xref: /linux/fs/nfsd/vfs.c (revision 5f4123be3cdb1dbd77fa9d6d2bb96bb9689a0a19)
1 #define MSNFS	/* HACK HACK */
2 /*
3  * linux/fs/nfsd/vfs.c
4  *
5  * File operations used by nfsd. Some of these have been ripped from
6  * other parts of the kernel because they weren't exported, others
7  * are partial duplicates with added or changed functionality.
8  *
9  * Note that several functions dget() the dentry upon which they want
10  * to act, most notably those that create directory entries. Response
11  * dentry's are dput()'d if necessary in the release callback.
12  * So if you notice code paths that apparently fail to dput() the
13  * dentry, don't worry--they have been taken care of.
14  *
15  * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16  * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17  */
18 
19 #include <linux/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/fs.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/splice.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
34 #include <linux/in.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
41 #ifdef CONFIG_NFSD_V3
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
51 #ifdef CONFIG_NFSD_V4
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
57 #include <linux/jhash.h>
58 
59 #include <asm/uaccess.h>
60 
61 #define NFSDDBG_FACILITY		NFSDDBG_FILEOP
62 
63 
64 /*
65  * This is a cache of readahead params that help us choose the proper
66  * readahead strategy. Initially, we set all readahead parameters to 0
67  * and let the VFS handle things.
68  * If you increase the number of cached files very much, you'll need to
69  * add a hash table here.
70  */
71 struct raparms {
72 	struct raparms		*p_next;
73 	unsigned int		p_count;
74 	ino_t			p_ino;
75 	dev_t			p_dev;
76 	int			p_set;
77 	struct file_ra_state	p_ra;
78 	unsigned int		p_hindex;
79 };
80 
81 struct raparm_hbucket {
82 	struct raparms		*pb_head;
83 	spinlock_t		pb_lock;
84 } ____cacheline_aligned_in_smp;
85 
86 #define RAPARM_HASH_BITS	4
87 #define RAPARM_HASH_SIZE	(1<<RAPARM_HASH_BITS)
88 #define RAPARM_HASH_MASK	(RAPARM_HASH_SIZE-1)
89 static struct raparm_hbucket	raparm_hash[RAPARM_HASH_SIZE];
90 
91 /*
92  * Called from nfsd_lookup and encode_dirent. Check if we have crossed
93  * a mount point.
94  * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
95  *  or nfs_ok having possibly changed *dpp and *expp
96  */
97 int
98 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
99 		        struct svc_export **expp)
100 {
101 	struct svc_export *exp = *expp, *exp2 = NULL;
102 	struct dentry *dentry = *dpp;
103 	struct vfsmount *mnt = mntget(exp->ex_path.mnt);
104 	struct dentry *mounts = dget(dentry);
105 	int err = 0;
106 
107 	while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
108 
109 	exp2 = rqst_exp_get_by_name(rqstp, mnt, mounts);
110 	if (IS_ERR(exp2)) {
111 		if (PTR_ERR(exp2) != -ENOENT)
112 			err = PTR_ERR(exp2);
113 		dput(mounts);
114 		mntput(mnt);
115 		goto out;
116 	}
117 	if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
118 		/* successfully crossed mount point */
119 		exp_put(exp);
120 		*expp = exp2;
121 		dput(dentry);
122 		*dpp = mounts;
123 	} else {
124 		exp_put(exp2);
125 		dput(mounts);
126 	}
127 	mntput(mnt);
128 out:
129 	return err;
130 }
131 
132 __be32
133 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
134 		   const char *name, unsigned int len,
135 		   struct svc_export **exp_ret, struct dentry **dentry_ret)
136 {
137 	struct svc_export	*exp;
138 	struct dentry		*dparent;
139 	struct dentry		*dentry;
140 	__be32			err;
141 	int			host_err;
142 
143 	dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
144 
145 	/* Obtain dentry and export. */
146 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
147 	if (err)
148 		return err;
149 
150 	dparent = fhp->fh_dentry;
151 	exp  = fhp->fh_export;
152 	exp_get(exp);
153 
154 	/* Lookup the name, but don't follow links */
155 	if (isdotent(name, len)) {
156 		if (len==1)
157 			dentry = dget(dparent);
158 		else if (dparent != exp->ex_path.dentry)
159 			dentry = dget_parent(dparent);
160 		else if (!EX_NOHIDE(exp))
161 			dentry = dget(dparent); /* .. == . just like at / */
162 		else {
163 			/* checking mountpoint crossing is very different when stepping up */
164 			struct svc_export *exp2 = NULL;
165 			struct dentry *dp;
166 			struct vfsmount *mnt = mntget(exp->ex_path.mnt);
167 			dentry = dget(dparent);
168 			while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
169 				;
170 			dp = dget_parent(dentry);
171 			dput(dentry);
172 			dentry = dp;
173 
174 			exp2 = rqst_exp_parent(rqstp, mnt, dentry);
175 			if (PTR_ERR(exp2) == -ENOENT) {
176 				dput(dentry);
177 				dentry = dget(dparent);
178 			} else if (IS_ERR(exp2)) {
179 				host_err = PTR_ERR(exp2);
180 				dput(dentry);
181 				mntput(mnt);
182 				goto out_nfserr;
183 			} else {
184 				exp_put(exp);
185 				exp = exp2;
186 			}
187 			mntput(mnt);
188 		}
189 	} else {
190 		fh_lock(fhp);
191 		dentry = lookup_one_len(name, dparent, len);
192 		host_err = PTR_ERR(dentry);
193 		if (IS_ERR(dentry))
194 			goto out_nfserr;
195 		/*
196 		 * check if we have crossed a mount point ...
197 		 */
198 		if (d_mountpoint(dentry)) {
199 			if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
200 				dput(dentry);
201 				goto out_nfserr;
202 			}
203 		}
204 	}
205 	*dentry_ret = dentry;
206 	*exp_ret = exp;
207 	return 0;
208 
209 out_nfserr:
210 	exp_put(exp);
211 	return nfserrno(host_err);
212 }
213 
214 /*
215  * Look up one component of a pathname.
216  * N.B. After this call _both_ fhp and resfh need an fh_put
217  *
218  * If the lookup would cross a mountpoint, and the mounted filesystem
219  * is exported to the client with NFSEXP_NOHIDE, then the lookup is
220  * accepted as it stands and the mounted directory is
221  * returned. Otherwise the covered directory is returned.
222  * NOTE: this mountpoint crossing is not supported properly by all
223  *   clients and is explicitly disallowed for NFSv3
224  *      NeilBrown <neilb@cse.unsw.edu.au>
225  */
226 __be32
227 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
228 				unsigned int len, struct svc_fh *resfh)
229 {
230 	struct svc_export	*exp;
231 	struct dentry		*dentry;
232 	__be32 err;
233 
234 	err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
235 	if (err)
236 		return err;
237 	err = check_nfsd_access(exp, rqstp);
238 	if (err)
239 		goto out;
240 	/*
241 	 * Note: we compose the file handle now, but as the
242 	 * dentry may be negative, it may need to be updated.
243 	 */
244 	err = fh_compose(resfh, exp, dentry, fhp);
245 	if (!err && !dentry->d_inode)
246 		err = nfserr_noent;
247 out:
248 	dput(dentry);
249 	exp_put(exp);
250 	return err;
251 }
252 
253 
254 /*
255  * Set various file attributes.
256  * N.B. After this call fhp needs an fh_put
257  */
258 __be32
259 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
260 	     int check_guard, time_t guardtime)
261 {
262 	struct dentry	*dentry;
263 	struct inode	*inode;
264 	int		accmode = NFSD_MAY_SATTR;
265 	int		ftype = 0;
266 	__be32		err;
267 	int		host_err;
268 	int		size_change = 0;
269 
270 	if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
271 		accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
272 	if (iap->ia_valid & ATTR_SIZE)
273 		ftype = S_IFREG;
274 
275 	/* Get inode */
276 	err = fh_verify(rqstp, fhp, ftype, accmode);
277 	if (err)
278 		goto out;
279 
280 	dentry = fhp->fh_dentry;
281 	inode = dentry->d_inode;
282 
283 	/* Ignore any mode updates on symlinks */
284 	if (S_ISLNK(inode->i_mode))
285 		iap->ia_valid &= ~ATTR_MODE;
286 
287 	if (!iap->ia_valid)
288 		goto out;
289 
290 	/*
291 	 * NFSv2 does not differentiate between "set-[ac]time-to-now"
292 	 * which only requires access, and "set-[ac]time-to-X" which
293 	 * requires ownership.
294 	 * So if it looks like it might be "set both to the same time which
295 	 * is close to now", and if inode_change_ok fails, then we
296 	 * convert to "set to now" instead of "set to explicit time"
297 	 *
298 	 * We only call inode_change_ok as the last test as technically
299 	 * it is not an interface that we should be using.  It is only
300 	 * valid if the filesystem does not define it's own i_op->setattr.
301 	 */
302 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
303 #define	MAX_TOUCH_TIME_ERROR (30*60)
304 	if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
305 	    iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
306 		/*
307 		 * Looks probable.
308 		 *
309 		 * Now just make sure time is in the right ballpark.
310 		 * Solaris, at least, doesn't seem to care what the time
311 		 * request is.  We require it be within 30 minutes of now.
312 		 */
313 		time_t delta = iap->ia_atime.tv_sec - get_seconds();
314 		if (delta < 0)
315 			delta = -delta;
316 		if (delta < MAX_TOUCH_TIME_ERROR &&
317 		    inode_change_ok(inode, iap) != 0) {
318 			/*
319 			 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
320 			 * This will cause notify_change to set these times
321 			 * to "now"
322 			 */
323 			iap->ia_valid &= ~BOTH_TIME_SET;
324 		}
325 	}
326 
327 	/*
328 	 * The size case is special.
329 	 * It changes the file as well as the attributes.
330 	 */
331 	if (iap->ia_valid & ATTR_SIZE) {
332 		if (iap->ia_size < inode->i_size) {
333 			err = nfsd_permission(rqstp, fhp->fh_export, dentry,
334 					NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
335 			if (err)
336 				goto out;
337 		}
338 
339 		/*
340 		 * If we are changing the size of the file, then
341 		 * we need to break all leases.
342 		 */
343 		host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
344 		if (host_err == -EWOULDBLOCK)
345 			host_err = -ETIMEDOUT;
346 		if (host_err) /* ENOMEM or EWOULDBLOCK */
347 			goto out_nfserr;
348 
349 		host_err = get_write_access(inode);
350 		if (host_err)
351 			goto out_nfserr;
352 
353 		size_change = 1;
354 		host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
355 		if (host_err) {
356 			put_write_access(inode);
357 			goto out_nfserr;
358 		}
359 		DQUOT_INIT(inode);
360 	}
361 
362 	/* sanitize the mode change */
363 	if (iap->ia_valid & ATTR_MODE) {
364 		iap->ia_mode &= S_IALLUGO;
365 		iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
366 	}
367 
368 	/* Revoke setuid/setgid on chown */
369 	if (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
370 	    ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)) {
371 		iap->ia_valid |= ATTR_KILL_PRIV;
372 		if (iap->ia_valid & ATTR_MODE) {
373 			/* we're setting mode too, just clear the s*id bits */
374 			iap->ia_mode &= ~S_ISUID;
375 			if (iap->ia_mode & S_IXGRP)
376 				iap->ia_mode &= ~S_ISGID;
377 		} else {
378 			/* set ATTR_KILL_* bits and let VFS handle it */
379 			iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
380 		}
381 	}
382 
383 	/* Change the attributes. */
384 
385 	iap->ia_valid |= ATTR_CTIME;
386 
387 	err = nfserr_notsync;
388 	if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
389 		fh_lock(fhp);
390 		host_err = notify_change(dentry, iap);
391 		err = nfserrno(host_err);
392 		fh_unlock(fhp);
393 	}
394 	if (size_change)
395 		put_write_access(inode);
396 	if (!err)
397 		if (EX_ISSYNC(fhp->fh_export))
398 			write_inode_now(inode, 1);
399 out:
400 	return err;
401 
402 out_nfserr:
403 	err = nfserrno(host_err);
404 	goto out;
405 }
406 
407 #if defined(CONFIG_NFSD_V2_ACL) || \
408     defined(CONFIG_NFSD_V3_ACL) || \
409     defined(CONFIG_NFSD_V4)
410 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
411 {
412 	ssize_t buflen;
413 
414 	buflen = vfs_getxattr(dentry, key, NULL, 0);
415 	if (buflen <= 0)
416 		return buflen;
417 
418 	*buf = kmalloc(buflen, GFP_KERNEL);
419 	if (!*buf)
420 		return -ENOMEM;
421 
422 	return vfs_getxattr(dentry, key, *buf, buflen);
423 }
424 #endif
425 
426 #if defined(CONFIG_NFSD_V4)
427 static int
428 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
429 {
430 	int len;
431 	size_t buflen;
432 	char *buf = NULL;
433 	int error = 0;
434 
435 	buflen = posix_acl_xattr_size(pacl->a_count);
436 	buf = kmalloc(buflen, GFP_KERNEL);
437 	error = -ENOMEM;
438 	if (buf == NULL)
439 		goto out;
440 
441 	len = posix_acl_to_xattr(pacl, buf, buflen);
442 	if (len < 0) {
443 		error = len;
444 		goto out;
445 	}
446 
447 	error = vfs_setxattr(dentry, key, buf, len, 0);
448 out:
449 	kfree(buf);
450 	return error;
451 }
452 
453 __be32
454 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
455     struct nfs4_acl *acl)
456 {
457 	__be32 error;
458 	int host_error;
459 	struct dentry *dentry;
460 	struct inode *inode;
461 	struct posix_acl *pacl = NULL, *dpacl = NULL;
462 	unsigned int flags = 0;
463 
464 	/* Get inode */
465 	error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
466 	if (error)
467 		return error;
468 
469 	dentry = fhp->fh_dentry;
470 	inode = dentry->d_inode;
471 	if (S_ISDIR(inode->i_mode))
472 		flags = NFS4_ACL_DIR;
473 
474 	host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
475 	if (host_error == -EINVAL) {
476 		return nfserr_attrnotsupp;
477 	} else if (host_error < 0)
478 		goto out_nfserr;
479 
480 	host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
481 	if (host_error < 0)
482 		goto out_release;
483 
484 	if (S_ISDIR(inode->i_mode))
485 		host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
486 
487 out_release:
488 	posix_acl_release(pacl);
489 	posix_acl_release(dpacl);
490 out_nfserr:
491 	if (host_error == -EOPNOTSUPP)
492 		return nfserr_attrnotsupp;
493 	else
494 		return nfserrno(host_error);
495 }
496 
497 static struct posix_acl *
498 _get_posix_acl(struct dentry *dentry, char *key)
499 {
500 	void *buf = NULL;
501 	struct posix_acl *pacl = NULL;
502 	int buflen;
503 
504 	buflen = nfsd_getxattr(dentry, key, &buf);
505 	if (!buflen)
506 		buflen = -ENODATA;
507 	if (buflen <= 0)
508 		return ERR_PTR(buflen);
509 
510 	pacl = posix_acl_from_xattr(buf, buflen);
511 	kfree(buf);
512 	return pacl;
513 }
514 
515 int
516 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
517 {
518 	struct inode *inode = dentry->d_inode;
519 	int error = 0;
520 	struct posix_acl *pacl = NULL, *dpacl = NULL;
521 	unsigned int flags = 0;
522 
523 	pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
524 	if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
525 		pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
526 	if (IS_ERR(pacl)) {
527 		error = PTR_ERR(pacl);
528 		pacl = NULL;
529 		goto out;
530 	}
531 
532 	if (S_ISDIR(inode->i_mode)) {
533 		dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
534 		if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
535 			dpacl = NULL;
536 		else if (IS_ERR(dpacl)) {
537 			error = PTR_ERR(dpacl);
538 			dpacl = NULL;
539 			goto out;
540 		}
541 		flags = NFS4_ACL_DIR;
542 	}
543 
544 	*acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
545 	if (IS_ERR(*acl)) {
546 		error = PTR_ERR(*acl);
547 		*acl = NULL;
548 	}
549  out:
550 	posix_acl_release(pacl);
551 	posix_acl_release(dpacl);
552 	return error;
553 }
554 
555 #endif /* defined(CONFIG_NFS_V4) */
556 
557 #ifdef CONFIG_NFSD_V3
558 /*
559  * Check server access rights to a file system object
560  */
561 struct accessmap {
562 	u32		access;
563 	int		how;
564 };
565 static struct accessmap	nfs3_regaccess[] = {
566     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
567     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
568     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_TRUNC	},
569     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE			},
570 
571     {	0,			0				}
572 };
573 
574 static struct accessmap	nfs3_diraccess[] = {
575     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
576     {	NFS3_ACCESS_LOOKUP,	NFSD_MAY_EXEC			},
577     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
578     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_EXEC|NFSD_MAY_WRITE	},
579     {	NFS3_ACCESS_DELETE,	NFSD_MAY_REMOVE			},
580 
581     {	0,			0				}
582 };
583 
584 static struct accessmap	nfs3_anyaccess[] = {
585 	/* Some clients - Solaris 2.6 at least, make an access call
586 	 * to the server to check for access for things like /dev/null
587 	 * (which really, the server doesn't care about).  So
588 	 * We provide simple access checking for them, looking
589 	 * mainly at mode bits, and we make sure to ignore read-only
590 	 * filesystem checks
591 	 */
592     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
593     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
594     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
595     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
596 
597     {	0,			0				}
598 };
599 
600 __be32
601 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
602 {
603 	struct accessmap	*map;
604 	struct svc_export	*export;
605 	struct dentry		*dentry;
606 	u32			query, result = 0, sresult = 0;
607 	__be32			error;
608 
609 	error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
610 	if (error)
611 		goto out;
612 
613 	export = fhp->fh_export;
614 	dentry = fhp->fh_dentry;
615 
616 	if (S_ISREG(dentry->d_inode->i_mode))
617 		map = nfs3_regaccess;
618 	else if (S_ISDIR(dentry->d_inode->i_mode))
619 		map = nfs3_diraccess;
620 	else
621 		map = nfs3_anyaccess;
622 
623 
624 	query = *access;
625 	for  (; map->access; map++) {
626 		if (map->access & query) {
627 			__be32 err2;
628 
629 			sresult |= map->access;
630 
631 			err2 = nfsd_permission(rqstp, export, dentry, map->how);
632 			switch (err2) {
633 			case nfs_ok:
634 				result |= map->access;
635 				break;
636 
637 			/* the following error codes just mean the access was not allowed,
638 			 * rather than an error occurred */
639 			case nfserr_rofs:
640 			case nfserr_acces:
641 			case nfserr_perm:
642 				/* simply don't "or" in the access bit. */
643 				break;
644 			default:
645 				error = err2;
646 				goto out;
647 			}
648 		}
649 	}
650 	*access = result;
651 	if (supported)
652 		*supported = sresult;
653 
654  out:
655 	return error;
656 }
657 #endif /* CONFIG_NFSD_V3 */
658 
659 
660 
661 /*
662  * Open an existing file or directory.
663  * The access argument indicates the type of open (read/write/lock)
664  * N.B. After this call fhp needs an fh_put
665  */
666 __be32
667 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
668 			int access, struct file **filp)
669 {
670 	struct dentry	*dentry;
671 	struct inode	*inode;
672 	int		flags = O_RDONLY|O_LARGEFILE;
673 	__be32		err;
674 	int		host_err;
675 
676 	/*
677 	 * If we get here, then the client has already done an "open",
678 	 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
679 	 * in case a chmod has now revoked permission.
680 	 */
681 	err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
682 	if (err)
683 		goto out;
684 
685 	dentry = fhp->fh_dentry;
686 	inode = dentry->d_inode;
687 
688 	/* Disallow write access to files with the append-only bit set
689 	 * or any access when mandatory locking enabled
690 	 */
691 	err = nfserr_perm;
692 	if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
693 		goto out;
694 	/*
695 	 * We must ignore files (but only files) which might have mandatory
696 	 * locks on them because there is no way to know if the accesser has
697 	 * the lock.
698 	 */
699 	if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
700 		goto out;
701 
702 	if (!inode->i_fop)
703 		goto out;
704 
705 	/*
706 	 * Check to see if there are any leases on this file.
707 	 * This may block while leases are broken.
708 	 */
709 	host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
710 	if (host_err == -EWOULDBLOCK)
711 		host_err = -ETIMEDOUT;
712 	if (host_err) /* NOMEM or WOULDBLOCK */
713 		goto out_nfserr;
714 
715 	if (access & NFSD_MAY_WRITE) {
716 		if (access & NFSD_MAY_READ)
717 			flags = O_RDWR|O_LARGEFILE;
718 		else
719 			flags = O_WRONLY|O_LARGEFILE;
720 
721 		DQUOT_INIT(inode);
722 	}
723 	*filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
724 				flags);
725 	if (IS_ERR(*filp))
726 		host_err = PTR_ERR(*filp);
727 out_nfserr:
728 	err = nfserrno(host_err);
729 out:
730 	return err;
731 }
732 
733 /*
734  * Close a file.
735  */
736 void
737 nfsd_close(struct file *filp)
738 {
739 	fput(filp);
740 }
741 
742 /*
743  * Sync a file
744  * As this calls fsync (not fdatasync) there is no need for a write_inode
745  * after it.
746  */
747 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
748 			      const struct file_operations *fop)
749 {
750 	struct inode *inode = dp->d_inode;
751 	int (*fsync) (struct file *, struct dentry *, int);
752 	int err;
753 
754 	err = filemap_fdatawrite(inode->i_mapping);
755 	if (err == 0 && fop && (fsync = fop->fsync))
756 		err = fsync(filp, dp, 0);
757 	if (err == 0)
758 		err = filemap_fdatawait(inode->i_mapping);
759 
760 	return err;
761 }
762 
763 
764 static int
765 nfsd_sync(struct file *filp)
766 {
767         int err;
768 	struct inode *inode = filp->f_path.dentry->d_inode;
769 	dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
770 	mutex_lock(&inode->i_mutex);
771 	err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
772 	mutex_unlock(&inode->i_mutex);
773 
774 	return err;
775 }
776 
777 int
778 nfsd_sync_dir(struct dentry *dp)
779 {
780 	return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
781 }
782 
783 /*
784  * Obtain the readahead parameters for the file
785  * specified by (dev, ino).
786  */
787 
788 static inline struct raparms *
789 nfsd_get_raparms(dev_t dev, ino_t ino)
790 {
791 	struct raparms	*ra, **rap, **frap = NULL;
792 	int depth = 0;
793 	unsigned int hash;
794 	struct raparm_hbucket *rab;
795 
796 	hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
797 	rab = &raparm_hash[hash];
798 
799 	spin_lock(&rab->pb_lock);
800 	for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
801 		if (ra->p_ino == ino && ra->p_dev == dev)
802 			goto found;
803 		depth++;
804 		if (ra->p_count == 0)
805 			frap = rap;
806 	}
807 	depth = nfsdstats.ra_size*11/10;
808 	if (!frap) {
809 		spin_unlock(&rab->pb_lock);
810 		return NULL;
811 	}
812 	rap = frap;
813 	ra = *frap;
814 	ra->p_dev = dev;
815 	ra->p_ino = ino;
816 	ra->p_set = 0;
817 	ra->p_hindex = hash;
818 found:
819 	if (rap != &rab->pb_head) {
820 		*rap = ra->p_next;
821 		ra->p_next   = rab->pb_head;
822 		rab->pb_head = ra;
823 	}
824 	ra->p_count++;
825 	nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
826 	spin_unlock(&rab->pb_lock);
827 	return ra;
828 }
829 
830 /*
831  * Grab and keep cached pages associated with a file in the svc_rqst
832  * so that they can be passed to the network sendmsg/sendpage routines
833  * directly. They will be released after the sending has completed.
834  */
835 static int
836 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
837 		  struct splice_desc *sd)
838 {
839 	struct svc_rqst *rqstp = sd->u.data;
840 	struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
841 	struct page *page = buf->page;
842 	size_t size;
843 	int ret;
844 
845 	ret = buf->ops->confirm(pipe, buf);
846 	if (unlikely(ret))
847 		return ret;
848 
849 	size = sd->len;
850 
851 	if (rqstp->rq_res.page_len == 0) {
852 		get_page(page);
853 		put_page(*pp);
854 		*pp = page;
855 		rqstp->rq_resused++;
856 		rqstp->rq_res.page_base = buf->offset;
857 		rqstp->rq_res.page_len = size;
858 	} else if (page != pp[-1]) {
859 		get_page(page);
860 		if (*pp)
861 			put_page(*pp);
862 		*pp = page;
863 		rqstp->rq_resused++;
864 		rqstp->rq_res.page_len += size;
865 	} else
866 		rqstp->rq_res.page_len += size;
867 
868 	return size;
869 }
870 
871 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
872 				    struct splice_desc *sd)
873 {
874 	return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
875 }
876 
877 static inline int svc_msnfs(struct svc_fh *ffhp)
878 {
879 #ifdef MSNFS
880 	return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
881 #else
882 	return 0;
883 #endif
884 }
885 
886 static __be32
887 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
888               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
889 {
890 	struct inode *inode;
891 	struct raparms	*ra;
892 	mm_segment_t	oldfs;
893 	__be32		err;
894 	int		host_err;
895 
896 	err = nfserr_perm;
897 	inode = file->f_path.dentry->d_inode;
898 
899 	if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
900 		goto out;
901 
902 	/* Get readahead parameters */
903 	ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
904 
905 	if (ra && ra->p_set)
906 		file->f_ra = ra->p_ra;
907 
908 	if (file->f_op->splice_read && rqstp->rq_splice_ok) {
909 		struct splice_desc sd = {
910 			.len		= 0,
911 			.total_len	= *count,
912 			.pos		= offset,
913 			.u.data		= rqstp,
914 		};
915 
916 		rqstp->rq_resused = 1;
917 		host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
918 	} else {
919 		oldfs = get_fs();
920 		set_fs(KERNEL_DS);
921 		host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
922 		set_fs(oldfs);
923 	}
924 
925 	/* Write back readahead params */
926 	if (ra) {
927 		struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
928 		spin_lock(&rab->pb_lock);
929 		ra->p_ra = file->f_ra;
930 		ra->p_set = 1;
931 		ra->p_count--;
932 		spin_unlock(&rab->pb_lock);
933 	}
934 
935 	if (host_err >= 0) {
936 		nfsdstats.io_read += host_err;
937 		*count = host_err;
938 		err = 0;
939 		fsnotify_access(file->f_path.dentry);
940 	} else
941 		err = nfserrno(host_err);
942 out:
943 	return err;
944 }
945 
946 static void kill_suid(struct dentry *dentry)
947 {
948 	struct iattr	ia;
949 	ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
950 
951 	mutex_lock(&dentry->d_inode->i_mutex);
952 	notify_change(dentry, &ia);
953 	mutex_unlock(&dentry->d_inode->i_mutex);
954 }
955 
956 static __be32
957 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
958 				loff_t offset, struct kvec *vec, int vlen,
959 	   			unsigned long cnt, int *stablep)
960 {
961 	struct svc_export	*exp;
962 	struct dentry		*dentry;
963 	struct inode		*inode;
964 	mm_segment_t		oldfs;
965 	__be32			err = 0;
966 	int			host_err;
967 	int			stable = *stablep;
968 
969 #ifdef MSNFS
970 	err = nfserr_perm;
971 
972 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
973 		(!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
974 		goto out;
975 #endif
976 
977 	dentry = file->f_path.dentry;
978 	inode = dentry->d_inode;
979 	exp   = fhp->fh_export;
980 
981 	/*
982 	 * Request sync writes if
983 	 *  -	the sync export option has been set, or
984 	 *  -	the client requested O_SYNC behavior (NFSv3 feature).
985 	 *  -   The file system doesn't support fsync().
986 	 * When gathered writes have been configured for this volume,
987 	 * flushing the data to disk is handled separately below.
988 	 */
989 
990 	if (!file->f_op->fsync) {/* COMMIT3 cannot work */
991 	       stable = 2;
992 	       *stablep = 2; /* FILE_SYNC */
993 	}
994 
995 	if (!EX_ISSYNC(exp))
996 		stable = 0;
997 	if (stable && !EX_WGATHER(exp))
998 		file->f_flags |= O_SYNC;
999 
1000 	/* Write the data. */
1001 	oldfs = get_fs(); set_fs(KERNEL_DS);
1002 	host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1003 	set_fs(oldfs);
1004 	if (host_err >= 0) {
1005 		nfsdstats.io_write += cnt;
1006 		fsnotify_modify(file->f_path.dentry);
1007 	}
1008 
1009 	/* clear setuid/setgid flag after write */
1010 	if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
1011 		kill_suid(dentry);
1012 
1013 	if (host_err >= 0 && stable) {
1014 		static ino_t	last_ino;
1015 		static dev_t	last_dev;
1016 
1017 		/*
1018 		 * Gathered writes: If another process is currently
1019 		 * writing to the file, there's a high chance
1020 		 * this is another nfsd (triggered by a bulk write
1021 		 * from a client's biod). Rather than syncing the
1022 		 * file with each write request, we sleep for 10 msec.
1023 		 *
1024 		 * I don't know if this roughly approximates
1025 		 * C. Juszak's idea of gathered writes, but it's a
1026 		 * nice and simple solution (IMHO), and it seems to
1027 		 * work:-)
1028 		 */
1029 		if (EX_WGATHER(exp)) {
1030 			if (atomic_read(&inode->i_writecount) > 1
1031 			    || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1032 				dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1033 				msleep(10);
1034 				dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1035 			}
1036 
1037 			if (inode->i_state & I_DIRTY) {
1038 				dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1039 				host_err=nfsd_sync(file);
1040 			}
1041 #if 0
1042 			wake_up(&inode->i_wait);
1043 #endif
1044 		}
1045 		last_ino = inode->i_ino;
1046 		last_dev = inode->i_sb->s_dev;
1047 	}
1048 
1049 	dprintk("nfsd: write complete host_err=%d\n", host_err);
1050 	if (host_err >= 0)
1051 		err = 0;
1052 	else
1053 		err = nfserrno(host_err);
1054 out:
1055 	return err;
1056 }
1057 
1058 /*
1059  * Read data from a file. count must contain the requested read count
1060  * on entry. On return, *count contains the number of bytes actually read.
1061  * N.B. After this call fhp needs an fh_put
1062  */
1063 __be32
1064 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1065 		loff_t offset, struct kvec *vec, int vlen,
1066 		unsigned long *count)
1067 {
1068 	__be32		err;
1069 
1070 	if (file) {
1071 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1072 				NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1073 		if (err)
1074 			goto out;
1075 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1076 	} else {
1077 		err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1078 		if (err)
1079 			goto out;
1080 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1081 		nfsd_close(file);
1082 	}
1083 out:
1084 	return err;
1085 }
1086 
1087 /*
1088  * Write data to a file.
1089  * The stable flag requests synchronous writes.
1090  * N.B. After this call fhp needs an fh_put
1091  */
1092 __be32
1093 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1094 		loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1095 		int *stablep)
1096 {
1097 	__be32			err = 0;
1098 
1099 	if (file) {
1100 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1101 				NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1102 		if (err)
1103 			goto out;
1104 		err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1105 				stablep);
1106 	} else {
1107 		err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1108 		if (err)
1109 			goto out;
1110 
1111 		if (cnt)
1112 			err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1113 					     cnt, stablep);
1114 		nfsd_close(file);
1115 	}
1116 out:
1117 	return err;
1118 }
1119 
1120 #ifdef CONFIG_NFSD_V3
1121 /*
1122  * Commit all pending writes to stable storage.
1123  * Strictly speaking, we could sync just the indicated file region here,
1124  * but there's currently no way we can ask the VFS to do so.
1125  *
1126  * Unfortunately we cannot lock the file to make sure we return full WCC
1127  * data to the client, as locking happens lower down in the filesystem.
1128  */
1129 __be32
1130 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1131                loff_t offset, unsigned long count)
1132 {
1133 	struct file	*file;
1134 	__be32		err;
1135 
1136 	if ((u64)count > ~(u64)offset)
1137 		return nfserr_inval;
1138 
1139 	err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1140 	if (err)
1141 		return err;
1142 	if (EX_ISSYNC(fhp->fh_export)) {
1143 		if (file->f_op && file->f_op->fsync) {
1144 			err = nfserrno(nfsd_sync(file));
1145 		} else {
1146 			err = nfserr_notsupp;
1147 		}
1148 	}
1149 
1150 	nfsd_close(file);
1151 	return err;
1152 }
1153 #endif /* CONFIG_NFSD_V3 */
1154 
1155 static __be32
1156 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1157 			struct iattr *iap)
1158 {
1159 	/*
1160 	 * Mode has already been set earlier in create:
1161 	 */
1162 	iap->ia_valid &= ~ATTR_MODE;
1163 	/*
1164 	 * Setting uid/gid works only for root.  Irix appears to
1165 	 * send along the gid on create when it tries to implement
1166 	 * setgid directories via NFS:
1167 	 */
1168 	if (current->fsuid != 0)
1169 		iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1170 	if (iap->ia_valid)
1171 		return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1172 	return 0;
1173 }
1174 
1175 /*
1176  * Create a file (regular, directory, device, fifo); UNIX sockets
1177  * not yet implemented.
1178  * If the response fh has been verified, the parent directory should
1179  * already be locked. Note that the parent directory is left locked.
1180  *
1181  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1182  */
1183 __be32
1184 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1185 		char *fname, int flen, struct iattr *iap,
1186 		int type, dev_t rdev, struct svc_fh *resfhp)
1187 {
1188 	struct dentry	*dentry, *dchild = NULL;
1189 	struct inode	*dirp;
1190 	__be32		err;
1191 	__be32		err2;
1192 	int		host_err;
1193 
1194 	err = nfserr_perm;
1195 	if (!flen)
1196 		goto out;
1197 	err = nfserr_exist;
1198 	if (isdotent(fname, flen))
1199 		goto out;
1200 
1201 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1202 	if (err)
1203 		goto out;
1204 
1205 	dentry = fhp->fh_dentry;
1206 	dirp = dentry->d_inode;
1207 
1208 	err = nfserr_notdir;
1209 	if(!dirp->i_op || !dirp->i_op->lookup)
1210 		goto out;
1211 	/*
1212 	 * Check whether the response file handle has been verified yet.
1213 	 * If it has, the parent directory should already be locked.
1214 	 */
1215 	if (!resfhp->fh_dentry) {
1216 		/* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1217 		fh_lock_nested(fhp, I_MUTEX_PARENT);
1218 		dchild = lookup_one_len(fname, dentry, flen);
1219 		host_err = PTR_ERR(dchild);
1220 		if (IS_ERR(dchild))
1221 			goto out_nfserr;
1222 		err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1223 		if (err)
1224 			goto out;
1225 	} else {
1226 		/* called from nfsd_proc_create */
1227 		dchild = dget(resfhp->fh_dentry);
1228 		if (!fhp->fh_locked) {
1229 			/* not actually possible */
1230 			printk(KERN_ERR
1231 				"nfsd_create: parent %s/%s not locked!\n",
1232 				dentry->d_parent->d_name.name,
1233 				dentry->d_name.name);
1234 			err = nfserr_io;
1235 			goto out;
1236 		}
1237 	}
1238 	/*
1239 	 * Make sure the child dentry is still negative ...
1240 	 */
1241 	err = nfserr_exist;
1242 	if (dchild->d_inode) {
1243 		dprintk("nfsd_create: dentry %s/%s not negative!\n",
1244 			dentry->d_name.name, dchild->d_name.name);
1245 		goto out;
1246 	}
1247 
1248 	if (!(iap->ia_valid & ATTR_MODE))
1249 		iap->ia_mode = 0;
1250 	iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1251 
1252 	err = nfserr_inval;
1253 	if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1254 		printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1255 		       type);
1256 		goto out;
1257 	}
1258 
1259 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1260 	if (host_err)
1261 		goto out_nfserr;
1262 
1263 	/*
1264 	 * Get the dir op function pointer.
1265 	 */
1266 	err = 0;
1267 	switch (type) {
1268 	case S_IFREG:
1269 		host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1270 		break;
1271 	case S_IFDIR:
1272 		host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1273 		break;
1274 	case S_IFCHR:
1275 	case S_IFBLK:
1276 	case S_IFIFO:
1277 	case S_IFSOCK:
1278 		host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1279 		break;
1280 	}
1281 	if (host_err < 0) {
1282 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1283 		goto out_nfserr;
1284 	}
1285 
1286 	if (EX_ISSYNC(fhp->fh_export)) {
1287 		err = nfserrno(nfsd_sync_dir(dentry));
1288 		write_inode_now(dchild->d_inode, 1);
1289 	}
1290 
1291 	err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1292 	if (err2)
1293 		err = err2;
1294 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1295 	/*
1296 	 * Update the file handle to get the new inode info.
1297 	 */
1298 	if (!err)
1299 		err = fh_update(resfhp);
1300 out:
1301 	if (dchild && !IS_ERR(dchild))
1302 		dput(dchild);
1303 	return err;
1304 
1305 out_nfserr:
1306 	err = nfserrno(host_err);
1307 	goto out;
1308 }
1309 
1310 #ifdef CONFIG_NFSD_V3
1311 /*
1312  * NFSv3 version of nfsd_create
1313  */
1314 __be32
1315 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1316 		char *fname, int flen, struct iattr *iap,
1317 		struct svc_fh *resfhp, int createmode, u32 *verifier,
1318 	        int *truncp, int *created)
1319 {
1320 	struct dentry	*dentry, *dchild = NULL;
1321 	struct inode	*dirp;
1322 	__be32		err;
1323 	__be32		err2;
1324 	int		host_err;
1325 	__u32		v_mtime=0, v_atime=0;
1326 
1327 	err = nfserr_perm;
1328 	if (!flen)
1329 		goto out;
1330 	err = nfserr_exist;
1331 	if (isdotent(fname, flen))
1332 		goto out;
1333 	if (!(iap->ia_valid & ATTR_MODE))
1334 		iap->ia_mode = 0;
1335 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1336 	if (err)
1337 		goto out;
1338 
1339 	dentry = fhp->fh_dentry;
1340 	dirp = dentry->d_inode;
1341 
1342 	/* Get all the sanity checks out of the way before
1343 	 * we lock the parent. */
1344 	err = nfserr_notdir;
1345 	if(!dirp->i_op || !dirp->i_op->lookup)
1346 		goto out;
1347 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1348 
1349 	/*
1350 	 * Compose the response file handle.
1351 	 */
1352 	dchild = lookup_one_len(fname, dentry, flen);
1353 	host_err = PTR_ERR(dchild);
1354 	if (IS_ERR(dchild))
1355 		goto out_nfserr;
1356 
1357 	err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1358 	if (err)
1359 		goto out;
1360 
1361 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1362 		/* solaris7 gets confused (bugid 4218508) if these have
1363 		 * the high bit set, so just clear the high bits. If this is
1364 		 * ever changed to use different attrs for storing the
1365 		 * verifier, then do_open_lookup() will also need to be fixed
1366 		 * accordingly.
1367 		 */
1368 		v_mtime = verifier[0]&0x7fffffff;
1369 		v_atime = verifier[1]&0x7fffffff;
1370 	}
1371 
1372 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1373 	if (host_err)
1374 		goto out_nfserr;
1375 	if (dchild->d_inode) {
1376 		err = 0;
1377 
1378 		switch (createmode) {
1379 		case NFS3_CREATE_UNCHECKED:
1380 			if (! S_ISREG(dchild->d_inode->i_mode))
1381 				err = nfserr_exist;
1382 			else if (truncp) {
1383 				/* in nfsv4, we need to treat this case a little
1384 				 * differently.  we don't want to truncate the
1385 				 * file now; this would be wrong if the OPEN
1386 				 * fails for some other reason.  furthermore,
1387 				 * if the size is nonzero, we should ignore it
1388 				 * according to spec!
1389 				 */
1390 				*truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1391 			}
1392 			else {
1393 				iap->ia_valid &= ATTR_SIZE;
1394 				goto set_attr;
1395 			}
1396 			break;
1397 		case NFS3_CREATE_EXCLUSIVE:
1398 			if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1399 			    && dchild->d_inode->i_atime.tv_sec == v_atime
1400 			    && dchild->d_inode->i_size  == 0 )
1401 				break;
1402 			 /* fallthru */
1403 		case NFS3_CREATE_GUARDED:
1404 			err = nfserr_exist;
1405 		}
1406 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1407 		goto out;
1408 	}
1409 
1410 	host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1411 	if (host_err < 0) {
1412 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1413 		goto out_nfserr;
1414 	}
1415 	if (created)
1416 		*created = 1;
1417 
1418 	if (EX_ISSYNC(fhp->fh_export)) {
1419 		err = nfserrno(nfsd_sync_dir(dentry));
1420 		/* setattr will sync the child (or not) */
1421 	}
1422 
1423 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1424 		/* Cram the verifier into atime/mtime */
1425 		iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1426 			| ATTR_MTIME_SET|ATTR_ATIME_SET;
1427 		/* XXX someone who knows this better please fix it for nsec */
1428 		iap->ia_mtime.tv_sec = v_mtime;
1429 		iap->ia_atime.tv_sec = v_atime;
1430 		iap->ia_mtime.tv_nsec = 0;
1431 		iap->ia_atime.tv_nsec = 0;
1432 	}
1433 
1434  set_attr:
1435 	err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1436 	if (err2)
1437 		err = err2;
1438 
1439 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1440 	/*
1441 	 * Update the filehandle to get the new inode info.
1442 	 */
1443 	if (!err)
1444 		err = fh_update(resfhp);
1445 
1446  out:
1447 	fh_unlock(fhp);
1448 	if (dchild && !IS_ERR(dchild))
1449 		dput(dchild);
1450  	return err;
1451 
1452  out_nfserr:
1453 	err = nfserrno(host_err);
1454 	goto out;
1455 }
1456 #endif /* CONFIG_NFSD_V3 */
1457 
1458 /*
1459  * Read a symlink. On entry, *lenp must contain the maximum path length that
1460  * fits into the buffer. On return, it contains the true length.
1461  * N.B. After this call fhp needs an fh_put
1462  */
1463 __be32
1464 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1465 {
1466 	struct dentry	*dentry;
1467 	struct inode	*inode;
1468 	mm_segment_t	oldfs;
1469 	__be32		err;
1470 	int		host_err;
1471 
1472 	err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1473 	if (err)
1474 		goto out;
1475 
1476 	dentry = fhp->fh_dentry;
1477 	inode = dentry->d_inode;
1478 
1479 	err = nfserr_inval;
1480 	if (!inode->i_op || !inode->i_op->readlink)
1481 		goto out;
1482 
1483 	touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1484 	/* N.B. Why does this call need a get_fs()??
1485 	 * Remove the set_fs and watch the fireworks:-) --okir
1486 	 */
1487 
1488 	oldfs = get_fs(); set_fs(KERNEL_DS);
1489 	host_err = inode->i_op->readlink(dentry, buf, *lenp);
1490 	set_fs(oldfs);
1491 
1492 	if (host_err < 0)
1493 		goto out_nfserr;
1494 	*lenp = host_err;
1495 	err = 0;
1496 out:
1497 	return err;
1498 
1499 out_nfserr:
1500 	err = nfserrno(host_err);
1501 	goto out;
1502 }
1503 
1504 /*
1505  * Create a symlink and look up its inode
1506  * N.B. After this call _both_ fhp and resfhp need an fh_put
1507  */
1508 __be32
1509 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1510 				char *fname, int flen,
1511 				char *path,  int plen,
1512 				struct svc_fh *resfhp,
1513 				struct iattr *iap)
1514 {
1515 	struct dentry	*dentry, *dnew;
1516 	__be32		err, cerr;
1517 	int		host_err;
1518 
1519 	err = nfserr_noent;
1520 	if (!flen || !plen)
1521 		goto out;
1522 	err = nfserr_exist;
1523 	if (isdotent(fname, flen))
1524 		goto out;
1525 
1526 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1527 	if (err)
1528 		goto out;
1529 	fh_lock(fhp);
1530 	dentry = fhp->fh_dentry;
1531 	dnew = lookup_one_len(fname, dentry, flen);
1532 	host_err = PTR_ERR(dnew);
1533 	if (IS_ERR(dnew))
1534 		goto out_nfserr;
1535 
1536 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1537 	if (host_err)
1538 		goto out_nfserr;
1539 
1540 	if (unlikely(path[plen] != 0)) {
1541 		char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1542 		if (path_alloced == NULL)
1543 			host_err = -ENOMEM;
1544 		else {
1545 			strncpy(path_alloced, path, plen);
1546 			path_alloced[plen] = 0;
1547 			host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1548 			kfree(path_alloced);
1549 		}
1550 	} else
1551 		host_err = vfs_symlink(dentry->d_inode, dnew, path);
1552 
1553 	if (!host_err) {
1554 		if (EX_ISSYNC(fhp->fh_export))
1555 			host_err = nfsd_sync_dir(dentry);
1556 	}
1557 	err = nfserrno(host_err);
1558 	fh_unlock(fhp);
1559 
1560 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1561 
1562 	cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1563 	dput(dnew);
1564 	if (err==0) err = cerr;
1565 out:
1566 	return err;
1567 
1568 out_nfserr:
1569 	err = nfserrno(host_err);
1570 	goto out;
1571 }
1572 
1573 /*
1574  * Create a hardlink
1575  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1576  */
1577 __be32
1578 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1579 				char *name, int len, struct svc_fh *tfhp)
1580 {
1581 	struct dentry	*ddir, *dnew, *dold;
1582 	struct inode	*dirp, *dest;
1583 	__be32		err;
1584 	int		host_err;
1585 
1586 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1587 	if (err)
1588 		goto out;
1589 	err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1590 	if (err)
1591 		goto out;
1592 
1593 	err = nfserr_perm;
1594 	if (!len)
1595 		goto out;
1596 	err = nfserr_exist;
1597 	if (isdotent(name, len))
1598 		goto out;
1599 
1600 	fh_lock_nested(ffhp, I_MUTEX_PARENT);
1601 	ddir = ffhp->fh_dentry;
1602 	dirp = ddir->d_inode;
1603 
1604 	dnew = lookup_one_len(name, ddir, len);
1605 	host_err = PTR_ERR(dnew);
1606 	if (IS_ERR(dnew))
1607 		goto out_nfserr;
1608 
1609 	dold = tfhp->fh_dentry;
1610 	dest = dold->d_inode;
1611 
1612 	host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1613 	if (host_err) {
1614 		err = nfserrno(host_err);
1615 		goto out_dput;
1616 	}
1617 	host_err = vfs_link(dold, dirp, dnew);
1618 	if (!host_err) {
1619 		if (EX_ISSYNC(ffhp->fh_export)) {
1620 			err = nfserrno(nfsd_sync_dir(ddir));
1621 			write_inode_now(dest, 1);
1622 		}
1623 		err = 0;
1624 	} else {
1625 		if (host_err == -EXDEV && rqstp->rq_vers == 2)
1626 			err = nfserr_acces;
1627 		else
1628 			err = nfserrno(host_err);
1629 	}
1630 	mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1631 out_dput:
1632 	dput(dnew);
1633 out_unlock:
1634 	fh_unlock(ffhp);
1635 out:
1636 	return err;
1637 
1638 out_nfserr:
1639 	err = nfserrno(host_err);
1640 	goto out_unlock;
1641 }
1642 
1643 /*
1644  * Rename a file
1645  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1646  */
1647 __be32
1648 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1649 			    struct svc_fh *tfhp, char *tname, int tlen)
1650 {
1651 	struct dentry	*fdentry, *tdentry, *odentry, *ndentry, *trap;
1652 	struct inode	*fdir, *tdir;
1653 	__be32		err;
1654 	int		host_err;
1655 
1656 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1657 	if (err)
1658 		goto out;
1659 	err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1660 	if (err)
1661 		goto out;
1662 
1663 	fdentry = ffhp->fh_dentry;
1664 	fdir = fdentry->d_inode;
1665 
1666 	tdentry = tfhp->fh_dentry;
1667 	tdir = tdentry->d_inode;
1668 
1669 	err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1670 	if (ffhp->fh_export != tfhp->fh_export)
1671 		goto out;
1672 
1673 	err = nfserr_perm;
1674 	if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1675 		goto out;
1676 
1677 	/* cannot use fh_lock as we need deadlock protective ordering
1678 	 * so do it by hand */
1679 	trap = lock_rename(tdentry, fdentry);
1680 	ffhp->fh_locked = tfhp->fh_locked = 1;
1681 	fill_pre_wcc(ffhp);
1682 	fill_pre_wcc(tfhp);
1683 
1684 	odentry = lookup_one_len(fname, fdentry, flen);
1685 	host_err = PTR_ERR(odentry);
1686 	if (IS_ERR(odentry))
1687 		goto out_nfserr;
1688 
1689 	host_err = -ENOENT;
1690 	if (!odentry->d_inode)
1691 		goto out_dput_old;
1692 	host_err = -EINVAL;
1693 	if (odentry == trap)
1694 		goto out_dput_old;
1695 
1696 	ndentry = lookup_one_len(tname, tdentry, tlen);
1697 	host_err = PTR_ERR(ndentry);
1698 	if (IS_ERR(ndentry))
1699 		goto out_dput_old;
1700 	host_err = -ENOTEMPTY;
1701 	if (ndentry == trap)
1702 		goto out_dput_new;
1703 
1704 	if (svc_msnfs(ffhp) &&
1705 		((atomic_read(&odentry->d_count) > 1)
1706 		 || (atomic_read(&ndentry->d_count) > 1))) {
1707 			host_err = -EPERM;
1708 			goto out_dput_new;
1709 	}
1710 
1711 	host_err = -EXDEV;
1712 	if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1713 		goto out_dput_new;
1714 	host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1715 	if (host_err)
1716 		goto out_dput_new;
1717 
1718 	host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1719 	if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1720 		host_err = nfsd_sync_dir(tdentry);
1721 		if (!host_err)
1722 			host_err = nfsd_sync_dir(fdentry);
1723 	}
1724 
1725 	mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1726 
1727  out_dput_new:
1728 	dput(ndentry);
1729  out_dput_old:
1730 	dput(odentry);
1731  out_nfserr:
1732 	err = nfserrno(host_err);
1733 
1734 	/* we cannot reply on fh_unlock on the two filehandles,
1735 	 * as that would do the wrong thing if the two directories
1736 	 * were the same, so again we do it by hand
1737 	 */
1738 	fill_post_wcc(ffhp);
1739 	fill_post_wcc(tfhp);
1740 	unlock_rename(tdentry, fdentry);
1741 	ffhp->fh_locked = tfhp->fh_locked = 0;
1742 
1743 out:
1744 	return err;
1745 }
1746 
1747 /*
1748  * Unlink a file or directory
1749  * N.B. After this call fhp needs an fh_put
1750  */
1751 __be32
1752 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1753 				char *fname, int flen)
1754 {
1755 	struct dentry	*dentry, *rdentry;
1756 	struct inode	*dirp;
1757 	__be32		err;
1758 	int		host_err;
1759 
1760 	err = nfserr_acces;
1761 	if (!flen || isdotent(fname, flen))
1762 		goto out;
1763 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1764 	if (err)
1765 		goto out;
1766 
1767 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1768 	dentry = fhp->fh_dentry;
1769 	dirp = dentry->d_inode;
1770 
1771 	rdentry = lookup_one_len(fname, dentry, flen);
1772 	host_err = PTR_ERR(rdentry);
1773 	if (IS_ERR(rdentry))
1774 		goto out_nfserr;
1775 
1776 	if (!rdentry->d_inode) {
1777 		dput(rdentry);
1778 		err = nfserr_noent;
1779 		goto out;
1780 	}
1781 
1782 	if (!type)
1783 		type = rdentry->d_inode->i_mode & S_IFMT;
1784 
1785 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1786 	if (host_err)
1787 		goto out_nfserr;
1788 
1789 	if (type != S_IFDIR) { /* It's UNLINK */
1790 #ifdef MSNFS
1791 		if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1792 			(atomic_read(&rdentry->d_count) > 1)) {
1793 			host_err = -EPERM;
1794 		} else
1795 #endif
1796 		host_err = vfs_unlink(dirp, rdentry);
1797 	} else { /* It's RMDIR */
1798 		host_err = vfs_rmdir(dirp, rdentry);
1799 	}
1800 
1801 	dput(rdentry);
1802 
1803 	if (host_err)
1804 		goto out_drop;
1805 	if (EX_ISSYNC(fhp->fh_export))
1806 		host_err = nfsd_sync_dir(dentry);
1807 
1808 out_drop:
1809 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1810 out_nfserr:
1811 	err = nfserrno(host_err);
1812 out:
1813 	return err;
1814 }
1815 
1816 /*
1817  * Read entries from a directory.
1818  * The  NFSv3/4 verifier we ignore for now.
1819  */
1820 __be32
1821 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1822 	     struct readdir_cd *cdp, filldir_t func)
1823 {
1824 	__be32		err;
1825 	int 		host_err;
1826 	struct file	*file;
1827 	loff_t		offset = *offsetp;
1828 
1829 	err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1830 	if (err)
1831 		goto out;
1832 
1833 	offset = vfs_llseek(file, offset, 0);
1834 	if (offset < 0) {
1835 		err = nfserrno((int)offset);
1836 		goto out_close;
1837 	}
1838 
1839 	/*
1840 	 * Read the directory entries. This silly loop is necessary because
1841 	 * readdir() is not guaranteed to fill up the entire buffer, but
1842 	 * may choose to do less.
1843 	 */
1844 
1845 	do {
1846 		cdp->err = nfserr_eof; /* will be cleared on successful read */
1847 		host_err = vfs_readdir(file, func, cdp);
1848 	} while (host_err >=0 && cdp->err == nfs_ok);
1849 	if (host_err)
1850 		err = nfserrno(host_err);
1851 	else
1852 		err = cdp->err;
1853 	*offsetp = vfs_llseek(file, 0, 1);
1854 
1855 	if (err == nfserr_eof || err == nfserr_toosmall)
1856 		err = nfs_ok; /* can still be found in ->err */
1857 out_close:
1858 	nfsd_close(file);
1859 out:
1860 	return err;
1861 }
1862 
1863 /*
1864  * Get file system stats
1865  * N.B. After this call fhp needs an fh_put
1866  */
1867 __be32
1868 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1869 {
1870 	__be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1871 	if (!err && vfs_statfs(fhp->fh_dentry,stat))
1872 		err = nfserr_io;
1873 	return err;
1874 }
1875 
1876 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1877 {
1878 	return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1879 }
1880 
1881 /*
1882  * Check for a user's access permissions to this inode.
1883  */
1884 __be32
1885 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1886 					struct dentry *dentry, int acc)
1887 {
1888 	struct inode	*inode = dentry->d_inode;
1889 	int		err;
1890 
1891 	if (acc == NFSD_MAY_NOP)
1892 		return 0;
1893 #if 0
1894 	dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1895 		acc,
1896 		(acc & NFSD_MAY_READ)?	" read"  : "",
1897 		(acc & NFSD_MAY_WRITE)?	" write" : "",
1898 		(acc & NFSD_MAY_EXEC)?	" exec"  : "",
1899 		(acc & NFSD_MAY_SATTR)?	" sattr" : "",
1900 		(acc & NFSD_MAY_TRUNC)?	" trunc" : "",
1901 		(acc & NFSD_MAY_LOCK)?	" lock"  : "",
1902 		(acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1903 		inode->i_mode,
1904 		IS_IMMUTABLE(inode)?	" immut" : "",
1905 		IS_APPEND(inode)?	" append" : "",
1906 		__mnt_is_readonly(exp->ex_path.mnt)?	" ro" : "");
1907 	dprintk("      owner %d/%d user %d/%d\n",
1908 		inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1909 #endif
1910 
1911 	/* Normally we reject any write/sattr etc access on a read-only file
1912 	 * system.  But if it is IRIX doing check on write-access for a
1913 	 * device special file, we ignore rofs.
1914 	 */
1915 	if (!(acc & NFSD_MAY_LOCAL_ACCESS))
1916 		if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
1917 			if (exp_rdonly(rqstp, exp) ||
1918 			    __mnt_is_readonly(exp->ex_path.mnt))
1919 				return nfserr_rofs;
1920 			if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
1921 				return nfserr_perm;
1922 		}
1923 	if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
1924 		return nfserr_perm;
1925 
1926 	if (acc & NFSD_MAY_LOCK) {
1927 		/* If we cannot rely on authentication in NLM requests,
1928 		 * just allow locks, otherwise require read permission, or
1929 		 * ownership
1930 		 */
1931 		if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1932 			return 0;
1933 		else
1934 			acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
1935 	}
1936 	/*
1937 	 * The file owner always gets access permission for accesses that
1938 	 * would normally be checked at open time. This is to make
1939 	 * file access work even when the client has done a fchmod(fd, 0).
1940 	 *
1941 	 * However, `cp foo bar' should fail nevertheless when bar is
1942 	 * readonly. A sensible way to do this might be to reject all
1943 	 * attempts to truncate a read-only file, because a creat() call
1944 	 * always implies file truncation.
1945 	 * ... but this isn't really fair.  A process may reasonably call
1946 	 * ftruncate on an open file descriptor on a file with perm 000.
1947 	 * We must trust the client to do permission checking - using "ACCESS"
1948 	 * with NFSv3.
1949 	 */
1950 	if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
1951 	    inode->i_uid == current->fsuid)
1952 		return 0;
1953 
1954 	/* This assumes  NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
1955 	err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
1956 
1957 	/* Allow read access to binaries even when mode 111 */
1958 	if (err == -EACCES && S_ISREG(inode->i_mode) &&
1959 	    acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
1960 		err = inode_permission(inode, MAY_EXEC);
1961 
1962 	return err? nfserrno(err) : 0;
1963 }
1964 
1965 void
1966 nfsd_racache_shutdown(void)
1967 {
1968 	struct raparms *raparm, *last_raparm;
1969 	unsigned int i;
1970 
1971 	dprintk("nfsd: freeing readahead buffers.\n");
1972 
1973 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
1974 		raparm = raparm_hash[i].pb_head;
1975 		while(raparm) {
1976 			last_raparm = raparm;
1977 			raparm = raparm->p_next;
1978 			kfree(last_raparm);
1979 		}
1980 		raparm_hash[i].pb_head = NULL;
1981 	}
1982 }
1983 /*
1984  * Initialize readahead param cache
1985  */
1986 int
1987 nfsd_racache_init(int cache_size)
1988 {
1989 	int	i;
1990 	int	j = 0;
1991 	int	nperbucket;
1992 	struct raparms **raparm = NULL;
1993 
1994 
1995 	if (raparm_hash[0].pb_head)
1996 		return 0;
1997 	nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
1998 	if (nperbucket < 2)
1999 		nperbucket = 2;
2000 	cache_size = nperbucket * RAPARM_HASH_SIZE;
2001 
2002 	dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2003 
2004 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2005 		spin_lock_init(&raparm_hash[i].pb_lock);
2006 
2007 		raparm = &raparm_hash[i].pb_head;
2008 		for (j = 0; j < nperbucket; j++) {
2009 			*raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2010 			if (!*raparm)
2011 				goto out_nomem;
2012 			raparm = &(*raparm)->p_next;
2013 		}
2014 		*raparm = NULL;
2015 	}
2016 
2017 	nfsdstats.ra_size = cache_size;
2018 	return 0;
2019 
2020 out_nomem:
2021 	dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2022 	nfsd_racache_shutdown();
2023 	return -ENOMEM;
2024 }
2025 
2026 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2027 struct posix_acl *
2028 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2029 {
2030 	struct inode *inode = fhp->fh_dentry->d_inode;
2031 	char *name;
2032 	void *value = NULL;
2033 	ssize_t size;
2034 	struct posix_acl *acl;
2035 
2036 	if (!IS_POSIXACL(inode))
2037 		return ERR_PTR(-EOPNOTSUPP);
2038 
2039 	switch (type) {
2040 	case ACL_TYPE_ACCESS:
2041 		name = POSIX_ACL_XATTR_ACCESS;
2042 		break;
2043 	case ACL_TYPE_DEFAULT:
2044 		name = POSIX_ACL_XATTR_DEFAULT;
2045 		break;
2046 	default:
2047 		return ERR_PTR(-EOPNOTSUPP);
2048 	}
2049 
2050 	size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2051 	if (size < 0)
2052 		return ERR_PTR(size);
2053 
2054 	acl = posix_acl_from_xattr(value, size);
2055 	kfree(value);
2056 	return acl;
2057 }
2058 
2059 int
2060 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2061 {
2062 	struct inode *inode = fhp->fh_dentry->d_inode;
2063 	char *name;
2064 	void *value = NULL;
2065 	size_t size;
2066 	int error;
2067 
2068 	if (!IS_POSIXACL(inode) || !inode->i_op ||
2069 	    !inode->i_op->setxattr || !inode->i_op->removexattr)
2070 		return -EOPNOTSUPP;
2071 	switch(type) {
2072 		case ACL_TYPE_ACCESS:
2073 			name = POSIX_ACL_XATTR_ACCESS;
2074 			break;
2075 		case ACL_TYPE_DEFAULT:
2076 			name = POSIX_ACL_XATTR_DEFAULT;
2077 			break;
2078 		default:
2079 			return -EOPNOTSUPP;
2080 	}
2081 
2082 	if (acl && acl->a_count) {
2083 		size = posix_acl_xattr_size(acl->a_count);
2084 		value = kmalloc(size, GFP_KERNEL);
2085 		if (!value)
2086 			return -ENOMEM;
2087 		error = posix_acl_to_xattr(acl, value, size);
2088 		if (error < 0)
2089 			goto getout;
2090 		size = error;
2091 	} else
2092 		size = 0;
2093 
2094 	error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2095 	if (error)
2096 		goto getout;
2097 	if (size)
2098 		error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2099 	else {
2100 		if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2101 			error = 0;
2102 		else {
2103 			error = vfs_removexattr(fhp->fh_dentry, name);
2104 			if (error == -ENODATA)
2105 				error = 0;
2106 		}
2107 	}
2108 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
2109 
2110 getout:
2111 	kfree(value);
2112 	return error;
2113 }
2114 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
2115