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_sb.h"
25#include "xfs_ag.h"
26#include "xfs_mount.h"
27#include "xfs_da_format.h"
28#include "xfs_inode.h"
29#include "xfs_bmap.h"
30#include "xfs_bmap_util.h"
31#include "xfs_acl.h"
32#include "xfs_quota.h"
33#include "xfs_error.h"
34#include "xfs_attr.h"
35#include "xfs_trans.h"
36#include "xfs_trace.h"
37#include "xfs_icache.h"
38#include "xfs_symlink.h"
39#include "xfs_da_btree.h"
40#include "xfs_dir2_priv.h"
41#include "xfs_dinode.h"
42#include "xfs_trans_space.h"
43
44#include <linux/capability.h>
45#include <linux/xattr.h>
46#include <linux/namei.h>
47#include <linux/posix_acl.h>
48#include <linux/security.h>
49#include <linux/fiemap.h>
50#include <linux/slab.h>
51
52/*
53 * Directories have different lock order w.r.t. mmap_sem compared to regular
54 * files. This is due to readdir potentially triggering page faults on a user
55 * buffer inside filldir(), and this happens with the ilock on the directory
56 * held. For regular files, the lock order is the other way around - the
57 * mmap_sem is taken during the page fault, and then we lock the ilock to do
58 * block mapping. Hence we need a different class for the directory ilock so
59 * that lockdep can tell them apart.
60 */
61static struct lock_class_key xfs_nondir_ilock_class;
62static struct lock_class_key xfs_dir_ilock_class;
63
64static int
65xfs_initxattrs(
66 struct inode *inode,
67 const struct xattr *xattr_array,
68 void *fs_info)
69{
70 const struct xattr *xattr;
71 struct xfs_inode *ip = XFS_I(inode);
72 int error = 0;
73
74 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
75 error = -xfs_attr_set(ip, xattr->name, xattr->value,
76 xattr->value_len, ATTR_SECURE);
77 if (error < 0)
78 break;
79 }
80 return error;
81}
82
83/*
84 * Hook in SELinux. This is not quite correct yet, what we really need
85 * here (as we do for default ACLs) is a mechanism by which creation of
86 * these attrs can be journalled at inode creation time (along with the
87 * inode, of course, such that log replay can't cause these to be lost).
88 */
89
90STATIC int
91xfs_init_security(
92 struct inode *inode,
93 struct inode *dir,
94 const struct qstr *qstr)
95{
96 return -security_inode_init_security(inode, dir, qstr,
97 &xfs_initxattrs, NULL);
98}
99
100static void
101xfs_dentry_to_name(
102 struct xfs_name *namep,
103 struct dentry *dentry,
104 int mode)
105{
106 namep->name = dentry->d_name.name;
107 namep->len = dentry->d_name.len;
108 namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
109}
110
111STATIC void
112xfs_cleanup_inode(
113 struct inode *dir,
114 struct inode *inode,
115 struct dentry *dentry)
116{
117 struct xfs_name teardown;
118
119 /* Oh, the horror.
120 * If we can't add the ACL or we fail in
121 * xfs_init_security we must back out.
122 * ENOSPC can hit here, among other things.
123 */
124 xfs_dentry_to_name(&teardown, dentry, 0);
125
126 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
127}
128
129STATIC int
130xfs_generic_create(
131 struct inode *dir,
132 struct dentry *dentry,
133 umode_t mode,
134 dev_t rdev,
135 bool tmpfile) /* unnamed file */
136{
137 struct inode *inode;
138 struct xfs_inode *ip = NULL;
139 struct posix_acl *default_acl, *acl;
140 struct xfs_name name;
141 int error;
142
143 /*
144 * Irix uses Missed'em'V split, but doesn't want to see
145 * the upper 5 bits of (14bit) major.
146 */
147 if (S_ISCHR(mode) || S_ISBLK(mode)) {
148 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
149 return -EINVAL;
150 rdev = sysv_encode_dev(rdev);
151 } else {
152 rdev = 0;
153 }
154
155 error = posix_acl_create(dir, &mode, &default_acl, &acl);
156 if (error)
157 return error;
158
159 if (!tmpfile) {
160 xfs_dentry_to_name(&name, dentry, mode);
161 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
162 } else {
163 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
164 }
165 if (unlikely(error))
166 goto out_free_acl;
167
168 inode = VFS_I(ip);
169
170 error = xfs_init_security(inode, dir, &dentry->d_name);
171 if (unlikely(error))
172 goto out_cleanup_inode;
173
174#ifdef CONFIG_XFS_POSIX_ACL
175 if (default_acl) {
176 error = -xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
177 if (error)
178 goto out_cleanup_inode;
179 }
180 if (acl) {
181 error = -xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
182 if (error)
183 goto out_cleanup_inode;
184 }
185#endif
186
187 if (tmpfile)
188 d_tmpfile(dentry, inode);
189 else
190 d_instantiate(dentry, inode);
191
192 out_free_acl:
193 if (default_acl)
194 posix_acl_release(default_acl);
195 if (acl)
196 posix_acl_release(acl);
197 return -error;
198
199 out_cleanup_inode:
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 = old_dentry->d_inode;
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(dentry->d_inode));
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 return 0;
373
374 out_cleanup_inode:
375 xfs_cleanup_inode(dir, inode, dentry);
376 iput(inode);
377 out:
378 return -error;
379}
380
381STATIC int
382xfs_vn_rename(
383 struct inode *odir,
384 struct dentry *odentry,
385 struct inode *ndir,
386 struct dentry *ndentry)
387{
388 struct inode *new_inode = ndentry->d_inode;
389 struct xfs_name oname;
390 struct xfs_name nname;
391
392 xfs_dentry_to_name(&oname, odentry, 0);
393 xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
394
395 return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
396 XFS_I(ndir), &nname, new_inode ?
397 XFS_I(new_inode) : NULL);
398}
399
400/*
401 * careful here - this function can get called recursively, so
402 * we need to be very careful about how much stack we use.
403 * uio is kmalloced for this reason...
404 */
405STATIC void *
406xfs_vn_follow_link(
407 struct dentry *dentry,
408 struct nameidata *nd)
409{
410 char *link;
411 int error = -ENOMEM;
412
413 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
414 if (!link)
415 goto out_err;
416
417 error = -xfs_readlink(XFS_I(dentry->d_inode), link);
418 if (unlikely(error))
419 goto out_kfree;
420
421 nd_set_link(nd, link);
422 return NULL;
423
424 out_kfree:
425 kfree(link);
426 out_err:
427 nd_set_link(nd, ERR_PTR(error));
428 return NULL;
429}
430
431STATIC int
432xfs_vn_getattr(
433 struct vfsmount *mnt,
434 struct dentry *dentry,
435 struct kstat *stat)
436{
437 struct inode *inode = dentry->d_inode;
438 struct xfs_inode *ip = XFS_I(inode);
439 struct xfs_mount *mp = ip->i_mount;
440
441 trace_xfs_getattr(ip);
442
443 if (XFS_FORCED_SHUTDOWN(mp))
444 return -XFS_ERROR(EIO);
445
446 stat->size = XFS_ISIZE(ip);
447 stat->dev = inode->i_sb->s_dev;
448 stat->mode = ip->i_d.di_mode;
449 stat->nlink = ip->i_d.di_nlink;
450 stat->uid = inode->i_uid;
451 stat->gid = inode->i_gid;
452 stat->ino = ip->i_ino;
453 stat->atime = inode->i_atime;
454 stat->mtime = inode->i_mtime;
455 stat->ctime = inode->i_ctime;
456 stat->blocks =
457 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
458
459
460 switch (inode->i_mode & S_IFMT) {
461 case S_IFBLK:
462 case S_IFCHR:
463 stat->blksize = BLKDEV_IOSIZE;
464 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
465 sysv_minor(ip->i_df.if_u2.if_rdev));
466 break;
467 default:
468 if (XFS_IS_REALTIME_INODE(ip)) {
469 /*
470 * If the file blocks are being allocated from a
471 * realtime volume, then return the inode's realtime
472 * extent size or the realtime volume's extent size.
473 */
474 stat->blksize =
475 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
476 } else
477 stat->blksize = xfs_preferred_iosize(mp);
478 stat->rdev = 0;
479 break;
480 }
481
482 return 0;
483}
484
485static void
486xfs_setattr_mode(
487 struct xfs_inode *ip,
488 struct iattr *iattr)
489{
490 struct inode *inode = VFS_I(ip);
491 umode_t mode = iattr->ia_mode;
492
493 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
494
495 ip->i_d.di_mode &= S_IFMT;
496 ip->i_d.di_mode |= mode & ~S_IFMT;
497
498 inode->i_mode &= S_IFMT;
499 inode->i_mode |= mode & ~S_IFMT;
500}
501
502static void
503xfs_setattr_time(
504 struct xfs_inode *ip,
505 struct iattr *iattr)
506{
507 struct inode *inode = VFS_I(ip);
508
509 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
510
511 if (iattr->ia_valid & ATTR_ATIME) {
512 inode->i_atime = iattr->ia_atime;
513 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
514 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
515 }
516 if (iattr->ia_valid & ATTR_CTIME) {
517 inode->i_ctime = iattr->ia_ctime;
518 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
519 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
520 }
521 if (iattr->ia_valid & ATTR_MTIME) {
522 inode->i_mtime = iattr->ia_mtime;
523 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
524 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
525 }
526}
527
528int
529xfs_setattr_nonsize(
530 struct xfs_inode *ip,
531 struct iattr *iattr,
532 int flags)
533{
534 xfs_mount_t *mp = ip->i_mount;
535 struct inode *inode = VFS_I(ip);
536 int mask = iattr->ia_valid;
537 xfs_trans_t *tp;
538 int error;
539 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
540 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
541 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
542 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
543
544 trace_xfs_setattr(ip);
545
546 /* If acls are being inherited, we already have this checked */
547 if (!(flags & XFS_ATTR_NOACL)) {
548 if (mp->m_flags & XFS_MOUNT_RDONLY)
549 return XFS_ERROR(EROFS);
550
551 if (XFS_FORCED_SHUTDOWN(mp))
552 return XFS_ERROR(EIO);
553
554 error = -inode_change_ok(inode, iattr);
555 if (error)
556 return XFS_ERROR(error);
557 }
558
559 ASSERT((mask & ATTR_SIZE) == 0);
560
561 /*
562 * If disk quotas is on, we make sure that the dquots do exist on disk,
563 * before we start any other transactions. Trying to do this later
564 * is messy. We don't care to take a readlock to look at the ids
565 * in inode here, because we can't hold it across the trans_reserve.
566 * If the IDs do change before we take the ilock, we're covered
567 * because the i_*dquot fields will get updated anyway.
568 */
569 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
570 uint qflags = 0;
571
572 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
573 uid = iattr->ia_uid;
574 qflags |= XFS_QMOPT_UQUOTA;
575 } else {
576 uid = inode->i_uid;
577 }
578 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
579 gid = iattr->ia_gid;
580 qflags |= XFS_QMOPT_GQUOTA;
581 } else {
582 gid = inode->i_gid;
583 }
584
585 /*
586 * We take a reference when we initialize udqp and gdqp,
587 * so it is important that we never blindly double trip on
588 * the same variable. See xfs_create() for an example.
589 */
590 ASSERT(udqp == NULL);
591 ASSERT(gdqp == NULL);
592 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
593 xfs_kgid_to_gid(gid),
594 xfs_get_projid(ip),
595 qflags, &udqp, &gdqp, NULL);
596 if (error)
597 return error;
598 }
599
600 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
601 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
602 if (error)
603 goto out_dqrele;
604
605 xfs_ilock(ip, XFS_ILOCK_EXCL);
606
607 /*
608 * Change file ownership. Must be the owner or privileged.
609 */
610 if (mask & (ATTR_UID|ATTR_GID)) {
611 /*
612 * These IDs could have changed since we last looked at them.
613 * But, we're assured that if the ownership did change
614 * while we didn't have the inode locked, inode's dquot(s)
615 * would have changed also.
616 */
617 iuid = inode->i_uid;
618 igid = inode->i_gid;
619 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
620 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
621
622 /*
623 * Do a quota reservation only if uid/gid is actually
624 * going to change.
625 */
626 if (XFS_IS_QUOTA_RUNNING(mp) &&
627 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
628 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
629 ASSERT(tp);
630 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
631 NULL, capable(CAP_FOWNER) ?
632 XFS_QMOPT_FORCE_RES : 0);
633 if (error) /* out of quota */
634 goto out_trans_cancel;
635 }
636 }
637
638 xfs_trans_ijoin(tp, ip, 0);
639
640 /*
641 * Change file ownership. Must be the owner or privileged.
642 */
643 if (mask & (ATTR_UID|ATTR_GID)) {
644 /*
645 * CAP_FSETID overrides the following restrictions:
646 *
647 * The set-user-ID and set-group-ID bits of a file will be
648 * cleared upon successful return from chown()
649 */
650 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
651 !capable(CAP_FSETID))
652 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
653
654 /*
655 * Change the ownerships and register quota modifications
656 * in the transaction.
657 */
658 if (!uid_eq(iuid, uid)) {
659 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
660 ASSERT(mask & ATTR_UID);
661 ASSERT(udqp);
662 olddquot1 = xfs_qm_vop_chown(tp, ip,
663 &ip->i_udquot, udqp);
664 }
665 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
666 inode->i_uid = uid;
667 }
668 if (!gid_eq(igid, gid)) {
669 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
670 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
671 !XFS_IS_PQUOTA_ON(mp));
672 ASSERT(mask & ATTR_GID);
673 ASSERT(gdqp);
674 olddquot2 = xfs_qm_vop_chown(tp, ip,
675 &ip->i_gdquot, gdqp);
676 }
677 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
678 inode->i_gid = gid;
679 }
680 }
681
682 if (mask & ATTR_MODE)
683 xfs_setattr_mode(ip, iattr);
684 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
685 xfs_setattr_time(ip, iattr);
686
687 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
688
689 XFS_STATS_INC(xs_ig_attrchg);
690
691 if (mp->m_flags & XFS_MOUNT_WSYNC)
692 xfs_trans_set_sync(tp);
693 error = xfs_trans_commit(tp, 0);
694
695 xfs_iunlock(ip, XFS_ILOCK_EXCL);
696
697 /*
698 * Release any dquot(s) the inode had kept before chown.
699 */
700 xfs_qm_dqrele(olddquot1);
701 xfs_qm_dqrele(olddquot2);
702 xfs_qm_dqrele(udqp);
703 xfs_qm_dqrele(gdqp);
704
705 if (error)
706 return XFS_ERROR(error);
707
708 /*
709 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
710 * update. We could avoid this with linked transactions
711 * and passing down the transaction pointer all the way
712 * to attr_set. No previous user of the generic
713 * Posix ACL code seems to care about this issue either.
714 */
715 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
716 error = -posix_acl_chmod(inode, inode->i_mode);
717 if (error)
718 return XFS_ERROR(error);
719 }
720
721 return 0;
722
723out_trans_cancel:
724 xfs_trans_cancel(tp, 0);
725 xfs_iunlock(ip, XFS_ILOCK_EXCL);
726out_dqrele:
727 xfs_qm_dqrele(udqp);
728 xfs_qm_dqrele(gdqp);
729 return error;
730}
731
732/*
733 * Truncate file. Must have write permission and not be a directory.
734 */
735int
736xfs_setattr_size(
737 struct xfs_inode *ip,
738 struct iattr *iattr)
739{
740 struct xfs_mount *mp = ip->i_mount;
741 struct inode *inode = VFS_I(ip);
742 xfs_off_t oldsize, newsize;
743 struct xfs_trans *tp;
744 int error;
745 uint lock_flags = 0;
746 uint commit_flags = 0;
747
748 trace_xfs_setattr(ip);
749
750 if (mp->m_flags & XFS_MOUNT_RDONLY)
751 return XFS_ERROR(EROFS);
752
753 if (XFS_FORCED_SHUTDOWN(mp))
754 return XFS_ERROR(EIO);
755
756 error = -inode_change_ok(inode, iattr);
757 if (error)
758 return XFS_ERROR(error);
759
760 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
761 ASSERT(S_ISREG(ip->i_d.di_mode));
762 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
763 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
764
765 oldsize = inode->i_size;
766 newsize = iattr->ia_size;
767
768 /*
769 * Short circuit the truncate case for zero length files.
770 */
771 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
772 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
773 return 0;
774
775 /*
776 * Use the regular setattr path to update the timestamps.
777 */
778 iattr->ia_valid &= ~ATTR_SIZE;
779 return xfs_setattr_nonsize(ip, iattr, 0);
780 }
781
782 /*
783 * Make sure that the dquots are attached to the inode.
784 */
785 error = xfs_qm_dqattach(ip, 0);
786 if (error)
787 return error;
788
789 /*
790 * Now we can make the changes. Before we join the inode to the
791 * transaction, take care of the part of the truncation that must be
792 * done without the inode lock. This needs to be done before joining
793 * the inode to the transaction, because the inode cannot be unlocked
794 * once it is a part of the transaction.
795 */
796 if (newsize > oldsize) {
797 /*
798 * Do the first part of growing a file: zero any data in the
799 * last block that is beyond the old EOF. We need to do this
800 * before the inode is joined to the transaction to modify
801 * i_size.
802 */
803 error = xfs_zero_eof(ip, newsize, oldsize);
804 if (error)
805 return error;
806 }
807
808 /*
809 * We are going to log the inode size change in this transaction so
810 * any previous writes that are beyond the on disk EOF and the new
811 * EOF that have not been written out need to be written here. If we
812 * do not write the data out, we expose ourselves to the null files
813 * problem.
814 *
815 * Only flush from the on disk size to the smaller of the in memory
816 * file size or the new size as that's the range we really care about
817 * here and prevents waiting for other data not within the range we
818 * care about here.
819 */
820 if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
821 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
822 ip->i_d.di_size, newsize);
823 if (error)
824 return error;
825 }
826
827 /*
828 * Wait for all direct I/O to complete.
829 */
830 inode_dio_wait(inode);
831
832 error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
833 if (error)
834 return error;
835
836 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
837 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
838 if (error)
839 goto out_trans_cancel;
840
841 truncate_setsize(inode, newsize);
842
843 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
844 lock_flags |= XFS_ILOCK_EXCL;
845
846 xfs_ilock(ip, XFS_ILOCK_EXCL);
847
848 xfs_trans_ijoin(tp, ip, 0);
849
850 /*
851 * Only change the c/mtime if we are changing the size or we are
852 * explicitly asked to change it. This handles the semantic difference
853 * between truncate() and ftruncate() as implemented in the VFS.
854 *
855 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
856 * special case where we need to update the times despite not having
857 * these flags set. For all other operations the VFS set these flags
858 * explicitly if it wants a timestamp update.
859 */
860 if (newsize != oldsize &&
861 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
862 iattr->ia_ctime = iattr->ia_mtime =
863 current_fs_time(inode->i_sb);
864 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
865 }
866
867 /*
868 * The first thing we do is set the size to new_size permanently on
869 * disk. This way we don't have to worry about anyone ever being able
870 * to look at the data being freed even in the face of a crash.
871 * What we're getting around here is the case where we free a block, it
872 * is allocated to another file, it is written to, and then we crash.
873 * If the new data gets written to the file but the log buffers
874 * containing the free and reallocation don't, then we'd end up with
875 * garbage in the blocks being freed. As long as we make the new size
876 * permanent before actually freeing any blocks it doesn't matter if
877 * they get written to.
878 */
879 ip->i_d.di_size = newsize;
880 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
881
882 if (newsize <= oldsize) {
883 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
884 if (error)
885 goto out_trans_abort;
886
887 /*
888 * Truncated "down", so we're removing references to old data
889 * here - if we delay flushing for a long time, we expose
890 * ourselves unduly to the notorious NULL files problem. So,
891 * we mark this inode and flush it when the file is closed,
892 * and do not wait the usual (long) time for writeout.
893 */
894 xfs_iflags_set(ip, XFS_ITRUNCATED);
895
896 /* A truncate down always removes post-EOF blocks. */
897 xfs_inode_clear_eofblocks_tag(ip);
898 }
899
900 if (iattr->ia_valid & ATTR_MODE)
901 xfs_setattr_mode(ip, iattr);
902 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
903 xfs_setattr_time(ip, iattr);
904
905 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
906
907 XFS_STATS_INC(xs_ig_attrchg);
908
909 if (mp->m_flags & XFS_MOUNT_WSYNC)
910 xfs_trans_set_sync(tp);
911
912 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
913out_unlock:
914 if (lock_flags)
915 xfs_iunlock(ip, lock_flags);
916 return error;
917
918out_trans_abort:
919 commit_flags |= XFS_TRANS_ABORT;
920out_trans_cancel:
921 xfs_trans_cancel(tp, commit_flags);
922 goto out_unlock;
923}
924
925STATIC int
926xfs_vn_setattr(
927 struct dentry *dentry,
928 struct iattr *iattr)
929{
930 struct xfs_inode *ip = XFS_I(dentry->d_inode);
931 int error;
932
933 if (iattr->ia_valid & ATTR_SIZE) {
934 xfs_ilock(ip, XFS_IOLOCK_EXCL);
935 error = xfs_setattr_size(ip, iattr);
936 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
937 } else {
938 error = xfs_setattr_nonsize(ip, iattr, 0);
939 }
940
941 return -error;
942}
943
944STATIC int
945xfs_vn_update_time(
946 struct inode *inode,
947 struct timespec *now,
948 int flags)
949{
950 struct xfs_inode *ip = XFS_I(inode);
951 struct xfs_mount *mp = ip->i_mount;
952 struct xfs_trans *tp;
953 int error;
954
955 trace_xfs_update_time(ip);
956
957 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
958 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
959 if (error) {
960 xfs_trans_cancel(tp, 0);
961 return -error;
962 }
963
964 xfs_ilock(ip, XFS_ILOCK_EXCL);
965 if (flags & S_CTIME) {
966 inode->i_ctime = *now;
967 ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
968 ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
969 }
970 if (flags & S_MTIME) {
971 inode->i_mtime = *now;
972 ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
973 ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
974 }
975 if (flags & S_ATIME) {
976 inode->i_atime = *now;
977 ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
978 ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
979 }
980 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
981 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
982 return -xfs_trans_commit(tp, 0);
983}
984
985#define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
986
987/*
988 * Call fiemap helper to fill in user data.
989 * Returns positive errors to xfs_getbmap.
990 */
991STATIC int
992xfs_fiemap_format(
993 void **arg,
994 struct getbmapx *bmv,
995 int *full)
996{
997 int error;
998 struct fiemap_extent_info *fieinfo = *arg;
999 u32 fiemap_flags = 0;
1000 u64 logical, physical, length;
1001
1002 /* Do nothing for a hole */
1003 if (bmv->bmv_block == -1LL)
1004 return 0;
1005
1006 logical = BBTOB(bmv->bmv_offset);
1007 physical = BBTOB(bmv->bmv_block);
1008 length = BBTOB(bmv->bmv_length);
1009
1010 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1011 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1012 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1013 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1014 FIEMAP_EXTENT_UNKNOWN);
1015 physical = 0; /* no block yet */
1016 }
1017 if (bmv->bmv_oflags & BMV_OF_LAST)
1018 fiemap_flags |= FIEMAP_EXTENT_LAST;
1019
1020 error = fiemap_fill_next_extent(fieinfo, logical, physical,
1021 length, fiemap_flags);
1022 if (error > 0) {
1023 error = 0;
1024 *full = 1; /* user array now full */
1025 }
1026
1027 return -error;
1028}
1029
1030STATIC int
1031xfs_vn_fiemap(
1032 struct inode *inode,
1033 struct fiemap_extent_info *fieinfo,
1034 u64 start,
1035 u64 length)
1036{
1037 xfs_inode_t *ip = XFS_I(inode);
1038 struct getbmapx bm;
1039 int error;
1040
1041 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1042 if (error)
1043 return error;
1044
1045 /* Set up bmap header for xfs internal routine */
1046 bm.bmv_offset = BTOBB(start);
1047 /* Special case for whole file */
1048 if (length == FIEMAP_MAX_OFFSET)
1049 bm.bmv_length = -1LL;
1050 else
1051 bm.bmv_length = BTOBB(length);
1052
1053 /* We add one because in getbmap world count includes the header */
1054 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1055 fieinfo->fi_extents_max + 1;
1056 bm.bmv_count = min_t(__s32, bm.bmv_count,
1057 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1058 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1059 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1060 bm.bmv_iflags |= BMV_IF_ATTRFORK;
1061 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1062 bm.bmv_iflags |= BMV_IF_DELALLOC;
1063
1064 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1065 if (error)
1066 return -error;
1067
1068 return 0;
1069}
1070
1071STATIC int
1072xfs_vn_tmpfile(
1073 struct inode *dir,
1074 struct dentry *dentry,
1075 umode_t mode)
1076{
1077 return xfs_generic_create(dir, dentry, mode, 0, true);
1078}
1079
1080static const struct inode_operations xfs_inode_operations = {
1081 .get_acl = xfs_get_acl,
1082 .set_acl = xfs_set_acl,
1083 .getattr = xfs_vn_getattr,
1084 .setattr = xfs_vn_setattr,
1085 .setxattr = generic_setxattr,
1086 .getxattr = generic_getxattr,
1087 .removexattr = generic_removexattr,
1088 .listxattr = xfs_vn_listxattr,
1089 .fiemap = xfs_vn_fiemap,
1090 .update_time = xfs_vn_update_time,
1091};
1092
1093static const struct inode_operations xfs_dir_inode_operations = {
1094 .create = xfs_vn_create,
1095 .lookup = xfs_vn_lookup,
1096 .link = xfs_vn_link,
1097 .unlink = xfs_vn_unlink,
1098 .symlink = xfs_vn_symlink,
1099 .mkdir = xfs_vn_mkdir,
1100 /*
1101 * Yes, XFS uses the same method for rmdir and unlink.
1102 *
1103 * There are some subtile differences deeper in the code,
1104 * but we use S_ISDIR to check for those.
1105 */
1106 .rmdir = xfs_vn_unlink,
1107 .mknod = xfs_vn_mknod,
1108 .rename = xfs_vn_rename,
1109 .get_acl = xfs_get_acl,
1110 .set_acl = xfs_set_acl,
1111 .getattr = xfs_vn_getattr,
1112 .setattr = xfs_vn_setattr,
1113 .setxattr = generic_setxattr,
1114 .getxattr = generic_getxattr,
1115 .removexattr = generic_removexattr,
1116 .listxattr = xfs_vn_listxattr,
1117 .update_time = xfs_vn_update_time,
1118 .tmpfile = xfs_vn_tmpfile,
1119};
1120
1121static const struct inode_operations xfs_dir_ci_inode_operations = {
1122 .create = xfs_vn_create,
1123 .lookup = xfs_vn_ci_lookup,
1124 .link = xfs_vn_link,
1125 .unlink = xfs_vn_unlink,
1126 .symlink = xfs_vn_symlink,
1127 .mkdir = xfs_vn_mkdir,
1128 /*
1129 * Yes, XFS uses the same method for rmdir and unlink.
1130 *
1131 * There are some subtile differences deeper in the code,
1132 * but we use S_ISDIR to check for those.
1133 */
1134 .rmdir = xfs_vn_unlink,
1135 .mknod = xfs_vn_mknod,
1136 .rename = xfs_vn_rename,
1137 .get_acl = xfs_get_acl,
1138 .set_acl = xfs_set_acl,
1139 .getattr = xfs_vn_getattr,
1140 .setattr = xfs_vn_setattr,
1141 .setxattr = generic_setxattr,
1142 .getxattr = generic_getxattr,
1143 .removexattr = generic_removexattr,
1144 .listxattr = xfs_vn_listxattr,
1145 .update_time = xfs_vn_update_time,
1146 .tmpfile = xfs_vn_tmpfile,
1147};
1148
1149static const struct inode_operations xfs_symlink_inode_operations = {
1150 .readlink = generic_readlink,
1151 .follow_link = xfs_vn_follow_link,
1152 .put_link = kfree_put_link,
1153 .getattr = xfs_vn_getattr,
1154 .setattr = xfs_vn_setattr,
1155 .setxattr = generic_setxattr,
1156 .getxattr = generic_getxattr,
1157 .removexattr = generic_removexattr,
1158 .listxattr = xfs_vn_listxattr,
1159 .update_time = xfs_vn_update_time,
1160};
1161
1162STATIC void
1163xfs_diflags_to_iflags(
1164 struct inode *inode,
1165 struct xfs_inode *ip)
1166{
1167 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1168 inode->i_flags |= S_IMMUTABLE;
1169 else
1170 inode->i_flags &= ~S_IMMUTABLE;
1171 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1172 inode->i_flags |= S_APPEND;
1173 else
1174 inode->i_flags &= ~S_APPEND;
1175 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1176 inode->i_flags |= S_SYNC;
1177 else
1178 inode->i_flags &= ~S_SYNC;
1179 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1180 inode->i_flags |= S_NOATIME;
1181 else
1182 inode->i_flags &= ~S_NOATIME;
1183}
1184
1185/*
1186 * Initialize the Linux inode, set up the operation vectors and
1187 * unlock the inode.
1188 *
1189 * When reading existing inodes from disk this is called directly
1190 * from xfs_iget, when creating a new inode it is called from
1191 * xfs_ialloc after setting up the inode.
1192 *
1193 * We are always called with an uninitialised linux inode here.
1194 * We need to initialise the necessary fields and take a reference
1195 * on it.
1196 */
1197void
1198xfs_setup_inode(
1199 struct xfs_inode *ip)
1200{
1201 struct inode *inode = &ip->i_vnode;
1202 gfp_t gfp_mask;
1203
1204 inode->i_ino = ip->i_ino;
1205 inode->i_state = I_NEW;
1206
1207 inode_sb_list_add(inode);
1208 /* make the inode look hashed for the writeback code */
1209 hlist_add_fake(&inode->i_hash);
1210
1211 inode->i_mode = ip->i_d.di_mode;
1212 set_nlink(inode, ip->i_d.di_nlink);
1213 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1214 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1215
1216 switch (inode->i_mode & S_IFMT) {
1217 case S_IFBLK:
1218 case S_IFCHR:
1219 inode->i_rdev =
1220 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1221 sysv_minor(ip->i_df.if_u2.if_rdev));
1222 break;
1223 default:
1224 inode->i_rdev = 0;
1225 break;
1226 }
1227
1228 inode->i_generation = ip->i_d.di_gen;
1229 i_size_write(inode, ip->i_d.di_size);
1230 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
1231 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
1232 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
1233 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
1234 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
1235 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
1236 xfs_diflags_to_iflags(inode, ip);
1237
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 switch (inode->i_mode & S_IFMT) {
1241 case S_IFREG:
1242 inode->i_op = &xfs_inode_operations;
1243 inode->i_fop = &xfs_file_operations;
1244 inode->i_mapping->a_ops = &xfs_address_space_operations;
1245 break;
1246 case S_IFDIR:
1247 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1248 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1249 inode->i_op = &xfs_dir_ci_inode_operations;
1250 else
1251 inode->i_op = &xfs_dir_inode_operations;
1252 inode->i_fop = &xfs_dir_file_operations;
1253 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1254 break;
1255 case S_IFLNK:
1256 inode->i_op = &xfs_symlink_inode_operations;
1257 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1258 inode->i_mapping->a_ops = &xfs_address_space_operations;
1259 break;
1260 default:
1261 inode->i_op = &xfs_inode_operations;
1262 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1263 break;
1264 }
1265
1266 /*
1267 * Ensure all page cache allocations are done from GFP_NOFS context to
1268 * prevent direct reclaim recursion back into the filesystem and blowing
1269 * stacks or deadlocking.
1270 */
1271 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1272 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1273
1274 /*
1275 * If there is no attribute fork no ACL can exist on this inode,
1276 * and it can't have any file capabilities attached to it either.
1277 */
1278 if (!XFS_IFORK_Q(ip)) {
1279 inode_has_no_xattr(inode);
1280 cache_no_acl(inode);
1281 }
1282
1283 xfs_iflags_clear(ip, XFS_INEW);
1284 barrier();
1285
1286 unlock_new_inode(inode);
1287}
1/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_acl.h"
21#include "xfs_log.h"
22#include "xfs_trans.h"
23#include "xfs_sb.h"
24#include "xfs_ag.h"
25#include "xfs_alloc.h"
26#include "xfs_quota.h"
27#include "xfs_mount.h"
28#include "xfs_bmap_btree.h"
29#include "xfs_dinode.h"
30#include "xfs_inode.h"
31#include "xfs_bmap.h"
32#include "xfs_rtalloc.h"
33#include "xfs_error.h"
34#include "xfs_itable.h"
35#include "xfs_attr.h"
36#include "xfs_buf_item.h"
37#include "xfs_utils.h"
38#include "xfs_vnodeops.h"
39#include "xfs_inode_item.h"
40#include "xfs_trace.h"
41
42#include <linux/capability.h>
43#include <linux/xattr.h>
44#include <linux/namei.h>
45#include <linux/posix_acl.h>
46#include <linux/security.h>
47#include <linux/fiemap.h>
48#include <linux/slab.h>
49
50static int
51xfs_initxattrs(
52 struct inode *inode,
53 const struct xattr *xattr_array,
54 void *fs_info)
55{
56 const struct xattr *xattr;
57 struct xfs_inode *ip = XFS_I(inode);
58 int error = 0;
59
60 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
61 error = xfs_attr_set(ip, xattr->name, xattr->value,
62 xattr->value_len, ATTR_SECURE);
63 if (error < 0)
64 break;
65 }
66 return error;
67}
68
69/*
70 * Hook in SELinux. This is not quite correct yet, what we really need
71 * here (as we do for default ACLs) is a mechanism by which creation of
72 * these attrs can be journalled at inode creation time (along with the
73 * inode, of course, such that log replay can't cause these to be lost).
74 */
75
76STATIC int
77xfs_init_security(
78 struct inode *inode,
79 struct inode *dir,
80 const struct qstr *qstr)
81{
82 return security_inode_init_security(inode, dir, qstr,
83 &xfs_initxattrs, NULL);
84}
85
86static void
87xfs_dentry_to_name(
88 struct xfs_name *namep,
89 struct dentry *dentry)
90{
91 namep->name = dentry->d_name.name;
92 namep->len = dentry->d_name.len;
93}
94
95STATIC void
96xfs_cleanup_inode(
97 struct inode *dir,
98 struct inode *inode,
99 struct dentry *dentry)
100{
101 struct xfs_name teardown;
102
103 /* Oh, the horror.
104 * If we can't add the ACL or we fail in
105 * xfs_init_security we must back out.
106 * ENOSPC can hit here, among other things.
107 */
108 xfs_dentry_to_name(&teardown, dentry);
109
110 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
111 iput(inode);
112}
113
114STATIC int
115xfs_vn_mknod(
116 struct inode *dir,
117 struct dentry *dentry,
118 umode_t mode,
119 dev_t rdev)
120{
121 struct inode *inode;
122 struct xfs_inode *ip = NULL;
123 struct posix_acl *default_acl = NULL;
124 struct xfs_name name;
125 int error;
126
127 /*
128 * Irix uses Missed'em'V split, but doesn't want to see
129 * the upper 5 bits of (14bit) major.
130 */
131 if (S_ISCHR(mode) || S_ISBLK(mode)) {
132 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
133 return -EINVAL;
134 rdev = sysv_encode_dev(rdev);
135 } else {
136 rdev = 0;
137 }
138
139 if (IS_POSIXACL(dir)) {
140 default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
141 if (IS_ERR(default_acl))
142 return PTR_ERR(default_acl);
143
144 if (!default_acl)
145 mode &= ~current_umask();
146 }
147
148 xfs_dentry_to_name(&name, dentry);
149 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
150 if (unlikely(error))
151 goto out_free_acl;
152
153 inode = VFS_I(ip);
154
155 error = xfs_init_security(inode, dir, &dentry->d_name);
156 if (unlikely(error))
157 goto out_cleanup_inode;
158
159 if (default_acl) {
160 error = -xfs_inherit_acl(inode, default_acl);
161 default_acl = NULL;
162 if (unlikely(error))
163 goto out_cleanup_inode;
164 }
165
166
167 d_instantiate(dentry, inode);
168 return -error;
169
170 out_cleanup_inode:
171 xfs_cleanup_inode(dir, inode, dentry);
172 out_free_acl:
173 posix_acl_release(default_acl);
174 return -error;
175}
176
177STATIC int
178xfs_vn_create(
179 struct inode *dir,
180 struct dentry *dentry,
181 umode_t mode,
182 struct nameidata *nd)
183{
184 return xfs_vn_mknod(dir, dentry, mode, 0);
185}
186
187STATIC int
188xfs_vn_mkdir(
189 struct inode *dir,
190 struct dentry *dentry,
191 umode_t mode)
192{
193 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
194}
195
196STATIC struct dentry *
197xfs_vn_lookup(
198 struct inode *dir,
199 struct dentry *dentry,
200 struct nameidata *nd)
201{
202 struct xfs_inode *cip;
203 struct xfs_name name;
204 int error;
205
206 if (dentry->d_name.len >= MAXNAMELEN)
207 return ERR_PTR(-ENAMETOOLONG);
208
209 xfs_dentry_to_name(&name, dentry);
210 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
211 if (unlikely(error)) {
212 if (unlikely(error != ENOENT))
213 return ERR_PTR(-error);
214 d_add(dentry, NULL);
215 return NULL;
216 }
217
218 return d_splice_alias(VFS_I(cip), dentry);
219}
220
221STATIC struct dentry *
222xfs_vn_ci_lookup(
223 struct inode *dir,
224 struct dentry *dentry,
225 struct nameidata *nd)
226{
227 struct xfs_inode *ip;
228 struct xfs_name xname;
229 struct xfs_name ci_name;
230 struct qstr dname;
231 int error;
232
233 if (dentry->d_name.len >= MAXNAMELEN)
234 return ERR_PTR(-ENAMETOOLONG);
235
236 xfs_dentry_to_name(&xname, dentry);
237 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
238 if (unlikely(error)) {
239 if (unlikely(error != ENOENT))
240 return ERR_PTR(-error);
241 /*
242 * call d_add(dentry, NULL) here when d_drop_negative_children
243 * is called in xfs_vn_mknod (ie. allow negative dentries
244 * with CI filesystems).
245 */
246 return NULL;
247 }
248
249 /* if exact match, just splice and exit */
250 if (!ci_name.name)
251 return d_splice_alias(VFS_I(ip), dentry);
252
253 /* else case-insensitive match... */
254 dname.name = ci_name.name;
255 dname.len = ci_name.len;
256 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
257 kmem_free(ci_name.name);
258 return dentry;
259}
260
261STATIC int
262xfs_vn_link(
263 struct dentry *old_dentry,
264 struct inode *dir,
265 struct dentry *dentry)
266{
267 struct inode *inode = old_dentry->d_inode;
268 struct xfs_name name;
269 int error;
270
271 xfs_dentry_to_name(&name, dentry);
272
273 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
274 if (unlikely(error))
275 return -error;
276
277 ihold(inode);
278 d_instantiate(dentry, inode);
279 return 0;
280}
281
282STATIC int
283xfs_vn_unlink(
284 struct inode *dir,
285 struct dentry *dentry)
286{
287 struct xfs_name name;
288 int error;
289
290 xfs_dentry_to_name(&name, dentry);
291
292 error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
293 if (error)
294 return error;
295
296 /*
297 * With unlink, the VFS makes the dentry "negative": no inode,
298 * but still hashed. This is incompatible with case-insensitive
299 * mode, so invalidate (unhash) the dentry in CI-mode.
300 */
301 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
302 d_invalidate(dentry);
303 return 0;
304}
305
306STATIC int
307xfs_vn_symlink(
308 struct inode *dir,
309 struct dentry *dentry,
310 const char *symname)
311{
312 struct inode *inode;
313 struct xfs_inode *cip = NULL;
314 struct xfs_name name;
315 int error;
316 umode_t mode;
317
318 mode = S_IFLNK |
319 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
320 xfs_dentry_to_name(&name, dentry);
321
322 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
323 if (unlikely(error))
324 goto out;
325
326 inode = VFS_I(cip);
327
328 error = xfs_init_security(inode, dir, &dentry->d_name);
329 if (unlikely(error))
330 goto out_cleanup_inode;
331
332 d_instantiate(dentry, inode);
333 return 0;
334
335 out_cleanup_inode:
336 xfs_cleanup_inode(dir, inode, dentry);
337 out:
338 return -error;
339}
340
341STATIC int
342xfs_vn_rename(
343 struct inode *odir,
344 struct dentry *odentry,
345 struct inode *ndir,
346 struct dentry *ndentry)
347{
348 struct inode *new_inode = ndentry->d_inode;
349 struct xfs_name oname;
350 struct xfs_name nname;
351
352 xfs_dentry_to_name(&oname, odentry);
353 xfs_dentry_to_name(&nname, ndentry);
354
355 return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
356 XFS_I(ndir), &nname, new_inode ?
357 XFS_I(new_inode) : NULL);
358}
359
360/*
361 * careful here - this function can get called recursively, so
362 * we need to be very careful about how much stack we use.
363 * uio is kmalloced for this reason...
364 */
365STATIC void *
366xfs_vn_follow_link(
367 struct dentry *dentry,
368 struct nameidata *nd)
369{
370 char *link;
371 int error = -ENOMEM;
372
373 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
374 if (!link)
375 goto out_err;
376
377 error = -xfs_readlink(XFS_I(dentry->d_inode), link);
378 if (unlikely(error))
379 goto out_kfree;
380
381 nd_set_link(nd, link);
382 return NULL;
383
384 out_kfree:
385 kfree(link);
386 out_err:
387 nd_set_link(nd, ERR_PTR(error));
388 return NULL;
389}
390
391STATIC void
392xfs_vn_put_link(
393 struct dentry *dentry,
394 struct nameidata *nd,
395 void *p)
396{
397 char *s = nd_get_link(nd);
398
399 if (!IS_ERR(s))
400 kfree(s);
401}
402
403STATIC int
404xfs_vn_getattr(
405 struct vfsmount *mnt,
406 struct dentry *dentry,
407 struct kstat *stat)
408{
409 struct inode *inode = dentry->d_inode;
410 struct xfs_inode *ip = XFS_I(inode);
411 struct xfs_mount *mp = ip->i_mount;
412
413 trace_xfs_getattr(ip);
414
415 if (XFS_FORCED_SHUTDOWN(mp))
416 return -XFS_ERROR(EIO);
417
418 stat->size = XFS_ISIZE(ip);
419 stat->dev = inode->i_sb->s_dev;
420 stat->mode = ip->i_d.di_mode;
421 stat->nlink = ip->i_d.di_nlink;
422 stat->uid = ip->i_d.di_uid;
423 stat->gid = ip->i_d.di_gid;
424 stat->ino = ip->i_ino;
425 stat->atime = inode->i_atime;
426 stat->mtime = inode->i_mtime;
427 stat->ctime = inode->i_ctime;
428 stat->blocks =
429 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
430
431
432 switch (inode->i_mode & S_IFMT) {
433 case S_IFBLK:
434 case S_IFCHR:
435 stat->blksize = BLKDEV_IOSIZE;
436 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
437 sysv_minor(ip->i_df.if_u2.if_rdev));
438 break;
439 default:
440 if (XFS_IS_REALTIME_INODE(ip)) {
441 /*
442 * If the file blocks are being allocated from a
443 * realtime volume, then return the inode's realtime
444 * extent size or the realtime volume's extent size.
445 */
446 stat->blksize =
447 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
448 } else
449 stat->blksize = xfs_preferred_iosize(mp);
450 stat->rdev = 0;
451 break;
452 }
453
454 return 0;
455}
456
457int
458xfs_setattr_nonsize(
459 struct xfs_inode *ip,
460 struct iattr *iattr,
461 int flags)
462{
463 xfs_mount_t *mp = ip->i_mount;
464 struct inode *inode = VFS_I(ip);
465 int mask = iattr->ia_valid;
466 xfs_trans_t *tp;
467 int error;
468 uid_t uid = 0, iuid = 0;
469 gid_t gid = 0, igid = 0;
470 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
471 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
472
473 trace_xfs_setattr(ip);
474
475 if (mp->m_flags & XFS_MOUNT_RDONLY)
476 return XFS_ERROR(EROFS);
477
478 if (XFS_FORCED_SHUTDOWN(mp))
479 return XFS_ERROR(EIO);
480
481 error = -inode_change_ok(inode, iattr);
482 if (error)
483 return XFS_ERROR(error);
484
485 ASSERT((mask & ATTR_SIZE) == 0);
486
487 /*
488 * If disk quotas is on, we make sure that the dquots do exist on disk,
489 * before we start any other transactions. Trying to do this later
490 * is messy. We don't care to take a readlock to look at the ids
491 * in inode here, because we can't hold it across the trans_reserve.
492 * If the IDs do change before we take the ilock, we're covered
493 * because the i_*dquot fields will get updated anyway.
494 */
495 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
496 uint qflags = 0;
497
498 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
499 uid = iattr->ia_uid;
500 qflags |= XFS_QMOPT_UQUOTA;
501 } else {
502 uid = ip->i_d.di_uid;
503 }
504 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
505 gid = iattr->ia_gid;
506 qflags |= XFS_QMOPT_GQUOTA;
507 } else {
508 gid = ip->i_d.di_gid;
509 }
510
511 /*
512 * We take a reference when we initialize udqp and gdqp,
513 * so it is important that we never blindly double trip on
514 * the same variable. See xfs_create() for an example.
515 */
516 ASSERT(udqp == NULL);
517 ASSERT(gdqp == NULL);
518 error = xfs_qm_vop_dqalloc(ip, uid, gid, xfs_get_projid(ip),
519 qflags, &udqp, &gdqp);
520 if (error)
521 return error;
522 }
523
524 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
525 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0);
526 if (error)
527 goto out_dqrele;
528
529 xfs_ilock(ip, XFS_ILOCK_EXCL);
530
531 /*
532 * Change file ownership. Must be the owner or privileged.
533 */
534 if (mask & (ATTR_UID|ATTR_GID)) {
535 /*
536 * These IDs could have changed since we last looked at them.
537 * But, we're assured that if the ownership did change
538 * while we didn't have the inode locked, inode's dquot(s)
539 * would have changed also.
540 */
541 iuid = ip->i_d.di_uid;
542 igid = ip->i_d.di_gid;
543 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
544 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
545
546 /*
547 * Do a quota reservation only if uid/gid is actually
548 * going to change.
549 */
550 if (XFS_IS_QUOTA_RUNNING(mp) &&
551 ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
552 (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
553 ASSERT(tp);
554 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
555 capable(CAP_FOWNER) ?
556 XFS_QMOPT_FORCE_RES : 0);
557 if (error) /* out of quota */
558 goto out_trans_cancel;
559 }
560 }
561
562 xfs_trans_ijoin(tp, ip, 0);
563
564 /*
565 * Change file ownership. Must be the owner or privileged.
566 */
567 if (mask & (ATTR_UID|ATTR_GID)) {
568 /*
569 * CAP_FSETID overrides the following restrictions:
570 *
571 * The set-user-ID and set-group-ID bits of a file will be
572 * cleared upon successful return from chown()
573 */
574 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
575 !capable(CAP_FSETID))
576 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
577
578 /*
579 * Change the ownerships and register quota modifications
580 * in the transaction.
581 */
582 if (iuid != uid) {
583 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
584 ASSERT(mask & ATTR_UID);
585 ASSERT(udqp);
586 olddquot1 = xfs_qm_vop_chown(tp, ip,
587 &ip->i_udquot, udqp);
588 }
589 ip->i_d.di_uid = uid;
590 inode->i_uid = uid;
591 }
592 if (igid != gid) {
593 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
594 ASSERT(!XFS_IS_PQUOTA_ON(mp));
595 ASSERT(mask & ATTR_GID);
596 ASSERT(gdqp);
597 olddquot2 = xfs_qm_vop_chown(tp, ip,
598 &ip->i_gdquot, gdqp);
599 }
600 ip->i_d.di_gid = gid;
601 inode->i_gid = gid;
602 }
603 }
604
605 /*
606 * Change file access modes.
607 */
608 if (mask & ATTR_MODE) {
609 umode_t mode = iattr->ia_mode;
610
611 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
612 mode &= ~S_ISGID;
613
614 ip->i_d.di_mode &= S_IFMT;
615 ip->i_d.di_mode |= mode & ~S_IFMT;
616
617 inode->i_mode &= S_IFMT;
618 inode->i_mode |= mode & ~S_IFMT;
619 }
620
621 /*
622 * Change file access or modified times.
623 */
624 if (mask & ATTR_ATIME) {
625 inode->i_atime = iattr->ia_atime;
626 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
627 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
628 }
629 if (mask & ATTR_CTIME) {
630 inode->i_ctime = iattr->ia_ctime;
631 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
632 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
633 }
634 if (mask & ATTR_MTIME) {
635 inode->i_mtime = iattr->ia_mtime;
636 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
637 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
638 }
639
640 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
641
642 XFS_STATS_INC(xs_ig_attrchg);
643
644 if (mp->m_flags & XFS_MOUNT_WSYNC)
645 xfs_trans_set_sync(tp);
646 error = xfs_trans_commit(tp, 0);
647
648 xfs_iunlock(ip, XFS_ILOCK_EXCL);
649
650 /*
651 * Release any dquot(s) the inode had kept before chown.
652 */
653 xfs_qm_dqrele(olddquot1);
654 xfs_qm_dqrele(olddquot2);
655 xfs_qm_dqrele(udqp);
656 xfs_qm_dqrele(gdqp);
657
658 if (error)
659 return XFS_ERROR(error);
660
661 /*
662 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
663 * update. We could avoid this with linked transactions
664 * and passing down the transaction pointer all the way
665 * to attr_set. No previous user of the generic
666 * Posix ACL code seems to care about this issue either.
667 */
668 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
669 error = -xfs_acl_chmod(inode);
670 if (error)
671 return XFS_ERROR(error);
672 }
673
674 return 0;
675
676out_trans_cancel:
677 xfs_trans_cancel(tp, 0);
678 xfs_iunlock(ip, XFS_ILOCK_EXCL);
679out_dqrele:
680 xfs_qm_dqrele(udqp);
681 xfs_qm_dqrele(gdqp);
682 return error;
683}
684
685/*
686 * Truncate file. Must have write permission and not be a directory.
687 */
688int
689xfs_setattr_size(
690 struct xfs_inode *ip,
691 struct iattr *iattr,
692 int flags)
693{
694 struct xfs_mount *mp = ip->i_mount;
695 struct inode *inode = VFS_I(ip);
696 int mask = iattr->ia_valid;
697 xfs_off_t oldsize, newsize;
698 struct xfs_trans *tp;
699 int error;
700 uint lock_flags = 0;
701 uint commit_flags = 0;
702
703 trace_xfs_setattr(ip);
704
705 if (mp->m_flags & XFS_MOUNT_RDONLY)
706 return XFS_ERROR(EROFS);
707
708 if (XFS_FORCED_SHUTDOWN(mp))
709 return XFS_ERROR(EIO);
710
711 error = -inode_change_ok(inode, iattr);
712 if (error)
713 return XFS_ERROR(error);
714
715 ASSERT(S_ISREG(ip->i_d.di_mode));
716 ASSERT((mask & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
717 ATTR_MTIME_SET|ATTR_KILL_SUID|ATTR_KILL_SGID|
718 ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
719
720 if (!(flags & XFS_ATTR_NOLOCK)) {
721 lock_flags |= XFS_IOLOCK_EXCL;
722 xfs_ilock(ip, lock_flags);
723 }
724
725 oldsize = inode->i_size;
726 newsize = iattr->ia_size;
727
728 /*
729 * Short circuit the truncate case for zero length files.
730 */
731 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
732 if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
733 goto out_unlock;
734
735 /*
736 * Use the regular setattr path to update the timestamps.
737 */
738 xfs_iunlock(ip, lock_flags);
739 iattr->ia_valid &= ~ATTR_SIZE;
740 return xfs_setattr_nonsize(ip, iattr, 0);
741 }
742
743 /*
744 * Make sure that the dquots are attached to the inode.
745 */
746 error = xfs_qm_dqattach(ip, 0);
747 if (error)
748 goto out_unlock;
749
750 /*
751 * Now we can make the changes. Before we join the inode to the
752 * transaction, take care of the part of the truncation that must be
753 * done without the inode lock. This needs to be done before joining
754 * the inode to the transaction, because the inode cannot be unlocked
755 * once it is a part of the transaction.
756 */
757 if (newsize > oldsize) {
758 /*
759 * Do the first part of growing a file: zero any data in the
760 * last block that is beyond the old EOF. We need to do this
761 * before the inode is joined to the transaction to modify
762 * i_size.
763 */
764 error = xfs_zero_eof(ip, newsize, oldsize);
765 if (error)
766 goto out_unlock;
767 }
768
769 /*
770 * We are going to log the inode size change in this transaction so
771 * any previous writes that are beyond the on disk EOF and the new
772 * EOF that have not been written out need to be written here. If we
773 * do not write the data out, we expose ourselves to the null files
774 * problem.
775 *
776 * Only flush from the on disk size to the smaller of the in memory
777 * file size or the new size as that's the range we really care about
778 * here and prevents waiting for other data not within the range we
779 * care about here.
780 */
781 if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
782 error = xfs_flush_pages(ip, ip->i_d.di_size, newsize, 0,
783 FI_NONE);
784 if (error)
785 goto out_unlock;
786 }
787
788 /*
789 * Wait for all direct I/O to complete.
790 */
791 inode_dio_wait(inode);
792
793 error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
794 if (error)
795 goto out_unlock;
796
797 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
798 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
799 XFS_TRANS_PERM_LOG_RES,
800 XFS_ITRUNCATE_LOG_COUNT);
801 if (error)
802 goto out_trans_cancel;
803
804 truncate_setsize(inode, newsize);
805
806 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
807 lock_flags |= XFS_ILOCK_EXCL;
808
809 xfs_ilock(ip, XFS_ILOCK_EXCL);
810
811 xfs_trans_ijoin(tp, ip, 0);
812
813 /*
814 * Only change the c/mtime if we are changing the size or we are
815 * explicitly asked to change it. This handles the semantic difference
816 * between truncate() and ftruncate() as implemented in the VFS.
817 *
818 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
819 * special case where we need to update the times despite not having
820 * these flags set. For all other operations the VFS set these flags
821 * explicitly if it wants a timestamp update.
822 */
823 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
824 iattr->ia_ctime = iattr->ia_mtime =
825 current_fs_time(inode->i_sb);
826 mask |= ATTR_CTIME | ATTR_MTIME;
827 }
828
829 /*
830 * The first thing we do is set the size to new_size permanently on
831 * disk. This way we don't have to worry about anyone ever being able
832 * to look at the data being freed even in the face of a crash.
833 * What we're getting around here is the case where we free a block, it
834 * is allocated to another file, it is written to, and then we crash.
835 * If the new data gets written to the file but the log buffers
836 * containing the free and reallocation don't, then we'd end up with
837 * garbage in the blocks being freed. As long as we make the new size
838 * permanent before actually freeing any blocks it doesn't matter if
839 * they get written to.
840 */
841 ip->i_d.di_size = newsize;
842 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
843
844 if (newsize <= oldsize) {
845 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
846 if (error)
847 goto out_trans_abort;
848
849 /*
850 * Truncated "down", so we're removing references to old data
851 * here - if we delay flushing for a long time, we expose
852 * ourselves unduly to the notorious NULL files problem. So,
853 * we mark this inode and flush it when the file is closed,
854 * and do not wait the usual (long) time for writeout.
855 */
856 xfs_iflags_set(ip, XFS_ITRUNCATED);
857 }
858
859 if (mask & ATTR_CTIME) {
860 inode->i_ctime = iattr->ia_ctime;
861 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
862 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
863 }
864 if (mask & ATTR_MTIME) {
865 inode->i_mtime = iattr->ia_mtime;
866 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
867 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
868 }
869
870 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
871
872 XFS_STATS_INC(xs_ig_attrchg);
873
874 if (mp->m_flags & XFS_MOUNT_WSYNC)
875 xfs_trans_set_sync(tp);
876
877 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
878out_unlock:
879 if (lock_flags)
880 xfs_iunlock(ip, lock_flags);
881 return error;
882
883out_trans_abort:
884 commit_flags |= XFS_TRANS_ABORT;
885out_trans_cancel:
886 xfs_trans_cancel(tp, commit_flags);
887 goto out_unlock;
888}
889
890STATIC int
891xfs_vn_setattr(
892 struct dentry *dentry,
893 struct iattr *iattr)
894{
895 if (iattr->ia_valid & ATTR_SIZE)
896 return -xfs_setattr_size(XFS_I(dentry->d_inode), iattr, 0);
897 return -xfs_setattr_nonsize(XFS_I(dentry->d_inode), iattr, 0);
898}
899
900#define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
901
902/*
903 * Call fiemap helper to fill in user data.
904 * Returns positive errors to xfs_getbmap.
905 */
906STATIC int
907xfs_fiemap_format(
908 void **arg,
909 struct getbmapx *bmv,
910 int *full)
911{
912 int error;
913 struct fiemap_extent_info *fieinfo = *arg;
914 u32 fiemap_flags = 0;
915 u64 logical, physical, length;
916
917 /* Do nothing for a hole */
918 if (bmv->bmv_block == -1LL)
919 return 0;
920
921 logical = BBTOB(bmv->bmv_offset);
922 physical = BBTOB(bmv->bmv_block);
923 length = BBTOB(bmv->bmv_length);
924
925 if (bmv->bmv_oflags & BMV_OF_PREALLOC)
926 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
927 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
928 fiemap_flags |= FIEMAP_EXTENT_DELALLOC;
929 physical = 0; /* no block yet */
930 }
931 if (bmv->bmv_oflags & BMV_OF_LAST)
932 fiemap_flags |= FIEMAP_EXTENT_LAST;
933
934 error = fiemap_fill_next_extent(fieinfo, logical, physical,
935 length, fiemap_flags);
936 if (error > 0) {
937 error = 0;
938 *full = 1; /* user array now full */
939 }
940
941 return -error;
942}
943
944STATIC int
945xfs_vn_fiemap(
946 struct inode *inode,
947 struct fiemap_extent_info *fieinfo,
948 u64 start,
949 u64 length)
950{
951 xfs_inode_t *ip = XFS_I(inode);
952 struct getbmapx bm;
953 int error;
954
955 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
956 if (error)
957 return error;
958
959 /* Set up bmap header for xfs internal routine */
960 bm.bmv_offset = BTOBB(start);
961 /* Special case for whole file */
962 if (length == FIEMAP_MAX_OFFSET)
963 bm.bmv_length = -1LL;
964 else
965 bm.bmv_length = BTOBB(length);
966
967 /* We add one because in getbmap world count includes the header */
968 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
969 fieinfo->fi_extents_max + 1;
970 bm.bmv_count = min_t(__s32, bm.bmv_count,
971 (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
972 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
973 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
974 bm.bmv_iflags |= BMV_IF_ATTRFORK;
975 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
976 bm.bmv_iflags |= BMV_IF_DELALLOC;
977
978 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
979 if (error)
980 return -error;
981
982 return 0;
983}
984
985static const struct inode_operations xfs_inode_operations = {
986 .get_acl = xfs_get_acl,
987 .getattr = xfs_vn_getattr,
988 .setattr = xfs_vn_setattr,
989 .setxattr = generic_setxattr,
990 .getxattr = generic_getxattr,
991 .removexattr = generic_removexattr,
992 .listxattr = xfs_vn_listxattr,
993 .fiemap = xfs_vn_fiemap,
994};
995
996static const struct inode_operations xfs_dir_inode_operations = {
997 .create = xfs_vn_create,
998 .lookup = xfs_vn_lookup,
999 .link = xfs_vn_link,
1000 .unlink = xfs_vn_unlink,
1001 .symlink = xfs_vn_symlink,
1002 .mkdir = xfs_vn_mkdir,
1003 /*
1004 * Yes, XFS uses the same method for rmdir and unlink.
1005 *
1006 * There are some subtile differences deeper in the code,
1007 * but we use S_ISDIR to check for those.
1008 */
1009 .rmdir = xfs_vn_unlink,
1010 .mknod = xfs_vn_mknod,
1011 .rename = xfs_vn_rename,
1012 .get_acl = xfs_get_acl,
1013 .getattr = xfs_vn_getattr,
1014 .setattr = xfs_vn_setattr,
1015 .setxattr = generic_setxattr,
1016 .getxattr = generic_getxattr,
1017 .removexattr = generic_removexattr,
1018 .listxattr = xfs_vn_listxattr,
1019};
1020
1021static const struct inode_operations xfs_dir_ci_inode_operations = {
1022 .create = xfs_vn_create,
1023 .lookup = xfs_vn_ci_lookup,
1024 .link = xfs_vn_link,
1025 .unlink = xfs_vn_unlink,
1026 .symlink = xfs_vn_symlink,
1027 .mkdir = xfs_vn_mkdir,
1028 /*
1029 * Yes, XFS uses the same method for rmdir and unlink.
1030 *
1031 * There are some subtile differences deeper in the code,
1032 * but we use S_ISDIR to check for those.
1033 */
1034 .rmdir = xfs_vn_unlink,
1035 .mknod = xfs_vn_mknod,
1036 .rename = xfs_vn_rename,
1037 .get_acl = xfs_get_acl,
1038 .getattr = xfs_vn_getattr,
1039 .setattr = xfs_vn_setattr,
1040 .setxattr = generic_setxattr,
1041 .getxattr = generic_getxattr,
1042 .removexattr = generic_removexattr,
1043 .listxattr = xfs_vn_listxattr,
1044};
1045
1046static const struct inode_operations xfs_symlink_inode_operations = {
1047 .readlink = generic_readlink,
1048 .follow_link = xfs_vn_follow_link,
1049 .put_link = xfs_vn_put_link,
1050 .get_acl = xfs_get_acl,
1051 .getattr = xfs_vn_getattr,
1052 .setattr = xfs_vn_setattr,
1053 .setxattr = generic_setxattr,
1054 .getxattr = generic_getxattr,
1055 .removexattr = generic_removexattr,
1056 .listxattr = xfs_vn_listxattr,
1057};
1058
1059STATIC void
1060xfs_diflags_to_iflags(
1061 struct inode *inode,
1062 struct xfs_inode *ip)
1063{
1064 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1065 inode->i_flags |= S_IMMUTABLE;
1066 else
1067 inode->i_flags &= ~S_IMMUTABLE;
1068 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1069 inode->i_flags |= S_APPEND;
1070 else
1071 inode->i_flags &= ~S_APPEND;
1072 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1073 inode->i_flags |= S_SYNC;
1074 else
1075 inode->i_flags &= ~S_SYNC;
1076 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1077 inode->i_flags |= S_NOATIME;
1078 else
1079 inode->i_flags &= ~S_NOATIME;
1080}
1081
1082/*
1083 * Initialize the Linux inode, set up the operation vectors and
1084 * unlock the inode.
1085 *
1086 * When reading existing inodes from disk this is called directly
1087 * from xfs_iget, when creating a new inode it is called from
1088 * xfs_ialloc after setting up the inode.
1089 *
1090 * We are always called with an uninitialised linux inode here.
1091 * We need to initialise the necessary fields and take a reference
1092 * on it.
1093 */
1094void
1095xfs_setup_inode(
1096 struct xfs_inode *ip)
1097{
1098 struct inode *inode = &ip->i_vnode;
1099
1100 inode->i_ino = ip->i_ino;
1101 inode->i_state = I_NEW;
1102
1103 inode_sb_list_add(inode);
1104 /* make the inode look hashed for the writeback code */
1105 hlist_add_fake(&inode->i_hash);
1106
1107 inode->i_mode = ip->i_d.di_mode;
1108 set_nlink(inode, ip->i_d.di_nlink);
1109 inode->i_uid = ip->i_d.di_uid;
1110 inode->i_gid = ip->i_d.di_gid;
1111
1112 switch (inode->i_mode & S_IFMT) {
1113 case S_IFBLK:
1114 case S_IFCHR:
1115 inode->i_rdev =
1116 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1117 sysv_minor(ip->i_df.if_u2.if_rdev));
1118 break;
1119 default:
1120 inode->i_rdev = 0;
1121 break;
1122 }
1123
1124 inode->i_generation = ip->i_d.di_gen;
1125 i_size_write(inode, ip->i_d.di_size);
1126 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
1127 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
1128 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
1129 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
1130 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
1131 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
1132 xfs_diflags_to_iflags(inode, ip);
1133
1134 switch (inode->i_mode & S_IFMT) {
1135 case S_IFREG:
1136 inode->i_op = &xfs_inode_operations;
1137 inode->i_fop = &xfs_file_operations;
1138 inode->i_mapping->a_ops = &xfs_address_space_operations;
1139 break;
1140 case S_IFDIR:
1141 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1142 inode->i_op = &xfs_dir_ci_inode_operations;
1143 else
1144 inode->i_op = &xfs_dir_inode_operations;
1145 inode->i_fop = &xfs_dir_file_operations;
1146 break;
1147 case S_IFLNK:
1148 inode->i_op = &xfs_symlink_inode_operations;
1149 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1150 inode->i_mapping->a_ops = &xfs_address_space_operations;
1151 break;
1152 default:
1153 inode->i_op = &xfs_inode_operations;
1154 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1155 break;
1156 }
1157
1158 /*
1159 * If there is no attribute fork no ACL can exist on this inode,
1160 * and it can't have any file capabilities attached to it either.
1161 */
1162 if (!XFS_IFORK_Q(ip)) {
1163 inode_has_no_xattr(inode);
1164 cache_no_acl(inode);
1165 }
1166
1167 xfs_iflags_clear(ip, XFS_INEW);
1168 barrier();
1169
1170 unlock_new_inode(inode);
1171}