mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / xfs / xfs_iops.c
blob42c670a313518ea1afb1da4d89c6b2154697040c
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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.
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.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_inode.h"
27 #include "xfs_bmap.h"
28 #include "xfs_bmap_util.h"
29 #include "xfs_acl.h"
30 #include "xfs_quota.h"
31 #include "xfs_error.h"
32 #include "xfs_attr.h"
33 #include "xfs_trans.h"
34 #include "xfs_trace.h"
35 #include "xfs_icache.h"
36 #include "xfs_symlink.h"
37 #include "xfs_da_btree.h"
38 #include "xfs_dir2.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_pnfs.h"
41 #include "xfs_iomap.h"
43 #include <linux/capability.h>
44 #include <linux/xattr.h>
45 #include <linux/posix_acl.h>
46 #include <linux/security.h>
47 #include <linux/iomap.h>
48 #include <linux/slab.h>
51 * Directories have different lock order w.r.t. mmap_sem compared to regular
52 * files. This is due to readdir potentially triggering page faults on a user
53 * buffer inside filldir(), and this happens with the ilock on the directory
54 * held. For regular files, the lock order is the other way around - the
55 * mmap_sem is taken during the page fault, and then we lock the ilock to do
56 * block mapping. Hence we need a different class for the directory ilock so
57 * that lockdep can tell them apart.
59 static struct lock_class_key xfs_nondir_ilock_class;
60 static struct lock_class_key xfs_dir_ilock_class;
62 static int
63 xfs_initxattrs(
64 struct inode *inode,
65 const struct xattr *xattr_array,
66 void *fs_info)
68 const struct xattr *xattr;
69 struct xfs_inode *ip = XFS_I(inode);
70 int error = 0;
72 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
73 error = xfs_attr_set(ip, xattr->name, xattr->value,
74 xattr->value_len, ATTR_SECURE);
75 if (error < 0)
76 break;
78 return error;
82 * Hook in SELinux. This is not quite correct yet, what we really need
83 * here (as we do for default ACLs) is a mechanism by which creation of
84 * these attrs can be journalled at inode creation time (along with the
85 * inode, of course, such that log replay can't cause these to be lost).
88 STATIC int
89 xfs_init_security(
90 struct inode *inode,
91 struct inode *dir,
92 const struct qstr *qstr)
94 return security_inode_init_security(inode, dir, qstr,
95 &xfs_initxattrs, NULL);
98 static void
99 xfs_dentry_to_name(
100 struct xfs_name *namep,
101 struct dentry *dentry)
103 namep->name = dentry->d_name.name;
104 namep->len = dentry->d_name.len;
105 namep->type = XFS_DIR3_FT_UNKNOWN;
108 static int
109 xfs_dentry_mode_to_name(
110 struct xfs_name *namep,
111 struct dentry *dentry,
112 int mode)
114 namep->name = dentry->d_name.name;
115 namep->len = dentry->d_name.len;
116 namep->type = xfs_mode_to_ftype(mode);
118 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
119 return -EFSCORRUPTED;
121 return 0;
124 STATIC void
125 xfs_cleanup_inode(
126 struct inode *dir,
127 struct inode *inode,
128 struct dentry *dentry)
130 struct xfs_name teardown;
132 /* Oh, the horror.
133 * If we can't add the ACL or we fail in
134 * xfs_init_security we must back out.
135 * ENOSPC can hit here, among other things.
137 xfs_dentry_to_name(&teardown, dentry);
139 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
142 STATIC int
143 xfs_generic_create(
144 struct inode *dir,
145 struct dentry *dentry,
146 umode_t mode,
147 dev_t rdev,
148 bool tmpfile) /* unnamed file */
150 struct inode *inode;
151 struct xfs_inode *ip = NULL;
152 struct posix_acl *default_acl, *acl;
153 struct xfs_name name;
154 int error;
157 * Irix uses Missed'em'V split, but doesn't want to see
158 * the upper 5 bits of (14bit) major.
160 if (S_ISCHR(mode) || S_ISBLK(mode)) {
161 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
162 return -EINVAL;
163 rdev = sysv_encode_dev(rdev);
164 } else {
165 rdev = 0;
168 error = posix_acl_create(dir, &mode, &default_acl, &acl);
169 if (error)
170 return error;
172 /* Verify mode is valid also for tmpfile case */
173 error = xfs_dentry_mode_to_name(&name, dentry, mode);
174 if (unlikely(error))
175 goto out_free_acl;
177 if (!tmpfile) {
178 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
179 } else {
180 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
182 if (unlikely(error))
183 goto out_free_acl;
185 inode = VFS_I(ip);
187 error = xfs_init_security(inode, dir, &dentry->d_name);
188 if (unlikely(error))
189 goto out_cleanup_inode;
191 #ifdef CONFIG_XFS_POSIX_ACL
192 if (default_acl) {
193 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
194 if (error)
195 goto out_cleanup_inode;
197 if (acl) {
198 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
199 if (error)
200 goto out_cleanup_inode;
202 #endif
204 xfs_setup_iops(ip);
206 if (tmpfile)
207 d_tmpfile(dentry, inode);
208 else
209 d_instantiate(dentry, inode);
211 xfs_finish_inode_setup(ip);
213 out_free_acl:
214 if (default_acl)
215 posix_acl_release(default_acl);
216 if (acl)
217 posix_acl_release(acl);
218 return error;
220 out_cleanup_inode:
221 xfs_finish_inode_setup(ip);
222 if (!tmpfile)
223 xfs_cleanup_inode(dir, inode, dentry);
224 iput(inode);
225 goto out_free_acl;
228 STATIC int
229 xfs_vn_mknod(
230 struct inode *dir,
231 struct dentry *dentry,
232 umode_t mode,
233 dev_t rdev)
235 return xfs_generic_create(dir, dentry, mode, rdev, false);
238 STATIC int
239 xfs_vn_create(
240 struct inode *dir,
241 struct dentry *dentry,
242 umode_t mode,
243 bool flags)
245 return xfs_vn_mknod(dir, dentry, mode, 0);
248 STATIC int
249 xfs_vn_mkdir(
250 struct inode *dir,
251 struct dentry *dentry,
252 umode_t mode)
254 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
257 STATIC struct dentry *
258 xfs_vn_lookup(
259 struct inode *dir,
260 struct dentry *dentry,
261 unsigned int flags)
263 struct xfs_inode *cip;
264 struct xfs_name name;
265 int error;
267 if (dentry->d_name.len >= MAXNAMELEN)
268 return ERR_PTR(-ENAMETOOLONG);
270 xfs_dentry_to_name(&name, dentry);
271 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
272 if (unlikely(error)) {
273 if (unlikely(error != -ENOENT))
274 return ERR_PTR(error);
275 d_add(dentry, NULL);
276 return NULL;
279 return d_splice_alias(VFS_I(cip), dentry);
282 STATIC struct dentry *
283 xfs_vn_ci_lookup(
284 struct inode *dir,
285 struct dentry *dentry,
286 unsigned int flags)
288 struct xfs_inode *ip;
289 struct xfs_name xname;
290 struct xfs_name ci_name;
291 struct qstr dname;
292 int error;
294 if (dentry->d_name.len >= MAXNAMELEN)
295 return ERR_PTR(-ENAMETOOLONG);
297 xfs_dentry_to_name(&xname, dentry);
298 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
299 if (unlikely(error)) {
300 if (unlikely(error != -ENOENT))
301 return ERR_PTR(error);
303 * call d_add(dentry, NULL) here when d_drop_negative_children
304 * is called in xfs_vn_mknod (ie. allow negative dentries
305 * with CI filesystems).
307 return NULL;
310 /* if exact match, just splice and exit */
311 if (!ci_name.name)
312 return d_splice_alias(VFS_I(ip), dentry);
314 /* else case-insensitive match... */
315 dname.name = ci_name.name;
316 dname.len = ci_name.len;
317 dentry = d_add_ci(dentry, VFS_I(ip), &dname);
318 kmem_free(ci_name.name);
319 return dentry;
322 STATIC int
323 xfs_vn_link(
324 struct dentry *old_dentry,
325 struct inode *dir,
326 struct dentry *dentry)
328 struct inode *inode = d_inode(old_dentry);
329 struct xfs_name name;
330 int error;
332 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
333 if (unlikely(error))
334 return error;
336 error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
337 if (unlikely(error))
338 return error;
340 ihold(inode);
341 d_instantiate(dentry, inode);
342 return 0;
345 STATIC int
346 xfs_vn_unlink(
347 struct inode *dir,
348 struct dentry *dentry)
350 struct xfs_name name;
351 int error;
353 xfs_dentry_to_name(&name, dentry);
355 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
356 if (error)
357 return error;
360 * With unlink, the VFS makes the dentry "negative": no inode,
361 * but still hashed. This is incompatible with case-insensitive
362 * mode, so invalidate (unhash) the dentry in CI-mode.
364 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
365 d_invalidate(dentry);
366 return 0;
369 STATIC int
370 xfs_vn_symlink(
371 struct inode *dir,
372 struct dentry *dentry,
373 const char *symname)
375 struct inode *inode;
376 struct xfs_inode *cip = NULL;
377 struct xfs_name name;
378 int error;
379 umode_t mode;
381 mode = S_IFLNK |
382 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
383 error = xfs_dentry_mode_to_name(&name, dentry, mode);
384 if (unlikely(error))
385 goto out;
387 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
388 if (unlikely(error))
389 goto out;
391 inode = VFS_I(cip);
393 error = xfs_init_security(inode, dir, &dentry->d_name);
394 if (unlikely(error))
395 goto out_cleanup_inode;
397 xfs_setup_iops(cip);
399 d_instantiate(dentry, inode);
400 xfs_finish_inode_setup(cip);
401 return 0;
403 out_cleanup_inode:
404 xfs_finish_inode_setup(cip);
405 xfs_cleanup_inode(dir, inode, dentry);
406 iput(inode);
407 out:
408 return error;
411 STATIC int
412 xfs_vn_rename(
413 struct inode *odir,
414 struct dentry *odentry,
415 struct inode *ndir,
416 struct dentry *ndentry,
417 unsigned int flags)
419 struct inode *new_inode = d_inode(ndentry);
420 int omode = 0;
421 int error;
422 struct xfs_name oname;
423 struct xfs_name nname;
425 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
426 return -EINVAL;
428 /* if we are exchanging files, we need to set i_mode of both files */
429 if (flags & RENAME_EXCHANGE)
430 omode = d_inode(ndentry)->i_mode;
432 error = xfs_dentry_mode_to_name(&oname, odentry, omode);
433 if (omode && unlikely(error))
434 return error;
436 error = xfs_dentry_mode_to_name(&nname, ndentry,
437 d_inode(odentry)->i_mode);
438 if (unlikely(error))
439 return error;
441 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
442 XFS_I(ndir), &nname,
443 new_inode ? XFS_I(new_inode) : NULL, flags);
447 * careful here - this function can get called recursively, so
448 * we need to be very careful about how much stack we use.
449 * uio is kmalloced for this reason...
451 STATIC const char *
452 xfs_vn_get_link(
453 struct dentry *dentry,
454 struct inode *inode,
455 struct delayed_call *done)
457 char *link;
458 int error = -ENOMEM;
460 if (!dentry)
461 return ERR_PTR(-ECHILD);
463 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
464 if (!link)
465 goto out_err;
467 error = xfs_readlink(XFS_I(d_inode(dentry)), link);
468 if (unlikely(error))
469 goto out_kfree;
471 set_delayed_call(done, kfree_link, link);
472 return link;
474 out_kfree:
475 kfree(link);
476 out_err:
477 return ERR_PTR(error);
480 STATIC const char *
481 xfs_vn_get_link_inline(
482 struct dentry *dentry,
483 struct inode *inode,
484 struct delayed_call *done)
486 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
487 return XFS_I(inode)->i_df.if_u1.if_data;
490 STATIC int
491 xfs_vn_getattr(
492 const struct path *path,
493 struct kstat *stat,
494 u32 request_mask,
495 unsigned int query_flags)
497 struct inode *inode = d_inode(path->dentry);
498 struct xfs_inode *ip = XFS_I(inode);
499 struct xfs_mount *mp = ip->i_mount;
501 trace_xfs_getattr(ip);
503 if (XFS_FORCED_SHUTDOWN(mp))
504 return -EIO;
506 stat->size = XFS_ISIZE(ip);
507 stat->dev = inode->i_sb->s_dev;
508 stat->mode = inode->i_mode;
509 stat->nlink = inode->i_nlink;
510 stat->uid = inode->i_uid;
511 stat->gid = inode->i_gid;
512 stat->ino = ip->i_ino;
513 stat->atime = inode->i_atime;
514 stat->mtime = inode->i_mtime;
515 stat->ctime = inode->i_ctime;
516 stat->blocks =
517 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
519 if (ip->i_d.di_version == 3) {
520 if (request_mask & STATX_BTIME) {
521 stat->result_mask |= STATX_BTIME;
522 stat->btime.tv_sec = ip->i_d.di_crtime.t_sec;
523 stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec;
527 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
528 stat->attributes |= STATX_ATTR_IMMUTABLE;
529 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
530 stat->attributes |= STATX_ATTR_APPEND;
531 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
532 stat->attributes |= STATX_ATTR_NODUMP;
534 switch (inode->i_mode & S_IFMT) {
535 case S_IFBLK:
536 case S_IFCHR:
537 stat->blksize = BLKDEV_IOSIZE;
538 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
539 sysv_minor(ip->i_df.if_u2.if_rdev));
540 break;
541 default:
542 if (XFS_IS_REALTIME_INODE(ip)) {
544 * If the file blocks are being allocated from a
545 * realtime volume, then return the inode's realtime
546 * extent size or the realtime volume's extent size.
548 stat->blksize =
549 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
550 } else
551 stat->blksize = xfs_preferred_iosize(mp);
552 stat->rdev = 0;
553 break;
556 return 0;
559 static void
560 xfs_setattr_mode(
561 struct xfs_inode *ip,
562 struct iattr *iattr)
564 struct inode *inode = VFS_I(ip);
565 umode_t mode = iattr->ia_mode;
567 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
569 inode->i_mode &= S_IFMT;
570 inode->i_mode |= mode & ~S_IFMT;
573 void
574 xfs_setattr_time(
575 struct xfs_inode *ip,
576 struct iattr *iattr)
578 struct inode *inode = VFS_I(ip);
580 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
582 if (iattr->ia_valid & ATTR_ATIME)
583 inode->i_atime = iattr->ia_atime;
584 if (iattr->ia_valid & ATTR_CTIME)
585 inode->i_ctime = iattr->ia_ctime;
586 if (iattr->ia_valid & ATTR_MTIME)
587 inode->i_mtime = iattr->ia_mtime;
590 static int
591 xfs_vn_change_ok(
592 struct dentry *dentry,
593 struct iattr *iattr)
595 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount;
597 if (mp->m_flags & XFS_MOUNT_RDONLY)
598 return -EROFS;
600 if (XFS_FORCED_SHUTDOWN(mp))
601 return -EIO;
603 return setattr_prepare(dentry, iattr);
607 * Set non-size attributes of an inode.
609 * Caution: The caller of this function is responsible for calling
610 * setattr_prepare() or otherwise verifying the change is fine.
613 xfs_setattr_nonsize(
614 struct xfs_inode *ip,
615 struct iattr *iattr,
616 int flags)
618 xfs_mount_t *mp = ip->i_mount;
619 struct inode *inode = VFS_I(ip);
620 int mask = iattr->ia_valid;
621 xfs_trans_t *tp;
622 int error;
623 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
624 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
625 struct xfs_dquot *udqp = NULL, *gdqp = NULL;
626 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL;
628 ASSERT((mask & ATTR_SIZE) == 0);
631 * If disk quotas is on, we make sure that the dquots do exist on disk,
632 * before we start any other transactions. Trying to do this later
633 * is messy. We don't care to take a readlock to look at the ids
634 * in inode here, because we can't hold it across the trans_reserve.
635 * If the IDs do change before we take the ilock, we're covered
636 * because the i_*dquot fields will get updated anyway.
638 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
639 uint qflags = 0;
641 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
642 uid = iattr->ia_uid;
643 qflags |= XFS_QMOPT_UQUOTA;
644 } else {
645 uid = inode->i_uid;
647 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
648 gid = iattr->ia_gid;
649 qflags |= XFS_QMOPT_GQUOTA;
650 } else {
651 gid = inode->i_gid;
655 * We take a reference when we initialize udqp and gdqp,
656 * so it is important that we never blindly double trip on
657 * the same variable. See xfs_create() for an example.
659 ASSERT(udqp == NULL);
660 ASSERT(gdqp == NULL);
661 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
662 xfs_kgid_to_gid(gid),
663 xfs_get_projid(ip),
664 qflags, &udqp, &gdqp, NULL);
665 if (error)
666 return error;
669 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
670 if (error)
671 goto out_dqrele;
673 xfs_ilock(ip, XFS_ILOCK_EXCL);
674 xfs_trans_ijoin(tp, ip, 0);
677 * Change file ownership. Must be the owner or privileged.
679 if (mask & (ATTR_UID|ATTR_GID)) {
681 * These IDs could have changed since we last looked at them.
682 * But, we're assured that if the ownership did change
683 * while we didn't have the inode locked, inode's dquot(s)
684 * would have changed also.
686 iuid = inode->i_uid;
687 igid = inode->i_gid;
688 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
689 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
692 * Do a quota reservation only if uid/gid is actually
693 * going to change.
695 if (XFS_IS_QUOTA_RUNNING(mp) &&
696 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
697 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
698 ASSERT(tp);
699 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
700 NULL, capable(CAP_FOWNER) ?
701 XFS_QMOPT_FORCE_RES : 0);
702 if (error) /* out of quota */
703 goto out_cancel;
708 * Change file ownership. Must be the owner or privileged.
710 if (mask & (ATTR_UID|ATTR_GID)) {
712 * CAP_FSETID overrides the following restrictions:
714 * The set-user-ID and set-group-ID bits of a file will be
715 * cleared upon successful return from chown()
717 if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
718 !capable(CAP_FSETID))
719 inode->i_mode &= ~(S_ISUID|S_ISGID);
722 * Change the ownerships and register quota modifications
723 * in the transaction.
725 if (!uid_eq(iuid, uid)) {
726 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
727 ASSERT(mask & ATTR_UID);
728 ASSERT(udqp);
729 olddquot1 = xfs_qm_vop_chown(tp, ip,
730 &ip->i_udquot, udqp);
732 ip->i_d.di_uid = xfs_kuid_to_uid(uid);
733 inode->i_uid = uid;
735 if (!gid_eq(igid, gid)) {
736 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
737 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
738 !XFS_IS_PQUOTA_ON(mp));
739 ASSERT(mask & ATTR_GID);
740 ASSERT(gdqp);
741 olddquot2 = xfs_qm_vop_chown(tp, ip,
742 &ip->i_gdquot, gdqp);
744 ip->i_d.di_gid = xfs_kgid_to_gid(gid);
745 inode->i_gid = gid;
749 if (mask & ATTR_MODE)
750 xfs_setattr_mode(ip, iattr);
751 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
752 xfs_setattr_time(ip, iattr);
754 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
756 XFS_STATS_INC(mp, xs_ig_attrchg);
758 if (mp->m_flags & XFS_MOUNT_WSYNC)
759 xfs_trans_set_sync(tp);
760 error = xfs_trans_commit(tp);
762 xfs_iunlock(ip, XFS_ILOCK_EXCL);
765 * Release any dquot(s) the inode had kept before chown.
767 xfs_qm_dqrele(olddquot1);
768 xfs_qm_dqrele(olddquot2);
769 xfs_qm_dqrele(udqp);
770 xfs_qm_dqrele(gdqp);
772 if (error)
773 return error;
776 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
777 * update. We could avoid this with linked transactions
778 * and passing down the transaction pointer all the way
779 * to attr_set. No previous user of the generic
780 * Posix ACL code seems to care about this issue either.
782 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
783 error = posix_acl_chmod(inode, inode->i_mode);
784 if (error)
785 return error;
788 return 0;
790 out_cancel:
791 xfs_trans_cancel(tp);
792 xfs_iunlock(ip, XFS_ILOCK_EXCL);
793 out_dqrele:
794 xfs_qm_dqrele(udqp);
795 xfs_qm_dqrele(gdqp);
796 return error;
800 xfs_vn_setattr_nonsize(
801 struct dentry *dentry,
802 struct iattr *iattr)
804 struct xfs_inode *ip = XFS_I(d_inode(dentry));
805 int error;
807 trace_xfs_setattr(ip);
809 error = xfs_vn_change_ok(dentry, iattr);
810 if (error)
811 return error;
812 return xfs_setattr_nonsize(ip, iattr, 0);
816 * Truncate file. Must have write permission and not be a directory.
818 * Caution: The caller of this function is responsible for calling
819 * setattr_prepare() or otherwise verifying the change is fine.
821 STATIC int
822 xfs_setattr_size(
823 struct xfs_inode *ip,
824 struct iattr *iattr)
826 struct xfs_mount *mp = ip->i_mount;
827 struct inode *inode = VFS_I(ip);
828 xfs_off_t oldsize, newsize;
829 struct xfs_trans *tp;
830 int error;
831 uint lock_flags = 0;
832 bool did_zeroing = false;
834 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
835 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
836 ASSERT(S_ISREG(inode->i_mode));
837 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
838 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
840 oldsize = inode->i_size;
841 newsize = iattr->ia_size;
844 * Short circuit the truncate case for zero length files.
846 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
847 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
848 return 0;
851 * Use the regular setattr path to update the timestamps.
853 iattr->ia_valid &= ~ATTR_SIZE;
854 return xfs_setattr_nonsize(ip, iattr, 0);
858 * Make sure that the dquots are attached to the inode.
860 error = xfs_qm_dqattach(ip, 0);
861 if (error)
862 return error;
865 * Wait for all direct I/O to complete.
867 inode_dio_wait(inode);
870 * File data changes must be complete before we start the transaction to
871 * modify the inode. This needs to be done before joining the inode to
872 * the transaction because the inode cannot be unlocked once it is a
873 * part of the transaction.
875 * Start with zeroing any data beyond EOF that we may expose on file
876 * extension, or zeroing out the rest of the block on a downward
877 * truncate.
879 if (newsize > oldsize) {
880 error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
881 } else {
883 * iomap won't detect a dirty page over an unwritten block (or a
884 * cow block over a hole) and subsequently skips zeroing the
885 * newly post-EOF portion of the page. Flush the new EOF to
886 * convert the block before the pagecache truncate.
888 error = filemap_write_and_wait_range(inode->i_mapping, newsize,
889 newsize);
890 if (error)
891 return error;
892 error = iomap_truncate_page(inode, newsize, &did_zeroing,
893 &xfs_iomap_ops);
896 if (error)
897 return error;
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.
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.
916 * And we update in-core i_size and truncate page cache beyond newsize
917 * before writeback the [di_size, newsize] range, so we're guaranteed
918 * not to write stale data past the new EOF on truncate down.
920 truncate_setsize(inode, newsize);
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.
930 if (did_zeroing ||
931 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
932 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
933 ip->i_d.di_size, newsize - 1);
934 if (error)
935 return error;
938 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
939 if (error)
940 return error;
942 lock_flags |= XFS_ILOCK_EXCL;
943 xfs_ilock(ip, XFS_ILOCK_EXCL);
944 xfs_trans_ijoin(tp, ip, 0);
947 * Only change the c/mtime if we are changing the size or we are
948 * explicitly asked to change it. This handles the semantic difference
949 * between truncate() and ftruncate() as implemented in the VFS.
951 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
952 * special case where we need to update the times despite not having
953 * these flags set. For all other operations the VFS set these flags
954 * explicitly if it wants a timestamp update.
956 if (newsize != oldsize &&
957 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
958 iattr->ia_ctime = iattr->ia_mtime =
959 current_time(inode);
960 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
964 * The first thing we do is set the size to new_size permanently on
965 * disk. This way we don't have to worry about anyone ever being able
966 * to look at the data being freed even in the face of a crash.
967 * What we're getting around here is the case where we free a block, it
968 * is allocated to another file, it is written to, and then we crash.
969 * If the new data gets written to the file but the log buffers
970 * containing the free and reallocation don't, then we'd end up with
971 * garbage in the blocks being freed. As long as we make the new size
972 * permanent before actually freeing any blocks it doesn't matter if
973 * they get written to.
975 ip->i_d.di_size = newsize;
976 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
978 if (newsize <= oldsize) {
979 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
980 if (error)
981 goto out_trans_cancel;
984 * Truncated "down", so we're removing references to old data
985 * here - if we delay flushing for a long time, we expose
986 * ourselves unduly to the notorious NULL files problem. So,
987 * we mark this inode and flush it when the file is closed,
988 * and do not wait the usual (long) time for writeout.
990 xfs_iflags_set(ip, XFS_ITRUNCATED);
992 /* A truncate down always removes post-EOF blocks. */
993 xfs_inode_clear_eofblocks_tag(ip);
996 if (iattr->ia_valid & ATTR_MODE)
997 xfs_setattr_mode(ip, iattr);
998 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
999 xfs_setattr_time(ip, iattr);
1001 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1003 XFS_STATS_INC(mp, xs_ig_attrchg);
1005 if (mp->m_flags & XFS_MOUNT_WSYNC)
1006 xfs_trans_set_sync(tp);
1008 error = xfs_trans_commit(tp);
1009 out_unlock:
1010 if (lock_flags)
1011 xfs_iunlock(ip, lock_flags);
1012 return error;
1014 out_trans_cancel:
1015 xfs_trans_cancel(tp);
1016 goto out_unlock;
1020 xfs_vn_setattr_size(
1021 struct dentry *dentry,
1022 struct iattr *iattr)
1024 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1025 int error;
1027 trace_xfs_setattr(ip);
1029 error = xfs_vn_change_ok(dentry, iattr);
1030 if (error)
1031 return error;
1032 return xfs_setattr_size(ip, iattr);
1035 STATIC int
1036 xfs_vn_setattr(
1037 struct dentry *dentry,
1038 struct iattr *iattr)
1040 int error;
1042 if (iattr->ia_valid & ATTR_SIZE) {
1043 struct xfs_inode *ip = XFS_I(d_inode(dentry));
1044 uint iolock = XFS_IOLOCK_EXCL;
1046 error = xfs_break_layouts(d_inode(dentry), &iolock);
1047 if (error)
1048 return error;
1050 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1051 error = xfs_vn_setattr_size(dentry, iattr);
1052 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1053 } else {
1054 error = xfs_vn_setattr_nonsize(dentry, iattr);
1057 return error;
1060 STATIC int
1061 xfs_vn_update_time(
1062 struct inode *inode,
1063 struct timespec *now,
1064 int flags)
1066 struct xfs_inode *ip = XFS_I(inode);
1067 struct xfs_mount *mp = ip->i_mount;
1068 struct xfs_trans *tp;
1069 int error;
1071 trace_xfs_update_time(ip);
1073 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1074 if (error)
1075 return error;
1077 xfs_ilock(ip, XFS_ILOCK_EXCL);
1078 if (flags & S_CTIME)
1079 inode->i_ctime = *now;
1080 if (flags & S_MTIME)
1081 inode->i_mtime = *now;
1082 if (flags & S_ATIME)
1083 inode->i_atime = *now;
1085 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1086 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1087 return xfs_trans_commit(tp);
1090 STATIC int
1091 xfs_vn_fiemap(
1092 struct inode *inode,
1093 struct fiemap_extent_info *fieinfo,
1094 u64 start,
1095 u64 length)
1097 int error;
1099 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1100 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1101 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1102 error = iomap_fiemap(inode, fieinfo, start, length,
1103 &xfs_xattr_iomap_ops);
1104 } else {
1105 error = iomap_fiemap(inode, fieinfo, start, length,
1106 &xfs_iomap_ops);
1108 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1110 return error;
1113 STATIC int
1114 xfs_vn_tmpfile(
1115 struct inode *dir,
1116 struct dentry *dentry,
1117 umode_t mode)
1119 return xfs_generic_create(dir, dentry, mode, 0, true);
1122 static const struct inode_operations xfs_inode_operations = {
1123 .get_acl = xfs_get_acl,
1124 .set_acl = xfs_set_acl,
1125 .getattr = xfs_vn_getattr,
1126 .setattr = xfs_vn_setattr,
1127 .listxattr = xfs_vn_listxattr,
1128 .fiemap = xfs_vn_fiemap,
1129 .update_time = xfs_vn_update_time,
1132 static const struct inode_operations xfs_dir_inode_operations = {
1133 .create = xfs_vn_create,
1134 .lookup = xfs_vn_lookup,
1135 .link = xfs_vn_link,
1136 .unlink = xfs_vn_unlink,
1137 .symlink = xfs_vn_symlink,
1138 .mkdir = xfs_vn_mkdir,
1140 * Yes, XFS uses the same method for rmdir and unlink.
1142 * There are some subtile differences deeper in the code,
1143 * but we use S_ISDIR to check for those.
1145 .rmdir = xfs_vn_unlink,
1146 .mknod = xfs_vn_mknod,
1147 .rename = xfs_vn_rename,
1148 .get_acl = xfs_get_acl,
1149 .set_acl = xfs_set_acl,
1150 .getattr = xfs_vn_getattr,
1151 .setattr = xfs_vn_setattr,
1152 .listxattr = xfs_vn_listxattr,
1153 .update_time = xfs_vn_update_time,
1154 .tmpfile = xfs_vn_tmpfile,
1157 static const struct inode_operations xfs_dir_ci_inode_operations = {
1158 .create = xfs_vn_create,
1159 .lookup = xfs_vn_ci_lookup,
1160 .link = xfs_vn_link,
1161 .unlink = xfs_vn_unlink,
1162 .symlink = xfs_vn_symlink,
1163 .mkdir = xfs_vn_mkdir,
1165 * Yes, XFS uses the same method for rmdir and unlink.
1167 * There are some subtile differences deeper in the code,
1168 * but we use S_ISDIR to check for those.
1170 .rmdir = xfs_vn_unlink,
1171 .mknod = xfs_vn_mknod,
1172 .rename = xfs_vn_rename,
1173 .get_acl = xfs_get_acl,
1174 .set_acl = xfs_set_acl,
1175 .getattr = xfs_vn_getattr,
1176 .setattr = xfs_vn_setattr,
1177 .listxattr = xfs_vn_listxattr,
1178 .update_time = xfs_vn_update_time,
1179 .tmpfile = xfs_vn_tmpfile,
1182 static const struct inode_operations xfs_symlink_inode_operations = {
1183 .get_link = xfs_vn_get_link,
1184 .getattr = xfs_vn_getattr,
1185 .setattr = xfs_vn_setattr,
1186 .listxattr = xfs_vn_listxattr,
1187 .update_time = xfs_vn_update_time,
1190 static const struct inode_operations xfs_inline_symlink_inode_operations = {
1191 .get_link = xfs_vn_get_link_inline,
1192 .getattr = xfs_vn_getattr,
1193 .setattr = xfs_vn_setattr,
1194 .listxattr = xfs_vn_listxattr,
1195 .update_time = xfs_vn_update_time,
1198 /* Figure out if this file actually supports DAX. */
1199 static bool
1200 xfs_inode_supports_dax(
1201 struct xfs_inode *ip)
1203 struct xfs_mount *mp = ip->i_mount;
1205 /* Only supported on non-reflinked files. */
1206 if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip))
1207 return false;
1209 /* DAX mount option or DAX iflag must be set. */
1210 if (!(mp->m_flags & XFS_MOUNT_DAX) &&
1211 !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1212 return false;
1214 /* Block size must match page size */
1215 if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1216 return false;
1218 /* Device has to support DAX too. */
1219 return xfs_find_daxdev_for_inode(VFS_I(ip)) != NULL;
1222 STATIC void
1223 xfs_diflags_to_iflags(
1224 struct inode *inode,
1225 struct xfs_inode *ip)
1227 uint16_t flags = ip->i_d.di_flags;
1229 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1230 S_NOATIME | S_DAX);
1232 if (flags & XFS_DIFLAG_IMMUTABLE)
1233 inode->i_flags |= S_IMMUTABLE;
1234 if (flags & XFS_DIFLAG_APPEND)
1235 inode->i_flags |= S_APPEND;
1236 if (flags & XFS_DIFLAG_SYNC)
1237 inode->i_flags |= S_SYNC;
1238 if (flags & XFS_DIFLAG_NOATIME)
1239 inode->i_flags |= S_NOATIME;
1240 if (xfs_inode_supports_dax(ip))
1241 inode->i_flags |= S_DAX;
1245 * Initialize the Linux inode.
1247 * When reading existing inodes from disk this is called directly from xfs_iget,
1248 * when creating a new inode it is called from xfs_ialloc after setting up the
1249 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1250 * it up to the caller to deal with unlocking the inode appropriately.
1252 void
1253 xfs_setup_inode(
1254 struct xfs_inode *ip)
1256 struct inode *inode = &ip->i_vnode;
1257 gfp_t gfp_mask;
1259 inode->i_ino = ip->i_ino;
1260 inode->i_state = I_NEW;
1262 inode_sb_list_add(inode);
1263 /* make the inode look hashed for the writeback code */
1264 hlist_add_fake(&inode->i_hash);
1266 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid);
1267 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid);
1269 switch (inode->i_mode & S_IFMT) {
1270 case S_IFBLK:
1271 case S_IFCHR:
1272 inode->i_rdev =
1273 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1274 sysv_minor(ip->i_df.if_u2.if_rdev));
1275 break;
1276 default:
1277 inode->i_rdev = 0;
1278 break;
1281 i_size_write(inode, ip->i_d.di_size);
1282 xfs_diflags_to_iflags(inode, ip);
1284 if (S_ISDIR(inode->i_mode)) {
1285 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1286 ip->d_ops = ip->i_mount->m_dir_inode_ops;
1287 } else {
1288 ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1289 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1293 * Ensure all page cache allocations are done from GFP_NOFS context to
1294 * prevent direct reclaim recursion back into the filesystem and blowing
1295 * stacks or deadlocking.
1297 gfp_mask = mapping_gfp_mask(inode->i_mapping);
1298 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1301 * If there is no attribute fork no ACL can exist on this inode,
1302 * and it can't have any file capabilities attached to it either.
1304 if (!XFS_IFORK_Q(ip)) {
1305 inode_has_no_xattr(inode);
1306 cache_no_acl(inode);
1310 void
1311 xfs_setup_iops(
1312 struct xfs_inode *ip)
1314 struct inode *inode = &ip->i_vnode;
1316 switch (inode->i_mode & S_IFMT) {
1317 case S_IFREG:
1318 inode->i_op = &xfs_inode_operations;
1319 inode->i_fop = &xfs_file_operations;
1320 inode->i_mapping->a_ops = &xfs_address_space_operations;
1321 break;
1322 case S_IFDIR:
1323 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1324 inode->i_op = &xfs_dir_ci_inode_operations;
1325 else
1326 inode->i_op = &xfs_dir_inode_operations;
1327 inode->i_fop = &xfs_dir_file_operations;
1328 break;
1329 case S_IFLNK:
1330 if (ip->i_df.if_flags & XFS_IFINLINE)
1331 inode->i_op = &xfs_inline_symlink_inode_operations;
1332 else
1333 inode->i_op = &xfs_symlink_inode_operations;
1334 break;
1335 default:
1336 inode->i_op = &xfs_inode_operations;
1337 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1338 break;