xref: /linux/fs/xfs/xfs_iops.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
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 int xfs_initxattrs(struct inode *inode, const struct xattr *xattr_array,
107 		   void *fs_info)
108 {
109 	const struct xattr *xattr;
110 	struct xfs_inode *ip = XFS_I(inode);
111 	int error = 0;
112 
113 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
114 		error = xfs_attr_set(ip, xattr->name, xattr->value,
115 				     xattr->value_len, ATTR_SECURE);
116 		if (error < 0)
117 			break;
118 	}
119 	return error;
120 }
121 
122 /*
123  * Hook in SELinux.  This is not quite correct yet, what we really need
124  * here (as we do for default ACLs) is a mechanism by which creation of
125  * these attrs can be journalled at inode creation time (along with the
126  * inode, of course, such that log replay can't cause these to be lost).
127  */
128 
129 STATIC int
130 xfs_init_security(
131 	struct inode	*inode,
132 	struct inode	*dir,
133 	const struct qstr *qstr)
134 {
135 	return security_inode_init_security(inode, dir, qstr,
136 					    &xfs_initxattrs, NULL);
137 }
138 
139 static void
140 xfs_dentry_to_name(
141 	struct xfs_name	*namep,
142 	struct dentry	*dentry)
143 {
144 	namep->name = dentry->d_name.name;
145 	namep->len = dentry->d_name.len;
146 }
147 
148 STATIC void
149 xfs_cleanup_inode(
150 	struct inode	*dir,
151 	struct inode	*inode,
152 	struct dentry	*dentry)
153 {
154 	struct xfs_name	teardown;
155 
156 	/* Oh, the horror.
157 	 * If we can't add the ACL or we fail in
158 	 * xfs_init_security we must back out.
159 	 * ENOSPC can hit here, among other things.
160 	 */
161 	xfs_dentry_to_name(&teardown, dentry);
162 
163 	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
164 	iput(inode);
165 }
166 
167 STATIC int
168 xfs_vn_mknod(
169 	struct inode	*dir,
170 	struct dentry	*dentry,
171 	int		mode,
172 	dev_t		rdev)
173 {
174 	struct inode	*inode;
175 	struct xfs_inode *ip = NULL;
176 	struct posix_acl *default_acl = NULL;
177 	struct xfs_name	name;
178 	int		error;
179 
180 	/*
181 	 * Irix uses Missed'em'V split, but doesn't want to see
182 	 * the upper 5 bits of (14bit) major.
183 	 */
184 	if (S_ISCHR(mode) || S_ISBLK(mode)) {
185 		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
186 			return -EINVAL;
187 		rdev = sysv_encode_dev(rdev);
188 	} else {
189 		rdev = 0;
190 	}
191 
192 	if (IS_POSIXACL(dir)) {
193 		default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
194 		if (IS_ERR(default_acl))
195 			return PTR_ERR(default_acl);
196 
197 		if (!default_acl)
198 			mode &= ~current_umask();
199 	}
200 
201 	xfs_dentry_to_name(&name, dentry);
202 	error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
203 	if (unlikely(error))
204 		goto out_free_acl;
205 
206 	inode = VFS_I(ip);
207 
208 	error = xfs_init_security(inode, dir, &dentry->d_name);
209 	if (unlikely(error))
210 		goto out_cleanup_inode;
211 
212 	if (default_acl) {
213 		error = -xfs_inherit_acl(inode, default_acl);
214 		default_acl = NULL;
215 		if (unlikely(error))
216 			goto out_cleanup_inode;
217 	}
218 
219 
220 	d_instantiate(dentry, inode);
221 	return -error;
222 
223  out_cleanup_inode:
224 	xfs_cleanup_inode(dir, inode, dentry);
225  out_free_acl:
226 	posix_acl_release(default_acl);
227 	return -error;
228 }
229 
230 STATIC int
231 xfs_vn_create(
232 	struct inode	*dir,
233 	struct dentry	*dentry,
234 	int		mode,
235 	struct nameidata *nd)
236 {
237 	return xfs_vn_mknod(dir, dentry, mode, 0);
238 }
239 
240 STATIC int
241 xfs_vn_mkdir(
242 	struct inode	*dir,
243 	struct dentry	*dentry,
244 	int		mode)
245 {
246 	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
247 }
248 
249 STATIC struct dentry *
250 xfs_vn_lookup(
251 	struct inode	*dir,
252 	struct dentry	*dentry,
253 	struct nameidata *nd)
254 {
255 	struct xfs_inode *cip;
256 	struct xfs_name	name;
257 	int		error;
258 
259 	if (dentry->d_name.len >= MAXNAMELEN)
260 		return ERR_PTR(-ENAMETOOLONG);
261 
262 	xfs_dentry_to_name(&name, dentry);
263 	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
264 	if (unlikely(error)) {
265 		if (unlikely(error != ENOENT))
266 			return ERR_PTR(-error);
267 		d_add(dentry, NULL);
268 		return NULL;
269 	}
270 
271 	return d_splice_alias(VFS_I(cip), dentry);
272 }
273 
274 STATIC struct dentry *
275 xfs_vn_ci_lookup(
276 	struct inode	*dir,
277 	struct dentry	*dentry,
278 	struct nameidata *nd)
279 {
280 	struct xfs_inode *ip;
281 	struct xfs_name	xname;
282 	struct xfs_name ci_name;
283 	struct qstr	dname;
284 	int		error;
285 
286 	if (dentry->d_name.len >= MAXNAMELEN)
287 		return ERR_PTR(-ENAMETOOLONG);
288 
289 	xfs_dentry_to_name(&xname, dentry);
290 	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
291 	if (unlikely(error)) {
292 		if (unlikely(error != ENOENT))
293 			return ERR_PTR(-error);
294 		/*
295 		 * call d_add(dentry, NULL) here when d_drop_negative_children
296 		 * is called in xfs_vn_mknod (ie. allow negative dentries
297 		 * with CI filesystems).
298 		 */
299 		return NULL;
300 	}
301 
302 	/* if exact match, just splice and exit */
303 	if (!ci_name.name)
304 		return d_splice_alias(VFS_I(ip), dentry);
305 
306 	/* else case-insensitive match... */
307 	dname.name = ci_name.name;
308 	dname.len = ci_name.len;
309 	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
310 	kmem_free(ci_name.name);
311 	return dentry;
312 }
313 
314 STATIC int
315 xfs_vn_link(
316 	struct dentry	*old_dentry,
317 	struct inode	*dir,
318 	struct dentry	*dentry)
319 {
320 	struct inode	*inode = old_dentry->d_inode;
321 	struct xfs_name	name;
322 	int		error;
323 
324 	xfs_dentry_to_name(&name, dentry);
325 
326 	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
327 	if (unlikely(error))
328 		return -error;
329 
330 	ihold(inode);
331 	d_instantiate(dentry, inode);
332 	return 0;
333 }
334 
335 STATIC int
336 xfs_vn_unlink(
337 	struct inode	*dir,
338 	struct dentry	*dentry)
339 {
340 	struct xfs_name	name;
341 	int		error;
342 
343 	xfs_dentry_to_name(&name, dentry);
344 
345 	error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
346 	if (error)
347 		return error;
348 
349 	/*
350 	 * With unlink, the VFS makes the dentry "negative": no inode,
351 	 * but still hashed. This is incompatible with case-insensitive
352 	 * mode, so invalidate (unhash) the dentry in CI-mode.
353 	 */
354 	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
355 		d_invalidate(dentry);
356 	return 0;
357 }
358 
359 STATIC int
360 xfs_vn_symlink(
361 	struct inode	*dir,
362 	struct dentry	*dentry,
363 	const char	*symname)
364 {
365 	struct inode	*inode;
366 	struct xfs_inode *cip = NULL;
367 	struct xfs_name	name;
368 	int		error;
369 	mode_t		mode;
370 
371 	mode = S_IFLNK |
372 		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
373 	xfs_dentry_to_name(&name, dentry);
374 
375 	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
376 	if (unlikely(error))
377 		goto out;
378 
379 	inode = VFS_I(cip);
380 
381 	error = xfs_init_security(inode, dir, &dentry->d_name);
382 	if (unlikely(error))
383 		goto out_cleanup_inode;
384 
385 	d_instantiate(dentry, inode);
386 	return 0;
387 
388  out_cleanup_inode:
389 	xfs_cleanup_inode(dir, inode, dentry);
390  out:
391 	return -error;
392 }
393 
394 STATIC int
395 xfs_vn_rename(
396 	struct inode	*odir,
397 	struct dentry	*odentry,
398 	struct inode	*ndir,
399 	struct dentry	*ndentry)
400 {
401 	struct inode	*new_inode = ndentry->d_inode;
402 	struct xfs_name	oname;
403 	struct xfs_name	nname;
404 
405 	xfs_dentry_to_name(&oname, odentry);
406 	xfs_dentry_to_name(&nname, ndentry);
407 
408 	return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
409 			   XFS_I(ndir), &nname, new_inode ?
410 			   			XFS_I(new_inode) : NULL);
411 }
412 
413 /*
414  * careful here - this function can get called recursively, so
415  * we need to be very careful about how much stack we use.
416  * uio is kmalloced for this reason...
417  */
418 STATIC void *
419 xfs_vn_follow_link(
420 	struct dentry		*dentry,
421 	struct nameidata	*nd)
422 {
423 	char			*link;
424 	int			error = -ENOMEM;
425 
426 	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
427 	if (!link)
428 		goto out_err;
429 
430 	error = -xfs_readlink(XFS_I(dentry->d_inode), link);
431 	if (unlikely(error))
432 		goto out_kfree;
433 
434 	nd_set_link(nd, link);
435 	return NULL;
436 
437  out_kfree:
438 	kfree(link);
439  out_err:
440 	nd_set_link(nd, ERR_PTR(error));
441 	return NULL;
442 }
443 
444 STATIC void
445 xfs_vn_put_link(
446 	struct dentry	*dentry,
447 	struct nameidata *nd,
448 	void		*p)
449 {
450 	char		*s = nd_get_link(nd);
451 
452 	if (!IS_ERR(s))
453 		kfree(s);
454 }
455 
456 STATIC int
457 xfs_vn_getattr(
458 	struct vfsmount		*mnt,
459 	struct dentry		*dentry,
460 	struct kstat		*stat)
461 {
462 	struct inode		*inode = dentry->d_inode;
463 	struct xfs_inode	*ip = XFS_I(inode);
464 	struct xfs_mount	*mp = ip->i_mount;
465 
466 	trace_xfs_getattr(ip);
467 
468 	if (XFS_FORCED_SHUTDOWN(mp))
469 		return -XFS_ERROR(EIO);
470 
471 	stat->size = XFS_ISIZE(ip);
472 	stat->dev = inode->i_sb->s_dev;
473 	stat->mode = ip->i_d.di_mode;
474 	stat->nlink = ip->i_d.di_nlink;
475 	stat->uid = ip->i_d.di_uid;
476 	stat->gid = ip->i_d.di_gid;
477 	stat->ino = ip->i_ino;
478 	stat->atime = inode->i_atime;
479 	stat->mtime = inode->i_mtime;
480 	stat->ctime = inode->i_ctime;
481 	stat->blocks =
482 		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
483 
484 
485 	switch (inode->i_mode & S_IFMT) {
486 	case S_IFBLK:
487 	case S_IFCHR:
488 		stat->blksize = BLKDEV_IOSIZE;
489 		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
490 				   sysv_minor(ip->i_df.if_u2.if_rdev));
491 		break;
492 	default:
493 		if (XFS_IS_REALTIME_INODE(ip)) {
494 			/*
495 			 * If the file blocks are being allocated from a
496 			 * realtime volume, then return the inode's realtime
497 			 * extent size or the realtime volume's extent size.
498 			 */
499 			stat->blksize =
500 				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
501 		} else
502 			stat->blksize = xfs_preferred_iosize(mp);
503 		stat->rdev = 0;
504 		break;
505 	}
506 
507 	return 0;
508 }
509 
510 int
511 xfs_setattr_nonsize(
512 	struct xfs_inode	*ip,
513 	struct iattr		*iattr,
514 	int			flags)
515 {
516 	xfs_mount_t		*mp = ip->i_mount;
517 	struct inode		*inode = VFS_I(ip);
518 	int			mask = iattr->ia_valid;
519 	xfs_trans_t		*tp;
520 	int			error;
521 	uid_t			uid = 0, iuid = 0;
522 	gid_t			gid = 0, igid = 0;
523 	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
524 	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
525 
526 	trace_xfs_setattr(ip);
527 
528 	if (mp->m_flags & XFS_MOUNT_RDONLY)
529 		return XFS_ERROR(EROFS);
530 
531 	if (XFS_FORCED_SHUTDOWN(mp))
532 		return XFS_ERROR(EIO);
533 
534 	error = -inode_change_ok(inode, iattr);
535 	if (error)
536 		return XFS_ERROR(error);
537 
538 	ASSERT((mask & ATTR_SIZE) == 0);
539 
540 	/*
541 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
542 	 * before we start any other transactions. Trying to do this later
543 	 * is messy. We don't care to take a readlock to look at the ids
544 	 * in inode here, because we can't hold it across the trans_reserve.
545 	 * If the IDs do change before we take the ilock, we're covered
546 	 * because the i_*dquot fields will get updated anyway.
547 	 */
548 	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
549 		uint	qflags = 0;
550 
551 		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
552 			uid = iattr->ia_uid;
553 			qflags |= XFS_QMOPT_UQUOTA;
554 		} else {
555 			uid = ip->i_d.di_uid;
556 		}
557 		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
558 			gid = iattr->ia_gid;
559 			qflags |= XFS_QMOPT_GQUOTA;
560 		}  else {
561 			gid = ip->i_d.di_gid;
562 		}
563 
564 		/*
565 		 * We take a reference when we initialize udqp and gdqp,
566 		 * so it is important that we never blindly double trip on
567 		 * the same variable. See xfs_create() for an example.
568 		 */
569 		ASSERT(udqp == NULL);
570 		ASSERT(gdqp == NULL);
571 		error = xfs_qm_vop_dqalloc(ip, uid, gid, xfs_get_projid(ip),
572 					 qflags, &udqp, &gdqp);
573 		if (error)
574 			return error;
575 	}
576 
577 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
578 	error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0);
579 	if (error)
580 		goto out_dqrele;
581 
582 	xfs_ilock(ip, XFS_ILOCK_EXCL);
583 
584 	/*
585 	 * Change file ownership.  Must be the owner or privileged.
586 	 */
587 	if (mask & (ATTR_UID|ATTR_GID)) {
588 		/*
589 		 * These IDs could have changed since we last looked at them.
590 		 * But, we're assured that if the ownership did change
591 		 * while we didn't have the inode locked, inode's dquot(s)
592 		 * would have changed also.
593 		 */
594 		iuid = ip->i_d.di_uid;
595 		igid = ip->i_d.di_gid;
596 		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
597 		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
598 
599 		/*
600 		 * Do a quota reservation only if uid/gid is actually
601 		 * going to change.
602 		 */
603 		if (XFS_IS_QUOTA_RUNNING(mp) &&
604 		    ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
605 		     (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
606 			ASSERT(tp);
607 			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
608 						capable(CAP_FOWNER) ?
609 						XFS_QMOPT_FORCE_RES : 0);
610 			if (error)	/* out of quota */
611 				goto out_trans_cancel;
612 		}
613 	}
614 
615 	xfs_trans_ijoin(tp, ip, 0);
616 
617 	/*
618 	 * Change file ownership.  Must be the owner or privileged.
619 	 */
620 	if (mask & (ATTR_UID|ATTR_GID)) {
621 		/*
622 		 * CAP_FSETID overrides the following restrictions:
623 		 *
624 		 * The set-user-ID and set-group-ID bits of a file will be
625 		 * cleared upon successful return from chown()
626 		 */
627 		if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
628 		    !capable(CAP_FSETID))
629 			ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
630 
631 		/*
632 		 * Change the ownerships and register quota modifications
633 		 * in the transaction.
634 		 */
635 		if (iuid != uid) {
636 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
637 				ASSERT(mask & ATTR_UID);
638 				ASSERT(udqp);
639 				olddquot1 = xfs_qm_vop_chown(tp, ip,
640 							&ip->i_udquot, udqp);
641 			}
642 			ip->i_d.di_uid = uid;
643 			inode->i_uid = uid;
644 		}
645 		if (igid != gid) {
646 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
647 				ASSERT(!XFS_IS_PQUOTA_ON(mp));
648 				ASSERT(mask & ATTR_GID);
649 				ASSERT(gdqp);
650 				olddquot2 = xfs_qm_vop_chown(tp, ip,
651 							&ip->i_gdquot, gdqp);
652 			}
653 			ip->i_d.di_gid = gid;
654 			inode->i_gid = gid;
655 		}
656 	}
657 
658 	/*
659 	 * Change file access modes.
660 	 */
661 	if (mask & ATTR_MODE) {
662 		umode_t mode = iattr->ia_mode;
663 
664 		if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
665 			mode &= ~S_ISGID;
666 
667 		ip->i_d.di_mode &= S_IFMT;
668 		ip->i_d.di_mode |= mode & ~S_IFMT;
669 
670 		inode->i_mode &= S_IFMT;
671 		inode->i_mode |= mode & ~S_IFMT;
672 	}
673 
674 	/*
675 	 * Change file access or modified times.
676 	 */
677 	if (mask & ATTR_ATIME) {
678 		inode->i_atime = iattr->ia_atime;
679 		ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
680 		ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
681 		ip->i_update_core = 1;
682 	}
683 	if (mask & ATTR_CTIME) {
684 		inode->i_ctime = iattr->ia_ctime;
685 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
686 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
687 		ip->i_update_core = 1;
688 	}
689 	if (mask & ATTR_MTIME) {
690 		inode->i_mtime = iattr->ia_mtime;
691 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
692 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
693 		ip->i_update_core = 1;
694 	}
695 
696 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
697 
698 	XFS_STATS_INC(xs_ig_attrchg);
699 
700 	if (mp->m_flags & XFS_MOUNT_WSYNC)
701 		xfs_trans_set_sync(tp);
702 	error = xfs_trans_commit(tp, 0);
703 
704 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
705 
706 	/*
707 	 * Release any dquot(s) the inode had kept before chown.
708 	 */
709 	xfs_qm_dqrele(olddquot1);
710 	xfs_qm_dqrele(olddquot2);
711 	xfs_qm_dqrele(udqp);
712 	xfs_qm_dqrele(gdqp);
713 
714 	if (error)
715 		return XFS_ERROR(error);
716 
717 	/*
718 	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
719 	 * 	     update.  We could avoid this with linked transactions
720 	 * 	     and passing down the transaction pointer all the way
721 	 *	     to attr_set.  No previous user of the generic
722 	 * 	     Posix ACL code seems to care about this issue either.
723 	 */
724 	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
725 		error = -xfs_acl_chmod(inode);
726 		if (error)
727 			return XFS_ERROR(error);
728 	}
729 
730 	return 0;
731 
732 out_trans_cancel:
733 	xfs_trans_cancel(tp, 0);
734 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
735 out_dqrele:
736 	xfs_qm_dqrele(udqp);
737 	xfs_qm_dqrele(gdqp);
738 	return error;
739 }
740 
741 /*
742  * Truncate file.  Must have write permission and not be a directory.
743  */
744 int
745 xfs_setattr_size(
746 	struct xfs_inode	*ip,
747 	struct iattr		*iattr,
748 	int			flags)
749 {
750 	struct xfs_mount	*mp = ip->i_mount;
751 	struct inode		*inode = VFS_I(ip);
752 	int			mask = iattr->ia_valid;
753 	struct xfs_trans	*tp;
754 	int			error;
755 	uint			lock_flags;
756 	uint			commit_flags = 0;
757 
758 	trace_xfs_setattr(ip);
759 
760 	if (mp->m_flags & XFS_MOUNT_RDONLY)
761 		return XFS_ERROR(EROFS);
762 
763 	if (XFS_FORCED_SHUTDOWN(mp))
764 		return XFS_ERROR(EIO);
765 
766 	error = -inode_change_ok(inode, iattr);
767 	if (error)
768 		return XFS_ERROR(error);
769 
770 	ASSERT(S_ISREG(ip->i_d.di_mode));
771 	ASSERT((mask & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
772 			ATTR_MTIME_SET|ATTR_KILL_SUID|ATTR_KILL_SGID|
773 			ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
774 
775 	lock_flags = XFS_ILOCK_EXCL;
776 	if (!(flags & XFS_ATTR_NOLOCK))
777 		lock_flags |= XFS_IOLOCK_EXCL;
778 	xfs_ilock(ip, lock_flags);
779 
780 	/*
781 	 * Short circuit the truncate case for zero length files.
782 	 */
783 	if (iattr->ia_size == 0 &&
784 	    ip->i_size == 0 && ip->i_d.di_nextents == 0) {
785 		if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
786 			goto out_unlock;
787 
788 		/*
789 		 * Use the regular setattr path to update the timestamps.
790 		 */
791 		xfs_iunlock(ip, lock_flags);
792 		iattr->ia_valid &= ~ATTR_SIZE;
793 		return xfs_setattr_nonsize(ip, iattr, 0);
794 	}
795 
796 	/*
797 	 * Make sure that the dquots are attached to the inode.
798 	 */
799 	error = xfs_qm_dqattach_locked(ip, 0);
800 	if (error)
801 		goto out_unlock;
802 
803 	/*
804 	 * Now we can make the changes.  Before we join the inode to the
805 	 * transaction, take care of the part of the truncation that must be
806 	 * done without the inode lock.  This needs to be done before joining
807 	 * the inode to the transaction, because the inode cannot be unlocked
808 	 * once it is a part of the transaction.
809 	 */
810 	if (iattr->ia_size > ip->i_size) {
811 		/*
812 		 * Do the first part of growing a file: zero any data in the
813 		 * last block that is beyond the old EOF.  We need to do this
814 		 * before the inode is joined to the transaction to modify
815 		 * i_size.
816 		 */
817 		error = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
818 		if (error)
819 			goto out_unlock;
820 	}
821 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
822 	lock_flags &= ~XFS_ILOCK_EXCL;
823 
824 	/*
825 	 * We are going to log the inode size change in this transaction so
826 	 * any previous writes that are beyond the on disk EOF and the new
827 	 * EOF that have not been written out need to be written here.  If we
828 	 * do not write the data out, we expose ourselves to the null files
829 	 * problem.
830 	 *
831 	 * Only flush from the on disk size to the smaller of the in memory
832 	 * file size or the new size as that's the range we really care about
833 	 * here and prevents waiting for other data not within the range we
834 	 * care about here.
835 	 */
836 	if (ip->i_size != ip->i_d.di_size && iattr->ia_size > ip->i_d.di_size) {
837 		error = xfs_flush_pages(ip, ip->i_d.di_size, iattr->ia_size, 0,
838 					FI_NONE);
839 		if (error)
840 			goto out_unlock;
841 	}
842 
843 	/*
844 	 * Wait for all direct I/O to complete.
845 	 */
846 	inode_dio_wait(inode);
847 
848 	error = -block_truncate_page(inode->i_mapping, iattr->ia_size,
849 				     xfs_get_blocks);
850 	if (error)
851 		goto out_unlock;
852 
853 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
854 	error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
855 				 XFS_TRANS_PERM_LOG_RES,
856 				 XFS_ITRUNCATE_LOG_COUNT);
857 	if (error)
858 		goto out_trans_cancel;
859 
860 	truncate_setsize(inode, iattr->ia_size);
861 
862 	commit_flags = XFS_TRANS_RELEASE_LOG_RES;
863 	lock_flags |= XFS_ILOCK_EXCL;
864 
865 	xfs_ilock(ip, XFS_ILOCK_EXCL);
866 
867 	xfs_trans_ijoin(tp, ip, 0);
868 
869 	/*
870 	 * Only change the c/mtime if we are changing the size or we are
871 	 * explicitly asked to change it.  This handles the semantic difference
872 	 * between truncate() and ftruncate() as implemented in the VFS.
873 	 *
874 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
875 	 * special case where we need to update the times despite not having
876 	 * these flags set.  For all other operations the VFS set these flags
877 	 * explicitly if it wants a timestamp update.
878 	 */
879 	if (iattr->ia_size != ip->i_size &&
880 	    (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
881 		iattr->ia_ctime = iattr->ia_mtime =
882 			current_fs_time(inode->i_sb);
883 		mask |= ATTR_CTIME | ATTR_MTIME;
884 	}
885 
886 	if (iattr->ia_size > ip->i_size) {
887 		ip->i_d.di_size = iattr->ia_size;
888 		ip->i_size = iattr->ia_size;
889 	} else if (iattr->ia_size <= ip->i_size ||
890 		   (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
891 		error = xfs_itruncate_data(&tp, ip, iattr->ia_size);
892 		if (error)
893 			goto out_trans_abort;
894 
895 		/*
896 		 * Truncated "down", so we're removing references to old data
897 		 * here - if we delay flushing for a long time, we expose
898 		 * ourselves unduly to the notorious NULL files problem.  So,
899 		 * we mark this inode and flush it when the file is closed,
900 		 * and do not wait the usual (long) time for writeout.
901 		 */
902 		xfs_iflags_set(ip, XFS_ITRUNCATED);
903 	}
904 
905 	if (mask & ATTR_CTIME) {
906 		inode->i_ctime = iattr->ia_ctime;
907 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
908 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
909 		ip->i_update_core = 1;
910 	}
911 	if (mask & ATTR_MTIME) {
912 		inode->i_mtime = iattr->ia_mtime;
913 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
914 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
915 		ip->i_update_core = 1;
916 	}
917 
918 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
919 
920 	XFS_STATS_INC(xs_ig_attrchg);
921 
922 	if (mp->m_flags & XFS_MOUNT_WSYNC)
923 		xfs_trans_set_sync(tp);
924 
925 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
926 out_unlock:
927 	if (lock_flags)
928 		xfs_iunlock(ip, lock_flags);
929 	return error;
930 
931 out_trans_abort:
932 	commit_flags |= XFS_TRANS_ABORT;
933 out_trans_cancel:
934 	xfs_trans_cancel(tp, commit_flags);
935 	goto out_unlock;
936 }
937 
938 STATIC int
939 xfs_vn_setattr(
940 	struct dentry	*dentry,
941 	struct iattr	*iattr)
942 {
943 	if (iattr->ia_valid & ATTR_SIZE)
944 		return -xfs_setattr_size(XFS_I(dentry->d_inode), iattr, 0);
945 	return -xfs_setattr_nonsize(XFS_I(dentry->d_inode), iattr, 0);
946 }
947 
948 #define XFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
949 
950 /*
951  * Call fiemap helper to fill in user data.
952  * Returns positive errors to xfs_getbmap.
953  */
954 STATIC int
955 xfs_fiemap_format(
956 	void			**arg,
957 	struct getbmapx		*bmv,
958 	int			*full)
959 {
960 	int			error;
961 	struct fiemap_extent_info *fieinfo = *arg;
962 	u32			fiemap_flags = 0;
963 	u64			logical, physical, length;
964 
965 	/* Do nothing for a hole */
966 	if (bmv->bmv_block == -1LL)
967 		return 0;
968 
969 	logical = BBTOB(bmv->bmv_offset);
970 	physical = BBTOB(bmv->bmv_block);
971 	length = BBTOB(bmv->bmv_length);
972 
973 	if (bmv->bmv_oflags & BMV_OF_PREALLOC)
974 		fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
975 	else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
976 		fiemap_flags |= FIEMAP_EXTENT_DELALLOC;
977 		physical = 0;   /* no block yet */
978 	}
979 	if (bmv->bmv_oflags & BMV_OF_LAST)
980 		fiemap_flags |= FIEMAP_EXTENT_LAST;
981 
982 	error = fiemap_fill_next_extent(fieinfo, logical, physical,
983 					length, fiemap_flags);
984 	if (error > 0) {
985 		error = 0;
986 		*full = 1;	/* user array now full */
987 	}
988 
989 	return -error;
990 }
991 
992 STATIC int
993 xfs_vn_fiemap(
994 	struct inode		*inode,
995 	struct fiemap_extent_info *fieinfo,
996 	u64			start,
997 	u64			length)
998 {
999 	xfs_inode_t		*ip = XFS_I(inode);
1000 	struct getbmapx		bm;
1001 	int			error;
1002 
1003 	error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1004 	if (error)
1005 		return error;
1006 
1007 	/* Set up bmap header for xfs internal routine */
1008 	bm.bmv_offset = BTOBB(start);
1009 	/* Special case for whole file */
1010 	if (length == FIEMAP_MAX_OFFSET)
1011 		bm.bmv_length = -1LL;
1012 	else
1013 		bm.bmv_length = BTOBB(length);
1014 
1015 	/* We add one because in getbmap world count includes the header */
1016 	bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1017 					fieinfo->fi_extents_max + 1;
1018 	bm.bmv_count = min_t(__s32, bm.bmv_count,
1019 			     (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1020 	bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1021 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1022 		bm.bmv_iflags |= BMV_IF_ATTRFORK;
1023 	if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1024 		bm.bmv_iflags |= BMV_IF_DELALLOC;
1025 
1026 	error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1027 	if (error)
1028 		return -error;
1029 
1030 	return 0;
1031 }
1032 
1033 static const struct inode_operations xfs_inode_operations = {
1034 	.get_acl		= xfs_get_acl,
1035 	.getattr		= xfs_vn_getattr,
1036 	.setattr		= xfs_vn_setattr,
1037 	.setxattr		= generic_setxattr,
1038 	.getxattr		= generic_getxattr,
1039 	.removexattr		= generic_removexattr,
1040 	.listxattr		= xfs_vn_listxattr,
1041 	.fiemap			= xfs_vn_fiemap,
1042 };
1043 
1044 static const struct inode_operations xfs_dir_inode_operations = {
1045 	.create			= xfs_vn_create,
1046 	.lookup			= xfs_vn_lookup,
1047 	.link			= xfs_vn_link,
1048 	.unlink			= xfs_vn_unlink,
1049 	.symlink		= xfs_vn_symlink,
1050 	.mkdir			= xfs_vn_mkdir,
1051 	/*
1052 	 * Yes, XFS uses the same method for rmdir and unlink.
1053 	 *
1054 	 * There are some subtile differences deeper in the code,
1055 	 * but we use S_ISDIR to check for those.
1056 	 */
1057 	.rmdir			= xfs_vn_unlink,
1058 	.mknod			= xfs_vn_mknod,
1059 	.rename			= xfs_vn_rename,
1060 	.get_acl		= xfs_get_acl,
1061 	.getattr		= xfs_vn_getattr,
1062 	.setattr		= xfs_vn_setattr,
1063 	.setxattr		= generic_setxattr,
1064 	.getxattr		= generic_getxattr,
1065 	.removexattr		= generic_removexattr,
1066 	.listxattr		= xfs_vn_listxattr,
1067 };
1068 
1069 static const struct inode_operations xfs_dir_ci_inode_operations = {
1070 	.create			= xfs_vn_create,
1071 	.lookup			= xfs_vn_ci_lookup,
1072 	.link			= xfs_vn_link,
1073 	.unlink			= xfs_vn_unlink,
1074 	.symlink		= xfs_vn_symlink,
1075 	.mkdir			= xfs_vn_mkdir,
1076 	/*
1077 	 * Yes, XFS uses the same method for rmdir and unlink.
1078 	 *
1079 	 * There are some subtile differences deeper in the code,
1080 	 * but we use S_ISDIR to check for those.
1081 	 */
1082 	.rmdir			= xfs_vn_unlink,
1083 	.mknod			= xfs_vn_mknod,
1084 	.rename			= xfs_vn_rename,
1085 	.get_acl		= xfs_get_acl,
1086 	.getattr		= xfs_vn_getattr,
1087 	.setattr		= xfs_vn_setattr,
1088 	.setxattr		= generic_setxattr,
1089 	.getxattr		= generic_getxattr,
1090 	.removexattr		= generic_removexattr,
1091 	.listxattr		= xfs_vn_listxattr,
1092 };
1093 
1094 static const struct inode_operations xfs_symlink_inode_operations = {
1095 	.readlink		= generic_readlink,
1096 	.follow_link		= xfs_vn_follow_link,
1097 	.put_link		= xfs_vn_put_link,
1098 	.get_acl		= xfs_get_acl,
1099 	.getattr		= xfs_vn_getattr,
1100 	.setattr		= xfs_vn_setattr,
1101 	.setxattr		= generic_setxattr,
1102 	.getxattr		= generic_getxattr,
1103 	.removexattr		= generic_removexattr,
1104 	.listxattr		= xfs_vn_listxattr,
1105 };
1106 
1107 STATIC void
1108 xfs_diflags_to_iflags(
1109 	struct inode		*inode,
1110 	struct xfs_inode	*ip)
1111 {
1112 	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1113 		inode->i_flags |= S_IMMUTABLE;
1114 	else
1115 		inode->i_flags &= ~S_IMMUTABLE;
1116 	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1117 		inode->i_flags |= S_APPEND;
1118 	else
1119 		inode->i_flags &= ~S_APPEND;
1120 	if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1121 		inode->i_flags |= S_SYNC;
1122 	else
1123 		inode->i_flags &= ~S_SYNC;
1124 	if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1125 		inode->i_flags |= S_NOATIME;
1126 	else
1127 		inode->i_flags &= ~S_NOATIME;
1128 }
1129 
1130 /*
1131  * Initialize the Linux inode, set up the operation vectors and
1132  * unlock the inode.
1133  *
1134  * When reading existing inodes from disk this is called directly
1135  * from xfs_iget, when creating a new inode it is called from
1136  * xfs_ialloc after setting up the inode.
1137  *
1138  * We are always called with an uninitialised linux inode here.
1139  * We need to initialise the necessary fields and take a reference
1140  * on it.
1141  */
1142 void
1143 xfs_setup_inode(
1144 	struct xfs_inode	*ip)
1145 {
1146 	struct inode		*inode = &ip->i_vnode;
1147 
1148 	inode->i_ino = ip->i_ino;
1149 	inode->i_state = I_NEW;
1150 
1151 	inode_sb_list_add(inode);
1152 	/* make the inode look hashed for the writeback code */
1153 	hlist_add_fake(&inode->i_hash);
1154 
1155 	inode->i_mode	= ip->i_d.di_mode;
1156 	set_nlink(inode, ip->i_d.di_nlink);
1157 	inode->i_uid	= ip->i_d.di_uid;
1158 	inode->i_gid	= ip->i_d.di_gid;
1159 
1160 	switch (inode->i_mode & S_IFMT) {
1161 	case S_IFBLK:
1162 	case S_IFCHR:
1163 		inode->i_rdev =
1164 			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1165 			      sysv_minor(ip->i_df.if_u2.if_rdev));
1166 		break;
1167 	default:
1168 		inode->i_rdev = 0;
1169 		break;
1170 	}
1171 
1172 	inode->i_generation = ip->i_d.di_gen;
1173 	i_size_write(inode, ip->i_d.di_size);
1174 	inode->i_atime.tv_sec	= ip->i_d.di_atime.t_sec;
1175 	inode->i_atime.tv_nsec	= ip->i_d.di_atime.t_nsec;
1176 	inode->i_mtime.tv_sec	= ip->i_d.di_mtime.t_sec;
1177 	inode->i_mtime.tv_nsec	= ip->i_d.di_mtime.t_nsec;
1178 	inode->i_ctime.tv_sec	= ip->i_d.di_ctime.t_sec;
1179 	inode->i_ctime.tv_nsec	= ip->i_d.di_ctime.t_nsec;
1180 	xfs_diflags_to_iflags(inode, ip);
1181 
1182 	switch (inode->i_mode & S_IFMT) {
1183 	case S_IFREG:
1184 		inode->i_op = &xfs_inode_operations;
1185 		inode->i_fop = &xfs_file_operations;
1186 		inode->i_mapping->a_ops = &xfs_address_space_operations;
1187 		break;
1188 	case S_IFDIR:
1189 		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1190 			inode->i_op = &xfs_dir_ci_inode_operations;
1191 		else
1192 			inode->i_op = &xfs_dir_inode_operations;
1193 		inode->i_fop = &xfs_dir_file_operations;
1194 		break;
1195 	case S_IFLNK:
1196 		inode->i_op = &xfs_symlink_inode_operations;
1197 		if (!(ip->i_df.if_flags & XFS_IFINLINE))
1198 			inode->i_mapping->a_ops = &xfs_address_space_operations;
1199 		break;
1200 	default:
1201 		inode->i_op = &xfs_inode_operations;
1202 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1203 		break;
1204 	}
1205 
1206 	/*
1207 	 * If there is no attribute fork no ACL can exist on this inode,
1208 	 * and it can't have any file capabilities attached to it either.
1209 	 */
1210 	if (!XFS_IFORK_Q(ip)) {
1211 		inode_has_no_xattr(inode);
1212 		cache_no_acl(inode);
1213 	}
1214 
1215 	xfs_iflags_clear(ip, XFS_INEW);
1216 	barrier();
1217 
1218 	unlock_new_inode(inode);
1219 }
1220