xref: /openbmc/linux/fs/xfs/xfs_iops.c (revision 92a2c6b2)
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_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_inode.h"
27 #include "xfs_bmap.h"
28 #include "xfs_bmap_util.h"
29 #include "xfs_acl.h"
30 #include "xfs_quota.h"
31 #include "xfs_error.h"
32 #include "xfs_attr.h"
33 #include "xfs_trans.h"
34 #include "xfs_trace.h"
35 #include "xfs_icache.h"
36 #include "xfs_symlink.h"
37 #include "xfs_da_btree.h"
38 #include "xfs_dir2.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_pnfs.h"
41 
42 #include <linux/capability.h>
43 #include <linux/xattr.h>
44 #include <linux/namei.h>
45 #include <linux/posix_acl.h>
46 #include <linux/security.h>
47 #include <linux/fiemap.h>
48 #include <linux/slab.h>
49 
50 /*
51  * Directories have different lock order w.r.t. mmap_sem compared to regular
52  * files. This is due to readdir potentially triggering page faults on a user
53  * buffer inside filldir(), and this happens with the ilock on the directory
54  * held. For regular files, the lock order is the other way around - the
55  * mmap_sem is taken during the page fault, and then we lock the ilock to do
56  * block mapping. Hence we need a different class for the directory ilock so
57  * that lockdep can tell them apart.
58  */
59 static struct lock_class_key xfs_nondir_ilock_class;
60 static struct lock_class_key xfs_dir_ilock_class;
61 
62 static int
63 xfs_initxattrs(
64 	struct inode		*inode,
65 	const struct xattr	*xattr_array,
66 	void			*fs_info)
67 {
68 	const struct xattr	*xattr;
69 	struct xfs_inode	*ip = XFS_I(inode);
70 	int			error = 0;
71 
72 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
73 		error = xfs_attr_set(ip, xattr->name, xattr->value,
74 				      xattr->value_len, ATTR_SECURE);
75 		if (error < 0)
76 			break;
77 	}
78 	return error;
79 }
80 
81 /*
82  * Hook in SELinux.  This is not quite correct yet, what we really need
83  * here (as we do for default ACLs) is a mechanism by which creation of
84  * these attrs can be journalled at inode creation time (along with the
85  * inode, of course, such that log replay can't cause these to be lost).
86  */
87 
88 STATIC int
89 xfs_init_security(
90 	struct inode	*inode,
91 	struct inode	*dir,
92 	const struct qstr *qstr)
93 {
94 	return security_inode_init_security(inode, dir, qstr,
95 					     &xfs_initxattrs, NULL);
96 }
97 
98 static void
99 xfs_dentry_to_name(
100 	struct xfs_name	*namep,
101 	struct dentry	*dentry,
102 	int		mode)
103 {
104 	namep->name = dentry->d_name.name;
105 	namep->len = dentry->d_name.len;
106 	namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
107 }
108 
109 STATIC void
110 xfs_cleanup_inode(
111 	struct inode	*dir,
112 	struct inode	*inode,
113 	struct dentry	*dentry)
114 {
115 	struct xfs_name	teardown;
116 
117 	/* Oh, the horror.
118 	 * If we can't add the ACL or we fail in
119 	 * xfs_init_security we must back out.
120 	 * ENOSPC can hit here, among other things.
121 	 */
122 	xfs_dentry_to_name(&teardown, dentry, 0);
123 
124 	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
125 }
126 
127 STATIC int
128 xfs_generic_create(
129 	struct inode	*dir,
130 	struct dentry	*dentry,
131 	umode_t		mode,
132 	dev_t		rdev,
133 	bool		tmpfile)	/* unnamed file */
134 {
135 	struct inode	*inode;
136 	struct xfs_inode *ip = NULL;
137 	struct posix_acl *default_acl, *acl;
138 	struct xfs_name	name;
139 	int		error;
140 
141 	/*
142 	 * Irix uses Missed'em'V split, but doesn't want to see
143 	 * the upper 5 bits of (14bit) major.
144 	 */
145 	if (S_ISCHR(mode) || S_ISBLK(mode)) {
146 		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
147 			return -EINVAL;
148 		rdev = sysv_encode_dev(rdev);
149 	} else {
150 		rdev = 0;
151 	}
152 
153 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
154 	if (error)
155 		return error;
156 
157 	if (!tmpfile) {
158 		xfs_dentry_to_name(&name, dentry, mode);
159 		error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
160 	} else {
161 		error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
162 	}
163 	if (unlikely(error))
164 		goto out_free_acl;
165 
166 	inode = VFS_I(ip);
167 
168 	error = xfs_init_security(inode, dir, &dentry->d_name);
169 	if (unlikely(error))
170 		goto out_cleanup_inode;
171 
172 #ifdef CONFIG_XFS_POSIX_ACL
173 	if (default_acl) {
174 		error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
175 		if (error)
176 			goto out_cleanup_inode;
177 	}
178 	if (acl) {
179 		error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
180 		if (error)
181 			goto out_cleanup_inode;
182 	}
183 #endif
184 
185 	if (tmpfile)
186 		d_tmpfile(dentry, inode);
187 	else
188 		d_instantiate(dentry, inode);
189 
190  out_free_acl:
191 	if (default_acl)
192 		posix_acl_release(default_acl);
193 	if (acl)
194 		posix_acl_release(acl);
195 	return error;
196 
197  out_cleanup_inode:
198 	if (!tmpfile)
199 		xfs_cleanup_inode(dir, inode, dentry);
200 	iput(inode);
201 	goto out_free_acl;
202 }
203 
204 STATIC int
205 xfs_vn_mknod(
206 	struct inode	*dir,
207 	struct dentry	*dentry,
208 	umode_t		mode,
209 	dev_t		rdev)
210 {
211 	return xfs_generic_create(dir, dentry, mode, rdev, false);
212 }
213 
214 STATIC int
215 xfs_vn_create(
216 	struct inode	*dir,
217 	struct dentry	*dentry,
218 	umode_t		mode,
219 	bool		flags)
220 {
221 	return xfs_vn_mknod(dir, dentry, mode, 0);
222 }
223 
224 STATIC int
225 xfs_vn_mkdir(
226 	struct inode	*dir,
227 	struct dentry	*dentry,
228 	umode_t		mode)
229 {
230 	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
231 }
232 
233 STATIC struct dentry *
234 xfs_vn_lookup(
235 	struct inode	*dir,
236 	struct dentry	*dentry,
237 	unsigned int flags)
238 {
239 	struct xfs_inode *cip;
240 	struct xfs_name	name;
241 	int		error;
242 
243 	if (dentry->d_name.len >= MAXNAMELEN)
244 		return ERR_PTR(-ENAMETOOLONG);
245 
246 	xfs_dentry_to_name(&name, dentry, 0);
247 	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
248 	if (unlikely(error)) {
249 		if (unlikely(error != -ENOENT))
250 			return ERR_PTR(error);
251 		d_add(dentry, NULL);
252 		return NULL;
253 	}
254 
255 	return d_splice_alias(VFS_I(cip), dentry);
256 }
257 
258 STATIC struct dentry *
259 xfs_vn_ci_lookup(
260 	struct inode	*dir,
261 	struct dentry	*dentry,
262 	unsigned int flags)
263 {
264 	struct xfs_inode *ip;
265 	struct xfs_name	xname;
266 	struct xfs_name ci_name;
267 	struct qstr	dname;
268 	int		error;
269 
270 	if (dentry->d_name.len >= MAXNAMELEN)
271 		return ERR_PTR(-ENAMETOOLONG);
272 
273 	xfs_dentry_to_name(&xname, dentry, 0);
274 	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
275 	if (unlikely(error)) {
276 		if (unlikely(error != -ENOENT))
277 			return ERR_PTR(error);
278 		/*
279 		 * call d_add(dentry, NULL) here when d_drop_negative_children
280 		 * is called in xfs_vn_mknod (ie. allow negative dentries
281 		 * with CI filesystems).
282 		 */
283 		return NULL;
284 	}
285 
286 	/* if exact match, just splice and exit */
287 	if (!ci_name.name)
288 		return d_splice_alias(VFS_I(ip), dentry);
289 
290 	/* else case-insensitive match... */
291 	dname.name = ci_name.name;
292 	dname.len = ci_name.len;
293 	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
294 	kmem_free(ci_name.name);
295 	return dentry;
296 }
297 
298 STATIC int
299 xfs_vn_link(
300 	struct dentry	*old_dentry,
301 	struct inode	*dir,
302 	struct dentry	*dentry)
303 {
304 	struct inode	*inode = old_dentry->d_inode;
305 	struct xfs_name	name;
306 	int		error;
307 
308 	xfs_dentry_to_name(&name, dentry, inode->i_mode);
309 
310 	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
311 	if (unlikely(error))
312 		return error;
313 
314 	ihold(inode);
315 	d_instantiate(dentry, inode);
316 	return 0;
317 }
318 
319 STATIC int
320 xfs_vn_unlink(
321 	struct inode	*dir,
322 	struct dentry	*dentry)
323 {
324 	struct xfs_name	name;
325 	int		error;
326 
327 	xfs_dentry_to_name(&name, dentry, 0);
328 
329 	error = xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
330 	if (error)
331 		return error;
332 
333 	/*
334 	 * With unlink, the VFS makes the dentry "negative": no inode,
335 	 * but still hashed. This is incompatible with case-insensitive
336 	 * mode, so invalidate (unhash) the dentry in CI-mode.
337 	 */
338 	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
339 		d_invalidate(dentry);
340 	return 0;
341 }
342 
343 STATIC int
344 xfs_vn_symlink(
345 	struct inode	*dir,
346 	struct dentry	*dentry,
347 	const char	*symname)
348 {
349 	struct inode	*inode;
350 	struct xfs_inode *cip = NULL;
351 	struct xfs_name	name;
352 	int		error;
353 	umode_t		mode;
354 
355 	mode = S_IFLNK |
356 		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
357 	xfs_dentry_to_name(&name, dentry, mode);
358 
359 	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
360 	if (unlikely(error))
361 		goto out;
362 
363 	inode = VFS_I(cip);
364 
365 	error = xfs_init_security(inode, dir, &dentry->d_name);
366 	if (unlikely(error))
367 		goto out_cleanup_inode;
368 
369 	d_instantiate(dentry, inode);
370 	return 0;
371 
372  out_cleanup_inode:
373 	xfs_cleanup_inode(dir, inode, dentry);
374 	iput(inode);
375  out:
376 	return error;
377 }
378 
379 STATIC int
380 xfs_vn_rename(
381 	struct inode	*odir,
382 	struct dentry	*odentry,
383 	struct inode	*ndir,
384 	struct dentry	*ndentry,
385 	unsigned int	flags)
386 {
387 	struct inode	*new_inode = ndentry->d_inode;
388 	int		omode = 0;
389 	struct xfs_name	oname;
390 	struct xfs_name	nname;
391 
392 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
393 		return -EINVAL;
394 
395 	/* if we are exchanging files, we need to set i_mode of both files */
396 	if (flags & RENAME_EXCHANGE)
397 		omode = ndentry->d_inode->i_mode;
398 
399 	xfs_dentry_to_name(&oname, odentry, omode);
400 	xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
401 
402 	return xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
403 			  XFS_I(ndir), &nname,
404 			  new_inode ? XFS_I(new_inode) : NULL, flags);
405 }
406 
407 /*
408  * careful here - this function can get called recursively, so
409  * we need to be very careful about how much stack we use.
410  * uio is kmalloced for this reason...
411  */
412 STATIC void *
413 xfs_vn_follow_link(
414 	struct dentry		*dentry,
415 	struct nameidata	*nd)
416 {
417 	char			*link;
418 	int			error = -ENOMEM;
419 
420 	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
421 	if (!link)
422 		goto out_err;
423 
424 	error = xfs_readlink(XFS_I(dentry->d_inode), link);
425 	if (unlikely(error))
426 		goto out_kfree;
427 
428 	nd_set_link(nd, link);
429 	return NULL;
430 
431  out_kfree:
432 	kfree(link);
433  out_err:
434 	nd_set_link(nd, ERR_PTR(error));
435 	return NULL;
436 }
437 
438 STATIC int
439 xfs_vn_getattr(
440 	struct vfsmount		*mnt,
441 	struct dentry		*dentry,
442 	struct kstat		*stat)
443 {
444 	struct inode		*inode = dentry->d_inode;
445 	struct xfs_inode	*ip = XFS_I(inode);
446 	struct xfs_mount	*mp = ip->i_mount;
447 
448 	trace_xfs_getattr(ip);
449 
450 	if (XFS_FORCED_SHUTDOWN(mp))
451 		return -EIO;
452 
453 	stat->size = XFS_ISIZE(ip);
454 	stat->dev = inode->i_sb->s_dev;
455 	stat->mode = ip->i_d.di_mode;
456 	stat->nlink = ip->i_d.di_nlink;
457 	stat->uid = inode->i_uid;
458 	stat->gid = inode->i_gid;
459 	stat->ino = ip->i_ino;
460 	stat->atime = inode->i_atime;
461 	stat->mtime = inode->i_mtime;
462 	stat->ctime = inode->i_ctime;
463 	stat->blocks =
464 		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
465 
466 
467 	switch (inode->i_mode & S_IFMT) {
468 	case S_IFBLK:
469 	case S_IFCHR:
470 		stat->blksize = BLKDEV_IOSIZE;
471 		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
472 				   sysv_minor(ip->i_df.if_u2.if_rdev));
473 		break;
474 	default:
475 		if (XFS_IS_REALTIME_INODE(ip)) {
476 			/*
477 			 * If the file blocks are being allocated from a
478 			 * realtime volume, then return the inode's realtime
479 			 * extent size or the realtime volume's extent size.
480 			 */
481 			stat->blksize =
482 				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
483 		} else
484 			stat->blksize = xfs_preferred_iosize(mp);
485 		stat->rdev = 0;
486 		break;
487 	}
488 
489 	return 0;
490 }
491 
492 static void
493 xfs_setattr_mode(
494 	struct xfs_inode	*ip,
495 	struct iattr		*iattr)
496 {
497 	struct inode		*inode = VFS_I(ip);
498 	umode_t			mode = iattr->ia_mode;
499 
500 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
501 
502 	ip->i_d.di_mode &= S_IFMT;
503 	ip->i_d.di_mode |= mode & ~S_IFMT;
504 
505 	inode->i_mode &= S_IFMT;
506 	inode->i_mode |= mode & ~S_IFMT;
507 }
508 
509 void
510 xfs_setattr_time(
511 	struct xfs_inode	*ip,
512 	struct iattr		*iattr)
513 {
514 	struct inode		*inode = VFS_I(ip);
515 
516 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
517 
518 	if (iattr->ia_valid & ATTR_ATIME) {
519 		inode->i_atime = iattr->ia_atime;
520 		ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
521 		ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
522 	}
523 	if (iattr->ia_valid & ATTR_CTIME) {
524 		inode->i_ctime = iattr->ia_ctime;
525 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
526 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
527 	}
528 	if (iattr->ia_valid & ATTR_MTIME) {
529 		inode->i_mtime = iattr->ia_mtime;
530 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
531 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
532 	}
533 }
534 
535 int
536 xfs_setattr_nonsize(
537 	struct xfs_inode	*ip,
538 	struct iattr		*iattr,
539 	int			flags)
540 {
541 	xfs_mount_t		*mp = ip->i_mount;
542 	struct inode		*inode = VFS_I(ip);
543 	int			mask = iattr->ia_valid;
544 	xfs_trans_t		*tp;
545 	int			error;
546 	kuid_t			uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
547 	kgid_t			gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
548 	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
549 	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
550 
551 	trace_xfs_setattr(ip);
552 
553 	/* If acls are being inherited, we already have this checked */
554 	if (!(flags & XFS_ATTR_NOACL)) {
555 		if (mp->m_flags & XFS_MOUNT_RDONLY)
556 			return -EROFS;
557 
558 		if (XFS_FORCED_SHUTDOWN(mp))
559 			return -EIO;
560 
561 		error = inode_change_ok(inode, iattr);
562 		if (error)
563 			return error;
564 	}
565 
566 	ASSERT((mask & ATTR_SIZE) == 0);
567 
568 	/*
569 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
570 	 * before we start any other transactions. Trying to do this later
571 	 * is messy. We don't care to take a readlock to look at the ids
572 	 * in inode here, because we can't hold it across the trans_reserve.
573 	 * If the IDs do change before we take the ilock, we're covered
574 	 * because the i_*dquot fields will get updated anyway.
575 	 */
576 	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
577 		uint	qflags = 0;
578 
579 		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
580 			uid = iattr->ia_uid;
581 			qflags |= XFS_QMOPT_UQUOTA;
582 		} else {
583 			uid = inode->i_uid;
584 		}
585 		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
586 			gid = iattr->ia_gid;
587 			qflags |= XFS_QMOPT_GQUOTA;
588 		}  else {
589 			gid = inode->i_gid;
590 		}
591 
592 		/*
593 		 * We take a reference when we initialize udqp and gdqp,
594 		 * so it is important that we never blindly double trip on
595 		 * the same variable. See xfs_create() for an example.
596 		 */
597 		ASSERT(udqp == NULL);
598 		ASSERT(gdqp == NULL);
599 		error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
600 					   xfs_kgid_to_gid(gid),
601 					   xfs_get_projid(ip),
602 					   qflags, &udqp, &gdqp, NULL);
603 		if (error)
604 			return error;
605 	}
606 
607 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
608 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
609 	if (error)
610 		goto out_dqrele;
611 
612 	xfs_ilock(ip, XFS_ILOCK_EXCL);
613 
614 	/*
615 	 * Change file ownership.  Must be the owner or privileged.
616 	 */
617 	if (mask & (ATTR_UID|ATTR_GID)) {
618 		/*
619 		 * These IDs could have changed since we last looked at them.
620 		 * But, we're assured that if the ownership did change
621 		 * while we didn't have the inode locked, inode's dquot(s)
622 		 * would have changed also.
623 		 */
624 		iuid = inode->i_uid;
625 		igid = inode->i_gid;
626 		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
627 		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
628 
629 		/*
630 		 * Do a quota reservation only if uid/gid is actually
631 		 * going to change.
632 		 */
633 		if (XFS_IS_QUOTA_RUNNING(mp) &&
634 		    ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
635 		     (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
636 			ASSERT(tp);
637 			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
638 						NULL, capable(CAP_FOWNER) ?
639 						XFS_QMOPT_FORCE_RES : 0);
640 			if (error)	/* out of quota */
641 				goto out_trans_cancel;
642 		}
643 	}
644 
645 	xfs_trans_ijoin(tp, ip, 0);
646 
647 	/*
648 	 * Change file ownership.  Must be the owner or privileged.
649 	 */
650 	if (mask & (ATTR_UID|ATTR_GID)) {
651 		/*
652 		 * CAP_FSETID overrides the following restrictions:
653 		 *
654 		 * The set-user-ID and set-group-ID bits of a file will be
655 		 * cleared upon successful return from chown()
656 		 */
657 		if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
658 		    !capable(CAP_FSETID))
659 			ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
660 
661 		/*
662 		 * Change the ownerships and register quota modifications
663 		 * in the transaction.
664 		 */
665 		if (!uid_eq(iuid, uid)) {
666 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
667 				ASSERT(mask & ATTR_UID);
668 				ASSERT(udqp);
669 				olddquot1 = xfs_qm_vop_chown(tp, ip,
670 							&ip->i_udquot, udqp);
671 			}
672 			ip->i_d.di_uid = xfs_kuid_to_uid(uid);
673 			inode->i_uid = uid;
674 		}
675 		if (!gid_eq(igid, gid)) {
676 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
677 				ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
678 				       !XFS_IS_PQUOTA_ON(mp));
679 				ASSERT(mask & ATTR_GID);
680 				ASSERT(gdqp);
681 				olddquot2 = xfs_qm_vop_chown(tp, ip,
682 							&ip->i_gdquot, gdqp);
683 			}
684 			ip->i_d.di_gid = xfs_kgid_to_gid(gid);
685 			inode->i_gid = gid;
686 		}
687 	}
688 
689 	if (mask & ATTR_MODE)
690 		xfs_setattr_mode(ip, iattr);
691 	if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
692 		xfs_setattr_time(ip, iattr);
693 
694 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
695 
696 	XFS_STATS_INC(xs_ig_attrchg);
697 
698 	if (mp->m_flags & XFS_MOUNT_WSYNC)
699 		xfs_trans_set_sync(tp);
700 	error = xfs_trans_commit(tp, 0);
701 
702 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
703 
704 	/*
705 	 * Release any dquot(s) the inode had kept before chown.
706 	 */
707 	xfs_qm_dqrele(olddquot1);
708 	xfs_qm_dqrele(olddquot2);
709 	xfs_qm_dqrele(udqp);
710 	xfs_qm_dqrele(gdqp);
711 
712 	if (error)
713 		return error;
714 
715 	/*
716 	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
717 	 * 	     update.  We could avoid this with linked transactions
718 	 * 	     and passing down the transaction pointer all the way
719 	 *	     to attr_set.  No previous user of the generic
720 	 * 	     Posix ACL code seems to care about this issue either.
721 	 */
722 	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
723 		error = posix_acl_chmod(inode, inode->i_mode);
724 		if (error)
725 			return error;
726 	}
727 
728 	return 0;
729 
730 out_trans_cancel:
731 	xfs_trans_cancel(tp, 0);
732 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
733 out_dqrele:
734 	xfs_qm_dqrele(udqp);
735 	xfs_qm_dqrele(gdqp);
736 	return error;
737 }
738 
739 /*
740  * Truncate file.  Must have write permission and not be a directory.
741  */
742 int
743 xfs_setattr_size(
744 	struct xfs_inode	*ip,
745 	struct iattr		*iattr)
746 {
747 	struct xfs_mount	*mp = ip->i_mount;
748 	struct inode		*inode = VFS_I(ip);
749 	xfs_off_t		oldsize, newsize;
750 	struct xfs_trans	*tp;
751 	int			error;
752 	uint			lock_flags = 0;
753 	uint			commit_flags = 0;
754 
755 	trace_xfs_setattr(ip);
756 
757 	if (mp->m_flags & XFS_MOUNT_RDONLY)
758 		return -EROFS;
759 
760 	if (XFS_FORCED_SHUTDOWN(mp))
761 		return -EIO;
762 
763 	error = inode_change_ok(inode, iattr);
764 	if (error)
765 		return error;
766 
767 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
768 	ASSERT(S_ISREG(ip->i_d.di_mode));
769 	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
770 		ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
771 
772 	oldsize = inode->i_size;
773 	newsize = iattr->ia_size;
774 
775 	/*
776 	 * Short circuit the truncate case for zero length files.
777 	 */
778 	if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
779 		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
780 			return 0;
781 
782 		/*
783 		 * Use the regular setattr path to update the timestamps.
784 		 */
785 		iattr->ia_valid &= ~ATTR_SIZE;
786 		return xfs_setattr_nonsize(ip, iattr, 0);
787 	}
788 
789 	/*
790 	 * Make sure that the dquots are attached to the inode.
791 	 */
792 	error = xfs_qm_dqattach(ip, 0);
793 	if (error)
794 		return error;
795 
796 	/*
797 	 * Now we can make the changes.  Before we join the inode to the
798 	 * transaction, take care of the part of the truncation that must be
799 	 * done without the inode lock.  This needs to be done before joining
800 	 * the inode to the transaction, because the inode cannot be unlocked
801 	 * once it is a part of the transaction.
802 	 */
803 	if (newsize > oldsize) {
804 		/*
805 		 * Do the first part of growing a file: zero any data in the
806 		 * last block that is beyond the old EOF.  We need to do this
807 		 * before the inode is joined to the transaction to modify
808 		 * i_size.
809 		 */
810 		error = xfs_zero_eof(ip, newsize, oldsize);
811 		if (error)
812 			return error;
813 	}
814 
815 	/*
816 	 * We are going to log the inode size change in this transaction so
817 	 * any previous writes that are beyond the on disk EOF and the new
818 	 * EOF that have not been written out need to be written here.  If we
819 	 * do not write the data out, we expose ourselves to the null files
820 	 * problem.
821 	 *
822 	 * Only flush from the on disk size to the smaller of the in memory
823 	 * file size or the new size as that's the range we really care about
824 	 * here and prevents waiting for other data not within the range we
825 	 * care about here.
826 	 */
827 	if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
828 		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
829 						      ip->i_d.di_size, newsize);
830 		if (error)
831 			return error;
832 	}
833 
834 	/*
835 	 * Wait for all direct I/O to complete.
836 	 */
837 	inode_dio_wait(inode);
838 
839 	/*
840 	 * Do all the page cache truncate work outside the transaction context
841 	 * as the "lock" order is page lock->log space reservation.  i.e.
842 	 * locking pages inside the transaction can ABBA deadlock with
843 	 * writeback. We have to do the VFS inode size update before we truncate
844 	 * the pagecache, however, to avoid racing with page faults beyond the
845 	 * new EOF they are not serialised against truncate operations except by
846 	 * page locks and size updates.
847 	 *
848 	 * Hence we are in a situation where a truncate can fail with ENOMEM
849 	 * from xfs_trans_reserve(), but having already truncated the in-memory
850 	 * version of the file (i.e. made user visible changes). There's not
851 	 * much we can do about this, except to hope that the caller sees ENOMEM
852 	 * and retries the truncate operation.
853 	 */
854 	error = block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
855 	if (error)
856 		return error;
857 	truncate_setsize(inode, newsize);
858 
859 	/*
860 	 * The "we can't serialise against page faults" pain gets worse.
861 	 *
862 	 * If the file is mapped then we have to clean the page at the old EOF
863 	 * when extending the file. Extending the file can expose changes the
864 	 * underlying page mapping (e.g. from beyond EOF to a hole or
865 	 * unwritten), and so on the next attempt to write to that page we need
866 	 * to remap it for write. i.e. we need .page_mkwrite() to be called.
867 	 * Hence we need to clean the page to clean the pte and so a new write
868 	 * fault will be triggered appropriately.
869 	 *
870 	 * If we do it before we change the inode size, then we can race with a
871 	 * page fault that maps the page with exactly the same problem. If we do
872 	 * it after we change the file size, then a new page fault can come in
873 	 * and allocate space before we've run the rest of the truncate
874 	 * transaction. That's kinda grotesque, but it's better than have data
875 	 * over a hole, and so that's the lesser evil that has been chosen here.
876 	 *
877 	 * The real solution, however, is to have some mechanism for locking out
878 	 * page faults while a truncate is in progress.
879 	 */
880 	if (newsize > oldsize && mapping_mapped(VFS_I(ip)->i_mapping)) {
881 		error = filemap_write_and_wait_range(
882 				VFS_I(ip)->i_mapping,
883 				round_down(oldsize, PAGE_CACHE_SIZE),
884 				round_up(oldsize, PAGE_CACHE_SIZE) - 1);
885 		if (error)
886 			return error;
887 	}
888 
889 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
890 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
891 	if (error)
892 		goto out_trans_cancel;
893 
894 	commit_flags = XFS_TRANS_RELEASE_LOG_RES;
895 	lock_flags |= XFS_ILOCK_EXCL;
896 	xfs_ilock(ip, XFS_ILOCK_EXCL);
897 	xfs_trans_ijoin(tp, ip, 0);
898 
899 	/*
900 	 * Only change the c/mtime if we are changing the size or we are
901 	 * explicitly asked to change it.  This handles the semantic difference
902 	 * between truncate() and ftruncate() as implemented in the VFS.
903 	 *
904 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
905 	 * special case where we need to update the times despite not having
906 	 * these flags set.  For all other operations the VFS set these flags
907 	 * explicitly if it wants a timestamp update.
908 	 */
909 	if (newsize != oldsize &&
910 	    !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
911 		iattr->ia_ctime = iattr->ia_mtime =
912 			current_fs_time(inode->i_sb);
913 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
914 	}
915 
916 	/*
917 	 * The first thing we do is set the size to new_size permanently on
918 	 * disk.  This way we don't have to worry about anyone ever being able
919 	 * to look at the data being freed even in the face of a crash.
920 	 * What we're getting around here is the case where we free a block, it
921 	 * is allocated to another file, it is written to, and then we crash.
922 	 * If the new data gets written to the file but the log buffers
923 	 * containing the free and reallocation don't, then we'd end up with
924 	 * garbage in the blocks being freed.  As long as we make the new size
925 	 * permanent before actually freeing any blocks it doesn't matter if
926 	 * they get written to.
927 	 */
928 	ip->i_d.di_size = newsize;
929 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
930 
931 	if (newsize <= oldsize) {
932 		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
933 		if (error)
934 			goto out_trans_abort;
935 
936 		/*
937 		 * Truncated "down", so we're removing references to old data
938 		 * here - if we delay flushing for a long time, we expose
939 		 * ourselves unduly to the notorious NULL files problem.  So,
940 		 * we mark this inode and flush it when the file is closed,
941 		 * and do not wait the usual (long) time for writeout.
942 		 */
943 		xfs_iflags_set(ip, XFS_ITRUNCATED);
944 
945 		/* A truncate down always removes post-EOF blocks. */
946 		xfs_inode_clear_eofblocks_tag(ip);
947 	}
948 
949 	if (iattr->ia_valid & ATTR_MODE)
950 		xfs_setattr_mode(ip, iattr);
951 	if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
952 		xfs_setattr_time(ip, iattr);
953 
954 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
955 
956 	XFS_STATS_INC(xs_ig_attrchg);
957 
958 	if (mp->m_flags & XFS_MOUNT_WSYNC)
959 		xfs_trans_set_sync(tp);
960 
961 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
962 out_unlock:
963 	if (lock_flags)
964 		xfs_iunlock(ip, lock_flags);
965 	return error;
966 
967 out_trans_abort:
968 	commit_flags |= XFS_TRANS_ABORT;
969 out_trans_cancel:
970 	xfs_trans_cancel(tp, commit_flags);
971 	goto out_unlock;
972 }
973 
974 STATIC int
975 xfs_vn_setattr(
976 	struct dentry		*dentry,
977 	struct iattr		*iattr)
978 {
979 	struct xfs_inode	*ip = XFS_I(dentry->d_inode);
980 	int			error;
981 
982 	if (iattr->ia_valid & ATTR_SIZE) {
983 		uint		iolock = XFS_IOLOCK_EXCL;
984 
985 		xfs_ilock(ip, iolock);
986 		error = xfs_break_layouts(dentry->d_inode, &iolock);
987 		if (!error)
988 			error = xfs_setattr_size(ip, iattr);
989 		xfs_iunlock(ip, iolock);
990 	} else {
991 		error = xfs_setattr_nonsize(ip, iattr, 0);
992 	}
993 
994 	return error;
995 }
996 
997 STATIC int
998 xfs_vn_update_time(
999 	struct inode		*inode,
1000 	struct timespec		*now,
1001 	int			flags)
1002 {
1003 	struct xfs_inode	*ip = XFS_I(inode);
1004 	struct xfs_mount	*mp = ip->i_mount;
1005 	struct xfs_trans	*tp;
1006 	int			error;
1007 
1008 	trace_xfs_update_time(ip);
1009 
1010 	tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
1011 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
1012 	if (error) {
1013 		xfs_trans_cancel(tp, 0);
1014 		return error;
1015 	}
1016 
1017 	xfs_ilock(ip, XFS_ILOCK_EXCL);
1018 	if (flags & S_CTIME) {
1019 		inode->i_ctime = *now;
1020 		ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
1021 		ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
1022 	}
1023 	if (flags & S_MTIME) {
1024 		inode->i_mtime = *now;
1025 		ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
1026 		ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
1027 	}
1028 	if (flags & S_ATIME) {
1029 		inode->i_atime = *now;
1030 		ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
1031 		ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
1032 	}
1033 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1034 	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1035 	return xfs_trans_commit(tp, 0);
1036 }
1037 
1038 #define XFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
1039 
1040 /*
1041  * Call fiemap helper to fill in user data.
1042  * Returns positive errors to xfs_getbmap.
1043  */
1044 STATIC int
1045 xfs_fiemap_format(
1046 	void			**arg,
1047 	struct getbmapx		*bmv,
1048 	int			*full)
1049 {
1050 	int			error;
1051 	struct fiemap_extent_info *fieinfo = *arg;
1052 	u32			fiemap_flags = 0;
1053 	u64			logical, physical, length;
1054 
1055 	/* Do nothing for a hole */
1056 	if (bmv->bmv_block == -1LL)
1057 		return 0;
1058 
1059 	logical = BBTOB(bmv->bmv_offset);
1060 	physical = BBTOB(bmv->bmv_block);
1061 	length = BBTOB(bmv->bmv_length);
1062 
1063 	if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1064 		fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1065 	else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1066 		fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1067 				 FIEMAP_EXTENT_UNKNOWN);
1068 		physical = 0;   /* no block yet */
1069 	}
1070 	if (bmv->bmv_oflags & BMV_OF_LAST)
1071 		fiemap_flags |= FIEMAP_EXTENT_LAST;
1072 
1073 	error = fiemap_fill_next_extent(fieinfo, logical, physical,
1074 					length, fiemap_flags);
1075 	if (error > 0) {
1076 		error = 0;
1077 		*full = 1;	/* user array now full */
1078 	}
1079 
1080 	return error;
1081 }
1082 
1083 STATIC int
1084 xfs_vn_fiemap(
1085 	struct inode		*inode,
1086 	struct fiemap_extent_info *fieinfo,
1087 	u64			start,
1088 	u64			length)
1089 {
1090 	xfs_inode_t		*ip = XFS_I(inode);
1091 	struct getbmapx		bm;
1092 	int			error;
1093 
1094 	error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1095 	if (error)
1096 		return error;
1097 
1098 	/* Set up bmap header for xfs internal routine */
1099 	bm.bmv_offset = BTOBBT(start);
1100 	/* Special case for whole file */
1101 	if (length == FIEMAP_MAX_OFFSET)
1102 		bm.bmv_length = -1LL;
1103 	else
1104 		bm.bmv_length = BTOBB(start + length) - bm.bmv_offset;
1105 
1106 	/* We add one because in getbmap world count includes the header */
1107 	bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1108 					fieinfo->fi_extents_max + 1;
1109 	bm.bmv_count = min_t(__s32, bm.bmv_count,
1110 			     (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1111 	bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1112 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1113 		bm.bmv_iflags |= BMV_IF_ATTRFORK;
1114 	if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1115 		bm.bmv_iflags |= BMV_IF_DELALLOC;
1116 
1117 	error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1118 	if (error)
1119 		return error;
1120 
1121 	return 0;
1122 }
1123 
1124 STATIC int
1125 xfs_vn_tmpfile(
1126 	struct inode	*dir,
1127 	struct dentry	*dentry,
1128 	umode_t		mode)
1129 {
1130 	return xfs_generic_create(dir, dentry, mode, 0, true);
1131 }
1132 
1133 static const struct inode_operations xfs_inode_operations = {
1134 	.get_acl		= xfs_get_acl,
1135 	.set_acl		= xfs_set_acl,
1136 	.getattr		= xfs_vn_getattr,
1137 	.setattr		= xfs_vn_setattr,
1138 	.setxattr		= generic_setxattr,
1139 	.getxattr		= generic_getxattr,
1140 	.removexattr		= generic_removexattr,
1141 	.listxattr		= xfs_vn_listxattr,
1142 	.fiemap			= xfs_vn_fiemap,
1143 	.update_time		= xfs_vn_update_time,
1144 };
1145 
1146 static const struct inode_operations xfs_dir_inode_operations = {
1147 	.create			= xfs_vn_create,
1148 	.lookup			= xfs_vn_lookup,
1149 	.link			= xfs_vn_link,
1150 	.unlink			= xfs_vn_unlink,
1151 	.symlink		= xfs_vn_symlink,
1152 	.mkdir			= xfs_vn_mkdir,
1153 	/*
1154 	 * Yes, XFS uses the same method for rmdir and unlink.
1155 	 *
1156 	 * There are some subtile differences deeper in the code,
1157 	 * but we use S_ISDIR to check for those.
1158 	 */
1159 	.rmdir			= xfs_vn_unlink,
1160 	.mknod			= xfs_vn_mknod,
1161 	.rename2		= xfs_vn_rename,
1162 	.get_acl		= xfs_get_acl,
1163 	.set_acl		= xfs_set_acl,
1164 	.getattr		= xfs_vn_getattr,
1165 	.setattr		= xfs_vn_setattr,
1166 	.setxattr		= generic_setxattr,
1167 	.getxattr		= generic_getxattr,
1168 	.removexattr		= generic_removexattr,
1169 	.listxattr		= xfs_vn_listxattr,
1170 	.update_time		= xfs_vn_update_time,
1171 	.tmpfile		= xfs_vn_tmpfile,
1172 };
1173 
1174 static const struct inode_operations xfs_dir_ci_inode_operations = {
1175 	.create			= xfs_vn_create,
1176 	.lookup			= xfs_vn_ci_lookup,
1177 	.link			= xfs_vn_link,
1178 	.unlink			= xfs_vn_unlink,
1179 	.symlink		= xfs_vn_symlink,
1180 	.mkdir			= xfs_vn_mkdir,
1181 	/*
1182 	 * Yes, XFS uses the same method for rmdir and unlink.
1183 	 *
1184 	 * There are some subtile differences deeper in the code,
1185 	 * but we use S_ISDIR to check for those.
1186 	 */
1187 	.rmdir			= xfs_vn_unlink,
1188 	.mknod			= xfs_vn_mknod,
1189 	.rename2		= xfs_vn_rename,
1190 	.get_acl		= xfs_get_acl,
1191 	.set_acl		= xfs_set_acl,
1192 	.getattr		= xfs_vn_getattr,
1193 	.setattr		= xfs_vn_setattr,
1194 	.setxattr		= generic_setxattr,
1195 	.getxattr		= generic_getxattr,
1196 	.removexattr		= generic_removexattr,
1197 	.listxattr		= xfs_vn_listxattr,
1198 	.update_time		= xfs_vn_update_time,
1199 	.tmpfile		= xfs_vn_tmpfile,
1200 };
1201 
1202 static const struct inode_operations xfs_symlink_inode_operations = {
1203 	.readlink		= generic_readlink,
1204 	.follow_link		= xfs_vn_follow_link,
1205 	.put_link		= kfree_put_link,
1206 	.getattr		= xfs_vn_getattr,
1207 	.setattr		= xfs_vn_setattr,
1208 	.setxattr		= generic_setxattr,
1209 	.getxattr		= generic_getxattr,
1210 	.removexattr		= generic_removexattr,
1211 	.listxattr		= xfs_vn_listxattr,
1212 	.update_time		= xfs_vn_update_time,
1213 };
1214 
1215 STATIC void
1216 xfs_diflags_to_iflags(
1217 	struct inode		*inode,
1218 	struct xfs_inode	*ip)
1219 {
1220 	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1221 		inode->i_flags |= S_IMMUTABLE;
1222 	else
1223 		inode->i_flags &= ~S_IMMUTABLE;
1224 	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1225 		inode->i_flags |= S_APPEND;
1226 	else
1227 		inode->i_flags &= ~S_APPEND;
1228 	if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1229 		inode->i_flags |= S_SYNC;
1230 	else
1231 		inode->i_flags &= ~S_SYNC;
1232 	if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1233 		inode->i_flags |= S_NOATIME;
1234 	else
1235 		inode->i_flags &= ~S_NOATIME;
1236 }
1237 
1238 /*
1239  * Initialize the Linux inode, set up the operation vectors and
1240  * unlock the inode.
1241  *
1242  * When reading existing inodes from disk this is called directly
1243  * from xfs_iget, when creating a new inode it is called from
1244  * xfs_ialloc after setting up the inode.
1245  *
1246  * We are always called with an uninitialised linux inode here.
1247  * We need to initialise the necessary fields and take a reference
1248  * on it.
1249  */
1250 void
1251 xfs_setup_inode(
1252 	struct xfs_inode	*ip)
1253 {
1254 	struct inode		*inode = &ip->i_vnode;
1255 	gfp_t			gfp_mask;
1256 
1257 	inode->i_ino = ip->i_ino;
1258 	inode->i_state = I_NEW;
1259 
1260 	inode_sb_list_add(inode);
1261 	/* make the inode look hashed for the writeback code */
1262 	hlist_add_fake(&inode->i_hash);
1263 
1264 	inode->i_mode	= ip->i_d.di_mode;
1265 	set_nlink(inode, ip->i_d.di_nlink);
1266 	inode->i_uid    = xfs_uid_to_kuid(ip->i_d.di_uid);
1267 	inode->i_gid    = xfs_gid_to_kgid(ip->i_d.di_gid);
1268 
1269 	switch (inode->i_mode & S_IFMT) {
1270 	case S_IFBLK:
1271 	case S_IFCHR:
1272 		inode->i_rdev =
1273 			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1274 			      sysv_minor(ip->i_df.if_u2.if_rdev));
1275 		break;
1276 	default:
1277 		inode->i_rdev = 0;
1278 		break;
1279 	}
1280 
1281 	inode->i_generation = ip->i_d.di_gen;
1282 	i_size_write(inode, ip->i_d.di_size);
1283 	inode->i_atime.tv_sec	= ip->i_d.di_atime.t_sec;
1284 	inode->i_atime.tv_nsec	= ip->i_d.di_atime.t_nsec;
1285 	inode->i_mtime.tv_sec	= ip->i_d.di_mtime.t_sec;
1286 	inode->i_mtime.tv_nsec	= ip->i_d.di_mtime.t_nsec;
1287 	inode->i_ctime.tv_sec	= ip->i_d.di_ctime.t_sec;
1288 	inode->i_ctime.tv_nsec	= ip->i_d.di_ctime.t_nsec;
1289 	xfs_diflags_to_iflags(inode, ip);
1290 
1291 	ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1292 	lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1293 	switch (inode->i_mode & S_IFMT) {
1294 	case S_IFREG:
1295 		inode->i_op = &xfs_inode_operations;
1296 		inode->i_fop = &xfs_file_operations;
1297 		inode->i_mapping->a_ops = &xfs_address_space_operations;
1298 		break;
1299 	case S_IFDIR:
1300 		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1301 		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1302 			inode->i_op = &xfs_dir_ci_inode_operations;
1303 		else
1304 			inode->i_op = &xfs_dir_inode_operations;
1305 		inode->i_fop = &xfs_dir_file_operations;
1306 		ip->d_ops = ip->i_mount->m_dir_inode_ops;
1307 		break;
1308 	case S_IFLNK:
1309 		inode->i_op = &xfs_symlink_inode_operations;
1310 		if (!(ip->i_df.if_flags & XFS_IFINLINE))
1311 			inode->i_mapping->a_ops = &xfs_address_space_operations;
1312 		break;
1313 	default:
1314 		inode->i_op = &xfs_inode_operations;
1315 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1316 		break;
1317 	}
1318 
1319 	/*
1320 	 * Ensure all page cache allocations are done from GFP_NOFS context to
1321 	 * prevent direct reclaim recursion back into the filesystem and blowing
1322 	 * stacks or deadlocking.
1323 	 */
1324 	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1325 	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1326 
1327 	/*
1328 	 * If there is no attribute fork no ACL can exist on this inode,
1329 	 * and it can't have any file capabilities attached to it either.
1330 	 */
1331 	if (!XFS_IFORK_Q(ip)) {
1332 		inode_has_no_xattr(inode);
1333 		cache_no_acl(inode);
1334 	}
1335 
1336 	xfs_iflags_clear(ip, XFS_INEW);
1337 	barrier();
1338 
1339 	unlock_new_inode(inode);
1340 }
1341