlock;
932}
933
934STATIC int
935xfs_vn_setattr(
936 struct dentry *dentry,
937 struct iattr *iattr)
938{
939 struct xfs_inode *ip = XFS_I(d_inode(dentry));
940 int error;
941
942 if (iattr->ia_valid & ATTR_SIZE) {
943 uint iolock = XFS_IOLOCK_EXCL;
944
945 xfs_ilock(ip, iolock);
946 error = xfs_break_layouts(d_inode(dentry), &iolock, true);
947 if (!error) {
948 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
949 iolock |= XFS_MMAPLOCK_EXCL;
950
951 error = xfs_setattr_size(ip, iattr);
952 }
953 xfs_iunlock(ip, iolock);
954 } else {
955 error = xfs_setattr_nonsize(ip, iattr, 0);
956 }
957
958 return error;
959}
960
961STATIC int
962xfs_vn_update_time(
963 struct inode *inode,
964 struct timespec *now,
965 int flags)
966{
967 struct xfs_inode *ip = XFS_I(inode);
968 struct xfs_mount *mp = ip->i_mount;
969 struct xfs_trans *tp;
970 int error;
971
972 trace_xfs_update_time(ip);
973
974 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
975 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
976 if (error) {
977 xfs_trans_cancel(tp);
978 return error;
979 }
980
981 xfs_ilock(ip, XFS_ILOCK_EXCL);
982 if (flags & S_CTIME)
983 inode->i_ctime = *now;
984 if (flags & S_MTIME)
985 inode->i_mtime = *now;
986 if (flags & S_ATIME)
987 inode->i_atime = *now;
988
989 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
990 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
991 return xfs_trans_commit(tp);
992}
993
994#define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
995
996/*
997 * Call fiemap helper to fill in user data.
998 * Returns positive errors to xfs_getbmap.
999 */
1000STATIC int
1001xfs_fiemap_format(
1002 void **arg,
1003 struct getbmapx *bmv,
1004 int *full)
1005{
1006 int error;
1007 struct fiemap_extent_info *fieinfo = *arg;
1008 u32 fiemap_flags = 0;
1009 u64 logical, physical, length;
1010
1011 /* Do nothing for a hole */
1012 if (bmv->bmv_block == -1LL)
1013 return 0;
1014
1015 logical = BBTOB(bmv->bmv_offset);
1016 physical = BBTOB(bmv->bmv_block);
1017 length = BBTOB(bmv->bmv_length);
1018
1019 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1020 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1021 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1022 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1023 FIEMAP_EXTENT_UNKNOWN);
1024 physical = 0; /* no block yet */
1025 }
1026 if (bmv->bmv_oflags & BMV_OF_LAST)
1027 fiemap_flags |= FIEMAP_EXTENT_LAST;
1028
1029 error = fiemap_fill_next_extent(fieinfo, logical, physical,
1030 length, fiemap_flags);
1031 if (error > 0) {
1032 error = 0;
1033 *full = 1; /* user array now full */
1034 }
1035
1036 return error;
1037}
1038
1039STATIC int
1040xfs_vn_fiemap(
1041 struct inode *inode,
1042 struct fiemap_extent_info *fieinfo,
1043 u64 start,
1044 u64 length)
1045{
1046 xfs_inode_t *ip = XFS_I(inode);
1047 struct getbmapx bm;
1048 int error;
1049
1050 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1051 if (error)
1052 return error;
1053
1054 /* Set up bmap header for xfs internal routine */
1055 bm.bmv_offset = BTOBBT(start);
1056 /* Special case for whole file */
1057 if (length == FIEMAP_MAX_OFFSET)
1058 bm.bmv_length = -1LL;
1059 else
1060 bm.bmv_length = BTOBB(start + length) - bm.bmv_offset;
1061
1062 /* We add one because in getbmap world count includes the header */
1063 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1064 fieinfo->fi_extents_max + 1;
1065 bm.bmv_count = min_t(__s32, bm.bmv_count,
1066 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1067 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1068 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1069 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1070 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1071 bm.bmv_iflags |= BMV_IF_DELALLOC;
1072
1073 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1074 if (error)
1075 return error;
1076
1077 return 0;
1078}
1079
1080STATIC int
1081xfs_vn_tmpfile(
1082 struct inode *dir,
1083 struct dentry *dentry,
1084 umode_t mode)
1085{
1086 return xfs_generic_create(dir, dentry, mode, 0, true);
1087}
1088
1089static const struct inode_operations xfs_inode_operations = {
1090 .get_acl = xfs_get_acl,
1091 .set_acl = xfs_set_acl,
1092 .getattr = xfs_vn_getattr,
1093 .setattr = xfs_vn_setattr,
1094 .setxattr = generic_setxattr,
1095 .getxattr = generic_getxattr,
1096 .removexattr = generic_removexattr,
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 .rename2 = 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 .setxattr = generic_setxattr,
1123 .getxattr = generic_getxattr,
1124 .removexattr = generic_removexattr,
1125 .listxattr = xfs_vn_listxattr,
1126 .update_time = xfs_vn_update_time,
1127 .tmpfile = xfs_vn_tmpfile,
1128};
1129
1130static const struct inode_operations xfs_dir_ci_inode_operations = {
1131 .create = xfs_vn_create,
1132 .lookup = xfs_vn_ci_lookup,
1133 .link = xfs_vn_link,
1134 .unlink = xfs_vn_unlink,
1135 .symlink = xfs_vn_symlink,
1136 .mkdir = xfs_vn_mkdir,
1137 /*
1138 * Yes, XFS uses the same method for rmdir and unlink.
1139 *
1140 * There are some subtile differences deeper in the code,
1141 * but we use S_ISDIR to check for those.
1142 */
1143 .rmdir = xfs_vn_unlink,
1144 .mknod = xfs_vn_mknod,
1145 .rename2 = xfs_vn_rename,
1146 .get_acl = xfs_get_acl,
1147 .set_acl = xfs_set_acl,
1148 .getattr = xfs_vn_getattr,
1149 .setattr = xfs_vn_setattr,
1150 .setxattr = generic_setxattr,
1151 .getxattr = generic_getxattr,
1152 .removexattr = generic_removexattr,
1153 .listxattr = xfs_vn_listxattr,
1154 .update_time = xfs_vn_update_time,
1155 .tmpfile = xfs_vn_tmpfile,
1156};
1157
1158static const struct inode_operations xfs_symlink_inode_operations = {
1159 .readlink = generic_readlink,
1160 .get_link = xfs_vn_get_link,
1161 .getattr = xfs_vn_getattr,
1162 .setattr = xfs_vn_setattr,
1163 .setxattr = generic_setxattr,
1164 .getxattr = generic_getxattr,
1165 .removexattr = generic_removexattr,
1166 .listxattr = xfs_vn_listxattr,
1167 .update_time = xfs_vn_update_time,
1168};
1169
1170STATIC void
1171xfs_diflags_to_iflags(
1172 struct inode *inode,
1173 struct xfs_inode *ip)
1174{
1175 uint16_t flags = ip->i_d.di_flags;
1176
1177 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1178 S_NOATIME | S_DAX);
1179
1180 if (flags & XFS_DIFLAG_IMMUTABLE)
1181 inode->i_flags |= S_IMMUTABLE;
1182 if (flags & XFS_DIFLAG_APPEND)
1183 inode->i_flags |= S_APPEND;
1184 if (flags & XFS_DIFLAG_SYNC)
1185 inode->i_flags |= S_SYNC;
1186 if (flags & XFS_DIFLAG_NOATIME)
1187 inode->i_flags |= S_NOATIME;
1188 if (S_ISREG(inode->i_mode) &&
1189 ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
1190 (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
1191 ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1192 inode->i_flags |= S_DAX;
1193}
1194
1195/*
1196 * Initialize the Linux inode and set up the operation vectors.
1197 *
1198 * When reading existing inodes from disk this is called directly from xfs_iget,
1199 * when creating a new inode it is called from xfs_ialloc after setting up the
1200 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1201 * it up to the caller to deal with unlocking the inode appropriately.
1202 */
1203void
1204xfs_setup_inode(
1205 struct xfs_inode *ip)
1206{
1207 struct inode *inode = &ip->i_vnode;
1208 gfp_t gfp_mask;
1209
1210 inode->i_ino = ip->i_ino;
1211 inode->i_state = I_NEW;
1212
1213 inode_sb_list_add(inode);
1214 /* make the inode look hashed for the writeback code */
1215 hlist_add_fake(&inode->i_hash);
1216
1217 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1218 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1219
1220 switch (inode->i_mode & S_IFMT) {
1221 case S_IFBLK:
1222 case S_IFCHR:
1223 inode->i_rdev =
1224 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1225 sysv_minor(ip->i_df.if_u2.if_rdev));
1226 break;
1227 default:
1228 inode->i_rdev = 0;
1229 break;
1230 }
1231
1232 i_size_write(inode, ip->i_d.di_size);
1233 xfs_diflags_to_iflags(inode, ip);
1234
1235 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1236 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1237 switch (inode->i_mode & S_IFMT) {
1238 case S_IFREG:
1239 inode->i_op = &xfs_inode_operations;
1240 inode->i_fop = &xfs_file_operations;
1241 inode->i_mapping->a_ops = &xfs_address_space_operations;
1242 break;
1243 case S_IFDIR:
1244 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1245 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1246 inode->i_op = &xfs_dir_ci_inode_operations;
1247 else
1248 inode->i_op = &xfs_dir_inode_operations;
1249 inode->i_fop = &xfs_dir_file_operations;
1250 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1251 break;
1252 case S_IFLNK:
1253 inode->i_op = &xfs_symlink_inode_operations;
1254 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1255 inode->i_mapping->a_ops = &xfs_address_space_operations;
1256 break;
1257 default:
1258 inode->i_op = &xfs_inode_operations;
1259 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1260 break;
1261 }
1262
1263 /*
1264 * Ensure all page cache allocations are done from GFP_NOFS context to
1265 * prevent direct reclaim recursion back into the filesystem and blowing
1266 * stacks or deadlocking.
1267 */
1268 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1269 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1270
1271 /*
1272 * If there is no attribute fork no ACL can exist on this inode,
1273 * and it can't have any file capabilities attached to it either.
1274 */
1275 if (!XFS_IFORK_Q(ip)) {
1276 inode_has_no_xattr(inode);
1277 cache_no_acl(inode);
1278 }
1279}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_inode.h"
14#include "xfs_acl.h"
15#include "xfs_quota.h"
16#include "xfs_da_format.h"
17#include "xfs_da_btree.h"
18#include "xfs_attr.h"
19#include "xfs_trans.h"
20#include "xfs_trans_space.h"
21#include "xfs_bmap_btree.h"
22#include "xfs_trace.h"
23#include "xfs_icache.h"
24#include "xfs_symlink.h"
25#include "xfs_dir2.h"
26#include "xfs_iomap.h"
27#include "xfs_error.h"
28#include "xfs_ioctl.h"
29#include "xfs_xattr.h"
30#include "xfs_file.h"
31#include "xfs_bmap.h"
32
33#include <linux/posix_acl.h>
34#include <linux/security.h>
35#include <linux/iversion.h>
36#include <linux/fiemap.h>
37
38/*
39 * Directories have different lock order w.r.t. mmap_lock compared to regular
40 * files. This is due to readdir potentially triggering page faults on a user
41 * buffer inside filldir(), and this happens with the ilock on the directory
42 * held. For regular files, the lock order is the other way around - the
43 * mmap_lock is taken during the page fault, and then we lock the ilock to do
44 * block mapping. Hence we need a different class for the directory ilock so
45 * that lockdep can tell them apart. Directories in the metadata directory
46 * tree get a separate class so that lockdep reports will warn us if someone
47 * ever tries to lock regular directories after locking metadata directories.
48 */
49static struct lock_class_key xfs_nondir_ilock_class;
50static struct lock_class_key xfs_dir_ilock_class;
51
52static int
53xfs_initxattrs(
54 struct inode *inode,
55 const struct xattr *xattr_array,
56 void *fs_info)
57{
58 const struct xattr *xattr;
59 struct xfs_inode *ip = XFS_I(inode);
60 int error = 0;
61
62 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
63 struct xfs_da_args args = {
64 .dp = ip,
65 .attr_filter = XFS_ATTR_SECURE,
66 .name = xattr->name,
67 .namelen = strlen(xattr->name),
68 .value = xattr->value,
69 .valuelen = xattr->value_len,
70 };
71 error = xfs_attr_change(&args, XFS_ATTRUPDATE_UPSERT);
72 if (error < 0)
73 break;
74 }
75 return error;
76}
77
78/*
79 * Hook in SELinux. This is not quite correct yet, what we really need
80 * here (as we do for default ACLs) is a mechanism by which creation of
81 * these attrs can be journalled at inode creation time (along with the
82 * inode, of course, such that log replay can't cause these to be lost).
83 */
84int
85xfs_inode_init_security(
86 struct inode *inode,
87 struct inode *dir,
88 const struct qstr *qstr)
89{
90 return security_inode_init_security(inode, dir, qstr,
91 &xfs_initxattrs, NULL);
92}
93
94static void
95xfs_dentry_to_name(
96 struct xfs_name *namep,
97 struct dentry *dentry)
98{
99 namep->name = dentry->d_name.name;
100 namep->len = dentry->d_name.len;
101 namep->type = XFS_DIR3_FT_UNKNOWN;
102}
103
104static int
105xfs_dentry_mode_to_name(
106 struct xfs_name *namep,
107 struct dentry *dentry,
108 int mode)
109{
110 namep->name = dentry->d_name.name;
111 namep->len = dentry->d_name.len;
112 namep->type = xfs_mode_to_ftype(mode);
113
114 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
115 return -EFSCORRUPTED;
116
117 return 0;
118}
119
120STATIC void
121xfs_cleanup_inode(
122 struct inode *dir,
123 struct inode *inode,
124 struct dentry *dentry)
125{
126 struct xfs_name teardown;
127
128 /* Oh, the horror.
129 * If we can't add the ACL or we fail in
130 * xfs_inode_init_security we must back out.
131 * ENOSPC can hit here, among other things.
132 */
133 xfs_dentry_to_name(&teardown, dentry);
134
135 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
136}
137
138/*
139 * Check to see if we are likely to need an extended attribute to be added to
140 * the inode we are about to allocate. This allows the attribute fork to be
141 * created during the inode allocation, reducing the number of transactions we
142 * need to do in this fast path.
143 *
144 * The security checks are optimistic, but not guaranteed. The two LSMs that
145 * require xattrs to be added here (selinux and smack) are also the only two
146 * LSMs that add a sb->s_security structure to the superblock. Hence if security
147 * is enabled and sb->s_security is set, we have a pretty good idea that we are
148 * going to be asked to add a security xattr immediately after allocating the
149 * xfs inode and instantiating the VFS inode.
150 */
151static inline bool
152xfs_create_need_xattr(
153 struct inode *dir,
154 struct posix_acl *default_acl,
155 struct posix_acl *acl)
156{
157 if (acl)
158 return true;
159 if (default_acl)
160 return true;
161#if IS_ENABLED(CONFIG_SECURITY)
162 if (dir->i_sb->s_security)
163 return true;
164#endif
165 return false;
166}
167
168
169STATIC int
170xfs_generic_create(
171 struct mnt_idmap *idmap,
172 struct inode *dir,
173 struct dentry *dentry,
174 umode_t mode,
175 dev_t rdev,
176 struct file *tmpfile) /* unnamed file */
177{
178 struct xfs_icreate_args args = {
179 .idmap = idmap,
180 .pip = XFS_I(dir),
181 .rdev = rdev,
182 .mode = mode,
183 };
184 struct inode *inode;
185 struct xfs_inode *ip = NULL;
186 struct posix_acl *default_acl, *acl;
187 struct xfs_name name;
188 int error;
189
190 /*
191 * Irix uses Missed'em'V split, but doesn't want to see
192 * the upper 5 bits of (14bit) major.
193 */
194 if (S_ISCHR(args.mode) || S_ISBLK(args.mode)) {
195 if (unlikely(!sysv_valid_dev(args.rdev) ||
196 MAJOR(args.rdev) & ~0x1ff))
197 return -EINVAL;
198 } else {
199 args.rdev = 0;
200 }
201
202 error = posix_acl_create(dir, &args.mode, &default_acl, &acl);
203 if (error)
204 return error;
205
206 /* Verify mode is valid also for tmpfile case */
207 error = xfs_dentry_mode_to_name(&name, dentry, args.mode);
208 if (unlikely(error))
209 goto out_free_acl;
210
211 if (!tmpfile) {
212 if (xfs_create_need_xattr(dir, default_acl, acl))
213 args.flags |= XFS_ICREATE_INIT_XATTRS;
214
215 error = xfs_create(&args, &name, &ip);
216 } else {
217 args.flags |= XFS_ICREATE_TMPFILE;
218
219 /*
220 * If this temporary file will not be linkable, don't bother
221 * creating an attr fork to receive a parent pointer.
222 */
223 if (tmpfile->f_flags & O_EXCL)
224 args.flags |= XFS_ICREATE_UNLINKABLE;
225
226 error = xfs_create_tmpfile(&args, &ip);
227 }
228 if (unlikely(error))
229 goto out_free_acl;
230
231 inode = VFS_I(ip);
232
233 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
234 if (unlikely(error))
235 goto out_cleanup_inode;
236
237 if (default_acl) {
238 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
239 if (error)
240 goto out_cleanup_inode;
241 }
242 if (acl) {
243 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
244 if (error)
245 goto out_cleanup_inode;
246 }
247
248 xfs_setup_iops(ip);
249
250 if (tmpfile) {
251 /*
252 * The VFS requires that any inode fed to d_tmpfile must have
253 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
254 * However, we created the temp file with nlink == 0 because
255 * we're not allowed to put an inode with nlink > 0 on the
256 * unlinked list. Therefore we have to set nlink to 1 so that
257 * d_tmpfile can immediately set it back to zero.
258 */
259 set_nlink(inode, 1);
260 d_tmpfile(tmpfile, inode);
261 } else
262 d_instantiate(dentry, inode);
263
264 xfs_finish_inode_setup(ip);
265
266 out_free_acl:
267 posix_acl_release(default_acl);
268 posix_acl_release(acl);
269 return error;
270
271 out_cleanup_inode:
272 xfs_finish_inode_setup(ip);
273 if (!tmpfile)
274 xfs_cleanup_inode(dir, inode, dentry);
275 xfs_irele(ip);
276 goto out_free_acl;
277}
278
279STATIC int
280xfs_vn_mknod(
281 struct mnt_idmap *idmap,
282 struct inode *dir,
283 struct dentry *dentry,
284 umode_t mode,
285 dev_t rdev)
286{
287 return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL);
288}
289
290STATIC int
291xfs_vn_create(
292 struct mnt_idmap *idmap,
293 struct inode *dir,
294 struct dentry *dentry,
295 umode_t mode,
296 bool flags)
297{
298 return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL);
299}
300
301STATIC int
302xfs_vn_mkdir(
303 struct mnt_idmap *idmap,
304 struct inode *dir,
305 struct dentry *dentry,
306 umode_t mode)
307{
308 return xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL);
309}
310
311STATIC struct dentry *
312xfs_vn_lookup(
313 struct inode *dir,
314 struct dentry *dentry,
315 unsigned int flags)
316{
317 struct inode *inode;
318 struct xfs_inode *cip;
319 struct xfs_name name;
320 int error;
321
322 if (dentry->d_name.len >= MAXNAMELEN)
323 return ERR_PTR(-ENAMETOOLONG);
324
325 xfs_dentry_to_name(&name, dentry);
326 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
327 if (likely(!error))
328 inode = VFS_I(cip);
329 else if (likely(error == -ENOENT))
330 inode = NULL;
331 else
332 inode = ERR_PTR(error);
333 return d_splice_alias(inode, dentry);
334}
335
336STATIC struct dentry *
337xfs_vn_ci_lookup(
338 struct inode *dir,
339 struct dentry *dentry,
340 unsigned int flags)
341{
342 struct xfs_inode *ip;
343 struct xfs_name xname;
344 struct xfs_name ci_name;
345 struct qstr dname;
346 int error;
347
348 if (dentry->d_name.len >= MAXNAMELEN)
349 return ERR_PTR(-ENAMETOOLONG);
350
351 xfs_dentry_to_name(&xname, dentry);
352 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
353 if (unlikely(error)) {
354 if (unlikely(error != -ENOENT))
355 return ERR_PTR(error);
356 /*
357 * call d_add(dentry, NULL) here when d_drop_negative_children
358 * is called in xfs_vn_mknod (ie. allow negative dentries
359 * with CI filesystems).
360 */
361 return NULL;
362 }
363
364 /* if exact match, just splice and exit */
365 if (!ci_name.name)
366 return d_splice_alias(VFS_I(ip), dentry);
367
368 /* else case-insensitive match... */
369 dname.name = ci_name.name;
370 dname.len = ci_name.len;
371 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
372 kfree(ci_name.name);
373 return dentry;
374}
375
376STATIC int
377xfs_vn_link(
378 struct dentry *old_dentry,
379 struct inode *dir,
380 struct dentry *dentry)
381{
382 struct inode *inode = d_inode(old_dentry);
383 struct xfs_name name;
384 int error;
385
386 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
387 if (unlikely(error))
388 return error;
389
390 if (IS_PRIVATE(inode))
391 return -EPERM;
392
393 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
394 if (unlikely(error))
395 return error;
396
397 ihold(inode);
398 d_instantiate(dentry, inode);
399 return 0;
400}
401
402STATIC int
403xfs_vn_unlink(
404 struct inode *dir,
405 struct dentry *dentry)
406{
407 struct xfs_name name;
408 int error;
409
410 xfs_dentry_to_name(&name, dentry);
411
412 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
413 if (error)
414 return error;
415
416 /*
417 * With unlink, the VFS makes the dentry "negative": no inode,
418 * but still hashed. This is incompatible with case-insensitive
419 * mode, so invalidate (unhash) the dentry in CI-mode.
420 */
421 if (xfs_has_asciici(XFS_M(dir->i_sb)))
422 d_invalidate(dentry);
423 return 0;
424}
425
426STATIC int
427xfs_vn_symlink(
428 struct mnt_idmap *idmap,
429 struct inode *dir,
430 struct dentry *dentry,
431 const char *symname)
432{
433 struct inode *inode;
434 struct xfs_inode *cip = NULL;
435 struct xfs_name name;
436 int error;
437 umode_t mode;
438
439 mode = S_IFLNK |
440 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
441 error = xfs_dentry_mode_to_name(&name, dentry, mode);
442 if (unlikely(error))
443 goto out;
444
445 error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip);
446 if (unlikely(error))
447 goto out;
448
449 inode = VFS_I(cip);
450
451 error = xfs_inode_init_security(inode, dir, &dentry->d_name);
452 if (unlikely(error))
453 goto out_cleanup_inode;
454
455 xfs_setup_iops(cip);
456
457 d_instantiate(dentry, inode);
458 xfs_finish_inode_setup(cip);
459 return 0;
460
461 out_cleanup_inode:
462 xfs_finish_inode_setup(cip);
463 xfs_cleanup_inode(dir, inode, dentry);
464 xfs_irele(cip);
465 out:
466 return error;
467}
468
469STATIC int
470xfs_vn_rename(
471 struct mnt_idmap *idmap,
472 struct inode *odir,
473 struct dentry *odentry,
474 struct inode *ndir,
475 struct dentry *ndentry,
476 unsigned int flags)
477{
478 struct inode *new_inode = d_inode(ndentry);
479 int omode = 0;
480 int error;
481 struct xfs_name oname;
482 struct xfs_name nname;
483
484 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
485 return -EINVAL;
486
487 /* if we are exchanging files, we need to set i_mode of both files */
488 if (flags & RENAME_EXCHANGE)
489 omode = d_inode(ndentry)->i_mode;
490
491 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
492 if (omode && unlikely(error))
493 return error;
494
495 error = xfs_dentry_mode_to_name(&nname, ndentry,
496 d_inode(odentry)->i_mode);
497 if (unlikely(error))
498 return error;
499
500 return xfs_rename(idmap, XFS_I(odir), &oname,
501 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
502 new_inode ? XFS_I(new_inode) : NULL, flags);
503}
504
505/*
506 * careful here - this function can get called recursively, so
507 * we need to be very careful about how much stack we use.
508 * uio is kmalloced for this reason...
509 */
510STATIC const char *
511xfs_vn_get_link(
512 struct dentry *dentry,
513 struct inode *inode,
514 struct delayed_call *done)
515{
516 char *link;
517 int error = -ENOMEM;
518
519 if (!dentry)
520 return ERR_PTR(-ECHILD);
521
522 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
523 if (!link)
524 goto out_err;
525
526 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
527 if (unlikely(error))
528 goto out_kfree;
529
530 set_delayed_call(done, kfree_link, link);
531 return link;
532
533 out_kfree:
534 kfree(link);
535 out_err:
536 return ERR_PTR(error);
537}
538
539static uint32_t
540xfs_stat_blksize(
541 struct xfs_inode *ip)
542{
543 struct xfs_mount *mp = ip->i_mount;
544
545 /*
546 * If the file blocks are being allocated from a realtime volume, then
547 * always return the realtime extent size.
548 */
549 if (XFS_IS_REALTIME_INODE(ip))
550 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip) ? : 1);
551
552 /*
553 * Allow large block sizes to be reported to userspace programs if the
554 * "largeio" mount option is used.
555 *
556 * If compatibility mode is specified, simply return the basic unit of
557 * caching so that we don't get inefficient read/modify/write I/O from
558 * user apps. Otherwise....
559 *
560 * If the underlying volume is a stripe, then return the stripe width in
561 * bytes as the recommended I/O size. It is not a stripe and we've set a
562 * default buffered I/O size, return that, otherwise return the compat
563 * default.
564 */
565 if (xfs_has_large_iosize(mp)) {
566 if (mp->m_swidth)
567 return XFS_FSB_TO_B(mp, mp->m_swidth);
568 if (xfs_has_allocsize(mp))
569 return 1U << mp->m_allocsize_log;
570 }
571
572 return max_t(uint32_t, PAGE_SIZE, mp->m_sb.sb_blocksize);
573}
574
575static void
576xfs_get_atomic_write_attr(
577 struct xfs_inode *ip,
578 unsigned int *unit_min,
579 unsigned int *unit_max)
580{
581 if (!xfs_inode_can_atomicwrite(ip)) {
582 *unit_min = *unit_max = 0;
583 return;
584 }
585
586 *unit_min = *unit_max = ip->i_mount->m_sb.sb_blocksize;
587}
588
589STATIC int
590xfs_vn_getattr(
591 struct mnt_idmap *idmap,
592 const struct path *path,
593 struct kstat *stat,
594 u32 request_mask,
595 unsigned int query_flags)
596{
597 struct inode *inode = d_inode(path->dentry);
598 struct xfs_inode *ip = XFS_I(inode);
599 struct xfs_mount *mp = ip->i_mount;
600 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode);
601 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
602
603 trace_xfs_getattr(ip);
604
605 if (xfs_is_shutdown(mp))
606 return -EIO;
607
608 stat->size = XFS_ISIZE(ip);
609 stat->dev = inode->i_sb->s_dev;
610 stat->mode = inode->i_mode;
611 stat->nlink = inode->i_nlink;
612 stat->uid = vfsuid_into_kuid(vfsuid);
613 stat->gid = vfsgid_into_kgid(vfsgid);
614 stat->ino = ip->i_ino;
615 stat->atime = inode_get_atime(inode);
616
617 fill_mg_cmtime(stat, request_mask, inode);
618
619 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
620
621 if (xfs_has_v3inodes(mp)) {
622 if (request_mask & STATX_BTIME) {
623 stat->result_mask |= STATX_BTIME;
624 stat->btime = ip->i_crtime;
625 }
626 }
627
628 /*
629 * Note: If you add another clause to set an attribute flag, please
630 * update attributes_mask below.
631 */
632 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
633 stat->attributes |= STATX_ATTR_IMMUTABLE;
634 if (ip->i_diflags & XFS_DIFLAG_APPEND)
635 stat->attributes |= STATX_ATTR_APPEND;
636 if (ip->i_diflags & XFS_DIFLAG_NODUMP)
637 stat->attributes |= STATX_ATTR_NODUMP;
638
639 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
640 STATX_ATTR_APPEND |
641 STATX_ATTR_NODUMP);
642
643 switch (inode->i_mode & S_IFMT) {
644 case S_IFBLK:
645 case S_IFCHR:
646 stat->blksize = BLKDEV_IOSIZE;
647 stat->rdev = inode->i_rdev;
648 break;
649 case S_IFREG:
650 if (request_mask & STATX_DIOALIGN) {
651 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
652 struct block_device *bdev = target->bt_bdev;
653
654 stat->result_mask |= STATX_DIOALIGN;
655 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
656 stat->dio_offset_align = bdev_logical_block_size(bdev);
657 }
658 if (request_mask & STATX_WRITE_ATOMIC) {
659 unsigned int unit_min, unit_max;
660
661 xfs_get_atomic_write_attr(ip, &unit_min,
662 &unit_max);
663 generic_fill_statx_atomic_writes(stat,
664 unit_min, unit_max);
665 }
666 fallthrough;
667 default:
668 stat->blksize = xfs_stat_blksize(ip);
669 stat->rdev = 0;
670 break;
671 }
672
673 return 0;
674}
675
676static int
677xfs_vn_change_ok(
678 struct mnt_idmap *idmap,
679 struct dentry *dentry,
680 struct iattr *iattr)
681{
682 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
683
684 if (xfs_is_readonly(mp))
685 return -EROFS;
686
687 if (xfs_is_shutdown(mp))
688 return -EIO;
689
690 return setattr_prepare(idmap, dentry, iattr);
691}
692
693/*
694 * Set non-size attributes of an inode.
695 *
696 * Caution: The caller of this function is responsible for calling
697 * setattr_prepare() or otherwise verifying the change is fine.
698 */
699static int
700xfs_setattr_nonsize(
701 struct mnt_idmap *idmap,
702 struct dentry *dentry,
703 struct xfs_inode *ip,
704 struct iattr *iattr)
705{
706 xfs_mount_t *mp = ip->i_mount;
707 struct inode *inode = VFS_I(ip);
708 int mask = iattr->ia_valid;
709 xfs_trans_t *tp;
710 int error;
711 kuid_t uid = GLOBAL_ROOT_UID;
712 kgid_t gid = GLOBAL_ROOT_GID;
713 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
714 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL;
715
716 ASSERT((mask & ATTR_SIZE) == 0);
717
718 /*
719 * If disk quotas is on, we make sure that the dquots do exist on disk,
720 * before we start any other transactions. Trying to do this later
721 * is messy. We don't care to take a readlock to look at the ids
722 * in inode here, because we can't hold it across the trans_reserve.
723 * If the IDs do change before we take the ilock, we're covered
724 * because the i_*dquot fields will get updated anyway.
725 */
726 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
727 uint qflags = 0;
728
729 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
730 uid = from_vfsuid(idmap, i_user_ns(inode),
731 iattr->ia_vfsuid);
732 qflags |= XFS_QMOPT_UQUOTA;
733 } else {
734 uid = inode->i_uid;
735 }
736 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
737 gid = from_vfsgid(idmap, i_user_ns(inode),
738 iattr->ia_vfsgid);
739 qflags |= XFS_QMOPT_GQUOTA;
740 } else {
741 gid = inode->i_gid;
742 }
743
744 /*
745 * We take a reference when we initialize udqp and gdqp,
746 * so it is important that we never blindly double trip on
747 * the same variable. See xfs_create() for an example.
748 */
749 ASSERT(udqp == NULL);
750 ASSERT(gdqp == NULL);
751 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
752 qflags, &udqp, &gdqp, NULL);
753 if (error)
754 return error;
755 }
756
757 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
758 has_capability_noaudit(current, CAP_FOWNER), &tp);
759 if (error)
760 goto out_dqrele;
761
762 /*
763 * Register quota modifications in the transaction. Must be the owner
764 * or privileged. These IDs could have changed since we last looked at
765 * them. But, we're assured that if the ownership did change while we
766 * didn't have the inode locked, inode's dquot(s) would have changed
767 * also.
768 */
769 if (XFS_IS_UQUOTA_ON(mp) &&
770 i_uid_needs_update(idmap, iattr, inode)) {
771 ASSERT(udqp);
772 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp);
773 }
774 if (XFS_IS_GQUOTA_ON(mp) &&
775 i_gid_needs_update(idmap, iattr, inode)) {
776 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp));
777 ASSERT(gdqp);
778 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp);
779 }
780
781 setattr_copy(idmap, inode, iattr);
782 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
783
784 XFS_STATS_INC(mp, xs_ig_attrchg);
785
786 if (xfs_has_wsync(mp))
787 xfs_trans_set_sync(tp);
788 error = xfs_trans_commit(tp);
789
790 /*
791 * Release any dquot(s) the inode had kept before chown.
792 */
793 xfs_qm_dqrele(old_udqp);
794 xfs_qm_dqrele(old_gdqp);
795 xfs_qm_dqrele(udqp);
796 xfs_qm_dqrele(gdqp);
797
798 if (error)
799 return error;
800
801 /*
802 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
803 * update. We could avoid this with linked transactions
804 * and passing down the transaction pointer all the way
805 * to attr_set. No previous user of the generic
806 * Posix ACL code seems to care about this issue either.
807 */
808 if (mask & ATTR_MODE) {
809 error = posix_acl_chmod(idmap, dentry, inode->i_mode);
810 if (error)
811 return error;
812 }
813
814 return 0;
815
816out_dqrele:
817 xfs_qm_dqrele(udqp);
818 xfs_qm_dqrele(gdqp);
819 return error;
820}
821
822/*
823 * Truncate file. Must have write permission and not be a directory.
824 *
825 * Caution: The caller of this function is responsible for calling
826 * setattr_prepare() or otherwise verifying the change is fine.
827 */
828STATIC int
829xfs_setattr_size(
830 struct mnt_idmap *idmap,
831 struct dentry *dentry,
832 struct xfs_inode *ip,
833 struct iattr *iattr)
834{
835 struct xfs_mount *mp = ip->i_mount;
836 struct inode *inode = VFS_I(ip);
837 xfs_off_t oldsize, newsize;
838 struct xfs_trans *tp;
839 int error;
840 uint lock_flags = 0;
841 uint resblks = 0;
842 bool did_zeroing = false;
843
844 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
845 ASSERT(S_ISREG(inode->i_mode));
846 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
847 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
848
849 oldsize = inode->i_size;
850 newsize = iattr->ia_size;
851
852 /*
853 * Short circuit the truncate case for zero length files.
854 */
855 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
856 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
857 return 0;
858
859 /*
860 * Use the regular setattr path to update the timestamps.
861 */
862 iattr->ia_valid &= ~ATTR_SIZE;
863 return xfs_setattr_nonsize(idmap, dentry, ip, iattr);
864 }
865
866 /*
867 * Make sure that the dquots are attached to the inode.
868 */
869 error = xfs_qm_dqattach(ip);
870 if (error)
871 return error;
872
873 /*
874 * Wait for all direct I/O to complete.
875 */
876 inode_dio_wait(inode);
877
878 /*
879 * File data changes must be complete before we start the transaction to
880 * modify the inode. This needs to be done before joining the inode to
881 * the transaction because the inode cannot be unlocked once it is a
882 * part of the transaction.
883 *
884 * Start with zeroing any data beyond EOF that we may expose on file
885 * extension, or zeroing out the rest of the block on a downward
886 * truncate.
887 */
888 if (newsize > oldsize) {
889 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
890 error = xfs_zero_range(ip, oldsize, newsize - oldsize,
891 &did_zeroing);
892 } else {
893 error = xfs_truncate_page(ip, newsize, &did_zeroing);
894 }
895
896 if (error)
897 return error;
898
899 /*
900 * We've already locked out new page faults, so now we can safely remove
901 * pages from the page cache knowing they won't get refaulted until we
902 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
903 * complete. The truncate_setsize() call also cleans partial EOF page
904 * PTEs on extending truncates and hence ensures sub-page block size
905 * filesystems are correctly handled, too.
906 *
907 * We have to do all the page cache truncate work outside the
908 * transaction context as the "lock" order is page lock->log space
909 * reservation as defined by extent allocation in the writeback path.
910 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
911 * having already truncated the in-memory version of the file (i.e. made
912 * user visible changes). There's not much we can do about this, except
913 * to hope that the caller sees ENOMEM and retries the truncate
914 * operation.
915 *
916 * And we update in-core i_size and truncate page cache beyond newsize
917 * before writeback the [i_disk_size, newsize] range, so we're
918 * guaranteed not to write stale data past the new EOF on truncate down.
919 */
920 truncate_setsize(inode, newsize);
921
922 /*
923 * We are going to log the inode size change in this transaction so
924 * any previous writes that are beyond the on disk EOF and the new
925 * EOF that have not been written out need to be written here. If we
926 * do not write the data out, we expose ourselves to the null files
927 * problem. Note that this includes any block zeroing we did above;
928 * otherwise those blocks may not be zeroed after a crash.
929 */
930 if (did_zeroing ||
931 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
932 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
933 ip->i_disk_size, newsize - 1);
934 if (error)
935 return error;
936 }
937
938 /*
939 * For realtime inode with more than one block rtextsize, we need the
940 * block reservation for bmap btree block allocations/splits that can
941 * happen since it could split the tail written extent and convert the
942 * right beyond EOF one to unwritten.
943 */
944 if (xfs_inode_has_bigrtalloc(ip))
945 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
946
947 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks,
948 0, 0, &tp);
949 if (error)
950 return error;
951
952 lock_flags |= XFS_ILOCK_EXCL;
953 xfs_ilock(ip, XFS_ILOCK_EXCL);
954 xfs_trans_ijoin(tp, ip, 0);
955
956 /*
957 * Only change the c/mtime if we are changing the size or we are
958 * explicitly asked to change it. This handles the semantic difference
959 * between truncate() and ftruncate() as implemented in the VFS.
960 *
961 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
962 * special case where we need to update the times despite not having
963 * these flags set. For all other operations the VFS set these flags
964 * explicitly if it wants a timestamp update.
965 */
966 if (newsize != oldsize &&
967 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
968 iattr->ia_ctime = iattr->ia_mtime =
969 current_time(inode);
970 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
971 }
972
973 /*
974 * The first thing we do is set the size to new_size permanently on
975 * disk. This way we don't have to worry about anyone ever being able
976 * to look at the data being freed even in the face of a crash.
977 * What we're getting around here is the case where we free a block, it
978 * is allocated to another file, it is written to, and then we crash.
979 * If the new data gets written to the file but the log buffers
980 * containing the free and reallocation don't, then we'd end up with
981 * garbage in the blocks being freed. As long as we make the new size
982 * permanent before actually freeing any blocks it doesn't matter if
983 * they get written to.
984 */
985 ip->i_disk_size = newsize;
986 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
987
988 if (newsize <= oldsize) {
989 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
990 if (error)
991 goto out_trans_cancel;
992
993 /*
994 * Truncated "down", so we're removing references to old data
995 * here - if we delay flushing for a long time, we expose
996 * ourselves unduly to the notorious NULL files problem. So,
997 * we mark this inode and flush it when the file is closed,
998 * and do not wait the usual (long) time for writeout.
999 */
1000 xfs_iflags_set(ip, XFS_ITRUNCATED);
1001
1002 /* A truncate down always removes post-EOF blocks. */
1003 xfs_inode_clear_eofblocks_tag(ip);
1004 }
1005
1006 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
1007 setattr_copy(idmap, inode, iattr);
1008 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1009
1010 XFS_STATS_INC(mp, xs_ig_attrchg);
1011
1012 if (xfs_has_wsync(mp))
1013 xfs_trans_set_sync(tp);
1014
1015 error = xfs_trans_commit(tp);
1016out_unlock:
1017 if (lock_flags)
1018 xfs_iunlock(ip, lock_flags);
1019 return error;
1020
1021out_trans_cancel:
1022 xfs_trans_cancel(tp);
1023 goto out_unlock;
1024}
1025
1026int
1027xfs_vn_setattr_size(
1028 struct mnt_idmap *idmap,
1029 struct dentry *dentry,
1030 struct iattr *iattr)
1031{
1032 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1033 int error;
1034
1035 trace_xfs_setattr(ip);
1036
1037 error = xfs_vn_change_ok(idmap, dentry, iattr);
1038 if (error)
1039 return error;
1040 return xfs_setattr_size(idmap, dentry, ip, iattr);
1041}
1042
1043STATIC int
1044xfs_vn_setattr(
1045 struct mnt_idmap *idmap,
1046 struct dentry *dentry,
1047 struct iattr *iattr)
1048{
1049 struct inode *inode = d_inode(dentry);
1050 struct xfs_inode *ip = XFS_I(inode);
1051 int error;
1052
1053 if (iattr->ia_valid & ATTR_SIZE) {
1054 uint iolock;
1055
1056 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1057 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1058
1059 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1060 if (error) {
1061 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1062 return error;
1063 }
1064
1065 error = xfs_vn_setattr_size(idmap, dentry, iattr);
1066 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1067 } else {
1068 trace_xfs_setattr(ip);
1069
1070 error = xfs_vn_change_ok(idmap, dentry, iattr);
1071 if (!error)
1072 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr);
1073 }
1074
1075 return error;
1076}
1077
1078STATIC int
1079xfs_vn_update_time(
1080 struct inode *inode,
1081 int flags)
1082{
1083 struct xfs_inode *ip = XFS_I(inode);
1084 struct xfs_mount *mp = ip->i_mount;
1085 int log_flags = XFS_ILOG_TIMESTAMP;
1086 struct xfs_trans *tp;
1087 int error;
1088 struct timespec64 now;
1089
1090 trace_xfs_update_time(ip);
1091
1092 if (inode->i_sb->s_flags & SB_LAZYTIME) {
1093 if (!((flags & S_VERSION) &&
1094 inode_maybe_inc_iversion(inode, false))) {
1095 generic_update_time(inode, flags);
1096 return 0;
1097 }
1098
1099 /* Capture the iversion update that just occurred */
1100 log_flags |= XFS_ILOG_CORE;
1101 }
1102
1103 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1104 if (error)
1105 return error;
1106
1107 xfs_ilock(ip, XFS_ILOCK_EXCL);
1108 if (flags & (S_CTIME|S_MTIME))
1109 now = inode_set_ctime_current(inode);
1110 else
1111 now = current_time(inode);
1112
1113 if (flags & S_MTIME)
1114 inode_set_mtime_to_ts(inode, now);
1115 if (flags & S_ATIME)
1116 inode_set_atime_to_ts(inode, now);
1117
1118 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1119 xfs_trans_log_inode(tp, ip, log_flags);
1120 return xfs_trans_commit(tp);
1121}
1122
1123STATIC int
1124xfs_vn_fiemap(
1125 struct inode *inode,
1126 struct fiemap_extent_info *fieinfo,
1127 u64 start,
1128 u64 length)
1129{
1130 int error;
1131
1132 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1133 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1134 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1135 error = iomap_fiemap(inode, fieinfo, start, length,
1136 &xfs_xattr_iomap_ops);
1137 } else {
1138 error = iomap_fiemap(inode, fieinfo, start, length,
1139 &xfs_read_iomap_ops);
1140 }
1141 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1142
1143 return error;
1144}
1145
1146STATIC int
1147xfs_vn_tmpfile(
1148 struct mnt_idmap *idmap,
1149 struct inode *dir,
1150 struct file *file,
1151 umode_t mode)
1152{
1153 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file);
1154
1155 return finish_open_simple(file, err);
1156}
1157
1158static const struct inode_operations xfs_inode_operations = {
1159 .get_inode_acl = xfs_get_acl,
1160 .set_acl = xfs_set_acl,
1161 .getattr = xfs_vn_getattr,
1162 .setattr = xfs_vn_setattr,
1163 .listxattr = xfs_vn_listxattr,
1164 .fiemap = xfs_vn_fiemap,
1165 .update_time = xfs_vn_update_time,
1166 .fileattr_get = xfs_fileattr_get,
1167 .fileattr_set = xfs_fileattr_set,
1168};
1169
1170static const struct inode_operations xfs_dir_inode_operations = {
1171 .create = xfs_vn_create,
1172 .lookup = xfs_vn_lookup,
1173 .link = xfs_vn_link,
1174 .unlink = xfs_vn_unlink,
1175 .symlink = xfs_vn_symlink,
1176 .mkdir = xfs_vn_mkdir,
1177 /*
1178 * Yes, XFS uses the same method for rmdir and unlink.
1179 *
1180 * There are some subtile differences deeper in the code,
1181 * but we use S_ISDIR to check for those.
1182 */
1183 .rmdir = xfs_vn_unlink,
1184 .mknod = xfs_vn_mknod,
1185 .rename = xfs_vn_rename,
1186 .get_inode_acl = xfs_get_acl,
1187 .set_acl = xfs_set_acl,
1188 .getattr = xfs_vn_getattr,
1189 .setattr = xfs_vn_setattr,
1190 .listxattr = xfs_vn_listxattr,
1191 .update_time = xfs_vn_update_time,
1192 .tmpfile = xfs_vn_tmpfile,
1193 .fileattr_get = xfs_fileattr_get,
1194 .fileattr_set = xfs_fileattr_set,
1195};
1196
1197static const struct inode_operations xfs_dir_ci_inode_operations = {
1198 .create = xfs_vn_create,
1199 .lookup = xfs_vn_ci_lookup,
1200 .link = xfs_vn_link,
1201 .unlink = xfs_vn_unlink,
1202 .symlink = xfs_vn_symlink,
1203 .mkdir = xfs_vn_mkdir,
1204 /*
1205 * Yes, XFS uses the same method for rmdir and unlink.
1206 *
1207 * There are some subtile differences deeper in the code,
1208 * but we use S_ISDIR to check for those.
1209 */
1210 .rmdir = xfs_vn_unlink,
1211 .mknod = xfs_vn_mknod,
1212 .rename = xfs_vn_rename,
1213 .get_inode_acl = xfs_get_acl,
1214 .set_acl = xfs_set_acl,
1215 .getattr = xfs_vn_getattr,
1216 .setattr = xfs_vn_setattr,
1217 .listxattr = xfs_vn_listxattr,
1218 .update_time = xfs_vn_update_time,
1219 .tmpfile = xfs_vn_tmpfile,
1220 .fileattr_get = xfs_fileattr_get,
1221 .fileattr_set = xfs_fileattr_set,
1222};
1223
1224static const struct inode_operations xfs_symlink_inode_operations = {
1225 .get_link = xfs_vn_get_link,
1226 .getattr = xfs_vn_getattr,
1227 .setattr = xfs_vn_setattr,
1228 .listxattr = xfs_vn_listxattr,
1229 .update_time = xfs_vn_update_time,
1230};
1231
1232/* Figure out if this file actually supports DAX. */
1233static bool
1234xfs_inode_supports_dax(
1235 struct xfs_inode *ip)
1236{
1237 struct xfs_mount *mp = ip->i_mount;
1238
1239 /* Only supported on regular files. */
1240 if (!S_ISREG(VFS_I(ip)->i_mode))
1241 return false;
1242
1243 /* Block size must match page size */
1244 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1245 return false;
1246
1247 /* Device has to support DAX too. */
1248 return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1249}
1250
1251static bool
1252xfs_inode_should_enable_dax(
1253 struct xfs_inode *ip)
1254{
1255 if (!IS_ENABLED(CONFIG_FS_DAX))
1256 return false;
1257 if (xfs_has_dax_never(ip->i_mount))
1258 return false;
1259 if (!xfs_inode_supports_dax(ip))
1260 return false;
1261 if (xfs_has_dax_always(ip->i_mount))
1262 return true;
1263 if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1264 return true;
1265 return false;
1266}
1267
1268void
1269xfs_diflags_to_iflags(
1270 struct xfs_inode *ip,
1271 bool init)
1272{
1273 struct inode *inode = VFS_I(ip);
1274 unsigned int xflags = xfs_ip2xflags(ip);
1275 unsigned int flags = 0;
1276
1277 ASSERT(!(IS_DAX(inode) && init));
1278
1279 if (xflags & FS_XFLAG_IMMUTABLE)
1280 flags |= S_IMMUTABLE;
1281 if (xflags & FS_XFLAG_APPEND)
1282 flags |= S_APPEND;
1283 if (xflags & FS_XFLAG_SYNC)
1284 flags |= S_SYNC;
1285 if (xflags & FS_XFLAG_NOATIME)
1286 flags |= S_NOATIME;
1287 if (init && xfs_inode_should_enable_dax(ip))
1288 flags |= S_DAX;
1289
1290 /*
1291 * S_DAX can only be set during inode initialization and is never set by
1292 * the VFS, so we cannot mask off S_DAX in i_flags.
1293 */
1294 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1295 inode->i_flags |= flags;
1296}
1297
1298/*
1299 * Initialize the Linux inode.
1300 *
1301 * When reading existing inodes from disk this is called directly from xfs_iget,
1302 * when creating a new inode it is called from xfs_init_new_inode after setting
1303 * up the inode. These callers have different criteria for clearing XFS_INEW, so
1304 * leave it up to the caller to deal with unlocking the inode appropriately.
1305 */
1306void
1307xfs_setup_inode(
1308 struct xfs_inode *ip)
1309{
1310 struct inode *inode = &ip->i_vnode;
1311 gfp_t gfp_mask;
1312 bool is_meta = xfs_is_internal_inode(ip);
1313
1314 inode->i_ino = ip->i_ino;
1315 inode->i_state |= I_NEW;
1316
1317 inode_sb_list_add(inode);
1318 /* make the inode look hashed for the writeback code */
1319 inode_fake_hash(inode);
1320
1321 i_size_write(inode, ip->i_disk_size);
1322 xfs_diflags_to_iflags(ip, true);
1323
1324 /*
1325 * Mark our metadata files as private so that LSMs and the ACL code
1326 * don't try to add their own metadata or reason about these files,
1327 * and users cannot ever obtain file handles to them.
1328 */
1329 if (is_meta) {
1330 inode->i_flags |= S_PRIVATE;
1331 inode->i_opflags &= ~IOP_XATTR;
1332 }
1333
1334 if (S_ISDIR(inode->i_mode)) {
1335 /*
1336 * We set the i_rwsem class here to avoid potential races with
1337 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1338 * after a filehandle lookup has already found the inode in
1339 * cache before it has been unlocked via unlock_new_inode().
1340 */
1341 lockdep_set_class(&inode->i_rwsem,
1342 &inode->i_sb->s_type->i_mutex_dir_key);
1343 lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class);
1344 } else {
1345 lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class);
1346 }
1347
1348 /*
1349 * Ensure all page cache allocations are done from GFP_NOFS context to
1350 * prevent direct reclaim recursion back into the filesystem and blowing
1351 * stacks or deadlocking.
1352 */
1353 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1354 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1355
1356 /*
1357 * For real-time inodes update the stable write flags to that of the RT
1358 * device instead of the data device.
1359 */
1360 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip))
1361 xfs_update_stable_writes(ip);
1362
1363 /*
1364 * If there is no attribute fork no ACL can exist on this inode,
1365 * and it can't have any file capabilities attached to it either.
1366 */
1367 if (!xfs_inode_has_attr_fork(ip)) {
1368 inode_has_no_xattr(inode);
1369 cache_no_acl(inode);
1370 }
1371}
1372
1373void
1374xfs_setup_iops(
1375 struct xfs_inode *ip)
1376{
1377 struct inode *inode = &ip->i_vnode;
1378
1379 switch (inode->i_mode & S_IFMT) {
1380 case S_IFREG:
1381 inode->i_op = &xfs_inode_operations;
1382 inode->i_fop = &xfs_file_operations;
1383 if (IS_DAX(inode))
1384 inode->i_mapping->a_ops = &xfs_dax_aops;
1385 else
1386 inode->i_mapping->a_ops = &xfs_address_space_operations;
1387 break;
1388 case S_IFDIR:
1389 if (xfs_has_asciici(XFS_M(inode->i_sb)))
1390 inode->i_op = &xfs_dir_ci_inode_operations;
1391 else
1392 inode->i_op = &xfs_dir_inode_operations;
1393 inode->i_fop = &xfs_dir_file_operations;
1394 break;
1395 case S_IFLNK:
1396 inode->i_op = &xfs_symlink_inode_operations;
1397 break;
1398 default:
1399 inode->i_op = &xfs_inode_operations;
1400 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1401 break;
1402 }
1403}