xref: /openbmc/linux/fs/xfs/xfs_iops.c (revision e1f7c9ee)
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 #include "xfs_trans_space.h"
43 
44 #include <linux/capability.h>
45 #include <linux/xattr.h>
46 #include <linux/namei.h>
47 #include <linux/posix_acl.h>
48 #include <linux/security.h>
49 #include <linux/fiemap.h>
50 #include <linux/slab.h>
51 
52 /*
53  * Directories have different lock order w.r.t. mmap_sem compared to regular
54  * files. This is due to readdir potentially triggering page faults on a user
55  * buffer inside filldir(), and this happens with the ilock on the directory
56  * held. For regular files, the lock order is the other way around - the
57  * mmap_sem is taken during the page fault, and then we lock the ilock to do
58  * block mapping. Hence we need a different class for the directory ilock so
59  * that lockdep can tell them apart.
60  */
61 static struct lock_class_key xfs_nondir_ilock_class;
62 static struct lock_class_key xfs_dir_ilock_class;
63 
64 static int
65 xfs_initxattrs(
66 	struct inode		*inode,
67 	const struct xattr	*xattr_array,
68 	void			*fs_info)
69 {
70 	const struct xattr	*xattr;
71 	struct xfs_inode	*ip = XFS_I(inode);
72 	int			error = 0;
73 
74 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
75 		error = xfs_attr_set(ip, xattr->name, xattr->value,
76 				      xattr->value_len, ATTR_SECURE);
77 		if (error < 0)
78 			break;
79 	}
80 	return error;
81 }
82 
83 /*
84  * Hook in SELinux.  This is not quite correct yet, what we really need
85  * here (as we do for default ACLs) is a mechanism by which creation of
86  * these attrs can be journalled at inode creation time (along with the
87  * inode, of course, such that log replay can't cause these to be lost).
88  */
89 
90 STATIC int
91 xfs_init_security(
92 	struct inode	*inode,
93 	struct inode	*dir,
94 	const struct qstr *qstr)
95 {
96 	return security_inode_init_security(inode, dir, qstr,
97 					     &xfs_initxattrs, NULL);
98 }
99 
100 static void
101 xfs_dentry_to_name(
102 	struct xfs_name	*namep,
103 	struct dentry	*dentry,
104 	int		mode)
105 {
106 	namep->name = dentry->d_name.name;
107 	namep->len = dentry->d_name.len;
108 	namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
109 }
110 
111 STATIC void
112 xfs_cleanup_inode(
113 	struct inode	*dir,
114 	struct inode	*inode,
115 	struct dentry	*dentry)
116 {
117 	struct xfs_name	teardown;
118 
119 	/* Oh, the horror.
120 	 * If we can't add the ACL or we fail in
121 	 * xfs_init_security we must back out.
122 	 * ENOSPC can hit here, among other things.
123 	 */
124 	xfs_dentry_to_name(&teardown, dentry, 0);
125 
126 	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
127 }
128 
129 STATIC int
130 xfs_generic_create(
131 	struct inode	*dir,
132 	struct dentry	*dentry,
133 	umode_t		mode,
134 	dev_t		rdev,
135 	bool		tmpfile)	/* unnamed file */
136 {
137 	struct inode	*inode;
138 	struct xfs_inode *ip = NULL;
139 	struct posix_acl *default_acl, *acl;
140 	struct xfs_name	name;
141 	int		error;
142 
143 	/*
144 	 * Irix uses Missed'em'V split, but doesn't want to see
145 	 * the upper 5 bits of (14bit) major.
146 	 */
147 	if (S_ISCHR(mode) || S_ISBLK(mode)) {
148 		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
149 			return -EINVAL;
150 		rdev = sysv_encode_dev(rdev);
151 	} else {
152 		rdev = 0;
153 	}
154 
155 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
156 	if (error)
157 		return error;
158 
159 	if (!tmpfile) {
160 		xfs_dentry_to_name(&name, dentry, mode);
161 		error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
162 	} else {
163 		error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
164 	}
165 	if (unlikely(error))
166 		goto out_free_acl;
167 
168 	inode = VFS_I(ip);
169 
170 	error = xfs_init_security(inode, dir, &dentry->d_name);
171 	if (unlikely(error))
172 		goto out_cleanup_inode;
173 
174 #ifdef CONFIG_XFS_POSIX_ACL
175 	if (default_acl) {
176 		error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
177 		if (error)
178 			goto out_cleanup_inode;
179 	}
180 	if (acl) {
181 		error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
182 		if (error)
183 			goto out_cleanup_inode;
184 	}
185 #endif
186 
187 	if (tmpfile)
188 		d_tmpfile(dentry, inode);
189 	else
190 		d_instantiate(dentry, inode);
191 
192  out_free_acl:
193 	if (default_acl)
194 		posix_acl_release(default_acl);
195 	if (acl)
196 		posix_acl_release(acl);
197 	return error;
198 
199  out_cleanup_inode:
200 	if (!tmpfile)
201 		xfs_cleanup_inode(dir, inode, dentry);
202 	iput(inode);
203 	goto out_free_acl;
204 }
205 
206 STATIC int
207 xfs_vn_mknod(
208 	struct inode	*dir,
209 	struct dentry	*dentry,
210 	umode_t		mode,
211 	dev_t		rdev)
212 {
213 	return xfs_generic_create(dir, dentry, mode, rdev, false);
214 }
215 
216 STATIC int
217 xfs_vn_create(
218 	struct inode	*dir,
219 	struct dentry	*dentry,
220 	umode_t		mode,
221 	bool		flags)
222 {
223 	return xfs_vn_mknod(dir, dentry, mode, 0);
224 }
225 
226 STATIC int
227 xfs_vn_mkdir(
228 	struct inode	*dir,
229 	struct dentry	*dentry,
230 	umode_t		mode)
231 {
232 	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
233 }
234 
235 STATIC struct dentry *
236 xfs_vn_lookup(
237 	struct inode	*dir,
238 	struct dentry	*dentry,
239 	unsigned int flags)
240 {
241 	struct xfs_inode *cip;
242 	struct xfs_name	name;
243 	int		error;
244 
245 	if (dentry->d_name.len >= MAXNAMELEN)
246 		return ERR_PTR(-ENAMETOOLONG);
247 
248 	xfs_dentry_to_name(&name, dentry, 0);
249 	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
250 	if (unlikely(error)) {
251 		if (unlikely(error != -ENOENT))
252 			return ERR_PTR(error);
253 		d_add(dentry, NULL);
254 		return NULL;
255 	}
256 
257 	return d_splice_alias(VFS_I(cip), dentry);
258 }
259 
260 STATIC struct dentry *
261 xfs_vn_ci_lookup(
262 	struct inode	*dir,
263 	struct dentry	*dentry,
264 	unsigned int flags)
265 {
266 	struct xfs_inode *ip;
267 	struct xfs_name	xname;
268 	struct xfs_name ci_name;
269 	struct qstr	dname;
270 	int		error;
271 
272 	if (dentry->d_name.len >= MAXNAMELEN)
273 		return ERR_PTR(-ENAMETOOLONG);
274 
275 	xfs_dentry_to_name(&xname, dentry, 0);
276 	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
277 	if (unlikely(error)) {
278 		if (unlikely(error != -ENOENT))
279 			return ERR_PTR(error);
280 		/*
281 		 * call d_add(dentry, NULL) here when d_drop_negative_children
282 		 * is called in xfs_vn_mknod (ie. allow negative dentries
283 		 * with CI filesystems).
284 		 */
285 		return NULL;
286 	}
287 
288 	/* if exact match, just splice and exit */
289 	if (!ci_name.name)
290 		return d_splice_alias(VFS_I(ip), dentry);
291 
292 	/* else case-insensitive match... */
293 	dname.name = ci_name.name;
294 	dname.len = ci_name.len;
295 	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
296 	kmem_free(ci_name.name);
297 	return dentry;
298 }
299 
300 STATIC int
301 xfs_vn_link(
302 	struct dentry	*old_dentry,
303 	struct inode	*dir,
304 	struct dentry	*dentry)
305 {
306 	struct inode	*inode = old_dentry->d_inode;
307 	struct xfs_name	name;
308 	int		error;
309 
310 	xfs_dentry_to_name(&name, dentry, inode->i_mode);
311 
312 	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
313 	if (unlikely(error))
314 		return error;
315 
316 	ihold(inode);
317 	d_instantiate(dentry, inode);
318 	return 0;
319 }
320 
321 STATIC int
322 xfs_vn_unlink(
323 	struct inode	*dir,
324 	struct dentry	*dentry)
325 {
326 	struct xfs_name	name;
327 	int		error;
328 
329 	xfs_dentry_to_name(&name, dentry, 0);
330 
331 	error = xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
332 	if (error)
333 		return error;
334 
335 	/*
336 	 * With unlink, the VFS makes the dentry "negative": no inode,
337 	 * but still hashed. This is incompatible with case-insensitive
338 	 * mode, so invalidate (unhash) the dentry in CI-mode.
339 	 */
340 	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
341 		d_invalidate(dentry);
342 	return 0;
343 }
344 
345 STATIC int
346 xfs_vn_symlink(
347 	struct inode	*dir,
348 	struct dentry	*dentry,
349 	const char	*symname)
350 {
351 	struct inode	*inode;
352 	struct xfs_inode *cip = NULL;
353 	struct xfs_name	name;
354 	int		error;
355 	umode_t		mode;
356 
357 	mode = S_IFLNK |
358 		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
359 	xfs_dentry_to_name(&name, dentry, mode);
360 
361 	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
362 	if (unlikely(error))
363 		goto out;
364 
365 	inode = VFS_I(cip);
366 
367 	error = xfs_init_security(inode, dir, &dentry->d_name);
368 	if (unlikely(error))
369 		goto out_cleanup_inode;
370 
371 	d_instantiate(dentry, inode);
372 	return 0;
373 
374  out_cleanup_inode:
375 	xfs_cleanup_inode(dir, inode, dentry);
376 	iput(inode);
377  out:
378 	return error;
379 }
380 
381 STATIC int
382 xfs_vn_rename(
383 	struct inode	*odir,
384 	struct dentry	*odentry,
385 	struct inode	*ndir,
386 	struct dentry	*ndentry)
387 {
388 	struct inode	*new_inode = ndentry->d_inode;
389 	struct xfs_name	oname;
390 	struct xfs_name	nname;
391 
392 	xfs_dentry_to_name(&oname, odentry, 0);
393 	xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
394 
395 	return xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
396 			  XFS_I(ndir), &nname, new_inode ?
397 						XFS_I(new_inode) : NULL);
398 }
399 
400 /*
401  * careful here - this function can get called recursively, so
402  * we need to be very careful about how much stack we use.
403  * uio is kmalloced for this reason...
404  */
405 STATIC void *
406 xfs_vn_follow_link(
407 	struct dentry		*dentry,
408 	struct nameidata	*nd)
409 {
410 	char			*link;
411 	int			error = -ENOMEM;
412 
413 	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
414 	if (!link)
415 		goto out_err;
416 
417 	error = xfs_readlink(XFS_I(dentry->d_inode), link);
418 	if (unlikely(error))
419 		goto out_kfree;
420 
421 	nd_set_link(nd, link);
422 	return NULL;
423 
424  out_kfree:
425 	kfree(link);
426  out_err:
427 	nd_set_link(nd, ERR_PTR(error));
428 	return NULL;
429 }
430 
431 STATIC int
432 xfs_vn_getattr(
433 	struct vfsmount		*mnt,
434 	struct dentry		*dentry,
435 	struct kstat		*stat)
436 {
437 	struct inode		*inode = dentry->d_inode;
438 	struct xfs_inode	*ip = XFS_I(inode);
439 	struct xfs_mount	*mp = ip->i_mount;
440 
441 	trace_xfs_getattr(ip);
442 
443 	if (XFS_FORCED_SHUTDOWN(mp))
444 		return -EIO;
445 
446 	stat->size = XFS_ISIZE(ip);
447 	stat->dev = inode->i_sb->s_dev;
448 	stat->mode = ip->i_d.di_mode;
449 	stat->nlink = ip->i_d.di_nlink;
450 	stat->uid = inode->i_uid;
451 	stat->gid = inode->i_gid;
452 	stat->ino = ip->i_ino;
453 	stat->atime = inode->i_atime;
454 	stat->mtime = inode->i_mtime;
455 	stat->ctime = inode->i_ctime;
456 	stat->blocks =
457 		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
458 
459 
460 	switch (inode->i_mode & S_IFMT) {
461 	case S_IFBLK:
462 	case S_IFCHR:
463 		stat->blksize = BLKDEV_IOSIZE;
464 		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
465 				   sysv_minor(ip->i_df.if_u2.if_rdev));
466 		break;
467 	default:
468 		if (XFS_IS_REALTIME_INODE(ip)) {
469 			/*
470 			 * If the file blocks are being allocated from a
471 			 * realtime volume, then return the inode's realtime
472 			 * extent size or the realtime volume's extent size.
473 			 */
474 			stat->blksize =
475 				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
476 		} else
477 			stat->blksize = xfs_preferred_iosize(mp);
478 		stat->rdev = 0;
479 		break;
480 	}
481 
482 	return 0;
483 }
484 
485 static void
486 xfs_setattr_mode(
487 	struct xfs_inode	*ip,
488 	struct iattr		*iattr)
489 {
490 	struct inode		*inode = VFS_I(ip);
491 	umode_t			mode = iattr->ia_mode;
492 
493 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
494 
495 	ip->i_d.di_mode &= S_IFMT;
496 	ip->i_d.di_mode |= mode & ~S_IFMT;
497 
498 	inode->i_mode &= S_IFMT;
499 	inode->i_mode |= mode & ~S_IFMT;
500 }
501 
502 static void
503 xfs_setattr_time(
504 	struct xfs_inode	*ip,
505 	struct iattr		*iattr)
506 {
507 	struct inode		*inode = VFS_I(ip);
508 
509 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
510 
511 	if (iattr->ia_valid & ATTR_ATIME) {
512 		inode->i_atime = iattr->ia_atime;
513 		ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
514 		ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
515 	}
516 	if (iattr->ia_valid & ATTR_CTIME) {
517 		inode->i_ctime = iattr->ia_ctime;
518 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
519 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
520 	}
521 	if (iattr->ia_valid & ATTR_MTIME) {
522 		inode->i_mtime = iattr->ia_mtime;
523 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
524 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
525 	}
526 }
527 
528 int
529 xfs_setattr_nonsize(
530 	struct xfs_inode	*ip,
531 	struct iattr		*iattr,
532 	int			flags)
533 {
534 	xfs_mount_t		*mp = ip->i_mount;
535 	struct inode		*inode = VFS_I(ip);
536 	int			mask = iattr->ia_valid;
537 	xfs_trans_t		*tp;
538 	int			error;
539 	kuid_t			uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
540 	kgid_t			gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
541 	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
542 	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
543 
544 	trace_xfs_setattr(ip);
545 
546 	/* If acls are being inherited, we already have this checked */
547 	if (!(flags & XFS_ATTR_NOACL)) {
548 		if (mp->m_flags & XFS_MOUNT_RDONLY)
549 			return -EROFS;
550 
551 		if (XFS_FORCED_SHUTDOWN(mp))
552 			return -EIO;
553 
554 		error = inode_change_ok(inode, iattr);
555 		if (error)
556 			return error;
557 	}
558 
559 	ASSERT((mask & ATTR_SIZE) == 0);
560 
561 	/*
562 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
563 	 * before we start any other transactions. Trying to do this later
564 	 * is messy. We don't care to take a readlock to look at the ids
565 	 * in inode here, because we can't hold it across the trans_reserve.
566 	 * If the IDs do change before we take the ilock, we're covered
567 	 * because the i_*dquot fields will get updated anyway.
568 	 */
569 	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
570 		uint	qflags = 0;
571 
572 		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
573 			uid = iattr->ia_uid;
574 			qflags |= XFS_QMOPT_UQUOTA;
575 		} else {
576 			uid = inode->i_uid;
577 		}
578 		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
579 			gid = iattr->ia_gid;
580 			qflags |= XFS_QMOPT_GQUOTA;
581 		}  else {
582 			gid = inode->i_gid;
583 		}
584 
585 		/*
586 		 * We take a reference when we initialize udqp and gdqp,
587 		 * so it is important that we never blindly double trip on
588 		 * the same variable. See xfs_create() for an example.
589 		 */
590 		ASSERT(udqp == NULL);
591 		ASSERT(gdqp == NULL);
592 		error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
593 					   xfs_kgid_to_gid(gid),
594 					   xfs_get_projid(ip),
595 					   qflags, &udqp, &gdqp, NULL);
596 		if (error)
597 			return error;
598 	}
599 
600 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
601 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
602 	if (error)
603 		goto out_dqrele;
604 
605 	xfs_ilock(ip, XFS_ILOCK_EXCL);
606 
607 	/*
608 	 * Change file ownership.  Must be the owner or privileged.
609 	 */
610 	if (mask & (ATTR_UID|ATTR_GID)) {
611 		/*
612 		 * These IDs could have changed since we last looked at them.
613 		 * But, we're assured that if the ownership did change
614 		 * while we didn't have the inode locked, inode's dquot(s)
615 		 * would have changed also.
616 		 */
617 		iuid = inode->i_uid;
618 		igid = inode->i_gid;
619 		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
620 		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
621 
622 		/*
623 		 * Do a quota reservation only if uid/gid is actually
624 		 * going to change.
625 		 */
626 		if (XFS_IS_QUOTA_RUNNING(mp) &&
627 		    ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
628 		     (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
629 			ASSERT(tp);
630 			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
631 						NULL, capable(CAP_FOWNER) ?
632 						XFS_QMOPT_FORCE_RES : 0);
633 			if (error)	/* out of quota */
634 				goto out_trans_cancel;
635 		}
636 	}
637 
638 	xfs_trans_ijoin(tp, ip, 0);
639 
640 	/*
641 	 * Change file ownership.  Must be the owner or privileged.
642 	 */
643 	if (mask & (ATTR_UID|ATTR_GID)) {
644 		/*
645 		 * CAP_FSETID overrides the following restrictions:
646 		 *
647 		 * The set-user-ID and set-group-ID bits of a file will be
648 		 * cleared upon successful return from chown()
649 		 */
650 		if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
651 		    !capable(CAP_FSETID))
652 			ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
653 
654 		/*
655 		 * Change the ownerships and register quota modifications
656 		 * in the transaction.
657 		 */
658 		if (!uid_eq(iuid, uid)) {
659 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
660 				ASSERT(mask & ATTR_UID);
661 				ASSERT(udqp);
662 				olddquot1 = xfs_qm_vop_chown(tp, ip,
663 							&ip->i_udquot, udqp);
664 			}
665 			ip->i_d.di_uid = xfs_kuid_to_uid(uid);
666 			inode->i_uid = uid;
667 		}
668 		if (!gid_eq(igid, gid)) {
669 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
670 				ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
671 				       !XFS_IS_PQUOTA_ON(mp));
672 				ASSERT(mask & ATTR_GID);
673 				ASSERT(gdqp);
674 				olddquot2 = xfs_qm_vop_chown(tp, ip,
675 							&ip->i_gdquot, gdqp);
676 			}
677 			ip->i_d.di_gid = xfs_kgid_to_gid(gid);
678 			inode->i_gid = gid;
679 		}
680 	}
681 
682 	if (mask & ATTR_MODE)
683 		xfs_setattr_mode(ip, iattr);
684 	if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
685 		xfs_setattr_time(ip, iattr);
686 
687 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
688 
689 	XFS_STATS_INC(xs_ig_attrchg);
690 
691 	if (mp->m_flags & XFS_MOUNT_WSYNC)
692 		xfs_trans_set_sync(tp);
693 	error = xfs_trans_commit(tp, 0);
694 
695 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
696 
697 	/*
698 	 * Release any dquot(s) the inode had kept before chown.
699 	 */
700 	xfs_qm_dqrele(olddquot1);
701 	xfs_qm_dqrele(olddquot2);
702 	xfs_qm_dqrele(udqp);
703 	xfs_qm_dqrele(gdqp);
704 
705 	if (error)
706 		return error;
707 
708 	/*
709 	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
710 	 * 	     update.  We could avoid this with linked transactions
711 	 * 	     and passing down the transaction pointer all the way
712 	 *	     to attr_set.  No previous user of the generic
713 	 * 	     Posix ACL code seems to care about this issue either.
714 	 */
715 	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
716 		error = posix_acl_chmod(inode, inode->i_mode);
717 		if (error)
718 			return error;
719 	}
720 
721 	return 0;
722 
723 out_trans_cancel:
724 	xfs_trans_cancel(tp, 0);
725 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
726 out_dqrele:
727 	xfs_qm_dqrele(udqp);
728 	xfs_qm_dqrele(gdqp);
729 	return error;
730 }
731 
732 /*
733  * Truncate file.  Must have write permission and not be a directory.
734  */
735 int
736 xfs_setattr_size(
737 	struct xfs_inode	*ip,
738 	struct iattr		*iattr)
739 {
740 	struct xfs_mount	*mp = ip->i_mount;
741 	struct inode		*inode = VFS_I(ip);
742 	xfs_off_t		oldsize, newsize;
743 	struct xfs_trans	*tp;
744 	int			error;
745 	uint			lock_flags = 0;
746 	uint			commit_flags = 0;
747 
748 	trace_xfs_setattr(ip);
749 
750 	if (mp->m_flags & XFS_MOUNT_RDONLY)
751 		return -EROFS;
752 
753 	if (XFS_FORCED_SHUTDOWN(mp))
754 		return -EIO;
755 
756 	error = inode_change_ok(inode, iattr);
757 	if (error)
758 		return error;
759 
760 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
761 	ASSERT(S_ISREG(ip->i_d.di_mode));
762 	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
763 		ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
764 
765 	oldsize = inode->i_size;
766 	newsize = iattr->ia_size;
767 
768 	/*
769 	 * Short circuit the truncate case for zero length files.
770 	 */
771 	if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
772 		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
773 			return 0;
774 
775 		/*
776 		 * Use the regular setattr path to update the timestamps.
777 		 */
778 		iattr->ia_valid &= ~ATTR_SIZE;
779 		return xfs_setattr_nonsize(ip, iattr, 0);
780 	}
781 
782 	/*
783 	 * Make sure that the dquots are attached to the inode.
784 	 */
785 	error = xfs_qm_dqattach(ip, 0);
786 	if (error)
787 		return error;
788 
789 	/*
790 	 * Now we can make the changes.  Before we join the inode to the
791 	 * transaction, take care of the part of the truncation that must be
792 	 * done without the inode lock.  This needs to be done before joining
793 	 * the inode to the transaction, because the inode cannot be unlocked
794 	 * once it is a part of the transaction.
795 	 */
796 	if (newsize > oldsize) {
797 		/*
798 		 * Do the first part of growing a file: zero any data in the
799 		 * last block that is beyond the old EOF.  We need to do this
800 		 * before the inode is joined to the transaction to modify
801 		 * i_size.
802 		 */
803 		error = xfs_zero_eof(ip, newsize, oldsize);
804 		if (error)
805 			return error;
806 	}
807 
808 	/*
809 	 * We are going to log the inode size change in this transaction so
810 	 * any previous writes that are beyond the on disk EOF and the new
811 	 * EOF that have not been written out need to be written here.  If we
812 	 * do not write the data out, we expose ourselves to the null files
813 	 * problem.
814 	 *
815 	 * Only flush from the on disk size to the smaller of the in memory
816 	 * file size or the new size as that's the range we really care about
817 	 * here and prevents waiting for other data not within the range we
818 	 * care about here.
819 	 */
820 	if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
821 		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
822 						      ip->i_d.di_size, newsize);
823 		if (error)
824 			return error;
825 	}
826 
827 	/*
828 	 * Wait for all direct I/O to complete.
829 	 */
830 	inode_dio_wait(inode);
831 
832 	/*
833 	 * Do all the page cache truncate work outside the transaction context
834 	 * as the "lock" order is page lock->log space reservation.  i.e.
835 	 * locking pages inside the transaction can ABBA deadlock with
836 	 * writeback. We have to do the VFS inode size update before we truncate
837 	 * the pagecache, however, to avoid racing with page faults beyond the
838 	 * new EOF they are not serialised against truncate operations except by
839 	 * page locks and size updates.
840 	 *
841 	 * Hence we are in a situation where a truncate can fail with ENOMEM
842 	 * from xfs_trans_reserve(), but having already truncated the in-memory
843 	 * version of the file (i.e. made user visible changes). There's not
844 	 * much we can do about this, except to hope that the caller sees ENOMEM
845 	 * and retries the truncate operation.
846 	 */
847 	error = block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
848 	if (error)
849 		return error;
850 	truncate_setsize(inode, newsize);
851 
852 	/*
853 	 * The "we can't serialise against page faults" pain gets worse.
854 	 *
855 	 * If the file is mapped then we have to clean the page at the old EOF
856 	 * when extending the file. Extending the file can expose changes the
857 	 * underlying page mapping (e.g. from beyond EOF to a hole or
858 	 * unwritten), and so on the next attempt to write to that page we need
859 	 * to remap it for write. i.e. we need .page_mkwrite() to be called.
860 	 * Hence we need to clean the page to clean the pte and so a new write
861 	 * fault will be triggered appropriately.
862 	 *
863 	 * If we do it before we change the inode size, then we can race with a
864 	 * page fault that maps the page with exactly the same problem. If we do
865 	 * it after we change the file size, then a new page fault can come in
866 	 * and allocate space before we've run the rest of the truncate
867 	 * transaction. That's kinda grotesque, but it's better than have data
868 	 * over a hole, and so that's the lesser evil that has been chosen here.
869 	 *
870 	 * The real solution, however, is to have some mechanism for locking out
871 	 * page faults while a truncate is in progress.
872 	 */
873 	if (newsize > oldsize && mapping_mapped(VFS_I(ip)->i_mapping)) {
874 		error = filemap_write_and_wait_range(
875 				VFS_I(ip)->i_mapping,
876 				round_down(oldsize, PAGE_CACHE_SIZE),
877 				round_up(oldsize, PAGE_CACHE_SIZE) - 1);
878 		if (error)
879 			return error;
880 	}
881 
882 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
883 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
884 	if (error)
885 		goto out_trans_cancel;
886 
887 	commit_flags = XFS_TRANS_RELEASE_LOG_RES;
888 	lock_flags |= XFS_ILOCK_EXCL;
889 	xfs_ilock(ip, XFS_ILOCK_EXCL);
890 	xfs_trans_ijoin(tp, ip, 0);
891 
892 	/*
893 	 * Only change the c/mtime if we are changing the size or we are
894 	 * explicitly asked to change it.  This handles the semantic difference
895 	 * between truncate() and ftruncate() as implemented in the VFS.
896 	 *
897 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
898 	 * special case where we need to update the times despite not having
899 	 * these flags set.  For all other operations the VFS set these flags
900 	 * explicitly if it wants a timestamp update.
901 	 */
902 	if (newsize != oldsize &&
903 	    !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
904 		iattr->ia_ctime = iattr->ia_mtime =
905 			current_fs_time(inode->i_sb);
906 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
907 	}
908 
909 	/*
910 	 * The first thing we do is set the size to new_size permanently on
911 	 * disk.  This way we don't have to worry about anyone ever being able
912 	 * to look at the data being freed even in the face of a crash.
913 	 * What we're getting around here is the case where we free a block, it
914 	 * is allocated to another file, it is written to, and then we crash.
915 	 * If the new data gets written to the file but the log buffers
916 	 * containing the free and reallocation don't, then we'd end up with
917 	 * garbage in the blocks being freed.  As long as we make the new size
918 	 * permanent before actually freeing any blocks it doesn't matter if
919 	 * they get written to.
920 	 */
921 	ip->i_d.di_size = newsize;
922 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
923 
924 	if (newsize <= oldsize) {
925 		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
926 		if (error)
927 			goto out_trans_abort;
928 
929 		/*
930 		 * Truncated "down", so we're removing references to old data
931 		 * here - if we delay flushing for a long time, we expose
932 		 * ourselves unduly to the notorious NULL files problem.  So,
933 		 * we mark this inode and flush it when the file is closed,
934 		 * and do not wait the usual (long) time for writeout.
935 		 */
936 		xfs_iflags_set(ip, XFS_ITRUNCATED);
937 
938 		/* A truncate down always removes post-EOF blocks. */
939 		xfs_inode_clear_eofblocks_tag(ip);
940 	}
941 
942 	if (iattr->ia_valid & ATTR_MODE)
943 		xfs_setattr_mode(ip, iattr);
944 	if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
945 		xfs_setattr_time(ip, iattr);
946 
947 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
948 
949 	XFS_STATS_INC(xs_ig_attrchg);
950 
951 	if (mp->m_flags & XFS_MOUNT_WSYNC)
952 		xfs_trans_set_sync(tp);
953 
954 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
955 out_unlock:
956 	if (lock_flags)
957 		xfs_iunlock(ip, lock_flags);
958 	return error;
959 
960 out_trans_abort:
961 	commit_flags |= XFS_TRANS_ABORT;
962 out_trans_cancel:
963 	xfs_trans_cancel(tp, commit_flags);
964 	goto out_unlock;
965 }
966 
967 STATIC int
968 xfs_vn_setattr(
969 	struct dentry		*dentry,
970 	struct iattr		*iattr)
971 {
972 	struct xfs_inode	*ip = XFS_I(dentry->d_inode);
973 	int			error;
974 
975 	if (iattr->ia_valid & ATTR_SIZE) {
976 		xfs_ilock(ip, XFS_IOLOCK_EXCL);
977 		error = xfs_setattr_size(ip, iattr);
978 		xfs_iunlock(ip, XFS_IOLOCK_EXCL);
979 	} else {
980 		error = xfs_setattr_nonsize(ip, iattr, 0);
981 	}
982 
983 	return error;
984 }
985 
986 STATIC int
987 xfs_vn_update_time(
988 	struct inode		*inode,
989 	struct timespec		*now,
990 	int			flags)
991 {
992 	struct xfs_inode	*ip = XFS_I(inode);
993 	struct xfs_mount	*mp = ip->i_mount;
994 	struct xfs_trans	*tp;
995 	int			error;
996 
997 	trace_xfs_update_time(ip);
998 
999 	tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
1000 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
1001 	if (error) {
1002 		xfs_trans_cancel(tp, 0);
1003 		return error;
1004 	}
1005 
1006 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1007 	if (flags & S_CTIME) {
1008 		inode->i_ctime = *now;
1009 		ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
1010 		ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
1011 	}
1012 	if (flags & S_MTIME) {
1013 		inode->i_mtime = *now;
1014 		ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
1015 		ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
1016 	}
1017 	if (flags & S_ATIME) {
1018 		inode->i_atime = *now;
1019 		ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
1020 		ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
1021 	}
1022 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1023 	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1024 	return xfs_trans_commit(tp, 0);
1025 }
1026 
1027 #define XFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
1028 
1029 /*
1030  * Call fiemap helper to fill in user data.
1031  * Returns positive errors to xfs_getbmap.
1032  */
1033 STATIC int
1034 xfs_fiemap_format(
1035 	void			**arg,
1036 	struct getbmapx		*bmv,
1037 	int			*full)
1038 {
1039 	int			error;
1040 	struct fiemap_extent_info *fieinfo = *arg;
1041 	u32			fiemap_flags = 0;
1042 	u64			logical, physical, length;
1043 
1044 	/* Do nothing for a hole */
1045 	if (bmv->bmv_block == -1LL)
1046 		return 0;
1047 
1048 	logical = BBTOB(bmv->bmv_offset);
1049 	physical = BBTOB(bmv->bmv_block);
1050 	length = BBTOB(bmv->bmv_length);
1051 
1052 	if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1053 		fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1054 	else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1055 		fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1056 				 FIEMAP_EXTENT_UNKNOWN);
1057 		physical = 0;   /* no block yet */
1058 	}
1059 	if (bmv->bmv_oflags & BMV_OF_LAST)
1060 		fiemap_flags |= FIEMAP_EXTENT_LAST;
1061 
1062 	error = fiemap_fill_next_extent(fieinfo, logical, physical,
1063 					length, fiemap_flags);
1064 	if (error > 0) {
1065 		error = 0;
1066 		*full = 1;	/* user array now full */
1067 	}
1068 
1069 	return error;
1070 }
1071 
1072 STATIC int
1073 xfs_vn_fiemap(
1074 	struct inode		*inode,
1075 	struct fiemap_extent_info *fieinfo,
1076 	u64			start,
1077 	u64			length)
1078 {
1079 	xfs_inode_t		*ip = XFS_I(inode);
1080 	struct getbmapx		bm;
1081 	int			error;
1082 
1083 	error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1084 	if (error)
1085 		return error;
1086 
1087 	/* Set up bmap header for xfs internal routine */
1088 	bm.bmv_offset = BTOBBT(start);
1089 	/* Special case for whole file */
1090 	if (length == FIEMAP_MAX_OFFSET)
1091 		bm.bmv_length = -1LL;
1092 	else
1093 		bm.bmv_length = BTOBB(start + length) - bm.bmv_offset;
1094 
1095 	/* We add one because in getbmap world count includes the header */
1096 	bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1097 					fieinfo->fi_extents_max + 1;
1098 	bm.bmv_count = min_t(__s32, bm.bmv_count,
1099 			     (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1100 	bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1101 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1102 		bm.bmv_iflags |= BMV_IF_ATTRFORK;
1103 	if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1104 		bm.bmv_iflags |= BMV_IF_DELALLOC;
1105 
1106 	error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1107 	if (error)
1108 		return error;
1109 
1110 	return 0;
1111 }
1112 
1113 STATIC int
1114 xfs_vn_tmpfile(
1115 	struct inode	*dir,
1116 	struct dentry	*dentry,
1117 	umode_t		mode)
1118 {
1119 	return xfs_generic_create(dir, dentry, mode, 0, true);
1120 }
1121 
1122 static const struct inode_operations xfs_inode_operations = {
1123 	.get_acl		= xfs_get_acl,
1124 	.set_acl		= xfs_set_acl,
1125 	.getattr		= xfs_vn_getattr,
1126 	.setattr		= xfs_vn_setattr,
1127 	.setxattr		= generic_setxattr,
1128 	.getxattr		= generic_getxattr,
1129 	.removexattr		= generic_removexattr,
1130 	.listxattr		= xfs_vn_listxattr,
1131 	.fiemap			= xfs_vn_fiemap,
1132 	.update_time		= xfs_vn_update_time,
1133 };
1134 
1135 static const struct inode_operations xfs_dir_inode_operations = {
1136 	.create			= xfs_vn_create,
1137 	.lookup			= xfs_vn_lookup,
1138 	.link			= xfs_vn_link,
1139 	.unlink			= xfs_vn_unlink,
1140 	.symlink		= xfs_vn_symlink,
1141 	.mkdir			= xfs_vn_mkdir,
1142 	/*
1143 	 * Yes, XFS uses the same method for rmdir and unlink.
1144 	 *
1145 	 * There are some subtile differences deeper in the code,
1146 	 * but we use S_ISDIR to check for those.
1147 	 */
1148 	.rmdir			= xfs_vn_unlink,
1149 	.mknod			= xfs_vn_mknod,
1150 	.rename			= xfs_vn_rename,
1151 	.get_acl		= xfs_get_acl,
1152 	.set_acl		= xfs_set_acl,
1153 	.getattr		= xfs_vn_getattr,
1154 	.setattr		= xfs_vn_setattr,
1155 	.setxattr		= generic_setxattr,
1156 	.getxattr		= generic_getxattr,
1157 	.removexattr		= generic_removexattr,
1158 	.listxattr		= xfs_vn_listxattr,
1159 	.update_time		= xfs_vn_update_time,
1160 	.tmpfile		= xfs_vn_tmpfile,
1161 };
1162 
1163 static const struct inode_operations xfs_dir_ci_inode_operations = {
1164 	.create			= xfs_vn_create,
1165 	.lookup			= xfs_vn_ci_lookup,
1166 	.link			= xfs_vn_link,
1167 	.unlink			= xfs_vn_unlink,
1168 	.symlink		= xfs_vn_symlink,
1169 	.mkdir			= xfs_vn_mkdir,
1170 	/*
1171 	 * Yes, XFS uses the same method for rmdir and unlink.
1172 	 *
1173 	 * There are some subtile differences deeper in the code,
1174 	 * but we use S_ISDIR to check for those.
1175 	 */
1176 	.rmdir			= xfs_vn_unlink,
1177 	.mknod			= xfs_vn_mknod,
1178 	.rename			= xfs_vn_rename,
1179 	.get_acl		= xfs_get_acl,
1180 	.set_acl		= xfs_set_acl,
1181 	.getattr		= xfs_vn_getattr,
1182 	.setattr		= xfs_vn_setattr,
1183 	.setxattr		= generic_setxattr,
1184 	.getxattr		= generic_getxattr,
1185 	.removexattr		= generic_removexattr,
1186 	.listxattr		= xfs_vn_listxattr,
1187 	.update_time		= xfs_vn_update_time,
1188 	.tmpfile		= xfs_vn_tmpfile,
1189 };
1190 
1191 static const struct inode_operations xfs_symlink_inode_operations = {
1192 	.readlink		= generic_readlink,
1193 	.follow_link		= xfs_vn_follow_link,
1194 	.put_link		= kfree_put_link,
1195 	.getattr		= xfs_vn_getattr,
1196 	.setattr		= xfs_vn_setattr,
1197 	.setxattr		= generic_setxattr,
1198 	.getxattr		= generic_getxattr,
1199 	.removexattr		= generic_removexattr,
1200 	.listxattr		= xfs_vn_listxattr,
1201 	.update_time		= xfs_vn_update_time,
1202 };
1203 
1204 STATIC void
1205 xfs_diflags_to_iflags(
1206 	struct inode		*inode,
1207 	struct xfs_inode	*ip)
1208 {
1209 	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1210 		inode->i_flags |= S_IMMUTABLE;
1211 	else
1212 		inode->i_flags &= ~S_IMMUTABLE;
1213 	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1214 		inode->i_flags |= S_APPEND;
1215 	else
1216 		inode->i_flags &= ~S_APPEND;
1217 	if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1218 		inode->i_flags |= S_SYNC;
1219 	else
1220 		inode->i_flags &= ~S_SYNC;
1221 	if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1222 		inode->i_flags |= S_NOATIME;
1223 	else
1224 		inode->i_flags &= ~S_NOATIME;
1225 }
1226 
1227 /*
1228  * Initialize the Linux inode, set up the operation vectors and
1229  * unlock the inode.
1230  *
1231  * When reading existing inodes from disk this is called directly
1232  * from xfs_iget, when creating a new inode it is called from
1233  * xfs_ialloc after setting up the inode.
1234  *
1235  * We are always called with an uninitialised linux inode here.
1236  * We need to initialise the necessary fields and take a reference
1237  * on it.
1238  */
1239 void
1240 xfs_setup_inode(
1241 	struct xfs_inode	*ip)
1242 {
1243 	struct inode		*inode = &ip->i_vnode;
1244 	gfp_t			gfp_mask;
1245 
1246 	inode->i_ino = ip->i_ino;
1247 	inode->i_state = I_NEW;
1248 
1249 	inode_sb_list_add(inode);
1250 	/* make the inode look hashed for the writeback code */
1251 	hlist_add_fake(&inode->i_hash);
1252 
1253 	inode->i_mode	= ip->i_d.di_mode;
1254 	set_nlink(inode, ip->i_d.di_nlink);
1255 	inode->i_uid    = xfs_uid_to_kuid(ip->i_d.di_uid);
1256 	inode->i_gid    = xfs_gid_to_kgid(ip->i_d.di_gid);
1257 
1258 	switch (inode->i_mode & S_IFMT) {
1259 	case S_IFBLK:
1260 	case S_IFCHR:
1261 		inode->i_rdev =
1262 			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1263 			      sysv_minor(ip->i_df.if_u2.if_rdev));
1264 		break;
1265 	default:
1266 		inode->i_rdev = 0;
1267 		break;
1268 	}
1269 
1270 	inode->i_generation = ip->i_d.di_gen;
1271 	i_size_write(inode, ip->i_d.di_size);
1272 	inode->i_atime.tv_sec	= ip->i_d.di_atime.t_sec;
1273 	inode->i_atime.tv_nsec	= ip->i_d.di_atime.t_nsec;
1274 	inode->i_mtime.tv_sec	= ip->i_d.di_mtime.t_sec;
1275 	inode->i_mtime.tv_nsec	= ip->i_d.di_mtime.t_nsec;
1276 	inode->i_ctime.tv_sec	= ip->i_d.di_ctime.t_sec;
1277 	inode->i_ctime.tv_nsec	= ip->i_d.di_ctime.t_nsec;
1278 	xfs_diflags_to_iflags(inode, ip);
1279 
1280 	ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1281 	lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1282 	switch (inode->i_mode & S_IFMT) {
1283 	case S_IFREG:
1284 		inode->i_op = &xfs_inode_operations;
1285 		inode->i_fop = &xfs_file_operations;
1286 		inode->i_mapping->a_ops = &xfs_address_space_operations;
1287 		break;
1288 	case S_IFDIR:
1289 		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1290 		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1291 			inode->i_op = &xfs_dir_ci_inode_operations;
1292 		else
1293 			inode->i_op = &xfs_dir_inode_operations;
1294 		inode->i_fop = &xfs_dir_file_operations;
1295 		ip->d_ops = ip->i_mount->m_dir_inode_ops;
1296 		break;
1297 	case S_IFLNK:
1298 		inode->i_op = &xfs_symlink_inode_operations;
1299 		if (!(ip->i_df.if_flags & XFS_IFINLINE))
1300 			inode->i_mapping->a_ops = &xfs_address_space_operations;
1301 		break;
1302 	default:
1303 		inode->i_op = &xfs_inode_operations;
1304 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1305 		break;
1306 	}
1307 
1308 	/*
1309 	 * Ensure all page cache allocations are done from GFP_NOFS context to
1310 	 * prevent direct reclaim recursion back into the filesystem and blowing
1311 	 * stacks or deadlocking.
1312 	 */
1313 	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1314 	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1315 
1316 	/*
1317 	 * If there is no attribute fork no ACL can exist on this inode,
1318 	 * and it can't have any file capabilities attached to it either.
1319 	 */
1320 	if (!XFS_IFORK_Q(ip)) {
1321 		inode_has_no_xattr(inode);
1322 		cache_no_acl(inode);
1323 	}
1324 
1325 	xfs_iflags_clear(ip, XFS_INEW);
1326 	barrier();
1327 
1328 	unlock_new_inode(inode);
1329 }
1330