xref: /openbmc/linux/fs/xfs/xfs_iops.c (revision 8e9356c6)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_inode.h"
29 #include "xfs_bmap.h"
30 #include "xfs_bmap_util.h"
31 #include "xfs_acl.h"
32 #include "xfs_quota.h"
33 #include "xfs_error.h"
34 #include "xfs_attr.h"
35 #include "xfs_trans.h"
36 #include "xfs_trace.h"
37 #include "xfs_icache.h"
38 #include "xfs_symlink.h"
39 #include "xfs_da_btree.h"
40 #include "xfs_dir2_priv.h"
41 #include "xfs_dinode.h"
42 
43 #include <linux/capability.h>
44 #include <linux/xattr.h>
45 #include <linux/namei.h>
46 #include <linux/posix_acl.h>
47 #include <linux/security.h>
48 #include <linux/fiemap.h>
49 #include <linux/slab.h>
50 
51 static int
52 xfs_initxattrs(
53 	struct inode		*inode,
54 	const struct xattr	*xattr_array,
55 	void			*fs_info)
56 {
57 	const struct xattr	*xattr;
58 	struct xfs_inode	*ip = XFS_I(inode);
59 	int			error = 0;
60 
61 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
62 		error = xfs_attr_set(ip, xattr->name, xattr->value,
63 				     xattr->value_len, ATTR_SECURE);
64 		if (error < 0)
65 			break;
66 	}
67 	return error;
68 }
69 
70 /*
71  * Hook in SELinux.  This is not quite correct yet, what we really need
72  * here (as we do for default ACLs) is a mechanism by which creation of
73  * these attrs can be journalled at inode creation time (along with the
74  * inode, of course, such that log replay can't cause these to be lost).
75  */
76 
77 STATIC int
78 xfs_init_security(
79 	struct inode	*inode,
80 	struct inode	*dir,
81 	const struct qstr *qstr)
82 {
83 	return security_inode_init_security(inode, dir, qstr,
84 					    &xfs_initxattrs, NULL);
85 }
86 
87 static void
88 xfs_dentry_to_name(
89 	struct xfs_name	*namep,
90 	struct dentry	*dentry,
91 	int		mode)
92 {
93 	namep->name = dentry->d_name.name;
94 	namep->len = dentry->d_name.len;
95 	namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
96 }
97 
98 STATIC void
99 xfs_cleanup_inode(
100 	struct inode	*dir,
101 	struct inode	*inode,
102 	struct dentry	*dentry)
103 {
104 	struct xfs_name	teardown;
105 
106 	/* Oh, the horror.
107 	 * If we can't add the ACL or we fail in
108 	 * xfs_init_security we must back out.
109 	 * ENOSPC can hit here, among other things.
110 	 */
111 	xfs_dentry_to_name(&teardown, dentry, 0);
112 
113 	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
114 	iput(inode);
115 }
116 
117 STATIC int
118 xfs_vn_mknod(
119 	struct inode	*dir,
120 	struct dentry	*dentry,
121 	umode_t		mode,
122 	dev_t		rdev)
123 {
124 	struct inode	*inode;
125 	struct xfs_inode *ip = NULL;
126 	struct posix_acl *default_acl, *acl;
127 	struct xfs_name	name;
128 	int		error;
129 
130 	/*
131 	 * Irix uses Missed'em'V split, but doesn't want to see
132 	 * the upper 5 bits of (14bit) major.
133 	 */
134 	if (S_ISCHR(mode) || S_ISBLK(mode)) {
135 		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
136 			return -EINVAL;
137 		rdev = sysv_encode_dev(rdev);
138 	} else {
139 		rdev = 0;
140 	}
141 
142 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
143 	if (error)
144 		return error;
145 
146 	xfs_dentry_to_name(&name, dentry, mode);
147 	error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
148 	if (unlikely(error))
149 		goto out_free_acl;
150 
151 	inode = VFS_I(ip);
152 
153 	error = xfs_init_security(inode, dir, &dentry->d_name);
154 	if (unlikely(error))
155 		goto out_cleanup_inode;
156 
157 #ifdef CONFIG_XFS_POSIX_ACL
158 	if (default_acl) {
159 		error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
160 		if (error)
161 			goto out_cleanup_inode;
162 	}
163 	if (acl) {
164 		error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
165 		if (error)
166 			goto out_cleanup_inode;
167 	}
168 #endif
169 
170 	d_instantiate(dentry, inode);
171  out_free_acl:
172 	if (default_acl)
173 		posix_acl_release(default_acl);
174 	if (acl)
175 		posix_acl_release(acl);
176 	return -error;
177 
178  out_cleanup_inode:
179 	xfs_cleanup_inode(dir, inode, dentry);
180 	goto out_free_acl;
181 }
182 
183 STATIC int
184 xfs_vn_create(
185 	struct inode	*dir,
186 	struct dentry	*dentry,
187 	umode_t		mode,
188 	bool		flags)
189 {
190 	return xfs_vn_mknod(dir, dentry, mode, 0);
191 }
192 
193 STATIC int
194 xfs_vn_mkdir(
195 	struct inode	*dir,
196 	struct dentry	*dentry,
197 	umode_t		mode)
198 {
199 	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
200 }
201 
202 STATIC struct dentry *
203 xfs_vn_lookup(
204 	struct inode	*dir,
205 	struct dentry	*dentry,
206 	unsigned int flags)
207 {
208 	struct xfs_inode *cip;
209 	struct xfs_name	name;
210 	int		error;
211 
212 	if (dentry->d_name.len >= MAXNAMELEN)
213 		return ERR_PTR(-ENAMETOOLONG);
214 
215 	xfs_dentry_to_name(&name, dentry, 0);
216 	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
217 	if (unlikely(error)) {
218 		if (unlikely(error != ENOENT))
219 			return ERR_PTR(-error);
220 		d_add(dentry, NULL);
221 		return NULL;
222 	}
223 
224 	return d_splice_alias(VFS_I(cip), dentry);
225 }
226 
227 STATIC struct dentry *
228 xfs_vn_ci_lookup(
229 	struct inode	*dir,
230 	struct dentry	*dentry,
231 	unsigned int flags)
232 {
233 	struct xfs_inode *ip;
234 	struct xfs_name	xname;
235 	struct xfs_name ci_name;
236 	struct qstr	dname;
237 	int		error;
238 
239 	if (dentry->d_name.len >= MAXNAMELEN)
240 		return ERR_PTR(-ENAMETOOLONG);
241 
242 	xfs_dentry_to_name(&xname, dentry, 0);
243 	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
244 	if (unlikely(error)) {
245 		if (unlikely(error != ENOENT))
246 			return ERR_PTR(-error);
247 		/*
248 		 * call d_add(dentry, NULL) here when d_drop_negative_children
249 		 * is called in xfs_vn_mknod (ie. allow negative dentries
250 		 * with CI filesystems).
251 		 */
252 		return NULL;
253 	}
254 
255 	/* if exact match, just splice and exit */
256 	if (!ci_name.name)
257 		return d_splice_alias(VFS_I(ip), dentry);
258 
259 	/* else case-insensitive match... */
260 	dname.name = ci_name.name;
261 	dname.len = ci_name.len;
262 	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
263 	kmem_free(ci_name.name);
264 	return dentry;
265 }
266 
267 STATIC int
268 xfs_vn_link(
269 	struct dentry	*old_dentry,
270 	struct inode	*dir,
271 	struct dentry	*dentry)
272 {
273 	struct inode	*inode = old_dentry->d_inode;
274 	struct xfs_name	name;
275 	int		error;
276 
277 	xfs_dentry_to_name(&name, dentry, inode->i_mode);
278 
279 	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
280 	if (unlikely(error))
281 		return -error;
282 
283 	ihold(inode);
284 	d_instantiate(dentry, inode);
285 	return 0;
286 }
287 
288 STATIC int
289 xfs_vn_unlink(
290 	struct inode	*dir,
291 	struct dentry	*dentry)
292 {
293 	struct xfs_name	name;
294 	int		error;
295 
296 	xfs_dentry_to_name(&name, dentry, 0);
297 
298 	error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
299 	if (error)
300 		return error;
301 
302 	/*
303 	 * With unlink, the VFS makes the dentry "negative": no inode,
304 	 * but still hashed. This is incompatible with case-insensitive
305 	 * mode, so invalidate (unhash) the dentry in CI-mode.
306 	 */
307 	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
308 		d_invalidate(dentry);
309 	return 0;
310 }
311 
312 STATIC int
313 xfs_vn_symlink(
314 	struct inode	*dir,
315 	struct dentry	*dentry,
316 	const char	*symname)
317 {
318 	struct inode	*inode;
319 	struct xfs_inode *cip = NULL;
320 	struct xfs_name	name;
321 	int		error;
322 	umode_t		mode;
323 
324 	mode = S_IFLNK |
325 		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
326 	xfs_dentry_to_name(&name, dentry, mode);
327 
328 	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
329 	if (unlikely(error))
330 		goto out;
331 
332 	inode = VFS_I(cip);
333 
334 	error = xfs_init_security(inode, dir, &dentry->d_name);
335 	if (unlikely(error))
336 		goto out_cleanup_inode;
337 
338 	d_instantiate(dentry, inode);
339 	return 0;
340 
341  out_cleanup_inode:
342 	xfs_cleanup_inode(dir, inode, dentry);
343  out:
344 	return -error;
345 }
346 
347 STATIC int
348 xfs_vn_rename(
349 	struct inode	*odir,
350 	struct dentry	*odentry,
351 	struct inode	*ndir,
352 	struct dentry	*ndentry)
353 {
354 	struct inode	*new_inode = ndentry->d_inode;
355 	struct xfs_name	oname;
356 	struct xfs_name	nname;
357 
358 	xfs_dentry_to_name(&oname, odentry, 0);
359 	xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode);
360 
361 	return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
362 			   XFS_I(ndir), &nname, new_inode ?
363 						XFS_I(new_inode) : NULL);
364 }
365 
366 /*
367  * careful here - this function can get called recursively, so
368  * we need to be very careful about how much stack we use.
369  * uio is kmalloced for this reason...
370  */
371 STATIC void *
372 xfs_vn_follow_link(
373 	struct dentry		*dentry,
374 	struct nameidata	*nd)
375 {
376 	char			*link;
377 	int			error = -ENOMEM;
378 
379 	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
380 	if (!link)
381 		goto out_err;
382 
383 	error = -xfs_readlink(XFS_I(dentry->d_inode), link);
384 	if (unlikely(error))
385 		goto out_kfree;
386 
387 	nd_set_link(nd, link);
388 	return NULL;
389 
390  out_kfree:
391 	kfree(link);
392  out_err:
393 	nd_set_link(nd, ERR_PTR(error));
394 	return NULL;
395 }
396 
397 STATIC int
398 xfs_vn_getattr(
399 	struct vfsmount		*mnt,
400 	struct dentry		*dentry,
401 	struct kstat		*stat)
402 {
403 	struct inode		*inode = dentry->d_inode;
404 	struct xfs_inode	*ip = XFS_I(inode);
405 	struct xfs_mount	*mp = ip->i_mount;
406 
407 	trace_xfs_getattr(ip);
408 
409 	if (XFS_FORCED_SHUTDOWN(mp))
410 		return -XFS_ERROR(EIO);
411 
412 	stat->size = XFS_ISIZE(ip);
413 	stat->dev = inode->i_sb->s_dev;
414 	stat->mode = ip->i_d.di_mode;
415 	stat->nlink = ip->i_d.di_nlink;
416 	stat->uid = inode->i_uid;
417 	stat->gid = inode->i_gid;
418 	stat->ino = ip->i_ino;
419 	stat->atime = inode->i_atime;
420 	stat->mtime = inode->i_mtime;
421 	stat->ctime = inode->i_ctime;
422 	stat->blocks =
423 		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
424 
425 
426 	switch (inode->i_mode & S_IFMT) {
427 	case S_IFBLK:
428 	case S_IFCHR:
429 		stat->blksize = BLKDEV_IOSIZE;
430 		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
431 				   sysv_minor(ip->i_df.if_u2.if_rdev));
432 		break;
433 	default:
434 		if (XFS_IS_REALTIME_INODE(ip)) {
435 			/*
436 			 * If the file blocks are being allocated from a
437 			 * realtime volume, then return the inode's realtime
438 			 * extent size or the realtime volume's extent size.
439 			 */
440 			stat->blksize =
441 				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
442 		} else
443 			stat->blksize = xfs_preferred_iosize(mp);
444 		stat->rdev = 0;
445 		break;
446 	}
447 
448 	return 0;
449 }
450 
451 static void
452 xfs_setattr_mode(
453 	struct xfs_inode	*ip,
454 	struct iattr		*iattr)
455 {
456 	struct inode		*inode = VFS_I(ip);
457 	umode_t			mode = iattr->ia_mode;
458 
459 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
460 
461 	ip->i_d.di_mode &= S_IFMT;
462 	ip->i_d.di_mode |= mode & ~S_IFMT;
463 
464 	inode->i_mode &= S_IFMT;
465 	inode->i_mode |= mode & ~S_IFMT;
466 }
467 
468 static void
469 xfs_setattr_time(
470 	struct xfs_inode	*ip,
471 	struct iattr		*iattr)
472 {
473 	struct inode		*inode = VFS_I(ip);
474 
475 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
476 
477 	if (iattr->ia_valid & ATTR_ATIME) {
478 		inode->i_atime = iattr->ia_atime;
479 		ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
480 		ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
481 	}
482 	if (iattr->ia_valid & ATTR_CTIME) {
483 		inode->i_ctime = iattr->ia_ctime;
484 		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
485 		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
486 	}
487 	if (iattr->ia_valid & ATTR_MTIME) {
488 		inode->i_mtime = iattr->ia_mtime;
489 		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
490 		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
491 	}
492 }
493 
494 int
495 xfs_setattr_nonsize(
496 	struct xfs_inode	*ip,
497 	struct iattr		*iattr,
498 	int			flags)
499 {
500 	xfs_mount_t		*mp = ip->i_mount;
501 	struct inode		*inode = VFS_I(ip);
502 	int			mask = iattr->ia_valid;
503 	xfs_trans_t		*tp;
504 	int			error;
505 	kuid_t			uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
506 	kgid_t			gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
507 	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
508 	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
509 
510 	trace_xfs_setattr(ip);
511 
512 	/* If acls are being inherited, we already have this checked */
513 	if (!(flags & XFS_ATTR_NOACL)) {
514 		if (mp->m_flags & XFS_MOUNT_RDONLY)
515 			return XFS_ERROR(EROFS);
516 
517 		if (XFS_FORCED_SHUTDOWN(mp))
518 			return XFS_ERROR(EIO);
519 
520 		error = -inode_change_ok(inode, iattr);
521 		if (error)
522 			return XFS_ERROR(error);
523 	}
524 
525 	ASSERT((mask & ATTR_SIZE) == 0);
526 
527 	/*
528 	 * If disk quotas is on, we make sure that the dquots do exist on disk,
529 	 * before we start any other transactions. Trying to do this later
530 	 * is messy. We don't care to take a readlock to look at the ids
531 	 * in inode here, because we can't hold it across the trans_reserve.
532 	 * If the IDs do change before we take the ilock, we're covered
533 	 * because the i_*dquot fields will get updated anyway.
534 	 */
535 	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
536 		uint	qflags = 0;
537 
538 		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
539 			uid = iattr->ia_uid;
540 			qflags |= XFS_QMOPT_UQUOTA;
541 		} else {
542 			uid = inode->i_uid;
543 		}
544 		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
545 			gid = iattr->ia_gid;
546 			qflags |= XFS_QMOPT_GQUOTA;
547 		}  else {
548 			gid = inode->i_gid;
549 		}
550 
551 		/*
552 		 * We take a reference when we initialize udqp and gdqp,
553 		 * so it is important that we never blindly double trip on
554 		 * the same variable. See xfs_create() for an example.
555 		 */
556 		ASSERT(udqp == NULL);
557 		ASSERT(gdqp == NULL);
558 		error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
559 					   xfs_kgid_to_gid(gid),
560 					   xfs_get_projid(ip),
561 					   qflags, &udqp, &gdqp, NULL);
562 		if (error)
563 			return error;
564 	}
565 
566 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
567 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
568 	if (error)
569 		goto out_dqrele;
570 
571 	xfs_ilock(ip, XFS_ILOCK_EXCL);
572 
573 	/*
574 	 * Change file ownership.  Must be the owner or privileged.
575 	 */
576 	if (mask & (ATTR_UID|ATTR_GID)) {
577 		/*
578 		 * These IDs could have changed since we last looked at them.
579 		 * But, we're assured that if the ownership did change
580 		 * while we didn't have the inode locked, inode's dquot(s)
581 		 * would have changed also.
582 		 */
583 		iuid = inode->i_uid;
584 		igid = inode->i_gid;
585 		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
586 		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
587 
588 		/*
589 		 * Do a quota reservation only if uid/gid is actually
590 		 * going to change.
591 		 */
592 		if (XFS_IS_QUOTA_RUNNING(mp) &&
593 		    ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
594 		     (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
595 			ASSERT(tp);
596 			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
597 						NULL, capable(CAP_FOWNER) ?
598 						XFS_QMOPT_FORCE_RES : 0);
599 			if (error)	/* out of quota */
600 				goto out_trans_cancel;
601 		}
602 	}
603 
604 	xfs_trans_ijoin(tp, ip, 0);
605 
606 	/*
607 	 * Change file ownership.  Must be the owner or privileged.
608 	 */
609 	if (mask & (ATTR_UID|ATTR_GID)) {
610 		/*
611 		 * CAP_FSETID overrides the following restrictions:
612 		 *
613 		 * The set-user-ID and set-group-ID bits of a file will be
614 		 * cleared upon successful return from chown()
615 		 */
616 		if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
617 		    !capable(CAP_FSETID))
618 			ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
619 
620 		/*
621 		 * Change the ownerships and register quota modifications
622 		 * in the transaction.
623 		 */
624 		if (!uid_eq(iuid, uid)) {
625 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
626 				ASSERT(mask & ATTR_UID);
627 				ASSERT(udqp);
628 				olddquot1 = xfs_qm_vop_chown(tp, ip,
629 							&ip->i_udquot, udqp);
630 			}
631 			ip->i_d.di_uid = xfs_kuid_to_uid(uid);
632 			inode->i_uid = uid;
633 		}
634 		if (!gid_eq(igid, gid)) {
635 			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
636 				ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
637 				       !XFS_IS_PQUOTA_ON(mp));
638 				ASSERT(mask & ATTR_GID);
639 				ASSERT(gdqp);
640 				olddquot2 = xfs_qm_vop_chown(tp, ip,
641 							&ip->i_gdquot, gdqp);
642 			}
643 			ip->i_d.di_gid = xfs_kgid_to_gid(gid);
644 			inode->i_gid = gid;
645 		}
646 	}
647 
648 	if (mask & ATTR_MODE)
649 		xfs_setattr_mode(ip, iattr);
650 	if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
651 		xfs_setattr_time(ip, iattr);
652 
653 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
654 
655 	XFS_STATS_INC(xs_ig_attrchg);
656 
657 	if (mp->m_flags & XFS_MOUNT_WSYNC)
658 		xfs_trans_set_sync(tp);
659 	error = xfs_trans_commit(tp, 0);
660 
661 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
662 
663 	/*
664 	 * Release any dquot(s) the inode had kept before chown.
665 	 */
666 	xfs_qm_dqrele(olddquot1);
667 	xfs_qm_dqrele(olddquot2);
668 	xfs_qm_dqrele(udqp);
669 	xfs_qm_dqrele(gdqp);
670 
671 	if (error)
672 		return XFS_ERROR(error);
673 
674 	/*
675 	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
676 	 * 	     update.  We could avoid this with linked transactions
677 	 * 	     and passing down the transaction pointer all the way
678 	 *	     to attr_set.  No previous user of the generic
679 	 * 	     Posix ACL code seems to care about this issue either.
680 	 */
681 	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
682 		error = -posix_acl_chmod(inode, inode->i_mode);
683 		if (error)
684 			return XFS_ERROR(error);
685 	}
686 
687 	return 0;
688 
689 out_trans_cancel:
690 	xfs_trans_cancel(tp, 0);
691 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
692 out_dqrele:
693 	xfs_qm_dqrele(udqp);
694 	xfs_qm_dqrele(gdqp);
695 	return error;
696 }
697 
698 /*
699  * Truncate file.  Must have write permission and not be a directory.
700  */
701 int
702 xfs_setattr_size(
703 	struct xfs_inode	*ip,
704 	struct iattr		*iattr)
705 {
706 	struct xfs_mount	*mp = ip->i_mount;
707 	struct inode		*inode = VFS_I(ip);
708 	int			mask = iattr->ia_valid;
709 	xfs_off_t		oldsize, newsize;
710 	struct xfs_trans	*tp;
711 	int			error;
712 	uint			lock_flags = 0;
713 	uint			commit_flags = 0;
714 
715 	trace_xfs_setattr(ip);
716 
717 	if (mp->m_flags & XFS_MOUNT_RDONLY)
718 		return XFS_ERROR(EROFS);
719 
720 	if (XFS_FORCED_SHUTDOWN(mp))
721 		return XFS_ERROR(EIO);
722 
723 	error = -inode_change_ok(inode, iattr);
724 	if (error)
725 		return XFS_ERROR(error);
726 
727 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
728 	ASSERT(S_ISREG(ip->i_d.di_mode));
729 	ASSERT((mask & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
730 			ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
731 
732 	oldsize = inode->i_size;
733 	newsize = iattr->ia_size;
734 
735 	/*
736 	 * Short circuit the truncate case for zero length files.
737 	 */
738 	if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
739 		if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
740 			return 0;
741 
742 		/*
743 		 * Use the regular setattr path to update the timestamps.
744 		 */
745 		iattr->ia_valid &= ~ATTR_SIZE;
746 		return xfs_setattr_nonsize(ip, iattr, 0);
747 	}
748 
749 	/*
750 	 * Make sure that the dquots are attached to the inode.
751 	 */
752 	error = xfs_qm_dqattach(ip, 0);
753 	if (error)
754 		return error;
755 
756 	/*
757 	 * Now we can make the changes.  Before we join the inode to the
758 	 * transaction, take care of the part of the truncation that must be
759 	 * done without the inode lock.  This needs to be done before joining
760 	 * the inode to the transaction, because the inode cannot be unlocked
761 	 * once it is a part of the transaction.
762 	 */
763 	if (newsize > oldsize) {
764 		/*
765 		 * Do the first part of growing a file: zero any data in the
766 		 * last block that is beyond the old EOF.  We need to do this
767 		 * before the inode is joined to the transaction to modify
768 		 * i_size.
769 		 */
770 		error = xfs_zero_eof(ip, newsize, oldsize);
771 		if (error)
772 			return error;
773 	}
774 
775 	/*
776 	 * We are going to log the inode size change in this transaction so
777 	 * any previous writes that are beyond the on disk EOF and the new
778 	 * EOF that have not been written out need to be written here.  If we
779 	 * do not write the data out, we expose ourselves to the null files
780 	 * problem.
781 	 *
782 	 * Only flush from the on disk size to the smaller of the in memory
783 	 * file size or the new size as that's the range we really care about
784 	 * here and prevents waiting for other data not within the range we
785 	 * care about here.
786 	 */
787 	if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
788 		error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
789 						      ip->i_d.di_size, newsize);
790 		if (error)
791 			return error;
792 	}
793 
794 	/*
795 	 * Wait for all direct I/O to complete.
796 	 */
797 	inode_dio_wait(inode);
798 
799 	error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
800 	if (error)
801 		return error;
802 
803 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
804 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
805 	if (error)
806 		goto out_trans_cancel;
807 
808 	truncate_setsize(inode, newsize);
809 
810 	commit_flags = XFS_TRANS_RELEASE_LOG_RES;
811 	lock_flags |= XFS_ILOCK_EXCL;
812 
813 	xfs_ilock(ip, XFS_ILOCK_EXCL);
814 
815 	xfs_trans_ijoin(tp, ip, 0);
816 
817 	/*
818 	 * Only change the c/mtime if we are changing the size or we are
819 	 * explicitly asked to change it.  This handles the semantic difference
820 	 * between truncate() and ftruncate() as implemented in the VFS.
821 	 *
822 	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
823 	 * special case where we need to update the times despite not having
824 	 * these flags set.  For all other operations the VFS set these flags
825 	 * explicitly if it wants a timestamp update.
826 	 */
827 	if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
828 		iattr->ia_ctime = iattr->ia_mtime =
829 			current_fs_time(inode->i_sb);
830 		mask |= ATTR_CTIME | ATTR_MTIME;
831 	}
832 
833 	/*
834 	 * The first thing we do is set the size to new_size permanently on
835 	 * disk.  This way we don't have to worry about anyone ever being able
836 	 * to look at the data being freed even in the face of a crash.
837 	 * What we're getting around here is the case where we free a block, it
838 	 * is allocated to another file, it is written to, and then we crash.
839 	 * If the new data gets written to the file but the log buffers
840 	 * containing the free and reallocation don't, then we'd end up with
841 	 * garbage in the blocks being freed.  As long as we make the new size
842 	 * permanent before actually freeing any blocks it doesn't matter if
843 	 * they get written to.
844 	 */
845 	ip->i_d.di_size = newsize;
846 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
847 
848 	if (newsize <= oldsize) {
849 		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
850 		if (error)
851 			goto out_trans_abort;
852 
853 		/*
854 		 * Truncated "down", so we're removing references to old data
855 		 * here - if we delay flushing for a long time, we expose
856 		 * ourselves unduly to the notorious NULL files problem.  So,
857 		 * we mark this inode and flush it when the file is closed,
858 		 * and do not wait the usual (long) time for writeout.
859 		 */
860 		xfs_iflags_set(ip, XFS_ITRUNCATED);
861 
862 		/* A truncate down always removes post-EOF blocks. */
863 		xfs_inode_clear_eofblocks_tag(ip);
864 	}
865 
866 	if (mask & ATTR_MODE)
867 		xfs_setattr_mode(ip, iattr);
868 	if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
869 		xfs_setattr_time(ip, iattr);
870 
871 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
872 
873 	XFS_STATS_INC(xs_ig_attrchg);
874 
875 	if (mp->m_flags & XFS_MOUNT_WSYNC)
876 		xfs_trans_set_sync(tp);
877 
878 	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
879 out_unlock:
880 	if (lock_flags)
881 		xfs_iunlock(ip, lock_flags);
882 	return error;
883 
884 out_trans_abort:
885 	commit_flags |= XFS_TRANS_ABORT;
886 out_trans_cancel:
887 	xfs_trans_cancel(tp, commit_flags);
888 	goto out_unlock;
889 }
890 
891 STATIC int
892 xfs_vn_setattr(
893 	struct dentry		*dentry,
894 	struct iattr		*iattr)
895 {
896 	struct xfs_inode	*ip = XFS_I(dentry->d_inode);
897 	int			error;
898 
899 	if (iattr->ia_valid & ATTR_SIZE) {
900 		xfs_ilock(ip, XFS_IOLOCK_EXCL);
901 		error = xfs_setattr_size(ip, iattr);
902 		xfs_iunlock(ip, XFS_IOLOCK_EXCL);
903 	} else {
904 		error = xfs_setattr_nonsize(ip, iattr, 0);
905 	}
906 
907 	return -error;
908 }
909 
910 STATIC int
911 xfs_vn_update_time(
912 	struct inode		*inode,
913 	struct timespec		*now,
914 	int			flags)
915 {
916 	struct xfs_inode	*ip = XFS_I(inode);
917 	struct xfs_mount	*mp = ip->i_mount;
918 	struct xfs_trans	*tp;
919 	int			error;
920 
921 	trace_xfs_update_time(ip);
922 
923 	tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
924 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
925 	if (error) {
926 		xfs_trans_cancel(tp, 0);
927 		return -error;
928 	}
929 
930 	xfs_ilock(ip, XFS_ILOCK_EXCL);
931 	if (flags & S_CTIME) {
932 		inode->i_ctime = *now;
933 		ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec;
934 		ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec;
935 	}
936 	if (flags & S_MTIME) {
937 		inode->i_mtime = *now;
938 		ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec;
939 		ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec;
940 	}
941 	if (flags & S_ATIME) {
942 		inode->i_atime = *now;
943 		ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec;
944 		ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec;
945 	}
946 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
947 	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
948 	return -xfs_trans_commit(tp, 0);
949 }
950 
951 #define XFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
952 
953 /*
954  * Call fiemap helper to fill in user data.
955  * Returns positive errors to xfs_getbmap.
956  */
957 STATIC int
958 xfs_fiemap_format(
959 	void			**arg,
960 	struct getbmapx		*bmv,
961 	int			*full)
962 {
963 	int			error;
964 	struct fiemap_extent_info *fieinfo = *arg;
965 	u32			fiemap_flags = 0;
966 	u64			logical, physical, length;
967 
968 	/* Do nothing for a hole */
969 	if (bmv->bmv_block == -1LL)
970 		return 0;
971 
972 	logical = BBTOB(bmv->bmv_offset);
973 	physical = BBTOB(bmv->bmv_block);
974 	length = BBTOB(bmv->bmv_length);
975 
976 	if (bmv->bmv_oflags & BMV_OF_PREALLOC)
977 		fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
978 	else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
979 		fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
980 				 FIEMAP_EXTENT_UNKNOWN);
981 		physical = 0;   /* no block yet */
982 	}
983 	if (bmv->bmv_oflags & BMV_OF_LAST)
984 		fiemap_flags |= FIEMAP_EXTENT_LAST;
985 
986 	error = fiemap_fill_next_extent(fieinfo, logical, physical,
987 					length, fiemap_flags);
988 	if (error > 0) {
989 		error = 0;
990 		*full = 1;	/* user array now full */
991 	}
992 
993 	return -error;
994 }
995 
996 STATIC int
997 xfs_vn_fiemap(
998 	struct inode		*inode,
999 	struct fiemap_extent_info *fieinfo,
1000 	u64			start,
1001 	u64			length)
1002 {
1003 	xfs_inode_t		*ip = XFS_I(inode);
1004 	struct getbmapx		bm;
1005 	int			error;
1006 
1007 	error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1008 	if (error)
1009 		return error;
1010 
1011 	/* Set up bmap header for xfs internal routine */
1012 	bm.bmv_offset = BTOBB(start);
1013 	/* Special case for whole file */
1014 	if (length == FIEMAP_MAX_OFFSET)
1015 		bm.bmv_length = -1LL;
1016 	else
1017 		bm.bmv_length = BTOBB(length);
1018 
1019 	/* We add one because in getbmap world count includes the header */
1020 	bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1021 					fieinfo->fi_extents_max + 1;
1022 	bm.bmv_count = min_t(__s32, bm.bmv_count,
1023 			     (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1024 	bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1025 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1026 		bm.bmv_iflags |= BMV_IF_ATTRFORK;
1027 	if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1028 		bm.bmv_iflags |= BMV_IF_DELALLOC;
1029 
1030 	error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1031 	if (error)
1032 		return -error;
1033 
1034 	return 0;
1035 }
1036 
1037 static const struct inode_operations xfs_inode_operations = {
1038 	.get_acl		= xfs_get_acl,
1039 	.set_acl		= xfs_set_acl,
1040 	.getattr		= xfs_vn_getattr,
1041 	.setattr		= xfs_vn_setattr,
1042 	.setxattr		= generic_setxattr,
1043 	.getxattr		= generic_getxattr,
1044 	.removexattr		= generic_removexattr,
1045 	.listxattr		= xfs_vn_listxattr,
1046 	.fiemap			= xfs_vn_fiemap,
1047 	.update_time		= xfs_vn_update_time,
1048 };
1049 
1050 static const struct inode_operations xfs_dir_inode_operations = {
1051 	.create			= xfs_vn_create,
1052 	.lookup			= xfs_vn_lookup,
1053 	.link			= xfs_vn_link,
1054 	.unlink			= xfs_vn_unlink,
1055 	.symlink		= xfs_vn_symlink,
1056 	.mkdir			= xfs_vn_mkdir,
1057 	/*
1058 	 * Yes, XFS uses the same method for rmdir and unlink.
1059 	 *
1060 	 * There are some subtile differences deeper in the code,
1061 	 * but we use S_ISDIR to check for those.
1062 	 */
1063 	.rmdir			= xfs_vn_unlink,
1064 	.mknod			= xfs_vn_mknod,
1065 	.rename			= xfs_vn_rename,
1066 	.get_acl		= xfs_get_acl,
1067 	.set_acl		= xfs_set_acl,
1068 	.getattr		= xfs_vn_getattr,
1069 	.setattr		= xfs_vn_setattr,
1070 	.setxattr		= generic_setxattr,
1071 	.getxattr		= generic_getxattr,
1072 	.removexattr		= generic_removexattr,
1073 	.listxattr		= xfs_vn_listxattr,
1074 	.update_time		= xfs_vn_update_time,
1075 };
1076 
1077 static const struct inode_operations xfs_dir_ci_inode_operations = {
1078 	.create			= xfs_vn_create,
1079 	.lookup			= xfs_vn_ci_lookup,
1080 	.link			= xfs_vn_link,
1081 	.unlink			= xfs_vn_unlink,
1082 	.symlink		= xfs_vn_symlink,
1083 	.mkdir			= xfs_vn_mkdir,
1084 	/*
1085 	 * Yes, XFS uses the same method for rmdir and unlink.
1086 	 *
1087 	 * There are some subtile differences deeper in the code,
1088 	 * but we use S_ISDIR to check for those.
1089 	 */
1090 	.rmdir			= xfs_vn_unlink,
1091 	.mknod			= xfs_vn_mknod,
1092 	.rename			= xfs_vn_rename,
1093 	.get_acl		= xfs_get_acl,
1094 	.set_acl		= xfs_set_acl,
1095 	.getattr		= xfs_vn_getattr,
1096 	.setattr		= xfs_vn_setattr,
1097 	.setxattr		= generic_setxattr,
1098 	.getxattr		= generic_getxattr,
1099 	.removexattr		= generic_removexattr,
1100 	.listxattr		= xfs_vn_listxattr,
1101 	.update_time		= xfs_vn_update_time,
1102 };
1103 
1104 static const struct inode_operations xfs_symlink_inode_operations = {
1105 	.readlink		= generic_readlink,
1106 	.follow_link		= xfs_vn_follow_link,
1107 	.put_link		= kfree_put_link,
1108 	.getattr		= xfs_vn_getattr,
1109 	.setattr		= xfs_vn_setattr,
1110 	.setxattr		= generic_setxattr,
1111 	.getxattr		= generic_getxattr,
1112 	.removexattr		= generic_removexattr,
1113 	.listxattr		= xfs_vn_listxattr,
1114 	.update_time		= xfs_vn_update_time,
1115 };
1116 
1117 STATIC void
1118 xfs_diflags_to_iflags(
1119 	struct inode		*inode,
1120 	struct xfs_inode	*ip)
1121 {
1122 	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1123 		inode->i_flags |= S_IMMUTABLE;
1124 	else
1125 		inode->i_flags &= ~S_IMMUTABLE;
1126 	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1127 		inode->i_flags |= S_APPEND;
1128 	else
1129 		inode->i_flags &= ~S_APPEND;
1130 	if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1131 		inode->i_flags |= S_SYNC;
1132 	else
1133 		inode->i_flags &= ~S_SYNC;
1134 	if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1135 		inode->i_flags |= S_NOATIME;
1136 	else
1137 		inode->i_flags &= ~S_NOATIME;
1138 }
1139 
1140 /*
1141  * Initialize the Linux inode, set up the operation vectors and
1142  * unlock the inode.
1143  *
1144  * When reading existing inodes from disk this is called directly
1145  * from xfs_iget, when creating a new inode it is called from
1146  * xfs_ialloc after setting up the inode.
1147  *
1148  * We are always called with an uninitialised linux inode here.
1149  * We need to initialise the necessary fields and take a reference
1150  * on it.
1151  */
1152 void
1153 xfs_setup_inode(
1154 	struct xfs_inode	*ip)
1155 {
1156 	struct inode		*inode = &ip->i_vnode;
1157 	gfp_t			gfp_mask;
1158 
1159 	inode->i_ino = ip->i_ino;
1160 	inode->i_state = I_NEW;
1161 
1162 	inode_sb_list_add(inode);
1163 	/* make the inode look hashed for the writeback code */
1164 	hlist_add_fake(&inode->i_hash);
1165 
1166 	inode->i_mode	= ip->i_d.di_mode;
1167 	set_nlink(inode, ip->i_d.di_nlink);
1168 	inode->i_uid    = xfs_uid_to_kuid(ip->i_d.di_uid);
1169 	inode->i_gid    = xfs_gid_to_kgid(ip->i_d.di_gid);
1170 
1171 	switch (inode->i_mode & S_IFMT) {
1172 	case S_IFBLK:
1173 	case S_IFCHR:
1174 		inode->i_rdev =
1175 			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1176 			      sysv_minor(ip->i_df.if_u2.if_rdev));
1177 		break;
1178 	default:
1179 		inode->i_rdev = 0;
1180 		break;
1181 	}
1182 
1183 	inode->i_generation = ip->i_d.di_gen;
1184 	i_size_write(inode, ip->i_d.di_size);
1185 	inode->i_atime.tv_sec	= ip->i_d.di_atime.t_sec;
1186 	inode->i_atime.tv_nsec	= ip->i_d.di_atime.t_nsec;
1187 	inode->i_mtime.tv_sec	= ip->i_d.di_mtime.t_sec;
1188 	inode->i_mtime.tv_nsec	= ip->i_d.di_mtime.t_nsec;
1189 	inode->i_ctime.tv_sec	= ip->i_d.di_ctime.t_sec;
1190 	inode->i_ctime.tv_nsec	= ip->i_d.di_ctime.t_nsec;
1191 	xfs_diflags_to_iflags(inode, ip);
1192 
1193 	ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1194 	switch (inode->i_mode & S_IFMT) {
1195 	case S_IFREG:
1196 		inode->i_op = &xfs_inode_operations;
1197 		inode->i_fop = &xfs_file_operations;
1198 		inode->i_mapping->a_ops = &xfs_address_space_operations;
1199 		break;
1200 	case S_IFDIR:
1201 		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1202 			inode->i_op = &xfs_dir_ci_inode_operations;
1203 		else
1204 			inode->i_op = &xfs_dir_inode_operations;
1205 		inode->i_fop = &xfs_dir_file_operations;
1206 		ip->d_ops = ip->i_mount->m_dir_inode_ops;
1207 		break;
1208 	case S_IFLNK:
1209 		inode->i_op = &xfs_symlink_inode_operations;
1210 		if (!(ip->i_df.if_flags & XFS_IFINLINE))
1211 			inode->i_mapping->a_ops = &xfs_address_space_operations;
1212 		break;
1213 	default:
1214 		inode->i_op = &xfs_inode_operations;
1215 		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1216 		break;
1217 	}
1218 
1219 	/*
1220 	 * Ensure all page cache allocations are done from GFP_NOFS context to
1221 	 * prevent direct reclaim recursion back into the filesystem and blowing
1222 	 * stacks or deadlocking.
1223 	 */
1224 	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1225 	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1226 
1227 	/*
1228 	 * If there is no attribute fork no ACL can exist on this inode,
1229 	 * and it can't have any file capabilities attached to it either.
1230 	 */
1231 	if (!XFS_IFORK_Q(ip)) {
1232 		inode_has_no_xattr(inode);
1233 		cache_no_acl(inode);
1234 	}
1235 
1236 	xfs_iflags_clear(ip, XFS_INEW);
1237 	barrier();
1238 
1239 	unlock_new_inode(inode);
1240 }
1241