xref: /linux/fs/nfsd/vfs.c (revision 9ce7677cfd7cd871adb457c80bea3b581b839641)
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 #ifdef CONFIG_NFSD_V4
52 #include <linux/xattr.h>
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_V4)
369 
370 static int
371 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
372 {
373 	int len;
374 	size_t buflen;
375 	char *buf = NULL;
376 	int error = 0;
377 	struct inode *inode = dentry->d_inode;
378 
379 	buflen = posix_acl_xattr_size(pacl->a_count);
380 	buf = kmalloc(buflen, GFP_KERNEL);
381 	error = -ENOMEM;
382 	if (buf == NULL)
383 		goto out;
384 
385 	len = posix_acl_to_xattr(pacl, buf, buflen);
386 	if (len < 0) {
387 		error = len;
388 		goto out;
389 	}
390 
391 	error = -EOPNOTSUPP;
392 	if (inode->i_op && inode->i_op->setxattr) {
393 		down(&inode->i_sem);
394 		security_inode_setxattr(dentry, key, buf, len, 0);
395 		error = inode->i_op->setxattr(dentry, key, buf, len, 0);
396 		if (!error)
397 			security_inode_post_setxattr(dentry, key, buf, len, 0);
398 		up(&inode->i_sem);
399 	}
400 out:
401 	kfree(buf);
402 	return error;
403 }
404 
405 int
406 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
407     struct nfs4_acl *acl)
408 {
409 	int error;
410 	struct dentry *dentry;
411 	struct inode *inode;
412 	struct posix_acl *pacl = NULL, *dpacl = NULL;
413 	unsigned int flags = 0;
414 
415 	/* Get inode */
416 	error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
417 	if (error)
418 		goto out;
419 
420 	dentry = fhp->fh_dentry;
421 	inode = dentry->d_inode;
422 	if (S_ISDIR(inode->i_mode))
423 		flags = NFS4_ACL_DIR;
424 
425 	error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
426 	if (error == -EINVAL) {
427 		error = nfserr_attrnotsupp;
428 		goto out;
429 	} else if (error < 0)
430 		goto out_nfserr;
431 
432 	if (pacl) {
433 		error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
434 		if (error < 0)
435 			goto out_nfserr;
436 	}
437 
438 	if (dpacl) {
439 		error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
440 		if (error < 0)
441 			goto out_nfserr;
442 	}
443 
444 	error = nfs_ok;
445 
446 out:
447 	posix_acl_release(pacl);
448 	posix_acl_release(dpacl);
449 	return (error);
450 out_nfserr:
451 	error = nfserrno(error);
452 	goto out;
453 }
454 
455 static struct posix_acl *
456 _get_posix_acl(struct dentry *dentry, char *key)
457 {
458 	struct inode *inode = dentry->d_inode;
459 	char *buf = NULL;
460 	int buflen, error = 0;
461 	struct posix_acl *pacl = NULL;
462 
463 	error = -EOPNOTSUPP;
464 	if (inode->i_op == NULL)
465 		goto out_err;
466 	if (inode->i_op->getxattr == NULL)
467 		goto out_err;
468 
469 	error = security_inode_getxattr(dentry, key);
470 	if (error)
471 		goto out_err;
472 
473 	buflen = inode->i_op->getxattr(dentry, key, NULL, 0);
474 	if (buflen <= 0) {
475 		error = buflen < 0 ? buflen : -ENODATA;
476 		goto out_err;
477 	}
478 
479 	buf = kmalloc(buflen, GFP_KERNEL);
480 	if (buf == NULL) {
481 		error = -ENOMEM;
482 		goto out_err;
483 	}
484 
485 	error = inode->i_op->getxattr(dentry, key, buf, buflen);
486 	if (error < 0)
487 		goto out_err;
488 
489 	pacl = posix_acl_from_xattr(buf, buflen);
490  out:
491 	kfree(buf);
492 	return pacl;
493  out_err:
494 	pacl = ERR_PTR(error);
495 	goto out;
496 }
497 
498 int
499 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
500 {
501 	struct inode *inode = dentry->d_inode;
502 	int error = 0;
503 	struct posix_acl *pacl = NULL, *dpacl = NULL;
504 	unsigned int flags = 0;
505 
506 	pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
507 	if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
508 		pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
509 	if (IS_ERR(pacl)) {
510 		error = PTR_ERR(pacl);
511 		pacl = NULL;
512 		goto out;
513 	}
514 
515 	if (S_ISDIR(inode->i_mode)) {
516 		dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
517 		if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
518 			dpacl = NULL;
519 		else if (IS_ERR(dpacl)) {
520 			error = PTR_ERR(dpacl);
521 			dpacl = NULL;
522 			goto out;
523 		}
524 		flags = NFS4_ACL_DIR;
525 	}
526 
527 	*acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
528 	if (IS_ERR(*acl)) {
529 		error = PTR_ERR(*acl);
530 		*acl = NULL;
531 	}
532  out:
533 	posix_acl_release(pacl);
534 	posix_acl_release(dpacl);
535 	return error;
536 }
537 
538 #endif /* defined(CONFIG_NFS_V4) */
539 
540 #ifdef CONFIG_NFSD_V3
541 /*
542  * Check server access rights to a file system object
543  */
544 struct accessmap {
545 	u32		access;
546 	int		how;
547 };
548 static struct accessmap	nfs3_regaccess[] = {
549     {	NFS3_ACCESS_READ,	MAY_READ			},
550     {	NFS3_ACCESS_EXECUTE,	MAY_EXEC			},
551     {	NFS3_ACCESS_MODIFY,	MAY_WRITE|MAY_TRUNC		},
552     {	NFS3_ACCESS_EXTEND,	MAY_WRITE			},
553 
554     {	0,			0				}
555 };
556 
557 static struct accessmap	nfs3_diraccess[] = {
558     {	NFS3_ACCESS_READ,	MAY_READ			},
559     {	NFS3_ACCESS_LOOKUP,	MAY_EXEC			},
560     {	NFS3_ACCESS_MODIFY,	MAY_EXEC|MAY_WRITE|MAY_TRUNC	},
561     {	NFS3_ACCESS_EXTEND,	MAY_EXEC|MAY_WRITE		},
562     {	NFS3_ACCESS_DELETE,	MAY_REMOVE			},
563 
564     {	0,			0				}
565 };
566 
567 static struct accessmap	nfs3_anyaccess[] = {
568 	/* Some clients - Solaris 2.6 at least, make an access call
569 	 * to the server to check for access for things like /dev/null
570 	 * (which really, the server doesn't care about).  So
571 	 * We provide simple access checking for them, looking
572 	 * mainly at mode bits, and we make sure to ignore read-only
573 	 * filesystem checks
574 	 */
575     {	NFS3_ACCESS_READ,	MAY_READ			},
576     {	NFS3_ACCESS_EXECUTE,	MAY_EXEC			},
577     {	NFS3_ACCESS_MODIFY,	MAY_WRITE|MAY_LOCAL_ACCESS	},
578     {	NFS3_ACCESS_EXTEND,	MAY_WRITE|MAY_LOCAL_ACCESS	},
579 
580     {	0,			0				}
581 };
582 
583 int
584 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
585 {
586 	struct accessmap	*map;
587 	struct svc_export	*export;
588 	struct dentry		*dentry;
589 	u32			query, result = 0, sresult = 0;
590 	unsigned int		error;
591 
592 	error = fh_verify(rqstp, fhp, 0, MAY_NOP);
593 	if (error)
594 		goto out;
595 
596 	export = fhp->fh_export;
597 	dentry = fhp->fh_dentry;
598 
599 	if (S_ISREG(dentry->d_inode->i_mode))
600 		map = nfs3_regaccess;
601 	else if (S_ISDIR(dentry->d_inode->i_mode))
602 		map = nfs3_diraccess;
603 	else
604 		map = nfs3_anyaccess;
605 
606 
607 	query = *access;
608 	for  (; map->access; map++) {
609 		if (map->access & query) {
610 			unsigned int err2;
611 
612 			sresult |= map->access;
613 
614 			err2 = nfsd_permission(export, dentry, map->how);
615 			switch (err2) {
616 			case nfs_ok:
617 				result |= map->access;
618 				break;
619 
620 			/* the following error codes just mean the access was not allowed,
621 			 * rather than an error occurred */
622 			case nfserr_rofs:
623 			case nfserr_acces:
624 			case nfserr_perm:
625 				/* simply don't "or" in the access bit. */
626 				break;
627 			default:
628 				error = err2;
629 				goto out;
630 			}
631 		}
632 	}
633 	*access = result;
634 	if (supported)
635 		*supported = sresult;
636 
637  out:
638 	return error;
639 }
640 #endif /* CONFIG_NFSD_V3 */
641 
642 
643 
644 /*
645  * Open an existing file or directory.
646  * The access argument indicates the type of open (read/write/lock)
647  * N.B. After this call fhp needs an fh_put
648  */
649 int
650 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
651 			int access, struct file **filp)
652 {
653 	struct dentry	*dentry;
654 	struct inode	*inode;
655 	int		flags = O_RDONLY|O_LARGEFILE, err;
656 
657 	/*
658 	 * If we get here, then the client has already done an "open",
659 	 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
660 	 * in case a chmod has now revoked permission.
661 	 */
662 	err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
663 	if (err)
664 		goto out;
665 
666 	dentry = fhp->fh_dentry;
667 	inode = dentry->d_inode;
668 
669 	/* Disallow write access to files with the append-only bit set
670 	 * or any access when mandatory locking enabled
671 	 */
672 	err = nfserr_perm;
673 	if (IS_APPEND(inode) && (access & MAY_WRITE))
674 		goto out;
675 	if (IS_ISMNDLK(inode))
676 		goto out;
677 
678 	if (!inode->i_fop)
679 		goto out;
680 
681 	/*
682 	 * Check to see if there are any leases on this file.
683 	 * This may block while leases are broken.
684 	 */
685 	err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
686 	if (err == -EWOULDBLOCK)
687 		err = -ETIMEDOUT;
688 	if (err) /* NOMEM or WOULDBLOCK */
689 		goto out_nfserr;
690 
691 	if (access & MAY_WRITE) {
692 		flags = O_WRONLY|O_LARGEFILE;
693 
694 		DQUOT_INIT(inode);
695 	}
696 	*filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
697 	if (IS_ERR(*filp))
698 		err = PTR_ERR(*filp);
699 out_nfserr:
700 	if (err)
701 		err = nfserrno(err);
702 out:
703 	return err;
704 }
705 
706 /*
707  * Close a file.
708  */
709 void
710 nfsd_close(struct file *filp)
711 {
712 	fput(filp);
713 }
714 
715 /*
716  * Sync a file
717  * As this calls fsync (not fdatasync) there is no need for a write_inode
718  * after it.
719  */
720 static inline void nfsd_dosync(struct file *filp, struct dentry *dp,
721 			       struct file_operations *fop)
722 {
723 	struct inode *inode = dp->d_inode;
724 	int (*fsync) (struct file *, struct dentry *, int);
725 
726 	filemap_fdatawrite(inode->i_mapping);
727 	if (fop && (fsync = fop->fsync))
728 		fsync(filp, dp, 0);
729 	filemap_fdatawait(inode->i_mapping);
730 }
731 
732 
733 static void
734 nfsd_sync(struct file *filp)
735 {
736 	struct inode *inode = filp->f_dentry->d_inode;
737 	dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
738 	down(&inode->i_sem);
739 	nfsd_dosync(filp, filp->f_dentry, filp->f_op);
740 	up(&inode->i_sem);
741 }
742 
743 void
744 nfsd_sync_dir(struct dentry *dp)
745 {
746 	nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
747 }
748 
749 /*
750  * Obtain the readahead parameters for the file
751  * specified by (dev, ino).
752  */
753 static DEFINE_SPINLOCK(ra_lock);
754 
755 static inline struct raparms *
756 nfsd_get_raparms(dev_t dev, ino_t ino)
757 {
758 	struct raparms	*ra, **rap, **frap = NULL;
759 	int depth = 0;
760 
761 	spin_lock(&ra_lock);
762 	for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
763 		if (ra->p_ino == ino && ra->p_dev == dev)
764 			goto found;
765 		depth++;
766 		if (ra->p_count == 0)
767 			frap = rap;
768 	}
769 	depth = nfsdstats.ra_size*11/10;
770 	if (!frap) {
771 		spin_unlock(&ra_lock);
772 		return NULL;
773 	}
774 	rap = frap;
775 	ra = *frap;
776 	ra->p_dev = dev;
777 	ra->p_ino = ino;
778 	ra->p_set = 0;
779 found:
780 	if (rap != &raparm_cache) {
781 		*rap = ra->p_next;
782 		ra->p_next   = raparm_cache;
783 		raparm_cache = ra;
784 	}
785 	ra->p_count++;
786 	nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
787 	spin_unlock(&ra_lock);
788 	return ra;
789 }
790 
791 /*
792  * Grab and keep cached pages assosiated with a file in the svc_rqst
793  * so that they can be passed to the netowork sendmsg/sendpage routines
794  * directrly. They will be released after the sending has completed.
795  */
796 static int
797 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
798 {
799 	unsigned long count = desc->count;
800 	struct svc_rqst *rqstp = desc->arg.data;
801 
802 	if (size > count)
803 		size = count;
804 
805 	if (rqstp->rq_res.page_len == 0) {
806 		get_page(page);
807 		rqstp->rq_respages[rqstp->rq_resused++] = page;
808 		rqstp->rq_res.page_base = offset;
809 		rqstp->rq_res.page_len = size;
810 	} else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
811 		get_page(page);
812 		rqstp->rq_respages[rqstp->rq_resused++] = page;
813 		rqstp->rq_res.page_len += size;
814 	} else {
815 		rqstp->rq_res.page_len += size;
816 	}
817 
818 	desc->count = count - size;
819 	desc->written += size;
820 	return size;
821 }
822 
823 static inline int
824 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
825               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
826 {
827 	struct inode *inode;
828 	struct raparms	*ra;
829 	mm_segment_t	oldfs;
830 	int		err;
831 
832 	err = nfserr_perm;
833 	inode = file->f_dentry->d_inode;
834 #ifdef MSNFS
835 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
836 		(!lock_may_read(inode, offset, *count)))
837 		goto out;
838 #endif
839 
840 	/* Get readahead parameters */
841 	ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
842 
843 	if (ra && ra->p_set)
844 		file->f_ra = ra->p_ra;
845 
846 	if (file->f_op->sendfile) {
847 		svc_pushback_unused_pages(rqstp);
848 		err = file->f_op->sendfile(file, &offset, *count,
849 						 nfsd_read_actor, rqstp);
850 	} else {
851 		oldfs = get_fs();
852 		set_fs(KERNEL_DS);
853 		err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
854 		set_fs(oldfs);
855 	}
856 
857 	/* Write back readahead params */
858 	if (ra) {
859 		spin_lock(&ra_lock);
860 		ra->p_ra = file->f_ra;
861 		ra->p_set = 1;
862 		ra->p_count--;
863 		spin_unlock(&ra_lock);
864 	}
865 
866 	if (err >= 0) {
867 		nfsdstats.io_read += err;
868 		*count = err;
869 		err = 0;
870 		fsnotify_access(file->f_dentry);
871 	} else
872 		err = nfserrno(err);
873 out:
874 	return err;
875 }
876 
877 static inline int
878 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
879 				loff_t offset, struct kvec *vec, int vlen,
880 	   			unsigned long cnt, int *stablep)
881 {
882 	struct svc_export	*exp;
883 	struct dentry		*dentry;
884 	struct inode		*inode;
885 	mm_segment_t		oldfs;
886 	int			err = 0;
887 	int			stable = *stablep;
888 
889 	err = nfserr_perm;
890 
891 #ifdef MSNFS
892 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
893 		(!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
894 		goto out;
895 #endif
896 
897 	dentry = file->f_dentry;
898 	inode = dentry->d_inode;
899 	exp   = fhp->fh_export;
900 
901 	/*
902 	 * Request sync writes if
903 	 *  -	the sync export option has been set, or
904 	 *  -	the client requested O_SYNC behavior (NFSv3 feature).
905 	 *  -   The file system doesn't support fsync().
906 	 * When gathered writes have been configured for this volume,
907 	 * flushing the data to disk is handled separately below.
908 	 */
909 
910 	if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
911 	       stable = 2;
912 	       *stablep = 2; /* FILE_SYNC */
913 	}
914 
915 	if (!EX_ISSYNC(exp))
916 		stable = 0;
917 	if (stable && !EX_WGATHER(exp))
918 		file->f_flags |= O_SYNC;
919 
920 	/* Write the data. */
921 	oldfs = get_fs(); set_fs(KERNEL_DS);
922 	err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
923 	set_fs(oldfs);
924 	if (err >= 0) {
925 		nfsdstats.io_write += cnt;
926 		fsnotify_modify(file->f_dentry);
927 	}
928 
929 	/* clear setuid/setgid flag after write */
930 	if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) {
931 		struct iattr	ia;
932 		ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
933 
934 		down(&inode->i_sem);
935 		notify_change(dentry, &ia);
936 		up(&inode->i_sem);
937 	}
938 
939 	if (err >= 0 && stable) {
940 		static ino_t	last_ino;
941 		static dev_t	last_dev;
942 
943 		/*
944 		 * Gathered writes: If another process is currently
945 		 * writing to the file, there's a high chance
946 		 * this is another nfsd (triggered by a bulk write
947 		 * from a client's biod). Rather than syncing the
948 		 * file with each write request, we sleep for 10 msec.
949 		 *
950 		 * I don't know if this roughly approximates
951 		 * C. Juszak's idea of gathered writes, but it's a
952 		 * nice and simple solution (IMHO), and it seems to
953 		 * work:-)
954 		 */
955 		if (EX_WGATHER(exp)) {
956 			if (atomic_read(&inode->i_writecount) > 1
957 			    || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
958 				dprintk("nfsd: write defer %d\n", current->pid);
959 				msleep(10);
960 				dprintk("nfsd: write resume %d\n", current->pid);
961 			}
962 
963 			if (inode->i_state & I_DIRTY) {
964 				dprintk("nfsd: write sync %d\n", current->pid);
965 				nfsd_sync(file);
966 			}
967 #if 0
968 			wake_up(&inode->i_wait);
969 #endif
970 		}
971 		last_ino = inode->i_ino;
972 		last_dev = inode->i_sb->s_dev;
973 	}
974 
975 	dprintk("nfsd: write complete err=%d\n", err);
976 	if (err >= 0)
977 		err = 0;
978 	else
979 		err = nfserrno(err);
980 out:
981 	return err;
982 }
983 
984 /*
985  * Read data from a file. count must contain the requested read count
986  * on entry. On return, *count contains the number of bytes actually read.
987  * N.B. After this call fhp needs an fh_put
988  */
989 int
990 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
991 		loff_t offset, struct kvec *vec, int vlen,
992 		unsigned long *count)
993 {
994 	int		err;
995 
996 	if (file) {
997 		err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
998 				MAY_READ|MAY_OWNER_OVERRIDE);
999 		if (err)
1000 			goto out;
1001 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1002 	} else {
1003 		err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1004 		if (err)
1005 			goto out;
1006 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1007 		nfsd_close(file);
1008 	}
1009 out:
1010 	return err;
1011 }
1012 
1013 /*
1014  * Write data to a file.
1015  * The stable flag requests synchronous writes.
1016  * N.B. After this call fhp needs an fh_put
1017  */
1018 int
1019 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1020 		loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1021 		int *stablep)
1022 {
1023 	int			err = 0;
1024 
1025 	if (file) {
1026 		err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1027 				MAY_WRITE|MAY_OWNER_OVERRIDE);
1028 		if (err)
1029 			goto out;
1030 		err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1031 				stablep);
1032 	} else {
1033 		err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1034 		if (err)
1035 			goto out;
1036 
1037 		if (cnt)
1038 			err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1039 					     cnt, stablep);
1040 		nfsd_close(file);
1041 	}
1042 out:
1043 	return err;
1044 }
1045 
1046 #ifdef CONFIG_NFSD_V3
1047 /*
1048  * Commit all pending writes to stable storage.
1049  * Strictly speaking, we could sync just the indicated file region here,
1050  * but there's currently no way we can ask the VFS to do so.
1051  *
1052  * Unfortunately we cannot lock the file to make sure we return full WCC
1053  * data to the client, as locking happens lower down in the filesystem.
1054  */
1055 int
1056 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1057                loff_t offset, unsigned long count)
1058 {
1059 	struct file	*file;
1060 	int		err;
1061 
1062 	if ((u64)count > ~(u64)offset)
1063 		return nfserr_inval;
1064 
1065 	if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1066 		return err;
1067 	if (EX_ISSYNC(fhp->fh_export)) {
1068 		if (file->f_op && file->f_op->fsync) {
1069 			nfsd_sync(file);
1070 		} else {
1071 			err = nfserr_notsupp;
1072 		}
1073 	}
1074 
1075 	nfsd_close(file);
1076 	return err;
1077 }
1078 #endif /* CONFIG_NFSD_V3 */
1079 
1080 /*
1081  * Create a file (regular, directory, device, fifo); UNIX sockets
1082  * not yet implemented.
1083  * If the response fh has been verified, the parent directory should
1084  * already be locked. Note that the parent directory is left locked.
1085  *
1086  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1087  */
1088 int
1089 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1090 		char *fname, int flen, struct iattr *iap,
1091 		int type, dev_t rdev, struct svc_fh *resfhp)
1092 {
1093 	struct dentry	*dentry, *dchild = NULL;
1094 	struct inode	*dirp;
1095 	int		err;
1096 
1097 	err = nfserr_perm;
1098 	if (!flen)
1099 		goto out;
1100 	err = nfserr_exist;
1101 	if (isdotent(fname, flen))
1102 		goto out;
1103 
1104 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1105 	if (err)
1106 		goto out;
1107 
1108 	dentry = fhp->fh_dentry;
1109 	dirp = dentry->d_inode;
1110 
1111 	err = nfserr_notdir;
1112 	if(!dirp->i_op || !dirp->i_op->lookup)
1113 		goto out;
1114 	/*
1115 	 * Check whether the response file handle has been verified yet.
1116 	 * If it has, the parent directory should already be locked.
1117 	 */
1118 	if (!resfhp->fh_dentry) {
1119 		/* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1120 		fh_lock(fhp);
1121 		dchild = lookup_one_len(fname, dentry, flen);
1122 		err = PTR_ERR(dchild);
1123 		if (IS_ERR(dchild))
1124 			goto out_nfserr;
1125 		err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1126 		if (err)
1127 			goto out;
1128 	} else {
1129 		/* called from nfsd_proc_create */
1130 		dchild = dget(resfhp->fh_dentry);
1131 		if (!fhp->fh_locked) {
1132 			/* not actually possible */
1133 			printk(KERN_ERR
1134 				"nfsd_create: parent %s/%s not locked!\n",
1135 				dentry->d_parent->d_name.name,
1136 				dentry->d_name.name);
1137 			err = -EIO;
1138 			goto out;
1139 		}
1140 	}
1141 	/*
1142 	 * Make sure the child dentry is still negative ...
1143 	 */
1144 	err = nfserr_exist;
1145 	if (dchild->d_inode) {
1146 		dprintk("nfsd_create: dentry %s/%s not negative!\n",
1147 			dentry->d_name.name, dchild->d_name.name);
1148 		goto out;
1149 	}
1150 
1151 	if (!(iap->ia_valid & ATTR_MODE))
1152 		iap->ia_mode = 0;
1153 	iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1154 
1155 	/*
1156 	 * Get the dir op function pointer.
1157 	 */
1158 	err = nfserr_perm;
1159 	switch (type) {
1160 	case S_IFREG:
1161 		err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1162 		break;
1163 	case S_IFDIR:
1164 		err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1165 		break;
1166 	case S_IFCHR:
1167 	case S_IFBLK:
1168 	case S_IFIFO:
1169 	case S_IFSOCK:
1170 		err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1171 		break;
1172 	default:
1173 	        printk("nfsd: bad file type %o in nfsd_create\n", type);
1174 		err = -EINVAL;
1175 	}
1176 	if (err < 0)
1177 		goto out_nfserr;
1178 
1179 	if (EX_ISSYNC(fhp->fh_export)) {
1180 		nfsd_sync_dir(dentry);
1181 		write_inode_now(dchild->d_inode, 1);
1182 	}
1183 
1184 
1185 	/* Set file attributes. Mode has already been set and
1186 	 * setting uid/gid works only for root. Irix appears to
1187 	 * send along the gid when it tries to implement setgid
1188 	 * directories via NFS.
1189 	 */
1190 	err = 0;
1191 	if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
1192 		err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
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 		nfsd_sync_dir(dentry);
1312 		/* setattr will sync the child (or not) */
1313 	}
1314 
1315 	/*
1316 	 * Update the filehandle to get the new inode info.
1317 	 */
1318 	err = fh_update(resfhp);
1319 	if (err)
1320 		goto out;
1321 
1322 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1323 		/* Cram the verifier into atime/mtime/mode */
1324 		iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1325 			| ATTR_MTIME_SET|ATTR_ATIME_SET
1326 			| ATTR_MODE;
1327 		/* XXX someone who knows this better please fix it for nsec */
1328 		iap->ia_mtime.tv_sec = v_mtime;
1329 		iap->ia_atime.tv_sec = v_atime;
1330 		iap->ia_mtime.tv_nsec = 0;
1331 		iap->ia_atime.tv_nsec = 0;
1332 		iap->ia_mode  = v_mode;
1333 	}
1334 
1335 	/* Set file attributes.
1336 	 * Mode has already been set but we might need to reset it
1337 	 * for CREATE_EXCLUSIVE
1338 	 * Irix appears to send along the gid when it tries to
1339 	 * implement setgid directories via NFS. Clear out all that cruft.
1340 	 */
1341  set_attr:
1342 	if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1343  		err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1344 
1345  out:
1346 	fh_unlock(fhp);
1347 	if (dchild && !IS_ERR(dchild))
1348 		dput(dchild);
1349  	return err;
1350 
1351  out_nfserr:
1352 	err = nfserrno(err);
1353 	goto out;
1354 }
1355 #endif /* CONFIG_NFSD_V3 */
1356 
1357 /*
1358  * Read a symlink. On entry, *lenp must contain the maximum path length that
1359  * fits into the buffer. On return, it contains the true length.
1360  * N.B. After this call fhp needs an fh_put
1361  */
1362 int
1363 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1364 {
1365 	struct dentry	*dentry;
1366 	struct inode	*inode;
1367 	mm_segment_t	oldfs;
1368 	int		err;
1369 
1370 	err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1371 	if (err)
1372 		goto out;
1373 
1374 	dentry = fhp->fh_dentry;
1375 	inode = dentry->d_inode;
1376 
1377 	err = nfserr_inval;
1378 	if (!inode->i_op || !inode->i_op->readlink)
1379 		goto out;
1380 
1381 	touch_atime(fhp->fh_export->ex_mnt, dentry);
1382 	/* N.B. Why does this call need a get_fs()??
1383 	 * Remove the set_fs and watch the fireworks:-) --okir
1384 	 */
1385 
1386 	oldfs = get_fs(); set_fs(KERNEL_DS);
1387 	err = inode->i_op->readlink(dentry, buf, *lenp);
1388 	set_fs(oldfs);
1389 
1390 	if (err < 0)
1391 		goto out_nfserr;
1392 	*lenp = err;
1393 	err = 0;
1394 out:
1395 	return err;
1396 
1397 out_nfserr:
1398 	err = nfserrno(err);
1399 	goto out;
1400 }
1401 
1402 /*
1403  * Create a symlink and look up its inode
1404  * N.B. After this call _both_ fhp and resfhp need an fh_put
1405  */
1406 int
1407 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1408 				char *fname, int flen,
1409 				char *path,  int plen,
1410 				struct svc_fh *resfhp,
1411 				struct iattr *iap)
1412 {
1413 	struct dentry	*dentry, *dnew;
1414 	int		err, cerr;
1415 	umode_t		mode;
1416 
1417 	err = nfserr_noent;
1418 	if (!flen || !plen)
1419 		goto out;
1420 	err = nfserr_exist;
1421 	if (isdotent(fname, flen))
1422 		goto out;
1423 
1424 	err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1425 	if (err)
1426 		goto out;
1427 	fh_lock(fhp);
1428 	dentry = fhp->fh_dentry;
1429 	dnew = lookup_one_len(fname, dentry, flen);
1430 	err = PTR_ERR(dnew);
1431 	if (IS_ERR(dnew))
1432 		goto out_nfserr;
1433 
1434 	mode = S_IALLUGO;
1435 	/* Only the MODE ATTRibute is even vaguely meaningful */
1436 	if (iap && (iap->ia_valid & ATTR_MODE))
1437 		mode = iap->ia_mode & S_IALLUGO;
1438 
1439 	if (unlikely(path[plen] != 0)) {
1440 		char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1441 		if (path_alloced == NULL)
1442 			err = -ENOMEM;
1443 		else {
1444 			strncpy(path_alloced, path, plen);
1445 			path_alloced[plen] = 0;
1446 			err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1447 			kfree(path_alloced);
1448 		}
1449 	} else
1450 		err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1451 
1452 	if (!err) {
1453 		if (EX_ISSYNC(fhp->fh_export))
1454 			nfsd_sync_dir(dentry);
1455 	} else
1456 		err = nfserrno(err);
1457 	fh_unlock(fhp);
1458 
1459 	cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1460 	dput(dnew);
1461 	if (err==0) err = cerr;
1462 out:
1463 	return err;
1464 
1465 out_nfserr:
1466 	err = nfserrno(err);
1467 	goto out;
1468 }
1469 
1470 /*
1471  * Create a hardlink
1472  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1473  */
1474 int
1475 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1476 				char *name, int len, struct svc_fh *tfhp)
1477 {
1478 	struct dentry	*ddir, *dnew, *dold;
1479 	struct inode	*dirp, *dest;
1480 	int		err;
1481 
1482 	err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1483 	if (err)
1484 		goto out;
1485 	err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1486 	if (err)
1487 		goto out;
1488 
1489 	err = nfserr_perm;
1490 	if (!len)
1491 		goto out;
1492 	err = nfserr_exist;
1493 	if (isdotent(name, len))
1494 		goto out;
1495 
1496 	fh_lock(ffhp);
1497 	ddir = ffhp->fh_dentry;
1498 	dirp = ddir->d_inode;
1499 
1500 	dnew = lookup_one_len(name, ddir, len);
1501 	err = PTR_ERR(dnew);
1502 	if (IS_ERR(dnew))
1503 		goto out_nfserr;
1504 
1505 	dold = tfhp->fh_dentry;
1506 	dest = dold->d_inode;
1507 
1508 	err = vfs_link(dold, dirp, dnew);
1509 	if (!err) {
1510 		if (EX_ISSYNC(ffhp->fh_export)) {
1511 			nfsd_sync_dir(ddir);
1512 			write_inode_now(dest, 1);
1513 		}
1514 	} else {
1515 		if (err == -EXDEV && rqstp->rq_vers == 2)
1516 			err = nfserr_acces;
1517 		else
1518 			err = nfserrno(err);
1519 	}
1520 
1521 	fh_unlock(ffhp);
1522 	dput(dnew);
1523 out:
1524 	return err;
1525 
1526 out_nfserr:
1527 	err = nfserrno(err);
1528 	goto out;
1529 }
1530 
1531 /*
1532  * Rename a file
1533  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1534  */
1535 int
1536 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1537 			    struct svc_fh *tfhp, char *tname, int tlen)
1538 {
1539 	struct dentry	*fdentry, *tdentry, *odentry, *ndentry, *trap;
1540 	struct inode	*fdir, *tdir;
1541 	int		err;
1542 
1543 	err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1544 	if (err)
1545 		goto out;
1546 	err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1547 	if (err)
1548 		goto out;
1549 
1550 	fdentry = ffhp->fh_dentry;
1551 	fdir = fdentry->d_inode;
1552 
1553 	tdentry = tfhp->fh_dentry;
1554 	tdir = tdentry->d_inode;
1555 
1556 	err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1557 	if (fdir->i_sb != tdir->i_sb)
1558 		goto out;
1559 
1560 	err = nfserr_perm;
1561 	if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1562 		goto out;
1563 
1564 	/* cannot use fh_lock as we need deadlock protective ordering
1565 	 * so do it by hand */
1566 	trap = lock_rename(tdentry, fdentry);
1567 	ffhp->fh_locked = tfhp->fh_locked = 1;
1568 	fill_pre_wcc(ffhp);
1569 	fill_pre_wcc(tfhp);
1570 
1571 	odentry = lookup_one_len(fname, fdentry, flen);
1572 	err = PTR_ERR(odentry);
1573 	if (IS_ERR(odentry))
1574 		goto out_nfserr;
1575 
1576 	err = -ENOENT;
1577 	if (!odentry->d_inode)
1578 		goto out_dput_old;
1579 	err = -EINVAL;
1580 	if (odentry == trap)
1581 		goto out_dput_old;
1582 
1583 	ndentry = lookup_one_len(tname, tdentry, tlen);
1584 	err = PTR_ERR(ndentry);
1585 	if (IS_ERR(ndentry))
1586 		goto out_dput_old;
1587 	err = -ENOTEMPTY;
1588 	if (ndentry == trap)
1589 		goto out_dput_new;
1590 
1591 #ifdef MSNFS
1592 	if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1593 		((atomic_read(&odentry->d_count) > 1)
1594 		 || (atomic_read(&ndentry->d_count) > 1))) {
1595 			err = nfserr_perm;
1596 	} else
1597 #endif
1598 	err = vfs_rename(fdir, odentry, tdir, ndentry);
1599 	if (!err && EX_ISSYNC(tfhp->fh_export)) {
1600 		nfsd_sync_dir(tdentry);
1601 		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 = nfserr_perm;
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)
1677 		goto out_nfserr;
1678 	if (EX_ISSYNC(fhp->fh_export))
1679 		nfsd_sync_dir(dentry);
1680 
1681 out:
1682 	return err;
1683 
1684 out_nfserr:
1685 	err = nfserrno(err);
1686 	goto out;
1687 }
1688 
1689 /*
1690  * Read entries from a directory.
1691  * The  NFSv3/4 verifier we ignore for now.
1692  */
1693 int
1694 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1695 	     struct readdir_cd *cdp, encode_dent_fn func)
1696 {
1697 	int		err;
1698 	struct file	*file;
1699 	loff_t		offset = *offsetp;
1700 
1701 	err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1702 	if (err)
1703 		goto out;
1704 
1705 	offset = vfs_llseek(file, offset, 0);
1706 	if (offset < 0) {
1707 		err = nfserrno((int)offset);
1708 		goto out_close;
1709 	}
1710 
1711 	/*
1712 	 * Read the directory entries. This silly loop is necessary because
1713 	 * readdir() is not guaranteed to fill up the entire buffer, but
1714 	 * may choose to do less.
1715 	 */
1716 
1717 	do {
1718 		cdp->err = nfserr_eof; /* will be cleared on successful read */
1719 		err = vfs_readdir(file, (filldir_t) func, cdp);
1720 	} while (err >=0 && cdp->err == nfs_ok);
1721 	if (err)
1722 		err = nfserrno(err);
1723 	else
1724 		err = cdp->err;
1725 	*offsetp = vfs_llseek(file, 0, 1);
1726 
1727 	if (err == nfserr_eof || err == nfserr_toosmall)
1728 		err = nfs_ok; /* can still be found in ->err */
1729 out_close:
1730 	nfsd_close(file);
1731 out:
1732 	return err;
1733 }
1734 
1735 /*
1736  * Get file system stats
1737  * N.B. After this call fhp needs an fh_put
1738  */
1739 int
1740 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1741 {
1742 	int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1743 	if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1744 		err = nfserr_io;
1745 	return err;
1746 }
1747 
1748 /*
1749  * Check for a user's access permissions to this inode.
1750  */
1751 int
1752 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1753 {
1754 	struct inode	*inode = dentry->d_inode;
1755 	int		err;
1756 
1757 	if (acc == MAY_NOP)
1758 		return 0;
1759 #if 0
1760 	dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1761 		acc,
1762 		(acc & MAY_READ)?	" read"  : "",
1763 		(acc & MAY_WRITE)?	" write" : "",
1764 		(acc & MAY_EXEC)?	" exec"  : "",
1765 		(acc & MAY_SATTR)?	" sattr" : "",
1766 		(acc & MAY_TRUNC)?	" trunc" : "",
1767 		(acc & MAY_LOCK)?	" lock"  : "",
1768 		(acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1769 		inode->i_mode,
1770 		IS_IMMUTABLE(inode)?	" immut" : "",
1771 		IS_APPEND(inode)?	" append" : "",
1772 		IS_RDONLY(inode)?	" ro" : "");
1773 	dprintk("      owner %d/%d user %d/%d\n",
1774 		inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1775 #endif
1776 
1777 	/* Normally we reject any write/sattr etc access on a read-only file
1778 	 * system.  But if it is IRIX doing check on write-access for a
1779 	 * device special file, we ignore rofs.
1780 	 */
1781 	if (!(acc & MAY_LOCAL_ACCESS))
1782 		if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1783 			if (EX_RDONLY(exp) || IS_RDONLY(inode))
1784 				return nfserr_rofs;
1785 			if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1786 				return nfserr_perm;
1787 		}
1788 	if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1789 		return nfserr_perm;
1790 
1791 	if (acc & MAY_LOCK) {
1792 		/* If we cannot rely on authentication in NLM requests,
1793 		 * just allow locks, otherwise require read permission, or
1794 		 * ownership
1795 		 */
1796 		if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1797 			return 0;
1798 		else
1799 			acc = MAY_READ | MAY_OWNER_OVERRIDE;
1800 	}
1801 	/*
1802 	 * The file owner always gets access permission for accesses that
1803 	 * would normally be checked at open time. This is to make
1804 	 * file access work even when the client has done a fchmod(fd, 0).
1805 	 *
1806 	 * However, `cp foo bar' should fail nevertheless when bar is
1807 	 * readonly. A sensible way to do this might be to reject all
1808 	 * attempts to truncate a read-only file, because a creat() call
1809 	 * always implies file truncation.
1810 	 * ... but this isn't really fair.  A process may reasonably call
1811 	 * ftruncate on an open file descriptor on a file with perm 000.
1812 	 * We must trust the client to do permission checking - using "ACCESS"
1813 	 * with NFSv3.
1814 	 */
1815 	if ((acc & MAY_OWNER_OVERRIDE) &&
1816 	    inode->i_uid == current->fsuid)
1817 		return 0;
1818 
1819 	err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1820 
1821 	/* Allow read access to binaries even when mode 111 */
1822 	if (err == -EACCES && S_ISREG(inode->i_mode) &&
1823 	    acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1824 		err = permission(inode, MAY_EXEC, NULL);
1825 
1826 	return err? nfserrno(err) : 0;
1827 }
1828 
1829 void
1830 nfsd_racache_shutdown(void)
1831 {
1832 	if (!raparm_cache)
1833 		return;
1834 	dprintk("nfsd: freeing readahead buffers.\n");
1835 	kfree(raparml);
1836 	raparm_cache = raparml = NULL;
1837 }
1838 /*
1839  * Initialize readahead param cache
1840  */
1841 int
1842 nfsd_racache_init(int cache_size)
1843 {
1844 	int	i;
1845 
1846 	if (raparm_cache)
1847 		return 0;
1848 	raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1849 
1850 	if (raparml != NULL) {
1851 		dprintk("nfsd: allocating %d readahead buffers.\n",
1852 			cache_size);
1853 		memset(raparml, 0, sizeof(struct raparms) * cache_size);
1854 		for (i = 0; i < cache_size - 1; i++) {
1855 			raparml[i].p_next = raparml + i + 1;
1856 		}
1857 		raparm_cache = raparml;
1858 	} else {
1859 		printk(KERN_WARNING
1860 		       "nfsd: Could not allocate memory read-ahead cache.\n");
1861 		return -ENOMEM;
1862 	}
1863 	nfsdstats.ra_size = cache_size;
1864 	return 0;
1865 }
1866 
1867 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1868 struct posix_acl *
1869 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1870 {
1871 	struct inode *inode = fhp->fh_dentry->d_inode;
1872 	char *name;
1873 	void *value = NULL;
1874 	ssize_t size;
1875 	struct posix_acl *acl;
1876 
1877 	if (!IS_POSIXACL(inode) || !inode->i_op || !inode->i_op->getxattr)
1878 		return ERR_PTR(-EOPNOTSUPP);
1879 	switch(type) {
1880 		case ACL_TYPE_ACCESS:
1881 			name = POSIX_ACL_XATTR_ACCESS;
1882 			break;
1883 		case ACL_TYPE_DEFAULT:
1884 			name = POSIX_ACL_XATTR_DEFAULT;
1885 			break;
1886 		default:
1887 			return ERR_PTR(-EOPNOTSUPP);
1888 	}
1889 
1890 	size = inode->i_op->getxattr(fhp->fh_dentry, name, NULL, 0);
1891 
1892 	if (size < 0) {
1893 		acl = ERR_PTR(size);
1894 		goto getout;
1895 	} else if (size > 0) {
1896 		value = kmalloc(size, GFP_KERNEL);
1897 		if (!value) {
1898 			acl = ERR_PTR(-ENOMEM);
1899 			goto getout;
1900 		}
1901 		size = inode->i_op->getxattr(fhp->fh_dentry, name, value, size);
1902 		if (size < 0) {
1903 			acl = ERR_PTR(size);
1904 			goto getout;
1905 		}
1906 	}
1907 	acl = posix_acl_from_xattr(value, size);
1908 
1909 getout:
1910 	kfree(value);
1911 	return acl;
1912 }
1913 
1914 int
1915 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1916 {
1917 	struct inode *inode = fhp->fh_dentry->d_inode;
1918 	char *name;
1919 	void *value = NULL;
1920 	size_t size;
1921 	int error;
1922 
1923 	if (!IS_POSIXACL(inode) || !inode->i_op ||
1924 	    !inode->i_op->setxattr || !inode->i_op->removexattr)
1925 		return -EOPNOTSUPP;
1926 	switch(type) {
1927 		case ACL_TYPE_ACCESS:
1928 			name = POSIX_ACL_XATTR_ACCESS;
1929 			break;
1930 		case ACL_TYPE_DEFAULT:
1931 			name = POSIX_ACL_XATTR_DEFAULT;
1932 			break;
1933 		default:
1934 			return -EOPNOTSUPP;
1935 	}
1936 
1937 	if (acl && acl->a_count) {
1938 		size = posix_acl_xattr_size(acl->a_count);
1939 		value = kmalloc(size, GFP_KERNEL);
1940 		if (!value)
1941 			return -ENOMEM;
1942 		size = posix_acl_to_xattr(acl, value, size);
1943 		if (size < 0) {
1944 			error = size;
1945 			goto getout;
1946 		}
1947 	} else
1948 		size = 0;
1949 
1950 	if (!fhp->fh_locked)
1951 		fh_lock(fhp);  /* unlocking is done automatically */
1952 	if (size)
1953 		error = inode->i_op->setxattr(fhp->fh_dentry, name,
1954 					      value, size, 0);
1955 	else {
1956 		if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1957 			error = 0;
1958 		else {
1959 			error = inode->i_op->removexattr(fhp->fh_dentry, name);
1960 			if (error == -ENODATA)
1961 				error = 0;
1962 		}
1963 	}
1964 
1965 getout:
1966 	kfree(value);
1967 	return error;
1968 }
1969 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
1970