xref: /linux/fs/nfsd/vfs.c (revision f3d9478b2ce468c3115b02ecae7e975990697f15)
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 		flags = O_WRONLY|O_LARGEFILE;
677 
678 		DQUOT_INIT(inode);
679 	}
680 	*filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
681 	if (IS_ERR(*filp))
682 		err = PTR_ERR(*filp);
683 out_nfserr:
684 	if (err)
685 		err = nfserrno(err);
686 out:
687 	return err;
688 }
689 
690 /*
691  * Close a file.
692  */
693 void
694 nfsd_close(struct file *filp)
695 {
696 	fput(filp);
697 }
698 
699 /*
700  * Sync a file
701  * As this calls fsync (not fdatasync) there is no need for a write_inode
702  * after it.
703  */
704 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
705 			      const struct file_operations *fop)
706 {
707 	struct inode *inode = dp->d_inode;
708 	int (*fsync) (struct file *, struct dentry *, int);
709 	int err;
710 
711 	err = filemap_fdatawrite(inode->i_mapping);
712 	if (err == 0 && fop && (fsync = fop->fsync))
713 		err = fsync(filp, dp, 0);
714 	if (err == 0)
715 		err = filemap_fdatawait(inode->i_mapping);
716 
717 	return err;
718 }
719 
720 
721 static int
722 nfsd_sync(struct file *filp)
723 {
724         int err;
725 	struct inode *inode = filp->f_dentry->d_inode;
726 	dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
727 	mutex_lock(&inode->i_mutex);
728 	err=nfsd_dosync(filp, filp->f_dentry, filp->f_op);
729 	mutex_unlock(&inode->i_mutex);
730 
731 	return err;
732 }
733 
734 int
735 nfsd_sync_dir(struct dentry *dp)
736 {
737 	return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
738 }
739 
740 /*
741  * Obtain the readahead parameters for the file
742  * specified by (dev, ino).
743  */
744 static DEFINE_SPINLOCK(ra_lock);
745 
746 static inline struct raparms *
747 nfsd_get_raparms(dev_t dev, ino_t ino)
748 {
749 	struct raparms	*ra, **rap, **frap = NULL;
750 	int depth = 0;
751 
752 	spin_lock(&ra_lock);
753 	for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
754 		if (ra->p_ino == ino && ra->p_dev == dev)
755 			goto found;
756 		depth++;
757 		if (ra->p_count == 0)
758 			frap = rap;
759 	}
760 	depth = nfsdstats.ra_size*11/10;
761 	if (!frap) {
762 		spin_unlock(&ra_lock);
763 		return NULL;
764 	}
765 	rap = frap;
766 	ra = *frap;
767 	ra->p_dev = dev;
768 	ra->p_ino = ino;
769 	ra->p_set = 0;
770 found:
771 	if (rap != &raparm_cache) {
772 		*rap = ra->p_next;
773 		ra->p_next   = raparm_cache;
774 		raparm_cache = ra;
775 	}
776 	ra->p_count++;
777 	nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
778 	spin_unlock(&ra_lock);
779 	return ra;
780 }
781 
782 /*
783  * Grab and keep cached pages assosiated with a file in the svc_rqst
784  * so that they can be passed to the netowork sendmsg/sendpage routines
785  * directrly. They will be released after the sending has completed.
786  */
787 static int
788 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
789 {
790 	unsigned long count = desc->count;
791 	struct svc_rqst *rqstp = desc->arg.data;
792 
793 	if (size > count)
794 		size = count;
795 
796 	if (rqstp->rq_res.page_len == 0) {
797 		get_page(page);
798 		rqstp->rq_respages[rqstp->rq_resused++] = page;
799 		rqstp->rq_res.page_base = offset;
800 		rqstp->rq_res.page_len = size;
801 	} else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
802 		get_page(page);
803 		rqstp->rq_respages[rqstp->rq_resused++] = page;
804 		rqstp->rq_res.page_len += size;
805 	} else {
806 		rqstp->rq_res.page_len += size;
807 	}
808 
809 	desc->count = count - size;
810 	desc->written += size;
811 	return size;
812 }
813 
814 static int
815 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
816               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
817 {
818 	struct inode *inode;
819 	struct raparms	*ra;
820 	mm_segment_t	oldfs;
821 	int		err;
822 
823 	err = nfserr_perm;
824 	inode = file->f_dentry->d_inode;
825 #ifdef MSNFS
826 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
827 		(!lock_may_read(inode, offset, *count)))
828 		goto out;
829 #endif
830 
831 	/* Get readahead parameters */
832 	ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
833 
834 	if (ra && ra->p_set)
835 		file->f_ra = ra->p_ra;
836 
837 	if (file->f_op->sendfile) {
838 		svc_pushback_unused_pages(rqstp);
839 		err = file->f_op->sendfile(file, &offset, *count,
840 						 nfsd_read_actor, rqstp);
841 	} else {
842 		oldfs = get_fs();
843 		set_fs(KERNEL_DS);
844 		err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
845 		set_fs(oldfs);
846 	}
847 
848 	/* Write back readahead params */
849 	if (ra) {
850 		spin_lock(&ra_lock);
851 		ra->p_ra = file->f_ra;
852 		ra->p_set = 1;
853 		ra->p_count--;
854 		spin_unlock(&ra_lock);
855 	}
856 
857 	if (err >= 0) {
858 		nfsdstats.io_read += err;
859 		*count = err;
860 		err = 0;
861 		fsnotify_access(file->f_dentry);
862 	} else
863 		err = nfserrno(err);
864 out:
865 	return err;
866 }
867 
868 static void kill_suid(struct dentry *dentry)
869 {
870 	struct iattr	ia;
871 	ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
872 
873 	mutex_lock(&dentry->d_inode->i_mutex);
874 	notify_change(dentry, &ia);
875 	mutex_unlock(&dentry->d_inode->i_mutex);
876 }
877 
878 static int
879 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
880 				loff_t offset, struct kvec *vec, int vlen,
881 	   			unsigned long cnt, int *stablep)
882 {
883 	struct svc_export	*exp;
884 	struct dentry		*dentry;
885 	struct inode		*inode;
886 	mm_segment_t		oldfs;
887 	int			err = 0;
888 	int			stable = *stablep;
889 
890 #ifdef MSNFS
891 	err = nfserr_perm;
892 
893 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
894 		(!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
895 		goto out;
896 #endif
897 
898 	dentry = file->f_dentry;
899 	inode = dentry->d_inode;
900 	exp   = fhp->fh_export;
901 
902 	/*
903 	 * Request sync writes if
904 	 *  -	the sync export option has been set, or
905 	 *  -	the client requested O_SYNC behavior (NFSv3 feature).
906 	 *  -   The file system doesn't support fsync().
907 	 * When gathered writes have been configured for this volume,
908 	 * flushing the data to disk is handled separately below.
909 	 */
910 
911 	if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
912 	       stable = 2;
913 	       *stablep = 2; /* FILE_SYNC */
914 	}
915 
916 	if (!EX_ISSYNC(exp))
917 		stable = 0;
918 	if (stable && !EX_WGATHER(exp))
919 		file->f_flags |= O_SYNC;
920 
921 	/* Write the data. */
922 	oldfs = get_fs(); set_fs(KERNEL_DS);
923 	err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
924 	set_fs(oldfs);
925 	if (err >= 0) {
926 		nfsdstats.io_write += cnt;
927 		fsnotify_modify(file->f_dentry);
928 	}
929 
930 	/* clear setuid/setgid flag after write */
931 	if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
932 		kill_suid(dentry);
933 
934 	if (err >= 0 && stable) {
935 		static ino_t	last_ino;
936 		static dev_t	last_dev;
937 
938 		/*
939 		 * Gathered writes: If another process is currently
940 		 * writing to the file, there's a high chance
941 		 * this is another nfsd (triggered by a bulk write
942 		 * from a client's biod). Rather than syncing the
943 		 * file with each write request, we sleep for 10 msec.
944 		 *
945 		 * I don't know if this roughly approximates
946 		 * C. Juszak's idea of gathered writes, but it's a
947 		 * nice and simple solution (IMHO), and it seems to
948 		 * work:-)
949 		 */
950 		if (EX_WGATHER(exp)) {
951 			if (atomic_read(&inode->i_writecount) > 1
952 			    || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
953 				dprintk("nfsd: write defer %d\n", current->pid);
954 				msleep(10);
955 				dprintk("nfsd: write resume %d\n", current->pid);
956 			}
957 
958 			if (inode->i_state & I_DIRTY) {
959 				dprintk("nfsd: write sync %d\n", current->pid);
960 				err=nfsd_sync(file);
961 			}
962 #if 0
963 			wake_up(&inode->i_wait);
964 #endif
965 		}
966 		last_ino = inode->i_ino;
967 		last_dev = inode->i_sb->s_dev;
968 	}
969 
970 	dprintk("nfsd: write complete err=%d\n", err);
971 	if (err >= 0)
972 		err = 0;
973 	else
974 		err = nfserrno(err);
975 out:
976 	return err;
977 }
978 
979 /*
980  * Read data from a file. count must contain the requested read count
981  * on entry. On return, *count contains the number of bytes actually read.
982  * N.B. After this call fhp needs an fh_put
983  */
984 int
985 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
986 		loff_t offset, struct kvec *vec, int vlen,
987 		unsigned long *count)
988 {
989 	int		err;
990 
991 	if (file) {
992 		err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
993 				MAY_READ|MAY_OWNER_OVERRIDE);
994 		if (err)
995 			goto out;
996 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
997 	} else {
998 		err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
999 		if (err)
1000 			goto out;
1001 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1002 		nfsd_close(file);
1003 	}
1004 out:
1005 	return err;
1006 }
1007 
1008 /*
1009  * Write data to a file.
1010  * The stable flag requests synchronous writes.
1011  * N.B. After this call fhp needs an fh_put
1012  */
1013 int
1014 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1015 		loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1016 		int *stablep)
1017 {
1018 	int			err = 0;
1019 
1020 	if (file) {
1021 		err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1022 				MAY_WRITE|MAY_OWNER_OVERRIDE);
1023 		if (err)
1024 			goto out;
1025 		err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1026 				stablep);
1027 	} else {
1028 		err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1029 		if (err)
1030 			goto out;
1031 
1032 		if (cnt)
1033 			err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1034 					     cnt, stablep);
1035 		nfsd_close(file);
1036 	}
1037 out:
1038 	return err;
1039 }
1040 
1041 #ifdef CONFIG_NFSD_V3
1042 /*
1043  * Commit all pending writes to stable storage.
1044  * Strictly speaking, we could sync just the indicated file region here,
1045  * but there's currently no way we can ask the VFS to do so.
1046  *
1047  * Unfortunately we cannot lock the file to make sure we return full WCC
1048  * data to the client, as locking happens lower down in the filesystem.
1049  */
1050 int
1051 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1052                loff_t offset, unsigned long count)
1053 {
1054 	struct file	*file;
1055 	int		err;
1056 
1057 	if ((u64)count > ~(u64)offset)
1058 		return nfserr_inval;
1059 
1060 	if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1061 		return err;
1062 	if (EX_ISSYNC(fhp->fh_export)) {
1063 		if (file->f_op && file->f_op->fsync) {
1064 			err = nfserrno(nfsd_sync(file));
1065 		} else {
1066 			err = nfserr_notsupp;
1067 		}
1068 	}
1069 
1070 	nfsd_close(file);
1071 	return err;
1072 }
1073 #endif /* CONFIG_NFSD_V3 */
1074 
1075 /*
1076  * Create a file (regular, directory, device, fifo); UNIX sockets
1077  * not yet implemented.
1078  * If the response fh has been verified, the parent directory should
1079  * already be locked. Note that the parent directory is left locked.
1080  *
1081  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1082  */
1083 int
1084 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1085 		char *fname, int flen, struct iattr *iap,
1086 		int type, dev_t rdev, struct svc_fh *resfhp)
1087 {
1088 	struct dentry	*dentry, *dchild = NULL;
1089 	struct inode	*dirp;
1090 	int		err;
1091 
1092 	err = nfserr_perm;
1093 	if (!flen)
1094 		goto out;
1095 	err = nfserr_exist;
1096 	if (isdotent(fname, flen))
1097 		goto out;
1098 
1099 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1100 	if (err)
1101 		goto out;
1102 
1103 	dentry = fhp->fh_dentry;
1104 	dirp = dentry->d_inode;
1105 
1106 	err = nfserr_notdir;
1107 	if(!dirp->i_op || !dirp->i_op->lookup)
1108 		goto out;
1109 	/*
1110 	 * Check whether the response file handle has been verified yet.
1111 	 * If it has, the parent directory should already be locked.
1112 	 */
1113 	if (!resfhp->fh_dentry) {
1114 		/* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1115 		fh_lock(fhp);
1116 		dchild = lookup_one_len(fname, dentry, flen);
1117 		err = PTR_ERR(dchild);
1118 		if (IS_ERR(dchild))
1119 			goto out_nfserr;
1120 		err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1121 		if (err)
1122 			goto out;
1123 	} else {
1124 		/* called from nfsd_proc_create */
1125 		dchild = dget(resfhp->fh_dentry);
1126 		if (!fhp->fh_locked) {
1127 			/* not actually possible */
1128 			printk(KERN_ERR
1129 				"nfsd_create: parent %s/%s not locked!\n",
1130 				dentry->d_parent->d_name.name,
1131 				dentry->d_name.name);
1132 			err = nfserr_io;
1133 			goto out;
1134 		}
1135 	}
1136 	/*
1137 	 * Make sure the child dentry is still negative ...
1138 	 */
1139 	err = nfserr_exist;
1140 	if (dchild->d_inode) {
1141 		dprintk("nfsd_create: dentry %s/%s not negative!\n",
1142 			dentry->d_name.name, dchild->d_name.name);
1143 		goto out;
1144 	}
1145 
1146 	if (!(iap->ia_valid & ATTR_MODE))
1147 		iap->ia_mode = 0;
1148 	iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1149 
1150 	/*
1151 	 * Get the dir op function pointer.
1152 	 */
1153 	err = nfserr_perm;
1154 	switch (type) {
1155 	case S_IFREG:
1156 		err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1157 		break;
1158 	case S_IFDIR:
1159 		err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1160 		break;
1161 	case S_IFCHR:
1162 	case S_IFBLK:
1163 	case S_IFIFO:
1164 	case S_IFSOCK:
1165 		err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1166 		break;
1167 	default:
1168 	        printk("nfsd: bad file type %o in nfsd_create\n", type);
1169 		err = -EINVAL;
1170 	}
1171 	if (err < 0)
1172 		goto out_nfserr;
1173 
1174 	if (EX_ISSYNC(fhp->fh_export)) {
1175 		err = nfserrno(nfsd_sync_dir(dentry));
1176 		write_inode_now(dchild->d_inode, 1);
1177 	}
1178 
1179 
1180 	/* Set file attributes. Mode has already been set and
1181 	 * setting uid/gid works only for root. Irix appears to
1182 	 * send along the gid when it tries to implement setgid
1183 	 * directories via NFS.
1184 	 */
1185 	if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1186 		int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1187 		if (err2)
1188 			err = err2;
1189 	}
1190 	/*
1191 	 * Update the file handle to get the new inode info.
1192 	 */
1193 	if (!err)
1194 		err = fh_update(resfhp);
1195 out:
1196 	if (dchild && !IS_ERR(dchild))
1197 		dput(dchild);
1198 	return err;
1199 
1200 out_nfserr:
1201 	err = nfserrno(err);
1202 	goto out;
1203 }
1204 
1205 #ifdef CONFIG_NFSD_V3
1206 /*
1207  * NFSv3 version of nfsd_create
1208  */
1209 int
1210 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1211 		char *fname, int flen, struct iattr *iap,
1212 		struct svc_fh *resfhp, int createmode, u32 *verifier,
1213 	        int *truncp)
1214 {
1215 	struct dentry	*dentry, *dchild = NULL;
1216 	struct inode	*dirp;
1217 	int		err;
1218 	__u32		v_mtime=0, v_atime=0;
1219 	int		v_mode=0;
1220 
1221 	err = nfserr_perm;
1222 	if (!flen)
1223 		goto out;
1224 	err = nfserr_exist;
1225 	if (isdotent(fname, flen))
1226 		goto out;
1227 	if (!(iap->ia_valid & ATTR_MODE))
1228 		iap->ia_mode = 0;
1229 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1230 	if (err)
1231 		goto out;
1232 
1233 	dentry = fhp->fh_dentry;
1234 	dirp = dentry->d_inode;
1235 
1236 	/* Get all the sanity checks out of the way before
1237 	 * we lock the parent. */
1238 	err = nfserr_notdir;
1239 	if(!dirp->i_op || !dirp->i_op->lookup)
1240 		goto out;
1241 	fh_lock(fhp);
1242 
1243 	/*
1244 	 * Compose the response file handle.
1245 	 */
1246 	dchild = lookup_one_len(fname, dentry, flen);
1247 	err = PTR_ERR(dchild);
1248 	if (IS_ERR(dchild))
1249 		goto out_nfserr;
1250 
1251 	err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1252 	if (err)
1253 		goto out;
1254 
1255 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1256 		/* while the verifier would fit in mtime+atime,
1257 		 * solaris7 gets confused (bugid 4218508) if these have
1258 		 * the high bit set, so we use the mode as well
1259 		 */
1260 		v_mtime = verifier[0]&0x7fffffff;
1261 		v_atime = verifier[1]&0x7fffffff;
1262 		v_mode  = S_IFREG
1263 			| ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1264 			| ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1265 			;
1266 	}
1267 
1268 	if (dchild->d_inode) {
1269 		err = 0;
1270 
1271 		switch (createmode) {
1272 		case NFS3_CREATE_UNCHECKED:
1273 			if (! S_ISREG(dchild->d_inode->i_mode))
1274 				err = nfserr_exist;
1275 			else if (truncp) {
1276 				/* in nfsv4, we need to treat this case a little
1277 				 * differently.  we don't want to truncate the
1278 				 * file now; this would be wrong if the OPEN
1279 				 * fails for some other reason.  furthermore,
1280 				 * if the size is nonzero, we should ignore it
1281 				 * according to spec!
1282 				 */
1283 				*truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1284 			}
1285 			else {
1286 				iap->ia_valid &= ATTR_SIZE;
1287 				goto set_attr;
1288 			}
1289 			break;
1290 		case NFS3_CREATE_EXCLUSIVE:
1291 			if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1292 			    && dchild->d_inode->i_atime.tv_sec == v_atime
1293 			    && dchild->d_inode->i_mode  == v_mode
1294 			    && dchild->d_inode->i_size  == 0 )
1295 				break;
1296 			 /* fallthru */
1297 		case NFS3_CREATE_GUARDED:
1298 			err = nfserr_exist;
1299 		}
1300 		goto out;
1301 	}
1302 
1303 	err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1304 	if (err < 0)
1305 		goto out_nfserr;
1306 
1307 	if (EX_ISSYNC(fhp->fh_export)) {
1308 		err = nfserrno(nfsd_sync_dir(dentry));
1309 		/* setattr will sync the child (or not) */
1310 	}
1311 
1312 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1313 		/* Cram the verifier into atime/mtime/mode */
1314 		iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1315 			| ATTR_MTIME_SET|ATTR_ATIME_SET
1316 			| ATTR_MODE;
1317 		/* XXX someone who knows this better please fix it for nsec */
1318 		iap->ia_mtime.tv_sec = v_mtime;
1319 		iap->ia_atime.tv_sec = v_atime;
1320 		iap->ia_mtime.tv_nsec = 0;
1321 		iap->ia_atime.tv_nsec = 0;
1322 		iap->ia_mode  = v_mode;
1323 	}
1324 
1325 	/* Set file attributes.
1326 	 * Mode has already been set but we might need to reset it
1327 	 * for CREATE_EXCLUSIVE
1328 	 * Irix appears to send along the gid when it tries to
1329 	 * implement setgid directories via NFS. Clear out all that cruft.
1330 	 */
1331  set_attr:
1332 	if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) {
1333  		int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1334 		if (err2)
1335 			err = err2;
1336 	}
1337 
1338 	/*
1339 	 * Update the filehandle to get the new inode info.
1340 	 */
1341 	if (!err)
1342 		err = fh_update(resfhp);
1343 
1344  out:
1345 	fh_unlock(fhp);
1346 	if (dchild && !IS_ERR(dchild))
1347 		dput(dchild);
1348  	return err;
1349 
1350  out_nfserr:
1351 	err = nfserrno(err);
1352 	goto out;
1353 }
1354 #endif /* CONFIG_NFSD_V3 */
1355 
1356 /*
1357  * Read a symlink. On entry, *lenp must contain the maximum path length that
1358  * fits into the buffer. On return, it contains the true length.
1359  * N.B. After this call fhp needs an fh_put
1360  */
1361 int
1362 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1363 {
1364 	struct dentry	*dentry;
1365 	struct inode	*inode;
1366 	mm_segment_t	oldfs;
1367 	int		err;
1368 
1369 	err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1370 	if (err)
1371 		goto out;
1372 
1373 	dentry = fhp->fh_dentry;
1374 	inode = dentry->d_inode;
1375 
1376 	err = nfserr_inval;
1377 	if (!inode->i_op || !inode->i_op->readlink)
1378 		goto out;
1379 
1380 	touch_atime(fhp->fh_export->ex_mnt, dentry);
1381 	/* N.B. Why does this call need a get_fs()??
1382 	 * Remove the set_fs and watch the fireworks:-) --okir
1383 	 */
1384 
1385 	oldfs = get_fs(); set_fs(KERNEL_DS);
1386 	err = inode->i_op->readlink(dentry, buf, *lenp);
1387 	set_fs(oldfs);
1388 
1389 	if (err < 0)
1390 		goto out_nfserr;
1391 	*lenp = err;
1392 	err = 0;
1393 out:
1394 	return err;
1395 
1396 out_nfserr:
1397 	err = nfserrno(err);
1398 	goto out;
1399 }
1400 
1401 /*
1402  * Create a symlink and look up its inode
1403  * N.B. After this call _both_ fhp and resfhp need an fh_put
1404  */
1405 int
1406 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1407 				char *fname, int flen,
1408 				char *path,  int plen,
1409 				struct svc_fh *resfhp,
1410 				struct iattr *iap)
1411 {
1412 	struct dentry	*dentry, *dnew;
1413 	int		err, cerr;
1414 	umode_t		mode;
1415 
1416 	err = nfserr_noent;
1417 	if (!flen || !plen)
1418 		goto out;
1419 	err = nfserr_exist;
1420 	if (isdotent(fname, flen))
1421 		goto out;
1422 
1423 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1424 	if (err)
1425 		goto out;
1426 	fh_lock(fhp);
1427 	dentry = fhp->fh_dentry;
1428 	dnew = lookup_one_len(fname, dentry, flen);
1429 	err = PTR_ERR(dnew);
1430 	if (IS_ERR(dnew))
1431 		goto out_nfserr;
1432 
1433 	mode = S_IALLUGO;
1434 	/* Only the MODE ATTRibute is even vaguely meaningful */
1435 	if (iap && (iap->ia_valid & ATTR_MODE))
1436 		mode = iap->ia_mode & S_IALLUGO;
1437 
1438 	if (unlikely(path[plen] != 0)) {
1439 		char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1440 		if (path_alloced == NULL)
1441 			err = -ENOMEM;
1442 		else {
1443 			strncpy(path_alloced, path, plen);
1444 			path_alloced[plen] = 0;
1445 			err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1446 			kfree(path_alloced);
1447 		}
1448 	} else
1449 		err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1450 
1451 	if (!err)
1452 		if (EX_ISSYNC(fhp->fh_export))
1453 			err = nfsd_sync_dir(dentry);
1454 	if (err)
1455 		err = nfserrno(err);
1456 	fh_unlock(fhp);
1457 
1458 	cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1459 	dput(dnew);
1460 	if (err==0) err = cerr;
1461 out:
1462 	return err;
1463 
1464 out_nfserr:
1465 	err = nfserrno(err);
1466 	goto out;
1467 }
1468 
1469 /*
1470  * Create a hardlink
1471  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1472  */
1473 int
1474 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1475 				char *name, int len, struct svc_fh *tfhp)
1476 {
1477 	struct dentry	*ddir, *dnew, *dold;
1478 	struct inode	*dirp, *dest;
1479 	int		err;
1480 
1481 	err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1482 	if (err)
1483 		goto out;
1484 	err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1485 	if (err)
1486 		goto out;
1487 
1488 	err = nfserr_perm;
1489 	if (!len)
1490 		goto out;
1491 	err = nfserr_exist;
1492 	if (isdotent(name, len))
1493 		goto out;
1494 
1495 	fh_lock(ffhp);
1496 	ddir = ffhp->fh_dentry;
1497 	dirp = ddir->d_inode;
1498 
1499 	dnew = lookup_one_len(name, ddir, len);
1500 	err = PTR_ERR(dnew);
1501 	if (IS_ERR(dnew))
1502 		goto out_nfserr;
1503 
1504 	dold = tfhp->fh_dentry;
1505 	dest = dold->d_inode;
1506 
1507 	err = vfs_link(dold, dirp, dnew);
1508 	if (!err) {
1509 		if (EX_ISSYNC(ffhp->fh_export)) {
1510 			err = nfserrno(nfsd_sync_dir(ddir));
1511 			write_inode_now(dest, 1);
1512 		}
1513 	} else {
1514 		if (err == -EXDEV && rqstp->rq_vers == 2)
1515 			err = nfserr_acces;
1516 		else
1517 			err = nfserrno(err);
1518 	}
1519 
1520 	fh_unlock(ffhp);
1521 	dput(dnew);
1522 out:
1523 	return err;
1524 
1525 out_nfserr:
1526 	err = nfserrno(err);
1527 	goto out;
1528 }
1529 
1530 /*
1531  * Rename a file
1532  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1533  */
1534 int
1535 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1536 			    struct svc_fh *tfhp, char *tname, int tlen)
1537 {
1538 	struct dentry	*fdentry, *tdentry, *odentry, *ndentry, *trap;
1539 	struct inode	*fdir, *tdir;
1540 	int		err;
1541 
1542 	err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1543 	if (err)
1544 		goto out;
1545 	err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1546 	if (err)
1547 		goto out;
1548 
1549 	fdentry = ffhp->fh_dentry;
1550 	fdir = fdentry->d_inode;
1551 
1552 	tdentry = tfhp->fh_dentry;
1553 	tdir = tdentry->d_inode;
1554 
1555 	err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1556 	if (fdir->i_sb != tdir->i_sb)
1557 		goto out;
1558 
1559 	err = nfserr_perm;
1560 	if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1561 		goto out;
1562 
1563 	/* cannot use fh_lock as we need deadlock protective ordering
1564 	 * so do it by hand */
1565 	trap = lock_rename(tdentry, fdentry);
1566 	ffhp->fh_locked = tfhp->fh_locked = 1;
1567 	fill_pre_wcc(ffhp);
1568 	fill_pre_wcc(tfhp);
1569 
1570 	odentry = lookup_one_len(fname, fdentry, flen);
1571 	err = PTR_ERR(odentry);
1572 	if (IS_ERR(odentry))
1573 		goto out_nfserr;
1574 
1575 	err = -ENOENT;
1576 	if (!odentry->d_inode)
1577 		goto out_dput_old;
1578 	err = -EINVAL;
1579 	if (odentry == trap)
1580 		goto out_dput_old;
1581 
1582 	ndentry = lookup_one_len(tname, tdentry, tlen);
1583 	err = PTR_ERR(ndentry);
1584 	if (IS_ERR(ndentry))
1585 		goto out_dput_old;
1586 	err = -ENOTEMPTY;
1587 	if (ndentry == trap)
1588 		goto out_dput_new;
1589 
1590 #ifdef MSNFS
1591 	if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1592 		((atomic_read(&odentry->d_count) > 1)
1593 		 || (atomic_read(&ndentry->d_count) > 1))) {
1594 			err = -EPERM;
1595 	} else
1596 #endif
1597 	err = vfs_rename(fdir, odentry, tdir, ndentry);
1598 	if (!err && EX_ISSYNC(tfhp->fh_export)) {
1599 		err = nfsd_sync_dir(tdentry);
1600 		if (!err)
1601 			err = nfsd_sync_dir(fdentry);
1602 	}
1603 
1604  out_dput_new:
1605 	dput(ndentry);
1606  out_dput_old:
1607 	dput(odentry);
1608  out_nfserr:
1609 	if (err)
1610 		err = nfserrno(err);
1611 
1612 	/* we cannot reply on fh_unlock on the two filehandles,
1613 	 * as that would do the wrong thing if the two directories
1614 	 * were the same, so again we do it by hand
1615 	 */
1616 	fill_post_wcc(ffhp);
1617 	fill_post_wcc(tfhp);
1618 	unlock_rename(tdentry, fdentry);
1619 	ffhp->fh_locked = tfhp->fh_locked = 0;
1620 
1621 out:
1622 	return err;
1623 }
1624 
1625 /*
1626  * Unlink a file or directory
1627  * N.B. After this call fhp needs an fh_put
1628  */
1629 int
1630 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1631 				char *fname, int flen)
1632 {
1633 	struct dentry	*dentry, *rdentry;
1634 	struct inode	*dirp;
1635 	int		err;
1636 
1637 	err = nfserr_acces;
1638 	if (!flen || isdotent(fname, flen))
1639 		goto out;
1640 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1641 	if (err)
1642 		goto out;
1643 
1644 	fh_lock(fhp);
1645 	dentry = fhp->fh_dentry;
1646 	dirp = dentry->d_inode;
1647 
1648 	rdentry = lookup_one_len(fname, dentry, flen);
1649 	err = PTR_ERR(rdentry);
1650 	if (IS_ERR(rdentry))
1651 		goto out_nfserr;
1652 
1653 	if (!rdentry->d_inode) {
1654 		dput(rdentry);
1655 		err = nfserr_noent;
1656 		goto out;
1657 	}
1658 
1659 	if (!type)
1660 		type = rdentry->d_inode->i_mode & S_IFMT;
1661 
1662 	if (type != S_IFDIR) { /* It's UNLINK */
1663 #ifdef MSNFS
1664 		if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1665 			(atomic_read(&rdentry->d_count) > 1)) {
1666 			err = -EPERM;
1667 		} else
1668 #endif
1669 		err = vfs_unlink(dirp, rdentry);
1670 	} else { /* It's RMDIR */
1671 		err = vfs_rmdir(dirp, rdentry);
1672 	}
1673 
1674 	dput(rdentry);
1675 
1676 	if (err == 0 &&
1677 	    EX_ISSYNC(fhp->fh_export))
1678 			err = nfsd_sync_dir(dentry);
1679 
1680 out_nfserr:
1681 	err = nfserrno(err);
1682 out:
1683 	return err;
1684 }
1685 
1686 /*
1687  * Read entries from a directory.
1688  * The  NFSv3/4 verifier we ignore for now.
1689  */
1690 int
1691 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1692 	     struct readdir_cd *cdp, encode_dent_fn func)
1693 {
1694 	int		err;
1695 	struct file	*file;
1696 	loff_t		offset = *offsetp;
1697 
1698 	err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1699 	if (err)
1700 		goto out;
1701 
1702 	offset = vfs_llseek(file, offset, 0);
1703 	if (offset < 0) {
1704 		err = nfserrno((int)offset);
1705 		goto out_close;
1706 	}
1707 
1708 	/*
1709 	 * Read the directory entries. This silly loop is necessary because
1710 	 * readdir() is not guaranteed to fill up the entire buffer, but
1711 	 * may choose to do less.
1712 	 */
1713 
1714 	do {
1715 		cdp->err = nfserr_eof; /* will be cleared on successful read */
1716 		err = vfs_readdir(file, (filldir_t) func, cdp);
1717 	} while (err >=0 && cdp->err == nfs_ok);
1718 	if (err)
1719 		err = nfserrno(err);
1720 	else
1721 		err = cdp->err;
1722 	*offsetp = vfs_llseek(file, 0, 1);
1723 
1724 	if (err == nfserr_eof || err == nfserr_toosmall)
1725 		err = nfs_ok; /* can still be found in ->err */
1726 out_close:
1727 	nfsd_close(file);
1728 out:
1729 	return err;
1730 }
1731 
1732 /*
1733  * Get file system stats
1734  * N.B. After this call fhp needs an fh_put
1735  */
1736 int
1737 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1738 {
1739 	int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1740 	if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1741 		err = nfserr_io;
1742 	return err;
1743 }
1744 
1745 /*
1746  * Check for a user's access permissions to this inode.
1747  */
1748 int
1749 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1750 {
1751 	struct inode	*inode = dentry->d_inode;
1752 	int		err;
1753 
1754 	if (acc == MAY_NOP)
1755 		return 0;
1756 #if 0
1757 	dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1758 		acc,
1759 		(acc & MAY_READ)?	" read"  : "",
1760 		(acc & MAY_WRITE)?	" write" : "",
1761 		(acc & MAY_EXEC)?	" exec"  : "",
1762 		(acc & MAY_SATTR)?	" sattr" : "",
1763 		(acc & MAY_TRUNC)?	" trunc" : "",
1764 		(acc & MAY_LOCK)?	" lock"  : "",
1765 		(acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1766 		inode->i_mode,
1767 		IS_IMMUTABLE(inode)?	" immut" : "",
1768 		IS_APPEND(inode)?	" append" : "",
1769 		IS_RDONLY(inode)?	" ro" : "");
1770 	dprintk("      owner %d/%d user %d/%d\n",
1771 		inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1772 #endif
1773 
1774 	/* Normally we reject any write/sattr etc access on a read-only file
1775 	 * system.  But if it is IRIX doing check on write-access for a
1776 	 * device special file, we ignore rofs.
1777 	 */
1778 	if (!(acc & MAY_LOCAL_ACCESS))
1779 		if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1780 			if (EX_RDONLY(exp) || IS_RDONLY(inode))
1781 				return nfserr_rofs;
1782 			if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1783 				return nfserr_perm;
1784 		}
1785 	if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1786 		return nfserr_perm;
1787 
1788 	if (acc & MAY_LOCK) {
1789 		/* If we cannot rely on authentication in NLM requests,
1790 		 * just allow locks, otherwise require read permission, or
1791 		 * ownership
1792 		 */
1793 		if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1794 			return 0;
1795 		else
1796 			acc = MAY_READ | MAY_OWNER_OVERRIDE;
1797 	}
1798 	/*
1799 	 * The file owner always gets access permission for accesses that
1800 	 * would normally be checked at open time. This is to make
1801 	 * file access work even when the client has done a fchmod(fd, 0).
1802 	 *
1803 	 * However, `cp foo bar' should fail nevertheless when bar is
1804 	 * readonly. A sensible way to do this might be to reject all
1805 	 * attempts to truncate a read-only file, because a creat() call
1806 	 * always implies file truncation.
1807 	 * ... but this isn't really fair.  A process may reasonably call
1808 	 * ftruncate on an open file descriptor on a file with perm 000.
1809 	 * We must trust the client to do permission checking - using "ACCESS"
1810 	 * with NFSv3.
1811 	 */
1812 	if ((acc & MAY_OWNER_OVERRIDE) &&
1813 	    inode->i_uid == current->fsuid)
1814 		return 0;
1815 
1816 	err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1817 
1818 	/* Allow read access to binaries even when mode 111 */
1819 	if (err == -EACCES && S_ISREG(inode->i_mode) &&
1820 	    acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1821 		err = permission(inode, MAY_EXEC, NULL);
1822 
1823 	return err? nfserrno(err) : 0;
1824 }
1825 
1826 void
1827 nfsd_racache_shutdown(void)
1828 {
1829 	if (!raparm_cache)
1830 		return;
1831 	dprintk("nfsd: freeing readahead buffers.\n");
1832 	kfree(raparml);
1833 	raparm_cache = raparml = NULL;
1834 }
1835 /*
1836  * Initialize readahead param cache
1837  */
1838 int
1839 nfsd_racache_init(int cache_size)
1840 {
1841 	int	i;
1842 
1843 	if (raparm_cache)
1844 		return 0;
1845 	raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1846 
1847 	if (raparml != NULL) {
1848 		dprintk("nfsd: allocating %d readahead buffers.\n",
1849 			cache_size);
1850 		memset(raparml, 0, sizeof(struct raparms) * cache_size);
1851 		for (i = 0; i < cache_size - 1; i++) {
1852 			raparml[i].p_next = raparml + i + 1;
1853 		}
1854 		raparm_cache = raparml;
1855 	} else {
1856 		printk(KERN_WARNING
1857 		       "nfsd: Could not allocate memory read-ahead cache.\n");
1858 		return -ENOMEM;
1859 	}
1860 	nfsdstats.ra_size = cache_size;
1861 	return 0;
1862 }
1863 
1864 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1865 struct posix_acl *
1866 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1867 {
1868 	struct inode *inode = fhp->fh_dentry->d_inode;
1869 	char *name;
1870 	void *value = NULL;
1871 	ssize_t size;
1872 	struct posix_acl *acl;
1873 
1874 	if (!IS_POSIXACL(inode))
1875 		return ERR_PTR(-EOPNOTSUPP);
1876 
1877 	switch (type) {
1878 	case ACL_TYPE_ACCESS:
1879 		name = POSIX_ACL_XATTR_ACCESS;
1880 		break;
1881 	case ACL_TYPE_DEFAULT:
1882 		name = POSIX_ACL_XATTR_DEFAULT;
1883 		break;
1884 	default:
1885 		return ERR_PTR(-EOPNOTSUPP);
1886 	}
1887 
1888 	size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1889 	if (size < 0)
1890 		return ERR_PTR(size);
1891 
1892 	acl = posix_acl_from_xattr(value, size);
1893 	kfree(value);
1894 	return acl;
1895 }
1896 
1897 int
1898 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1899 {
1900 	struct inode *inode = fhp->fh_dentry->d_inode;
1901 	char *name;
1902 	void *value = NULL;
1903 	size_t size;
1904 	int error;
1905 
1906 	if (!IS_POSIXACL(inode) || !inode->i_op ||
1907 	    !inode->i_op->setxattr || !inode->i_op->removexattr)
1908 		return -EOPNOTSUPP;
1909 	switch(type) {
1910 		case ACL_TYPE_ACCESS:
1911 			name = POSIX_ACL_XATTR_ACCESS;
1912 			break;
1913 		case ACL_TYPE_DEFAULT:
1914 			name = POSIX_ACL_XATTR_DEFAULT;
1915 			break;
1916 		default:
1917 			return -EOPNOTSUPP;
1918 	}
1919 
1920 	if (acl && acl->a_count) {
1921 		size = posix_acl_xattr_size(acl->a_count);
1922 		value = kmalloc(size, GFP_KERNEL);
1923 		if (!value)
1924 			return -ENOMEM;
1925 		error = posix_acl_to_xattr(acl, value, size);
1926 		if (error < 0)
1927 			goto getout;
1928 		size = error;
1929 	} else
1930 		size = 0;
1931 
1932 	if (size)
1933 		error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1934 	else {
1935 		if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1936 			error = 0;
1937 		else {
1938 			error = vfs_removexattr(fhp->fh_dentry, name);
1939 			if (error == -ENODATA)
1940 				error = 0;
1941 		}
1942 	}
1943 
1944 getout:
1945 	kfree(value);
1946 	return error;
1947 }
1948 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
1949