xref: /linux/fs/nfsd/vfs.c (revision 5e8d780d745c1619aba81fe7166c5a4b5cad2b84)
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/config.h>
20 #include <linux/string.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/fs.h>
24 #include <linux/file.h>
25 #include <linux/mount.h>
26 #include <linux/major.h>
27 #include <linux/ext2_fs.h>
28 #include <linux/proc_fs.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/net.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/pagemap.h>
35 #include <linux/in.h>
36 #include <linux/module.h>
37 #include <linux/namei.h>
38 #include <linux/vfs.h>
39 #include <linux/delay.h>
40 #include <linux/sunrpc/svc.h>
41 #include <linux/nfsd/nfsd.h>
42 #ifdef CONFIG_NFSD_V3
43 #include <linux/nfs3.h>
44 #include <linux/nfsd/xdr3.h>
45 #endif /* CONFIG_NFSD_V3 */
46 #include <linux/nfsd/nfsfh.h>
47 #include <linux/quotaops.h>
48 #include <linux/fsnotify.h>
49 #include <linux/posix_acl.h>
50 #include <linux/posix_acl_xattr.h>
51 #include <linux/xattr.h>
52 #ifdef CONFIG_NFSD_V4
53 #include <linux/nfs4.h>
54 #include <linux/nfs4_acl.h>
55 #include <linux/nfsd_idmap.h>
56 #include <linux/security.h>
57 #endif /* CONFIG_NFSD_V4 */
58 
59 #include <asm/uaccess.h>
60 
61 #define NFSDDBG_FACILITY		NFSDDBG_FILEOP
62 #define NFSD_PARANOIA
63 
64 
65 /* We must ignore files (but only files) which might have mandatory
66  * locks on them because there is no way to know if the accesser has
67  * the lock.
68  */
69 #define IS_ISMNDLK(i)	(S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
70 
71 /*
72  * This is a cache of readahead params that help us choose the proper
73  * readahead strategy. Initially, we set all readahead parameters to 0
74  * and let the VFS handle things.
75  * If you increase the number of cached files very much, you'll need to
76  * add a hash table here.
77  */
78 struct raparms {
79 	struct raparms		*p_next;
80 	unsigned int		p_count;
81 	ino_t			p_ino;
82 	dev_t			p_dev;
83 	int			p_set;
84 	struct file_ra_state	p_ra;
85 };
86 
87 static struct raparms *		raparml;
88 static struct raparms *		raparm_cache;
89 
90 /*
91  * Called from nfsd_lookup and encode_dirent. Check if we have crossed
92  * a mount point.
93  * Returns -EAGAIN leaving *dpp and *expp unchanged,
94  *  or nfs_ok having possibly changed *dpp and *expp
95  */
96 int
97 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
98 		        struct svc_export **expp)
99 {
100 	struct svc_export *exp = *expp, *exp2 = NULL;
101 	struct dentry *dentry = *dpp;
102 	struct vfsmount *mnt = mntget(exp->ex_mnt);
103 	struct dentry *mounts = dget(dentry);
104 	int err = nfs_ok;
105 
106 	while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
107 
108 	exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
109 	if (IS_ERR(exp2)) {
110 		err = PTR_ERR(exp2);
111 		dput(mounts);
112 		mntput(mnt);
113 		goto out;
114 	}
115 	if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
116 		/* successfully crossed mount point */
117 		exp_put(exp);
118 		*expp = exp2;
119 		dput(dentry);
120 		*dpp = mounts;
121 	} else {
122 		if (exp2) exp_put(exp2);
123 		dput(mounts);
124 	}
125 	mntput(mnt);
126 out:
127 	return err;
128 }
129 
130 /*
131  * Look up one component of a pathname.
132  * N.B. After this call _both_ fhp and resfh need an fh_put
133  *
134  * If the lookup would cross a mountpoint, and the mounted filesystem
135  * is exported to the client with NFSEXP_NOHIDE, then the lookup is
136  * accepted as it stands and the mounted directory is
137  * returned. Otherwise the covered directory is returned.
138  * NOTE: this mountpoint crossing is not supported properly by all
139  *   clients and is explicitly disallowed for NFSv3
140  *      NeilBrown <neilb@cse.unsw.edu.au>
141  */
142 int
143 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
144 					int len, struct svc_fh *resfh)
145 {
146 	struct svc_export	*exp;
147 	struct dentry		*dparent;
148 	struct dentry		*dentry;
149 	int			err;
150 
151 	dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
152 
153 	/* Obtain dentry and export. */
154 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
155 	if (err)
156 		return err;
157 
158 	dparent = fhp->fh_dentry;
159 	exp  = fhp->fh_export;
160 	exp_get(exp);
161 
162 	err = nfserr_acces;
163 
164 	/* Lookup the name, but don't follow links */
165 	if (isdotent(name, len)) {
166 		if (len==1)
167 			dentry = dget(dparent);
168 		else if (dparent != exp->ex_dentry) {
169 			dentry = dget_parent(dparent);
170 		} else  if (!EX_NOHIDE(exp))
171 			dentry = dget(dparent); /* .. == . just like at / */
172 		else {
173 			/* checking mountpoint crossing is very different when stepping up */
174 			struct svc_export *exp2 = NULL;
175 			struct dentry *dp;
176 			struct vfsmount *mnt = mntget(exp->ex_mnt);
177 			dentry = dget(dparent);
178 			while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
179 				;
180 			dp = dget_parent(dentry);
181 			dput(dentry);
182 			dentry = dp;
183 
184 			exp2 = exp_parent(exp->ex_client, mnt, dentry,
185 					  &rqstp->rq_chandle);
186 			if (IS_ERR(exp2)) {
187 				err = PTR_ERR(exp2);
188 				dput(dentry);
189 				mntput(mnt);
190 				goto out_nfserr;
191 			}
192 			if (!exp2) {
193 				dput(dentry);
194 				dentry = dget(dparent);
195 			} else {
196 				exp_put(exp);
197 				exp = exp2;
198 			}
199 			mntput(mnt);
200 		}
201 	} else {
202 		fh_lock(fhp);
203 		dentry = lookup_one_len(name, dparent, len);
204 		err = PTR_ERR(dentry);
205 		if (IS_ERR(dentry))
206 			goto out_nfserr;
207 		/*
208 		 * check if we have crossed a mount point ...
209 		 */
210 		if (d_mountpoint(dentry)) {
211 			if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
212 				dput(dentry);
213 				goto out_nfserr;
214 			}
215 		}
216 	}
217 	/*
218 	 * Note: we compose the file handle now, but as the
219 	 * dentry may be negative, it may need to be updated.
220 	 */
221 	err = fh_compose(resfh, exp, dentry, fhp);
222 	if (!err && !dentry->d_inode)
223 		err = nfserr_noent;
224 	dput(dentry);
225 out:
226 	exp_put(exp);
227 	return err;
228 
229 out_nfserr:
230 	err = nfserrno(err);
231 	goto out;
232 }
233 
234 /*
235  * Set various file attributes.
236  * N.B. After this call fhp needs an fh_put
237  */
238 int
239 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
240 	     int check_guard, time_t guardtime)
241 {
242 	struct dentry	*dentry;
243 	struct inode	*inode;
244 	int		accmode = MAY_SATTR;
245 	int		ftype = 0;
246 	int		imode;
247 	int		err;
248 	int		size_change = 0;
249 
250 	if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
251 		accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
252 	if (iap->ia_valid & ATTR_SIZE)
253 		ftype = S_IFREG;
254 
255 	/* Get inode */
256 	err = fh_verify(rqstp, fhp, ftype, accmode);
257 	if (err)
258 		goto out;
259 
260 	dentry = fhp->fh_dentry;
261 	inode = dentry->d_inode;
262 
263 	/* Ignore any mode updates on symlinks */
264 	if (S_ISLNK(inode->i_mode))
265 		iap->ia_valid &= ~ATTR_MODE;
266 
267 	if (!iap->ia_valid)
268 		goto out;
269 
270 	/* NFSv2 does not differentiate between "set-[ac]time-to-now"
271 	 * which only requires access, and "set-[ac]time-to-X" which
272 	 * requires ownership.
273 	 * So if it looks like it might be "set both to the same time which
274 	 * is close to now", and if inode_change_ok fails, then we
275 	 * convert to "set to now" instead of "set to explicit time"
276 	 *
277 	 * We only call inode_change_ok as the last test as technically
278 	 * it is not an interface that we should be using.  It is only
279 	 * valid if the filesystem does not define it's own i_op->setattr.
280 	 */
281 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
282 #define	MAX_TOUCH_TIME_ERROR (30*60)
283 	if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
284 	    && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
285 	    ) {
286 	    /* Looks probable.  Now just make sure time is in the right ballpark.
287 	     * Solaris, at least, doesn't seem to care what the time request is.
288 	     * We require it be within 30 minutes of now.
289 	     */
290 	    time_t delta = iap->ia_atime.tv_sec - get_seconds();
291 	    if (delta<0) delta = -delta;
292 	    if (delta < MAX_TOUCH_TIME_ERROR &&
293 		inode_change_ok(inode, iap) != 0) {
294 		/* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
295 		 * this will cause notify_change to set these times to "now"
296 		 */
297 		iap->ia_valid &= ~BOTH_TIME_SET;
298 	    }
299 	}
300 
301 	/* The size case is special. It changes the file as well as the attributes.  */
302 	if (iap->ia_valid & ATTR_SIZE) {
303 		if (iap->ia_size < inode->i_size) {
304 			err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
305 			if (err)
306 				goto out;
307 		}
308 
309 		/*
310 		 * If we are changing the size of the file, then
311 		 * we need to break all leases.
312 		 */
313 		err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
314 		if (err == -EWOULDBLOCK)
315 			err = -ETIMEDOUT;
316 		if (err) /* ENOMEM or EWOULDBLOCK */
317 			goto out_nfserr;
318 
319 		err = get_write_access(inode);
320 		if (err)
321 			goto out_nfserr;
322 
323 		size_change = 1;
324 		err = locks_verify_truncate(inode, NULL, iap->ia_size);
325 		if (err) {
326 			put_write_access(inode);
327 			goto out_nfserr;
328 		}
329 		DQUOT_INIT(inode);
330 	}
331 
332 	imode = inode->i_mode;
333 	if (iap->ia_valid & ATTR_MODE) {
334 		iap->ia_mode &= S_IALLUGO;
335 		imode = iap->ia_mode |= (imode & ~S_IALLUGO);
336 	}
337 
338 	/* Revoke setuid/setgid bit on chown/chgrp */
339 	if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
340 		iap->ia_valid |= ATTR_KILL_SUID;
341 	if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
342 		iap->ia_valid |= ATTR_KILL_SGID;
343 
344 	/* Change the attributes. */
345 
346 	iap->ia_valid |= ATTR_CTIME;
347 
348 	err = nfserr_notsync;
349 	if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
350 		fh_lock(fhp);
351 		err = notify_change(dentry, iap);
352 		err = nfserrno(err);
353 		fh_unlock(fhp);
354 	}
355 	if (size_change)
356 		put_write_access(inode);
357 	if (!err)
358 		if (EX_ISSYNC(fhp->fh_export))
359 			write_inode_now(inode, 1);
360 out:
361 	return err;
362 
363 out_nfserr:
364 	err = nfserrno(err);
365 	goto out;
366 }
367 
368 #if defined(CONFIG_NFSD_V2_ACL) || \
369     defined(CONFIG_NFSD_V3_ACL) || \
370     defined(CONFIG_NFSD_V4)
371 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
372 {
373 	ssize_t buflen;
374 
375 	buflen = vfs_getxattr(dentry, key, NULL, 0);
376 	if (buflen <= 0)
377 		return buflen;
378 
379 	*buf = kmalloc(buflen, GFP_KERNEL);
380 	if (!*buf)
381 		return -ENOMEM;
382 
383 	return vfs_getxattr(dentry, key, *buf, buflen);
384 }
385 #endif
386 
387 #if defined(CONFIG_NFSD_V4)
388 static int
389 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
390 {
391 	int len;
392 	size_t buflen;
393 	char *buf = NULL;
394 	int error = 0;
395 
396 	buflen = posix_acl_xattr_size(pacl->a_count);
397 	buf = kmalloc(buflen, GFP_KERNEL);
398 	error = -ENOMEM;
399 	if (buf == NULL)
400 		goto out;
401 
402 	len = posix_acl_to_xattr(pacl, buf, buflen);
403 	if (len < 0) {
404 		error = len;
405 		goto out;
406 	}
407 
408 	error = vfs_setxattr(dentry, key, buf, len, 0);
409 out:
410 	kfree(buf);
411 	return error;
412 }
413 
414 int
415 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
416     struct nfs4_acl *acl)
417 {
418 	int error;
419 	struct dentry *dentry;
420 	struct inode *inode;
421 	struct posix_acl *pacl = NULL, *dpacl = NULL;
422 	unsigned int flags = 0;
423 
424 	/* Get inode */
425 	error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
426 	if (error)
427 		goto out;
428 
429 	dentry = fhp->fh_dentry;
430 	inode = dentry->d_inode;
431 	if (S_ISDIR(inode->i_mode))
432 		flags = NFS4_ACL_DIR;
433 
434 	error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
435 	if (error == -EINVAL) {
436 		error = nfserr_attrnotsupp;
437 		goto out;
438 	} else if (error < 0)
439 		goto out_nfserr;
440 
441 	if (pacl) {
442 		error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
443 		if (error < 0)
444 			goto out_nfserr;
445 	}
446 
447 	if (dpacl) {
448 		error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
449 		if (error < 0)
450 			goto out_nfserr;
451 	}
452 
453 	error = nfs_ok;
454 
455 out:
456 	posix_acl_release(pacl);
457 	posix_acl_release(dpacl);
458 	return (error);
459 out_nfserr:
460 	error = nfserrno(error);
461 	goto out;
462 }
463 
464 static struct posix_acl *
465 _get_posix_acl(struct dentry *dentry, char *key)
466 {
467 	void *buf = NULL;
468 	struct posix_acl *pacl = NULL;
469 	int buflen;
470 
471 	buflen = nfsd_getxattr(dentry, key, &buf);
472 	if (!buflen)
473 		buflen = -ENODATA;
474 	if (buflen <= 0)
475 		return ERR_PTR(buflen);
476 
477 	pacl = posix_acl_from_xattr(buf, buflen);
478 	kfree(buf);
479 	return pacl;
480 }
481 
482 int
483 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
484 {
485 	struct inode *inode = dentry->d_inode;
486 	int error = 0;
487 	struct posix_acl *pacl = NULL, *dpacl = NULL;
488 	unsigned int flags = 0;
489 
490 	pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
491 	if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
492 		pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
493 	if (IS_ERR(pacl)) {
494 		error = PTR_ERR(pacl);
495 		pacl = NULL;
496 		goto out;
497 	}
498 
499 	if (S_ISDIR(inode->i_mode)) {
500 		dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
501 		if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
502 			dpacl = NULL;
503 		else if (IS_ERR(dpacl)) {
504 			error = PTR_ERR(dpacl);
505 			dpacl = NULL;
506 			goto out;
507 		}
508 		flags = NFS4_ACL_DIR;
509 	}
510 
511 	*acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
512 	if (IS_ERR(*acl)) {
513 		error = PTR_ERR(*acl);
514 		*acl = NULL;
515 	}
516  out:
517 	posix_acl_release(pacl);
518 	posix_acl_release(dpacl);
519 	return error;
520 }
521 
522 #endif /* defined(CONFIG_NFS_V4) */
523 
524 #ifdef CONFIG_NFSD_V3
525 /*
526  * Check server access rights to a file system object
527  */
528 struct accessmap {
529 	u32		access;
530 	int		how;
531 };
532 static struct accessmap	nfs3_regaccess[] = {
533     {	NFS3_ACCESS_READ,	MAY_READ			},
534     {	NFS3_ACCESS_EXECUTE,	MAY_EXEC			},
535     {	NFS3_ACCESS_MODIFY,	MAY_WRITE|MAY_TRUNC		},
536     {	NFS3_ACCESS_EXTEND,	MAY_WRITE			},
537 
538     {	0,			0				}
539 };
540 
541 static struct accessmap	nfs3_diraccess[] = {
542     {	NFS3_ACCESS_READ,	MAY_READ			},
543     {	NFS3_ACCESS_LOOKUP,	MAY_EXEC			},
544     {	NFS3_ACCESS_MODIFY,	MAY_EXEC|MAY_WRITE|MAY_TRUNC	},
545     {	NFS3_ACCESS_EXTEND,	MAY_EXEC|MAY_WRITE		},
546     {	NFS3_ACCESS_DELETE,	MAY_REMOVE			},
547 
548     {	0,			0				}
549 };
550 
551 static struct accessmap	nfs3_anyaccess[] = {
552 	/* Some clients - Solaris 2.6 at least, make an access call
553 	 * to the server to check for access for things like /dev/null
554 	 * (which really, the server doesn't care about).  So
555 	 * We provide simple access checking for them, looking
556 	 * mainly at mode bits, and we make sure to ignore read-only
557 	 * filesystem checks
558 	 */
559     {	NFS3_ACCESS_READ,	MAY_READ			},
560     {	NFS3_ACCESS_EXECUTE,	MAY_EXEC			},
561     {	NFS3_ACCESS_MODIFY,	MAY_WRITE|MAY_LOCAL_ACCESS	},
562     {	NFS3_ACCESS_EXTEND,	MAY_WRITE|MAY_LOCAL_ACCESS	},
563 
564     {	0,			0				}
565 };
566 
567 int
568 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
569 {
570 	struct accessmap	*map;
571 	struct svc_export	*export;
572 	struct dentry		*dentry;
573 	u32			query, result = 0, sresult = 0;
574 	unsigned int		error;
575 
576 	error = fh_verify(rqstp, fhp, 0, MAY_NOP);
577 	if (error)
578 		goto out;
579 
580 	export = fhp->fh_export;
581 	dentry = fhp->fh_dentry;
582 
583 	if (S_ISREG(dentry->d_inode->i_mode))
584 		map = nfs3_regaccess;
585 	else if (S_ISDIR(dentry->d_inode->i_mode))
586 		map = nfs3_diraccess;
587 	else
588 		map = nfs3_anyaccess;
589 
590 
591 	query = *access;
592 	for  (; map->access; map++) {
593 		if (map->access & query) {
594 			unsigned int err2;
595 
596 			sresult |= map->access;
597 
598 			err2 = nfsd_permission(export, dentry, map->how);
599 			switch (err2) {
600 			case nfs_ok:
601 				result |= map->access;
602 				break;
603 
604 			/* the following error codes just mean the access was not allowed,
605 			 * rather than an error occurred */
606 			case nfserr_rofs:
607 			case nfserr_acces:
608 			case nfserr_perm:
609 				/* simply don't "or" in the access bit. */
610 				break;
611 			default:
612 				error = err2;
613 				goto out;
614 			}
615 		}
616 	}
617 	*access = result;
618 	if (supported)
619 		*supported = sresult;
620 
621  out:
622 	return error;
623 }
624 #endif /* CONFIG_NFSD_V3 */
625 
626 
627 
628 /*
629  * Open an existing file or directory.
630  * The access argument indicates the type of open (read/write/lock)
631  * N.B. After this call fhp needs an fh_put
632  */
633 int
634 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
635 			int access, struct file **filp)
636 {
637 	struct dentry	*dentry;
638 	struct inode	*inode;
639 	int		flags = O_RDONLY|O_LARGEFILE, err;
640 
641 	/*
642 	 * If we get here, then the client has already done an "open",
643 	 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
644 	 * in case a chmod has now revoked permission.
645 	 */
646 	err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
647 	if (err)
648 		goto out;
649 
650 	dentry = fhp->fh_dentry;
651 	inode = dentry->d_inode;
652 
653 	/* Disallow write access to files with the append-only bit set
654 	 * or any access when mandatory locking enabled
655 	 */
656 	err = nfserr_perm;
657 	if (IS_APPEND(inode) && (access & MAY_WRITE))
658 		goto out;
659 	if (IS_ISMNDLK(inode))
660 		goto out;
661 
662 	if (!inode->i_fop)
663 		goto out;
664 
665 	/*
666 	 * Check to see if there are any leases on this file.
667 	 * This may block while leases are broken.
668 	 */
669 	err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
670 	if (err == -EWOULDBLOCK)
671 		err = -ETIMEDOUT;
672 	if (err) /* NOMEM or WOULDBLOCK */
673 		goto out_nfserr;
674 
675 	if (access & MAY_WRITE) {
676 		if (access & MAY_READ)
677 			flags = O_RDWR|O_LARGEFILE;
678 		else
679 			flags = O_WRONLY|O_LARGEFILE;
680 
681 		DQUOT_INIT(inode);
682 	}
683 	*filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
684 	if (IS_ERR(*filp))
685 		err = PTR_ERR(*filp);
686 out_nfserr:
687 	if (err)
688 		err = nfserrno(err);
689 out:
690 	return err;
691 }
692 
693 /*
694  * Close a file.
695  */
696 void
697 nfsd_close(struct file *filp)
698 {
699 	fput(filp);
700 }
701 
702 /*
703  * Sync a file
704  * As this calls fsync (not fdatasync) there is no need for a write_inode
705  * after it.
706  */
707 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
708 			      const struct file_operations *fop)
709 {
710 	struct inode *inode = dp->d_inode;
711 	int (*fsync) (struct file *, struct dentry *, int);
712 	int err;
713 
714 	err = filemap_fdatawrite(inode->i_mapping);
715 	if (err == 0 && fop && (fsync = fop->fsync))
716 		err = fsync(filp, dp, 0);
717 	if (err == 0)
718 		err = filemap_fdatawait(inode->i_mapping);
719 
720 	return err;
721 }
722 
723 
724 static int
725 nfsd_sync(struct file *filp)
726 {
727         int err;
728 	struct inode *inode = filp->f_dentry->d_inode;
729 	dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
730 	mutex_lock(&inode->i_mutex);
731 	err=nfsd_dosync(filp, filp->f_dentry, filp->f_op);
732 	mutex_unlock(&inode->i_mutex);
733 
734 	return err;
735 }
736 
737 int
738 nfsd_sync_dir(struct dentry *dp)
739 {
740 	return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
741 }
742 
743 /*
744  * Obtain the readahead parameters for the file
745  * specified by (dev, ino).
746  */
747 static DEFINE_SPINLOCK(ra_lock);
748 
749 static inline struct raparms *
750 nfsd_get_raparms(dev_t dev, ino_t ino)
751 {
752 	struct raparms	*ra, **rap, **frap = NULL;
753 	int depth = 0;
754 
755 	spin_lock(&ra_lock);
756 	for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
757 		if (ra->p_ino == ino && ra->p_dev == dev)
758 			goto found;
759 		depth++;
760 		if (ra->p_count == 0)
761 			frap = rap;
762 	}
763 	depth = nfsdstats.ra_size*11/10;
764 	if (!frap) {
765 		spin_unlock(&ra_lock);
766 		return NULL;
767 	}
768 	rap = frap;
769 	ra = *frap;
770 	ra->p_dev = dev;
771 	ra->p_ino = ino;
772 	ra->p_set = 0;
773 found:
774 	if (rap != &raparm_cache) {
775 		*rap = ra->p_next;
776 		ra->p_next   = raparm_cache;
777 		raparm_cache = ra;
778 	}
779 	ra->p_count++;
780 	nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
781 	spin_unlock(&ra_lock);
782 	return ra;
783 }
784 
785 /*
786  * Grab and keep cached pages assosiated with a file in the svc_rqst
787  * so that they can be passed to the netowork sendmsg/sendpage routines
788  * directrly. They will be released after the sending has completed.
789  */
790 static int
791 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
792 {
793 	unsigned long count = desc->count;
794 	struct svc_rqst *rqstp = desc->arg.data;
795 
796 	if (size > count)
797 		size = count;
798 
799 	if (rqstp->rq_res.page_len == 0) {
800 		get_page(page);
801 		rqstp->rq_respages[rqstp->rq_resused++] = page;
802 		rqstp->rq_res.page_base = offset;
803 		rqstp->rq_res.page_len = size;
804 	} else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
805 		get_page(page);
806 		rqstp->rq_respages[rqstp->rq_resused++] = page;
807 		rqstp->rq_res.page_len += size;
808 	} else {
809 		rqstp->rq_res.page_len += size;
810 	}
811 
812 	desc->count = count - size;
813 	desc->written += size;
814 	return size;
815 }
816 
817 static int
818 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
819               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
820 {
821 	struct inode *inode;
822 	struct raparms	*ra;
823 	mm_segment_t	oldfs;
824 	int		err;
825 
826 	err = nfserr_perm;
827 	inode = file->f_dentry->d_inode;
828 #ifdef MSNFS
829 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
830 		(!lock_may_read(inode, offset, *count)))
831 		goto out;
832 #endif
833 
834 	/* Get readahead parameters */
835 	ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
836 
837 	if (ra && ra->p_set)
838 		file->f_ra = ra->p_ra;
839 
840 	if (file->f_op->sendfile && rqstp->rq_sendfile_ok) {
841 		svc_pushback_unused_pages(rqstp);
842 		err = file->f_op->sendfile(file, &offset, *count,
843 						 nfsd_read_actor, rqstp);
844 	} else {
845 		oldfs = get_fs();
846 		set_fs(KERNEL_DS);
847 		err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
848 		set_fs(oldfs);
849 	}
850 
851 	/* Write back readahead params */
852 	if (ra) {
853 		spin_lock(&ra_lock);
854 		ra->p_ra = file->f_ra;
855 		ra->p_set = 1;
856 		ra->p_count--;
857 		spin_unlock(&ra_lock);
858 	}
859 
860 	if (err >= 0) {
861 		nfsdstats.io_read += err;
862 		*count = err;
863 		err = 0;
864 		fsnotify_access(file->f_dentry);
865 	} else
866 		err = nfserrno(err);
867 out:
868 	return err;
869 }
870 
871 static void kill_suid(struct dentry *dentry)
872 {
873 	struct iattr	ia;
874 	ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
875 
876 	mutex_lock(&dentry->d_inode->i_mutex);
877 	notify_change(dentry, &ia);
878 	mutex_unlock(&dentry->d_inode->i_mutex);
879 }
880 
881 static int
882 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
883 				loff_t offset, struct kvec *vec, int vlen,
884 	   			unsigned long cnt, int *stablep)
885 {
886 	struct svc_export	*exp;
887 	struct dentry		*dentry;
888 	struct inode		*inode;
889 	mm_segment_t		oldfs;
890 	int			err = 0;
891 	int			stable = *stablep;
892 
893 #ifdef MSNFS
894 	err = nfserr_perm;
895 
896 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
897 		(!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
898 		goto out;
899 #endif
900 
901 	dentry = file->f_dentry;
902 	inode = dentry->d_inode;
903 	exp   = fhp->fh_export;
904 
905 	/*
906 	 * Request sync writes if
907 	 *  -	the sync export option has been set, or
908 	 *  -	the client requested O_SYNC behavior (NFSv3 feature).
909 	 *  -   The file system doesn't support fsync().
910 	 * When gathered writes have been configured for this volume,
911 	 * flushing the data to disk is handled separately below.
912 	 */
913 
914 	if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
915 	       stable = 2;
916 	       *stablep = 2; /* FILE_SYNC */
917 	}
918 
919 	if (!EX_ISSYNC(exp))
920 		stable = 0;
921 	if (stable && !EX_WGATHER(exp))
922 		file->f_flags |= O_SYNC;
923 
924 	/* Write the data. */
925 	oldfs = get_fs(); set_fs(KERNEL_DS);
926 	err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
927 	set_fs(oldfs);
928 	if (err >= 0) {
929 		nfsdstats.io_write += cnt;
930 		fsnotify_modify(file->f_dentry);
931 	}
932 
933 	/* clear setuid/setgid flag after write */
934 	if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
935 		kill_suid(dentry);
936 
937 	if (err >= 0 && stable) {
938 		static ino_t	last_ino;
939 		static dev_t	last_dev;
940 
941 		/*
942 		 * Gathered writes: If another process is currently
943 		 * writing to the file, there's a high chance
944 		 * this is another nfsd (triggered by a bulk write
945 		 * from a client's biod). Rather than syncing the
946 		 * file with each write request, we sleep for 10 msec.
947 		 *
948 		 * I don't know if this roughly approximates
949 		 * C. Juszak's idea of gathered writes, but it's a
950 		 * nice and simple solution (IMHO), and it seems to
951 		 * work:-)
952 		 */
953 		if (EX_WGATHER(exp)) {
954 			if (atomic_read(&inode->i_writecount) > 1
955 			    || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
956 				dprintk("nfsd: write defer %d\n", current->pid);
957 				msleep(10);
958 				dprintk("nfsd: write resume %d\n", current->pid);
959 			}
960 
961 			if (inode->i_state & I_DIRTY) {
962 				dprintk("nfsd: write sync %d\n", current->pid);
963 				err=nfsd_sync(file);
964 			}
965 #if 0
966 			wake_up(&inode->i_wait);
967 #endif
968 		}
969 		last_ino = inode->i_ino;
970 		last_dev = inode->i_sb->s_dev;
971 	}
972 
973 	dprintk("nfsd: write complete err=%d\n", err);
974 	if (err >= 0)
975 		err = 0;
976 	else
977 		err = nfserrno(err);
978 out:
979 	return err;
980 }
981 
982 /*
983  * Read data from a file. count must contain the requested read count
984  * on entry. On return, *count contains the number of bytes actually read.
985  * N.B. After this call fhp needs an fh_put
986  */
987 int
988 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
989 		loff_t offset, struct kvec *vec, int vlen,
990 		unsigned long *count)
991 {
992 	int		err;
993 
994 	if (file) {
995 		err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
996 				MAY_READ|MAY_OWNER_OVERRIDE);
997 		if (err)
998 			goto out;
999 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1000 	} else {
1001 		err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1002 		if (err)
1003 			goto out;
1004 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1005 		nfsd_close(file);
1006 	}
1007 out:
1008 	return err;
1009 }
1010 
1011 /*
1012  * Write data to a file.
1013  * The stable flag requests synchronous writes.
1014  * N.B. After this call fhp needs an fh_put
1015  */
1016 int
1017 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1018 		loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1019 		int *stablep)
1020 {
1021 	int			err = 0;
1022 
1023 	if (file) {
1024 		err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1025 				MAY_WRITE|MAY_OWNER_OVERRIDE);
1026 		if (err)
1027 			goto out;
1028 		err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1029 				stablep);
1030 	} else {
1031 		err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1032 		if (err)
1033 			goto out;
1034 
1035 		if (cnt)
1036 			err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1037 					     cnt, stablep);
1038 		nfsd_close(file);
1039 	}
1040 out:
1041 	return err;
1042 }
1043 
1044 #ifdef CONFIG_NFSD_V3
1045 /*
1046  * Commit all pending writes to stable storage.
1047  * Strictly speaking, we could sync just the indicated file region here,
1048  * but there's currently no way we can ask the VFS to do so.
1049  *
1050  * Unfortunately we cannot lock the file to make sure we return full WCC
1051  * data to the client, as locking happens lower down in the filesystem.
1052  */
1053 int
1054 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1055                loff_t offset, unsigned long count)
1056 {
1057 	struct file	*file;
1058 	int		err;
1059 
1060 	if ((u64)count > ~(u64)offset)
1061 		return nfserr_inval;
1062 
1063 	if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1064 		return err;
1065 	if (EX_ISSYNC(fhp->fh_export)) {
1066 		if (file->f_op && file->f_op->fsync) {
1067 			err = nfserrno(nfsd_sync(file));
1068 		} else {
1069 			err = nfserr_notsupp;
1070 		}
1071 	}
1072 
1073 	nfsd_close(file);
1074 	return err;
1075 }
1076 #endif /* CONFIG_NFSD_V3 */
1077 
1078 /*
1079  * Create a file (regular, directory, device, fifo); UNIX sockets
1080  * not yet implemented.
1081  * If the response fh has been verified, the parent directory should
1082  * already be locked. Note that the parent directory is left locked.
1083  *
1084  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1085  */
1086 int
1087 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1088 		char *fname, int flen, struct iattr *iap,
1089 		int type, dev_t rdev, struct svc_fh *resfhp)
1090 {
1091 	struct dentry	*dentry, *dchild = NULL;
1092 	struct inode	*dirp;
1093 	int		err;
1094 
1095 	err = nfserr_perm;
1096 	if (!flen)
1097 		goto out;
1098 	err = nfserr_exist;
1099 	if (isdotent(fname, flen))
1100 		goto out;
1101 
1102 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1103 	if (err)
1104 		goto out;
1105 
1106 	dentry = fhp->fh_dentry;
1107 	dirp = dentry->d_inode;
1108 
1109 	err = nfserr_notdir;
1110 	if(!dirp->i_op || !dirp->i_op->lookup)
1111 		goto out;
1112 	/*
1113 	 * Check whether the response file handle has been verified yet.
1114 	 * If it has, the parent directory should already be locked.
1115 	 */
1116 	if (!resfhp->fh_dentry) {
1117 		/* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1118 		fh_lock(fhp);
1119 		dchild = lookup_one_len(fname, dentry, flen);
1120 		err = PTR_ERR(dchild);
1121 		if (IS_ERR(dchild))
1122 			goto out_nfserr;
1123 		err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1124 		if (err)
1125 			goto out;
1126 	} else {
1127 		/* called from nfsd_proc_create */
1128 		dchild = dget(resfhp->fh_dentry);
1129 		if (!fhp->fh_locked) {
1130 			/* not actually possible */
1131 			printk(KERN_ERR
1132 				"nfsd_create: parent %s/%s not locked!\n",
1133 				dentry->d_parent->d_name.name,
1134 				dentry->d_name.name);
1135 			err = nfserr_io;
1136 			goto out;
1137 		}
1138 	}
1139 	/*
1140 	 * Make sure the child dentry is still negative ...
1141 	 */
1142 	err = nfserr_exist;
1143 	if (dchild->d_inode) {
1144 		dprintk("nfsd_create: dentry %s/%s not negative!\n",
1145 			dentry->d_name.name, dchild->d_name.name);
1146 		goto out;
1147 	}
1148 
1149 	if (!(iap->ia_valid & ATTR_MODE))
1150 		iap->ia_mode = 0;
1151 	iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1152 
1153 	/*
1154 	 * Get the dir op function pointer.
1155 	 */
1156 	err = nfserr_perm;
1157 	switch (type) {
1158 	case S_IFREG:
1159 		err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1160 		break;
1161 	case S_IFDIR:
1162 		err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1163 		break;
1164 	case S_IFCHR:
1165 	case S_IFBLK:
1166 	case S_IFIFO:
1167 	case S_IFSOCK:
1168 		err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1169 		break;
1170 	default:
1171 	        printk("nfsd: bad file type %o in nfsd_create\n", type);
1172 		err = -EINVAL;
1173 	}
1174 	if (err < 0)
1175 		goto out_nfserr;
1176 
1177 	if (EX_ISSYNC(fhp->fh_export)) {
1178 		err = nfserrno(nfsd_sync_dir(dentry));
1179 		write_inode_now(dchild->d_inode, 1);
1180 	}
1181 
1182 
1183 	/* Set file attributes. Mode has already been set and
1184 	 * setting uid/gid works only for root. Irix appears to
1185 	 * send along the gid when it tries to implement setgid
1186 	 * directories via NFS.
1187 	 */
1188 	if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1189 		int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1190 		if (err2)
1191 			err = err2;
1192 	}
1193 	/*
1194 	 * Update the file handle to get the new inode info.
1195 	 */
1196 	if (!err)
1197 		err = fh_update(resfhp);
1198 out:
1199 	if (dchild && !IS_ERR(dchild))
1200 		dput(dchild);
1201 	return err;
1202 
1203 out_nfserr:
1204 	err = nfserrno(err);
1205 	goto out;
1206 }
1207 
1208 #ifdef CONFIG_NFSD_V3
1209 /*
1210  * NFSv3 version of nfsd_create
1211  */
1212 int
1213 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1214 		char *fname, int flen, struct iattr *iap,
1215 		struct svc_fh *resfhp, int createmode, u32 *verifier,
1216 	        int *truncp)
1217 {
1218 	struct dentry	*dentry, *dchild = NULL;
1219 	struct inode	*dirp;
1220 	int		err;
1221 	__u32		v_mtime=0, v_atime=0;
1222 	int		v_mode=0;
1223 
1224 	err = nfserr_perm;
1225 	if (!flen)
1226 		goto out;
1227 	err = nfserr_exist;
1228 	if (isdotent(fname, flen))
1229 		goto out;
1230 	if (!(iap->ia_valid & ATTR_MODE))
1231 		iap->ia_mode = 0;
1232 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1233 	if (err)
1234 		goto out;
1235 
1236 	dentry = fhp->fh_dentry;
1237 	dirp = dentry->d_inode;
1238 
1239 	/* Get all the sanity checks out of the way before
1240 	 * we lock the parent. */
1241 	err = nfserr_notdir;
1242 	if(!dirp->i_op || !dirp->i_op->lookup)
1243 		goto out;
1244 	fh_lock(fhp);
1245 
1246 	/*
1247 	 * Compose the response file handle.
1248 	 */
1249 	dchild = lookup_one_len(fname, dentry, flen);
1250 	err = PTR_ERR(dchild);
1251 	if (IS_ERR(dchild))
1252 		goto out_nfserr;
1253 
1254 	err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1255 	if (err)
1256 		goto out;
1257 
1258 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1259 		/* while the verifier would fit in mtime+atime,
1260 		 * solaris7 gets confused (bugid 4218508) if these have
1261 		 * the high bit set, so we use the mode as well
1262 		 */
1263 		v_mtime = verifier[0]&0x7fffffff;
1264 		v_atime = verifier[1]&0x7fffffff;
1265 		v_mode  = S_IFREG
1266 			| ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1267 			| ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1268 			;
1269 	}
1270 
1271 	if (dchild->d_inode) {
1272 		err = 0;
1273 
1274 		switch (createmode) {
1275 		case NFS3_CREATE_UNCHECKED:
1276 			if (! S_ISREG(dchild->d_inode->i_mode))
1277 				err = nfserr_exist;
1278 			else if (truncp) {
1279 				/* in nfsv4, we need to treat this case a little
1280 				 * differently.  we don't want to truncate the
1281 				 * file now; this would be wrong if the OPEN
1282 				 * fails for some other reason.  furthermore,
1283 				 * if the size is nonzero, we should ignore it
1284 				 * according to spec!
1285 				 */
1286 				*truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1287 			}
1288 			else {
1289 				iap->ia_valid &= ATTR_SIZE;
1290 				goto set_attr;
1291 			}
1292 			break;
1293 		case NFS3_CREATE_EXCLUSIVE:
1294 			if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1295 			    && dchild->d_inode->i_atime.tv_sec == v_atime
1296 			    && dchild->d_inode->i_mode  == v_mode
1297 			    && dchild->d_inode->i_size  == 0 )
1298 				break;
1299 			 /* fallthru */
1300 		case NFS3_CREATE_GUARDED:
1301 			err = nfserr_exist;
1302 		}
1303 		goto out;
1304 	}
1305 
1306 	err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1307 	if (err < 0)
1308 		goto out_nfserr;
1309 
1310 	if (EX_ISSYNC(fhp->fh_export)) {
1311 		err = nfserrno(nfsd_sync_dir(dentry));
1312 		/* setattr will sync the child (or not) */
1313 	}
1314 
1315 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1316 		/* Cram the verifier into atime/mtime/mode */
1317 		iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1318 			| ATTR_MTIME_SET|ATTR_ATIME_SET
1319 			| ATTR_MODE;
1320 		/* XXX someone who knows this better please fix it for nsec */
1321 		iap->ia_mtime.tv_sec = v_mtime;
1322 		iap->ia_atime.tv_sec = v_atime;
1323 		iap->ia_mtime.tv_nsec = 0;
1324 		iap->ia_atime.tv_nsec = 0;
1325 		iap->ia_mode  = v_mode;
1326 	}
1327 
1328 	/* Set file attributes.
1329 	 * Mode has already been set but we might need to reset it
1330 	 * for CREATE_EXCLUSIVE
1331 	 * Irix appears to send along the gid when it tries to
1332 	 * implement setgid directories via NFS. Clear out all that cruft.
1333 	 */
1334  set_attr:
1335 	if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) {
1336  		int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1337 		if (err2)
1338 			err = err2;
1339 	}
1340 
1341 	/*
1342 	 * Update the filehandle to get the new inode info.
1343 	 */
1344 	if (!err)
1345 		err = fh_update(resfhp);
1346 
1347  out:
1348 	fh_unlock(fhp);
1349 	if (dchild && !IS_ERR(dchild))
1350 		dput(dchild);
1351  	return err;
1352 
1353  out_nfserr:
1354 	err = nfserrno(err);
1355 	goto out;
1356 }
1357 #endif /* CONFIG_NFSD_V3 */
1358 
1359 /*
1360  * Read a symlink. On entry, *lenp must contain the maximum path length that
1361  * fits into the buffer. On return, it contains the true length.
1362  * N.B. After this call fhp needs an fh_put
1363  */
1364 int
1365 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1366 {
1367 	struct dentry	*dentry;
1368 	struct inode	*inode;
1369 	mm_segment_t	oldfs;
1370 	int		err;
1371 
1372 	err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1373 	if (err)
1374 		goto out;
1375 
1376 	dentry = fhp->fh_dentry;
1377 	inode = dentry->d_inode;
1378 
1379 	err = nfserr_inval;
1380 	if (!inode->i_op || !inode->i_op->readlink)
1381 		goto out;
1382 
1383 	touch_atime(fhp->fh_export->ex_mnt, dentry);
1384 	/* N.B. Why does this call need a get_fs()??
1385 	 * Remove the set_fs and watch the fireworks:-) --okir
1386 	 */
1387 
1388 	oldfs = get_fs(); set_fs(KERNEL_DS);
1389 	err = inode->i_op->readlink(dentry, buf, *lenp);
1390 	set_fs(oldfs);
1391 
1392 	if (err < 0)
1393 		goto out_nfserr;
1394 	*lenp = err;
1395 	err = 0;
1396 out:
1397 	return err;
1398 
1399 out_nfserr:
1400 	err = nfserrno(err);
1401 	goto out;
1402 }
1403 
1404 /*
1405  * Create a symlink and look up its inode
1406  * N.B. After this call _both_ fhp and resfhp need an fh_put
1407  */
1408 int
1409 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1410 				char *fname, int flen,
1411 				char *path,  int plen,
1412 				struct svc_fh *resfhp,
1413 				struct iattr *iap)
1414 {
1415 	struct dentry	*dentry, *dnew;
1416 	int		err, cerr;
1417 	umode_t		mode;
1418 
1419 	err = nfserr_noent;
1420 	if (!flen || !plen)
1421 		goto out;
1422 	err = nfserr_exist;
1423 	if (isdotent(fname, flen))
1424 		goto out;
1425 
1426 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1427 	if (err)
1428 		goto out;
1429 	fh_lock(fhp);
1430 	dentry = fhp->fh_dentry;
1431 	dnew = lookup_one_len(fname, dentry, flen);
1432 	err = PTR_ERR(dnew);
1433 	if (IS_ERR(dnew))
1434 		goto out_nfserr;
1435 
1436 	mode = S_IALLUGO;
1437 	/* Only the MODE ATTRibute is even vaguely meaningful */
1438 	if (iap && (iap->ia_valid & ATTR_MODE))
1439 		mode = iap->ia_mode & S_IALLUGO;
1440 
1441 	if (unlikely(path[plen] != 0)) {
1442 		char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1443 		if (path_alloced == NULL)
1444 			err = -ENOMEM;
1445 		else {
1446 			strncpy(path_alloced, path, plen);
1447 			path_alloced[plen] = 0;
1448 			err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1449 			kfree(path_alloced);
1450 		}
1451 	} else
1452 		err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1453 
1454 	if (!err)
1455 		if (EX_ISSYNC(fhp->fh_export))
1456 			err = nfsd_sync_dir(dentry);
1457 	if (err)
1458 		err = nfserrno(err);
1459 	fh_unlock(fhp);
1460 
1461 	cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1462 	dput(dnew);
1463 	if (err==0) err = cerr;
1464 out:
1465 	return err;
1466 
1467 out_nfserr:
1468 	err = nfserrno(err);
1469 	goto out;
1470 }
1471 
1472 /*
1473  * Create a hardlink
1474  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1475  */
1476 int
1477 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1478 				char *name, int len, struct svc_fh *tfhp)
1479 {
1480 	struct dentry	*ddir, *dnew, *dold;
1481 	struct inode	*dirp, *dest;
1482 	int		err;
1483 
1484 	err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1485 	if (err)
1486 		goto out;
1487 	err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1488 	if (err)
1489 		goto out;
1490 
1491 	err = nfserr_perm;
1492 	if (!len)
1493 		goto out;
1494 	err = nfserr_exist;
1495 	if (isdotent(name, len))
1496 		goto out;
1497 
1498 	fh_lock(ffhp);
1499 	ddir = ffhp->fh_dentry;
1500 	dirp = ddir->d_inode;
1501 
1502 	dnew = lookup_one_len(name, ddir, len);
1503 	err = PTR_ERR(dnew);
1504 	if (IS_ERR(dnew))
1505 		goto out_nfserr;
1506 
1507 	dold = tfhp->fh_dentry;
1508 	dest = dold->d_inode;
1509 
1510 	err = vfs_link(dold, dirp, dnew);
1511 	if (!err) {
1512 		if (EX_ISSYNC(ffhp->fh_export)) {
1513 			err = nfserrno(nfsd_sync_dir(ddir));
1514 			write_inode_now(dest, 1);
1515 		}
1516 	} else {
1517 		if (err == -EXDEV && rqstp->rq_vers == 2)
1518 			err = nfserr_acces;
1519 		else
1520 			err = nfserrno(err);
1521 	}
1522 
1523 	dput(dnew);
1524 out_unlock:
1525 	fh_unlock(ffhp);
1526 out:
1527 	return err;
1528 
1529 out_nfserr:
1530 	err = nfserrno(err);
1531 	goto out_unlock;
1532 }
1533 
1534 /*
1535  * Rename a file
1536  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1537  */
1538 int
1539 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1540 			    struct svc_fh *tfhp, char *tname, int tlen)
1541 {
1542 	struct dentry	*fdentry, *tdentry, *odentry, *ndentry, *trap;
1543 	struct inode	*fdir, *tdir;
1544 	int		err;
1545 
1546 	err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1547 	if (err)
1548 		goto out;
1549 	err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1550 	if (err)
1551 		goto out;
1552 
1553 	fdentry = ffhp->fh_dentry;
1554 	fdir = fdentry->d_inode;
1555 
1556 	tdentry = tfhp->fh_dentry;
1557 	tdir = tdentry->d_inode;
1558 
1559 	err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1560 	if (ffhp->fh_export != tfhp->fh_export)
1561 		goto out;
1562 
1563 	err = nfserr_perm;
1564 	if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1565 		goto out;
1566 
1567 	/* cannot use fh_lock as we need deadlock protective ordering
1568 	 * so do it by hand */
1569 	trap = lock_rename(tdentry, fdentry);
1570 	ffhp->fh_locked = tfhp->fh_locked = 1;
1571 	fill_pre_wcc(ffhp);
1572 	fill_pre_wcc(tfhp);
1573 
1574 	odentry = lookup_one_len(fname, fdentry, flen);
1575 	err = PTR_ERR(odentry);
1576 	if (IS_ERR(odentry))
1577 		goto out_nfserr;
1578 
1579 	err = -ENOENT;
1580 	if (!odentry->d_inode)
1581 		goto out_dput_old;
1582 	err = -EINVAL;
1583 	if (odentry == trap)
1584 		goto out_dput_old;
1585 
1586 	ndentry = lookup_one_len(tname, tdentry, tlen);
1587 	err = PTR_ERR(ndentry);
1588 	if (IS_ERR(ndentry))
1589 		goto out_dput_old;
1590 	err = -ENOTEMPTY;
1591 	if (ndentry == trap)
1592 		goto out_dput_new;
1593 
1594 #ifdef MSNFS
1595 	if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1596 		((atomic_read(&odentry->d_count) > 1)
1597 		 || (atomic_read(&ndentry->d_count) > 1))) {
1598 			err = -EPERM;
1599 	} else
1600 #endif
1601 	err = vfs_rename(fdir, odentry, tdir, ndentry);
1602 	if (!err && EX_ISSYNC(tfhp->fh_export)) {
1603 		err = nfsd_sync_dir(tdentry);
1604 		if (!err)
1605 			err = nfsd_sync_dir(fdentry);
1606 	}
1607 
1608  out_dput_new:
1609 	dput(ndentry);
1610  out_dput_old:
1611 	dput(odentry);
1612  out_nfserr:
1613 	if (err)
1614 		err = nfserrno(err);
1615 
1616 	/* we cannot reply on fh_unlock on the two filehandles,
1617 	 * as that would do the wrong thing if the two directories
1618 	 * were the same, so again we do it by hand
1619 	 */
1620 	fill_post_wcc(ffhp);
1621 	fill_post_wcc(tfhp);
1622 	unlock_rename(tdentry, fdentry);
1623 	ffhp->fh_locked = tfhp->fh_locked = 0;
1624 
1625 out:
1626 	return err;
1627 }
1628 
1629 /*
1630  * Unlink a file or directory
1631  * N.B. After this call fhp needs an fh_put
1632  */
1633 int
1634 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1635 				char *fname, int flen)
1636 {
1637 	struct dentry	*dentry, *rdentry;
1638 	struct inode	*dirp;
1639 	int		err;
1640 
1641 	err = nfserr_acces;
1642 	if (!flen || isdotent(fname, flen))
1643 		goto out;
1644 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1645 	if (err)
1646 		goto out;
1647 
1648 	fh_lock(fhp);
1649 	dentry = fhp->fh_dentry;
1650 	dirp = dentry->d_inode;
1651 
1652 	rdentry = lookup_one_len(fname, dentry, flen);
1653 	err = PTR_ERR(rdentry);
1654 	if (IS_ERR(rdentry))
1655 		goto out_nfserr;
1656 
1657 	if (!rdentry->d_inode) {
1658 		dput(rdentry);
1659 		err = nfserr_noent;
1660 		goto out;
1661 	}
1662 
1663 	if (!type)
1664 		type = rdentry->d_inode->i_mode & S_IFMT;
1665 
1666 	if (type != S_IFDIR) { /* It's UNLINK */
1667 #ifdef MSNFS
1668 		if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1669 			(atomic_read(&rdentry->d_count) > 1)) {
1670 			err = -EPERM;
1671 		} else
1672 #endif
1673 		err = vfs_unlink(dirp, rdentry);
1674 	} else { /* It's RMDIR */
1675 		err = vfs_rmdir(dirp, rdentry);
1676 	}
1677 
1678 	dput(rdentry);
1679 
1680 	if (err == 0 &&
1681 	    EX_ISSYNC(fhp->fh_export))
1682 			err = nfsd_sync_dir(dentry);
1683 
1684 out_nfserr:
1685 	err = nfserrno(err);
1686 out:
1687 	return err;
1688 }
1689 
1690 /*
1691  * Read entries from a directory.
1692  * The  NFSv3/4 verifier we ignore for now.
1693  */
1694 int
1695 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1696 	     struct readdir_cd *cdp, encode_dent_fn func)
1697 {
1698 	int		err;
1699 	struct file	*file;
1700 	loff_t		offset = *offsetp;
1701 
1702 	err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1703 	if (err)
1704 		goto out;
1705 
1706 	offset = vfs_llseek(file, offset, 0);
1707 	if (offset < 0) {
1708 		err = nfserrno((int)offset);
1709 		goto out_close;
1710 	}
1711 
1712 	/*
1713 	 * Read the directory entries. This silly loop is necessary because
1714 	 * readdir() is not guaranteed to fill up the entire buffer, but
1715 	 * may choose to do less.
1716 	 */
1717 
1718 	do {
1719 		cdp->err = nfserr_eof; /* will be cleared on successful read */
1720 		err = vfs_readdir(file, (filldir_t) func, cdp);
1721 	} while (err >=0 && cdp->err == nfs_ok);
1722 	if (err)
1723 		err = nfserrno(err);
1724 	else
1725 		err = cdp->err;
1726 	*offsetp = vfs_llseek(file, 0, 1);
1727 
1728 	if (err == nfserr_eof || err == nfserr_toosmall)
1729 		err = nfs_ok; /* can still be found in ->err */
1730 out_close:
1731 	nfsd_close(file);
1732 out:
1733 	return err;
1734 }
1735 
1736 /*
1737  * Get file system stats
1738  * N.B. After this call fhp needs an fh_put
1739  */
1740 int
1741 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1742 {
1743 	int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1744 	if (!err && vfs_statfs(fhp->fh_dentry,stat))
1745 		err = nfserr_io;
1746 	return err;
1747 }
1748 
1749 /*
1750  * Check for a user's access permissions to this inode.
1751  */
1752 int
1753 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1754 {
1755 	struct inode	*inode = dentry->d_inode;
1756 	int		err;
1757 
1758 	if (acc == MAY_NOP)
1759 		return 0;
1760 #if 0
1761 	dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1762 		acc,
1763 		(acc & MAY_READ)?	" read"  : "",
1764 		(acc & MAY_WRITE)?	" write" : "",
1765 		(acc & MAY_EXEC)?	" exec"  : "",
1766 		(acc & MAY_SATTR)?	" sattr" : "",
1767 		(acc & MAY_TRUNC)?	" trunc" : "",
1768 		(acc & MAY_LOCK)?	" lock"  : "",
1769 		(acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1770 		inode->i_mode,
1771 		IS_IMMUTABLE(inode)?	" immut" : "",
1772 		IS_APPEND(inode)?	" append" : "",
1773 		IS_RDONLY(inode)?	" ro" : "");
1774 	dprintk("      owner %d/%d user %d/%d\n",
1775 		inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1776 #endif
1777 
1778 	/* Normally we reject any write/sattr etc access on a read-only file
1779 	 * system.  But if it is IRIX doing check on write-access for a
1780 	 * device special file, we ignore rofs.
1781 	 */
1782 	if (!(acc & MAY_LOCAL_ACCESS))
1783 		if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1784 			if (EX_RDONLY(exp) || IS_RDONLY(inode))
1785 				return nfserr_rofs;
1786 			if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1787 				return nfserr_perm;
1788 		}
1789 	if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1790 		return nfserr_perm;
1791 
1792 	if (acc & MAY_LOCK) {
1793 		/* If we cannot rely on authentication in NLM requests,
1794 		 * just allow locks, otherwise require read permission, or
1795 		 * ownership
1796 		 */
1797 		if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1798 			return 0;
1799 		else
1800 			acc = MAY_READ | MAY_OWNER_OVERRIDE;
1801 	}
1802 	/*
1803 	 * The file owner always gets access permission for accesses that
1804 	 * would normally be checked at open time. This is to make
1805 	 * file access work even when the client has done a fchmod(fd, 0).
1806 	 *
1807 	 * However, `cp foo bar' should fail nevertheless when bar is
1808 	 * readonly. A sensible way to do this might be to reject all
1809 	 * attempts to truncate a read-only file, because a creat() call
1810 	 * always implies file truncation.
1811 	 * ... but this isn't really fair.  A process may reasonably call
1812 	 * ftruncate on an open file descriptor on a file with perm 000.
1813 	 * We must trust the client to do permission checking - using "ACCESS"
1814 	 * with NFSv3.
1815 	 */
1816 	if ((acc & MAY_OWNER_OVERRIDE) &&
1817 	    inode->i_uid == current->fsuid)
1818 		return 0;
1819 
1820 	err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1821 
1822 	/* Allow read access to binaries even when mode 111 */
1823 	if (err == -EACCES && S_ISREG(inode->i_mode) &&
1824 	    acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1825 		err = permission(inode, MAY_EXEC, NULL);
1826 
1827 	return err? nfserrno(err) : 0;
1828 }
1829 
1830 void
1831 nfsd_racache_shutdown(void)
1832 {
1833 	if (!raparm_cache)
1834 		return;
1835 	dprintk("nfsd: freeing readahead buffers.\n");
1836 	kfree(raparml);
1837 	raparm_cache = raparml = NULL;
1838 }
1839 /*
1840  * Initialize readahead param cache
1841  */
1842 int
1843 nfsd_racache_init(int cache_size)
1844 {
1845 	int	i;
1846 
1847 	if (raparm_cache)
1848 		return 0;
1849 	raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1850 
1851 	if (raparml != NULL) {
1852 		dprintk("nfsd: allocating %d readahead buffers.\n",
1853 			cache_size);
1854 		memset(raparml, 0, sizeof(struct raparms) * cache_size);
1855 		for (i = 0; i < cache_size - 1; i++) {
1856 			raparml[i].p_next = raparml + i + 1;
1857 		}
1858 		raparm_cache = raparml;
1859 	} else {
1860 		printk(KERN_WARNING
1861 		       "nfsd: Could not allocate memory read-ahead cache.\n");
1862 		return -ENOMEM;
1863 	}
1864 	nfsdstats.ra_size = cache_size;
1865 	return 0;
1866 }
1867 
1868 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1869 struct posix_acl *
1870 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1871 {
1872 	struct inode *inode = fhp->fh_dentry->d_inode;
1873 	char *name;
1874 	void *value = NULL;
1875 	ssize_t size;
1876 	struct posix_acl *acl;
1877 
1878 	if (!IS_POSIXACL(inode))
1879 		return ERR_PTR(-EOPNOTSUPP);
1880 
1881 	switch (type) {
1882 	case ACL_TYPE_ACCESS:
1883 		name = POSIX_ACL_XATTR_ACCESS;
1884 		break;
1885 	case ACL_TYPE_DEFAULT:
1886 		name = POSIX_ACL_XATTR_DEFAULT;
1887 		break;
1888 	default:
1889 		return ERR_PTR(-EOPNOTSUPP);
1890 	}
1891 
1892 	size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1893 	if (size < 0)
1894 		return ERR_PTR(size);
1895 
1896 	acl = posix_acl_from_xattr(value, size);
1897 	kfree(value);
1898 	return acl;
1899 }
1900 
1901 int
1902 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1903 {
1904 	struct inode *inode = fhp->fh_dentry->d_inode;
1905 	char *name;
1906 	void *value = NULL;
1907 	size_t size;
1908 	int error;
1909 
1910 	if (!IS_POSIXACL(inode) || !inode->i_op ||
1911 	    !inode->i_op->setxattr || !inode->i_op->removexattr)
1912 		return -EOPNOTSUPP;
1913 	switch(type) {
1914 		case ACL_TYPE_ACCESS:
1915 			name = POSIX_ACL_XATTR_ACCESS;
1916 			break;
1917 		case ACL_TYPE_DEFAULT:
1918 			name = POSIX_ACL_XATTR_DEFAULT;
1919 			break;
1920 		default:
1921 			return -EOPNOTSUPP;
1922 	}
1923 
1924 	if (acl && acl->a_count) {
1925 		size = posix_acl_xattr_size(acl->a_count);
1926 		value = kmalloc(size, GFP_KERNEL);
1927 		if (!value)
1928 			return -ENOMEM;
1929 		error = posix_acl_to_xattr(acl, value, size);
1930 		if (error < 0)
1931 			goto getout;
1932 		size = error;
1933 	} else
1934 		size = 0;
1935 
1936 	if (size)
1937 		error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1938 	else {
1939 		if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1940 			error = 0;
1941 		else {
1942 			error = vfs_removexattr(fhp->fh_dentry, name);
1943 			if (error == -ENODATA)
1944 				error = 0;
1945 		}
1946 	}
1947 
1948 getout:
1949 	kfree(value);
1950 	return error;
1951 }
1952 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
1953