xref: /openbmc/linux/fs/xfs/xfs_iops.c (revision 9c1f8594)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_acl.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_alloc.h"
28 #include "xfs_quota.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_bmap.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_error.h"
36 #include "xfs_itable.h"
37 #include "xfs_rw.h"
38 #include "xfs_attr.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_utils.h"
41 #include "xfs_vnodeops.h"
42 #include "xfs_inode_item.h"
43 #include "xfs_trace.h"
44 
45 #include <linux/capability.h>
46 #include <linux/xattr.h>
47 #include <linux/namei.h>
48 #include <linux/posix_acl.h>
49 #include <linux/security.h>
50 #include <linux/fiemap.h>
51 #include <linux/slab.h>
52 
53 /*
54  * Bring the timestamps in the XFS inode uptodate.
55  *
56  * Used before writing the inode to disk.
57  */
58 void
59 xfs_synchronize_times(
60 	xfs_inode_t	*ip)
61 {
62 	struct inode	*inode = VFS_I(ip);
63 
64 	ip->i_d.di_atime.t_sec = (__int32_t)inode->i_atime.tv_sec;
65 	ip->i_d.di_atime.t_nsec = (__int32_t)inode->i_atime.tv_nsec;
66 	ip->i_d.di_ctime.t_sec = (__int32_t)inode->i_ctime.tv_sec;
67 	ip->i_d.di_ctime.t_nsec = (__int32_t)inode->i_ctime.tv_nsec;
68 	ip->i_d.di_mtime.t_sec = (__int32_t)inode->i_mtime.tv_sec;
69 	ip->i_d.di_mtime.t_nsec = (__int32_t)inode->i_mtime.tv_nsec;
70 }
71 
72 /*
73  * If the linux inode is valid, mark it dirty, else mark the dirty state
74  * in the XFS inode to make sure we pick it up when reclaiming the inode.
75  */
76 void
77 xfs_mark_inode_dirty_sync(
78 	xfs_inode_t	*ip)
79 {
80 	struct inode	*inode = VFS_I(ip);
81 
82 	if (!(inode->i_state & (I_WILL_FREE|I_FREEING)))
83 		mark_inode_dirty_sync(inode);
84 	else {
85 		barrier();
86 		ip->i_update_core = 1;
87 	}
88 }
89 
90 void
91 xfs_mark_inode_dirty(
92 	xfs_inode_t	*ip)
93 {
94 	struct inode	*inode = VFS_I(ip);
95 
96 	if (!(inode->i_state & (I_WILL_FREE|I_FREEING)))
97 		mark_inode_dirty(inode);
98 	else {
99 		barrier();
100 		ip->i_update_core = 1;
101 	}
102 
103 }
104 
105 /*
106  * Hook in SELinux.  This is not quite correct yet, what we really need
107  * here (as we do for default ACLs) is a mechanism by which creation of
108  * these attrs can be journalled at inode creation time (along with the
109  * inode, of course, such that log replay can't cause these to be lost).
110  */
111 STATIC int
112 xfs_init_security(
113 	struct inode	*inode,
114 	struct inode	*dir,
115 	const struct qstr *qstr)
116 {
117 	struct xfs_inode *ip = XFS_I(inode);
118 	size_t		length;
119 	void		*value;
120 	unsigned char	*name;
121 	int		error;
122 
123 	error = security_inode_init_security(inode, dir, qstr, (char **)&name,
124 					     &value, &length);
125 	if (error) {
126 		if (error == -EOPNOTSUPP)
127 			return 0;
128 		return -error;
129 	}
130 
131 	error = xfs_attr_set(ip, name, value, length, ATTR_SECURE);
132 
133 	kfree(name);
134 	kfree(value);
135 	return error;
136 }
137 
138 static void
139 xfs_dentry_to_name(
140 	struct xfs_name	*namep,
141 	struct dentry	*dentry)
142 {
143 	namep->name = dentry->d_name.name;
144 	namep->len = dentry->d_name.len;
145 }
146 
147 STATIC void
148 xfs_cleanup_inode(
149 	struct inode	*dir,
150 	struct inode	*inode,
151 	struct dentry	*dentry)
152 {
153 	struct xfs_name	teardown;
154 
155 	/* Oh, the horror.
156 	 * If we can't add the ACL or we fail in
157 	 * xfs_init_security we must back out.
158 	 * ENOSPC can hit here, among other things.
159 	 */
160 	xfs_dentry_to_name(&teardown, dentry);
161 
162 	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
163 	iput(inode);
164 }
165 
166 STATIC int
167 xfs_vn_mknod(
168 	struct inode	*dir,
169 	struct dentry	*dentry,
170 	int		mode,
171 	dev_t		rdev)
172 {
173 	struct inode	*inode;
174 	struct xfs_inode *ip = NULL;
175 	struct posix_acl *default_acl = NULL;
176 	struct xfs_name	name;
177 	int		error;
178 
179 	/*
180 	 * Irix uses Missed'em'V split, but doesn't want to see
181 	 * the upper 5 bits of (14bit) major.
182 	 */
183 	if (S_ISCHR(mode) || S_ISBLK(mode)) {
184 		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
185 			return -EINVAL;
186 		rdev = sysv_encode_dev(rdev);
187 	} else {
188 		rdev = 0;
189 	}
190 
191 	if (IS_POSIXACL(dir)) {
192 		default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
193 		if (IS_ERR(default_acl))
194 			return PTR_ERR(default_acl);
195 
196 		if (!default_acl)
197 			mode &= ~current_umask();
198 	}
199 
200 	xfs_dentry_to_name(&name, dentry);
201 	error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
202 	if (unlikely(error))
203 		goto out_free_acl;
204 
205 	inode = VFS_I(ip);
206 
207 	error = xfs_init_security(inode, dir, &dentry->d_name);
208 	if (unlikely(error))
209 		goto out_cleanup_inode;
210 
211 	if (default_acl) {
212 		error = -xfs_inherit_acl(inode, default_acl);
213 		default_acl = NULL;
214 		if (unlikely(error))
215 			goto out_cleanup_inode;
216 	}
217 
218 
219 	d_instantiate(dentry, inode);
220 	return -error;
221 
222  out_cleanup_inode:
223 	xfs_cleanup_inode(dir, inode, dentry);
224  out_free_acl:
225 	posix_acl_release(default_acl);
226 	return -error;
227 }
228 
229 STATIC int
230 xfs_vn_create(
231 	struct inode	*dir,
232 	struct dentry	*dentry,
233 	int		mode,
234 	struct nameidata *nd)
235 {
236 	return xfs_vn_mknod(dir, dentry, mode, 0);
237 }
238 
239 STATIC int
240 xfs_vn_mkdir(
241 	struct inode	*dir,
242 	struct dentry	*dentry,
243 	int		mode)
244 {
245 	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
246 }
247 
248 STATIC struct dentry *
249 xfs_vn_lookup(
250 	struct inode	*dir,
251 	struct dentry	*dentry,
252 	struct nameidata *nd)
253 {
254 	struct xfs_inode *cip;
255 	struct xfs_name	name;
256 	int		error;
257 
258 	if (dentry->d_name.len >= MAXNAMELEN)
259 		return ERR_PTR(-ENAMETOOLONG);
260 
261 	xfs_dentry_to_name(&name, dentry);
262 	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
263 	if (unlikely(error)) {
264 		if (unlikely(error != ENOENT))
265 			return ERR_PTR(-error);
266 		d_add(dentry, NULL);
267 		return NULL;
268 	}
269 
270 	return d_splice_alias(VFS_I(cip), dentry);
271 }
272 
273 STATIC struct dentry *
274 xfs_vn_ci_lookup(
275 	struct inode	*dir,
276 	struct dentry	*dentry,
277 	struct nameidata *nd)
278 {
279 	struct xfs_inode *ip;
280 	struct xfs_name	xname;
281 	struct xfs_name ci_name;
282 	struct qstr	dname;
283 	int		error;
284 
285 	if (dentry->d_name.len >= MAXNAMELEN)
286 		return ERR_PTR(-ENAMETOOLONG);
287 
288 	xfs_dentry_to_name(&xname, dentry);
289 	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
290 	if (unlikely(error)) {
291 		if (unlikely(error != ENOENT))
292 			return ERR_PTR(-error);
293 		/*
294 		 * call d_add(dentry, NULL) here when d_drop_negative_children
295 		 * is called in xfs_vn_mknod (ie. allow negative dentries
296 		 * with CI filesystems).
297 		 */
298 		return NULL;
299 	}
300 
301 	/* if exact match, just splice and exit */
302 	if (!ci_name.name)
303 		return d_splice_alias(VFS_I(ip), dentry);
304 
305 	/* else case-insensitive match... */
306 	dname.name = ci_name.name;
307 	dname.len = ci_name.len;
308 	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
309 	kmem_free(ci_name.name);
310 	return dentry;
311 }
312 
313 STATIC int
314 xfs_vn_link(
315 	struct dentry	*old_dentry,
316 	struct inode	*dir,
317 	struct dentry	*dentry)
318 {
319 	struct inode	*inode = old_dentry->d_inode;
320 	struct xfs_name	name;
321 	int		error;
322 
323 	xfs_dentry_to_name(&name, dentry);
324 
325 	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
326 	if (unlikely(error))
327 		return -error;
328 
329 	ihold(inode);
330 	d_instantiate(dentry, inode);
331 	return 0;
332 }
333 
334 STATIC int
335 xfs_vn_unlink(
336 	struct inode	*dir,
337 	struct dentry	*dentry)
338 {
339 	struct xfs_name	name;
340 	int		error;
341 
342 	xfs_dentry_to_name(&name, dentry);
343 
344 	error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
345 	if (error)
346 		return error;
347 
348 	/*
349 	 * With unlink, the VFS makes the dentry "negative": no inode,
350 	 * but still hashed. This is incompatible with case-insensitive
351 	 * mode, so invalidate (unhash) the dentry in CI-mode.
352 	 */
353 	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
354 		d_invalidate(dentry);
355 	return 0;
356 }
357 
358 STATIC int
359 xfs_vn_symlink(
360 	struct inode	*dir,
361 	struct dentry	*dentry,
362 	const char	*symname)
363 {
364 	struct inode	*inode;
365 	struct xfs_inode *cip = NULL;
366 	struct xfs_name	name;
367 	int		error;
368 	mode_t		mode;
369 
370 	mode = S_IFLNK |
371 		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
372 	xfs_dentry_to_name(&name, dentry);
373 
374 	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
375 	if (unlikely(error))
376 		goto out;
377 
378 	inode = VFS_I(cip);
379 
380 	error = xfs_init_security(inode, dir, &dentry->d_name);
381 	if (unlikely(error))
382 		goto out_cleanup_inode;
383 
384 	d_instantiate(dentry, inode);
385 	return 0;
386 
387  out_cleanup_inode:
388 	xfs_cleanup_inode(dir, inode, dentry);
389  out:
390 	return -error;
391 }
392 
393 STATIC int
394 xfs_vn_rename(
395 	struct inode	*odir,
396 	struct dentry	*odentry,
397 	struct inode	*ndir,
398 	struct dentry	*ndentry)
399 {
400 	struct inode	*new_inode = ndentry->d_inode;
401 	struct xfs_name	oname;
402 	struct xfs_name	nname;
403 
404 	xfs_dentry_to_name(&oname, odentry);
405 	xfs_dentry_to_name(&nname, ndentry);
406 
407 	return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
408 			   XFS_I(ndir), &nname, new_inode ?
409 			   			XFS_I(new_inode) : NULL);
410 }
411 
412 /*
413  * careful here - this function can get called recursively, so
414  * we need to be very careful about how much stack we use.
415  * uio is kmalloced for this reason...
416  */
417 STATIC void *
418 xfs_vn_follow_link(
419 	struct dentry		*dentry,
420 	struct nameidata	*nd)
421 {
422 	char			*link;
423 	int			error = -ENOMEM;
424 
425 	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
426 	if (!link)
427 		goto out_err;
428 
429 	error = -xfs_readlink(XFS_I(dentry->d_inode), link);
430 	if (unlikely(error))
431 		goto out_kfree;
432 
433 	nd_set_link(nd, link);
434 	return NULL;
435 
436  out_kfree:
437 	kfree(link);
438  out_err:
439 	nd_set_link(nd, ERR_PTR(error));
440 	return NULL;
441 }
442 
443 STATIC void
444 xfs_vn_put_link(
445 	struct dentry	*dentry,
446 	struct nameidata *nd,
447 	void		*p)
448 {
449 	char		*s = nd_get_link(nd);
450 
451 	if (!IS_ERR(s))
452 		kfree(s);
453 }
454 
455 STATIC int
456 xfs_vn_getattr(
457 	struct vfsmount		*mnt,
458 	struct dentry		*dentry,
459 	struct kstat		*stat)
460 {
461 	struct inode		*inode = dentry->d_inode;
462 	struct xfs_inode	*ip = XFS_I(inode);
463 	struct xfs_mount	*mp = ip->i_mount;
464 
465 	trace_xfs_getattr(ip);
466 
467 	if (XFS_FORCED_SHUTDOWN(mp))
468 		return XFS_ERROR(EIO);
469 
470 	stat->size = XFS_ISIZE(ip);
471 	stat->dev = inode->i_sb->s_dev;
472 	stat->mode = ip->i_d.di_mode;
473 	stat->nlink = ip->i_d.di_nlink;
474 	stat->uid = ip->i_d.di_uid;
475 	stat->gid = ip->i_d.di_gid;
476 	stat->ino = ip->i_ino;
477 	stat->atime = inode->i_atime;
478 	stat->mtime = inode->i_mtime;
479 	stat->ctime = inode->i_ctime;
480 	stat->blocks =
481 		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
482 
483 
484 	switch (inode->i_mode & S_IFMT) {
485 	case S_IFBLK:
486 	case S_IFCHR:
487 		stat->blksize = BLKDEV_IOSIZE;
488 		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
489 				   sysv_minor(ip->i_df.if_u2.if_rdev));
490 		break;
491 	default:
492 		if (XFS_IS_REALTIME_INODE(ip)) {
493 			/*
494 			 * If the file blocks are being allocated from a
495 			 * realtime volume, then return the inode's realtime
496 			 * extent size or the realtime volume's extent size.
497 			 */
498 			stat->blksize =
499 				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
500 		} else
501 			stat->blksize = xfs_preferred_iosize(mp);
502 		stat->rdev = 0;
503 		break;
504 	}
505 
506 	return 0;
507 }
508 
509 int
510 xfs_setattr_nonsize(
511 	struct xfs_inode	*ip,
512 	struct iattr		*iattr,
513 	int			flags)
514 {
515 	xfs_mount_t		*mp = ip->i_mount;
516 	struct inode		*inode = VFS_I(ip);
517 	int			mask = iattr->ia_valid;
518 	xfs_trans_t		*tp;
519 	int			error;
520 	uid_t			uid = 0, iuid = 0;
521 	gid_t			gid = 0, igid = 0;
522 	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
523 	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
524 
525 	trace_xfs_setattr(ip);
526 
527 	if (mp->m_flags & XFS_MOUNT_RDONLY)
528 		return XFS_ERROR(EROFS);
529 
530 	if (XFS_FORCED_SHUTDOWN(mp))
531 		return XFS_ERROR(EIO);
532 
533 	error = -inode_change_ok(inode, iattr);
534 	if (error)
535 		return XFS_ERROR(error);
536 
537 	ASSERT((mask & ATTR_SIZE) == 0);
538 
539 	/*
540 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
541 	 * before we start any other transactions. Trying to do this later
542 	 * is messy. We don't care to take a readlock to look at the ids
543 	 * in inode here, because we can't hold it across the trans_reserve.
544 	 * If the IDs do change before we take the ilock, we're covered
545 	 * because the i_*dquot fields will get updated anyway.
546 	 */
547 	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
548 		uint	qflags = 0;
549 
550 		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
551 			uid = iattr->ia_uid;
552 			qflags |= XFS_QMOPT_UQUOTA;
553 		} else {
554 			uid = ip->i_d.di_uid;
555 		}
556 		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
557 			gid = iattr->ia_gid;
558 			qflags |= XFS_QMOPT_GQUOTA;
559 		}  else {
560 			gid = ip->i_d.di_gid;
561 		}
562 
563 		/*
564 		 * We take a reference when we initialize udqp and gdqp,
565 		 * so it is important that we never blindly double trip on
566 		 * the same variable. See xfs_create() for an example.
567 		 */
568 		ASSERT(udqp == NULL);
569 		ASSERT(gdqp == NULL);
570 		error = xfs_qm_vop_dqalloc(ip, uid, gid, xfs_get_projid(ip),
571 					 qflags, &udqp, &gdqp);
572 		if (error)
573 			return error;
574 	}
575 
576 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
577 	error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0);
578 	if (error)
579 		goto out_dqrele;
580 
581 	xfs_ilock(ip, XFS_ILOCK_EXCL);
582 
583 	/*
584 	 * Change file ownership.  Must be the owner or privileged.
585 	 */
586 	if (mask & (ATTR_UID|ATTR_GID)) {
587 		/*
588 		 * These IDs could have changed since we last looked at them.
589 		 * But, we're assured that if the ownership did change
590 		 * while we didn't have the inode locked, inode's dquot(s)
591 		 * would have changed also.
592 		 */
593 		iuid = ip->i_d.di_uid;
594 		igid = ip->i_d.di_gid;
595 		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
596 		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
597 
598 		/*
599 		 * Do a quota reservation only if uid/gid is actually
600 		 * going to change.
601 		 */
602 		if (XFS_IS_QUOTA_RUNNING(mp) &&
603 		    ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
604 		     (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
605 			ASSERT(tp);
606 			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
607 						capable(CAP_FOWNER) ?
608 						XFS_QMOPT_FORCE_RES : 0);
609 			if (error)	/* out of quota */
610 				goto out_trans_cancel;
611 		}
612 	}
613 
614 	xfs_trans_ijoin(tp, ip);
615 
616 	/*
617 	 * Change file ownership.  Must be the owner or privileged.
618 	 */
619 	if (mask & (ATTR_UID|ATTR_GID)) {
620 		/*
621 		 * CAP_FSETID overrides the following restrictions:
622 		 *
623 		 * The set-user-ID and set-group-ID bits of a file will be
624 		 * cleared upon successful return from chown()
625 		 */
626 		if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
627 		    !capable(CAP_FSETID))
628 			ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
629 
630 		/*
631 		 * Change the ownerships and register quota modifications
632 		 * in the transaction.
633 		 */
634 		if (iuid != uid) {
635 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
636 				ASSERT(mask & ATTR_UID);
637 				ASSERT(udqp);
638 				olddquot1 = xfs_qm_vop_chown(tp, ip,
639 							&ip->i_udquot, udqp);
640 			}
641 			ip->i_d.di_uid = uid;
642 			inode->i_uid = uid;
643 		}
644 		if (igid != gid) {
645 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
646 				ASSERT(!XFS_IS_PQUOTA_ON(mp));
647 				ASSERT(mask & ATTR_GID);
648 				ASSERT(gdqp);
649 				olddquot2 = xfs_qm_vop_chown(tp, ip,
650 							&ip->i_gdquot, gdqp);
651 			}
652 			ip->i_d.di_gid = gid;
653 			inode->i_gid = gid;
654 		}
655 	}
656 
657 	/*
658 	 * Change file access modes.
659 	 */
660 	if (mask & ATTR_MODE) {
661 		umode_t mode = iattr->ia_mode;
662 
663 		if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
664 			mode &= ~S_ISGID;
665 
666 		ip->i_d.di_mode &= S_IFMT;
667 		ip->i_d.di_mode |= mode & ~S_IFMT;
668 
669 		inode->i_mode &= S_IFMT;
670 		inode->i_mode |= mode & ~S_IFMT;
671 	}
672 
673 	/*
674 	 * Change file access or modified times.
675 	 */
676 	if (mask & ATTR_ATIME) {
677 		inode->i_atime = iattr->ia_atime;
678 		ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
679 		ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
680 		ip->i_update_core = 1;
681 	}
682 	if (mask & ATTR_CTIME) {
683 		inode->i_ctime = iattr->ia_ctime;
684 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
685 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
686 		ip->i_update_core = 1;
687 	}
688 	if (mask & ATTR_MTIME) {
689 		inode->i_mtime = iattr->ia_mtime;
690 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
691 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
692 		ip->i_update_core = 1;
693 	}
694 
695 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
696 
697 	XFS_STATS_INC(xs_ig_attrchg);
698 
699 	if (mp->m_flags & XFS_MOUNT_WSYNC)
700 		xfs_trans_set_sync(tp);
701 	error = xfs_trans_commit(tp, 0);
702 
703 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
704 
705 	/*
706 	 * Release any dquot(s) the inode had kept before chown.
707 	 */
708 	xfs_qm_dqrele(olddquot1);
709 	xfs_qm_dqrele(olddquot2);
710 	xfs_qm_dqrele(udqp);
711 	xfs_qm_dqrele(gdqp);
712 
713 	if (error)
714 		return XFS_ERROR(error);
715 
716 	/*
717 	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
718 	 * 	     update.  We could avoid this with linked transactions
719 	 * 	     and passing down the transaction pointer all the way
720 	 *	     to attr_set.  No previous user of the generic
721 	 * 	     Posix ACL code seems to care about this issue either.
722 	 */
723 	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
724 		error = -xfs_acl_chmod(inode);
725 		if (error)
726 			return XFS_ERROR(error);
727 	}
728 
729 	return 0;
730 
731 out_trans_cancel:
732 	xfs_trans_cancel(tp, 0);
733 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
734 out_dqrele:
735 	xfs_qm_dqrele(udqp);
736 	xfs_qm_dqrele(gdqp);
737 	return error;
738 }
739 
740 /*
741  * Truncate file.  Must have write permission and not be a directory.
742  */
743 int
744 xfs_setattr_size(
745 	struct xfs_inode	*ip,
746 	struct iattr		*iattr,
747 	int			flags)
748 {
749 	struct xfs_mount	*mp = ip->i_mount;
750 	struct inode		*inode = VFS_I(ip);
751 	int			mask = iattr->ia_valid;
752 	struct xfs_trans	*tp;
753 	int			error;
754 	uint			lock_flags;
755 	uint			commit_flags = 0;
756 
757 	trace_xfs_setattr(ip);
758 
759 	if (mp->m_flags & XFS_MOUNT_RDONLY)
760 		return XFS_ERROR(EROFS);
761 
762 	if (XFS_FORCED_SHUTDOWN(mp))
763 		return XFS_ERROR(EIO);
764 
765 	error = -inode_change_ok(inode, iattr);
766 	if (error)
767 		return XFS_ERROR(error);
768 
769 	ASSERT(S_ISREG(ip->i_d.di_mode));
770 	ASSERT((mask & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
771 			ATTR_MTIME_SET|ATTR_KILL_SUID|ATTR_KILL_SGID|
772 			ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
773 
774 	lock_flags = XFS_ILOCK_EXCL;
775 	if (!(flags & XFS_ATTR_NOLOCK))
776 		lock_flags |= XFS_IOLOCK_EXCL;
777 	xfs_ilock(ip, lock_flags);
778 
779 	/*
780 	 * Short circuit the truncate case for zero length files.
781 	 */
782 	if (iattr->ia_size == 0 &&
783 	    ip->i_size == 0 && ip->i_d.di_nextents == 0) {
784 		if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
785 			goto out_unlock;
786 
787 		/*
788 		 * Use the regular setattr path to update the timestamps.
789 		 */
790 		xfs_iunlock(ip, lock_flags);
791 		iattr->ia_valid &= ~ATTR_SIZE;
792 		return xfs_setattr_nonsize(ip, iattr, 0);
793 	}
794 
795 	/*
796 	 * Make sure that the dquots are attached to the inode.
797 	 */
798 	error = xfs_qm_dqattach_locked(ip, 0);
799 	if (error)
800 		goto out_unlock;
801 
802 	/*
803 	 * Now we can make the changes.  Before we join the inode to the
804 	 * transaction, take care of the part of the truncation that must be
805 	 * done without the inode lock.  This needs to be done before joining
806 	 * the inode to the transaction, because the inode cannot be unlocked
807 	 * once it is a part of the transaction.
808 	 */
809 	if (iattr->ia_size > ip->i_size) {
810 		/*
811 		 * Do the first part of growing a file: zero any data in the
812 		 * last block that is beyond the old EOF.  We need to do this
813 		 * before the inode is joined to the transaction to modify
814 		 * i_size.
815 		 */
816 		error = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
817 		if (error)
818 			goto out_unlock;
819 	}
820 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
821 	lock_flags &= ~XFS_ILOCK_EXCL;
822 
823 	/*
824 	 * We are going to log the inode size change in this transaction so
825 	 * any previous writes that are beyond the on disk EOF and the new
826 	 * EOF that have not been written out need to be written here.  If we
827 	 * do not write the data out, we expose ourselves to the null files
828 	 * problem.
829 	 *
830 	 * Only flush from the on disk size to the smaller of the in memory
831 	 * file size or the new size as that's the range we really care about
832 	 * here and prevents waiting for other data not within the range we
833 	 * care about here.
834 	 */
835 	if (ip->i_size != ip->i_d.di_size && iattr->ia_size > ip->i_d.di_size) {
836 		error = xfs_flush_pages(ip, ip->i_d.di_size, iattr->ia_size,
837 					XBF_ASYNC, FI_NONE);
838 		if (error)
839 			goto out_unlock;
840 	}
841 
842 	/*
843 	 * Wait for all I/O to complete.
844 	 */
845 	xfs_ioend_wait(ip);
846 
847 	error = -block_truncate_page(inode->i_mapping, iattr->ia_size,
848 				     xfs_get_blocks);
849 	if (error)
850 		goto out_unlock;
851 
852 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
853 	error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
854 				 XFS_TRANS_PERM_LOG_RES,
855 				 XFS_ITRUNCATE_LOG_COUNT);
856 	if (error)
857 		goto out_trans_cancel;
858 
859 	truncate_setsize(inode, iattr->ia_size);
860 
861 	commit_flags = XFS_TRANS_RELEASE_LOG_RES;
862 	lock_flags |= XFS_ILOCK_EXCL;
863 
864 	xfs_ilock(ip, XFS_ILOCK_EXCL);
865 
866 	xfs_trans_ijoin(tp, ip);
867 
868 	/*
869 	 * Only change the c/mtime if we are changing the size or we are
870 	 * explicitly asked to change it.  This handles the semantic difference
871 	 * between truncate() and ftruncate() as implemented in the VFS.
872 	 *
873 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
874 	 * special case where we need to update the times despite not having
875 	 * these flags set.  For all other operations the VFS set these flags
876 	 * explicitly if it wants a timestamp update.
877 	 */
878 	if (iattr->ia_size != ip->i_size &&
879 	    (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
880 		iattr->ia_ctime = iattr->ia_mtime =
881 			current_fs_time(inode->i_sb);
882 		mask |= ATTR_CTIME | ATTR_MTIME;
883 	}
884 
885 	if (iattr->ia_size > ip->i_size) {
886 		ip->i_d.di_size = iattr->ia_size;
887 		ip->i_size = iattr->ia_size;
888 	} else if (iattr->ia_size <= ip->i_size ||
889 		   (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
890 		error = xfs_itruncate_data(&tp, ip, iattr->ia_size);
891 		if (error)
892 			goto out_trans_abort;
893 
894 		/*
895 		 * Truncated "down", so we're removing references to old data
896 		 * here - if we delay flushing for a long time, we expose
897 		 * ourselves unduly to the notorious NULL files problem.  So,
898 		 * we mark this inode and flush it when the file is closed,
899 		 * and do not wait the usual (long) time for writeout.
900 		 */
901 		xfs_iflags_set(ip, XFS_ITRUNCATED);
902 	}
903 
904 	if (mask & ATTR_CTIME) {
905 		inode->i_ctime = iattr->ia_ctime;
906 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
907 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
908 		ip->i_update_core = 1;
909 	}
910 	if (mask & ATTR_MTIME) {
911 		inode->i_mtime = iattr->ia_mtime;
912 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
913 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
914 		ip->i_update_core = 1;
915 	}
916 
917 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
918 
919 	XFS_STATS_INC(xs_ig_attrchg);
920 
921 	if (mp->m_flags & XFS_MOUNT_WSYNC)
922 		xfs_trans_set_sync(tp);
923 
924 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
925 out_unlock:
926 	if (lock_flags)
927 		xfs_iunlock(ip, lock_flags);
928 	return error;
929 
930 out_trans_abort:
931 	commit_flags |= XFS_TRANS_ABORT;
932 out_trans_cancel:
933 	xfs_trans_cancel(tp, commit_flags);
934 	goto out_unlock;
935 }
936 
937 STATIC int
938 xfs_vn_setattr(
939 	struct dentry	*dentry,
940 	struct iattr	*iattr)
941 {
942 	if (iattr->ia_valid & ATTR_SIZE)
943 		return -xfs_setattr_size(XFS_I(dentry->d_inode), iattr, 0);
944 	return -xfs_setattr_nonsize(XFS_I(dentry->d_inode), iattr, 0);
945 }
946 
947 #define XFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
948 
949 /*
950  * Call fiemap helper to fill in user data.
951  * Returns positive errors to xfs_getbmap.
952  */
953 STATIC int
954 xfs_fiemap_format(
955 	void			**arg,
956 	struct getbmapx		*bmv,
957 	int			*full)
958 {
959 	int			error;
960 	struct fiemap_extent_info *fieinfo = *arg;
961 	u32			fiemap_flags = 0;
962 	u64			logical, physical, length;
963 
964 	/* Do nothing for a hole */
965 	if (bmv->bmv_block == -1LL)
966 		return 0;
967 
968 	logical = BBTOB(bmv->bmv_offset);
969 	physical = BBTOB(bmv->bmv_block);
970 	length = BBTOB(bmv->bmv_length);
971 
972 	if (bmv->bmv_oflags & BMV_OF_PREALLOC)
973 		fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
974 	else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
975 		fiemap_flags |= FIEMAP_EXTENT_DELALLOC;
976 		physical = 0;   /* no block yet */
977 	}
978 	if (bmv->bmv_oflags & BMV_OF_LAST)
979 		fiemap_flags |= FIEMAP_EXTENT_LAST;
980 
981 	error = fiemap_fill_next_extent(fieinfo, logical, physical,
982 					length, fiemap_flags);
983 	if (error > 0) {
984 		error = 0;
985 		*full = 1;	/* user array now full */
986 	}
987 
988 	return -error;
989 }
990 
991 STATIC int
992 xfs_vn_fiemap(
993 	struct inode		*inode,
994 	struct fiemap_extent_info *fieinfo,
995 	u64			start,
996 	u64			length)
997 {
998 	xfs_inode_t		*ip = XFS_I(inode);
999 	struct getbmapx		bm;
1000 	int			error;
1001 
1002 	error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1003 	if (error)
1004 		return error;
1005 
1006 	/* Set up bmap header for xfs internal routine */
1007 	bm.bmv_offset = BTOBB(start);
1008 	/* Special case for whole file */
1009 	if (length == FIEMAP_MAX_OFFSET)
1010 		bm.bmv_length = -1LL;
1011 	else
1012 		bm.bmv_length = BTOBB(length);
1013 
1014 	/* We add one because in getbmap world count includes the header */
1015 	bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1016 					fieinfo->fi_extents_max + 1;
1017 	bm.bmv_count = min_t(__s32, bm.bmv_count,
1018 			     (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1019 	bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1020 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1021 		bm.bmv_iflags |= BMV_IF_ATTRFORK;
1022 	if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1023 		bm.bmv_iflags |= BMV_IF_DELALLOC;
1024 
1025 	error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1026 	if (error)
1027 		return -error;
1028 
1029 	return 0;
1030 }
1031 
1032 static const struct inode_operations xfs_inode_operations = {
1033 	.get_acl		= xfs_get_acl,
1034 	.getattr		= xfs_vn_getattr,
1035 	.setattr		= xfs_vn_setattr,
1036 	.setxattr		= generic_setxattr,
1037 	.getxattr		= generic_getxattr,
1038 	.removexattr		= generic_removexattr,
1039 	.listxattr		= xfs_vn_listxattr,
1040 	.fiemap			= xfs_vn_fiemap,
1041 };
1042 
1043 static const struct inode_operations xfs_dir_inode_operations = {
1044 	.create			= xfs_vn_create,
1045 	.lookup			= xfs_vn_lookup,
1046 	.link			= xfs_vn_link,
1047 	.unlink			= xfs_vn_unlink,
1048 	.symlink		= xfs_vn_symlink,
1049 	.mkdir			= xfs_vn_mkdir,
1050 	/*
1051 	 * Yes, XFS uses the same method for rmdir and unlink.
1052 	 *
1053 	 * There are some subtile differences deeper in the code,
1054 	 * but we use S_ISDIR to check for those.
1055 	 */
1056 	.rmdir			= xfs_vn_unlink,
1057 	.mknod			= xfs_vn_mknod,
1058 	.rename			= xfs_vn_rename,
1059 	.get_acl		= xfs_get_acl,
1060 	.getattr		= xfs_vn_getattr,
1061 	.setattr		= xfs_vn_setattr,
1062 	.setxattr		= generic_setxattr,
1063 	.getxattr		= generic_getxattr,
1064 	.removexattr		= generic_removexattr,
1065 	.listxattr		= xfs_vn_listxattr,
1066 };
1067 
1068 static const struct inode_operations xfs_dir_ci_inode_operations = {
1069 	.create			= xfs_vn_create,
1070 	.lookup			= xfs_vn_ci_lookup,
1071 	.link			= xfs_vn_link,
1072 	.unlink			= xfs_vn_unlink,
1073 	.symlink		= xfs_vn_symlink,
1074 	.mkdir			= xfs_vn_mkdir,
1075 	/*
1076 	 * Yes, XFS uses the same method for rmdir and unlink.
1077 	 *
1078 	 * There are some subtile differences deeper in the code,
1079 	 * but we use S_ISDIR to check for those.
1080 	 */
1081 	.rmdir			= xfs_vn_unlink,
1082 	.mknod			= xfs_vn_mknod,
1083 	.rename			= xfs_vn_rename,
1084 	.get_acl		= xfs_get_acl,
1085 	.getattr		= xfs_vn_getattr,
1086 	.setattr		= xfs_vn_setattr,
1087 	.setxattr		= generic_setxattr,
1088 	.getxattr		= generic_getxattr,
1089 	.removexattr		= generic_removexattr,
1090 	.listxattr		= xfs_vn_listxattr,
1091 };
1092 
1093 static const struct inode_operations xfs_symlink_inode_operations = {
1094 	.readlink		= generic_readlink,
1095 	.follow_link		= xfs_vn_follow_link,
1096 	.put_link		= xfs_vn_put_link,
1097 	.get_acl		= xfs_get_acl,
1098 	.getattr		= xfs_vn_getattr,
1099 	.setattr		= xfs_vn_setattr,
1100 	.setxattr		= generic_setxattr,
1101 	.getxattr		= generic_getxattr,
1102 	.removexattr		= generic_removexattr,
1103 	.listxattr		= xfs_vn_listxattr,
1104 };
1105 
1106 STATIC void
1107 xfs_diflags_to_iflags(
1108 	struct inode		*inode,
1109 	struct xfs_inode	*ip)
1110 {
1111 	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1112 		inode->i_flags |= S_IMMUTABLE;
1113 	else
1114 		inode->i_flags &= ~S_IMMUTABLE;
1115 	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1116 		inode->i_flags |= S_APPEND;
1117 	else
1118 		inode->i_flags &= ~S_APPEND;
1119 	if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1120 		inode->i_flags |= S_SYNC;
1121 	else
1122 		inode->i_flags &= ~S_SYNC;
1123 	if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1124 		inode->i_flags |= S_NOATIME;
1125 	else
1126 		inode->i_flags &= ~S_NOATIME;
1127 }
1128 
1129 /*
1130  * Initialize the Linux inode, set up the operation vectors and
1131  * unlock the inode.
1132  *
1133  * When reading existing inodes from disk this is called directly
1134  * from xfs_iget, when creating a new inode it is called from
1135  * xfs_ialloc after setting up the inode.
1136  *
1137  * We are always called with an uninitialised linux inode here.
1138  * We need to initialise the necessary fields and take a reference
1139  * on it.
1140  */
1141 void
1142 xfs_setup_inode(
1143 	struct xfs_inode	*ip)
1144 {
1145 	struct inode		*inode = &ip->i_vnode;
1146 
1147 	inode->i_ino = ip->i_ino;
1148 	inode->i_state = I_NEW;
1149 
1150 	inode_sb_list_add(inode);
1151 	/* make the inode look hashed for the writeback code */
1152 	hlist_add_fake(&inode->i_hash);
1153 
1154 	inode->i_mode	= ip->i_d.di_mode;
1155 	inode->i_nlink	= ip->i_d.di_nlink;
1156 	inode->i_uid	= ip->i_d.di_uid;
1157 	inode->i_gid	= ip->i_d.di_gid;
1158 
1159 	switch (inode->i_mode & S_IFMT) {
1160 	case S_IFBLK:
1161 	case S_IFCHR:
1162 		inode->i_rdev =
1163 			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1164 			      sysv_minor(ip->i_df.if_u2.if_rdev));
1165 		break;
1166 	default:
1167 		inode->i_rdev = 0;
1168 		break;
1169 	}
1170 
1171 	inode->i_generation = ip->i_d.di_gen;
1172 	i_size_write(inode, ip->i_d.di_size);
1173 	inode->i_atime.tv_sec	= ip->i_d.di_atime.t_sec;
1174 	inode->i_atime.tv_nsec	= ip->i_d.di_atime.t_nsec;
1175 	inode->i_mtime.tv_sec	= ip->i_d.di_mtime.t_sec;
1176 	inode->i_mtime.tv_nsec	= ip->i_d.di_mtime.t_nsec;
1177 	inode->i_ctime.tv_sec	= ip->i_d.di_ctime.t_sec;
1178 	inode->i_ctime.tv_nsec	= ip->i_d.di_ctime.t_nsec;
1179 	xfs_diflags_to_iflags(inode, ip);
1180 
1181 	switch (inode->i_mode & S_IFMT) {
1182 	case S_IFREG:
1183 		inode->i_op = &xfs_inode_operations;
1184 		inode->i_fop = &xfs_file_operations;
1185 		inode->i_mapping->a_ops = &xfs_address_space_operations;
1186 		break;
1187 	case S_IFDIR:
1188 		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1189 			inode->i_op = &xfs_dir_ci_inode_operations;
1190 		else
1191 			inode->i_op = &xfs_dir_inode_operations;
1192 		inode->i_fop = &xfs_dir_file_operations;
1193 		break;
1194 	case S_IFLNK:
1195 		inode->i_op = &xfs_symlink_inode_operations;
1196 		if (!(ip->i_df.if_flags & XFS_IFINLINE))
1197 			inode->i_mapping->a_ops = &xfs_address_space_operations;
1198 		break;
1199 	default:
1200 		inode->i_op = &xfs_inode_operations;
1201 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1202 		break;
1203 	}
1204 
1205 	/*
1206 	 * If there is no attribute fork no ACL can exist on this inode,
1207 	 * and it can't have any file capabilities attached to it either.
1208 	 */
1209 	if (!XFS_IFORK_Q(ip)) {
1210 		inode_has_no_xattr(inode);
1211 		cache_no_acl(inode);
1212 	}
1213 
1214 	xfs_iflags_clear(ip, XFS_INEW);
1215 	barrier();
1216 
1217 	unlock_new_inode(inode);
1218 }
1219