xref: /linux/fs/nfsd/vfs.c (revision 14b42963f64b98ab61fa9723c03d71aa5ef4f862)
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/ext2_fs.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 
58 #include <asm/uaccess.h>
59 
60 #define NFSDDBG_FACILITY		NFSDDBG_FILEOP
61 #define NFSD_PARANOIA
62 
63 
64 /* We must ignore files (but only files) which might have mandatory
65  * locks on them because there is no way to know if the accesser has
66  * the lock.
67  */
68 #define IS_ISMNDLK(i)	(S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
69 
70 /*
71  * This is a cache of readahead params that help us choose the proper
72  * readahead strategy. Initially, we set all readahead parameters to 0
73  * and let the VFS handle things.
74  * If you increase the number of cached files very much, you'll need to
75  * add a hash table here.
76  */
77 struct raparms {
78 	struct raparms		*p_next;
79 	unsigned int		p_count;
80 	ino_t			p_ino;
81 	dev_t			p_dev;
82 	int			p_set;
83 	struct file_ra_state	p_ra;
84 };
85 
86 static struct raparms *		raparml;
87 static struct raparms *		raparm_cache;
88 
89 /*
90  * Called from nfsd_lookup and encode_dirent. Check if we have crossed
91  * a mount point.
92  * Returns -EAGAIN leaving *dpp and *expp unchanged,
93  *  or nfs_ok having possibly changed *dpp and *expp
94  */
95 int
96 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
97 		        struct svc_export **expp)
98 {
99 	struct svc_export *exp = *expp, *exp2 = NULL;
100 	struct dentry *dentry = *dpp;
101 	struct vfsmount *mnt = mntget(exp->ex_mnt);
102 	struct dentry *mounts = dget(dentry);
103 	int err = nfs_ok;
104 
105 	while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
106 
107 	exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
108 	if (IS_ERR(exp2)) {
109 		err = PTR_ERR(exp2);
110 		dput(mounts);
111 		mntput(mnt);
112 		goto out;
113 	}
114 	if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
115 		/* successfully crossed mount point */
116 		exp_put(exp);
117 		*expp = exp2;
118 		dput(dentry);
119 		*dpp = mounts;
120 	} else {
121 		if (exp2) exp_put(exp2);
122 		dput(mounts);
123 	}
124 	mntput(mnt);
125 out:
126 	return err;
127 }
128 
129 /*
130  * Look up one component of a pathname.
131  * N.B. After this call _both_ fhp and resfh need an fh_put
132  *
133  * If the lookup would cross a mountpoint, and the mounted filesystem
134  * is exported to the client with NFSEXP_NOHIDE, then the lookup is
135  * accepted as it stands and the mounted directory is
136  * returned. Otherwise the covered directory is returned.
137  * NOTE: this mountpoint crossing is not supported properly by all
138  *   clients and is explicitly disallowed for NFSv3
139  *      NeilBrown <neilb@cse.unsw.edu.au>
140  */
141 int
142 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
143 					int len, struct svc_fh *resfh)
144 {
145 	struct svc_export	*exp;
146 	struct dentry		*dparent;
147 	struct dentry		*dentry;
148 	int			err;
149 
150 	dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
151 
152 	/* Obtain dentry and export. */
153 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
154 	if (err)
155 		return err;
156 
157 	dparent = fhp->fh_dentry;
158 	exp  = fhp->fh_export;
159 	exp_get(exp);
160 
161 	err = nfserr_acces;
162 
163 	/* Lookup the name, but don't follow links */
164 	if (isdotent(name, len)) {
165 		if (len==1)
166 			dentry = dget(dparent);
167 		else if (dparent != exp->ex_dentry) {
168 			dentry = dget_parent(dparent);
169 		} else  if (!EX_NOHIDE(exp))
170 			dentry = dget(dparent); /* .. == . just like at / */
171 		else {
172 			/* checking mountpoint crossing is very different when stepping up */
173 			struct svc_export *exp2 = NULL;
174 			struct dentry *dp;
175 			struct vfsmount *mnt = mntget(exp->ex_mnt);
176 			dentry = dget(dparent);
177 			while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
178 				;
179 			dp = dget_parent(dentry);
180 			dput(dentry);
181 			dentry = dp;
182 
183 			exp2 = exp_parent(exp->ex_client, mnt, dentry,
184 					  &rqstp->rq_chandle);
185 			if (IS_ERR(exp2)) {
186 				err = PTR_ERR(exp2);
187 				dput(dentry);
188 				mntput(mnt);
189 				goto out_nfserr;
190 			}
191 			if (!exp2) {
192 				dput(dentry);
193 				dentry = dget(dparent);
194 			} else {
195 				exp_put(exp);
196 				exp = exp2;
197 			}
198 			mntput(mnt);
199 		}
200 	} else {
201 		fh_lock(fhp);
202 		dentry = lookup_one_len(name, dparent, len);
203 		err = PTR_ERR(dentry);
204 		if (IS_ERR(dentry))
205 			goto out_nfserr;
206 		/*
207 		 * check if we have crossed a mount point ...
208 		 */
209 		if (d_mountpoint(dentry)) {
210 			if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
211 				dput(dentry);
212 				goto out_nfserr;
213 			}
214 		}
215 	}
216 	/*
217 	 * Note: we compose the file handle now, but as the
218 	 * dentry may be negative, it may need to be updated.
219 	 */
220 	err = fh_compose(resfh, exp, dentry, fhp);
221 	if (!err && !dentry->d_inode)
222 		err = nfserr_noent;
223 	dput(dentry);
224 out:
225 	exp_put(exp);
226 	return err;
227 
228 out_nfserr:
229 	err = nfserrno(err);
230 	goto out;
231 }
232 
233 /*
234  * Set various file attributes.
235  * N.B. After this call fhp needs an fh_put
236  */
237 int
238 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
239 	     int check_guard, time_t guardtime)
240 {
241 	struct dentry	*dentry;
242 	struct inode	*inode;
243 	int		accmode = MAY_SATTR;
244 	int		ftype = 0;
245 	int		imode;
246 	int		err;
247 	int		size_change = 0;
248 
249 	if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
250 		accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
251 	if (iap->ia_valid & ATTR_SIZE)
252 		ftype = S_IFREG;
253 
254 	/* Get inode */
255 	err = fh_verify(rqstp, fhp, ftype, accmode);
256 	if (err)
257 		goto out;
258 
259 	dentry = fhp->fh_dentry;
260 	inode = dentry->d_inode;
261 
262 	/* Ignore any mode updates on symlinks */
263 	if (S_ISLNK(inode->i_mode))
264 		iap->ia_valid &= ~ATTR_MODE;
265 
266 	if (!iap->ia_valid)
267 		goto out;
268 
269 	/* NFSv2 does not differentiate between "set-[ac]time-to-now"
270 	 * which only requires access, and "set-[ac]time-to-X" which
271 	 * requires ownership.
272 	 * So if it looks like it might be "set both to the same time which
273 	 * is close to now", and if inode_change_ok fails, then we
274 	 * convert to "set to now" instead of "set to explicit time"
275 	 *
276 	 * We only call inode_change_ok as the last test as technically
277 	 * it is not an interface that we should be using.  It is only
278 	 * valid if the filesystem does not define it's own i_op->setattr.
279 	 */
280 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
281 #define	MAX_TOUCH_TIME_ERROR (30*60)
282 	if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
283 	    && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
284 	    ) {
285 	    /* Looks probable.  Now just make sure time is in the right ballpark.
286 	     * Solaris, at least, doesn't seem to care what the time request is.
287 	     * We require it be within 30 minutes of now.
288 	     */
289 	    time_t delta = iap->ia_atime.tv_sec - get_seconds();
290 	    if (delta<0) delta = -delta;
291 	    if (delta < MAX_TOUCH_TIME_ERROR &&
292 		inode_change_ok(inode, iap) != 0) {
293 		/* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
294 		 * this will cause notify_change to set these times to "now"
295 		 */
296 		iap->ia_valid &= ~BOTH_TIME_SET;
297 	    }
298 	}
299 
300 	/* The size case is special. It changes the file as well as the attributes.  */
301 	if (iap->ia_valid & ATTR_SIZE) {
302 		if (iap->ia_size < inode->i_size) {
303 			err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
304 			if (err)
305 				goto out;
306 		}
307 
308 		/*
309 		 * If we are changing the size of the file, then
310 		 * we need to break all leases.
311 		 */
312 		err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
313 		if (err == -EWOULDBLOCK)
314 			err = -ETIMEDOUT;
315 		if (err) /* ENOMEM or EWOULDBLOCK */
316 			goto out_nfserr;
317 
318 		err = get_write_access(inode);
319 		if (err)
320 			goto out_nfserr;
321 
322 		size_change = 1;
323 		err = locks_verify_truncate(inode, NULL, iap->ia_size);
324 		if (err) {
325 			put_write_access(inode);
326 			goto out_nfserr;
327 		}
328 		DQUOT_INIT(inode);
329 	}
330 
331 	imode = inode->i_mode;
332 	if (iap->ia_valid & ATTR_MODE) {
333 		iap->ia_mode &= S_IALLUGO;
334 		imode = iap->ia_mode |= (imode & ~S_IALLUGO);
335 	}
336 
337 	/* Revoke setuid/setgid bit on chown/chgrp */
338 	if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
339 		iap->ia_valid |= ATTR_KILL_SUID;
340 	if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
341 		iap->ia_valid |= ATTR_KILL_SGID;
342 
343 	/* Change the attributes. */
344 
345 	iap->ia_valid |= ATTR_CTIME;
346 
347 	err = nfserr_notsync;
348 	if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
349 		fh_lock(fhp);
350 		err = notify_change(dentry, iap);
351 		err = nfserrno(err);
352 		fh_unlock(fhp);
353 	}
354 	if (size_change)
355 		put_write_access(inode);
356 	if (!err)
357 		if (EX_ISSYNC(fhp->fh_export))
358 			write_inode_now(inode, 1);
359 out:
360 	return err;
361 
362 out_nfserr:
363 	err = nfserrno(err);
364 	goto out;
365 }
366 
367 #if defined(CONFIG_NFSD_V2_ACL) || \
368     defined(CONFIG_NFSD_V3_ACL) || \
369     defined(CONFIG_NFSD_V4)
370 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
371 {
372 	ssize_t buflen;
373 
374 	buflen = vfs_getxattr(dentry, key, NULL, 0);
375 	if (buflen <= 0)
376 		return buflen;
377 
378 	*buf = kmalloc(buflen, GFP_KERNEL);
379 	if (!*buf)
380 		return -ENOMEM;
381 
382 	return vfs_getxattr(dentry, key, *buf, buflen);
383 }
384 #endif
385 
386 #if defined(CONFIG_NFSD_V4)
387 static int
388 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
389 {
390 	int len;
391 	size_t buflen;
392 	char *buf = NULL;
393 	int error = 0;
394 
395 	buflen = posix_acl_xattr_size(pacl->a_count);
396 	buf = kmalloc(buflen, GFP_KERNEL);
397 	error = -ENOMEM;
398 	if (buf == NULL)
399 		goto out;
400 
401 	len = posix_acl_to_xattr(pacl, buf, buflen);
402 	if (len < 0) {
403 		error = len;
404 		goto out;
405 	}
406 
407 	error = vfs_setxattr(dentry, key, buf, len, 0);
408 out:
409 	kfree(buf);
410 	return error;
411 }
412 
413 int
414 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
415     struct nfs4_acl *acl)
416 {
417 	int error;
418 	struct dentry *dentry;
419 	struct inode *inode;
420 	struct posix_acl *pacl = NULL, *dpacl = NULL;
421 	unsigned int flags = 0;
422 
423 	/* Get inode */
424 	error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
425 	if (error)
426 		goto out;
427 
428 	dentry = fhp->fh_dentry;
429 	inode = dentry->d_inode;
430 	if (S_ISDIR(inode->i_mode))
431 		flags = NFS4_ACL_DIR;
432 
433 	error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
434 	if (error == -EINVAL) {
435 		error = nfserr_attrnotsupp;
436 		goto out;
437 	} else if (error < 0)
438 		goto out_nfserr;
439 
440 	if (pacl) {
441 		error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
442 		if (error < 0)
443 			goto out_nfserr;
444 	}
445 
446 	if (dpacl) {
447 		error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
448 		if (error < 0)
449 			goto out_nfserr;
450 	}
451 
452 	error = nfs_ok;
453 
454 out:
455 	posix_acl_release(pacl);
456 	posix_acl_release(dpacl);
457 	return (error);
458 out_nfserr:
459 	error = nfserrno(error);
460 	goto out;
461 }
462 
463 static struct posix_acl *
464 _get_posix_acl(struct dentry *dentry, char *key)
465 {
466 	void *buf = NULL;
467 	struct posix_acl *pacl = NULL;
468 	int buflen;
469 
470 	buflen = nfsd_getxattr(dentry, key, &buf);
471 	if (!buflen)
472 		buflen = -ENODATA;
473 	if (buflen <= 0)
474 		return ERR_PTR(buflen);
475 
476 	pacl = posix_acl_from_xattr(buf, buflen);
477 	kfree(buf);
478 	return pacl;
479 }
480 
481 int
482 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
483 {
484 	struct inode *inode = dentry->d_inode;
485 	int error = 0;
486 	struct posix_acl *pacl = NULL, *dpacl = NULL;
487 	unsigned int flags = 0;
488 
489 	pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
490 	if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
491 		pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
492 	if (IS_ERR(pacl)) {
493 		error = PTR_ERR(pacl);
494 		pacl = NULL;
495 		goto out;
496 	}
497 
498 	if (S_ISDIR(inode->i_mode)) {
499 		dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
500 		if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
501 			dpacl = NULL;
502 		else if (IS_ERR(dpacl)) {
503 			error = PTR_ERR(dpacl);
504 			dpacl = NULL;
505 			goto out;
506 		}
507 		flags = NFS4_ACL_DIR;
508 	}
509 
510 	*acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
511 	if (IS_ERR(*acl)) {
512 		error = PTR_ERR(*acl);
513 		*acl = NULL;
514 	}
515  out:
516 	posix_acl_release(pacl);
517 	posix_acl_release(dpacl);
518 	return error;
519 }
520 
521 #endif /* defined(CONFIG_NFS_V4) */
522 
523 #ifdef CONFIG_NFSD_V3
524 /*
525  * Check server access rights to a file system object
526  */
527 struct accessmap {
528 	u32		access;
529 	int		how;
530 };
531 static struct accessmap	nfs3_regaccess[] = {
532     {	NFS3_ACCESS_READ,	MAY_READ			},
533     {	NFS3_ACCESS_EXECUTE,	MAY_EXEC			},
534     {	NFS3_ACCESS_MODIFY,	MAY_WRITE|MAY_TRUNC		},
535     {	NFS3_ACCESS_EXTEND,	MAY_WRITE			},
536 
537     {	0,			0				}
538 };
539 
540 static struct accessmap	nfs3_diraccess[] = {
541     {	NFS3_ACCESS_READ,	MAY_READ			},
542     {	NFS3_ACCESS_LOOKUP,	MAY_EXEC			},
543     {	NFS3_ACCESS_MODIFY,	MAY_EXEC|MAY_WRITE|MAY_TRUNC	},
544     {	NFS3_ACCESS_EXTEND,	MAY_EXEC|MAY_WRITE		},
545     {	NFS3_ACCESS_DELETE,	MAY_REMOVE			},
546 
547     {	0,			0				}
548 };
549 
550 static struct accessmap	nfs3_anyaccess[] = {
551 	/* Some clients - Solaris 2.6 at least, make an access call
552 	 * to the server to check for access for things like /dev/null
553 	 * (which really, the server doesn't care about).  So
554 	 * We provide simple access checking for them, looking
555 	 * mainly at mode bits, and we make sure to ignore read-only
556 	 * filesystem checks
557 	 */
558     {	NFS3_ACCESS_READ,	MAY_READ			},
559     {	NFS3_ACCESS_EXECUTE,	MAY_EXEC			},
560     {	NFS3_ACCESS_MODIFY,	MAY_WRITE|MAY_LOCAL_ACCESS	},
561     {	NFS3_ACCESS_EXTEND,	MAY_WRITE|MAY_LOCAL_ACCESS	},
562 
563     {	0,			0				}
564 };
565 
566 int
567 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
568 {
569 	struct accessmap	*map;
570 	struct svc_export	*export;
571 	struct dentry		*dentry;
572 	u32			query, result = 0, sresult = 0;
573 	unsigned int		error;
574 
575 	error = fh_verify(rqstp, fhp, 0, MAY_NOP);
576 	if (error)
577 		goto out;
578 
579 	export = fhp->fh_export;
580 	dentry = fhp->fh_dentry;
581 
582 	if (S_ISREG(dentry->d_inode->i_mode))
583 		map = nfs3_regaccess;
584 	else if (S_ISDIR(dentry->d_inode->i_mode))
585 		map = nfs3_diraccess;
586 	else
587 		map = nfs3_anyaccess;
588 
589 
590 	query = *access;
591 	for  (; map->access; map++) {
592 		if (map->access & query) {
593 			unsigned int err2;
594 
595 			sresult |= map->access;
596 
597 			err2 = nfsd_permission(export, dentry, map->how);
598 			switch (err2) {
599 			case nfs_ok:
600 				result |= map->access;
601 				break;
602 
603 			/* the following error codes just mean the access was not allowed,
604 			 * rather than an error occurred */
605 			case nfserr_rofs:
606 			case nfserr_acces:
607 			case nfserr_perm:
608 				/* simply don't "or" in the access bit. */
609 				break;
610 			default:
611 				error = err2;
612 				goto out;
613 			}
614 		}
615 	}
616 	*access = result;
617 	if (supported)
618 		*supported = sresult;
619 
620  out:
621 	return error;
622 }
623 #endif /* CONFIG_NFSD_V3 */
624 
625 
626 
627 /*
628  * Open an existing file or directory.
629  * The access argument indicates the type of open (read/write/lock)
630  * N.B. After this call fhp needs an fh_put
631  */
632 int
633 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
634 			int access, struct file **filp)
635 {
636 	struct dentry	*dentry;
637 	struct inode	*inode;
638 	int		flags = O_RDONLY|O_LARGEFILE, err;
639 
640 	/*
641 	 * If we get here, then the client has already done an "open",
642 	 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
643 	 * in case a chmod has now revoked permission.
644 	 */
645 	err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
646 	if (err)
647 		goto out;
648 
649 	dentry = fhp->fh_dentry;
650 	inode = dentry->d_inode;
651 
652 	/* Disallow write access to files with the append-only bit set
653 	 * or any access when mandatory locking enabled
654 	 */
655 	err = nfserr_perm;
656 	if (IS_APPEND(inode) && (access & MAY_WRITE))
657 		goto out;
658 	if (IS_ISMNDLK(inode))
659 		goto out;
660 
661 	if (!inode->i_fop)
662 		goto out;
663 
664 	/*
665 	 * Check to see if there are any leases on this file.
666 	 * This may block while leases are broken.
667 	 */
668 	err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
669 	if (err == -EWOULDBLOCK)
670 		err = -ETIMEDOUT;
671 	if (err) /* NOMEM or WOULDBLOCK */
672 		goto out_nfserr;
673 
674 	if (access & MAY_WRITE) {
675 		if (access & MAY_READ)
676 			flags = O_RDWR|O_LARGEFILE;
677 		else
678 			flags = O_WRONLY|O_LARGEFILE;
679 
680 		DQUOT_INIT(inode);
681 	}
682 	*filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
683 	if (IS_ERR(*filp))
684 		err = PTR_ERR(*filp);
685 out_nfserr:
686 	if (err)
687 		err = nfserrno(err);
688 out:
689 	return err;
690 }
691 
692 /*
693  * Close a file.
694  */
695 void
696 nfsd_close(struct file *filp)
697 {
698 	fput(filp);
699 }
700 
701 /*
702  * Sync a file
703  * As this calls fsync (not fdatasync) there is no need for a write_inode
704  * after it.
705  */
706 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
707 			      const struct file_operations *fop)
708 {
709 	struct inode *inode = dp->d_inode;
710 	int (*fsync) (struct file *, struct dentry *, int);
711 	int err;
712 
713 	err = filemap_fdatawrite(inode->i_mapping);
714 	if (err == 0 && fop && (fsync = fop->fsync))
715 		err = fsync(filp, dp, 0);
716 	if (err == 0)
717 		err = filemap_fdatawait(inode->i_mapping);
718 
719 	return err;
720 }
721 
722 
723 static int
724 nfsd_sync(struct file *filp)
725 {
726         int err;
727 	struct inode *inode = filp->f_dentry->d_inode;
728 	dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
729 	mutex_lock(&inode->i_mutex);
730 	err=nfsd_dosync(filp, filp->f_dentry, filp->f_op);
731 	mutex_unlock(&inode->i_mutex);
732 
733 	return err;
734 }
735 
736 int
737 nfsd_sync_dir(struct dentry *dp)
738 {
739 	return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
740 }
741 
742 /*
743  * Obtain the readahead parameters for the file
744  * specified by (dev, ino).
745  */
746 static DEFINE_SPINLOCK(ra_lock);
747 
748 static inline struct raparms *
749 nfsd_get_raparms(dev_t dev, ino_t ino)
750 {
751 	struct raparms	*ra, **rap, **frap = NULL;
752 	int depth = 0;
753 
754 	spin_lock(&ra_lock);
755 	for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
756 		if (ra->p_ino == ino && ra->p_dev == dev)
757 			goto found;
758 		depth++;
759 		if (ra->p_count == 0)
760 			frap = rap;
761 	}
762 	depth = nfsdstats.ra_size*11/10;
763 	if (!frap) {
764 		spin_unlock(&ra_lock);
765 		return NULL;
766 	}
767 	rap = frap;
768 	ra = *frap;
769 	ra->p_dev = dev;
770 	ra->p_ino = ino;
771 	ra->p_set = 0;
772 found:
773 	if (rap != &raparm_cache) {
774 		*rap = ra->p_next;
775 		ra->p_next   = raparm_cache;
776 		raparm_cache = ra;
777 	}
778 	ra->p_count++;
779 	nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
780 	spin_unlock(&ra_lock);
781 	return ra;
782 }
783 
784 /*
785  * Grab and keep cached pages assosiated with a file in the svc_rqst
786  * so that they can be passed to the netowork sendmsg/sendpage routines
787  * directrly. They will be released after the sending has completed.
788  */
789 static int
790 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
791 {
792 	unsigned long count = desc->count;
793 	struct svc_rqst *rqstp = desc->arg.data;
794 
795 	if (size > count)
796 		size = count;
797 
798 	if (rqstp->rq_res.page_len == 0) {
799 		get_page(page);
800 		rqstp->rq_respages[rqstp->rq_resused++] = page;
801 		rqstp->rq_res.page_base = offset;
802 		rqstp->rq_res.page_len = size;
803 	} else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
804 		get_page(page);
805 		rqstp->rq_respages[rqstp->rq_resused++] = page;
806 		rqstp->rq_res.page_len += size;
807 	} else {
808 		rqstp->rq_res.page_len += size;
809 	}
810 
811 	desc->count = count - size;
812 	desc->written += size;
813 	return size;
814 }
815 
816 static int
817 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
818               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
819 {
820 	struct inode *inode;
821 	struct raparms	*ra;
822 	mm_segment_t	oldfs;
823 	int		err;
824 
825 	err = nfserr_perm;
826 	inode = file->f_dentry->d_inode;
827 #ifdef MSNFS
828 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
829 		(!lock_may_read(inode, offset, *count)))
830 		goto out;
831 #endif
832 
833 	/* Get readahead parameters */
834 	ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
835 
836 	if (ra && ra->p_set)
837 		file->f_ra = ra->p_ra;
838 
839 	if (file->f_op->sendfile && rqstp->rq_sendfile_ok) {
840 		svc_pushback_unused_pages(rqstp);
841 		err = file->f_op->sendfile(file, &offset, *count,
842 						 nfsd_read_actor, rqstp);
843 	} else {
844 		oldfs = get_fs();
845 		set_fs(KERNEL_DS);
846 		err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
847 		set_fs(oldfs);
848 	}
849 
850 	/* Write back readahead params */
851 	if (ra) {
852 		spin_lock(&ra_lock);
853 		ra->p_ra = file->f_ra;
854 		ra->p_set = 1;
855 		ra->p_count--;
856 		spin_unlock(&ra_lock);
857 	}
858 
859 	if (err >= 0) {
860 		nfsdstats.io_read += err;
861 		*count = err;
862 		err = 0;
863 		fsnotify_access(file->f_dentry);
864 	} else
865 		err = nfserrno(err);
866 out:
867 	return err;
868 }
869 
870 static void kill_suid(struct dentry *dentry)
871 {
872 	struct iattr	ia;
873 	ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
874 
875 	mutex_lock(&dentry->d_inode->i_mutex);
876 	notify_change(dentry, &ia);
877 	mutex_unlock(&dentry->d_inode->i_mutex);
878 }
879 
880 static int
881 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
882 				loff_t offset, struct kvec *vec, int vlen,
883 	   			unsigned long cnt, int *stablep)
884 {
885 	struct svc_export	*exp;
886 	struct dentry		*dentry;
887 	struct inode		*inode;
888 	mm_segment_t		oldfs;
889 	int			err = 0;
890 	int			stable = *stablep;
891 
892 #ifdef MSNFS
893 	err = nfserr_perm;
894 
895 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
896 		(!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
897 		goto out;
898 #endif
899 
900 	dentry = file->f_dentry;
901 	inode = dentry->d_inode;
902 	exp   = fhp->fh_export;
903 
904 	/*
905 	 * Request sync writes if
906 	 *  -	the sync export option has been set, or
907 	 *  -	the client requested O_SYNC behavior (NFSv3 feature).
908 	 *  -   The file system doesn't support fsync().
909 	 * When gathered writes have been configured for this volume,
910 	 * flushing the data to disk is handled separately below.
911 	 */
912 
913 	if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
914 	       stable = 2;
915 	       *stablep = 2; /* FILE_SYNC */
916 	}
917 
918 	if (!EX_ISSYNC(exp))
919 		stable = 0;
920 	if (stable && !EX_WGATHER(exp))
921 		file->f_flags |= O_SYNC;
922 
923 	/* Write the data. */
924 	oldfs = get_fs(); set_fs(KERNEL_DS);
925 	err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
926 	set_fs(oldfs);
927 	if (err >= 0) {
928 		nfsdstats.io_write += cnt;
929 		fsnotify_modify(file->f_dentry);
930 	}
931 
932 	/* clear setuid/setgid flag after write */
933 	if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
934 		kill_suid(dentry);
935 
936 	if (err >= 0 && stable) {
937 		static ino_t	last_ino;
938 		static dev_t	last_dev;
939 
940 		/*
941 		 * Gathered writes: If another process is currently
942 		 * writing to the file, there's a high chance
943 		 * this is another nfsd (triggered by a bulk write
944 		 * from a client's biod). Rather than syncing the
945 		 * file with each write request, we sleep for 10 msec.
946 		 *
947 		 * I don't know if this roughly approximates
948 		 * C. Juszak's idea of gathered writes, but it's a
949 		 * nice and simple solution (IMHO), and it seems to
950 		 * work:-)
951 		 */
952 		if (EX_WGATHER(exp)) {
953 			if (atomic_read(&inode->i_writecount) > 1
954 			    || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
955 				dprintk("nfsd: write defer %d\n", current->pid);
956 				msleep(10);
957 				dprintk("nfsd: write resume %d\n", current->pid);
958 			}
959 
960 			if (inode->i_state & I_DIRTY) {
961 				dprintk("nfsd: write sync %d\n", current->pid);
962 				err=nfsd_sync(file);
963 			}
964 #if 0
965 			wake_up(&inode->i_wait);
966 #endif
967 		}
968 		last_ino = inode->i_ino;
969 		last_dev = inode->i_sb->s_dev;
970 	}
971 
972 	dprintk("nfsd: write complete err=%d\n", err);
973 	if (err >= 0)
974 		err = 0;
975 	else
976 		err = nfserrno(err);
977 out:
978 	return err;
979 }
980 
981 /*
982  * Read data from a file. count must contain the requested read count
983  * on entry. On return, *count contains the number of bytes actually read.
984  * N.B. After this call fhp needs an fh_put
985  */
986 int
987 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
988 		loff_t offset, struct kvec *vec, int vlen,
989 		unsigned long *count)
990 {
991 	int		err;
992 
993 	if (file) {
994 		err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
995 				MAY_READ|MAY_OWNER_OVERRIDE);
996 		if (err)
997 			goto out;
998 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
999 	} else {
1000 		err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1001 		if (err)
1002 			goto out;
1003 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1004 		nfsd_close(file);
1005 	}
1006 out:
1007 	return err;
1008 }
1009 
1010 /*
1011  * Write data to a file.
1012  * The stable flag requests synchronous writes.
1013  * N.B. After this call fhp needs an fh_put
1014  */
1015 int
1016 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1017 		loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1018 		int *stablep)
1019 {
1020 	int			err = 0;
1021 
1022 	if (file) {
1023 		err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1024 				MAY_WRITE|MAY_OWNER_OVERRIDE);
1025 		if (err)
1026 			goto out;
1027 		err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1028 				stablep);
1029 	} else {
1030 		err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1031 		if (err)
1032 			goto out;
1033 
1034 		if (cnt)
1035 			err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1036 					     cnt, stablep);
1037 		nfsd_close(file);
1038 	}
1039 out:
1040 	return err;
1041 }
1042 
1043 #ifdef CONFIG_NFSD_V3
1044 /*
1045  * Commit all pending writes to stable storage.
1046  * Strictly speaking, we could sync just the indicated file region here,
1047  * but there's currently no way we can ask the VFS to do so.
1048  *
1049  * Unfortunately we cannot lock the file to make sure we return full WCC
1050  * data to the client, as locking happens lower down in the filesystem.
1051  */
1052 int
1053 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1054                loff_t offset, unsigned long count)
1055 {
1056 	struct file	*file;
1057 	int		err;
1058 
1059 	if ((u64)count > ~(u64)offset)
1060 		return nfserr_inval;
1061 
1062 	if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1063 		return err;
1064 	if (EX_ISSYNC(fhp->fh_export)) {
1065 		if (file->f_op && file->f_op->fsync) {
1066 			err = nfserrno(nfsd_sync(file));
1067 		} else {
1068 			err = nfserr_notsupp;
1069 		}
1070 	}
1071 
1072 	nfsd_close(file);
1073 	return err;
1074 }
1075 #endif /* CONFIG_NFSD_V3 */
1076 
1077 /*
1078  * Create a file (regular, directory, device, fifo); UNIX sockets
1079  * not yet implemented.
1080  * If the response fh has been verified, the parent directory should
1081  * already be locked. Note that the parent directory is left locked.
1082  *
1083  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1084  */
1085 int
1086 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1087 		char *fname, int flen, struct iattr *iap,
1088 		int type, dev_t rdev, struct svc_fh *resfhp)
1089 {
1090 	struct dentry	*dentry, *dchild = NULL;
1091 	struct inode	*dirp;
1092 	int		err;
1093 
1094 	err = nfserr_perm;
1095 	if (!flen)
1096 		goto out;
1097 	err = nfserr_exist;
1098 	if (isdotent(fname, flen))
1099 		goto out;
1100 
1101 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1102 	if (err)
1103 		goto out;
1104 
1105 	dentry = fhp->fh_dentry;
1106 	dirp = dentry->d_inode;
1107 
1108 	err = nfserr_notdir;
1109 	if(!dirp->i_op || !dirp->i_op->lookup)
1110 		goto out;
1111 	/*
1112 	 * Check whether the response file handle has been verified yet.
1113 	 * If it has, the parent directory should already be locked.
1114 	 */
1115 	if (!resfhp->fh_dentry) {
1116 		/* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1117 		fh_lock(fhp);
1118 		dchild = lookup_one_len(fname, dentry, flen);
1119 		err = PTR_ERR(dchild);
1120 		if (IS_ERR(dchild))
1121 			goto out_nfserr;
1122 		err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1123 		if (err)
1124 			goto out;
1125 	} else {
1126 		/* called from nfsd_proc_create */
1127 		dchild = dget(resfhp->fh_dentry);
1128 		if (!fhp->fh_locked) {
1129 			/* not actually possible */
1130 			printk(KERN_ERR
1131 				"nfsd_create: parent %s/%s not locked!\n",
1132 				dentry->d_parent->d_name.name,
1133 				dentry->d_name.name);
1134 			err = nfserr_io;
1135 			goto out;
1136 		}
1137 	}
1138 	/*
1139 	 * Make sure the child dentry is still negative ...
1140 	 */
1141 	err = nfserr_exist;
1142 	if (dchild->d_inode) {
1143 		dprintk("nfsd_create: dentry %s/%s not negative!\n",
1144 			dentry->d_name.name, dchild->d_name.name);
1145 		goto out;
1146 	}
1147 
1148 	if (!(iap->ia_valid & ATTR_MODE))
1149 		iap->ia_mode = 0;
1150 	iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1151 
1152 	/*
1153 	 * Get the dir op function pointer.
1154 	 */
1155 	err = nfserr_perm;
1156 	switch (type) {
1157 	case S_IFREG:
1158 		err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1159 		break;
1160 	case S_IFDIR:
1161 		err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1162 		break;
1163 	case S_IFCHR:
1164 	case S_IFBLK:
1165 	case S_IFIFO:
1166 	case S_IFSOCK:
1167 		err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1168 		break;
1169 	default:
1170 	        printk("nfsd: bad file type %o in nfsd_create\n", type);
1171 		err = -EINVAL;
1172 	}
1173 	if (err < 0)
1174 		goto out_nfserr;
1175 
1176 	if (EX_ISSYNC(fhp->fh_export)) {
1177 		err = nfserrno(nfsd_sync_dir(dentry));
1178 		write_inode_now(dchild->d_inode, 1);
1179 	}
1180 
1181 
1182 	/* Set file attributes. Mode has already been set and
1183 	 * setting uid/gid works only for root. Irix appears to
1184 	 * send along the gid when it tries to implement setgid
1185 	 * directories via NFS.
1186 	 */
1187 	if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1188 		int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1189 		if (err2)
1190 			err = err2;
1191 	}
1192 	/*
1193 	 * Update the file handle to get the new inode info.
1194 	 */
1195 	if (!err)
1196 		err = fh_update(resfhp);
1197 out:
1198 	if (dchild && !IS_ERR(dchild))
1199 		dput(dchild);
1200 	return err;
1201 
1202 out_nfserr:
1203 	err = nfserrno(err);
1204 	goto out;
1205 }
1206 
1207 #ifdef CONFIG_NFSD_V3
1208 /*
1209  * NFSv3 version of nfsd_create
1210  */
1211 int
1212 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1213 		char *fname, int flen, struct iattr *iap,
1214 		struct svc_fh *resfhp, int createmode, u32 *verifier,
1215 	        int *truncp)
1216 {
1217 	struct dentry	*dentry, *dchild = NULL;
1218 	struct inode	*dirp;
1219 	int		err;
1220 	__u32		v_mtime=0, v_atime=0;
1221 	int		v_mode=0;
1222 
1223 	err = nfserr_perm;
1224 	if (!flen)
1225 		goto out;
1226 	err = nfserr_exist;
1227 	if (isdotent(fname, flen))
1228 		goto out;
1229 	if (!(iap->ia_valid & ATTR_MODE))
1230 		iap->ia_mode = 0;
1231 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1232 	if (err)
1233 		goto out;
1234 
1235 	dentry = fhp->fh_dentry;
1236 	dirp = dentry->d_inode;
1237 
1238 	/* Get all the sanity checks out of the way before
1239 	 * we lock the parent. */
1240 	err = nfserr_notdir;
1241 	if(!dirp->i_op || !dirp->i_op->lookup)
1242 		goto out;
1243 	fh_lock(fhp);
1244 
1245 	/*
1246 	 * Compose the response file handle.
1247 	 */
1248 	dchild = lookup_one_len(fname, dentry, flen);
1249 	err = PTR_ERR(dchild);
1250 	if (IS_ERR(dchild))
1251 		goto out_nfserr;
1252 
1253 	err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1254 	if (err)
1255 		goto out;
1256 
1257 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1258 		/* while the verifier would fit in mtime+atime,
1259 		 * solaris7 gets confused (bugid 4218508) if these have
1260 		 * the high bit set, so we use the mode as well
1261 		 */
1262 		v_mtime = verifier[0]&0x7fffffff;
1263 		v_atime = verifier[1]&0x7fffffff;
1264 		v_mode  = S_IFREG
1265 			| ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1266 			| ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1267 			;
1268 	}
1269 
1270 	if (dchild->d_inode) {
1271 		err = 0;
1272 
1273 		switch (createmode) {
1274 		case NFS3_CREATE_UNCHECKED:
1275 			if (! S_ISREG(dchild->d_inode->i_mode))
1276 				err = nfserr_exist;
1277 			else if (truncp) {
1278 				/* in nfsv4, we need to treat this case a little
1279 				 * differently.  we don't want to truncate the
1280 				 * file now; this would be wrong if the OPEN
1281 				 * fails for some other reason.  furthermore,
1282 				 * if the size is nonzero, we should ignore it
1283 				 * according to spec!
1284 				 */
1285 				*truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1286 			}
1287 			else {
1288 				iap->ia_valid &= ATTR_SIZE;
1289 				goto set_attr;
1290 			}
1291 			break;
1292 		case NFS3_CREATE_EXCLUSIVE:
1293 			if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1294 			    && dchild->d_inode->i_atime.tv_sec == v_atime
1295 			    && dchild->d_inode->i_mode  == v_mode
1296 			    && dchild->d_inode->i_size  == 0 )
1297 				break;
1298 			 /* fallthru */
1299 		case NFS3_CREATE_GUARDED:
1300 			err = nfserr_exist;
1301 		}
1302 		goto out;
1303 	}
1304 
1305 	err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1306 	if (err < 0)
1307 		goto out_nfserr;
1308 
1309 	if (EX_ISSYNC(fhp->fh_export)) {
1310 		err = nfserrno(nfsd_sync_dir(dentry));
1311 		/* setattr will sync the child (or not) */
1312 	}
1313 
1314 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1315 		/* Cram the verifier into atime/mtime/mode */
1316 		iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1317 			| ATTR_MTIME_SET|ATTR_ATIME_SET
1318 			| ATTR_MODE;
1319 		/* XXX someone who knows this better please fix it for nsec */
1320 		iap->ia_mtime.tv_sec = v_mtime;
1321 		iap->ia_atime.tv_sec = v_atime;
1322 		iap->ia_mtime.tv_nsec = 0;
1323 		iap->ia_atime.tv_nsec = 0;
1324 		iap->ia_mode  = v_mode;
1325 	}
1326 
1327 	/* Set file attributes.
1328 	 * Mode has already been set but we might need to reset it
1329 	 * for CREATE_EXCLUSIVE
1330 	 * Irix appears to send along the gid when it tries to
1331 	 * implement setgid directories via NFS. Clear out all that cruft.
1332 	 */
1333  set_attr:
1334 	if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) {
1335  		int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1336 		if (err2)
1337 			err = err2;
1338 	}
1339 
1340 	/*
1341 	 * Update the filehandle to get the new inode info.
1342 	 */
1343 	if (!err)
1344 		err = fh_update(resfhp);
1345 
1346  out:
1347 	fh_unlock(fhp);
1348 	if (dchild && !IS_ERR(dchild))
1349 		dput(dchild);
1350  	return err;
1351 
1352  out_nfserr:
1353 	err = nfserrno(err);
1354 	goto out;
1355 }
1356 #endif /* CONFIG_NFSD_V3 */
1357 
1358 /*
1359  * Read a symlink. On entry, *lenp must contain the maximum path length that
1360  * fits into the buffer. On return, it contains the true length.
1361  * N.B. After this call fhp needs an fh_put
1362  */
1363 int
1364 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1365 {
1366 	struct dentry	*dentry;
1367 	struct inode	*inode;
1368 	mm_segment_t	oldfs;
1369 	int		err;
1370 
1371 	err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1372 	if (err)
1373 		goto out;
1374 
1375 	dentry = fhp->fh_dentry;
1376 	inode = dentry->d_inode;
1377 
1378 	err = nfserr_inval;
1379 	if (!inode->i_op || !inode->i_op->readlink)
1380 		goto out;
1381 
1382 	touch_atime(fhp->fh_export->ex_mnt, dentry);
1383 	/* N.B. Why does this call need a get_fs()??
1384 	 * Remove the set_fs and watch the fireworks:-) --okir
1385 	 */
1386 
1387 	oldfs = get_fs(); set_fs(KERNEL_DS);
1388 	err = inode->i_op->readlink(dentry, buf, *lenp);
1389 	set_fs(oldfs);
1390 
1391 	if (err < 0)
1392 		goto out_nfserr;
1393 	*lenp = err;
1394 	err = 0;
1395 out:
1396 	return err;
1397 
1398 out_nfserr:
1399 	err = nfserrno(err);
1400 	goto out;
1401 }
1402 
1403 /*
1404  * Create a symlink and look up its inode
1405  * N.B. After this call _both_ fhp and resfhp need an fh_put
1406  */
1407 int
1408 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1409 				char *fname, int flen,
1410 				char *path,  int plen,
1411 				struct svc_fh *resfhp,
1412 				struct iattr *iap)
1413 {
1414 	struct dentry	*dentry, *dnew;
1415 	int		err, cerr;
1416 	umode_t		mode;
1417 
1418 	err = nfserr_noent;
1419 	if (!flen || !plen)
1420 		goto out;
1421 	err = nfserr_exist;
1422 	if (isdotent(fname, flen))
1423 		goto out;
1424 
1425 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1426 	if (err)
1427 		goto out;
1428 	fh_lock(fhp);
1429 	dentry = fhp->fh_dentry;
1430 	dnew = lookup_one_len(fname, dentry, flen);
1431 	err = PTR_ERR(dnew);
1432 	if (IS_ERR(dnew))
1433 		goto out_nfserr;
1434 
1435 	mode = S_IALLUGO;
1436 	/* Only the MODE ATTRibute is even vaguely meaningful */
1437 	if (iap && (iap->ia_valid & ATTR_MODE))
1438 		mode = iap->ia_mode & S_IALLUGO;
1439 
1440 	if (unlikely(path[plen] != 0)) {
1441 		char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1442 		if (path_alloced == NULL)
1443 			err = -ENOMEM;
1444 		else {
1445 			strncpy(path_alloced, path, plen);
1446 			path_alloced[plen] = 0;
1447 			err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1448 			kfree(path_alloced);
1449 		}
1450 	} else
1451 		err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1452 
1453 	if (!err)
1454 		if (EX_ISSYNC(fhp->fh_export))
1455 			err = nfsd_sync_dir(dentry);
1456 	if (err)
1457 		err = nfserrno(err);
1458 	fh_unlock(fhp);
1459 
1460 	cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1461 	dput(dnew);
1462 	if (err==0) err = cerr;
1463 out:
1464 	return err;
1465 
1466 out_nfserr:
1467 	err = nfserrno(err);
1468 	goto out;
1469 }
1470 
1471 /*
1472  * Create a hardlink
1473  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1474  */
1475 int
1476 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1477 				char *name, int len, struct svc_fh *tfhp)
1478 {
1479 	struct dentry	*ddir, *dnew, *dold;
1480 	struct inode	*dirp, *dest;
1481 	int		err;
1482 
1483 	err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1484 	if (err)
1485 		goto out;
1486 	err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1487 	if (err)
1488 		goto out;
1489 
1490 	err = nfserr_perm;
1491 	if (!len)
1492 		goto out;
1493 	err = nfserr_exist;
1494 	if (isdotent(name, len))
1495 		goto out;
1496 
1497 	fh_lock(ffhp);
1498 	ddir = ffhp->fh_dentry;
1499 	dirp = ddir->d_inode;
1500 
1501 	dnew = lookup_one_len(name, ddir, len);
1502 	err = PTR_ERR(dnew);
1503 	if (IS_ERR(dnew))
1504 		goto out_nfserr;
1505 
1506 	dold = tfhp->fh_dentry;
1507 	dest = dold->d_inode;
1508 
1509 	err = vfs_link(dold, dirp, dnew);
1510 	if (!err) {
1511 		if (EX_ISSYNC(ffhp->fh_export)) {
1512 			err = nfserrno(nfsd_sync_dir(ddir));
1513 			write_inode_now(dest, 1);
1514 		}
1515 	} else {
1516 		if (err == -EXDEV && rqstp->rq_vers == 2)
1517 			err = nfserr_acces;
1518 		else
1519 			err = nfserrno(err);
1520 	}
1521 
1522 	dput(dnew);
1523 out_unlock:
1524 	fh_unlock(ffhp);
1525 out:
1526 	return err;
1527 
1528 out_nfserr:
1529 	err = nfserrno(err);
1530 	goto out_unlock;
1531 }
1532 
1533 /*
1534  * Rename a file
1535  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1536  */
1537 int
1538 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1539 			    struct svc_fh *tfhp, char *tname, int tlen)
1540 {
1541 	struct dentry	*fdentry, *tdentry, *odentry, *ndentry, *trap;
1542 	struct inode	*fdir, *tdir;
1543 	int		err;
1544 
1545 	err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1546 	if (err)
1547 		goto out;
1548 	err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1549 	if (err)
1550 		goto out;
1551 
1552 	fdentry = ffhp->fh_dentry;
1553 	fdir = fdentry->d_inode;
1554 
1555 	tdentry = tfhp->fh_dentry;
1556 	tdir = tdentry->d_inode;
1557 
1558 	err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1559 	if (ffhp->fh_export != tfhp->fh_export)
1560 		goto out;
1561 
1562 	err = nfserr_perm;
1563 	if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1564 		goto out;
1565 
1566 	/* cannot use fh_lock as we need deadlock protective ordering
1567 	 * so do it by hand */
1568 	trap = lock_rename(tdentry, fdentry);
1569 	ffhp->fh_locked = tfhp->fh_locked = 1;
1570 	fill_pre_wcc(ffhp);
1571 	fill_pre_wcc(tfhp);
1572 
1573 	odentry = lookup_one_len(fname, fdentry, flen);
1574 	err = PTR_ERR(odentry);
1575 	if (IS_ERR(odentry))
1576 		goto out_nfserr;
1577 
1578 	err = -ENOENT;
1579 	if (!odentry->d_inode)
1580 		goto out_dput_old;
1581 	err = -EINVAL;
1582 	if (odentry == trap)
1583 		goto out_dput_old;
1584 
1585 	ndentry = lookup_one_len(tname, tdentry, tlen);
1586 	err = PTR_ERR(ndentry);
1587 	if (IS_ERR(ndentry))
1588 		goto out_dput_old;
1589 	err = -ENOTEMPTY;
1590 	if (ndentry == trap)
1591 		goto out_dput_new;
1592 
1593 #ifdef MSNFS
1594 	if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1595 		((atomic_read(&odentry->d_count) > 1)
1596 		 || (atomic_read(&ndentry->d_count) > 1))) {
1597 			err = -EPERM;
1598 	} else
1599 #endif
1600 	err = vfs_rename(fdir, odentry, tdir, ndentry);
1601 	if (!err && EX_ISSYNC(tfhp->fh_export)) {
1602 		err = nfsd_sync_dir(tdentry);
1603 		if (!err)
1604 			err = nfsd_sync_dir(fdentry);
1605 	}
1606 
1607  out_dput_new:
1608 	dput(ndentry);
1609  out_dput_old:
1610 	dput(odentry);
1611  out_nfserr:
1612 	if (err)
1613 		err = nfserrno(err);
1614 
1615 	/* we cannot reply on fh_unlock on the two filehandles,
1616 	 * as that would do the wrong thing if the two directories
1617 	 * were the same, so again we do it by hand
1618 	 */
1619 	fill_post_wcc(ffhp);
1620 	fill_post_wcc(tfhp);
1621 	unlock_rename(tdentry, fdentry);
1622 	ffhp->fh_locked = tfhp->fh_locked = 0;
1623 
1624 out:
1625 	return err;
1626 }
1627 
1628 /*
1629  * Unlink a file or directory
1630  * N.B. After this call fhp needs an fh_put
1631  */
1632 int
1633 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1634 				char *fname, int flen)
1635 {
1636 	struct dentry	*dentry, *rdentry;
1637 	struct inode	*dirp;
1638 	int		err;
1639 
1640 	err = nfserr_acces;
1641 	if (!flen || isdotent(fname, flen))
1642 		goto out;
1643 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1644 	if (err)
1645 		goto out;
1646 
1647 	fh_lock(fhp);
1648 	dentry = fhp->fh_dentry;
1649 	dirp = dentry->d_inode;
1650 
1651 	rdentry = lookup_one_len(fname, dentry, flen);
1652 	err = PTR_ERR(rdentry);
1653 	if (IS_ERR(rdentry))
1654 		goto out_nfserr;
1655 
1656 	if (!rdentry->d_inode) {
1657 		dput(rdentry);
1658 		err = nfserr_noent;
1659 		goto out;
1660 	}
1661 
1662 	if (!type)
1663 		type = rdentry->d_inode->i_mode & S_IFMT;
1664 
1665 	if (type != S_IFDIR) { /* It's UNLINK */
1666 #ifdef MSNFS
1667 		if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1668 			(atomic_read(&rdentry->d_count) > 1)) {
1669 			err = -EPERM;
1670 		} else
1671 #endif
1672 		err = vfs_unlink(dirp, rdentry);
1673 	} else { /* It's RMDIR */
1674 		err = vfs_rmdir(dirp, rdentry);
1675 	}
1676 
1677 	dput(rdentry);
1678 
1679 	if (err == 0 &&
1680 	    EX_ISSYNC(fhp->fh_export))
1681 			err = nfsd_sync_dir(dentry);
1682 
1683 out_nfserr:
1684 	err = nfserrno(err);
1685 out:
1686 	return err;
1687 }
1688 
1689 /*
1690  * Read entries from a directory.
1691  * The  NFSv3/4 verifier we ignore for now.
1692  */
1693 int
1694 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1695 	     struct readdir_cd *cdp, encode_dent_fn func)
1696 {
1697 	int		err;
1698 	struct file	*file;
1699 	loff_t		offset = *offsetp;
1700 
1701 	err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1702 	if (err)
1703 		goto out;
1704 
1705 	offset = vfs_llseek(file, offset, 0);
1706 	if (offset < 0) {
1707 		err = nfserrno((int)offset);
1708 		goto out_close;
1709 	}
1710 
1711 	/*
1712 	 * Read the directory entries. This silly loop is necessary because
1713 	 * readdir() is not guaranteed to fill up the entire buffer, but
1714 	 * may choose to do less.
1715 	 */
1716 
1717 	do {
1718 		cdp->err = nfserr_eof; /* will be cleared on successful read */
1719 		err = vfs_readdir(file, (filldir_t) func, cdp);
1720 	} while (err >=0 && cdp->err == nfs_ok);
1721 	if (err)
1722 		err = nfserrno(err);
1723 	else
1724 		err = cdp->err;
1725 	*offsetp = vfs_llseek(file, 0, 1);
1726 
1727 	if (err == nfserr_eof || err == nfserr_toosmall)
1728 		err = nfs_ok; /* can still be found in ->err */
1729 out_close:
1730 	nfsd_close(file);
1731 out:
1732 	return err;
1733 }
1734 
1735 /*
1736  * Get file system stats
1737  * N.B. After this call fhp needs an fh_put
1738  */
1739 int
1740 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1741 {
1742 	int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1743 	if (!err && vfs_statfs(fhp->fh_dentry,stat))
1744 		err = nfserr_io;
1745 	return err;
1746 }
1747 
1748 /*
1749  * Check for a user's access permissions to this inode.
1750  */
1751 int
1752 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1753 {
1754 	struct inode	*inode = dentry->d_inode;
1755 	int		err;
1756 
1757 	if (acc == MAY_NOP)
1758 		return 0;
1759 #if 0
1760 	dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1761 		acc,
1762 		(acc & MAY_READ)?	" read"  : "",
1763 		(acc & MAY_WRITE)?	" write" : "",
1764 		(acc & MAY_EXEC)?	" exec"  : "",
1765 		(acc & MAY_SATTR)?	" sattr" : "",
1766 		(acc & MAY_TRUNC)?	" trunc" : "",
1767 		(acc & MAY_LOCK)?	" lock"  : "",
1768 		(acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1769 		inode->i_mode,
1770 		IS_IMMUTABLE(inode)?	" immut" : "",
1771 		IS_APPEND(inode)?	" append" : "",
1772 		IS_RDONLY(inode)?	" ro" : "");
1773 	dprintk("      owner %d/%d user %d/%d\n",
1774 		inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1775 #endif
1776 
1777 	/* Normally we reject any write/sattr etc access on a read-only file
1778 	 * system.  But if it is IRIX doing check on write-access for a
1779 	 * device special file, we ignore rofs.
1780 	 */
1781 	if (!(acc & MAY_LOCAL_ACCESS))
1782 		if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1783 			if (EX_RDONLY(exp) || IS_RDONLY(inode))
1784 				return nfserr_rofs;
1785 			if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1786 				return nfserr_perm;
1787 		}
1788 	if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1789 		return nfserr_perm;
1790 
1791 	if (acc & MAY_LOCK) {
1792 		/* If we cannot rely on authentication in NLM requests,
1793 		 * just allow locks, otherwise require read permission, or
1794 		 * ownership
1795 		 */
1796 		if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1797 			return 0;
1798 		else
1799 			acc = MAY_READ | MAY_OWNER_OVERRIDE;
1800 	}
1801 	/*
1802 	 * The file owner always gets access permission for accesses that
1803 	 * would normally be checked at open time. This is to make
1804 	 * file access work even when the client has done a fchmod(fd, 0).
1805 	 *
1806 	 * However, `cp foo bar' should fail nevertheless when bar is
1807 	 * readonly. A sensible way to do this might be to reject all
1808 	 * attempts to truncate a read-only file, because a creat() call
1809 	 * always implies file truncation.
1810 	 * ... but this isn't really fair.  A process may reasonably call
1811 	 * ftruncate on an open file descriptor on a file with perm 000.
1812 	 * We must trust the client to do permission checking - using "ACCESS"
1813 	 * with NFSv3.
1814 	 */
1815 	if ((acc & MAY_OWNER_OVERRIDE) &&
1816 	    inode->i_uid == current->fsuid)
1817 		return 0;
1818 
1819 	err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1820 
1821 	/* Allow read access to binaries even when mode 111 */
1822 	if (err == -EACCES && S_ISREG(inode->i_mode) &&
1823 	    acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1824 		err = permission(inode, MAY_EXEC, NULL);
1825 
1826 	return err? nfserrno(err) : 0;
1827 }
1828 
1829 void
1830 nfsd_racache_shutdown(void)
1831 {
1832 	if (!raparm_cache)
1833 		return;
1834 	dprintk("nfsd: freeing readahead buffers.\n");
1835 	kfree(raparml);
1836 	raparm_cache = raparml = NULL;
1837 }
1838 /*
1839  * Initialize readahead param cache
1840  */
1841 int
1842 nfsd_racache_init(int cache_size)
1843 {
1844 	int	i;
1845 
1846 	if (raparm_cache)
1847 		return 0;
1848 	raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1849 
1850 	if (raparml != NULL) {
1851 		dprintk("nfsd: allocating %d readahead buffers.\n",
1852 			cache_size);
1853 		memset(raparml, 0, sizeof(struct raparms) * cache_size);
1854 		for (i = 0; i < cache_size - 1; i++) {
1855 			raparml[i].p_next = raparml + i + 1;
1856 		}
1857 		raparm_cache = raparml;
1858 	} else {
1859 		printk(KERN_WARNING
1860 		       "nfsd: Could not allocate memory read-ahead cache.\n");
1861 		return -ENOMEM;
1862 	}
1863 	nfsdstats.ra_size = cache_size;
1864 	return 0;
1865 }
1866 
1867 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1868 struct posix_acl *
1869 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1870 {
1871 	struct inode *inode = fhp->fh_dentry->d_inode;
1872 	char *name;
1873 	void *value = NULL;
1874 	ssize_t size;
1875 	struct posix_acl *acl;
1876 
1877 	if (!IS_POSIXACL(inode))
1878 		return ERR_PTR(-EOPNOTSUPP);
1879 
1880 	switch (type) {
1881 	case ACL_TYPE_ACCESS:
1882 		name = POSIX_ACL_XATTR_ACCESS;
1883 		break;
1884 	case ACL_TYPE_DEFAULT:
1885 		name = POSIX_ACL_XATTR_DEFAULT;
1886 		break;
1887 	default:
1888 		return ERR_PTR(-EOPNOTSUPP);
1889 	}
1890 
1891 	size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1892 	if (size < 0)
1893 		return ERR_PTR(size);
1894 
1895 	acl = posix_acl_from_xattr(value, size);
1896 	kfree(value);
1897 	return acl;
1898 }
1899 
1900 int
1901 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1902 {
1903 	struct inode *inode = fhp->fh_dentry->d_inode;
1904 	char *name;
1905 	void *value = NULL;
1906 	size_t size;
1907 	int error;
1908 
1909 	if (!IS_POSIXACL(inode) || !inode->i_op ||
1910 	    !inode->i_op->setxattr || !inode->i_op->removexattr)
1911 		return -EOPNOTSUPP;
1912 	switch(type) {
1913 		case ACL_TYPE_ACCESS:
1914 			name = POSIX_ACL_XATTR_ACCESS;
1915 			break;
1916 		case ACL_TYPE_DEFAULT:
1917 			name = POSIX_ACL_XATTR_DEFAULT;
1918 			break;
1919 		default:
1920 			return -EOPNOTSUPP;
1921 	}
1922 
1923 	if (acl && acl->a_count) {
1924 		size = posix_acl_xattr_size(acl->a_count);
1925 		value = kmalloc(size, GFP_KERNEL);
1926 		if (!value)
1927 			return -ENOMEM;
1928 		error = posix_acl_to_xattr(acl, value, size);
1929 		if (error < 0)
1930 			goto getout;
1931 		size = error;
1932 	} else
1933 		size = 0;
1934 
1935 	if (size)
1936 		error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1937 	else {
1938 		if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1939 			error = 0;
1940 		else {
1941 			error = vfs_removexattr(fhp->fh_dentry, name);
1942 			if (error == -ENODATA)
1943 				error = 0;
1944 		}
1945 	}
1946 
1947 getout:
1948 	kfree(value);
1949 	return error;
1950 }
1951 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
1952