xref: /linux/fs/nfsd/vfs.c (revision d524dac9279b6a41ffdf7ff7958c577f2e387db6)
1 /*
2  * File operations used by nfsd. Some of these have been ripped from
3  * other parts of the kernel because they weren't exported, others
4  * are partial duplicates with added or changed functionality.
5  *
6  * Note that several functions dget() the dentry upon which they want
7  * to act, most notably those that create directory entries. Response
8  * dentry's are dput()'d if necessary in the release callback.
9  * So if you notice code paths that apparently fail to dput() the
10  * dentry, don't worry--they have been taken care of.
11  *
12  * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13  * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
14  */
15 
16 #include <linux/fs.h>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
31 
32 #ifdef CONFIG_NFSD_V3
33 #include "xdr3.h"
34 #endif /* CONFIG_NFSD_V3 */
35 
36 #ifdef CONFIG_NFSD_V4
37 #include "acl.h"
38 #include "idmap.h"
39 #endif /* CONFIG_NFSD_V4 */
40 
41 #include "nfsd.h"
42 #include "vfs.h"
43 
44 #define NFSDDBG_FACILITY		NFSDDBG_FILEOP
45 
46 
47 /*
48  * This is a cache of readahead params that help us choose the proper
49  * readahead strategy. Initially, we set all readahead parameters to 0
50  * and let the VFS handle things.
51  * If you increase the number of cached files very much, you'll need to
52  * add a hash table here.
53  */
54 struct raparms {
55 	struct raparms		*p_next;
56 	unsigned int		p_count;
57 	ino_t			p_ino;
58 	dev_t			p_dev;
59 	int			p_set;
60 	struct file_ra_state	p_ra;
61 	unsigned int		p_hindex;
62 };
63 
64 struct raparm_hbucket {
65 	struct raparms		*pb_head;
66 	spinlock_t		pb_lock;
67 } ____cacheline_aligned_in_smp;
68 
69 #define RAPARM_HASH_BITS	4
70 #define RAPARM_HASH_SIZE	(1<<RAPARM_HASH_BITS)
71 #define RAPARM_HASH_MASK	(RAPARM_HASH_SIZE-1)
72 static struct raparm_hbucket	raparm_hash[RAPARM_HASH_SIZE];
73 
74 /*
75  * Called from nfsd_lookup and encode_dirent. Check if we have crossed
76  * a mount point.
77  * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
78  *  or nfs_ok having possibly changed *dpp and *expp
79  */
80 int
81 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
82 		        struct svc_export **expp)
83 {
84 	struct svc_export *exp = *expp, *exp2 = NULL;
85 	struct dentry *dentry = *dpp;
86 	struct path path = {.mnt = mntget(exp->ex_path.mnt),
87 			    .dentry = dget(dentry)};
88 	int err = 0;
89 
90 	err = follow_down(&path, false);
91 	if (err < 0)
92 		goto out;
93 
94 	exp2 = rqst_exp_get_by_name(rqstp, &path);
95 	if (IS_ERR(exp2)) {
96 		err = PTR_ERR(exp2);
97 		/*
98 		 * We normally allow NFS clients to continue
99 		 * "underneath" a mountpoint that is not exported.
100 		 * The exception is V4ROOT, where no traversal is ever
101 		 * allowed without an explicit export of the new
102 		 * directory.
103 		 */
104 		if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
105 			err = 0;
106 		path_put(&path);
107 		goto out;
108 	}
109 	if (nfsd_v4client(rqstp) ||
110 		(exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
111 		/* successfully crossed mount point */
112 		/*
113 		 * This is subtle: path.dentry is *not* on path.mnt
114 		 * at this point.  The only reason we are safe is that
115 		 * original mnt is pinned down by exp, so we should
116 		 * put path *before* putting exp
117 		 */
118 		*dpp = path.dentry;
119 		path.dentry = dentry;
120 		*expp = exp2;
121 		exp2 = exp;
122 	}
123 	path_put(&path);
124 	exp_put(exp2);
125 out:
126 	return err;
127 }
128 
129 static void follow_to_parent(struct path *path)
130 {
131 	struct dentry *dp;
132 
133 	while (path->dentry == path->mnt->mnt_root && follow_up(path))
134 		;
135 	dp = dget_parent(path->dentry);
136 	dput(path->dentry);
137 	path->dentry = dp;
138 }
139 
140 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
141 {
142 	struct svc_export *exp2;
143 	struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
144 			    .dentry = dget(dparent)};
145 
146 	follow_to_parent(&path);
147 
148 	exp2 = rqst_exp_parent(rqstp, &path);
149 	if (PTR_ERR(exp2) == -ENOENT) {
150 		*dentryp = dget(dparent);
151 	} else if (IS_ERR(exp2)) {
152 		path_put(&path);
153 		return PTR_ERR(exp2);
154 	} else {
155 		*dentryp = dget(path.dentry);
156 		exp_put(*exp);
157 		*exp = exp2;
158 	}
159 	path_put(&path);
160 	return 0;
161 }
162 
163 /*
164  * For nfsd purposes, we treat V4ROOT exports as though there was an
165  * export at *every* directory.
166  */
167 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
168 {
169 	if (d_mountpoint(dentry))
170 		return 1;
171 	if (!(exp->ex_flags & NFSEXP_V4ROOT))
172 		return 0;
173 	return dentry->d_inode != NULL;
174 }
175 
176 __be32
177 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
178 		   const char *name, unsigned int len,
179 		   struct svc_export **exp_ret, struct dentry **dentry_ret)
180 {
181 	struct svc_export	*exp;
182 	struct dentry		*dparent;
183 	struct dentry		*dentry;
184 	__be32			err;
185 	int			host_err;
186 
187 	dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
188 
189 	/* Obtain dentry and export. */
190 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
191 	if (err)
192 		return err;
193 
194 	dparent = fhp->fh_dentry;
195 	exp  = fhp->fh_export;
196 	exp_get(exp);
197 
198 	/* Lookup the name, but don't follow links */
199 	if (isdotent(name, len)) {
200 		if (len==1)
201 			dentry = dget(dparent);
202 		else if (dparent != exp->ex_path.dentry)
203 			dentry = dget_parent(dparent);
204 		else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
205 			dentry = dget(dparent); /* .. == . just like at / */
206 		else {
207 			/* checking mountpoint crossing is very different when stepping up */
208 			host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
209 			if (host_err)
210 				goto out_nfserr;
211 		}
212 	} else {
213 		fh_lock(fhp);
214 		dentry = lookup_one_len(name, dparent, len);
215 		host_err = PTR_ERR(dentry);
216 		if (IS_ERR(dentry))
217 			goto out_nfserr;
218 		/*
219 		 * check if we have crossed a mount point ...
220 		 */
221 		if (nfsd_mountpoint(dentry, exp)) {
222 			if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
223 				dput(dentry);
224 				goto out_nfserr;
225 			}
226 		}
227 	}
228 	*dentry_ret = dentry;
229 	*exp_ret = exp;
230 	return 0;
231 
232 out_nfserr:
233 	exp_put(exp);
234 	return nfserrno(host_err);
235 }
236 
237 /*
238  * Look up one component of a pathname.
239  * N.B. After this call _both_ fhp and resfh need an fh_put
240  *
241  * If the lookup would cross a mountpoint, and the mounted filesystem
242  * is exported to the client with NFSEXP_NOHIDE, then the lookup is
243  * accepted as it stands and the mounted directory is
244  * returned. Otherwise the covered directory is returned.
245  * NOTE: this mountpoint crossing is not supported properly by all
246  *   clients and is explicitly disallowed for NFSv3
247  *      NeilBrown <neilb@cse.unsw.edu.au>
248  */
249 __be32
250 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
251 				unsigned int len, struct svc_fh *resfh)
252 {
253 	struct svc_export	*exp;
254 	struct dentry		*dentry;
255 	__be32 err;
256 
257 	err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
258 	if (err)
259 		return err;
260 	err = check_nfsd_access(exp, rqstp);
261 	if (err)
262 		goto out;
263 	/*
264 	 * Note: we compose the file handle now, but as the
265 	 * dentry may be negative, it may need to be updated.
266 	 */
267 	err = fh_compose(resfh, exp, dentry, fhp);
268 	if (!err && !dentry->d_inode)
269 		err = nfserr_noent;
270 out:
271 	dput(dentry);
272 	exp_put(exp);
273 	return err;
274 }
275 
276 static int nfsd_break_lease(struct inode *inode)
277 {
278 	if (!S_ISREG(inode->i_mode))
279 		return 0;
280 	return break_lease(inode, O_WRONLY | O_NONBLOCK);
281 }
282 
283 /*
284  * Commit metadata changes to stable storage.
285  */
286 static int
287 commit_metadata(struct svc_fh *fhp)
288 {
289 	struct inode *inode = fhp->fh_dentry->d_inode;
290 	const struct export_operations *export_ops = inode->i_sb->s_export_op;
291 
292 	if (!EX_ISSYNC(fhp->fh_export))
293 		return 0;
294 
295 	if (export_ops->commit_metadata)
296 		return export_ops->commit_metadata(inode);
297 	return sync_inode_metadata(inode, 1);
298 }
299 
300 /*
301  * Set various file attributes.
302  * N.B. After this call fhp needs an fh_put
303  */
304 __be32
305 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
306 	     int check_guard, time_t guardtime)
307 {
308 	struct dentry	*dentry;
309 	struct inode	*inode;
310 	int		accmode = NFSD_MAY_SATTR;
311 	int		ftype = 0;
312 	__be32		err;
313 	int		host_err;
314 	int		size_change = 0;
315 
316 	if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
317 		accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
318 	if (iap->ia_valid & ATTR_SIZE)
319 		ftype = S_IFREG;
320 
321 	/* Get inode */
322 	err = fh_verify(rqstp, fhp, ftype, accmode);
323 	if (err)
324 		goto out;
325 
326 	dentry = fhp->fh_dentry;
327 	inode = dentry->d_inode;
328 
329 	/* Ignore any mode updates on symlinks */
330 	if (S_ISLNK(inode->i_mode))
331 		iap->ia_valid &= ~ATTR_MODE;
332 
333 	if (!iap->ia_valid)
334 		goto out;
335 
336 	/*
337 	 * NFSv2 does not differentiate between "set-[ac]time-to-now"
338 	 * which only requires access, and "set-[ac]time-to-X" which
339 	 * requires ownership.
340 	 * So if it looks like it might be "set both to the same time which
341 	 * is close to now", and if inode_change_ok fails, then we
342 	 * convert to "set to now" instead of "set to explicit time"
343 	 *
344 	 * We only call inode_change_ok as the last test as technically
345 	 * it is not an interface that we should be using.  It is only
346 	 * valid if the filesystem does not define it's own i_op->setattr.
347 	 */
348 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
349 #define	MAX_TOUCH_TIME_ERROR (30*60)
350 	if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
351 	    iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
352 		/*
353 		 * Looks probable.
354 		 *
355 		 * Now just make sure time is in the right ballpark.
356 		 * Solaris, at least, doesn't seem to care what the time
357 		 * request is.  We require it be within 30 minutes of now.
358 		 */
359 		time_t delta = iap->ia_atime.tv_sec - get_seconds();
360 		if (delta < 0)
361 			delta = -delta;
362 		if (delta < MAX_TOUCH_TIME_ERROR &&
363 		    inode_change_ok(inode, iap) != 0) {
364 			/*
365 			 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
366 			 * This will cause notify_change to set these times
367 			 * to "now"
368 			 */
369 			iap->ia_valid &= ~BOTH_TIME_SET;
370 		}
371 	}
372 
373 	/*
374 	 * The size case is special.
375 	 * It changes the file as well as the attributes.
376 	 */
377 	if (iap->ia_valid & ATTR_SIZE) {
378 		if (iap->ia_size < inode->i_size) {
379 			err = nfsd_permission(rqstp, fhp->fh_export, dentry,
380 					NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
381 			if (err)
382 				goto out;
383 		}
384 
385 		host_err = get_write_access(inode);
386 		if (host_err)
387 			goto out_nfserr;
388 
389 		size_change = 1;
390 		host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
391 		if (host_err) {
392 			put_write_access(inode);
393 			goto out_nfserr;
394 		}
395 	}
396 
397 	/* sanitize the mode change */
398 	if (iap->ia_valid & ATTR_MODE) {
399 		iap->ia_mode &= S_IALLUGO;
400 		iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
401 	}
402 
403 	/* Revoke setuid/setgid on chown */
404 	if (!S_ISDIR(inode->i_mode) &&
405 	    (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
406 	     ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
407 		iap->ia_valid |= ATTR_KILL_PRIV;
408 		if (iap->ia_valid & ATTR_MODE) {
409 			/* we're setting mode too, just clear the s*id bits */
410 			iap->ia_mode &= ~S_ISUID;
411 			if (iap->ia_mode & S_IXGRP)
412 				iap->ia_mode &= ~S_ISGID;
413 		} else {
414 			/* set ATTR_KILL_* bits and let VFS handle it */
415 			iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
416 		}
417 	}
418 
419 	/* Change the attributes. */
420 
421 	iap->ia_valid |= ATTR_CTIME;
422 
423 	err = nfserr_notsync;
424 	if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
425 		host_err = nfsd_break_lease(inode);
426 		if (host_err)
427 			goto out_nfserr;
428 		fh_lock(fhp);
429 
430 		host_err = notify_change(dentry, iap);
431 		err = nfserrno(host_err);
432 		fh_unlock(fhp);
433 	}
434 	if (size_change)
435 		put_write_access(inode);
436 	if (!err)
437 		commit_metadata(fhp);
438 out:
439 	return err;
440 
441 out_nfserr:
442 	err = nfserrno(host_err);
443 	goto out;
444 }
445 
446 #if defined(CONFIG_NFSD_V2_ACL) || \
447     defined(CONFIG_NFSD_V3_ACL) || \
448     defined(CONFIG_NFSD_V4)
449 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
450 {
451 	ssize_t buflen;
452 	ssize_t ret;
453 
454 	buflen = vfs_getxattr(dentry, key, NULL, 0);
455 	if (buflen <= 0)
456 		return buflen;
457 
458 	*buf = kmalloc(buflen, GFP_KERNEL);
459 	if (!*buf)
460 		return -ENOMEM;
461 
462 	ret = vfs_getxattr(dentry, key, *buf, buflen);
463 	if (ret < 0)
464 		kfree(*buf);
465 	return ret;
466 }
467 #endif
468 
469 #if defined(CONFIG_NFSD_V4)
470 static int
471 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
472 {
473 	int len;
474 	size_t buflen;
475 	char *buf = NULL;
476 	int error = 0;
477 
478 	buflen = posix_acl_xattr_size(pacl->a_count);
479 	buf = kmalloc(buflen, GFP_KERNEL);
480 	error = -ENOMEM;
481 	if (buf == NULL)
482 		goto out;
483 
484 	len = posix_acl_to_xattr(pacl, buf, buflen);
485 	if (len < 0) {
486 		error = len;
487 		goto out;
488 	}
489 
490 	error = vfs_setxattr(dentry, key, buf, len, 0);
491 out:
492 	kfree(buf);
493 	return error;
494 }
495 
496 __be32
497 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
498     struct nfs4_acl *acl)
499 {
500 	__be32 error;
501 	int host_error;
502 	struct dentry *dentry;
503 	struct inode *inode;
504 	struct posix_acl *pacl = NULL, *dpacl = NULL;
505 	unsigned int flags = 0;
506 
507 	/* Get inode */
508 	error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
509 	if (error)
510 		return error;
511 
512 	dentry = fhp->fh_dentry;
513 	inode = dentry->d_inode;
514 	if (S_ISDIR(inode->i_mode))
515 		flags = NFS4_ACL_DIR;
516 
517 	host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
518 	if (host_error == -EINVAL) {
519 		return nfserr_attrnotsupp;
520 	} else if (host_error < 0)
521 		goto out_nfserr;
522 
523 	host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
524 	if (host_error < 0)
525 		goto out_release;
526 
527 	if (S_ISDIR(inode->i_mode))
528 		host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
529 
530 out_release:
531 	posix_acl_release(pacl);
532 	posix_acl_release(dpacl);
533 out_nfserr:
534 	if (host_error == -EOPNOTSUPP)
535 		return nfserr_attrnotsupp;
536 	else
537 		return nfserrno(host_error);
538 }
539 
540 static struct posix_acl *
541 _get_posix_acl(struct dentry *dentry, char *key)
542 {
543 	void *buf = NULL;
544 	struct posix_acl *pacl = NULL;
545 	int buflen;
546 
547 	buflen = nfsd_getxattr(dentry, key, &buf);
548 	if (!buflen)
549 		buflen = -ENODATA;
550 	if (buflen <= 0)
551 		return ERR_PTR(buflen);
552 
553 	pacl = posix_acl_from_xattr(buf, buflen);
554 	kfree(buf);
555 	return pacl;
556 }
557 
558 int
559 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
560 {
561 	struct inode *inode = dentry->d_inode;
562 	int error = 0;
563 	struct posix_acl *pacl = NULL, *dpacl = NULL;
564 	unsigned int flags = 0;
565 
566 	pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
567 	if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
568 		pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
569 	if (IS_ERR(pacl)) {
570 		error = PTR_ERR(pacl);
571 		pacl = NULL;
572 		goto out;
573 	}
574 
575 	if (S_ISDIR(inode->i_mode)) {
576 		dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
577 		if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
578 			dpacl = NULL;
579 		else if (IS_ERR(dpacl)) {
580 			error = PTR_ERR(dpacl);
581 			dpacl = NULL;
582 			goto out;
583 		}
584 		flags = NFS4_ACL_DIR;
585 	}
586 
587 	*acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
588 	if (IS_ERR(*acl)) {
589 		error = PTR_ERR(*acl);
590 		*acl = NULL;
591 	}
592  out:
593 	posix_acl_release(pacl);
594 	posix_acl_release(dpacl);
595 	return error;
596 }
597 
598 #endif /* defined(CONFIG_NFSD_V4) */
599 
600 #ifdef CONFIG_NFSD_V3
601 /*
602  * Check server access rights to a file system object
603  */
604 struct accessmap {
605 	u32		access;
606 	int		how;
607 };
608 static struct accessmap	nfs3_regaccess[] = {
609     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
610     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
611     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_TRUNC	},
612     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE			},
613 
614     {	0,			0				}
615 };
616 
617 static struct accessmap	nfs3_diraccess[] = {
618     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
619     {	NFS3_ACCESS_LOOKUP,	NFSD_MAY_EXEC			},
620     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
621     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_EXEC|NFSD_MAY_WRITE	},
622     {	NFS3_ACCESS_DELETE,	NFSD_MAY_REMOVE			},
623 
624     {	0,			0				}
625 };
626 
627 static struct accessmap	nfs3_anyaccess[] = {
628 	/* Some clients - Solaris 2.6 at least, make an access call
629 	 * to the server to check for access for things like /dev/null
630 	 * (which really, the server doesn't care about).  So
631 	 * We provide simple access checking for them, looking
632 	 * mainly at mode bits, and we make sure to ignore read-only
633 	 * filesystem checks
634 	 */
635     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
636     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
637     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
638     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
639 
640     {	0,			0				}
641 };
642 
643 __be32
644 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
645 {
646 	struct accessmap	*map;
647 	struct svc_export	*export;
648 	struct dentry		*dentry;
649 	u32			query, result = 0, sresult = 0;
650 	__be32			error;
651 
652 	error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
653 	if (error)
654 		goto out;
655 
656 	export = fhp->fh_export;
657 	dentry = fhp->fh_dentry;
658 
659 	if (S_ISREG(dentry->d_inode->i_mode))
660 		map = nfs3_regaccess;
661 	else if (S_ISDIR(dentry->d_inode->i_mode))
662 		map = nfs3_diraccess;
663 	else
664 		map = nfs3_anyaccess;
665 
666 
667 	query = *access;
668 	for  (; map->access; map++) {
669 		if (map->access & query) {
670 			__be32 err2;
671 
672 			sresult |= map->access;
673 
674 			err2 = nfsd_permission(rqstp, export, dentry, map->how);
675 			switch (err2) {
676 			case nfs_ok:
677 				result |= map->access;
678 				break;
679 
680 			/* the following error codes just mean the access was not allowed,
681 			 * rather than an error occurred */
682 			case nfserr_rofs:
683 			case nfserr_acces:
684 			case nfserr_perm:
685 				/* simply don't "or" in the access bit. */
686 				break;
687 			default:
688 				error = err2;
689 				goto out;
690 			}
691 		}
692 	}
693 	*access = result;
694 	if (supported)
695 		*supported = sresult;
696 
697  out:
698 	return error;
699 }
700 #endif /* CONFIG_NFSD_V3 */
701 
702 
703 
704 /*
705  * Open an existing file or directory.
706  * The access argument indicates the type of open (read/write/lock)
707  * N.B. After this call fhp needs an fh_put
708  */
709 __be32
710 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
711 			int access, struct file **filp)
712 {
713 	struct dentry	*dentry;
714 	struct inode	*inode;
715 	int		flags = O_RDONLY|O_LARGEFILE;
716 	__be32		err;
717 	int		host_err = 0;
718 
719 	validate_process_creds();
720 
721 	/*
722 	 * If we get here, then the client has already done an "open",
723 	 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
724 	 * in case a chmod has now revoked permission.
725 	 */
726 	err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
727 	if (err)
728 		goto out;
729 
730 	dentry = fhp->fh_dentry;
731 	inode = dentry->d_inode;
732 
733 	/* Disallow write access to files with the append-only bit set
734 	 * or any access when mandatory locking enabled
735 	 */
736 	err = nfserr_perm;
737 	if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
738 		goto out;
739 	/*
740 	 * We must ignore files (but only files) which might have mandatory
741 	 * locks on them because there is no way to know if the accesser has
742 	 * the lock.
743 	 */
744 	if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
745 		goto out;
746 
747 	if (!inode->i_fop)
748 		goto out;
749 
750 	/*
751 	 * Check to see if there are any leases on this file.
752 	 * This may block while leases are broken.
753 	 */
754 	if (!(access & NFSD_MAY_NOT_BREAK_LEASE))
755 		host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
756 	if (host_err) /* NOMEM or WOULDBLOCK */
757 		goto out_nfserr;
758 
759 	if (access & NFSD_MAY_WRITE) {
760 		if (access & NFSD_MAY_READ)
761 			flags = O_RDWR|O_LARGEFILE;
762 		else
763 			flags = O_WRONLY|O_LARGEFILE;
764 	}
765 	*filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
766 			    flags, current_cred());
767 	if (IS_ERR(*filp))
768 		host_err = PTR_ERR(*filp);
769 	else
770 		host_err = ima_file_check(*filp, access);
771 out_nfserr:
772 	err = nfserrno(host_err);
773 out:
774 	validate_process_creds();
775 	return err;
776 }
777 
778 /*
779  * Close a file.
780  */
781 void
782 nfsd_close(struct file *filp)
783 {
784 	fput(filp);
785 }
786 
787 /*
788  * Obtain the readahead parameters for the file
789  * specified by (dev, ino).
790  */
791 
792 static inline struct raparms *
793 nfsd_get_raparms(dev_t dev, ino_t ino)
794 {
795 	struct raparms	*ra, **rap, **frap = NULL;
796 	int depth = 0;
797 	unsigned int hash;
798 	struct raparm_hbucket *rab;
799 
800 	hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
801 	rab = &raparm_hash[hash];
802 
803 	spin_lock(&rab->pb_lock);
804 	for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
805 		if (ra->p_ino == ino && ra->p_dev == dev)
806 			goto found;
807 		depth++;
808 		if (ra->p_count == 0)
809 			frap = rap;
810 	}
811 	depth = nfsdstats.ra_size*11/10;
812 	if (!frap) {
813 		spin_unlock(&rab->pb_lock);
814 		return NULL;
815 	}
816 	rap = frap;
817 	ra = *frap;
818 	ra->p_dev = dev;
819 	ra->p_ino = ino;
820 	ra->p_set = 0;
821 	ra->p_hindex = hash;
822 found:
823 	if (rap != &rab->pb_head) {
824 		*rap = ra->p_next;
825 		ra->p_next   = rab->pb_head;
826 		rab->pb_head = ra;
827 	}
828 	ra->p_count++;
829 	nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
830 	spin_unlock(&rab->pb_lock);
831 	return ra;
832 }
833 
834 /*
835  * Grab and keep cached pages associated with a file in the svc_rqst
836  * so that they can be passed to the network sendmsg/sendpage routines
837  * directly. They will be released after the sending has completed.
838  */
839 static int
840 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
841 		  struct splice_desc *sd)
842 {
843 	struct svc_rqst *rqstp = sd->u.data;
844 	struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
845 	struct page *page = buf->page;
846 	size_t size;
847 
848 	size = sd->len;
849 
850 	if (rqstp->rq_res.page_len == 0) {
851 		get_page(page);
852 		put_page(*pp);
853 		*pp = page;
854 		rqstp->rq_resused++;
855 		rqstp->rq_res.page_base = buf->offset;
856 		rqstp->rq_res.page_len = size;
857 	} else if (page != pp[-1]) {
858 		get_page(page);
859 		if (*pp)
860 			put_page(*pp);
861 		*pp = page;
862 		rqstp->rq_resused++;
863 		rqstp->rq_res.page_len += size;
864 	} else
865 		rqstp->rq_res.page_len += size;
866 
867 	return size;
868 }
869 
870 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
871 				    struct splice_desc *sd)
872 {
873 	return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
874 }
875 
876 static __be32
877 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
878               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
879 {
880 	struct inode *inode;
881 	mm_segment_t	oldfs;
882 	__be32		err;
883 	int		host_err;
884 
885 	err = nfserr_perm;
886 	inode = file->f_path.dentry->d_inode;
887 
888 	if (file->f_op->splice_read && rqstp->rq_splice_ok) {
889 		struct splice_desc sd = {
890 			.len		= 0,
891 			.total_len	= *count,
892 			.pos		= offset,
893 			.u.data		= rqstp,
894 		};
895 
896 		rqstp->rq_resused = 1;
897 		host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
898 	} else {
899 		oldfs = get_fs();
900 		set_fs(KERNEL_DS);
901 		host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
902 		set_fs(oldfs);
903 	}
904 
905 	if (host_err >= 0) {
906 		nfsdstats.io_read += host_err;
907 		*count = host_err;
908 		err = 0;
909 		fsnotify_access(file);
910 	} else
911 		err = nfserrno(host_err);
912 	return err;
913 }
914 
915 static void kill_suid(struct dentry *dentry)
916 {
917 	struct iattr	ia;
918 	ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
919 
920 	mutex_lock(&dentry->d_inode->i_mutex);
921 	notify_change(dentry, &ia);
922 	mutex_unlock(&dentry->d_inode->i_mutex);
923 }
924 
925 /*
926  * Gathered writes: If another process is currently writing to the file,
927  * there's a high chance this is another nfsd (triggered by a bulk write
928  * from a client's biod). Rather than syncing the file with each write
929  * request, we sleep for 10 msec.
930  *
931  * I don't know if this roughly approximates C. Juszak's idea of
932  * gathered writes, but it's a nice and simple solution (IMHO), and it
933  * seems to work:-)
934  *
935  * Note: we do this only in the NFSv2 case, since v3 and higher have a
936  * better tool (separate unstable writes and commits) for solving this
937  * problem.
938  */
939 static int wait_for_concurrent_writes(struct file *file)
940 {
941 	struct inode *inode = file->f_path.dentry->d_inode;
942 	static ino_t last_ino;
943 	static dev_t last_dev;
944 	int err = 0;
945 
946 	if (atomic_read(&inode->i_writecount) > 1
947 	    || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
948 		dprintk("nfsd: write defer %d\n", task_pid_nr(current));
949 		msleep(10);
950 		dprintk("nfsd: write resume %d\n", task_pid_nr(current));
951 	}
952 
953 	if (inode->i_state & I_DIRTY) {
954 		dprintk("nfsd: write sync %d\n", task_pid_nr(current));
955 		err = vfs_fsync(file, 0);
956 	}
957 	last_ino = inode->i_ino;
958 	last_dev = inode->i_sb->s_dev;
959 	return err;
960 }
961 
962 static __be32
963 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
964 				loff_t offset, struct kvec *vec, int vlen,
965 				unsigned long *cnt, int *stablep)
966 {
967 	struct svc_export	*exp;
968 	struct dentry		*dentry;
969 	struct inode		*inode;
970 	mm_segment_t		oldfs;
971 	__be32			err = 0;
972 	int			host_err;
973 	int			stable = *stablep;
974 	int			use_wgather;
975 
976 	dentry = file->f_path.dentry;
977 	inode = dentry->d_inode;
978 	exp   = fhp->fh_export;
979 
980 	/*
981 	 * Request sync writes if
982 	 *  -	the sync export option has been set, or
983 	 *  -	the client requested O_SYNC behavior (NFSv3 feature).
984 	 *  -   The file system doesn't support fsync().
985 	 * When NFSv2 gathered writes have been configured for this volume,
986 	 * flushing the data to disk is handled separately below.
987 	 */
988 	use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
989 
990 	if (!file->f_op->fsync) {/* COMMIT3 cannot work */
991 	       stable = 2;
992 	       *stablep = 2; /* FILE_SYNC */
993 	}
994 
995 	if (!EX_ISSYNC(exp))
996 		stable = 0;
997 	if (stable && !use_wgather) {
998 		spin_lock(&file->f_lock);
999 		file->f_flags |= O_SYNC;
1000 		spin_unlock(&file->f_lock);
1001 	}
1002 
1003 	/* Write the data. */
1004 	oldfs = get_fs(); set_fs(KERNEL_DS);
1005 	host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1006 	set_fs(oldfs);
1007 	if (host_err < 0)
1008 		goto out_nfserr;
1009 	*cnt = host_err;
1010 	nfsdstats.io_write += host_err;
1011 	fsnotify_modify(file);
1012 
1013 	/* clear setuid/setgid flag after write */
1014 	if (inode->i_mode & (S_ISUID | S_ISGID))
1015 		kill_suid(dentry);
1016 
1017 	if (stable && use_wgather)
1018 		host_err = wait_for_concurrent_writes(file);
1019 
1020 out_nfserr:
1021 	dprintk("nfsd: write complete host_err=%d\n", host_err);
1022 	if (host_err >= 0)
1023 		err = 0;
1024 	else
1025 		err = nfserrno(host_err);
1026 	return err;
1027 }
1028 
1029 /*
1030  * Read data from a file. count must contain the requested read count
1031  * on entry. On return, *count contains the number of bytes actually read.
1032  * N.B. After this call fhp needs an fh_put
1033  */
1034 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1035 	loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1036 {
1037 	struct file *file;
1038 	struct inode *inode;
1039 	struct raparms	*ra;
1040 	__be32 err;
1041 
1042 	err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1043 	if (err)
1044 		return err;
1045 
1046 	inode = file->f_path.dentry->d_inode;
1047 
1048 	/* Get readahead parameters */
1049 	ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1050 
1051 	if (ra && ra->p_set)
1052 		file->f_ra = ra->p_ra;
1053 
1054 	err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1055 
1056 	/* Write back readahead params */
1057 	if (ra) {
1058 		struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1059 		spin_lock(&rab->pb_lock);
1060 		ra->p_ra = file->f_ra;
1061 		ra->p_set = 1;
1062 		ra->p_count--;
1063 		spin_unlock(&rab->pb_lock);
1064 	}
1065 
1066 	nfsd_close(file);
1067 	return err;
1068 }
1069 
1070 /* As above, but use the provided file descriptor. */
1071 __be32
1072 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1073 		loff_t offset, struct kvec *vec, int vlen,
1074 		unsigned long *count)
1075 {
1076 	__be32		err;
1077 
1078 	if (file) {
1079 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1080 				NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1081 		if (err)
1082 			goto out;
1083 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1084 	} else /* Note file may still be NULL in NFSv4 special stateid case: */
1085 		err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1086 out:
1087 	return err;
1088 }
1089 
1090 /*
1091  * Write data to a file.
1092  * The stable flag requests synchronous writes.
1093  * N.B. After this call fhp needs an fh_put
1094  */
1095 __be32
1096 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1097 		loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1098 		int *stablep)
1099 {
1100 	__be32			err = 0;
1101 
1102 	if (file) {
1103 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1104 				NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1105 		if (err)
1106 			goto out;
1107 		err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1108 				stablep);
1109 	} else {
1110 		err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1111 		if (err)
1112 			goto out;
1113 
1114 		if (cnt)
1115 			err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1116 					     cnt, stablep);
1117 		nfsd_close(file);
1118 	}
1119 out:
1120 	return err;
1121 }
1122 
1123 #ifdef CONFIG_NFSD_V3
1124 /*
1125  * Commit all pending writes to stable storage.
1126  *
1127  * Note: we only guarantee that data that lies within the range specified
1128  * by the 'offset' and 'count' parameters will be synced.
1129  *
1130  * Unfortunately we cannot lock the file to make sure we return full WCC
1131  * data to the client, as locking happens lower down in the filesystem.
1132  */
1133 __be32
1134 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1135                loff_t offset, unsigned long count)
1136 {
1137 	struct file	*file;
1138 	loff_t		end = LLONG_MAX;
1139 	__be32		err = nfserr_inval;
1140 
1141 	if (offset < 0)
1142 		goto out;
1143 	if (count != 0) {
1144 		end = offset + (loff_t)count - 1;
1145 		if (end < offset)
1146 			goto out;
1147 	}
1148 
1149 	err = nfsd_open(rqstp, fhp, S_IFREG,
1150 			NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1151 	if (err)
1152 		goto out;
1153 	if (EX_ISSYNC(fhp->fh_export)) {
1154 		int err2 = vfs_fsync_range(file, offset, end, 0);
1155 
1156 		if (err2 != -EINVAL)
1157 			err = nfserrno(err2);
1158 		else
1159 			err = nfserr_notsupp;
1160 	}
1161 
1162 	nfsd_close(file);
1163 out:
1164 	return err;
1165 }
1166 #endif /* CONFIG_NFSD_V3 */
1167 
1168 static __be32
1169 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1170 			struct iattr *iap)
1171 {
1172 	/*
1173 	 * Mode has already been set earlier in create:
1174 	 */
1175 	iap->ia_valid &= ~ATTR_MODE;
1176 	/*
1177 	 * Setting uid/gid works only for root.  Irix appears to
1178 	 * send along the gid on create when it tries to implement
1179 	 * setgid directories via NFS:
1180 	 */
1181 	if (current_fsuid() != 0)
1182 		iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1183 	if (iap->ia_valid)
1184 		return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1185 	return 0;
1186 }
1187 
1188 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1189  * setting size to 0 may fail for some specific file systems by the permission
1190  * checking which requires WRITE permission but the mode is 000.
1191  * we ignore the resizing(to 0) on the just new created file, since the size is
1192  * 0 after file created.
1193  *
1194  * call this only after vfs_create() is called.
1195  * */
1196 static void
1197 nfsd_check_ignore_resizing(struct iattr *iap)
1198 {
1199 	if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1200 		iap->ia_valid &= ~ATTR_SIZE;
1201 }
1202 
1203 /*
1204  * Create a file (regular, directory, device, fifo); UNIX sockets
1205  * not yet implemented.
1206  * If the response fh has been verified, the parent directory should
1207  * already be locked. Note that the parent directory is left locked.
1208  *
1209  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1210  */
1211 __be32
1212 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1213 		char *fname, int flen, struct iattr *iap,
1214 		int type, dev_t rdev, struct svc_fh *resfhp)
1215 {
1216 	struct dentry	*dentry, *dchild = NULL;
1217 	struct inode	*dirp;
1218 	__be32		err;
1219 	__be32		err2;
1220 	int		host_err;
1221 
1222 	err = nfserr_perm;
1223 	if (!flen)
1224 		goto out;
1225 	err = nfserr_exist;
1226 	if (isdotent(fname, flen))
1227 		goto out;
1228 
1229 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1230 	if (err)
1231 		goto out;
1232 
1233 	dentry = fhp->fh_dentry;
1234 	dirp = dentry->d_inode;
1235 
1236 	err = nfserr_notdir;
1237 	if (!dirp->i_op->lookup)
1238 		goto out;
1239 	/*
1240 	 * Check whether the response file handle has been verified yet.
1241 	 * If it has, the parent directory should already be locked.
1242 	 */
1243 	if (!resfhp->fh_dentry) {
1244 		/* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1245 		fh_lock_nested(fhp, I_MUTEX_PARENT);
1246 		dchild = lookup_one_len(fname, dentry, flen);
1247 		host_err = PTR_ERR(dchild);
1248 		if (IS_ERR(dchild))
1249 			goto out_nfserr;
1250 		err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1251 		if (err)
1252 			goto out;
1253 	} else {
1254 		/* called from nfsd_proc_create */
1255 		dchild = dget(resfhp->fh_dentry);
1256 		if (!fhp->fh_locked) {
1257 			/* not actually possible */
1258 			printk(KERN_ERR
1259 				"nfsd_create: parent %s/%s not locked!\n",
1260 				dentry->d_parent->d_name.name,
1261 				dentry->d_name.name);
1262 			err = nfserr_io;
1263 			goto out;
1264 		}
1265 	}
1266 	/*
1267 	 * Make sure the child dentry is still negative ...
1268 	 */
1269 	err = nfserr_exist;
1270 	if (dchild->d_inode) {
1271 		dprintk("nfsd_create: dentry %s/%s not negative!\n",
1272 			dentry->d_name.name, dchild->d_name.name);
1273 		goto out;
1274 	}
1275 
1276 	if (!(iap->ia_valid & ATTR_MODE))
1277 		iap->ia_mode = 0;
1278 	iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1279 
1280 	err = nfserr_inval;
1281 	if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1282 		printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1283 		       type);
1284 		goto out;
1285 	}
1286 
1287 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1288 	if (host_err)
1289 		goto out_nfserr;
1290 
1291 	/*
1292 	 * Get the dir op function pointer.
1293 	 */
1294 	err = 0;
1295 	switch (type) {
1296 	case S_IFREG:
1297 		host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1298 		if (!host_err)
1299 			nfsd_check_ignore_resizing(iap);
1300 		break;
1301 	case S_IFDIR:
1302 		host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1303 		break;
1304 	case S_IFCHR:
1305 	case S_IFBLK:
1306 	case S_IFIFO:
1307 	case S_IFSOCK:
1308 		host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1309 		break;
1310 	}
1311 	if (host_err < 0) {
1312 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1313 		goto out_nfserr;
1314 	}
1315 
1316 	err = nfsd_create_setattr(rqstp, resfhp, iap);
1317 
1318 	/*
1319 	 * nfsd_setattr already committed the child.  Transactional filesystems
1320 	 * had a chance to commit changes for both parent and child
1321 	 * simultaneously making the following commit_metadata a noop.
1322 	 */
1323 	err2 = nfserrno(commit_metadata(fhp));
1324 	if (err2)
1325 		err = err2;
1326 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1327 	/*
1328 	 * Update the file handle to get the new inode info.
1329 	 */
1330 	if (!err)
1331 		err = fh_update(resfhp);
1332 out:
1333 	if (dchild && !IS_ERR(dchild))
1334 		dput(dchild);
1335 	return err;
1336 
1337 out_nfserr:
1338 	err = nfserrno(host_err);
1339 	goto out;
1340 }
1341 
1342 #ifdef CONFIG_NFSD_V3
1343 /*
1344  * NFSv3 version of nfsd_create
1345  */
1346 __be32
1347 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1348 		char *fname, int flen, struct iattr *iap,
1349 		struct svc_fh *resfhp, int createmode, u32 *verifier,
1350 	        int *truncp, int *created)
1351 {
1352 	struct dentry	*dentry, *dchild = NULL;
1353 	struct inode	*dirp;
1354 	__be32		err;
1355 	int		host_err;
1356 	__u32		v_mtime=0, v_atime=0;
1357 
1358 	err = nfserr_perm;
1359 	if (!flen)
1360 		goto out;
1361 	err = nfserr_exist;
1362 	if (isdotent(fname, flen))
1363 		goto out;
1364 	if (!(iap->ia_valid & ATTR_MODE))
1365 		iap->ia_mode = 0;
1366 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1367 	if (err)
1368 		goto out;
1369 
1370 	dentry = fhp->fh_dentry;
1371 	dirp = dentry->d_inode;
1372 
1373 	/* Get all the sanity checks out of the way before
1374 	 * we lock the parent. */
1375 	err = nfserr_notdir;
1376 	if (!dirp->i_op->lookup)
1377 		goto out;
1378 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1379 
1380 	/*
1381 	 * Compose the response file handle.
1382 	 */
1383 	dchild = lookup_one_len(fname, dentry, flen);
1384 	host_err = PTR_ERR(dchild);
1385 	if (IS_ERR(dchild))
1386 		goto out_nfserr;
1387 
1388 	err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1389 	if (err)
1390 		goto out;
1391 
1392 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1393 		/* solaris7 gets confused (bugid 4218508) if these have
1394 		 * the high bit set, so just clear the high bits. If this is
1395 		 * ever changed to use different attrs for storing the
1396 		 * verifier, then do_open_lookup() will also need to be fixed
1397 		 * accordingly.
1398 		 */
1399 		v_mtime = verifier[0]&0x7fffffff;
1400 		v_atime = verifier[1]&0x7fffffff;
1401 	}
1402 
1403 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1404 	if (host_err)
1405 		goto out_nfserr;
1406 	if (dchild->d_inode) {
1407 		err = 0;
1408 
1409 		switch (createmode) {
1410 		case NFS3_CREATE_UNCHECKED:
1411 			if (! S_ISREG(dchild->d_inode->i_mode))
1412 				err = nfserr_exist;
1413 			else if (truncp) {
1414 				/* in nfsv4, we need to treat this case a little
1415 				 * differently.  we don't want to truncate the
1416 				 * file now; this would be wrong if the OPEN
1417 				 * fails for some other reason.  furthermore,
1418 				 * if the size is nonzero, we should ignore it
1419 				 * according to spec!
1420 				 */
1421 				*truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1422 			}
1423 			else {
1424 				iap->ia_valid &= ATTR_SIZE;
1425 				goto set_attr;
1426 			}
1427 			break;
1428 		case NFS3_CREATE_EXCLUSIVE:
1429 			if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1430 			    && dchild->d_inode->i_atime.tv_sec == v_atime
1431 			    && dchild->d_inode->i_size  == 0 )
1432 				break;
1433 			 /* fallthru */
1434 		case NFS3_CREATE_GUARDED:
1435 			err = nfserr_exist;
1436 		}
1437 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1438 		goto out;
1439 	}
1440 
1441 	host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1442 	if (host_err < 0) {
1443 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1444 		goto out_nfserr;
1445 	}
1446 	if (created)
1447 		*created = 1;
1448 
1449 	nfsd_check_ignore_resizing(iap);
1450 
1451 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1452 		/* Cram the verifier into atime/mtime */
1453 		iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1454 			| ATTR_MTIME_SET|ATTR_ATIME_SET;
1455 		/* XXX someone who knows this better please fix it for nsec */
1456 		iap->ia_mtime.tv_sec = v_mtime;
1457 		iap->ia_atime.tv_sec = v_atime;
1458 		iap->ia_mtime.tv_nsec = 0;
1459 		iap->ia_atime.tv_nsec = 0;
1460 	}
1461 
1462  set_attr:
1463 	err = nfsd_create_setattr(rqstp, resfhp, iap);
1464 
1465 	/*
1466 	 * nfsd_setattr already committed the child (and possibly also the parent).
1467 	 */
1468 	if (!err)
1469 		err = nfserrno(commit_metadata(fhp));
1470 
1471 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1472 	/*
1473 	 * Update the filehandle to get the new inode info.
1474 	 */
1475 	if (!err)
1476 		err = fh_update(resfhp);
1477 
1478  out:
1479 	fh_unlock(fhp);
1480 	if (dchild && !IS_ERR(dchild))
1481 		dput(dchild);
1482  	return err;
1483 
1484  out_nfserr:
1485 	err = nfserrno(host_err);
1486 	goto out;
1487 }
1488 #endif /* CONFIG_NFSD_V3 */
1489 
1490 /*
1491  * Read a symlink. On entry, *lenp must contain the maximum path length that
1492  * fits into the buffer. On return, it contains the true length.
1493  * N.B. After this call fhp needs an fh_put
1494  */
1495 __be32
1496 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1497 {
1498 	struct dentry	*dentry;
1499 	struct inode	*inode;
1500 	mm_segment_t	oldfs;
1501 	__be32		err;
1502 	int		host_err;
1503 
1504 	err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1505 	if (err)
1506 		goto out;
1507 
1508 	dentry = fhp->fh_dentry;
1509 	inode = dentry->d_inode;
1510 
1511 	err = nfserr_inval;
1512 	if (!inode->i_op->readlink)
1513 		goto out;
1514 
1515 	touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1516 	/* N.B. Why does this call need a get_fs()??
1517 	 * Remove the set_fs and watch the fireworks:-) --okir
1518 	 */
1519 
1520 	oldfs = get_fs(); set_fs(KERNEL_DS);
1521 	host_err = inode->i_op->readlink(dentry, buf, *lenp);
1522 	set_fs(oldfs);
1523 
1524 	if (host_err < 0)
1525 		goto out_nfserr;
1526 	*lenp = host_err;
1527 	err = 0;
1528 out:
1529 	return err;
1530 
1531 out_nfserr:
1532 	err = nfserrno(host_err);
1533 	goto out;
1534 }
1535 
1536 /*
1537  * Create a symlink and look up its inode
1538  * N.B. After this call _both_ fhp and resfhp need an fh_put
1539  */
1540 __be32
1541 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1542 				char *fname, int flen,
1543 				char *path,  int plen,
1544 				struct svc_fh *resfhp,
1545 				struct iattr *iap)
1546 {
1547 	struct dentry	*dentry, *dnew;
1548 	__be32		err, cerr;
1549 	int		host_err;
1550 
1551 	err = nfserr_noent;
1552 	if (!flen || !plen)
1553 		goto out;
1554 	err = nfserr_exist;
1555 	if (isdotent(fname, flen))
1556 		goto out;
1557 
1558 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1559 	if (err)
1560 		goto out;
1561 	fh_lock(fhp);
1562 	dentry = fhp->fh_dentry;
1563 	dnew = lookup_one_len(fname, dentry, flen);
1564 	host_err = PTR_ERR(dnew);
1565 	if (IS_ERR(dnew))
1566 		goto out_nfserr;
1567 
1568 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1569 	if (host_err)
1570 		goto out_nfserr;
1571 
1572 	if (unlikely(path[plen] != 0)) {
1573 		char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1574 		if (path_alloced == NULL)
1575 			host_err = -ENOMEM;
1576 		else {
1577 			strncpy(path_alloced, path, plen);
1578 			path_alloced[plen] = 0;
1579 			host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1580 			kfree(path_alloced);
1581 		}
1582 	} else
1583 		host_err = vfs_symlink(dentry->d_inode, dnew, path);
1584 	err = nfserrno(host_err);
1585 	if (!err)
1586 		err = nfserrno(commit_metadata(fhp));
1587 	fh_unlock(fhp);
1588 
1589 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1590 
1591 	cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1592 	dput(dnew);
1593 	if (err==0) err = cerr;
1594 out:
1595 	return err;
1596 
1597 out_nfserr:
1598 	err = nfserrno(host_err);
1599 	goto out;
1600 }
1601 
1602 /*
1603  * Create a hardlink
1604  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1605  */
1606 __be32
1607 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1608 				char *name, int len, struct svc_fh *tfhp)
1609 {
1610 	struct dentry	*ddir, *dnew, *dold;
1611 	struct inode	*dirp;
1612 	__be32		err;
1613 	int		host_err;
1614 
1615 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1616 	if (err)
1617 		goto out;
1618 	err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1619 	if (err)
1620 		goto out;
1621 
1622 	err = nfserr_perm;
1623 	if (!len)
1624 		goto out;
1625 	err = nfserr_exist;
1626 	if (isdotent(name, len))
1627 		goto out;
1628 
1629 	fh_lock_nested(ffhp, I_MUTEX_PARENT);
1630 	ddir = ffhp->fh_dentry;
1631 	dirp = ddir->d_inode;
1632 
1633 	dnew = lookup_one_len(name, ddir, len);
1634 	host_err = PTR_ERR(dnew);
1635 	if (IS_ERR(dnew))
1636 		goto out_nfserr;
1637 
1638 	dold = tfhp->fh_dentry;
1639 
1640 	host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1641 	if (host_err) {
1642 		err = nfserrno(host_err);
1643 		goto out_dput;
1644 	}
1645 	err = nfserr_noent;
1646 	if (!dold->d_inode)
1647 		goto out_drop_write;
1648 	host_err = nfsd_break_lease(dold->d_inode);
1649 	if (host_err)
1650 		goto out_drop_write;
1651 	host_err = vfs_link(dold, dirp, dnew);
1652 	if (!host_err) {
1653 		err = nfserrno(commit_metadata(ffhp));
1654 		if (!err)
1655 			err = nfserrno(commit_metadata(tfhp));
1656 	} else {
1657 		if (host_err == -EXDEV && rqstp->rq_vers == 2)
1658 			err = nfserr_acces;
1659 		else
1660 			err = nfserrno(host_err);
1661 	}
1662 out_drop_write:
1663 	mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1664 out_dput:
1665 	dput(dnew);
1666 out_unlock:
1667 	fh_unlock(ffhp);
1668 out:
1669 	return err;
1670 
1671 out_nfserr:
1672 	err = nfserrno(host_err);
1673 	goto out_unlock;
1674 }
1675 
1676 /*
1677  * Rename a file
1678  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1679  */
1680 __be32
1681 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1682 			    struct svc_fh *tfhp, char *tname, int tlen)
1683 {
1684 	struct dentry	*fdentry, *tdentry, *odentry, *ndentry, *trap;
1685 	struct inode	*fdir, *tdir;
1686 	__be32		err;
1687 	int		host_err;
1688 
1689 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1690 	if (err)
1691 		goto out;
1692 	err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1693 	if (err)
1694 		goto out;
1695 
1696 	fdentry = ffhp->fh_dentry;
1697 	fdir = fdentry->d_inode;
1698 
1699 	tdentry = tfhp->fh_dentry;
1700 	tdir = tdentry->d_inode;
1701 
1702 	err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1703 	if (ffhp->fh_export != tfhp->fh_export)
1704 		goto out;
1705 
1706 	err = nfserr_perm;
1707 	if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1708 		goto out;
1709 
1710 	/* cannot use fh_lock as we need deadlock protective ordering
1711 	 * so do it by hand */
1712 	trap = lock_rename(tdentry, fdentry);
1713 	ffhp->fh_locked = tfhp->fh_locked = 1;
1714 	fill_pre_wcc(ffhp);
1715 	fill_pre_wcc(tfhp);
1716 
1717 	odentry = lookup_one_len(fname, fdentry, flen);
1718 	host_err = PTR_ERR(odentry);
1719 	if (IS_ERR(odentry))
1720 		goto out_nfserr;
1721 
1722 	host_err = -ENOENT;
1723 	if (!odentry->d_inode)
1724 		goto out_dput_old;
1725 	host_err = -EINVAL;
1726 	if (odentry == trap)
1727 		goto out_dput_old;
1728 
1729 	ndentry = lookup_one_len(tname, tdentry, tlen);
1730 	host_err = PTR_ERR(ndentry);
1731 	if (IS_ERR(ndentry))
1732 		goto out_dput_old;
1733 	host_err = -ENOTEMPTY;
1734 	if (ndentry == trap)
1735 		goto out_dput_new;
1736 
1737 	host_err = -EXDEV;
1738 	if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1739 		goto out_dput_new;
1740 	host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1741 	if (host_err)
1742 		goto out_dput_new;
1743 
1744 	host_err = nfsd_break_lease(odentry->d_inode);
1745 	if (host_err)
1746 		goto out_drop_write;
1747 	host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1748 	if (!host_err) {
1749 		host_err = commit_metadata(tfhp);
1750 		if (!host_err)
1751 			host_err = commit_metadata(ffhp);
1752 	}
1753 out_drop_write:
1754 	mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1755  out_dput_new:
1756 	dput(ndentry);
1757  out_dput_old:
1758 	dput(odentry);
1759  out_nfserr:
1760 	err = nfserrno(host_err);
1761 
1762 	/* we cannot reply on fh_unlock on the two filehandles,
1763 	 * as that would do the wrong thing if the two directories
1764 	 * were the same, so again we do it by hand
1765 	 */
1766 	fill_post_wcc(ffhp);
1767 	fill_post_wcc(tfhp);
1768 	unlock_rename(tdentry, fdentry);
1769 	ffhp->fh_locked = tfhp->fh_locked = 0;
1770 
1771 out:
1772 	return err;
1773 }
1774 
1775 /*
1776  * Unlink a file or directory
1777  * N.B. After this call fhp needs an fh_put
1778  */
1779 __be32
1780 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1781 				char *fname, int flen)
1782 {
1783 	struct dentry	*dentry, *rdentry;
1784 	struct inode	*dirp;
1785 	__be32		err;
1786 	int		host_err;
1787 
1788 	err = nfserr_acces;
1789 	if (!flen || isdotent(fname, flen))
1790 		goto out;
1791 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1792 	if (err)
1793 		goto out;
1794 
1795 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1796 	dentry = fhp->fh_dentry;
1797 	dirp = dentry->d_inode;
1798 
1799 	rdentry = lookup_one_len(fname, dentry, flen);
1800 	host_err = PTR_ERR(rdentry);
1801 	if (IS_ERR(rdentry))
1802 		goto out_nfserr;
1803 
1804 	if (!rdentry->d_inode) {
1805 		dput(rdentry);
1806 		err = nfserr_noent;
1807 		goto out;
1808 	}
1809 
1810 	if (!type)
1811 		type = rdentry->d_inode->i_mode & S_IFMT;
1812 
1813 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1814 	if (host_err)
1815 		goto out_nfserr;
1816 
1817 	host_err = nfsd_break_lease(rdentry->d_inode);
1818 	if (host_err)
1819 		goto out_put;
1820 	if (type != S_IFDIR)
1821 		host_err = vfs_unlink(dirp, rdentry);
1822 	else
1823 		host_err = vfs_rmdir(dirp, rdentry);
1824 out_put:
1825 	dput(rdentry);
1826 
1827 	if (!host_err)
1828 		host_err = commit_metadata(fhp);
1829 
1830 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1831 out_nfserr:
1832 	err = nfserrno(host_err);
1833 out:
1834 	return err;
1835 }
1836 
1837 /*
1838  * We do this buffering because we must not call back into the file
1839  * system's ->lookup() method from the filldir callback. That may well
1840  * deadlock a number of file systems.
1841  *
1842  * This is based heavily on the implementation of same in XFS.
1843  */
1844 struct buffered_dirent {
1845 	u64		ino;
1846 	loff_t		offset;
1847 	int		namlen;
1848 	unsigned int	d_type;
1849 	char		name[];
1850 };
1851 
1852 struct readdir_data {
1853 	char		*dirent;
1854 	size_t		used;
1855 	int		full;
1856 };
1857 
1858 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1859 				 loff_t offset, u64 ino, unsigned int d_type)
1860 {
1861 	struct readdir_data *buf = __buf;
1862 	struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1863 	unsigned int reclen;
1864 
1865 	reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1866 	if (buf->used + reclen > PAGE_SIZE) {
1867 		buf->full = 1;
1868 		return -EINVAL;
1869 	}
1870 
1871 	de->namlen = namlen;
1872 	de->offset = offset;
1873 	de->ino = ino;
1874 	de->d_type = d_type;
1875 	memcpy(de->name, name, namlen);
1876 	buf->used += reclen;
1877 
1878 	return 0;
1879 }
1880 
1881 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1882 				    struct readdir_cd *cdp, loff_t *offsetp)
1883 {
1884 	struct readdir_data buf;
1885 	struct buffered_dirent *de;
1886 	int host_err;
1887 	int size;
1888 	loff_t offset;
1889 
1890 	buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1891 	if (!buf.dirent)
1892 		return nfserrno(-ENOMEM);
1893 
1894 	offset = *offsetp;
1895 
1896 	while (1) {
1897 		struct inode *dir_inode = file->f_path.dentry->d_inode;
1898 		unsigned int reclen;
1899 
1900 		cdp->err = nfserr_eof; /* will be cleared on successful read */
1901 		buf.used = 0;
1902 		buf.full = 0;
1903 
1904 		host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1905 		if (buf.full)
1906 			host_err = 0;
1907 
1908 		if (host_err < 0)
1909 			break;
1910 
1911 		size = buf.used;
1912 
1913 		if (!size)
1914 			break;
1915 
1916 		/*
1917 		 * Various filldir functions may end up calling back into
1918 		 * lookup_one_len() and the file system's ->lookup() method.
1919 		 * These expect i_mutex to be held, as it would within readdir.
1920 		 */
1921 		host_err = mutex_lock_killable(&dir_inode->i_mutex);
1922 		if (host_err)
1923 			break;
1924 
1925 		de = (struct buffered_dirent *)buf.dirent;
1926 		while (size > 0) {
1927 			offset = de->offset;
1928 
1929 			if (func(cdp, de->name, de->namlen, de->offset,
1930 				 de->ino, de->d_type))
1931 				break;
1932 
1933 			if (cdp->err != nfs_ok)
1934 				break;
1935 
1936 			reclen = ALIGN(sizeof(*de) + de->namlen,
1937 				       sizeof(u64));
1938 			size -= reclen;
1939 			de = (struct buffered_dirent *)((char *)de + reclen);
1940 		}
1941 		mutex_unlock(&dir_inode->i_mutex);
1942 		if (size > 0) /* We bailed out early */
1943 			break;
1944 
1945 		offset = vfs_llseek(file, 0, SEEK_CUR);
1946 	}
1947 
1948 	free_page((unsigned long)(buf.dirent));
1949 
1950 	if (host_err)
1951 		return nfserrno(host_err);
1952 
1953 	*offsetp = offset;
1954 	return cdp->err;
1955 }
1956 
1957 /*
1958  * Read entries from a directory.
1959  * The  NFSv3/4 verifier we ignore for now.
1960  */
1961 __be32
1962 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1963 	     struct readdir_cd *cdp, filldir_t func)
1964 {
1965 	__be32		err;
1966 	struct file	*file;
1967 	loff_t		offset = *offsetp;
1968 
1969 	err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1970 	if (err)
1971 		goto out;
1972 
1973 	offset = vfs_llseek(file, offset, 0);
1974 	if (offset < 0) {
1975 		err = nfserrno((int)offset);
1976 		goto out_close;
1977 	}
1978 
1979 	err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1980 
1981 	if (err == nfserr_eof || err == nfserr_toosmall)
1982 		err = nfs_ok; /* can still be found in ->err */
1983 out_close:
1984 	nfsd_close(file);
1985 out:
1986 	return err;
1987 }
1988 
1989 /*
1990  * Get file system stats
1991  * N.B. After this call fhp needs an fh_put
1992  */
1993 __be32
1994 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1995 {
1996 	__be32 err;
1997 
1998 	err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1999 	if (!err) {
2000 		struct path path = {
2001 			.mnt	= fhp->fh_export->ex_path.mnt,
2002 			.dentry	= fhp->fh_dentry,
2003 		};
2004 		if (vfs_statfs(&path, stat))
2005 			err = nfserr_io;
2006 	}
2007 	return err;
2008 }
2009 
2010 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2011 {
2012 	return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2013 }
2014 
2015 /*
2016  * Check for a user's access permissions to this inode.
2017  */
2018 __be32
2019 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2020 					struct dentry *dentry, int acc)
2021 {
2022 	struct inode	*inode = dentry->d_inode;
2023 	int		err;
2024 
2025 	if (acc == NFSD_MAY_NOP)
2026 		return 0;
2027 #if 0
2028 	dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2029 		acc,
2030 		(acc & NFSD_MAY_READ)?	" read"  : "",
2031 		(acc & NFSD_MAY_WRITE)?	" write" : "",
2032 		(acc & NFSD_MAY_EXEC)?	" exec"  : "",
2033 		(acc & NFSD_MAY_SATTR)?	" sattr" : "",
2034 		(acc & NFSD_MAY_TRUNC)?	" trunc" : "",
2035 		(acc & NFSD_MAY_LOCK)?	" lock"  : "",
2036 		(acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2037 		inode->i_mode,
2038 		IS_IMMUTABLE(inode)?	" immut" : "",
2039 		IS_APPEND(inode)?	" append" : "",
2040 		__mnt_is_readonly(exp->ex_path.mnt)?	" ro" : "");
2041 	dprintk("      owner %d/%d user %d/%d\n",
2042 		inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2043 #endif
2044 
2045 	/* Normally we reject any write/sattr etc access on a read-only file
2046 	 * system.  But if it is IRIX doing check on write-access for a
2047 	 * device special file, we ignore rofs.
2048 	 */
2049 	if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2050 		if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2051 			if (exp_rdonly(rqstp, exp) ||
2052 			    __mnt_is_readonly(exp->ex_path.mnt))
2053 				return nfserr_rofs;
2054 			if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2055 				return nfserr_perm;
2056 		}
2057 	if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2058 		return nfserr_perm;
2059 
2060 	if (acc & NFSD_MAY_LOCK) {
2061 		/* If we cannot rely on authentication in NLM requests,
2062 		 * just allow locks, otherwise require read permission, or
2063 		 * ownership
2064 		 */
2065 		if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2066 			return 0;
2067 		else
2068 			acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2069 	}
2070 	/*
2071 	 * The file owner always gets access permission for accesses that
2072 	 * would normally be checked at open time. This is to make
2073 	 * file access work even when the client has done a fchmod(fd, 0).
2074 	 *
2075 	 * However, `cp foo bar' should fail nevertheless when bar is
2076 	 * readonly. A sensible way to do this might be to reject all
2077 	 * attempts to truncate a read-only file, because a creat() call
2078 	 * always implies file truncation.
2079 	 * ... but this isn't really fair.  A process may reasonably call
2080 	 * ftruncate on an open file descriptor on a file with perm 000.
2081 	 * We must trust the client to do permission checking - using "ACCESS"
2082 	 * with NFSv3.
2083 	 */
2084 	if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2085 	    inode->i_uid == current_fsuid())
2086 		return 0;
2087 
2088 	/* This assumes  NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2089 	err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2090 
2091 	/* Allow read access to binaries even when mode 111 */
2092 	if (err == -EACCES && S_ISREG(inode->i_mode) &&
2093 	    acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2094 		err = inode_permission(inode, MAY_EXEC);
2095 
2096 	return err? nfserrno(err) : 0;
2097 }
2098 
2099 void
2100 nfsd_racache_shutdown(void)
2101 {
2102 	struct raparms *raparm, *last_raparm;
2103 	unsigned int i;
2104 
2105 	dprintk("nfsd: freeing readahead buffers.\n");
2106 
2107 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2108 		raparm = raparm_hash[i].pb_head;
2109 		while(raparm) {
2110 			last_raparm = raparm;
2111 			raparm = raparm->p_next;
2112 			kfree(last_raparm);
2113 		}
2114 		raparm_hash[i].pb_head = NULL;
2115 	}
2116 }
2117 /*
2118  * Initialize readahead param cache
2119  */
2120 int
2121 nfsd_racache_init(int cache_size)
2122 {
2123 	int	i;
2124 	int	j = 0;
2125 	int	nperbucket;
2126 	struct raparms **raparm = NULL;
2127 
2128 
2129 	if (raparm_hash[0].pb_head)
2130 		return 0;
2131 	nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2132 	if (nperbucket < 2)
2133 		nperbucket = 2;
2134 	cache_size = nperbucket * RAPARM_HASH_SIZE;
2135 
2136 	dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2137 
2138 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2139 		spin_lock_init(&raparm_hash[i].pb_lock);
2140 
2141 		raparm = &raparm_hash[i].pb_head;
2142 		for (j = 0; j < nperbucket; j++) {
2143 			*raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2144 			if (!*raparm)
2145 				goto out_nomem;
2146 			raparm = &(*raparm)->p_next;
2147 		}
2148 		*raparm = NULL;
2149 	}
2150 
2151 	nfsdstats.ra_size = cache_size;
2152 	return 0;
2153 
2154 out_nomem:
2155 	dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2156 	nfsd_racache_shutdown();
2157 	return -ENOMEM;
2158 }
2159 
2160 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2161 struct posix_acl *
2162 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2163 {
2164 	struct inode *inode = fhp->fh_dentry->d_inode;
2165 	char *name;
2166 	void *value = NULL;
2167 	ssize_t size;
2168 	struct posix_acl *acl;
2169 
2170 	if (!IS_POSIXACL(inode))
2171 		return ERR_PTR(-EOPNOTSUPP);
2172 
2173 	switch (type) {
2174 	case ACL_TYPE_ACCESS:
2175 		name = POSIX_ACL_XATTR_ACCESS;
2176 		break;
2177 	case ACL_TYPE_DEFAULT:
2178 		name = POSIX_ACL_XATTR_DEFAULT;
2179 		break;
2180 	default:
2181 		return ERR_PTR(-EOPNOTSUPP);
2182 	}
2183 
2184 	size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2185 	if (size < 0)
2186 		return ERR_PTR(size);
2187 
2188 	acl = posix_acl_from_xattr(value, size);
2189 	kfree(value);
2190 	return acl;
2191 }
2192 
2193 int
2194 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2195 {
2196 	struct inode *inode = fhp->fh_dentry->d_inode;
2197 	char *name;
2198 	void *value = NULL;
2199 	size_t size;
2200 	int error;
2201 
2202 	if (!IS_POSIXACL(inode) ||
2203 	    !inode->i_op->setxattr || !inode->i_op->removexattr)
2204 		return -EOPNOTSUPP;
2205 	switch(type) {
2206 		case ACL_TYPE_ACCESS:
2207 			name = POSIX_ACL_XATTR_ACCESS;
2208 			break;
2209 		case ACL_TYPE_DEFAULT:
2210 			name = POSIX_ACL_XATTR_DEFAULT;
2211 			break;
2212 		default:
2213 			return -EOPNOTSUPP;
2214 	}
2215 
2216 	if (acl && acl->a_count) {
2217 		size = posix_acl_xattr_size(acl->a_count);
2218 		value = kmalloc(size, GFP_KERNEL);
2219 		if (!value)
2220 			return -ENOMEM;
2221 		error = posix_acl_to_xattr(acl, value, size);
2222 		if (error < 0)
2223 			goto getout;
2224 		size = error;
2225 	} else
2226 		size = 0;
2227 
2228 	error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2229 	if (error)
2230 		goto getout;
2231 	if (size)
2232 		error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2233 	else {
2234 		if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2235 			error = 0;
2236 		else {
2237 			error = vfs_removexattr(fhp->fh_dentry, name);
2238 			if (error == -ENODATA)
2239 				error = 0;
2240 		}
2241 	}
2242 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
2243 
2244 getout:
2245 	kfree(value);
2246 	return error;
2247 }
2248 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
2249