xref: /openbmc/linux/fs/xfs/xfs_iops.c (revision 930beb5a)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_inode.h"
29 #include "xfs_bmap.h"
30 #include "xfs_bmap_util.h"
31 #include "xfs_acl.h"
32 #include "xfs_quota.h"
33 #include "xfs_error.h"
34 #include "xfs_attr.h"
35 #include "xfs_trans.h"
36 #include "xfs_trace.h"
37 #include "xfs_icache.h"
38 #include "xfs_symlink.h"
39 #include "xfs_da_btree.h"
40 #include "xfs_dir2_priv.h"
41 #include "xfs_dinode.h"
42 
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 	int		mode)
92 {
93 	namep->name = dentry->d_name.name;
94 	namep->len = dentry->d_name.len;
95 	namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
96 }
97 
98 STATIC void
99 xfs_cleanup_inode(
100 	struct inode	*dir,
101 	struct inode	*inode,
102 	struct dentry	*dentry)
103 {
104 	struct xfs_name	teardown;
105 
106 	/* Oh, the horror.
107 	 * If we can't add the ACL or we fail in
108 	 * xfs_init_security we must back out.
109 	 * ENOSPC can hit here, among other things.
110 	 */
111 	xfs_dentry_to_name(&teardown, dentry, 0);
112 
113 	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
114 	iput(inode);
115 }
116 
117 STATIC int
118 xfs_vn_mknod(
119 	struct inode	*dir,
120 	struct dentry	*dentry,
121 	umode_t		mode,
122 	dev_t		rdev)
123 {
124 	struct inode	*inode;
125 	struct xfs_inode *ip = NULL;
126 	struct posix_acl *default_acl = NULL;
127 	struct xfs_name	name;
128 	int		error;
129 
130 	/*
131 	 * Irix uses Missed'em'V split, but doesn't want to see
132 	 * the upper 5 bits of (14bit) major.
133 	 */
134 	if (S_ISCHR(mode) || S_ISBLK(mode)) {
135 		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
136 			return -EINVAL;
137 		rdev = sysv_encode_dev(rdev);
138 	} else {
139 		rdev = 0;
140 	}
141 
142 	if (IS_POSIXACL(dir)) {
143 		default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
144 		if (IS_ERR(default_acl))
145 			return PTR_ERR(default_acl);
146 
147 		if (!default_acl)
148 			mode &= ~current_umask();
149 	}
150 
151 	xfs_dentry_to_name(&name, dentry, mode);
152 	error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
153 	if (unlikely(error))
154 		goto out_free_acl;
155 
156 	inode = VFS_I(ip);
157 
158 	error = xfs_init_security(inode, dir, &dentry->d_name);
159 	if (unlikely(error))
160 		goto out_cleanup_inode;
161 
162 	if (default_acl) {
163 		error = -xfs_inherit_acl(inode, default_acl);
164 		default_acl = NULL;
165 		if (unlikely(error))
166 			goto out_cleanup_inode;
167 	}
168 
169 
170 	d_instantiate(dentry, inode);
171 	return -error;
172 
173  out_cleanup_inode:
174 	xfs_cleanup_inode(dir, inode, dentry);
175  out_free_acl:
176 	posix_acl_release(default_acl);
177 	return -error;
178 }
179 
180 STATIC int
181 xfs_vn_create(
182 	struct inode	*dir,
183 	struct dentry	*dentry,
184 	umode_t		mode,
185 	bool		flags)
186 {
187 	return xfs_vn_mknod(dir, dentry, mode, 0);
188 }
189 
190 STATIC int
191 xfs_vn_mkdir(
192 	struct inode	*dir,
193 	struct dentry	*dentry,
194 	umode_t		mode)
195 {
196 	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
197 }
198 
199 STATIC struct dentry *
200 xfs_vn_lookup(
201 	struct inode	*dir,
202 	struct dentry	*dentry,
203 	unsigned int flags)
204 {
205 	struct xfs_inode *cip;
206 	struct xfs_name	name;
207 	int		error;
208 
209 	if (dentry->d_name.len >= MAXNAMELEN)
210 		return ERR_PTR(-ENAMETOOLONG);
211 
212 	xfs_dentry_to_name(&name, dentry, 0);
213 	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
214 	if (unlikely(error)) {
215 		if (unlikely(error != ENOENT))
216 			return ERR_PTR(-error);
217 		d_add(dentry, NULL);
218 		return NULL;
219 	}
220 
221 	return d_splice_alias(VFS_I(cip), dentry);
222 }
223 
224 STATIC struct dentry *
225 xfs_vn_ci_lookup(
226 	struct inode	*dir,
227 	struct dentry	*dentry,
228 	unsigned int flags)
229 {
230 	struct xfs_inode *ip;
231 	struct xfs_name	xname;
232 	struct xfs_name ci_name;
233 	struct qstr	dname;
234 	int		error;
235 
236 	if (dentry->d_name.len >= MAXNAMELEN)
237 		return ERR_PTR(-ENAMETOOLONG);
238 
239 	xfs_dentry_to_name(&xname, dentry, 0);
240 	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
241 	if (unlikely(error)) {
242 		if (unlikely(error != ENOENT))
243 			return ERR_PTR(-error);
244 		/*
245 		 * call d_add(dentry, NULL) here when d_drop_negative_children
246 		 * is called in xfs_vn_mknod (ie. allow negative dentries
247 		 * with CI filesystems).
248 		 */
249 		return NULL;
250 	}
251 
252 	/* if exact match, just splice and exit */
253 	if (!ci_name.name)
254 		return d_splice_alias(VFS_I(ip), dentry);
255 
256 	/* else case-insensitive match... */
257 	dname.name = ci_name.name;
258 	dname.len = ci_name.len;
259 	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
260 	kmem_free(ci_name.name);
261 	return dentry;
262 }
263 
264 STATIC int
265 xfs_vn_link(
266 	struct dentry	*old_dentry,
267 	struct inode	*dir,
268 	struct dentry	*dentry)
269 {
270 	struct inode	*inode = old_dentry->d_inode;
271 	struct xfs_name	name;
272 	int		error;
273 
274 	xfs_dentry_to_name(&name, dentry, inode->i_mode);
275 
276 	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
277 	if (unlikely(error))
278 		return -error;
279 
280 	ihold(inode);
281 	d_instantiate(dentry, inode);
282 	return 0;
283 }
284 
285 STATIC int
286 xfs_vn_unlink(
287 	struct inode	*dir,
288 	struct dentry	*dentry)
289 {
290 	struct xfs_name	name;
291 	int		error;
292 
293 	xfs_dentry_to_name(&name, dentry, 0);
294 
295 	error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
296 	if (error)
297 		return error;
298 
299 	/*
300 	 * With unlink, the VFS makes the dentry "negative": no inode,
301 	 * but still hashed. This is incompatible with case-insensitive
302 	 * mode, so invalidate (unhash) the dentry in CI-mode.
303 	 */
304 	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
305 		d_invalidate(dentry);
306 	return 0;
307 }
308 
309 STATIC int
310 xfs_vn_symlink(
311 	struct inode	*dir,
312 	struct dentry	*dentry,
313 	const char	*symname)
314 {
315 	struct inode	*inode;
316 	struct xfs_inode *cip = NULL;
317 	struct xfs_name	name;
318 	int		error;
319 	umode_t		mode;
320 
321 	mode = S_IFLNK |
322 		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
323 	xfs_dentry_to_name(&name, dentry, mode);
324 
325 	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
326 	if (unlikely(error))
327 		goto out;
328 
329 	inode = VFS_I(cip);
330 
331 	error = xfs_init_security(inode, dir, &dentry->d_name);
332 	if (unlikely(error))
333 		goto out_cleanup_inode;
334 
335 	d_instantiate(dentry, inode);
336 	return 0;
337 
338  out_cleanup_inode:
339 	xfs_cleanup_inode(dir, inode, dentry);
340  out:
341 	return -error;
342 }
343 
344 STATIC int
345 xfs_vn_rename(
346 	struct inode	*odir,
347 	struct dentry	*odentry,
348 	struct inode	*ndir,
349 	struct dentry	*ndentry)
350 {
351 	struct inode	*new_inode = ndentry->d_inode;
352 	struct xfs_name	oname;
353 	struct xfs_name	nname;
354 
355 	xfs_dentry_to_name(&oname, odentry, 0);
356 	xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
357 
358 	return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
359 			   XFS_I(ndir), &nname, new_inode ?
360 						XFS_I(new_inode) : NULL);
361 }
362 
363 /*
364  * careful here - this function can get called recursively, so
365  * we need to be very careful about how much stack we use.
366  * uio is kmalloced for this reason...
367  */
368 STATIC void *
369 xfs_vn_follow_link(
370 	struct dentry		*dentry,
371 	struct nameidata	*nd)
372 {
373 	char			*link;
374 	int			error = -ENOMEM;
375 
376 	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
377 	if (!link)
378 		goto out_err;
379 
380 	error = -xfs_readlink(XFS_I(dentry->d_inode), link);
381 	if (unlikely(error))
382 		goto out_kfree;
383 
384 	nd_set_link(nd, link);
385 	return NULL;
386 
387  out_kfree:
388 	kfree(link);
389  out_err:
390 	nd_set_link(nd, ERR_PTR(error));
391 	return NULL;
392 }
393 
394 STATIC void
395 xfs_vn_put_link(
396 	struct dentry	*dentry,
397 	struct nameidata *nd,
398 	void		*p)
399 {
400 	char		*s = nd_get_link(nd);
401 
402 	if (!IS_ERR(s))
403 		kfree(s);
404 }
405 
406 STATIC int
407 xfs_vn_getattr(
408 	struct vfsmount		*mnt,
409 	struct dentry		*dentry,
410 	struct kstat		*stat)
411 {
412 	struct inode		*inode = dentry->d_inode;
413 	struct xfs_inode	*ip = XFS_I(inode);
414 	struct xfs_mount	*mp = ip->i_mount;
415 
416 	trace_xfs_getattr(ip);
417 
418 	if (XFS_FORCED_SHUTDOWN(mp))
419 		return -XFS_ERROR(EIO);
420 
421 	stat->size = XFS_ISIZE(ip);
422 	stat->dev = inode->i_sb->s_dev;
423 	stat->mode = ip->i_d.di_mode;
424 	stat->nlink = ip->i_d.di_nlink;
425 	stat->uid = inode->i_uid;
426 	stat->gid = inode->i_gid;
427 	stat->ino = ip->i_ino;
428 	stat->atime = inode->i_atime;
429 	stat->mtime = inode->i_mtime;
430 	stat->ctime = inode->i_ctime;
431 	stat->blocks =
432 		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
433 
434 
435 	switch (inode->i_mode & S_IFMT) {
436 	case S_IFBLK:
437 	case S_IFCHR:
438 		stat->blksize = BLKDEV_IOSIZE;
439 		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
440 				   sysv_minor(ip->i_df.if_u2.if_rdev));
441 		break;
442 	default:
443 		if (XFS_IS_REALTIME_INODE(ip)) {
444 			/*
445 			 * If the file blocks are being allocated from a
446 			 * realtime volume, then return the inode's realtime
447 			 * extent size or the realtime volume's extent size.
448 			 */
449 			stat->blksize =
450 				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
451 		} else
452 			stat->blksize = xfs_preferred_iosize(mp);
453 		stat->rdev = 0;
454 		break;
455 	}
456 
457 	return 0;
458 }
459 
460 static void
461 xfs_setattr_mode(
462 	struct xfs_trans	*tp,
463 	struct xfs_inode	*ip,
464 	struct iattr		*iattr)
465 {
466 	struct inode	*inode = VFS_I(ip);
467 	umode_t		mode = iattr->ia_mode;
468 
469 	ASSERT(tp);
470 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
471 
472 	ip->i_d.di_mode &= S_IFMT;
473 	ip->i_d.di_mode |= mode & ~S_IFMT;
474 
475 	inode->i_mode &= S_IFMT;
476 	inode->i_mode |= mode & ~S_IFMT;
477 }
478 
479 int
480 xfs_setattr_nonsize(
481 	struct xfs_inode	*ip,
482 	struct iattr		*iattr,
483 	int			flags)
484 {
485 	xfs_mount_t		*mp = ip->i_mount;
486 	struct inode		*inode = VFS_I(ip);
487 	int			mask = iattr->ia_valid;
488 	xfs_trans_t		*tp;
489 	int			error;
490 	kuid_t			uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
491 	kgid_t			gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
492 	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
493 	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
494 
495 	trace_xfs_setattr(ip);
496 
497 	/* If acls are being inherited, we already have this checked */
498 	if (!(flags & XFS_ATTR_NOACL)) {
499 		if (mp->m_flags & XFS_MOUNT_RDONLY)
500 			return XFS_ERROR(EROFS);
501 
502 		if (XFS_FORCED_SHUTDOWN(mp))
503 			return XFS_ERROR(EIO);
504 
505 		error = -inode_change_ok(inode, iattr);
506 		if (error)
507 			return XFS_ERROR(error);
508 	}
509 
510 	ASSERT((mask & ATTR_SIZE) == 0);
511 
512 	/*
513 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
514 	 * before we start any other transactions. Trying to do this later
515 	 * is messy. We don't care to take a readlock to look at the ids
516 	 * in inode here, because we can't hold it across the trans_reserve.
517 	 * If the IDs do change before we take the ilock, we're covered
518 	 * because the i_*dquot fields will get updated anyway.
519 	 */
520 	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
521 		uint	qflags = 0;
522 
523 		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
524 			uid = iattr->ia_uid;
525 			qflags |= XFS_QMOPT_UQUOTA;
526 		} else {
527 			uid = inode->i_uid;
528 		}
529 		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
530 			gid = iattr->ia_gid;
531 			qflags |= XFS_QMOPT_GQUOTA;
532 		}  else {
533 			gid = inode->i_gid;
534 		}
535 
536 		/*
537 		 * We take a reference when we initialize udqp and gdqp,
538 		 * so it is important that we never blindly double trip on
539 		 * the same variable. See xfs_create() for an example.
540 		 */
541 		ASSERT(udqp == NULL);
542 		ASSERT(gdqp == NULL);
543 		error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
544 					   xfs_kgid_to_gid(gid),
545 					   xfs_get_projid(ip),
546 					   qflags, &udqp, &gdqp, NULL);
547 		if (error)
548 			return error;
549 	}
550 
551 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
552 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
553 	if (error)
554 		goto out_dqrele;
555 
556 	xfs_ilock(ip, XFS_ILOCK_EXCL);
557 
558 	/*
559 	 * Change file ownership.  Must be the owner or privileged.
560 	 */
561 	if (mask & (ATTR_UID|ATTR_GID)) {
562 		/*
563 		 * These IDs could have changed since we last looked at them.
564 		 * But, we're assured that if the ownership did change
565 		 * while we didn't have the inode locked, inode's dquot(s)
566 		 * would have changed also.
567 		 */
568 		iuid = inode->i_uid;
569 		igid = inode->i_gid;
570 		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
571 		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
572 
573 		/*
574 		 * Do a quota reservation only if uid/gid is actually
575 		 * going to change.
576 		 */
577 		if (XFS_IS_QUOTA_RUNNING(mp) &&
578 		    ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
579 		     (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
580 			ASSERT(tp);
581 			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
582 						NULL, capable(CAP_FOWNER) ?
583 						XFS_QMOPT_FORCE_RES : 0);
584 			if (error)	/* out of quota */
585 				goto out_trans_cancel;
586 		}
587 	}
588 
589 	xfs_trans_ijoin(tp, ip, 0);
590 
591 	/*
592 	 * Change file ownership.  Must be the owner or privileged.
593 	 */
594 	if (mask & (ATTR_UID|ATTR_GID)) {
595 		/*
596 		 * CAP_FSETID overrides the following restrictions:
597 		 *
598 		 * The set-user-ID and set-group-ID bits of a file will be
599 		 * cleared upon successful return from chown()
600 		 */
601 		if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
602 		    !capable(CAP_FSETID))
603 			ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
604 
605 		/*
606 		 * Change the ownerships and register quota modifications
607 		 * in the transaction.
608 		 */
609 		if (!uid_eq(iuid, uid)) {
610 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
611 				ASSERT(mask & ATTR_UID);
612 				ASSERT(udqp);
613 				olddquot1 = xfs_qm_vop_chown(tp, ip,
614 							&ip->i_udquot, udqp);
615 			}
616 			ip->i_d.di_uid = xfs_kuid_to_uid(uid);
617 			inode->i_uid = uid;
618 		}
619 		if (!gid_eq(igid, gid)) {
620 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
621 				ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
622 				       !XFS_IS_PQUOTA_ON(mp));
623 				ASSERT(mask & ATTR_GID);
624 				ASSERT(gdqp);
625 				olddquot2 = xfs_qm_vop_chown(tp, ip,
626 							&ip->i_gdquot, gdqp);
627 			}
628 			ip->i_d.di_gid = xfs_kgid_to_gid(gid);
629 			inode->i_gid = gid;
630 		}
631 	}
632 
633 	/*
634 	 * Change file access modes.
635 	 */
636 	if (mask & ATTR_MODE)
637 		xfs_setattr_mode(tp, ip, iattr);
638 
639 	/*
640 	 * Change file access or modified times.
641 	 */
642 	if (mask & ATTR_ATIME) {
643 		inode->i_atime = iattr->ia_atime;
644 		ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
645 		ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
646 	}
647 	if (mask & ATTR_CTIME) {
648 		inode->i_ctime = iattr->ia_ctime;
649 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
650 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
651 	}
652 	if (mask & ATTR_MTIME) {
653 		inode->i_mtime = iattr->ia_mtime;
654 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
655 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
656 	}
657 
658 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
659 
660 	XFS_STATS_INC(xs_ig_attrchg);
661 
662 	if (mp->m_flags & XFS_MOUNT_WSYNC)
663 		xfs_trans_set_sync(tp);
664 	error = xfs_trans_commit(tp, 0);
665 
666 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
667 
668 	/*
669 	 * Release any dquot(s) the inode had kept before chown.
670 	 */
671 	xfs_qm_dqrele(olddquot1);
672 	xfs_qm_dqrele(olddquot2);
673 	xfs_qm_dqrele(udqp);
674 	xfs_qm_dqrele(gdqp);
675 
676 	if (error)
677 		return XFS_ERROR(error);
678 
679 	/*
680 	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
681 	 * 	     update.  We could avoid this with linked transactions
682 	 * 	     and passing down the transaction pointer all the way
683 	 *	     to attr_set.  No previous user of the generic
684 	 * 	     Posix ACL code seems to care about this issue either.
685 	 */
686 	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
687 		error = -xfs_acl_chmod(inode);
688 		if (error)
689 			return XFS_ERROR(error);
690 	}
691 
692 	return 0;
693 
694 out_trans_cancel:
695 	xfs_trans_cancel(tp, 0);
696 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
697 out_dqrele:
698 	xfs_qm_dqrele(udqp);
699 	xfs_qm_dqrele(gdqp);
700 	return error;
701 }
702 
703 /*
704  * Truncate file.  Must have write permission and not be a directory.
705  */
706 int
707 xfs_setattr_size(
708 	struct xfs_inode	*ip,
709 	struct iattr		*iattr)
710 {
711 	struct xfs_mount	*mp = ip->i_mount;
712 	struct inode		*inode = VFS_I(ip);
713 	int			mask = iattr->ia_valid;
714 	xfs_off_t		oldsize, newsize;
715 	struct xfs_trans	*tp;
716 	int			error;
717 	uint			lock_flags = 0;
718 	uint			commit_flags = 0;
719 
720 	trace_xfs_setattr(ip);
721 
722 	if (mp->m_flags & XFS_MOUNT_RDONLY)
723 		return XFS_ERROR(EROFS);
724 
725 	if (XFS_FORCED_SHUTDOWN(mp))
726 		return XFS_ERROR(EIO);
727 
728 	error = -inode_change_ok(inode, iattr);
729 	if (error)
730 		return XFS_ERROR(error);
731 
732 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
733 	ASSERT(S_ISREG(ip->i_d.di_mode));
734 	ASSERT((mask & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
735 			ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
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 			return 0;
746 
747 		/*
748 		 * Use the regular setattr path to update the timestamps.
749 		 */
750 		iattr->ia_valid &= ~ATTR_SIZE;
751 		return xfs_setattr_nonsize(ip, iattr, 0);
752 	}
753 
754 	/*
755 	 * Make sure that the dquots are attached to the inode.
756 	 */
757 	error = xfs_qm_dqattach(ip, 0);
758 	if (error)
759 		return error;
760 
761 	/*
762 	 * Now we can make the changes.  Before we join the inode to the
763 	 * transaction, take care of the part of the truncation that must be
764 	 * done without the inode lock.  This needs to be done before joining
765 	 * the inode to the transaction, because the inode cannot be unlocked
766 	 * once it is a part of the transaction.
767 	 */
768 	if (newsize > oldsize) {
769 		/*
770 		 * Do the first part of growing a file: zero any data in the
771 		 * last block that is beyond the old EOF.  We need to do this
772 		 * before the inode is joined to the transaction to modify
773 		 * i_size.
774 		 */
775 		error = xfs_zero_eof(ip, newsize, oldsize);
776 		if (error)
777 			return error;
778 	}
779 
780 	/*
781 	 * We are going to log the inode size change in this transaction so
782 	 * any previous writes that are beyond the on disk EOF and the new
783 	 * EOF that have not been written out need to be written here.  If we
784 	 * do not write the data out, we expose ourselves to the null files
785 	 * problem.
786 	 *
787 	 * Only flush from the on disk size to the smaller of the in memory
788 	 * file size or the new size as that's the range we really care about
789 	 * here and prevents waiting for other data not within the range we
790 	 * care about here.
791 	 */
792 	if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
793 		error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
794 						      ip->i_d.di_size, newsize);
795 		if (error)
796 			return error;
797 	}
798 
799 	/*
800 	 * Wait for all direct I/O to complete.
801 	 */
802 	inode_dio_wait(inode);
803 
804 	error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
805 	if (error)
806 		return error;
807 
808 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
809 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
810 	if (error)
811 		goto out_trans_cancel;
812 
813 	truncate_setsize(inode, newsize);
814 
815 	commit_flags = XFS_TRANS_RELEASE_LOG_RES;
816 	lock_flags |= XFS_ILOCK_EXCL;
817 
818 	xfs_ilock(ip, XFS_ILOCK_EXCL);
819 
820 	xfs_trans_ijoin(tp, ip, 0);
821 
822 	/*
823 	 * Only change the c/mtime if we are changing the size or we are
824 	 * explicitly asked to change it.  This handles the semantic difference
825 	 * between truncate() and ftruncate() as implemented in the VFS.
826 	 *
827 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
828 	 * special case where we need to update the times despite not having
829 	 * these flags set.  For all other operations the VFS set these flags
830 	 * explicitly if it wants a timestamp update.
831 	 */
832 	if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
833 		iattr->ia_ctime = iattr->ia_mtime =
834 			current_fs_time(inode->i_sb);
835 		mask |= ATTR_CTIME | ATTR_MTIME;
836 	}
837 
838 	/*
839 	 * The first thing we do is set the size to new_size permanently on
840 	 * disk.  This way we don't have to worry about anyone ever being able
841 	 * to look at the data being freed even in the face of a crash.
842 	 * What we're getting around here is the case where we free a block, it
843 	 * is allocated to another file, it is written to, and then we crash.
844 	 * If the new data gets written to the file but the log buffers
845 	 * containing the free and reallocation don't, then we'd end up with
846 	 * garbage in the blocks being freed.  As long as we make the new size
847 	 * permanent before actually freeing any blocks it doesn't matter if
848 	 * they get written to.
849 	 */
850 	ip->i_d.di_size = newsize;
851 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
852 
853 	if (newsize <= oldsize) {
854 		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
855 		if (error)
856 			goto out_trans_abort;
857 
858 		/*
859 		 * Truncated "down", so we're removing references to old data
860 		 * here - if we delay flushing for a long time, we expose
861 		 * ourselves unduly to the notorious NULL files problem.  So,
862 		 * we mark this inode and flush it when the file is closed,
863 		 * and do not wait the usual (long) time for writeout.
864 		 */
865 		xfs_iflags_set(ip, XFS_ITRUNCATED);
866 
867 		/* A truncate down always removes post-EOF blocks. */
868 		xfs_inode_clear_eofblocks_tag(ip);
869 	}
870 
871 	/*
872 	 * Change file access modes.
873 	 */
874 	if (mask & ATTR_MODE)
875 		xfs_setattr_mode(tp, ip, iattr);
876 
877 	if (mask & ATTR_CTIME) {
878 		inode->i_ctime = iattr->ia_ctime;
879 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
880 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
881 	}
882 	if (mask & ATTR_MTIME) {
883 		inode->i_mtime = iattr->ia_mtime;
884 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
885 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
886 	}
887 
888 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
889 
890 	XFS_STATS_INC(xs_ig_attrchg);
891 
892 	if (mp->m_flags & XFS_MOUNT_WSYNC)
893 		xfs_trans_set_sync(tp);
894 
895 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
896 out_unlock:
897 	if (lock_flags)
898 		xfs_iunlock(ip, lock_flags);
899 	return error;
900 
901 out_trans_abort:
902 	commit_flags |= XFS_TRANS_ABORT;
903 out_trans_cancel:
904 	xfs_trans_cancel(tp, commit_flags);
905 	goto out_unlock;
906 }
907 
908 STATIC int
909 xfs_vn_setattr(
910 	struct dentry		*dentry,
911 	struct iattr		*iattr)
912 {
913 	struct xfs_inode	*ip = XFS_I(dentry->d_inode);
914 	int			error;
915 
916 	if (iattr->ia_valid & ATTR_SIZE) {
917 		xfs_ilock(ip, XFS_IOLOCK_EXCL);
918 		error = xfs_setattr_size(ip, iattr);
919 		xfs_iunlock(ip, XFS_IOLOCK_EXCL);
920 	} else {
921 		error = xfs_setattr_nonsize(ip, iattr, 0);
922 	}
923 
924 	return -error;
925 }
926 
927 STATIC int
928 xfs_vn_update_time(
929 	struct inode		*inode,
930 	struct timespec		*now,
931 	int			flags)
932 {
933 	struct xfs_inode	*ip = XFS_I(inode);
934 	struct xfs_mount	*mp = ip->i_mount;
935 	struct xfs_trans	*tp;
936 	int			error;
937 
938 	trace_xfs_update_time(ip);
939 
940 	tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
941 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
942 	if (error) {
943 		xfs_trans_cancel(tp, 0);
944 		return -error;
945 	}
946 
947 	xfs_ilock(ip, XFS_ILOCK_EXCL);
948 	if (flags & S_CTIME) {
949 		inode->i_ctime = *now;
950 		ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
951 		ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
952 	}
953 	if (flags & S_MTIME) {
954 		inode->i_mtime = *now;
955 		ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
956 		ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
957 	}
958 	if (flags & S_ATIME) {
959 		inode->i_atime = *now;
960 		ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
961 		ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
962 	}
963 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
964 	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
965 	return -xfs_trans_commit(tp, 0);
966 }
967 
968 #define XFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
969 
970 /*
971  * Call fiemap helper to fill in user data.
972  * Returns positive errors to xfs_getbmap.
973  */
974 STATIC int
975 xfs_fiemap_format(
976 	void			**arg,
977 	struct getbmapx		*bmv,
978 	int			*full)
979 {
980 	int			error;
981 	struct fiemap_extent_info *fieinfo = *arg;
982 	u32			fiemap_flags = 0;
983 	u64			logical, physical, length;
984 
985 	/* Do nothing for a hole */
986 	if (bmv->bmv_block == -1LL)
987 		return 0;
988 
989 	logical = BBTOB(bmv->bmv_offset);
990 	physical = BBTOB(bmv->bmv_block);
991 	length = BBTOB(bmv->bmv_length);
992 
993 	if (bmv->bmv_oflags & BMV_OF_PREALLOC)
994 		fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
995 	else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
996 		fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
997 				 FIEMAP_EXTENT_UNKNOWN);
998 		physical = 0;   /* no block yet */
999 	}
1000 	if (bmv->bmv_oflags & BMV_OF_LAST)
1001 		fiemap_flags |= FIEMAP_EXTENT_LAST;
1002 
1003 	error = fiemap_fill_next_extent(fieinfo, logical, physical,
1004 					length, fiemap_flags);
1005 	if (error > 0) {
1006 		error = 0;
1007 		*full = 1;	/* user array now full */
1008 	}
1009 
1010 	return -error;
1011 }
1012 
1013 STATIC int
1014 xfs_vn_fiemap(
1015 	struct inode		*inode,
1016 	struct fiemap_extent_info *fieinfo,
1017 	u64			start,
1018 	u64			length)
1019 {
1020 	xfs_inode_t		*ip = XFS_I(inode);
1021 	struct getbmapx		bm;
1022 	int			error;
1023 
1024 	error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1025 	if (error)
1026 		return error;
1027 
1028 	/* Set up bmap header for xfs internal routine */
1029 	bm.bmv_offset = BTOBB(start);
1030 	/* Special case for whole file */
1031 	if (length == FIEMAP_MAX_OFFSET)
1032 		bm.bmv_length = -1LL;
1033 	else
1034 		bm.bmv_length = BTOBB(length);
1035 
1036 	/* We add one because in getbmap world count includes the header */
1037 	bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1038 					fieinfo->fi_extents_max + 1;
1039 	bm.bmv_count = min_t(__s32, bm.bmv_count,
1040 			     (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1041 	bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1042 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1043 		bm.bmv_iflags |= BMV_IF_ATTRFORK;
1044 	if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1045 		bm.bmv_iflags |= BMV_IF_DELALLOC;
1046 
1047 	error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1048 	if (error)
1049 		return -error;
1050 
1051 	return 0;
1052 }
1053 
1054 static const struct inode_operations xfs_inode_operations = {
1055 	.get_acl		= xfs_get_acl,
1056 	.getattr		= xfs_vn_getattr,
1057 	.setattr		= xfs_vn_setattr,
1058 	.setxattr		= generic_setxattr,
1059 	.getxattr		= generic_getxattr,
1060 	.removexattr		= generic_removexattr,
1061 	.listxattr		= xfs_vn_listxattr,
1062 	.fiemap			= xfs_vn_fiemap,
1063 	.update_time		= xfs_vn_update_time,
1064 };
1065 
1066 static const struct inode_operations xfs_dir_inode_operations = {
1067 	.create			= xfs_vn_create,
1068 	.lookup			= xfs_vn_lookup,
1069 	.link			= xfs_vn_link,
1070 	.unlink			= xfs_vn_unlink,
1071 	.symlink		= xfs_vn_symlink,
1072 	.mkdir			= xfs_vn_mkdir,
1073 	/*
1074 	 * Yes, XFS uses the same method for rmdir and unlink.
1075 	 *
1076 	 * There are some subtile differences deeper in the code,
1077 	 * but we use S_ISDIR to check for those.
1078 	 */
1079 	.rmdir			= xfs_vn_unlink,
1080 	.mknod			= xfs_vn_mknod,
1081 	.rename			= xfs_vn_rename,
1082 	.get_acl		= xfs_get_acl,
1083 	.getattr		= xfs_vn_getattr,
1084 	.setattr		= xfs_vn_setattr,
1085 	.setxattr		= generic_setxattr,
1086 	.getxattr		= generic_getxattr,
1087 	.removexattr		= generic_removexattr,
1088 	.listxattr		= xfs_vn_listxattr,
1089 	.update_time		= xfs_vn_update_time,
1090 };
1091 
1092 static const struct inode_operations xfs_dir_ci_inode_operations = {
1093 	.create			= xfs_vn_create,
1094 	.lookup			= xfs_vn_ci_lookup,
1095 	.link			= xfs_vn_link,
1096 	.unlink			= xfs_vn_unlink,
1097 	.symlink		= xfs_vn_symlink,
1098 	.mkdir			= xfs_vn_mkdir,
1099 	/*
1100 	 * Yes, XFS uses the same method for rmdir and unlink.
1101 	 *
1102 	 * There are some subtile differences deeper in the code,
1103 	 * but we use S_ISDIR to check for those.
1104 	 */
1105 	.rmdir			= xfs_vn_unlink,
1106 	.mknod			= xfs_vn_mknod,
1107 	.rename			= xfs_vn_rename,
1108 	.get_acl		= xfs_get_acl,
1109 	.getattr		= xfs_vn_getattr,
1110 	.setattr		= xfs_vn_setattr,
1111 	.setxattr		= generic_setxattr,
1112 	.getxattr		= generic_getxattr,
1113 	.removexattr		= generic_removexattr,
1114 	.listxattr		= xfs_vn_listxattr,
1115 	.update_time		= xfs_vn_update_time,
1116 };
1117 
1118 static const struct inode_operations xfs_symlink_inode_operations = {
1119 	.readlink		= generic_readlink,
1120 	.follow_link		= xfs_vn_follow_link,
1121 	.put_link		= xfs_vn_put_link,
1122 	.get_acl		= xfs_get_acl,
1123 	.getattr		= xfs_vn_getattr,
1124 	.setattr		= xfs_vn_setattr,
1125 	.setxattr		= generic_setxattr,
1126 	.getxattr		= generic_getxattr,
1127 	.removexattr		= generic_removexattr,
1128 	.listxattr		= xfs_vn_listxattr,
1129 	.update_time		= xfs_vn_update_time,
1130 };
1131 
1132 STATIC void
1133 xfs_diflags_to_iflags(
1134 	struct inode		*inode,
1135 	struct xfs_inode	*ip)
1136 {
1137 	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1138 		inode->i_flags |= S_IMMUTABLE;
1139 	else
1140 		inode->i_flags &= ~S_IMMUTABLE;
1141 	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1142 		inode->i_flags |= S_APPEND;
1143 	else
1144 		inode->i_flags &= ~S_APPEND;
1145 	if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1146 		inode->i_flags |= S_SYNC;
1147 	else
1148 		inode->i_flags &= ~S_SYNC;
1149 	if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1150 		inode->i_flags |= S_NOATIME;
1151 	else
1152 		inode->i_flags &= ~S_NOATIME;
1153 }
1154 
1155 /*
1156  * Initialize the Linux inode, set up the operation vectors and
1157  * unlock the inode.
1158  *
1159  * When reading existing inodes from disk this is called directly
1160  * from xfs_iget, when creating a new inode it is called from
1161  * xfs_ialloc after setting up the inode.
1162  *
1163  * We are always called with an uninitialised linux inode here.
1164  * We need to initialise the necessary fields and take a reference
1165  * on it.
1166  */
1167 void
1168 xfs_setup_inode(
1169 	struct xfs_inode	*ip)
1170 {
1171 	struct inode		*inode = &ip->i_vnode;
1172 	gfp_t			gfp_mask;
1173 
1174 	inode->i_ino = ip->i_ino;
1175 	inode->i_state = I_NEW;
1176 
1177 	inode_sb_list_add(inode);
1178 	/* make the inode look hashed for the writeback code */
1179 	hlist_add_fake(&inode->i_hash);
1180 
1181 	inode->i_mode	= ip->i_d.di_mode;
1182 	set_nlink(inode, ip->i_d.di_nlink);
1183 	inode->i_uid    = xfs_uid_to_kuid(ip->i_d.di_uid);
1184 	inode->i_gid    = xfs_gid_to_kgid(ip->i_d.di_gid);
1185 
1186 	switch (inode->i_mode & S_IFMT) {
1187 	case S_IFBLK:
1188 	case S_IFCHR:
1189 		inode->i_rdev =
1190 			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1191 			      sysv_minor(ip->i_df.if_u2.if_rdev));
1192 		break;
1193 	default:
1194 		inode->i_rdev = 0;
1195 		break;
1196 	}
1197 
1198 	inode->i_generation = ip->i_d.di_gen;
1199 	i_size_write(inode, ip->i_d.di_size);
1200 	inode->i_atime.tv_sec	= ip->i_d.di_atime.t_sec;
1201 	inode->i_atime.tv_nsec	= ip->i_d.di_atime.t_nsec;
1202 	inode->i_mtime.tv_sec	= ip->i_d.di_mtime.t_sec;
1203 	inode->i_mtime.tv_nsec	= ip->i_d.di_mtime.t_nsec;
1204 	inode->i_ctime.tv_sec	= ip->i_d.di_ctime.t_sec;
1205 	inode->i_ctime.tv_nsec	= ip->i_d.di_ctime.t_nsec;
1206 	xfs_diflags_to_iflags(inode, ip);
1207 
1208 	ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1209 	switch (inode->i_mode & S_IFMT) {
1210 	case S_IFREG:
1211 		inode->i_op = &xfs_inode_operations;
1212 		inode->i_fop = &xfs_file_operations;
1213 		inode->i_mapping->a_ops = &xfs_address_space_operations;
1214 		break;
1215 	case S_IFDIR:
1216 		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1217 			inode->i_op = &xfs_dir_ci_inode_operations;
1218 		else
1219 			inode->i_op = &xfs_dir_inode_operations;
1220 		inode->i_fop = &xfs_dir_file_operations;
1221 		ip->d_ops = ip->i_mount->m_dir_inode_ops;
1222 		break;
1223 	case S_IFLNK:
1224 		inode->i_op = &xfs_symlink_inode_operations;
1225 		if (!(ip->i_df.if_flags & XFS_IFINLINE))
1226 			inode->i_mapping->a_ops = &xfs_address_space_operations;
1227 		break;
1228 	default:
1229 		inode->i_op = &xfs_inode_operations;
1230 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1231 		break;
1232 	}
1233 
1234 	/*
1235 	 * Ensure all page cache allocations are done from GFP_NOFS context to
1236 	 * prevent direct reclaim recursion back into the filesystem and blowing
1237 	 * stacks or deadlocking.
1238 	 */
1239 	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1240 	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1241 
1242 	/*
1243 	 * If there is no attribute fork no ACL can exist on this inode,
1244 	 * and it can't have any file capabilities attached to it either.
1245 	 */
1246 	if (!XFS_IFORK_Q(ip)) {
1247 		inode_has_no_xattr(inode);
1248 		cache_no_acl(inode);
1249 	}
1250 
1251 	xfs_iflags_clear(ip, XFS_INEW);
1252 	barrier();
1253 
1254 	unlock_new_inode(inode);
1255 }
1256