xref: /openbmc/linux/fs/xfs/xfs_iops.c (revision 39b6f3aa)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_acl.h"
21 #include "xfs_log.h"
22 #include "xfs_trans.h"
23 #include "xfs_sb.h"
24 #include "xfs_ag.h"
25 #include "xfs_alloc.h"
26 #include "xfs_quota.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
31 #include "xfs_bmap.h"
32 #include "xfs_rtalloc.h"
33 #include "xfs_error.h"
34 #include "xfs_itable.h"
35 #include "xfs_attr.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_utils.h"
38 #include "xfs_vnodeops.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_trace.h"
41 #include "xfs_icache.h"
42 
43 #include <linux/capability.h>
44 #include <linux/xattr.h>
45 #include <linux/namei.h>
46 #include <linux/posix_acl.h>
47 #include <linux/security.h>
48 #include <linux/fiemap.h>
49 #include <linux/slab.h>
50 
51 static int
52 xfs_initxattrs(
53 	struct inode		*inode,
54 	const struct xattr	*xattr_array,
55 	void			*fs_info)
56 {
57 	const struct xattr	*xattr;
58 	struct xfs_inode	*ip = XFS_I(inode);
59 	int			error = 0;
60 
61 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
62 		error = xfs_attr_set(ip, xattr->name, xattr->value,
63 				     xattr->value_len, ATTR_SECURE);
64 		if (error < 0)
65 			break;
66 	}
67 	return error;
68 }
69 
70 /*
71  * Hook in SELinux.  This is not quite correct yet, what we really need
72  * here (as we do for default ACLs) is a mechanism by which creation of
73  * these attrs can be journalled at inode creation time (along with the
74  * inode, of course, such that log replay can't cause these to be lost).
75  */
76 
77 STATIC int
78 xfs_init_security(
79 	struct inode	*inode,
80 	struct inode	*dir,
81 	const struct qstr *qstr)
82 {
83 	return security_inode_init_security(inode, dir, qstr,
84 					    &xfs_initxattrs, NULL);
85 }
86 
87 static void
88 xfs_dentry_to_name(
89 	struct xfs_name	*namep,
90 	struct dentry	*dentry)
91 {
92 	namep->name = dentry->d_name.name;
93 	namep->len = dentry->d_name.len;
94 }
95 
96 STATIC void
97 xfs_cleanup_inode(
98 	struct inode	*dir,
99 	struct inode	*inode,
100 	struct dentry	*dentry)
101 {
102 	struct xfs_name	teardown;
103 
104 	/* Oh, the horror.
105 	 * If we can't add the ACL or we fail in
106 	 * xfs_init_security we must back out.
107 	 * ENOSPC can hit here, among other things.
108 	 */
109 	xfs_dentry_to_name(&teardown, dentry);
110 
111 	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
112 	iput(inode);
113 }
114 
115 STATIC int
116 xfs_vn_mknod(
117 	struct inode	*dir,
118 	struct dentry	*dentry,
119 	umode_t		mode,
120 	dev_t		rdev)
121 {
122 	struct inode	*inode;
123 	struct xfs_inode *ip = NULL;
124 	struct posix_acl *default_acl = NULL;
125 	struct xfs_name	name;
126 	int		error;
127 
128 	/*
129 	 * Irix uses Missed'em'V split, but doesn't want to see
130 	 * the upper 5 bits of (14bit) major.
131 	 */
132 	if (S_ISCHR(mode) || S_ISBLK(mode)) {
133 		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
134 			return -EINVAL;
135 		rdev = sysv_encode_dev(rdev);
136 	} else {
137 		rdev = 0;
138 	}
139 
140 	if (IS_POSIXACL(dir)) {
141 		default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
142 		if (IS_ERR(default_acl))
143 			return PTR_ERR(default_acl);
144 
145 		if (!default_acl)
146 			mode &= ~current_umask();
147 	}
148 
149 	xfs_dentry_to_name(&name, dentry);
150 	error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
151 	if (unlikely(error))
152 		goto out_free_acl;
153 
154 	inode = VFS_I(ip);
155 
156 	error = xfs_init_security(inode, dir, &dentry->d_name);
157 	if (unlikely(error))
158 		goto out_cleanup_inode;
159 
160 	if (default_acl) {
161 		error = -xfs_inherit_acl(inode, default_acl);
162 		default_acl = NULL;
163 		if (unlikely(error))
164 			goto out_cleanup_inode;
165 	}
166 
167 
168 	d_instantiate(dentry, inode);
169 	return -error;
170 
171  out_cleanup_inode:
172 	xfs_cleanup_inode(dir, inode, dentry);
173  out_free_acl:
174 	posix_acl_release(default_acl);
175 	return -error;
176 }
177 
178 STATIC int
179 xfs_vn_create(
180 	struct inode	*dir,
181 	struct dentry	*dentry,
182 	umode_t		mode,
183 	bool		flags)
184 {
185 	return xfs_vn_mknod(dir, dentry, mode, 0);
186 }
187 
188 STATIC int
189 xfs_vn_mkdir(
190 	struct inode	*dir,
191 	struct dentry	*dentry,
192 	umode_t		mode)
193 {
194 	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
195 }
196 
197 STATIC struct dentry *
198 xfs_vn_lookup(
199 	struct inode	*dir,
200 	struct dentry	*dentry,
201 	unsigned int flags)
202 {
203 	struct xfs_inode *cip;
204 	struct xfs_name	name;
205 	int		error;
206 
207 	if (dentry->d_name.len >= MAXNAMELEN)
208 		return ERR_PTR(-ENAMETOOLONG);
209 
210 	xfs_dentry_to_name(&name, dentry);
211 	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
212 	if (unlikely(error)) {
213 		if (unlikely(error != ENOENT))
214 			return ERR_PTR(-error);
215 		d_add(dentry, NULL);
216 		return NULL;
217 	}
218 
219 	return d_splice_alias(VFS_I(cip), dentry);
220 }
221 
222 STATIC struct dentry *
223 xfs_vn_ci_lookup(
224 	struct inode	*dir,
225 	struct dentry	*dentry,
226 	unsigned int flags)
227 {
228 	struct xfs_inode *ip;
229 	struct xfs_name	xname;
230 	struct xfs_name ci_name;
231 	struct qstr	dname;
232 	int		error;
233 
234 	if (dentry->d_name.len >= MAXNAMELEN)
235 		return ERR_PTR(-ENAMETOOLONG);
236 
237 	xfs_dentry_to_name(&xname, dentry);
238 	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
239 	if (unlikely(error)) {
240 		if (unlikely(error != ENOENT))
241 			return ERR_PTR(-error);
242 		/*
243 		 * call d_add(dentry, NULL) here when d_drop_negative_children
244 		 * is called in xfs_vn_mknod (ie. allow negative dentries
245 		 * with CI filesystems).
246 		 */
247 		return NULL;
248 	}
249 
250 	/* if exact match, just splice and exit */
251 	if (!ci_name.name)
252 		return d_splice_alias(VFS_I(ip), dentry);
253 
254 	/* else case-insensitive match... */
255 	dname.name = ci_name.name;
256 	dname.len = ci_name.len;
257 	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
258 	kmem_free(ci_name.name);
259 	return dentry;
260 }
261 
262 STATIC int
263 xfs_vn_link(
264 	struct dentry	*old_dentry,
265 	struct inode	*dir,
266 	struct dentry	*dentry)
267 {
268 	struct inode	*inode = old_dentry->d_inode;
269 	struct xfs_name	name;
270 	int		error;
271 
272 	xfs_dentry_to_name(&name, dentry);
273 
274 	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
275 	if (unlikely(error))
276 		return -error;
277 
278 	ihold(inode);
279 	d_instantiate(dentry, inode);
280 	return 0;
281 }
282 
283 STATIC int
284 xfs_vn_unlink(
285 	struct inode	*dir,
286 	struct dentry	*dentry)
287 {
288 	struct xfs_name	name;
289 	int		error;
290 
291 	xfs_dentry_to_name(&name, dentry);
292 
293 	error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
294 	if (error)
295 		return error;
296 
297 	/*
298 	 * With unlink, the VFS makes the dentry "negative": no inode,
299 	 * but still hashed. This is incompatible with case-insensitive
300 	 * mode, so invalidate (unhash) the dentry in CI-mode.
301 	 */
302 	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
303 		d_invalidate(dentry);
304 	return 0;
305 }
306 
307 STATIC int
308 xfs_vn_symlink(
309 	struct inode	*dir,
310 	struct dentry	*dentry,
311 	const char	*symname)
312 {
313 	struct inode	*inode;
314 	struct xfs_inode *cip = NULL;
315 	struct xfs_name	name;
316 	int		error;
317 	umode_t		mode;
318 
319 	mode = S_IFLNK |
320 		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
321 	xfs_dentry_to_name(&name, dentry);
322 
323 	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
324 	if (unlikely(error))
325 		goto out;
326 
327 	inode = VFS_I(cip);
328 
329 	error = xfs_init_security(inode, dir, &dentry->d_name);
330 	if (unlikely(error))
331 		goto out_cleanup_inode;
332 
333 	d_instantiate(dentry, inode);
334 	return 0;
335 
336  out_cleanup_inode:
337 	xfs_cleanup_inode(dir, inode, dentry);
338  out:
339 	return -error;
340 }
341 
342 STATIC int
343 xfs_vn_rename(
344 	struct inode	*odir,
345 	struct dentry	*odentry,
346 	struct inode	*ndir,
347 	struct dentry	*ndentry)
348 {
349 	struct inode	*new_inode = ndentry->d_inode;
350 	struct xfs_name	oname;
351 	struct xfs_name	nname;
352 
353 	xfs_dentry_to_name(&oname, odentry);
354 	xfs_dentry_to_name(&nname, ndentry);
355 
356 	return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
357 			   XFS_I(ndir), &nname, new_inode ?
358 			   			XFS_I(new_inode) : NULL);
359 }
360 
361 /*
362  * careful here - this function can get called recursively, so
363  * we need to be very careful about how much stack we use.
364  * uio is kmalloced for this reason...
365  */
366 STATIC void *
367 xfs_vn_follow_link(
368 	struct dentry		*dentry,
369 	struct nameidata	*nd)
370 {
371 	char			*link;
372 	int			error = -ENOMEM;
373 
374 	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
375 	if (!link)
376 		goto out_err;
377 
378 	error = -xfs_readlink(XFS_I(dentry->d_inode), link);
379 	if (unlikely(error))
380 		goto out_kfree;
381 
382 	nd_set_link(nd, link);
383 	return NULL;
384 
385  out_kfree:
386 	kfree(link);
387  out_err:
388 	nd_set_link(nd, ERR_PTR(error));
389 	return NULL;
390 }
391 
392 STATIC void
393 xfs_vn_put_link(
394 	struct dentry	*dentry,
395 	struct nameidata *nd,
396 	void		*p)
397 {
398 	char		*s = nd_get_link(nd);
399 
400 	if (!IS_ERR(s))
401 		kfree(s);
402 }
403 
404 STATIC int
405 xfs_vn_getattr(
406 	struct vfsmount		*mnt,
407 	struct dentry		*dentry,
408 	struct kstat		*stat)
409 {
410 	struct inode		*inode = dentry->d_inode;
411 	struct xfs_inode	*ip = XFS_I(inode);
412 	struct xfs_mount	*mp = ip->i_mount;
413 
414 	trace_xfs_getattr(ip);
415 
416 	if (XFS_FORCED_SHUTDOWN(mp))
417 		return -XFS_ERROR(EIO);
418 
419 	stat->size = XFS_ISIZE(ip);
420 	stat->dev = inode->i_sb->s_dev;
421 	stat->mode = ip->i_d.di_mode;
422 	stat->nlink = ip->i_d.di_nlink;
423 	stat->uid = ip->i_d.di_uid;
424 	stat->gid = ip->i_d.di_gid;
425 	stat->ino = ip->i_ino;
426 	stat->atime = inode->i_atime;
427 	stat->mtime = inode->i_mtime;
428 	stat->ctime = inode->i_ctime;
429 	stat->blocks =
430 		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
431 
432 
433 	switch (inode->i_mode & S_IFMT) {
434 	case S_IFBLK:
435 	case S_IFCHR:
436 		stat->blksize = BLKDEV_IOSIZE;
437 		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
438 				   sysv_minor(ip->i_df.if_u2.if_rdev));
439 		break;
440 	default:
441 		if (XFS_IS_REALTIME_INODE(ip)) {
442 			/*
443 			 * If the file blocks are being allocated from a
444 			 * realtime volume, then return the inode's realtime
445 			 * extent size or the realtime volume's extent size.
446 			 */
447 			stat->blksize =
448 				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
449 		} else
450 			stat->blksize = xfs_preferred_iosize(mp);
451 		stat->rdev = 0;
452 		break;
453 	}
454 
455 	return 0;
456 }
457 
458 static void
459 xfs_setattr_mode(
460 	struct xfs_trans	*tp,
461 	struct xfs_inode	*ip,
462 	struct iattr		*iattr)
463 {
464 	struct inode	*inode = VFS_I(ip);
465 	umode_t		mode = iattr->ia_mode;
466 
467 	ASSERT(tp);
468 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
469 
470 	if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
471 		mode &= ~S_ISGID;
472 
473 	ip->i_d.di_mode &= S_IFMT;
474 	ip->i_d.di_mode |= mode & ~S_IFMT;
475 
476 	inode->i_mode &= S_IFMT;
477 	inode->i_mode |= mode & ~S_IFMT;
478 }
479 
480 int
481 xfs_setattr_nonsize(
482 	struct xfs_inode	*ip,
483 	struct iattr		*iattr,
484 	int			flags)
485 {
486 	xfs_mount_t		*mp = ip->i_mount;
487 	struct inode		*inode = VFS_I(ip);
488 	int			mask = iattr->ia_valid;
489 	xfs_trans_t		*tp;
490 	int			error;
491 	uid_t			uid = 0, iuid = 0;
492 	gid_t			gid = 0, igid = 0;
493 	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
494 	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
495 
496 	trace_xfs_setattr(ip);
497 
498 	if (mp->m_flags & XFS_MOUNT_RDONLY)
499 		return XFS_ERROR(EROFS);
500 
501 	if (XFS_FORCED_SHUTDOWN(mp))
502 		return XFS_ERROR(EIO);
503 
504 	error = -inode_change_ok(inode, iattr);
505 	if (error)
506 		return XFS_ERROR(error);
507 
508 	ASSERT((mask & ATTR_SIZE) == 0);
509 
510 	/*
511 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
512 	 * before we start any other transactions. Trying to do this later
513 	 * is messy. We don't care to take a readlock to look at the ids
514 	 * in inode here, because we can't hold it across the trans_reserve.
515 	 * If the IDs do change before we take the ilock, we're covered
516 	 * because the i_*dquot fields will get updated anyway.
517 	 */
518 	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
519 		uint	qflags = 0;
520 
521 		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
522 			uid = iattr->ia_uid;
523 			qflags |= XFS_QMOPT_UQUOTA;
524 		} else {
525 			uid = ip->i_d.di_uid;
526 		}
527 		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
528 			gid = iattr->ia_gid;
529 			qflags |= XFS_QMOPT_GQUOTA;
530 		}  else {
531 			gid = ip->i_d.di_gid;
532 		}
533 
534 		/*
535 		 * We take a reference when we initialize udqp and gdqp,
536 		 * so it is important that we never blindly double trip on
537 		 * the same variable. See xfs_create() for an example.
538 		 */
539 		ASSERT(udqp == NULL);
540 		ASSERT(gdqp == NULL);
541 		error = xfs_qm_vop_dqalloc(ip, uid, gid, xfs_get_projid(ip),
542 					 qflags, &udqp, &gdqp);
543 		if (error)
544 			return error;
545 	}
546 
547 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
548 	error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0);
549 	if (error)
550 		goto out_dqrele;
551 
552 	xfs_ilock(ip, XFS_ILOCK_EXCL);
553 
554 	/*
555 	 * Change file ownership.  Must be the owner or privileged.
556 	 */
557 	if (mask & (ATTR_UID|ATTR_GID)) {
558 		/*
559 		 * These IDs could have changed since we last looked at them.
560 		 * But, we're assured that if the ownership did change
561 		 * while we didn't have the inode locked, inode's dquot(s)
562 		 * would have changed also.
563 		 */
564 		iuid = ip->i_d.di_uid;
565 		igid = ip->i_d.di_gid;
566 		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
567 		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
568 
569 		/*
570 		 * Do a quota reservation only if uid/gid is actually
571 		 * going to change.
572 		 */
573 		if (XFS_IS_QUOTA_RUNNING(mp) &&
574 		    ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
575 		     (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
576 			ASSERT(tp);
577 			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
578 						capable(CAP_FOWNER) ?
579 						XFS_QMOPT_FORCE_RES : 0);
580 			if (error)	/* out of quota */
581 				goto out_trans_cancel;
582 		}
583 	}
584 
585 	xfs_trans_ijoin(tp, ip, 0);
586 
587 	/*
588 	 * Change file ownership.  Must be the owner or privileged.
589 	 */
590 	if (mask & (ATTR_UID|ATTR_GID)) {
591 		/*
592 		 * CAP_FSETID overrides the following restrictions:
593 		 *
594 		 * The set-user-ID and set-group-ID bits of a file will be
595 		 * cleared upon successful return from chown()
596 		 */
597 		if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
598 		    !capable(CAP_FSETID))
599 			ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
600 
601 		/*
602 		 * Change the ownerships and register quota modifications
603 		 * in the transaction.
604 		 */
605 		if (iuid != uid) {
606 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
607 				ASSERT(mask & ATTR_UID);
608 				ASSERT(udqp);
609 				olddquot1 = xfs_qm_vop_chown(tp, ip,
610 							&ip->i_udquot, udqp);
611 			}
612 			ip->i_d.di_uid = uid;
613 			inode->i_uid = uid;
614 		}
615 		if (igid != gid) {
616 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
617 				ASSERT(!XFS_IS_PQUOTA_ON(mp));
618 				ASSERT(mask & ATTR_GID);
619 				ASSERT(gdqp);
620 				olddquot2 = xfs_qm_vop_chown(tp, ip,
621 							&ip->i_gdquot, gdqp);
622 			}
623 			ip->i_d.di_gid = gid;
624 			inode->i_gid = gid;
625 		}
626 	}
627 
628 	/*
629 	 * Change file access modes.
630 	 */
631 	if (mask & ATTR_MODE)
632 		xfs_setattr_mode(tp, ip, iattr);
633 
634 	/*
635 	 * Change file access or modified times.
636 	 */
637 	if (mask & ATTR_ATIME) {
638 		inode->i_atime = iattr->ia_atime;
639 		ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
640 		ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
641 	}
642 	if (mask & ATTR_CTIME) {
643 		inode->i_ctime = iattr->ia_ctime;
644 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
645 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
646 	}
647 	if (mask & ATTR_MTIME) {
648 		inode->i_mtime = iattr->ia_mtime;
649 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
650 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
651 	}
652 
653 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
654 
655 	XFS_STATS_INC(xs_ig_attrchg);
656 
657 	if (mp->m_flags & XFS_MOUNT_WSYNC)
658 		xfs_trans_set_sync(tp);
659 	error = xfs_trans_commit(tp, 0);
660 
661 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
662 
663 	/*
664 	 * Release any dquot(s) the inode had kept before chown.
665 	 */
666 	xfs_qm_dqrele(olddquot1);
667 	xfs_qm_dqrele(olddquot2);
668 	xfs_qm_dqrele(udqp);
669 	xfs_qm_dqrele(gdqp);
670 
671 	if (error)
672 		return XFS_ERROR(error);
673 
674 	/*
675 	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
676 	 * 	     update.  We could avoid this with linked transactions
677 	 * 	     and passing down the transaction pointer all the way
678 	 *	     to attr_set.  No previous user of the generic
679 	 * 	     Posix ACL code seems to care about this issue either.
680 	 */
681 	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
682 		error = -xfs_acl_chmod(inode);
683 		if (error)
684 			return XFS_ERROR(error);
685 	}
686 
687 	return 0;
688 
689 out_trans_cancel:
690 	xfs_trans_cancel(tp, 0);
691 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
692 out_dqrele:
693 	xfs_qm_dqrele(udqp);
694 	xfs_qm_dqrele(gdqp);
695 	return error;
696 }
697 
698 /*
699  * Truncate file.  Must have write permission and not be a directory.
700  */
701 int
702 xfs_setattr_size(
703 	struct xfs_inode	*ip,
704 	struct iattr		*iattr,
705 	int			flags)
706 {
707 	struct xfs_mount	*mp = ip->i_mount;
708 	struct inode		*inode = VFS_I(ip);
709 	int			mask = iattr->ia_valid;
710 	xfs_off_t		oldsize, newsize;
711 	struct xfs_trans	*tp;
712 	int			error;
713 	uint			lock_flags = 0;
714 	uint			commit_flags = 0;
715 
716 	trace_xfs_setattr(ip);
717 
718 	if (mp->m_flags & XFS_MOUNT_RDONLY)
719 		return XFS_ERROR(EROFS);
720 
721 	if (XFS_FORCED_SHUTDOWN(mp))
722 		return XFS_ERROR(EIO);
723 
724 	error = -inode_change_ok(inode, iattr);
725 	if (error)
726 		return XFS_ERROR(error);
727 
728 	ASSERT(S_ISREG(ip->i_d.di_mode));
729 	ASSERT((mask & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
730 			ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
731 
732 	if (!(flags & XFS_ATTR_NOLOCK)) {
733 		lock_flags |= XFS_IOLOCK_EXCL;
734 		xfs_ilock(ip, lock_flags);
735 	}
736 
737 	oldsize = inode->i_size;
738 	newsize = iattr->ia_size;
739 
740 	/*
741 	 * Short circuit the truncate case for zero length files.
742 	 */
743 	if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
744 		if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
745 			goto out_unlock;
746 
747 		/*
748 		 * Use the regular setattr path to update the timestamps.
749 		 */
750 		xfs_iunlock(ip, lock_flags);
751 		iattr->ia_valid &= ~ATTR_SIZE;
752 		return xfs_setattr_nonsize(ip, iattr, 0);
753 	}
754 
755 	/*
756 	 * Make sure that the dquots are attached to the inode.
757 	 */
758 	error = xfs_qm_dqattach(ip, 0);
759 	if (error)
760 		goto out_unlock;
761 
762 	/*
763 	 * Now we can make the changes.  Before we join the inode to the
764 	 * transaction, take care of the part of the truncation that must be
765 	 * done without the inode lock.  This needs to be done before joining
766 	 * the inode to the transaction, because the inode cannot be unlocked
767 	 * once it is a part of the transaction.
768 	 */
769 	if (newsize > oldsize) {
770 		/*
771 		 * Do the first part of growing a file: zero any data in the
772 		 * last block that is beyond the old EOF.  We need to do this
773 		 * before the inode is joined to the transaction to modify
774 		 * i_size.
775 		 */
776 		error = xfs_zero_eof(ip, newsize, oldsize);
777 		if (error)
778 			goto out_unlock;
779 	}
780 
781 	/*
782 	 * We are going to log the inode size change in this transaction so
783 	 * any previous writes that are beyond the on disk EOF and the new
784 	 * EOF that have not been written out need to be written here.  If we
785 	 * do not write the data out, we expose ourselves to the null files
786 	 * problem.
787 	 *
788 	 * Only flush from the on disk size to the smaller of the in memory
789 	 * file size or the new size as that's the range we really care about
790 	 * here and prevents waiting for other data not within the range we
791 	 * care about here.
792 	 */
793 	if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
794 		error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
795 						      ip->i_d.di_size, newsize);
796 		if (error)
797 			goto out_unlock;
798 	}
799 
800 	/*
801 	 * Wait for all direct I/O to complete.
802 	 */
803 	inode_dio_wait(inode);
804 
805 	error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
806 	if (error)
807 		goto out_unlock;
808 
809 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
810 	error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
811 				 XFS_TRANS_PERM_LOG_RES,
812 				 XFS_ITRUNCATE_LOG_COUNT);
813 	if (error)
814 		goto out_trans_cancel;
815 
816 	truncate_setsize(inode, newsize);
817 
818 	commit_flags = XFS_TRANS_RELEASE_LOG_RES;
819 	lock_flags |= XFS_ILOCK_EXCL;
820 
821 	xfs_ilock(ip, XFS_ILOCK_EXCL);
822 
823 	xfs_trans_ijoin(tp, ip, 0);
824 
825 	/*
826 	 * Only change the c/mtime if we are changing the size or we are
827 	 * explicitly asked to change it.  This handles the semantic difference
828 	 * between truncate() and ftruncate() as implemented in the VFS.
829 	 *
830 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
831 	 * special case where we need to update the times despite not having
832 	 * these flags set.  For all other operations the VFS set these flags
833 	 * explicitly if it wants a timestamp update.
834 	 */
835 	if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
836 		iattr->ia_ctime = iattr->ia_mtime =
837 			current_fs_time(inode->i_sb);
838 		mask |= ATTR_CTIME | ATTR_MTIME;
839 	}
840 
841 	/*
842 	 * The first thing we do is set the size to new_size permanently on
843 	 * disk.  This way we don't have to worry about anyone ever being able
844 	 * to look at the data being freed even in the face of a crash.
845 	 * What we're getting around here is the case where we free a block, it
846 	 * is allocated to another file, it is written to, and then we crash.
847 	 * If the new data gets written to the file but the log buffers
848 	 * containing the free and reallocation don't, then we'd end up with
849 	 * garbage in the blocks being freed.  As long as we make the new size
850 	 * permanent before actually freeing any blocks it doesn't matter if
851 	 * they get written to.
852 	 */
853 	ip->i_d.di_size = newsize;
854 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
855 
856 	if (newsize <= oldsize) {
857 		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
858 		if (error)
859 			goto out_trans_abort;
860 
861 		/*
862 		 * Truncated "down", so we're removing references to old data
863 		 * here - if we delay flushing for a long time, we expose
864 		 * ourselves unduly to the notorious NULL files problem.  So,
865 		 * we mark this inode and flush it when the file is closed,
866 		 * and do not wait the usual (long) time for writeout.
867 		 */
868 		xfs_iflags_set(ip, XFS_ITRUNCATED);
869 
870 		/* A truncate down always removes post-EOF blocks. */
871 		xfs_inode_clear_eofblocks_tag(ip);
872 	}
873 
874 	/*
875 	 * Change file access modes.
876 	 */
877 	if (mask & ATTR_MODE)
878 		xfs_setattr_mode(tp, ip, iattr);
879 
880 	if (mask & ATTR_CTIME) {
881 		inode->i_ctime = iattr->ia_ctime;
882 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
883 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
884 	}
885 	if (mask & ATTR_MTIME) {
886 		inode->i_mtime = iattr->ia_mtime;
887 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
888 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
889 	}
890 
891 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
892 
893 	XFS_STATS_INC(xs_ig_attrchg);
894 
895 	if (mp->m_flags & XFS_MOUNT_WSYNC)
896 		xfs_trans_set_sync(tp);
897 
898 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
899 out_unlock:
900 	if (lock_flags)
901 		xfs_iunlock(ip, lock_flags);
902 	return error;
903 
904 out_trans_abort:
905 	commit_flags |= XFS_TRANS_ABORT;
906 out_trans_cancel:
907 	xfs_trans_cancel(tp, commit_flags);
908 	goto out_unlock;
909 }
910 
911 STATIC int
912 xfs_vn_setattr(
913 	struct dentry	*dentry,
914 	struct iattr	*iattr)
915 {
916 	if (iattr->ia_valid & ATTR_SIZE)
917 		return -xfs_setattr_size(XFS_I(dentry->d_inode), iattr, 0);
918 	return -xfs_setattr_nonsize(XFS_I(dentry->d_inode), iattr, 0);
919 }
920 
921 STATIC int
922 xfs_vn_update_time(
923 	struct inode		*inode,
924 	struct timespec		*now,
925 	int			flags)
926 {
927 	struct xfs_inode	*ip = XFS_I(inode);
928 	struct xfs_mount	*mp = ip->i_mount;
929 	struct xfs_trans	*tp;
930 	int			error;
931 
932 	trace_xfs_update_time(ip);
933 
934 	tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
935 	error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
936 	if (error) {
937 		xfs_trans_cancel(tp, 0);
938 		return -error;
939 	}
940 
941 	xfs_ilock(ip, XFS_ILOCK_EXCL);
942 	if (flags & S_CTIME) {
943 		inode->i_ctime = *now;
944 		ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
945 		ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
946 	}
947 	if (flags & S_MTIME) {
948 		inode->i_mtime = *now;
949 		ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
950 		ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
951 	}
952 	if (flags & S_ATIME) {
953 		inode->i_atime = *now;
954 		ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
955 		ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
956 	}
957 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
958 	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
959 	return -xfs_trans_commit(tp, 0);
960 }
961 
962 #define XFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
963 
964 /*
965  * Call fiemap helper to fill in user data.
966  * Returns positive errors to xfs_getbmap.
967  */
968 STATIC int
969 xfs_fiemap_format(
970 	void			**arg,
971 	struct getbmapx		*bmv,
972 	int			*full)
973 {
974 	int			error;
975 	struct fiemap_extent_info *fieinfo = *arg;
976 	u32			fiemap_flags = 0;
977 	u64			logical, physical, length;
978 
979 	/* Do nothing for a hole */
980 	if (bmv->bmv_block == -1LL)
981 		return 0;
982 
983 	logical = BBTOB(bmv->bmv_offset);
984 	physical = BBTOB(bmv->bmv_block);
985 	length = BBTOB(bmv->bmv_length);
986 
987 	if (bmv->bmv_oflags & BMV_OF_PREALLOC)
988 		fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
989 	else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
990 		fiemap_flags |= FIEMAP_EXTENT_DELALLOC;
991 		physical = 0;   /* no block yet */
992 	}
993 	if (bmv->bmv_oflags & BMV_OF_LAST)
994 		fiemap_flags |= FIEMAP_EXTENT_LAST;
995 
996 	error = fiemap_fill_next_extent(fieinfo, logical, physical,
997 					length, fiemap_flags);
998 	if (error > 0) {
999 		error = 0;
1000 		*full = 1;	/* user array now full */
1001 	}
1002 
1003 	return -error;
1004 }
1005 
1006 STATIC int
1007 xfs_vn_fiemap(
1008 	struct inode		*inode,
1009 	struct fiemap_extent_info *fieinfo,
1010 	u64			start,
1011 	u64			length)
1012 {
1013 	xfs_inode_t		*ip = XFS_I(inode);
1014 	struct getbmapx		bm;
1015 	int			error;
1016 
1017 	error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1018 	if (error)
1019 		return error;
1020 
1021 	/* Set up bmap header for xfs internal routine */
1022 	bm.bmv_offset = BTOBB(start);
1023 	/* Special case for whole file */
1024 	if (length == FIEMAP_MAX_OFFSET)
1025 		bm.bmv_length = -1LL;
1026 	else
1027 		bm.bmv_length = BTOBB(length);
1028 
1029 	/* We add one because in getbmap world count includes the header */
1030 	bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1031 					fieinfo->fi_extents_max + 1;
1032 	bm.bmv_count = min_t(__s32, bm.bmv_count,
1033 			     (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1034 	bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1035 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1036 		bm.bmv_iflags |= BMV_IF_ATTRFORK;
1037 	if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1038 		bm.bmv_iflags |= BMV_IF_DELALLOC;
1039 
1040 	error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1041 	if (error)
1042 		return -error;
1043 
1044 	return 0;
1045 }
1046 
1047 static const struct inode_operations xfs_inode_operations = {
1048 	.get_acl		= xfs_get_acl,
1049 	.getattr		= xfs_vn_getattr,
1050 	.setattr		= xfs_vn_setattr,
1051 	.setxattr		= generic_setxattr,
1052 	.getxattr		= generic_getxattr,
1053 	.removexattr		= generic_removexattr,
1054 	.listxattr		= xfs_vn_listxattr,
1055 	.fiemap			= xfs_vn_fiemap,
1056 	.update_time		= xfs_vn_update_time,
1057 };
1058 
1059 static const struct inode_operations xfs_dir_inode_operations = {
1060 	.create			= xfs_vn_create,
1061 	.lookup			= xfs_vn_lookup,
1062 	.link			= xfs_vn_link,
1063 	.unlink			= xfs_vn_unlink,
1064 	.symlink		= xfs_vn_symlink,
1065 	.mkdir			= xfs_vn_mkdir,
1066 	/*
1067 	 * Yes, XFS uses the same method for rmdir and unlink.
1068 	 *
1069 	 * There are some subtile differences deeper in the code,
1070 	 * but we use S_ISDIR to check for those.
1071 	 */
1072 	.rmdir			= xfs_vn_unlink,
1073 	.mknod			= xfs_vn_mknod,
1074 	.rename			= xfs_vn_rename,
1075 	.get_acl		= xfs_get_acl,
1076 	.getattr		= xfs_vn_getattr,
1077 	.setattr		= xfs_vn_setattr,
1078 	.setxattr		= generic_setxattr,
1079 	.getxattr		= generic_getxattr,
1080 	.removexattr		= generic_removexattr,
1081 	.listxattr		= xfs_vn_listxattr,
1082 	.update_time		= xfs_vn_update_time,
1083 };
1084 
1085 static const struct inode_operations xfs_dir_ci_inode_operations = {
1086 	.create			= xfs_vn_create,
1087 	.lookup			= xfs_vn_ci_lookup,
1088 	.link			= xfs_vn_link,
1089 	.unlink			= xfs_vn_unlink,
1090 	.symlink		= xfs_vn_symlink,
1091 	.mkdir			= xfs_vn_mkdir,
1092 	/*
1093 	 * Yes, XFS uses the same method for rmdir and unlink.
1094 	 *
1095 	 * There are some subtile differences deeper in the code,
1096 	 * but we use S_ISDIR to check for those.
1097 	 */
1098 	.rmdir			= xfs_vn_unlink,
1099 	.mknod			= xfs_vn_mknod,
1100 	.rename			= xfs_vn_rename,
1101 	.get_acl		= xfs_get_acl,
1102 	.getattr		= xfs_vn_getattr,
1103 	.setattr		= xfs_vn_setattr,
1104 	.setxattr		= generic_setxattr,
1105 	.getxattr		= generic_getxattr,
1106 	.removexattr		= generic_removexattr,
1107 	.listxattr		= xfs_vn_listxattr,
1108 	.update_time		= xfs_vn_update_time,
1109 };
1110 
1111 static const struct inode_operations xfs_symlink_inode_operations = {
1112 	.readlink		= generic_readlink,
1113 	.follow_link		= xfs_vn_follow_link,
1114 	.put_link		= xfs_vn_put_link,
1115 	.get_acl		= xfs_get_acl,
1116 	.getattr		= xfs_vn_getattr,
1117 	.setattr		= xfs_vn_setattr,
1118 	.setxattr		= generic_setxattr,
1119 	.getxattr		= generic_getxattr,
1120 	.removexattr		= generic_removexattr,
1121 	.listxattr		= xfs_vn_listxattr,
1122 	.update_time		= xfs_vn_update_time,
1123 };
1124 
1125 STATIC void
1126 xfs_diflags_to_iflags(
1127 	struct inode		*inode,
1128 	struct xfs_inode	*ip)
1129 {
1130 	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1131 		inode->i_flags |= S_IMMUTABLE;
1132 	else
1133 		inode->i_flags &= ~S_IMMUTABLE;
1134 	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1135 		inode->i_flags |= S_APPEND;
1136 	else
1137 		inode->i_flags &= ~S_APPEND;
1138 	if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1139 		inode->i_flags |= S_SYNC;
1140 	else
1141 		inode->i_flags &= ~S_SYNC;
1142 	if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1143 		inode->i_flags |= S_NOATIME;
1144 	else
1145 		inode->i_flags &= ~S_NOATIME;
1146 }
1147 
1148 /*
1149  * Initialize the Linux inode, set up the operation vectors and
1150  * unlock the inode.
1151  *
1152  * When reading existing inodes from disk this is called directly
1153  * from xfs_iget, when creating a new inode it is called from
1154  * xfs_ialloc after setting up the inode.
1155  *
1156  * We are always called with an uninitialised linux inode here.
1157  * We need to initialise the necessary fields and take a reference
1158  * on it.
1159  */
1160 void
1161 xfs_setup_inode(
1162 	struct xfs_inode	*ip)
1163 {
1164 	struct inode		*inode = &ip->i_vnode;
1165 
1166 	inode->i_ino = ip->i_ino;
1167 	inode->i_state = I_NEW;
1168 
1169 	inode_sb_list_add(inode);
1170 	/* make the inode look hashed for the writeback code */
1171 	hlist_add_fake(&inode->i_hash);
1172 
1173 	inode->i_mode	= ip->i_d.di_mode;
1174 	set_nlink(inode, ip->i_d.di_nlink);
1175 	inode->i_uid	= ip->i_d.di_uid;
1176 	inode->i_gid	= ip->i_d.di_gid;
1177 
1178 	switch (inode->i_mode & S_IFMT) {
1179 	case S_IFBLK:
1180 	case S_IFCHR:
1181 		inode->i_rdev =
1182 			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1183 			      sysv_minor(ip->i_df.if_u2.if_rdev));
1184 		break;
1185 	default:
1186 		inode->i_rdev = 0;
1187 		break;
1188 	}
1189 
1190 	inode->i_generation = ip->i_d.di_gen;
1191 	i_size_write(inode, ip->i_d.di_size);
1192 	inode->i_atime.tv_sec	= ip->i_d.di_atime.t_sec;
1193 	inode->i_atime.tv_nsec	= ip->i_d.di_atime.t_nsec;
1194 	inode->i_mtime.tv_sec	= ip->i_d.di_mtime.t_sec;
1195 	inode->i_mtime.tv_nsec	= ip->i_d.di_mtime.t_nsec;
1196 	inode->i_ctime.tv_sec	= ip->i_d.di_ctime.t_sec;
1197 	inode->i_ctime.tv_nsec	= ip->i_d.di_ctime.t_nsec;
1198 	xfs_diflags_to_iflags(inode, ip);
1199 
1200 	switch (inode->i_mode & S_IFMT) {
1201 	case S_IFREG:
1202 		inode->i_op = &xfs_inode_operations;
1203 		inode->i_fop = &xfs_file_operations;
1204 		inode->i_mapping->a_ops = &xfs_address_space_operations;
1205 		break;
1206 	case S_IFDIR:
1207 		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1208 			inode->i_op = &xfs_dir_ci_inode_operations;
1209 		else
1210 			inode->i_op = &xfs_dir_inode_operations;
1211 		inode->i_fop = &xfs_dir_file_operations;
1212 		break;
1213 	case S_IFLNK:
1214 		inode->i_op = &xfs_symlink_inode_operations;
1215 		if (!(ip->i_df.if_flags & XFS_IFINLINE))
1216 			inode->i_mapping->a_ops = &xfs_address_space_operations;
1217 		break;
1218 	default:
1219 		inode->i_op = &xfs_inode_operations;
1220 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1221 		break;
1222 	}
1223 
1224 	/*
1225 	 * If there is no attribute fork no ACL can exist on this inode,
1226 	 * and it can't have any file capabilities attached to it either.
1227 	 */
1228 	if (!XFS_IFORK_Q(ip)) {
1229 		inode_has_no_xattr(inode);
1230 		cache_no_acl(inode);
1231 	}
1232 
1233 	xfs_iflags_clear(ip, XFS_INEW);
1234 	barrier();
1235 
1236 	unlock_new_inode(inode);
1237 }
1238