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