xref: /linux/fs/nfsd/vfs.c (revision ecba1060583635ab55092072441ff903b5e9a659)
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 #include <linux/ima.h>
59 
60 #include <asm/uaccess.h>
61 
62 #define NFSDDBG_FACILITY		NFSDDBG_FILEOP
63 
64 
65 /*
66  * This is a cache of readahead params that help us choose the proper
67  * readahead strategy. Initially, we set all readahead parameters to 0
68  * and let the VFS handle things.
69  * If you increase the number of cached files very much, you'll need to
70  * add a hash table here.
71  */
72 struct raparms {
73 	struct raparms		*p_next;
74 	unsigned int		p_count;
75 	ino_t			p_ino;
76 	dev_t			p_dev;
77 	int			p_set;
78 	struct file_ra_state	p_ra;
79 	unsigned int		p_hindex;
80 };
81 
82 struct raparm_hbucket {
83 	struct raparms		*pb_head;
84 	spinlock_t		pb_lock;
85 } ____cacheline_aligned_in_smp;
86 
87 #define RAPARM_HASH_BITS	4
88 #define RAPARM_HASH_SIZE	(1<<RAPARM_HASH_BITS)
89 #define RAPARM_HASH_MASK	(RAPARM_HASH_SIZE-1)
90 static struct raparm_hbucket	raparm_hash[RAPARM_HASH_SIZE];
91 
92 /*
93  * Called from nfsd_lookup and encode_dirent. Check if we have crossed
94  * a mount point.
95  * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
96  *  or nfs_ok having possibly changed *dpp and *expp
97  */
98 int
99 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
100 		        struct svc_export **expp)
101 {
102 	struct svc_export *exp = *expp, *exp2 = NULL;
103 	struct dentry *dentry = *dpp;
104 	struct path path = {.mnt = mntget(exp->ex_path.mnt),
105 			    .dentry = dget(dentry)};
106 	int err = 0;
107 
108 	while (d_mountpoint(path.dentry) && follow_down(&path))
109 		;
110 
111 	exp2 = rqst_exp_get_by_name(rqstp, &path);
112 	if (IS_ERR(exp2)) {
113 		if (PTR_ERR(exp2) != -ENOENT)
114 			err = PTR_ERR(exp2);
115 		path_put(&path);
116 		goto out;
117 	}
118 	if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
119 		/* successfully crossed mount point */
120 		/*
121 		 * This is subtle: path.dentry is *not* on path.mnt
122 		 * at this point.  The only reason we are safe is that
123 		 * original mnt is pinned down by exp, so we should
124 		 * put path *before* putting exp
125 		 */
126 		*dpp = path.dentry;
127 		path.dentry = dentry;
128 		*expp = exp2;
129 		exp2 = exp;
130 	}
131 	path_put(&path);
132 	exp_put(exp2);
133 out:
134 	return err;
135 }
136 
137 __be32
138 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
139 		   const char *name, unsigned int len,
140 		   struct svc_export **exp_ret, struct dentry **dentry_ret)
141 {
142 	struct svc_export	*exp;
143 	struct dentry		*dparent;
144 	struct dentry		*dentry;
145 	__be32			err;
146 	int			host_err;
147 
148 	dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
149 
150 	/* Obtain dentry and export. */
151 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
152 	if (err)
153 		return err;
154 
155 	dparent = fhp->fh_dentry;
156 	exp  = fhp->fh_export;
157 	exp_get(exp);
158 
159 	/* Lookup the name, but don't follow links */
160 	if (isdotent(name, len)) {
161 		if (len==1)
162 			dentry = dget(dparent);
163 		else if (dparent != exp->ex_path.dentry)
164 			dentry = dget_parent(dparent);
165 		else if (!EX_NOHIDE(exp))
166 			dentry = dget(dparent); /* .. == . just like at / */
167 		else {
168 			/* checking mountpoint crossing is very different when stepping up */
169 			struct svc_export *exp2 = NULL;
170 			struct dentry *dp;
171 			struct path path = {.mnt = mntget(exp->ex_path.mnt),
172 					    .dentry = dget(dparent)};
173 
174 			while (path.dentry == path.mnt->mnt_root &&
175 			       follow_up(&path))
176 				;
177 			dp = dget_parent(path.dentry);
178 			dput(path.dentry);
179 			path.dentry = dp;
180 
181 			exp2 = rqst_exp_parent(rqstp, &path);
182 			if (PTR_ERR(exp2) == -ENOENT) {
183 				dentry = dget(dparent);
184 			} else if (IS_ERR(exp2)) {
185 				host_err = PTR_ERR(exp2);
186 				path_put(&path);
187 				goto out_nfserr;
188 			} else {
189 				dentry = dget(path.dentry);
190 				exp_put(exp);
191 				exp = exp2;
192 			}
193 			path_put(&path);
194 		}
195 	} else {
196 		fh_lock(fhp);
197 		dentry = lookup_one_len(name, dparent, len);
198 		host_err = PTR_ERR(dentry);
199 		if (IS_ERR(dentry))
200 			goto out_nfserr;
201 		/*
202 		 * check if we have crossed a mount point ...
203 		 */
204 		if (d_mountpoint(dentry)) {
205 			if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
206 				dput(dentry);
207 				goto out_nfserr;
208 			}
209 		}
210 	}
211 	*dentry_ret = dentry;
212 	*exp_ret = exp;
213 	return 0;
214 
215 out_nfserr:
216 	exp_put(exp);
217 	return nfserrno(host_err);
218 }
219 
220 /*
221  * Look up one component of a pathname.
222  * N.B. After this call _both_ fhp and resfh need an fh_put
223  *
224  * If the lookup would cross a mountpoint, and the mounted filesystem
225  * is exported to the client with NFSEXP_NOHIDE, then the lookup is
226  * accepted as it stands and the mounted directory is
227  * returned. Otherwise the covered directory is returned.
228  * NOTE: this mountpoint crossing is not supported properly by all
229  *   clients and is explicitly disallowed for NFSv3
230  *      NeilBrown <neilb@cse.unsw.edu.au>
231  */
232 __be32
233 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
234 				unsigned int len, struct svc_fh *resfh)
235 {
236 	struct svc_export	*exp;
237 	struct dentry		*dentry;
238 	__be32 err;
239 
240 	err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
241 	if (err)
242 		return err;
243 	err = check_nfsd_access(exp, rqstp);
244 	if (err)
245 		goto out;
246 	/*
247 	 * Note: we compose the file handle now, but as the
248 	 * dentry may be negative, it may need to be updated.
249 	 */
250 	err = fh_compose(resfh, exp, dentry, fhp);
251 	if (!err && !dentry->d_inode)
252 		err = nfserr_noent;
253 out:
254 	dput(dentry);
255 	exp_put(exp);
256 	return err;
257 }
258 
259 
260 /*
261  * Set various file attributes.
262  * N.B. After this call fhp needs an fh_put
263  */
264 __be32
265 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
266 	     int check_guard, time_t guardtime)
267 {
268 	struct dentry	*dentry;
269 	struct inode	*inode;
270 	int		accmode = NFSD_MAY_SATTR;
271 	int		ftype = 0;
272 	__be32		err;
273 	int		host_err;
274 	int		size_change = 0;
275 
276 	if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
277 		accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
278 	if (iap->ia_valid & ATTR_SIZE)
279 		ftype = S_IFREG;
280 
281 	/* Get inode */
282 	err = fh_verify(rqstp, fhp, ftype, accmode);
283 	if (err)
284 		goto out;
285 
286 	dentry = fhp->fh_dentry;
287 	inode = dentry->d_inode;
288 
289 	/* Ignore any mode updates on symlinks */
290 	if (S_ISLNK(inode->i_mode))
291 		iap->ia_valid &= ~ATTR_MODE;
292 
293 	if (!iap->ia_valid)
294 		goto out;
295 
296 	/*
297 	 * NFSv2 does not differentiate between "set-[ac]time-to-now"
298 	 * which only requires access, and "set-[ac]time-to-X" which
299 	 * requires ownership.
300 	 * So if it looks like it might be "set both to the same time which
301 	 * is close to now", and if inode_change_ok fails, then we
302 	 * convert to "set to now" instead of "set to explicit time"
303 	 *
304 	 * We only call inode_change_ok as the last test as technically
305 	 * it is not an interface that we should be using.  It is only
306 	 * valid if the filesystem does not define it's own i_op->setattr.
307 	 */
308 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
309 #define	MAX_TOUCH_TIME_ERROR (30*60)
310 	if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
311 	    iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
312 		/*
313 		 * Looks probable.
314 		 *
315 		 * Now just make sure time is in the right ballpark.
316 		 * Solaris, at least, doesn't seem to care what the time
317 		 * request is.  We require it be within 30 minutes of now.
318 		 */
319 		time_t delta = iap->ia_atime.tv_sec - get_seconds();
320 		if (delta < 0)
321 			delta = -delta;
322 		if (delta < MAX_TOUCH_TIME_ERROR &&
323 		    inode_change_ok(inode, iap) != 0) {
324 			/*
325 			 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
326 			 * This will cause notify_change to set these times
327 			 * to "now"
328 			 */
329 			iap->ia_valid &= ~BOTH_TIME_SET;
330 		}
331 	}
332 
333 	/*
334 	 * The size case is special.
335 	 * It changes the file as well as the attributes.
336 	 */
337 	if (iap->ia_valid & ATTR_SIZE) {
338 		if (iap->ia_size < inode->i_size) {
339 			err = nfsd_permission(rqstp, fhp->fh_export, dentry,
340 					NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
341 			if (err)
342 				goto out;
343 		}
344 
345 		/*
346 		 * If we are changing the size of the file, then
347 		 * we need to break all leases.
348 		 */
349 		host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
350 		if (host_err == -EWOULDBLOCK)
351 			host_err = -ETIMEDOUT;
352 		if (host_err) /* ENOMEM or EWOULDBLOCK */
353 			goto out_nfserr;
354 
355 		host_err = get_write_access(inode);
356 		if (host_err)
357 			goto out_nfserr;
358 
359 		size_change = 1;
360 		host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
361 		if (host_err) {
362 			put_write_access(inode);
363 			goto out_nfserr;
364 		}
365 		vfs_dq_init(inode);
366 	}
367 
368 	/* sanitize the mode change */
369 	if (iap->ia_valid & ATTR_MODE) {
370 		iap->ia_mode &= S_IALLUGO;
371 		iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
372 	}
373 
374 	/* Revoke setuid/setgid on chown */
375 	if (!S_ISDIR(inode->i_mode) &&
376 	    (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
377 	     ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
378 		iap->ia_valid |= ATTR_KILL_PRIV;
379 		if (iap->ia_valid & ATTR_MODE) {
380 			/* we're setting mode too, just clear the s*id bits */
381 			iap->ia_mode &= ~S_ISUID;
382 			if (iap->ia_mode & S_IXGRP)
383 				iap->ia_mode &= ~S_ISGID;
384 		} else {
385 			/* set ATTR_KILL_* bits and let VFS handle it */
386 			iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
387 		}
388 	}
389 
390 	/* Change the attributes. */
391 
392 	iap->ia_valid |= ATTR_CTIME;
393 
394 	err = nfserr_notsync;
395 	if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
396 		fh_lock(fhp);
397 		host_err = notify_change(dentry, iap);
398 		err = nfserrno(host_err);
399 		fh_unlock(fhp);
400 	}
401 	if (size_change)
402 		put_write_access(inode);
403 	if (!err)
404 		if (EX_ISSYNC(fhp->fh_export))
405 			write_inode_now(inode, 1);
406 out:
407 	return err;
408 
409 out_nfserr:
410 	err = nfserrno(host_err);
411 	goto out;
412 }
413 
414 #if defined(CONFIG_NFSD_V2_ACL) || \
415     defined(CONFIG_NFSD_V3_ACL) || \
416     defined(CONFIG_NFSD_V4)
417 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
418 {
419 	ssize_t buflen;
420 	ssize_t ret;
421 
422 	buflen = vfs_getxattr(dentry, key, NULL, 0);
423 	if (buflen <= 0)
424 		return buflen;
425 
426 	*buf = kmalloc(buflen, GFP_KERNEL);
427 	if (!*buf)
428 		return -ENOMEM;
429 
430 	ret = vfs_getxattr(dentry, key, *buf, buflen);
431 	if (ret < 0)
432 		kfree(*buf);
433 	return ret;
434 }
435 #endif
436 
437 #if defined(CONFIG_NFSD_V4)
438 static int
439 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
440 {
441 	int len;
442 	size_t buflen;
443 	char *buf = NULL;
444 	int error = 0;
445 
446 	buflen = posix_acl_xattr_size(pacl->a_count);
447 	buf = kmalloc(buflen, GFP_KERNEL);
448 	error = -ENOMEM;
449 	if (buf == NULL)
450 		goto out;
451 
452 	len = posix_acl_to_xattr(pacl, buf, buflen);
453 	if (len < 0) {
454 		error = len;
455 		goto out;
456 	}
457 
458 	error = vfs_setxattr(dentry, key, buf, len, 0);
459 out:
460 	kfree(buf);
461 	return error;
462 }
463 
464 __be32
465 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
466     struct nfs4_acl *acl)
467 {
468 	__be32 error;
469 	int host_error;
470 	struct dentry *dentry;
471 	struct inode *inode;
472 	struct posix_acl *pacl = NULL, *dpacl = NULL;
473 	unsigned int flags = 0;
474 
475 	/* Get inode */
476 	error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
477 	if (error)
478 		return error;
479 
480 	dentry = fhp->fh_dentry;
481 	inode = dentry->d_inode;
482 	if (S_ISDIR(inode->i_mode))
483 		flags = NFS4_ACL_DIR;
484 
485 	host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
486 	if (host_error == -EINVAL) {
487 		return nfserr_attrnotsupp;
488 	} else if (host_error < 0)
489 		goto out_nfserr;
490 
491 	host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
492 	if (host_error < 0)
493 		goto out_release;
494 
495 	if (S_ISDIR(inode->i_mode))
496 		host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
497 
498 out_release:
499 	posix_acl_release(pacl);
500 	posix_acl_release(dpacl);
501 out_nfserr:
502 	if (host_error == -EOPNOTSUPP)
503 		return nfserr_attrnotsupp;
504 	else
505 		return nfserrno(host_error);
506 }
507 
508 static struct posix_acl *
509 _get_posix_acl(struct dentry *dentry, char *key)
510 {
511 	void *buf = NULL;
512 	struct posix_acl *pacl = NULL;
513 	int buflen;
514 
515 	buflen = nfsd_getxattr(dentry, key, &buf);
516 	if (!buflen)
517 		buflen = -ENODATA;
518 	if (buflen <= 0)
519 		return ERR_PTR(buflen);
520 
521 	pacl = posix_acl_from_xattr(buf, buflen);
522 	kfree(buf);
523 	return pacl;
524 }
525 
526 int
527 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
528 {
529 	struct inode *inode = dentry->d_inode;
530 	int error = 0;
531 	struct posix_acl *pacl = NULL, *dpacl = NULL;
532 	unsigned int flags = 0;
533 
534 	pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
535 	if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
536 		pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
537 	if (IS_ERR(pacl)) {
538 		error = PTR_ERR(pacl);
539 		pacl = NULL;
540 		goto out;
541 	}
542 
543 	if (S_ISDIR(inode->i_mode)) {
544 		dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
545 		if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
546 			dpacl = NULL;
547 		else if (IS_ERR(dpacl)) {
548 			error = PTR_ERR(dpacl);
549 			dpacl = NULL;
550 			goto out;
551 		}
552 		flags = NFS4_ACL_DIR;
553 	}
554 
555 	*acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
556 	if (IS_ERR(*acl)) {
557 		error = PTR_ERR(*acl);
558 		*acl = NULL;
559 	}
560  out:
561 	posix_acl_release(pacl);
562 	posix_acl_release(dpacl);
563 	return error;
564 }
565 
566 #endif /* defined(CONFIG_NFS_V4) */
567 
568 #ifdef CONFIG_NFSD_V3
569 /*
570  * Check server access rights to a file system object
571  */
572 struct accessmap {
573 	u32		access;
574 	int		how;
575 };
576 static struct accessmap	nfs3_regaccess[] = {
577     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
578     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
579     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_TRUNC	},
580     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE			},
581 
582     {	0,			0				}
583 };
584 
585 static struct accessmap	nfs3_diraccess[] = {
586     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
587     {	NFS3_ACCESS_LOOKUP,	NFSD_MAY_EXEC			},
588     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
589     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_EXEC|NFSD_MAY_WRITE	},
590     {	NFS3_ACCESS_DELETE,	NFSD_MAY_REMOVE			},
591 
592     {	0,			0				}
593 };
594 
595 static struct accessmap	nfs3_anyaccess[] = {
596 	/* Some clients - Solaris 2.6 at least, make an access call
597 	 * to the server to check for access for things like /dev/null
598 	 * (which really, the server doesn't care about).  So
599 	 * We provide simple access checking for them, looking
600 	 * mainly at mode bits, and we make sure to ignore read-only
601 	 * filesystem checks
602 	 */
603     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
604     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
605     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
606     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
607 
608     {	0,			0				}
609 };
610 
611 __be32
612 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
613 {
614 	struct accessmap	*map;
615 	struct svc_export	*export;
616 	struct dentry		*dentry;
617 	u32			query, result = 0, sresult = 0;
618 	__be32			error;
619 
620 	error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
621 	if (error)
622 		goto out;
623 
624 	export = fhp->fh_export;
625 	dentry = fhp->fh_dentry;
626 
627 	if (S_ISREG(dentry->d_inode->i_mode))
628 		map = nfs3_regaccess;
629 	else if (S_ISDIR(dentry->d_inode->i_mode))
630 		map = nfs3_diraccess;
631 	else
632 		map = nfs3_anyaccess;
633 
634 
635 	query = *access;
636 	for  (; map->access; map++) {
637 		if (map->access & query) {
638 			__be32 err2;
639 
640 			sresult |= map->access;
641 
642 			err2 = nfsd_permission(rqstp, export, dentry, map->how);
643 			switch (err2) {
644 			case nfs_ok:
645 				result |= map->access;
646 				break;
647 
648 			/* the following error codes just mean the access was not allowed,
649 			 * rather than an error occurred */
650 			case nfserr_rofs:
651 			case nfserr_acces:
652 			case nfserr_perm:
653 				/* simply don't "or" in the access bit. */
654 				break;
655 			default:
656 				error = err2;
657 				goto out;
658 			}
659 		}
660 	}
661 	*access = result;
662 	if (supported)
663 		*supported = sresult;
664 
665  out:
666 	return error;
667 }
668 #endif /* CONFIG_NFSD_V3 */
669 
670 
671 
672 /*
673  * Open an existing file or directory.
674  * The access argument indicates the type of open (read/write/lock)
675  * N.B. After this call fhp needs an fh_put
676  */
677 __be32
678 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
679 			int access, struct file **filp)
680 {
681 	struct dentry	*dentry;
682 	struct inode	*inode;
683 	int		flags = O_RDONLY|O_LARGEFILE;
684 	__be32		err;
685 	int		host_err;
686 
687 	/*
688 	 * If we get here, then the client has already done an "open",
689 	 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
690 	 * in case a chmod has now revoked permission.
691 	 */
692 	err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
693 	if (err)
694 		goto out;
695 
696 	dentry = fhp->fh_dentry;
697 	inode = dentry->d_inode;
698 
699 	/* Disallow write access to files with the append-only bit set
700 	 * or any access when mandatory locking enabled
701 	 */
702 	err = nfserr_perm;
703 	if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
704 		goto out;
705 	/*
706 	 * We must ignore files (but only files) which might have mandatory
707 	 * locks on them because there is no way to know if the accesser has
708 	 * the lock.
709 	 */
710 	if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
711 		goto out;
712 
713 	if (!inode->i_fop)
714 		goto out;
715 
716 	/*
717 	 * Check to see if there are any leases on this file.
718 	 * This may block while leases are broken.
719 	 */
720 	host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
721 	if (host_err == -EWOULDBLOCK)
722 		host_err = -ETIMEDOUT;
723 	if (host_err) /* NOMEM or WOULDBLOCK */
724 		goto out_nfserr;
725 
726 	if (access & NFSD_MAY_WRITE) {
727 		if (access & NFSD_MAY_READ)
728 			flags = O_RDWR|O_LARGEFILE;
729 		else
730 			flags = O_WRONLY|O_LARGEFILE;
731 
732 		vfs_dq_init(inode);
733 	}
734 	*filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
735 			    flags, current_cred());
736 	if (IS_ERR(*filp))
737 		host_err = PTR_ERR(*filp);
738 	else
739 		ima_counts_get(*filp);
740 out_nfserr:
741 	err = nfserrno(host_err);
742 out:
743 	return err;
744 }
745 
746 /*
747  * Close a file.
748  */
749 void
750 nfsd_close(struct file *filp)
751 {
752 	fput(filp);
753 }
754 
755 /*
756  * Sync a file
757  * As this calls fsync (not fdatasync) there is no need for a write_inode
758  * after it.
759  */
760 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
761 			      const struct file_operations *fop)
762 {
763 	struct inode *inode = dp->d_inode;
764 	int (*fsync) (struct file *, struct dentry *, int);
765 	int err;
766 
767 	err = filemap_fdatawrite(inode->i_mapping);
768 	if (err == 0 && fop && (fsync = fop->fsync))
769 		err = fsync(filp, dp, 0);
770 	if (err == 0)
771 		err = filemap_fdatawait(inode->i_mapping);
772 
773 	return err;
774 }
775 
776 static int
777 nfsd_sync(struct file *filp)
778 {
779         int err;
780 	struct inode *inode = filp->f_path.dentry->d_inode;
781 	dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
782 	mutex_lock(&inode->i_mutex);
783 	err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
784 	mutex_unlock(&inode->i_mutex);
785 
786 	return err;
787 }
788 
789 int
790 nfsd_sync_dir(struct dentry *dp)
791 {
792 	return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
793 }
794 
795 /*
796  * Obtain the readahead parameters for the file
797  * specified by (dev, ino).
798  */
799 
800 static inline struct raparms *
801 nfsd_get_raparms(dev_t dev, ino_t ino)
802 {
803 	struct raparms	*ra, **rap, **frap = NULL;
804 	int depth = 0;
805 	unsigned int hash;
806 	struct raparm_hbucket *rab;
807 
808 	hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
809 	rab = &raparm_hash[hash];
810 
811 	spin_lock(&rab->pb_lock);
812 	for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
813 		if (ra->p_ino == ino && ra->p_dev == dev)
814 			goto found;
815 		depth++;
816 		if (ra->p_count == 0)
817 			frap = rap;
818 	}
819 	depth = nfsdstats.ra_size*11/10;
820 	if (!frap) {
821 		spin_unlock(&rab->pb_lock);
822 		return NULL;
823 	}
824 	rap = frap;
825 	ra = *frap;
826 	ra->p_dev = dev;
827 	ra->p_ino = ino;
828 	ra->p_set = 0;
829 	ra->p_hindex = hash;
830 found:
831 	if (rap != &rab->pb_head) {
832 		*rap = ra->p_next;
833 		ra->p_next   = rab->pb_head;
834 		rab->pb_head = ra;
835 	}
836 	ra->p_count++;
837 	nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
838 	spin_unlock(&rab->pb_lock);
839 	return ra;
840 }
841 
842 /*
843  * Grab and keep cached pages associated with a file in the svc_rqst
844  * so that they can be passed to the network sendmsg/sendpage routines
845  * directly. They will be released after the sending has completed.
846  */
847 static int
848 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
849 		  struct splice_desc *sd)
850 {
851 	struct svc_rqst *rqstp = sd->u.data;
852 	struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
853 	struct page *page = buf->page;
854 	size_t size;
855 	int ret;
856 
857 	ret = buf->ops->confirm(pipe, buf);
858 	if (unlikely(ret))
859 		return ret;
860 
861 	size = sd->len;
862 
863 	if (rqstp->rq_res.page_len == 0) {
864 		get_page(page);
865 		put_page(*pp);
866 		*pp = page;
867 		rqstp->rq_resused++;
868 		rqstp->rq_res.page_base = buf->offset;
869 		rqstp->rq_res.page_len = size;
870 	} else if (page != pp[-1]) {
871 		get_page(page);
872 		if (*pp)
873 			put_page(*pp);
874 		*pp = page;
875 		rqstp->rq_resused++;
876 		rqstp->rq_res.page_len += size;
877 	} else
878 		rqstp->rq_res.page_len += size;
879 
880 	return size;
881 }
882 
883 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
884 				    struct splice_desc *sd)
885 {
886 	return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
887 }
888 
889 static inline int svc_msnfs(struct svc_fh *ffhp)
890 {
891 #ifdef MSNFS
892 	return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
893 #else
894 	return 0;
895 #endif
896 }
897 
898 static __be32
899 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
900               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
901 {
902 	struct inode *inode;
903 	struct raparms	*ra;
904 	mm_segment_t	oldfs;
905 	__be32		err;
906 	int		host_err;
907 
908 	err = nfserr_perm;
909 	inode = file->f_path.dentry->d_inode;
910 
911 	if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
912 		goto out;
913 
914 	/* Get readahead parameters */
915 	ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
916 
917 	if (ra && ra->p_set)
918 		file->f_ra = ra->p_ra;
919 
920 	if (file->f_op->splice_read && rqstp->rq_splice_ok) {
921 		struct splice_desc sd = {
922 			.len		= 0,
923 			.total_len	= *count,
924 			.pos		= offset,
925 			.u.data		= rqstp,
926 		};
927 
928 		rqstp->rq_resused = 1;
929 		host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
930 	} else {
931 		oldfs = get_fs();
932 		set_fs(KERNEL_DS);
933 		host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
934 		set_fs(oldfs);
935 	}
936 
937 	/* Write back readahead params */
938 	if (ra) {
939 		struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
940 		spin_lock(&rab->pb_lock);
941 		ra->p_ra = file->f_ra;
942 		ra->p_set = 1;
943 		ra->p_count--;
944 		spin_unlock(&rab->pb_lock);
945 	}
946 
947 	if (host_err >= 0) {
948 		nfsdstats.io_read += host_err;
949 		*count = host_err;
950 		err = 0;
951 		fsnotify_access(file->f_path.dentry);
952 	} else
953 		err = nfserrno(host_err);
954 out:
955 	return err;
956 }
957 
958 static void kill_suid(struct dentry *dentry)
959 {
960 	struct iattr	ia;
961 	ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
962 
963 	mutex_lock(&dentry->d_inode->i_mutex);
964 	notify_change(dentry, &ia);
965 	mutex_unlock(&dentry->d_inode->i_mutex);
966 }
967 
968 /*
969  * Gathered writes: If another process is currently writing to the file,
970  * there's a high chance this is another nfsd (triggered by a bulk write
971  * from a client's biod). Rather than syncing the file with each write
972  * request, we sleep for 10 msec.
973  *
974  * I don't know if this roughly approximates C. Juszak's idea of
975  * gathered writes, but it's a nice and simple solution (IMHO), and it
976  * seems to work:-)
977  *
978  * Note: we do this only in the NFSv2 case, since v3 and higher have a
979  * better tool (separate unstable writes and commits) for solving this
980  * problem.
981  */
982 static int wait_for_concurrent_writes(struct file *file)
983 {
984 	struct inode *inode = file->f_path.dentry->d_inode;
985 	static ino_t last_ino;
986 	static dev_t last_dev;
987 	int err = 0;
988 
989 	if (atomic_read(&inode->i_writecount) > 1
990 	    || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
991 		dprintk("nfsd: write defer %d\n", task_pid_nr(current));
992 		msleep(10);
993 		dprintk("nfsd: write resume %d\n", task_pid_nr(current));
994 	}
995 
996 	if (inode->i_state & I_DIRTY) {
997 		dprintk("nfsd: write sync %d\n", task_pid_nr(current));
998 		err = nfsd_sync(file);
999 	}
1000 	last_ino = inode->i_ino;
1001 	last_dev = inode->i_sb->s_dev;
1002 	return err;
1003 }
1004 
1005 static __be32
1006 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1007 				loff_t offset, struct kvec *vec, int vlen,
1008 				unsigned long *cnt, int *stablep)
1009 {
1010 	struct svc_export	*exp;
1011 	struct dentry		*dentry;
1012 	struct inode		*inode;
1013 	mm_segment_t		oldfs;
1014 	__be32			err = 0;
1015 	int			host_err;
1016 	int			stable = *stablep;
1017 	int			use_wgather;
1018 
1019 #ifdef MSNFS
1020 	err = nfserr_perm;
1021 
1022 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1023 		(!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1024 		goto out;
1025 #endif
1026 
1027 	dentry = file->f_path.dentry;
1028 	inode = dentry->d_inode;
1029 	exp   = fhp->fh_export;
1030 
1031 	/*
1032 	 * Request sync writes if
1033 	 *  -	the sync export option has been set, or
1034 	 *  -	the client requested O_SYNC behavior (NFSv3 feature).
1035 	 *  -   The file system doesn't support fsync().
1036 	 * When NFSv2 gathered writes have been configured for this volume,
1037 	 * flushing the data to disk is handled separately below.
1038 	 */
1039 	use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1040 
1041 	if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1042 	       stable = 2;
1043 	       *stablep = 2; /* FILE_SYNC */
1044 	}
1045 
1046 	if (!EX_ISSYNC(exp))
1047 		stable = 0;
1048 	if (stable && !use_wgather) {
1049 		spin_lock(&file->f_lock);
1050 		file->f_flags |= O_SYNC;
1051 		spin_unlock(&file->f_lock);
1052 	}
1053 
1054 	/* Write the data. */
1055 	oldfs = get_fs(); set_fs(KERNEL_DS);
1056 	host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1057 	set_fs(oldfs);
1058 	if (host_err < 0)
1059 		goto out_nfserr;
1060 	*cnt = host_err;
1061 	nfsdstats.io_write += host_err;
1062 	fsnotify_modify(file->f_path.dentry);
1063 
1064 	/* clear setuid/setgid flag after write */
1065 	if (inode->i_mode & (S_ISUID | S_ISGID))
1066 		kill_suid(dentry);
1067 
1068 	if (stable && use_wgather)
1069 		host_err = wait_for_concurrent_writes(file);
1070 
1071 out_nfserr:
1072 	dprintk("nfsd: write complete host_err=%d\n", host_err);
1073 	if (host_err >= 0)
1074 		err = 0;
1075 	else
1076 		err = nfserrno(host_err);
1077 out:
1078 	return err;
1079 }
1080 
1081 /*
1082  * Read data from a file. count must contain the requested read count
1083  * on entry. On return, *count contains the number of bytes actually read.
1084  * N.B. After this call fhp needs an fh_put
1085  */
1086 __be32
1087 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1088 		loff_t offset, struct kvec *vec, int vlen,
1089 		unsigned long *count)
1090 {
1091 	__be32		err;
1092 
1093 	if (file) {
1094 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1095 				NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1096 		if (err)
1097 			goto out;
1098 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1099 	} else {
1100 		err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1101 		if (err)
1102 			goto out;
1103 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1104 		nfsd_close(file);
1105 	}
1106 out:
1107 	return err;
1108 }
1109 
1110 /*
1111  * Write data to a file.
1112  * The stable flag requests synchronous writes.
1113  * N.B. After this call fhp needs an fh_put
1114  */
1115 __be32
1116 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1117 		loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1118 		int *stablep)
1119 {
1120 	__be32			err = 0;
1121 
1122 	if (file) {
1123 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1124 				NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1125 		if (err)
1126 			goto out;
1127 		err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1128 				stablep);
1129 	} else {
1130 		err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1131 		if (err)
1132 			goto out;
1133 
1134 		if (cnt)
1135 			err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1136 					     cnt, stablep);
1137 		nfsd_close(file);
1138 	}
1139 out:
1140 	return err;
1141 }
1142 
1143 #ifdef CONFIG_NFSD_V3
1144 /*
1145  * Commit all pending writes to stable storage.
1146  * Strictly speaking, we could sync just the indicated file region here,
1147  * but there's currently no way we can ask the VFS to do so.
1148  *
1149  * Unfortunately we cannot lock the file to make sure we return full WCC
1150  * data to the client, as locking happens lower down in the filesystem.
1151  */
1152 __be32
1153 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1154                loff_t offset, unsigned long count)
1155 {
1156 	struct file	*file;
1157 	__be32		err;
1158 
1159 	if ((u64)count > ~(u64)offset)
1160 		return nfserr_inval;
1161 
1162 	err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1163 	if (err)
1164 		return err;
1165 	if (EX_ISSYNC(fhp->fh_export)) {
1166 		if (file->f_op && file->f_op->fsync) {
1167 			err = nfserrno(nfsd_sync(file));
1168 		} else {
1169 			err = nfserr_notsupp;
1170 		}
1171 	}
1172 
1173 	nfsd_close(file);
1174 	return err;
1175 }
1176 #endif /* CONFIG_NFSD_V3 */
1177 
1178 static __be32
1179 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1180 			struct iattr *iap)
1181 {
1182 	/*
1183 	 * Mode has already been set earlier in create:
1184 	 */
1185 	iap->ia_valid &= ~ATTR_MODE;
1186 	/*
1187 	 * Setting uid/gid works only for root.  Irix appears to
1188 	 * send along the gid on create when it tries to implement
1189 	 * setgid directories via NFS:
1190 	 */
1191 	if (current_fsuid() != 0)
1192 		iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1193 	if (iap->ia_valid)
1194 		return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1195 	return 0;
1196 }
1197 
1198 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1199  * setting size to 0 may fail for some specific file systems by the permission
1200  * checking which requires WRITE permission but the mode is 000.
1201  * we ignore the resizing(to 0) on the just new created file, since the size is
1202  * 0 after file created.
1203  *
1204  * call this only after vfs_create() is called.
1205  * */
1206 static void
1207 nfsd_check_ignore_resizing(struct iattr *iap)
1208 {
1209 	if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1210 		iap->ia_valid &= ~ATTR_SIZE;
1211 }
1212 
1213 /*
1214  * Create a file (regular, directory, device, fifo); UNIX sockets
1215  * not yet implemented.
1216  * If the response fh has been verified, the parent directory should
1217  * already be locked. Note that the parent directory is left locked.
1218  *
1219  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1220  */
1221 __be32
1222 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1223 		char *fname, int flen, struct iattr *iap,
1224 		int type, dev_t rdev, struct svc_fh *resfhp)
1225 {
1226 	struct dentry	*dentry, *dchild = NULL;
1227 	struct inode	*dirp;
1228 	__be32		err;
1229 	__be32		err2;
1230 	int		host_err;
1231 
1232 	err = nfserr_perm;
1233 	if (!flen)
1234 		goto out;
1235 	err = nfserr_exist;
1236 	if (isdotent(fname, flen))
1237 		goto out;
1238 
1239 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1240 	if (err)
1241 		goto out;
1242 
1243 	dentry = fhp->fh_dentry;
1244 	dirp = dentry->d_inode;
1245 
1246 	err = nfserr_notdir;
1247 	if (!dirp->i_op->lookup)
1248 		goto out;
1249 	/*
1250 	 * Check whether the response file handle has been verified yet.
1251 	 * If it has, the parent directory should already be locked.
1252 	 */
1253 	if (!resfhp->fh_dentry) {
1254 		/* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1255 		fh_lock_nested(fhp, I_MUTEX_PARENT);
1256 		dchild = lookup_one_len(fname, dentry, flen);
1257 		host_err = PTR_ERR(dchild);
1258 		if (IS_ERR(dchild))
1259 			goto out_nfserr;
1260 		err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1261 		if (err)
1262 			goto out;
1263 	} else {
1264 		/* called from nfsd_proc_create */
1265 		dchild = dget(resfhp->fh_dentry);
1266 		if (!fhp->fh_locked) {
1267 			/* not actually possible */
1268 			printk(KERN_ERR
1269 				"nfsd_create: parent %s/%s not locked!\n",
1270 				dentry->d_parent->d_name.name,
1271 				dentry->d_name.name);
1272 			err = nfserr_io;
1273 			goto out;
1274 		}
1275 	}
1276 	/*
1277 	 * Make sure the child dentry is still negative ...
1278 	 */
1279 	err = nfserr_exist;
1280 	if (dchild->d_inode) {
1281 		dprintk("nfsd_create: dentry %s/%s not negative!\n",
1282 			dentry->d_name.name, dchild->d_name.name);
1283 		goto out;
1284 	}
1285 
1286 	if (!(iap->ia_valid & ATTR_MODE))
1287 		iap->ia_mode = 0;
1288 	iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1289 
1290 	err = nfserr_inval;
1291 	if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1292 		printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1293 		       type);
1294 		goto out;
1295 	}
1296 
1297 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1298 	if (host_err)
1299 		goto out_nfserr;
1300 
1301 	/*
1302 	 * Get the dir op function pointer.
1303 	 */
1304 	err = 0;
1305 	switch (type) {
1306 	case S_IFREG:
1307 		host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1308 		if (!host_err)
1309 			nfsd_check_ignore_resizing(iap);
1310 		break;
1311 	case S_IFDIR:
1312 		host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1313 		break;
1314 	case S_IFCHR:
1315 	case S_IFBLK:
1316 	case S_IFIFO:
1317 	case S_IFSOCK:
1318 		host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1319 		break;
1320 	}
1321 	if (host_err < 0) {
1322 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1323 		goto out_nfserr;
1324 	}
1325 
1326 	if (EX_ISSYNC(fhp->fh_export)) {
1327 		err = nfserrno(nfsd_sync_dir(dentry));
1328 		write_inode_now(dchild->d_inode, 1);
1329 	}
1330 
1331 	err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1332 	if (err2)
1333 		err = err2;
1334 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1335 	/*
1336 	 * Update the file handle to get the new inode info.
1337 	 */
1338 	if (!err)
1339 		err = fh_update(resfhp);
1340 out:
1341 	if (dchild && !IS_ERR(dchild))
1342 		dput(dchild);
1343 	return err;
1344 
1345 out_nfserr:
1346 	err = nfserrno(host_err);
1347 	goto out;
1348 }
1349 
1350 #ifdef CONFIG_NFSD_V3
1351 /*
1352  * NFSv3 version of nfsd_create
1353  */
1354 __be32
1355 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1356 		char *fname, int flen, struct iattr *iap,
1357 		struct svc_fh *resfhp, int createmode, u32 *verifier,
1358 	        int *truncp, int *created)
1359 {
1360 	struct dentry	*dentry, *dchild = NULL;
1361 	struct inode	*dirp;
1362 	__be32		err;
1363 	__be32		err2;
1364 	int		host_err;
1365 	__u32		v_mtime=0, v_atime=0;
1366 
1367 	err = nfserr_perm;
1368 	if (!flen)
1369 		goto out;
1370 	err = nfserr_exist;
1371 	if (isdotent(fname, flen))
1372 		goto out;
1373 	if (!(iap->ia_valid & ATTR_MODE))
1374 		iap->ia_mode = 0;
1375 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1376 	if (err)
1377 		goto out;
1378 
1379 	dentry = fhp->fh_dentry;
1380 	dirp = dentry->d_inode;
1381 
1382 	/* Get all the sanity checks out of the way before
1383 	 * we lock the parent. */
1384 	err = nfserr_notdir;
1385 	if (!dirp->i_op->lookup)
1386 		goto out;
1387 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1388 
1389 	/*
1390 	 * Compose the response file handle.
1391 	 */
1392 	dchild = lookup_one_len(fname, dentry, flen);
1393 	host_err = PTR_ERR(dchild);
1394 	if (IS_ERR(dchild))
1395 		goto out_nfserr;
1396 
1397 	err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1398 	if (err)
1399 		goto out;
1400 
1401 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1402 		/* solaris7 gets confused (bugid 4218508) if these have
1403 		 * the high bit set, so just clear the high bits. If this is
1404 		 * ever changed to use different attrs for storing the
1405 		 * verifier, then do_open_lookup() will also need to be fixed
1406 		 * accordingly.
1407 		 */
1408 		v_mtime = verifier[0]&0x7fffffff;
1409 		v_atime = verifier[1]&0x7fffffff;
1410 	}
1411 
1412 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1413 	if (host_err)
1414 		goto out_nfserr;
1415 	if (dchild->d_inode) {
1416 		err = 0;
1417 
1418 		switch (createmode) {
1419 		case NFS3_CREATE_UNCHECKED:
1420 			if (! S_ISREG(dchild->d_inode->i_mode))
1421 				err = nfserr_exist;
1422 			else if (truncp) {
1423 				/* in nfsv4, we need to treat this case a little
1424 				 * differently.  we don't want to truncate the
1425 				 * file now; this would be wrong if the OPEN
1426 				 * fails for some other reason.  furthermore,
1427 				 * if the size is nonzero, we should ignore it
1428 				 * according to spec!
1429 				 */
1430 				*truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1431 			}
1432 			else {
1433 				iap->ia_valid &= ATTR_SIZE;
1434 				goto set_attr;
1435 			}
1436 			break;
1437 		case NFS3_CREATE_EXCLUSIVE:
1438 			if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1439 			    && dchild->d_inode->i_atime.tv_sec == v_atime
1440 			    && dchild->d_inode->i_size  == 0 )
1441 				break;
1442 			 /* fallthru */
1443 		case NFS3_CREATE_GUARDED:
1444 			err = nfserr_exist;
1445 		}
1446 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1447 		goto out;
1448 	}
1449 
1450 	host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1451 	if (host_err < 0) {
1452 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1453 		goto out_nfserr;
1454 	}
1455 	if (created)
1456 		*created = 1;
1457 
1458 	if (EX_ISSYNC(fhp->fh_export)) {
1459 		err = nfserrno(nfsd_sync_dir(dentry));
1460 		/* setattr will sync the child (or not) */
1461 	}
1462 
1463 	nfsd_check_ignore_resizing(iap);
1464 
1465 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1466 		/* Cram the verifier into atime/mtime */
1467 		iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1468 			| ATTR_MTIME_SET|ATTR_ATIME_SET;
1469 		/* XXX someone who knows this better please fix it for nsec */
1470 		iap->ia_mtime.tv_sec = v_mtime;
1471 		iap->ia_atime.tv_sec = v_atime;
1472 		iap->ia_mtime.tv_nsec = 0;
1473 		iap->ia_atime.tv_nsec = 0;
1474 	}
1475 
1476  set_attr:
1477 	err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1478 	if (err2)
1479 		err = err2;
1480 
1481 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1482 	/*
1483 	 * Update the filehandle to get the new inode info.
1484 	 */
1485 	if (!err)
1486 		err = fh_update(resfhp);
1487 
1488  out:
1489 	fh_unlock(fhp);
1490 	if (dchild && !IS_ERR(dchild))
1491 		dput(dchild);
1492  	return err;
1493 
1494  out_nfserr:
1495 	err = nfserrno(host_err);
1496 	goto out;
1497 }
1498 #endif /* CONFIG_NFSD_V3 */
1499 
1500 /*
1501  * Read a symlink. On entry, *lenp must contain the maximum path length that
1502  * fits into the buffer. On return, it contains the true length.
1503  * N.B. After this call fhp needs an fh_put
1504  */
1505 __be32
1506 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1507 {
1508 	struct dentry	*dentry;
1509 	struct inode	*inode;
1510 	mm_segment_t	oldfs;
1511 	__be32		err;
1512 	int		host_err;
1513 
1514 	err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1515 	if (err)
1516 		goto out;
1517 
1518 	dentry = fhp->fh_dentry;
1519 	inode = dentry->d_inode;
1520 
1521 	err = nfserr_inval;
1522 	if (!inode->i_op->readlink)
1523 		goto out;
1524 
1525 	touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1526 	/* N.B. Why does this call need a get_fs()??
1527 	 * Remove the set_fs and watch the fireworks:-) --okir
1528 	 */
1529 
1530 	oldfs = get_fs(); set_fs(KERNEL_DS);
1531 	host_err = inode->i_op->readlink(dentry, buf, *lenp);
1532 	set_fs(oldfs);
1533 
1534 	if (host_err < 0)
1535 		goto out_nfserr;
1536 	*lenp = host_err;
1537 	err = 0;
1538 out:
1539 	return err;
1540 
1541 out_nfserr:
1542 	err = nfserrno(host_err);
1543 	goto out;
1544 }
1545 
1546 /*
1547  * Create a symlink and look up its inode
1548  * N.B. After this call _both_ fhp and resfhp need an fh_put
1549  */
1550 __be32
1551 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1552 				char *fname, int flen,
1553 				char *path,  int plen,
1554 				struct svc_fh *resfhp,
1555 				struct iattr *iap)
1556 {
1557 	struct dentry	*dentry, *dnew;
1558 	__be32		err, cerr;
1559 	int		host_err;
1560 
1561 	err = nfserr_noent;
1562 	if (!flen || !plen)
1563 		goto out;
1564 	err = nfserr_exist;
1565 	if (isdotent(fname, flen))
1566 		goto out;
1567 
1568 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1569 	if (err)
1570 		goto out;
1571 	fh_lock(fhp);
1572 	dentry = fhp->fh_dentry;
1573 	dnew = lookup_one_len(fname, dentry, flen);
1574 	host_err = PTR_ERR(dnew);
1575 	if (IS_ERR(dnew))
1576 		goto out_nfserr;
1577 
1578 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1579 	if (host_err)
1580 		goto out_nfserr;
1581 
1582 	if (unlikely(path[plen] != 0)) {
1583 		char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1584 		if (path_alloced == NULL)
1585 			host_err = -ENOMEM;
1586 		else {
1587 			strncpy(path_alloced, path, plen);
1588 			path_alloced[plen] = 0;
1589 			host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1590 			kfree(path_alloced);
1591 		}
1592 	} else
1593 		host_err = vfs_symlink(dentry->d_inode, dnew, path);
1594 
1595 	if (!host_err) {
1596 		if (EX_ISSYNC(fhp->fh_export))
1597 			host_err = nfsd_sync_dir(dentry);
1598 	}
1599 	err = nfserrno(host_err);
1600 	fh_unlock(fhp);
1601 
1602 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1603 
1604 	cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1605 	dput(dnew);
1606 	if (err==0) err = cerr;
1607 out:
1608 	return err;
1609 
1610 out_nfserr:
1611 	err = nfserrno(host_err);
1612 	goto out;
1613 }
1614 
1615 /*
1616  * Create a hardlink
1617  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1618  */
1619 __be32
1620 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1621 				char *name, int len, struct svc_fh *tfhp)
1622 {
1623 	struct dentry	*ddir, *dnew, *dold;
1624 	struct inode	*dirp, *dest;
1625 	__be32		err;
1626 	int		host_err;
1627 
1628 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1629 	if (err)
1630 		goto out;
1631 	err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1632 	if (err)
1633 		goto out;
1634 
1635 	err = nfserr_perm;
1636 	if (!len)
1637 		goto out;
1638 	err = nfserr_exist;
1639 	if (isdotent(name, len))
1640 		goto out;
1641 
1642 	fh_lock_nested(ffhp, I_MUTEX_PARENT);
1643 	ddir = ffhp->fh_dentry;
1644 	dirp = ddir->d_inode;
1645 
1646 	dnew = lookup_one_len(name, ddir, len);
1647 	host_err = PTR_ERR(dnew);
1648 	if (IS_ERR(dnew))
1649 		goto out_nfserr;
1650 
1651 	dold = tfhp->fh_dentry;
1652 	dest = dold->d_inode;
1653 
1654 	host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1655 	if (host_err) {
1656 		err = nfserrno(host_err);
1657 		goto out_dput;
1658 	}
1659 	host_err = vfs_link(dold, dirp, dnew);
1660 	if (!host_err) {
1661 		if (EX_ISSYNC(ffhp->fh_export)) {
1662 			err = nfserrno(nfsd_sync_dir(ddir));
1663 			write_inode_now(dest, 1);
1664 		}
1665 		err = 0;
1666 	} else {
1667 		if (host_err == -EXDEV && rqstp->rq_vers == 2)
1668 			err = nfserr_acces;
1669 		else
1670 			err = nfserrno(host_err);
1671 	}
1672 	mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1673 out_dput:
1674 	dput(dnew);
1675 out_unlock:
1676 	fh_unlock(ffhp);
1677 out:
1678 	return err;
1679 
1680 out_nfserr:
1681 	err = nfserrno(host_err);
1682 	goto out_unlock;
1683 }
1684 
1685 /*
1686  * Rename a file
1687  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1688  */
1689 __be32
1690 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1691 			    struct svc_fh *tfhp, char *tname, int tlen)
1692 {
1693 	struct dentry	*fdentry, *tdentry, *odentry, *ndentry, *trap;
1694 	struct inode	*fdir, *tdir;
1695 	__be32		err;
1696 	int		host_err;
1697 
1698 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1699 	if (err)
1700 		goto out;
1701 	err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1702 	if (err)
1703 		goto out;
1704 
1705 	fdentry = ffhp->fh_dentry;
1706 	fdir = fdentry->d_inode;
1707 
1708 	tdentry = tfhp->fh_dentry;
1709 	tdir = tdentry->d_inode;
1710 
1711 	err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1712 	if (ffhp->fh_export != tfhp->fh_export)
1713 		goto out;
1714 
1715 	err = nfserr_perm;
1716 	if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1717 		goto out;
1718 
1719 	/* cannot use fh_lock as we need deadlock protective ordering
1720 	 * so do it by hand */
1721 	trap = lock_rename(tdentry, fdentry);
1722 	ffhp->fh_locked = tfhp->fh_locked = 1;
1723 	fill_pre_wcc(ffhp);
1724 	fill_pre_wcc(tfhp);
1725 
1726 	odentry = lookup_one_len(fname, fdentry, flen);
1727 	host_err = PTR_ERR(odentry);
1728 	if (IS_ERR(odentry))
1729 		goto out_nfserr;
1730 
1731 	host_err = -ENOENT;
1732 	if (!odentry->d_inode)
1733 		goto out_dput_old;
1734 	host_err = -EINVAL;
1735 	if (odentry == trap)
1736 		goto out_dput_old;
1737 
1738 	ndentry = lookup_one_len(tname, tdentry, tlen);
1739 	host_err = PTR_ERR(ndentry);
1740 	if (IS_ERR(ndentry))
1741 		goto out_dput_old;
1742 	host_err = -ENOTEMPTY;
1743 	if (ndentry == trap)
1744 		goto out_dput_new;
1745 
1746 	if (svc_msnfs(ffhp) &&
1747 		((atomic_read(&odentry->d_count) > 1)
1748 		 || (atomic_read(&ndentry->d_count) > 1))) {
1749 			host_err = -EPERM;
1750 			goto out_dput_new;
1751 	}
1752 
1753 	host_err = -EXDEV;
1754 	if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1755 		goto out_dput_new;
1756 	host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1757 	if (host_err)
1758 		goto out_dput_new;
1759 
1760 	host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1761 	if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1762 		host_err = nfsd_sync_dir(tdentry);
1763 		if (!host_err)
1764 			host_err = nfsd_sync_dir(fdentry);
1765 	}
1766 
1767 	mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1768 
1769  out_dput_new:
1770 	dput(ndentry);
1771  out_dput_old:
1772 	dput(odentry);
1773  out_nfserr:
1774 	err = nfserrno(host_err);
1775 
1776 	/* we cannot reply on fh_unlock on the two filehandles,
1777 	 * as that would do the wrong thing if the two directories
1778 	 * were the same, so again we do it by hand
1779 	 */
1780 	fill_post_wcc(ffhp);
1781 	fill_post_wcc(tfhp);
1782 	unlock_rename(tdentry, fdentry);
1783 	ffhp->fh_locked = tfhp->fh_locked = 0;
1784 
1785 out:
1786 	return err;
1787 }
1788 
1789 /*
1790  * Unlink a file or directory
1791  * N.B. After this call fhp needs an fh_put
1792  */
1793 __be32
1794 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1795 				char *fname, int flen)
1796 {
1797 	struct dentry	*dentry, *rdentry;
1798 	struct inode	*dirp;
1799 	__be32		err;
1800 	int		host_err;
1801 
1802 	err = nfserr_acces;
1803 	if (!flen || isdotent(fname, flen))
1804 		goto out;
1805 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1806 	if (err)
1807 		goto out;
1808 
1809 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1810 	dentry = fhp->fh_dentry;
1811 	dirp = dentry->d_inode;
1812 
1813 	rdentry = lookup_one_len(fname, dentry, flen);
1814 	host_err = PTR_ERR(rdentry);
1815 	if (IS_ERR(rdentry))
1816 		goto out_nfserr;
1817 
1818 	if (!rdentry->d_inode) {
1819 		dput(rdentry);
1820 		err = nfserr_noent;
1821 		goto out;
1822 	}
1823 
1824 	if (!type)
1825 		type = rdentry->d_inode->i_mode & S_IFMT;
1826 
1827 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1828 	if (host_err)
1829 		goto out_nfserr;
1830 
1831 	if (type != S_IFDIR) { /* It's UNLINK */
1832 #ifdef MSNFS
1833 		if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1834 			(atomic_read(&rdentry->d_count) > 1)) {
1835 			host_err = -EPERM;
1836 		} else
1837 #endif
1838 		host_err = vfs_unlink(dirp, rdentry);
1839 	} else { /* It's RMDIR */
1840 		host_err = vfs_rmdir(dirp, rdentry);
1841 	}
1842 
1843 	dput(rdentry);
1844 
1845 	if (host_err)
1846 		goto out_drop;
1847 	if (EX_ISSYNC(fhp->fh_export))
1848 		host_err = nfsd_sync_dir(dentry);
1849 
1850 out_drop:
1851 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1852 out_nfserr:
1853 	err = nfserrno(host_err);
1854 out:
1855 	return err;
1856 }
1857 
1858 /*
1859  * We do this buffering because we must not call back into the file
1860  * system's ->lookup() method from the filldir callback. That may well
1861  * deadlock a number of file systems.
1862  *
1863  * This is based heavily on the implementation of same in XFS.
1864  */
1865 struct buffered_dirent {
1866 	u64		ino;
1867 	loff_t		offset;
1868 	int		namlen;
1869 	unsigned int	d_type;
1870 	char		name[];
1871 };
1872 
1873 struct readdir_data {
1874 	char		*dirent;
1875 	size_t		used;
1876 	int		full;
1877 };
1878 
1879 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1880 				 loff_t offset, u64 ino, unsigned int d_type)
1881 {
1882 	struct readdir_data *buf = __buf;
1883 	struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1884 	unsigned int reclen;
1885 
1886 	reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1887 	if (buf->used + reclen > PAGE_SIZE) {
1888 		buf->full = 1;
1889 		return -EINVAL;
1890 	}
1891 
1892 	de->namlen = namlen;
1893 	de->offset = offset;
1894 	de->ino = ino;
1895 	de->d_type = d_type;
1896 	memcpy(de->name, name, namlen);
1897 	buf->used += reclen;
1898 
1899 	return 0;
1900 }
1901 
1902 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1903 				    struct readdir_cd *cdp, loff_t *offsetp)
1904 {
1905 	struct readdir_data buf;
1906 	struct buffered_dirent *de;
1907 	int host_err;
1908 	int size;
1909 	loff_t offset;
1910 
1911 	buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1912 	if (!buf.dirent)
1913 		return nfserrno(-ENOMEM);
1914 
1915 	offset = *offsetp;
1916 
1917 	while (1) {
1918 		struct inode *dir_inode = file->f_path.dentry->d_inode;
1919 		unsigned int reclen;
1920 
1921 		cdp->err = nfserr_eof; /* will be cleared on successful read */
1922 		buf.used = 0;
1923 		buf.full = 0;
1924 
1925 		host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1926 		if (buf.full)
1927 			host_err = 0;
1928 
1929 		if (host_err < 0)
1930 			break;
1931 
1932 		size = buf.used;
1933 
1934 		if (!size)
1935 			break;
1936 
1937 		/*
1938 		 * Various filldir functions may end up calling back into
1939 		 * lookup_one_len() and the file system's ->lookup() method.
1940 		 * These expect i_mutex to be held, as it would within readdir.
1941 		 */
1942 		host_err = mutex_lock_killable(&dir_inode->i_mutex);
1943 		if (host_err)
1944 			break;
1945 
1946 		de = (struct buffered_dirent *)buf.dirent;
1947 		while (size > 0) {
1948 			offset = de->offset;
1949 
1950 			if (func(cdp, de->name, de->namlen, de->offset,
1951 				 de->ino, de->d_type))
1952 				break;
1953 
1954 			if (cdp->err != nfs_ok)
1955 				break;
1956 
1957 			reclen = ALIGN(sizeof(*de) + de->namlen,
1958 				       sizeof(u64));
1959 			size -= reclen;
1960 			de = (struct buffered_dirent *)((char *)de + reclen);
1961 		}
1962 		mutex_unlock(&dir_inode->i_mutex);
1963 		if (size > 0) /* We bailed out early */
1964 			break;
1965 
1966 		offset = vfs_llseek(file, 0, SEEK_CUR);
1967 	}
1968 
1969 	free_page((unsigned long)(buf.dirent));
1970 
1971 	if (host_err)
1972 		return nfserrno(host_err);
1973 
1974 	*offsetp = offset;
1975 	return cdp->err;
1976 }
1977 
1978 /*
1979  * Read entries from a directory.
1980  * The  NFSv3/4 verifier we ignore for now.
1981  */
1982 __be32
1983 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1984 	     struct readdir_cd *cdp, filldir_t func)
1985 {
1986 	__be32		err;
1987 	struct file	*file;
1988 	loff_t		offset = *offsetp;
1989 
1990 	err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1991 	if (err)
1992 		goto out;
1993 
1994 	offset = vfs_llseek(file, offset, 0);
1995 	if (offset < 0) {
1996 		err = nfserrno((int)offset);
1997 		goto out_close;
1998 	}
1999 
2000 	err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2001 
2002 	if (err == nfserr_eof || err == nfserr_toosmall)
2003 		err = nfs_ok; /* can still be found in ->err */
2004 out_close:
2005 	nfsd_close(file);
2006 out:
2007 	return err;
2008 }
2009 
2010 /*
2011  * Get file system stats
2012  * N.B. After this call fhp needs an fh_put
2013  */
2014 __be32
2015 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2016 {
2017 	__be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2018 	if (!err && vfs_statfs(fhp->fh_dentry,stat))
2019 		err = nfserr_io;
2020 	return err;
2021 }
2022 
2023 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2024 {
2025 	return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2026 }
2027 
2028 /*
2029  * Check for a user's access permissions to this inode.
2030  */
2031 __be32
2032 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2033 					struct dentry *dentry, int acc)
2034 {
2035 	struct inode	*inode = dentry->d_inode;
2036 	struct path	path;
2037 	int		err;
2038 
2039 	if (acc == NFSD_MAY_NOP)
2040 		return 0;
2041 #if 0
2042 	dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2043 		acc,
2044 		(acc & NFSD_MAY_READ)?	" read"  : "",
2045 		(acc & NFSD_MAY_WRITE)?	" write" : "",
2046 		(acc & NFSD_MAY_EXEC)?	" exec"  : "",
2047 		(acc & NFSD_MAY_SATTR)?	" sattr" : "",
2048 		(acc & NFSD_MAY_TRUNC)?	" trunc" : "",
2049 		(acc & NFSD_MAY_LOCK)?	" lock"  : "",
2050 		(acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2051 		inode->i_mode,
2052 		IS_IMMUTABLE(inode)?	" immut" : "",
2053 		IS_APPEND(inode)?	" append" : "",
2054 		__mnt_is_readonly(exp->ex_path.mnt)?	" ro" : "");
2055 	dprintk("      owner %d/%d user %d/%d\n",
2056 		inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2057 #endif
2058 
2059 	/* Normally we reject any write/sattr etc access on a read-only file
2060 	 * system.  But if it is IRIX doing check on write-access for a
2061 	 * device special file, we ignore rofs.
2062 	 */
2063 	if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2064 		if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2065 			if (exp_rdonly(rqstp, exp) ||
2066 			    __mnt_is_readonly(exp->ex_path.mnt))
2067 				return nfserr_rofs;
2068 			if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2069 				return nfserr_perm;
2070 		}
2071 	if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2072 		return nfserr_perm;
2073 
2074 	if (acc & NFSD_MAY_LOCK) {
2075 		/* If we cannot rely on authentication in NLM requests,
2076 		 * just allow locks, otherwise require read permission, or
2077 		 * ownership
2078 		 */
2079 		if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2080 			return 0;
2081 		else
2082 			acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2083 	}
2084 	/*
2085 	 * The file owner always gets access permission for accesses that
2086 	 * would normally be checked at open time. This is to make
2087 	 * file access work even when the client has done a fchmod(fd, 0).
2088 	 *
2089 	 * However, `cp foo bar' should fail nevertheless when bar is
2090 	 * readonly. A sensible way to do this might be to reject all
2091 	 * attempts to truncate a read-only file, because a creat() call
2092 	 * always implies file truncation.
2093 	 * ... but this isn't really fair.  A process may reasonably call
2094 	 * ftruncate on an open file descriptor on a file with perm 000.
2095 	 * We must trust the client to do permission checking - using "ACCESS"
2096 	 * with NFSv3.
2097 	 */
2098 	if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2099 	    inode->i_uid == current_fsuid())
2100 		return 0;
2101 
2102 	/* This assumes  NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2103 	err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2104 
2105 	/* Allow read access to binaries even when mode 111 */
2106 	if (err == -EACCES && S_ISREG(inode->i_mode) &&
2107 	    acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2108 		err = inode_permission(inode, MAY_EXEC);
2109 	if (err)
2110 		goto nfsd_out;
2111 
2112 	/* Do integrity (permission) checking now, but defer incrementing
2113 	 * IMA counts to the actual file open.
2114 	 */
2115 	path.mnt = exp->ex_path.mnt;
2116 	path.dentry = dentry;
2117 	err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC),
2118 			     IMA_COUNT_LEAVE);
2119 nfsd_out:
2120 	return err? nfserrno(err) : 0;
2121 }
2122 
2123 void
2124 nfsd_racache_shutdown(void)
2125 {
2126 	struct raparms *raparm, *last_raparm;
2127 	unsigned int i;
2128 
2129 	dprintk("nfsd: freeing readahead buffers.\n");
2130 
2131 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2132 		raparm = raparm_hash[i].pb_head;
2133 		while(raparm) {
2134 			last_raparm = raparm;
2135 			raparm = raparm->p_next;
2136 			kfree(last_raparm);
2137 		}
2138 		raparm_hash[i].pb_head = NULL;
2139 	}
2140 }
2141 /*
2142  * Initialize readahead param cache
2143  */
2144 int
2145 nfsd_racache_init(int cache_size)
2146 {
2147 	int	i;
2148 	int	j = 0;
2149 	int	nperbucket;
2150 	struct raparms **raparm = NULL;
2151 
2152 
2153 	if (raparm_hash[0].pb_head)
2154 		return 0;
2155 	nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2156 	if (nperbucket < 2)
2157 		nperbucket = 2;
2158 	cache_size = nperbucket * RAPARM_HASH_SIZE;
2159 
2160 	dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2161 
2162 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2163 		spin_lock_init(&raparm_hash[i].pb_lock);
2164 
2165 		raparm = &raparm_hash[i].pb_head;
2166 		for (j = 0; j < nperbucket; j++) {
2167 			*raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2168 			if (!*raparm)
2169 				goto out_nomem;
2170 			raparm = &(*raparm)->p_next;
2171 		}
2172 		*raparm = NULL;
2173 	}
2174 
2175 	nfsdstats.ra_size = cache_size;
2176 	return 0;
2177 
2178 out_nomem:
2179 	dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2180 	nfsd_racache_shutdown();
2181 	return -ENOMEM;
2182 }
2183 
2184 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2185 struct posix_acl *
2186 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2187 {
2188 	struct inode *inode = fhp->fh_dentry->d_inode;
2189 	char *name;
2190 	void *value = NULL;
2191 	ssize_t size;
2192 	struct posix_acl *acl;
2193 
2194 	if (!IS_POSIXACL(inode))
2195 		return ERR_PTR(-EOPNOTSUPP);
2196 
2197 	switch (type) {
2198 	case ACL_TYPE_ACCESS:
2199 		name = POSIX_ACL_XATTR_ACCESS;
2200 		break;
2201 	case ACL_TYPE_DEFAULT:
2202 		name = POSIX_ACL_XATTR_DEFAULT;
2203 		break;
2204 	default:
2205 		return ERR_PTR(-EOPNOTSUPP);
2206 	}
2207 
2208 	size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2209 	if (size < 0)
2210 		return ERR_PTR(size);
2211 
2212 	acl = posix_acl_from_xattr(value, size);
2213 	kfree(value);
2214 	return acl;
2215 }
2216 
2217 int
2218 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2219 {
2220 	struct inode *inode = fhp->fh_dentry->d_inode;
2221 	char *name;
2222 	void *value = NULL;
2223 	size_t size;
2224 	int error;
2225 
2226 	if (!IS_POSIXACL(inode) ||
2227 	    !inode->i_op->setxattr || !inode->i_op->removexattr)
2228 		return -EOPNOTSUPP;
2229 	switch(type) {
2230 		case ACL_TYPE_ACCESS:
2231 			name = POSIX_ACL_XATTR_ACCESS;
2232 			break;
2233 		case ACL_TYPE_DEFAULT:
2234 			name = POSIX_ACL_XATTR_DEFAULT;
2235 			break;
2236 		default:
2237 			return -EOPNOTSUPP;
2238 	}
2239 
2240 	if (acl && acl->a_count) {
2241 		size = posix_acl_xattr_size(acl->a_count);
2242 		value = kmalloc(size, GFP_KERNEL);
2243 		if (!value)
2244 			return -ENOMEM;
2245 		error = posix_acl_to_xattr(acl, value, size);
2246 		if (error < 0)
2247 			goto getout;
2248 		size = error;
2249 	} else
2250 		size = 0;
2251 
2252 	error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2253 	if (error)
2254 		goto getout;
2255 	if (size)
2256 		error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2257 	else {
2258 		if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2259 			error = 0;
2260 		else {
2261 			error = vfs_removexattr(fhp->fh_dentry, name);
2262 			if (error == -ENODATA)
2263 				error = 0;
2264 		}
2265 	}
2266 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
2267 
2268 getout:
2269 	kfree(value);
2270 	return error;
2271 }
2272 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
2273