xref: /linux/fs/nfsd/vfs.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
1 #define MSNFS	/* HACK HACK */
2 /*
3  * File operations used by nfsd. Some of these have been ripped from
4  * other parts of the kernel because they weren't exported, others
5  * are partial duplicates with added or changed functionality.
6  *
7  * Note that several functions dget() the dentry upon which they want
8  * to act, most notably those that create directory entries. Response
9  * dentry's are dput()'d if necessary in the release callback.
10  * So if you notice code paths that apparently fail to dput() the
11  * dentry, don't worry--they have been taken care of.
12  *
13  * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14  * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
15  */
16 
17 #include <linux/fs.h>
18 #include <linux/file.h>
19 #include <linux/splice.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <linux/slab.h>
29 #include <asm/uaccess.h>
30 #include <linux/exportfs.h>
31 #include <linux/writeback.h>
32 
33 #ifdef CONFIG_NFSD_V3
34 #include "xdr3.h"
35 #endif /* CONFIG_NFSD_V3 */
36 
37 #ifdef CONFIG_NFSD_V4
38 #include <linux/nfs4_acl.h>
39 #include <linux/nfsd_idmap.h>
40 #endif /* CONFIG_NFSD_V4 */
41 
42 #include "nfsd.h"
43 #include "vfs.h"
44 
45 #define NFSDDBG_FACILITY		NFSDDBG_FILEOP
46 
47 
48 /*
49  * This is a cache of readahead params that help us choose the proper
50  * readahead strategy. Initially, we set all readahead parameters to 0
51  * and let the VFS handle things.
52  * If you increase the number of cached files very much, you'll need to
53  * add a hash table here.
54  */
55 struct raparms {
56 	struct raparms		*p_next;
57 	unsigned int		p_count;
58 	ino_t			p_ino;
59 	dev_t			p_dev;
60 	int			p_set;
61 	struct file_ra_state	p_ra;
62 	unsigned int		p_hindex;
63 };
64 
65 struct raparm_hbucket {
66 	struct raparms		*pb_head;
67 	spinlock_t		pb_lock;
68 } ____cacheline_aligned_in_smp;
69 
70 #define RAPARM_HASH_BITS	4
71 #define RAPARM_HASH_SIZE	(1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK	(RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket	raparm_hash[RAPARM_HASH_SIZE];
74 
75 /*
76  * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77  * a mount point.
78  * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79  *  or nfs_ok having possibly changed *dpp and *expp
80  */
81 int
82 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
83 		        struct svc_export **expp)
84 {
85 	struct svc_export *exp = *expp, *exp2 = NULL;
86 	struct dentry *dentry = *dpp;
87 	struct path path = {.mnt = mntget(exp->ex_path.mnt),
88 			    .dentry = dget(dentry)};
89 	int err = 0;
90 
91 	while (d_mountpoint(path.dentry) && follow_down(&path))
92 		;
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 /*
277  * Commit metadata changes to stable storage.
278  */
279 static int
280 commit_metadata(struct svc_fh *fhp)
281 {
282 	struct inode *inode = fhp->fh_dentry->d_inode;
283 	const struct export_operations *export_ops = inode->i_sb->s_export_op;
284 	int error = 0;
285 
286 	if (!EX_ISSYNC(fhp->fh_export))
287 		return 0;
288 
289 	if (export_ops->commit_metadata) {
290 		error = export_ops->commit_metadata(inode);
291 	} else {
292 		struct writeback_control wbc = {
293 			.sync_mode = WB_SYNC_ALL,
294 			.nr_to_write = 0, /* metadata only */
295 		};
296 
297 		error = sync_inode(inode, &wbc);
298 	}
299 
300 	return error;
301 }
302 
303 /*
304  * Set various file attributes.
305  * N.B. After this call fhp needs an fh_put
306  */
307 __be32
308 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
309 	     int check_guard, time_t guardtime)
310 {
311 	struct dentry	*dentry;
312 	struct inode	*inode;
313 	int		accmode = NFSD_MAY_SATTR;
314 	int		ftype = 0;
315 	__be32		err;
316 	int		host_err;
317 	int		size_change = 0;
318 
319 	if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
320 		accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
321 	if (iap->ia_valid & ATTR_SIZE)
322 		ftype = S_IFREG;
323 
324 	/* Get inode */
325 	err = fh_verify(rqstp, fhp, ftype, accmode);
326 	if (err)
327 		goto out;
328 
329 	dentry = fhp->fh_dentry;
330 	inode = dentry->d_inode;
331 
332 	/* Ignore any mode updates on symlinks */
333 	if (S_ISLNK(inode->i_mode))
334 		iap->ia_valid &= ~ATTR_MODE;
335 
336 	if (!iap->ia_valid)
337 		goto out;
338 
339 	/*
340 	 * NFSv2 does not differentiate between "set-[ac]time-to-now"
341 	 * which only requires access, and "set-[ac]time-to-X" which
342 	 * requires ownership.
343 	 * So if it looks like it might be "set both to the same time which
344 	 * is close to now", and if inode_change_ok fails, then we
345 	 * convert to "set to now" instead of "set to explicit time"
346 	 *
347 	 * We only call inode_change_ok as the last test as technically
348 	 * it is not an interface that we should be using.  It is only
349 	 * valid if the filesystem does not define it's own i_op->setattr.
350 	 */
351 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
352 #define	MAX_TOUCH_TIME_ERROR (30*60)
353 	if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
354 	    iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
355 		/*
356 		 * Looks probable.
357 		 *
358 		 * Now just make sure time is in the right ballpark.
359 		 * Solaris, at least, doesn't seem to care what the time
360 		 * request is.  We require it be within 30 minutes of now.
361 		 */
362 		time_t delta = iap->ia_atime.tv_sec - get_seconds();
363 		if (delta < 0)
364 			delta = -delta;
365 		if (delta < MAX_TOUCH_TIME_ERROR &&
366 		    inode_change_ok(inode, iap) != 0) {
367 			/*
368 			 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
369 			 * This will cause notify_change to set these times
370 			 * to "now"
371 			 */
372 			iap->ia_valid &= ~BOTH_TIME_SET;
373 		}
374 	}
375 
376 	/*
377 	 * The size case is special.
378 	 * It changes the file as well as the attributes.
379 	 */
380 	if (iap->ia_valid & ATTR_SIZE) {
381 		if (iap->ia_size < inode->i_size) {
382 			err = nfsd_permission(rqstp, fhp->fh_export, dentry,
383 					NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
384 			if (err)
385 				goto out;
386 		}
387 
388 		/*
389 		 * If we are changing the size of the file, then
390 		 * we need to break all leases.
391 		 */
392 		host_err = break_lease(inode, O_WRONLY | O_NONBLOCK);
393 		if (host_err == -EWOULDBLOCK)
394 			host_err = -ETIMEDOUT;
395 		if (host_err) /* ENOMEM or EWOULDBLOCK */
396 			goto out_nfserr;
397 
398 		host_err = get_write_access(inode);
399 		if (host_err)
400 			goto out_nfserr;
401 
402 		size_change = 1;
403 		host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
404 		if (host_err) {
405 			put_write_access(inode);
406 			goto out_nfserr;
407 		}
408 	}
409 
410 	/* sanitize the mode change */
411 	if (iap->ia_valid & ATTR_MODE) {
412 		iap->ia_mode &= S_IALLUGO;
413 		iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
414 	}
415 
416 	/* Revoke setuid/setgid on chown */
417 	if (!S_ISDIR(inode->i_mode) &&
418 	    (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
419 	     ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
420 		iap->ia_valid |= ATTR_KILL_PRIV;
421 		if (iap->ia_valid & ATTR_MODE) {
422 			/* we're setting mode too, just clear the s*id bits */
423 			iap->ia_mode &= ~S_ISUID;
424 			if (iap->ia_mode & S_IXGRP)
425 				iap->ia_mode &= ~S_ISGID;
426 		} else {
427 			/* set ATTR_KILL_* bits and let VFS handle it */
428 			iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
429 		}
430 	}
431 
432 	/* Change the attributes. */
433 
434 	iap->ia_valid |= ATTR_CTIME;
435 
436 	err = nfserr_notsync;
437 	if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
438 		fh_lock(fhp);
439 		host_err = notify_change(dentry, iap);
440 		err = nfserrno(host_err);
441 		fh_unlock(fhp);
442 	}
443 	if (size_change)
444 		put_write_access(inode);
445 	if (!err)
446 		commit_metadata(fhp);
447 out:
448 	return err;
449 
450 out_nfserr:
451 	err = nfserrno(host_err);
452 	goto out;
453 }
454 
455 #if defined(CONFIG_NFSD_V2_ACL) || \
456     defined(CONFIG_NFSD_V3_ACL) || \
457     defined(CONFIG_NFSD_V4)
458 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
459 {
460 	ssize_t buflen;
461 	ssize_t ret;
462 
463 	buflen = vfs_getxattr(dentry, key, NULL, 0);
464 	if (buflen <= 0)
465 		return buflen;
466 
467 	*buf = kmalloc(buflen, GFP_KERNEL);
468 	if (!*buf)
469 		return -ENOMEM;
470 
471 	ret = vfs_getxattr(dentry, key, *buf, buflen);
472 	if (ret < 0)
473 		kfree(*buf);
474 	return ret;
475 }
476 #endif
477 
478 #if defined(CONFIG_NFSD_V4)
479 static int
480 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
481 {
482 	int len;
483 	size_t buflen;
484 	char *buf = NULL;
485 	int error = 0;
486 
487 	buflen = posix_acl_xattr_size(pacl->a_count);
488 	buf = kmalloc(buflen, GFP_KERNEL);
489 	error = -ENOMEM;
490 	if (buf == NULL)
491 		goto out;
492 
493 	len = posix_acl_to_xattr(pacl, buf, buflen);
494 	if (len < 0) {
495 		error = len;
496 		goto out;
497 	}
498 
499 	error = vfs_setxattr(dentry, key, buf, len, 0);
500 out:
501 	kfree(buf);
502 	return error;
503 }
504 
505 __be32
506 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
507     struct nfs4_acl *acl)
508 {
509 	__be32 error;
510 	int host_error;
511 	struct dentry *dentry;
512 	struct inode *inode;
513 	struct posix_acl *pacl = NULL, *dpacl = NULL;
514 	unsigned int flags = 0;
515 
516 	/* Get inode */
517 	error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
518 	if (error)
519 		return error;
520 
521 	dentry = fhp->fh_dentry;
522 	inode = dentry->d_inode;
523 	if (S_ISDIR(inode->i_mode))
524 		flags = NFS4_ACL_DIR;
525 
526 	host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
527 	if (host_error == -EINVAL) {
528 		return nfserr_attrnotsupp;
529 	} else if (host_error < 0)
530 		goto out_nfserr;
531 
532 	host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
533 	if (host_error < 0)
534 		goto out_release;
535 
536 	if (S_ISDIR(inode->i_mode))
537 		host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
538 
539 out_release:
540 	posix_acl_release(pacl);
541 	posix_acl_release(dpacl);
542 out_nfserr:
543 	if (host_error == -EOPNOTSUPP)
544 		return nfserr_attrnotsupp;
545 	else
546 		return nfserrno(host_error);
547 }
548 
549 static struct posix_acl *
550 _get_posix_acl(struct dentry *dentry, char *key)
551 {
552 	void *buf = NULL;
553 	struct posix_acl *pacl = NULL;
554 	int buflen;
555 
556 	buflen = nfsd_getxattr(dentry, key, &buf);
557 	if (!buflen)
558 		buflen = -ENODATA;
559 	if (buflen <= 0)
560 		return ERR_PTR(buflen);
561 
562 	pacl = posix_acl_from_xattr(buf, buflen);
563 	kfree(buf);
564 	return pacl;
565 }
566 
567 int
568 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
569 {
570 	struct inode *inode = dentry->d_inode;
571 	int error = 0;
572 	struct posix_acl *pacl = NULL, *dpacl = NULL;
573 	unsigned int flags = 0;
574 
575 	pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
576 	if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
577 		pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
578 	if (IS_ERR(pacl)) {
579 		error = PTR_ERR(pacl);
580 		pacl = NULL;
581 		goto out;
582 	}
583 
584 	if (S_ISDIR(inode->i_mode)) {
585 		dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
586 		if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
587 			dpacl = NULL;
588 		else if (IS_ERR(dpacl)) {
589 			error = PTR_ERR(dpacl);
590 			dpacl = NULL;
591 			goto out;
592 		}
593 		flags = NFS4_ACL_DIR;
594 	}
595 
596 	*acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
597 	if (IS_ERR(*acl)) {
598 		error = PTR_ERR(*acl);
599 		*acl = NULL;
600 	}
601  out:
602 	posix_acl_release(pacl);
603 	posix_acl_release(dpacl);
604 	return error;
605 }
606 
607 #endif /* defined(CONFIG_NFSD_V4) */
608 
609 #ifdef CONFIG_NFSD_V3
610 /*
611  * Check server access rights to a file system object
612  */
613 struct accessmap {
614 	u32		access;
615 	int		how;
616 };
617 static struct accessmap	nfs3_regaccess[] = {
618     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
619     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
620     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_TRUNC	},
621     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE			},
622 
623     {	0,			0				}
624 };
625 
626 static struct accessmap	nfs3_diraccess[] = {
627     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
628     {	NFS3_ACCESS_LOOKUP,	NFSD_MAY_EXEC			},
629     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
630     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_EXEC|NFSD_MAY_WRITE	},
631     {	NFS3_ACCESS_DELETE,	NFSD_MAY_REMOVE			},
632 
633     {	0,			0				}
634 };
635 
636 static struct accessmap	nfs3_anyaccess[] = {
637 	/* Some clients - Solaris 2.6 at least, make an access call
638 	 * to the server to check for access for things like /dev/null
639 	 * (which really, the server doesn't care about).  So
640 	 * We provide simple access checking for them, looking
641 	 * mainly at mode bits, and we make sure to ignore read-only
642 	 * filesystem checks
643 	 */
644     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
645     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
646     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
647     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
648 
649     {	0,			0				}
650 };
651 
652 __be32
653 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
654 {
655 	struct accessmap	*map;
656 	struct svc_export	*export;
657 	struct dentry		*dentry;
658 	u32			query, result = 0, sresult = 0;
659 	__be32			error;
660 
661 	error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
662 	if (error)
663 		goto out;
664 
665 	export = fhp->fh_export;
666 	dentry = fhp->fh_dentry;
667 
668 	if (S_ISREG(dentry->d_inode->i_mode))
669 		map = nfs3_regaccess;
670 	else if (S_ISDIR(dentry->d_inode->i_mode))
671 		map = nfs3_diraccess;
672 	else
673 		map = nfs3_anyaccess;
674 
675 
676 	query = *access;
677 	for  (; map->access; map++) {
678 		if (map->access & query) {
679 			__be32 err2;
680 
681 			sresult |= map->access;
682 
683 			err2 = nfsd_permission(rqstp, export, dentry, map->how);
684 			switch (err2) {
685 			case nfs_ok:
686 				result |= map->access;
687 				break;
688 
689 			/* the following error codes just mean the access was not allowed,
690 			 * rather than an error occurred */
691 			case nfserr_rofs:
692 			case nfserr_acces:
693 			case nfserr_perm:
694 				/* simply don't "or" in the access bit. */
695 				break;
696 			default:
697 				error = err2;
698 				goto out;
699 			}
700 		}
701 	}
702 	*access = result;
703 	if (supported)
704 		*supported = sresult;
705 
706  out:
707 	return error;
708 }
709 #endif /* CONFIG_NFSD_V3 */
710 
711 
712 
713 /*
714  * Open an existing file or directory.
715  * The access argument indicates the type of open (read/write/lock)
716  * N.B. After this call fhp needs an fh_put
717  */
718 __be32
719 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
720 			int access, struct file **filp)
721 {
722 	struct dentry	*dentry;
723 	struct inode	*inode;
724 	int		flags = O_RDONLY|O_LARGEFILE;
725 	__be32		err;
726 	int		host_err = 0;
727 
728 	validate_process_creds();
729 
730 	/*
731 	 * If we get here, then the client has already done an "open",
732 	 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
733 	 * in case a chmod has now revoked permission.
734 	 */
735 	err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
736 	if (err)
737 		goto out;
738 
739 	dentry = fhp->fh_dentry;
740 	inode = dentry->d_inode;
741 
742 	/* Disallow write access to files with the append-only bit set
743 	 * or any access when mandatory locking enabled
744 	 */
745 	err = nfserr_perm;
746 	if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
747 		goto out;
748 	/*
749 	 * We must ignore files (but only files) which might have mandatory
750 	 * locks on them because there is no way to know if the accesser has
751 	 * the lock.
752 	 */
753 	if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
754 		goto out;
755 
756 	if (!inode->i_fop)
757 		goto out;
758 
759 	/*
760 	 * Check to see if there are any leases on this file.
761 	 * This may block while leases are broken.
762 	 */
763 	if (!(access & NFSD_MAY_NOT_BREAK_LEASE))
764 		host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
765 	if (host_err == -EWOULDBLOCK)
766 		host_err = -ETIMEDOUT;
767 	if (host_err) /* NOMEM or WOULDBLOCK */
768 		goto out_nfserr;
769 
770 	if (access & NFSD_MAY_WRITE) {
771 		if (access & NFSD_MAY_READ)
772 			flags = O_RDWR|O_LARGEFILE;
773 		else
774 			flags = O_WRONLY|O_LARGEFILE;
775 	}
776 	*filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
777 			    flags, current_cred());
778 	if (IS_ERR(*filp))
779 		host_err = PTR_ERR(*filp);
780 	else
781 		host_err = ima_file_check(*filp, access);
782 out_nfserr:
783 	err = nfserrno(host_err);
784 out:
785 	validate_process_creds();
786 	return err;
787 }
788 
789 /*
790  * Close a file.
791  */
792 void
793 nfsd_close(struct file *filp)
794 {
795 	fput(filp);
796 }
797 
798 /*
799  * Obtain the readahead parameters for the file
800  * specified by (dev, ino).
801  */
802 
803 static inline struct raparms *
804 nfsd_get_raparms(dev_t dev, ino_t ino)
805 {
806 	struct raparms	*ra, **rap, **frap = NULL;
807 	int depth = 0;
808 	unsigned int hash;
809 	struct raparm_hbucket *rab;
810 
811 	hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
812 	rab = &raparm_hash[hash];
813 
814 	spin_lock(&rab->pb_lock);
815 	for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
816 		if (ra->p_ino == ino && ra->p_dev == dev)
817 			goto found;
818 		depth++;
819 		if (ra->p_count == 0)
820 			frap = rap;
821 	}
822 	depth = nfsdstats.ra_size*11/10;
823 	if (!frap) {
824 		spin_unlock(&rab->pb_lock);
825 		return NULL;
826 	}
827 	rap = frap;
828 	ra = *frap;
829 	ra->p_dev = dev;
830 	ra->p_ino = ino;
831 	ra->p_set = 0;
832 	ra->p_hindex = hash;
833 found:
834 	if (rap != &rab->pb_head) {
835 		*rap = ra->p_next;
836 		ra->p_next   = rab->pb_head;
837 		rab->pb_head = ra;
838 	}
839 	ra->p_count++;
840 	nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
841 	spin_unlock(&rab->pb_lock);
842 	return ra;
843 }
844 
845 /*
846  * Grab and keep cached pages associated with a file in the svc_rqst
847  * so that they can be passed to the network sendmsg/sendpage routines
848  * directly. They will be released after the sending has completed.
849  */
850 static int
851 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
852 		  struct splice_desc *sd)
853 {
854 	struct svc_rqst *rqstp = sd->u.data;
855 	struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
856 	struct page *page = buf->page;
857 	size_t size;
858 	int ret;
859 
860 	ret = buf->ops->confirm(pipe, buf);
861 	if (unlikely(ret))
862 		return ret;
863 
864 	size = sd->len;
865 
866 	if (rqstp->rq_res.page_len == 0) {
867 		get_page(page);
868 		put_page(*pp);
869 		*pp = page;
870 		rqstp->rq_resused++;
871 		rqstp->rq_res.page_base = buf->offset;
872 		rqstp->rq_res.page_len = size;
873 	} else if (page != pp[-1]) {
874 		get_page(page);
875 		if (*pp)
876 			put_page(*pp);
877 		*pp = page;
878 		rqstp->rq_resused++;
879 		rqstp->rq_res.page_len += size;
880 	} else
881 		rqstp->rq_res.page_len += size;
882 
883 	return size;
884 }
885 
886 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
887 				    struct splice_desc *sd)
888 {
889 	return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
890 }
891 
892 static inline int svc_msnfs(struct svc_fh *ffhp)
893 {
894 #ifdef MSNFS
895 	return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
896 #else
897 	return 0;
898 #endif
899 }
900 
901 static __be32
902 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
903               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
904 {
905 	struct inode *inode;
906 	mm_segment_t	oldfs;
907 	__be32		err;
908 	int		host_err;
909 
910 	err = nfserr_perm;
911 	inode = file->f_path.dentry->d_inode;
912 
913 	if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
914 		goto out;
915 
916 	if (file->f_op->splice_read && rqstp->rq_splice_ok) {
917 		struct splice_desc sd = {
918 			.len		= 0,
919 			.total_len	= *count,
920 			.pos		= offset,
921 			.u.data		= rqstp,
922 		};
923 
924 		rqstp->rq_resused = 1;
925 		host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
926 	} else {
927 		oldfs = get_fs();
928 		set_fs(KERNEL_DS);
929 		host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
930 		set_fs(oldfs);
931 	}
932 
933 	if (host_err >= 0) {
934 		nfsdstats.io_read += host_err;
935 		*count = host_err;
936 		err = 0;
937 		fsnotify_access(file);
938 	} else
939 		err = nfserrno(host_err);
940 out:
941 	return err;
942 }
943 
944 static void kill_suid(struct dentry *dentry)
945 {
946 	struct iattr	ia;
947 	ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
948 
949 	mutex_lock(&dentry->d_inode->i_mutex);
950 	notify_change(dentry, &ia);
951 	mutex_unlock(&dentry->d_inode->i_mutex);
952 }
953 
954 /*
955  * Gathered writes: If another process is currently writing to the file,
956  * there's a high chance this is another nfsd (triggered by a bulk write
957  * from a client's biod). Rather than syncing the file with each write
958  * request, we sleep for 10 msec.
959  *
960  * I don't know if this roughly approximates C. Juszak's idea of
961  * gathered writes, but it's a nice and simple solution (IMHO), and it
962  * seems to work:-)
963  *
964  * Note: we do this only in the NFSv2 case, since v3 and higher have a
965  * better tool (separate unstable writes and commits) for solving this
966  * problem.
967  */
968 static int wait_for_concurrent_writes(struct file *file)
969 {
970 	struct inode *inode = file->f_path.dentry->d_inode;
971 	static ino_t last_ino;
972 	static dev_t last_dev;
973 	int err = 0;
974 
975 	if (atomic_read(&inode->i_writecount) > 1
976 	    || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
977 		dprintk("nfsd: write defer %d\n", task_pid_nr(current));
978 		msleep(10);
979 		dprintk("nfsd: write resume %d\n", task_pid_nr(current));
980 	}
981 
982 	if (inode->i_state & I_DIRTY) {
983 		dprintk("nfsd: write sync %d\n", task_pid_nr(current));
984 		err = vfs_fsync(file, 0);
985 	}
986 	last_ino = inode->i_ino;
987 	last_dev = inode->i_sb->s_dev;
988 	return err;
989 }
990 
991 static __be32
992 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
993 				loff_t offset, struct kvec *vec, int vlen,
994 				unsigned long *cnt, int *stablep)
995 {
996 	struct svc_export	*exp;
997 	struct dentry		*dentry;
998 	struct inode		*inode;
999 	mm_segment_t		oldfs;
1000 	__be32			err = 0;
1001 	int			host_err;
1002 	int			stable = *stablep;
1003 	int			use_wgather;
1004 
1005 #ifdef MSNFS
1006 	err = nfserr_perm;
1007 
1008 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1009 		(!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1010 		goto out;
1011 #endif
1012 
1013 	dentry = file->f_path.dentry;
1014 	inode = dentry->d_inode;
1015 	exp   = fhp->fh_export;
1016 
1017 	/*
1018 	 * Request sync writes if
1019 	 *  -	the sync export option has been set, or
1020 	 *  -	the client requested O_SYNC behavior (NFSv3 feature).
1021 	 *  -   The file system doesn't support fsync().
1022 	 * When NFSv2 gathered writes have been configured for this volume,
1023 	 * flushing the data to disk is handled separately below.
1024 	 */
1025 	use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1026 
1027 	if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1028 	       stable = 2;
1029 	       *stablep = 2; /* FILE_SYNC */
1030 	}
1031 
1032 	if (!EX_ISSYNC(exp))
1033 		stable = 0;
1034 	if (stable && !use_wgather) {
1035 		spin_lock(&file->f_lock);
1036 		file->f_flags |= O_SYNC;
1037 		spin_unlock(&file->f_lock);
1038 	}
1039 
1040 	/* Write the data. */
1041 	oldfs = get_fs(); set_fs(KERNEL_DS);
1042 	host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1043 	set_fs(oldfs);
1044 	if (host_err < 0)
1045 		goto out_nfserr;
1046 	*cnt = host_err;
1047 	nfsdstats.io_write += host_err;
1048 	fsnotify_modify(file);
1049 
1050 	/* clear setuid/setgid flag after write */
1051 	if (inode->i_mode & (S_ISUID | S_ISGID))
1052 		kill_suid(dentry);
1053 
1054 	if (stable && use_wgather)
1055 		host_err = wait_for_concurrent_writes(file);
1056 
1057 out_nfserr:
1058 	dprintk("nfsd: write complete host_err=%d\n", host_err);
1059 	if (host_err >= 0)
1060 		err = 0;
1061 	else
1062 		err = nfserrno(host_err);
1063 out:
1064 	return err;
1065 }
1066 
1067 /*
1068  * Read data from a file. count must contain the requested read count
1069  * on entry. On return, *count contains the number of bytes actually read.
1070  * N.B. After this call fhp needs an fh_put
1071  */
1072 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1073 	loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1074 {
1075 	struct file *file;
1076 	struct inode *inode;
1077 	struct raparms	*ra;
1078 	__be32 err;
1079 
1080 	err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1081 	if (err)
1082 		return err;
1083 
1084 	inode = file->f_path.dentry->d_inode;
1085 
1086 	/* Get readahead parameters */
1087 	ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1088 
1089 	if (ra && ra->p_set)
1090 		file->f_ra = ra->p_ra;
1091 
1092 	err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1093 
1094 	/* Write back readahead params */
1095 	if (ra) {
1096 		struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1097 		spin_lock(&rab->pb_lock);
1098 		ra->p_ra = file->f_ra;
1099 		ra->p_set = 1;
1100 		ra->p_count--;
1101 		spin_unlock(&rab->pb_lock);
1102 	}
1103 
1104 	nfsd_close(file);
1105 	return err;
1106 }
1107 
1108 /* As above, but use the provided file descriptor. */
1109 __be32
1110 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1111 		loff_t offset, struct kvec *vec, int vlen,
1112 		unsigned long *count)
1113 {
1114 	__be32		err;
1115 
1116 	if (file) {
1117 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1118 				NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1119 		if (err)
1120 			goto out;
1121 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1122 	} else /* Note file may still be NULL in NFSv4 special stateid case: */
1123 		err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1124 out:
1125 	return err;
1126 }
1127 
1128 /*
1129  * Write data to a file.
1130  * The stable flag requests synchronous writes.
1131  * N.B. After this call fhp needs an fh_put
1132  */
1133 __be32
1134 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1135 		loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1136 		int *stablep)
1137 {
1138 	__be32			err = 0;
1139 
1140 	if (file) {
1141 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1142 				NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1143 		if (err)
1144 			goto out;
1145 		err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1146 				stablep);
1147 	} else {
1148 		err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1149 		if (err)
1150 			goto out;
1151 
1152 		if (cnt)
1153 			err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1154 					     cnt, stablep);
1155 		nfsd_close(file);
1156 	}
1157 out:
1158 	return err;
1159 }
1160 
1161 #ifdef CONFIG_NFSD_V3
1162 /*
1163  * Commit all pending writes to stable storage.
1164  *
1165  * Note: we only guarantee that data that lies within the range specified
1166  * by the 'offset' and 'count' parameters will be synced.
1167  *
1168  * Unfortunately we cannot lock the file to make sure we return full WCC
1169  * data to the client, as locking happens lower down in the filesystem.
1170  */
1171 __be32
1172 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1173                loff_t offset, unsigned long count)
1174 {
1175 	struct file	*file;
1176 	loff_t		end = LLONG_MAX;
1177 	__be32		err = nfserr_inval;
1178 
1179 	if (offset < 0)
1180 		goto out;
1181 	if (count != 0) {
1182 		end = offset + (loff_t)count - 1;
1183 		if (end < offset)
1184 			goto out;
1185 	}
1186 
1187 	err = nfsd_open(rqstp, fhp, S_IFREG,
1188 			NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1189 	if (err)
1190 		goto out;
1191 	if (EX_ISSYNC(fhp->fh_export)) {
1192 		int err2 = vfs_fsync_range(file, offset, end, 0);
1193 
1194 		if (err2 != -EINVAL)
1195 			err = nfserrno(err2);
1196 		else
1197 			err = nfserr_notsupp;
1198 	}
1199 
1200 	nfsd_close(file);
1201 out:
1202 	return err;
1203 }
1204 #endif /* CONFIG_NFSD_V3 */
1205 
1206 static __be32
1207 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1208 			struct iattr *iap)
1209 {
1210 	/*
1211 	 * Mode has already been set earlier in create:
1212 	 */
1213 	iap->ia_valid &= ~ATTR_MODE;
1214 	/*
1215 	 * Setting uid/gid works only for root.  Irix appears to
1216 	 * send along the gid on create when it tries to implement
1217 	 * setgid directories via NFS:
1218 	 */
1219 	if (current_fsuid() != 0)
1220 		iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1221 	if (iap->ia_valid)
1222 		return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1223 	return 0;
1224 }
1225 
1226 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1227  * setting size to 0 may fail for some specific file systems by the permission
1228  * checking which requires WRITE permission but the mode is 000.
1229  * we ignore the resizing(to 0) on the just new created file, since the size is
1230  * 0 after file created.
1231  *
1232  * call this only after vfs_create() is called.
1233  * */
1234 static void
1235 nfsd_check_ignore_resizing(struct iattr *iap)
1236 {
1237 	if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1238 		iap->ia_valid &= ~ATTR_SIZE;
1239 }
1240 
1241 /*
1242  * Create a file (regular, directory, device, fifo); UNIX sockets
1243  * not yet implemented.
1244  * If the response fh has been verified, the parent directory should
1245  * already be locked. Note that the parent directory is left locked.
1246  *
1247  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1248  */
1249 __be32
1250 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1251 		char *fname, int flen, struct iattr *iap,
1252 		int type, dev_t rdev, struct svc_fh *resfhp)
1253 {
1254 	struct dentry	*dentry, *dchild = NULL;
1255 	struct inode	*dirp;
1256 	__be32		err;
1257 	__be32		err2;
1258 	int		host_err;
1259 
1260 	err = nfserr_perm;
1261 	if (!flen)
1262 		goto out;
1263 	err = nfserr_exist;
1264 	if (isdotent(fname, flen))
1265 		goto out;
1266 
1267 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1268 	if (err)
1269 		goto out;
1270 
1271 	dentry = fhp->fh_dentry;
1272 	dirp = dentry->d_inode;
1273 
1274 	err = nfserr_notdir;
1275 	if (!dirp->i_op->lookup)
1276 		goto out;
1277 	/*
1278 	 * Check whether the response file handle has been verified yet.
1279 	 * If it has, the parent directory should already be locked.
1280 	 */
1281 	if (!resfhp->fh_dentry) {
1282 		/* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1283 		fh_lock_nested(fhp, I_MUTEX_PARENT);
1284 		dchild = lookup_one_len(fname, dentry, flen);
1285 		host_err = PTR_ERR(dchild);
1286 		if (IS_ERR(dchild))
1287 			goto out_nfserr;
1288 		err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1289 		if (err)
1290 			goto out;
1291 	} else {
1292 		/* called from nfsd_proc_create */
1293 		dchild = dget(resfhp->fh_dentry);
1294 		if (!fhp->fh_locked) {
1295 			/* not actually possible */
1296 			printk(KERN_ERR
1297 				"nfsd_create: parent %s/%s not locked!\n",
1298 				dentry->d_parent->d_name.name,
1299 				dentry->d_name.name);
1300 			err = nfserr_io;
1301 			goto out;
1302 		}
1303 	}
1304 	/*
1305 	 * Make sure the child dentry is still negative ...
1306 	 */
1307 	err = nfserr_exist;
1308 	if (dchild->d_inode) {
1309 		dprintk("nfsd_create: dentry %s/%s not negative!\n",
1310 			dentry->d_name.name, dchild->d_name.name);
1311 		goto out;
1312 	}
1313 
1314 	if (!(iap->ia_valid & ATTR_MODE))
1315 		iap->ia_mode = 0;
1316 	iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1317 
1318 	err = nfserr_inval;
1319 	if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1320 		printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1321 		       type);
1322 		goto out;
1323 	}
1324 
1325 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1326 	if (host_err)
1327 		goto out_nfserr;
1328 
1329 	/*
1330 	 * Get the dir op function pointer.
1331 	 */
1332 	err = 0;
1333 	switch (type) {
1334 	case S_IFREG:
1335 		host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1336 		if (!host_err)
1337 			nfsd_check_ignore_resizing(iap);
1338 		break;
1339 	case S_IFDIR:
1340 		host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1341 		break;
1342 	case S_IFCHR:
1343 	case S_IFBLK:
1344 	case S_IFIFO:
1345 	case S_IFSOCK:
1346 		host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1347 		break;
1348 	}
1349 	if (host_err < 0) {
1350 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1351 		goto out_nfserr;
1352 	}
1353 
1354 	err = nfsd_create_setattr(rqstp, resfhp, iap);
1355 
1356 	/*
1357 	 * nfsd_setattr already committed the child.  Transactional filesystems
1358 	 * had a chance to commit changes for both parent and child
1359 	 * simultaneously making the following commit_metadata a noop.
1360 	 */
1361 	err2 = nfserrno(commit_metadata(fhp));
1362 	if (err2)
1363 		err = err2;
1364 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1365 	/*
1366 	 * Update the file handle to get the new inode info.
1367 	 */
1368 	if (!err)
1369 		err = fh_update(resfhp);
1370 out:
1371 	if (dchild && !IS_ERR(dchild))
1372 		dput(dchild);
1373 	return err;
1374 
1375 out_nfserr:
1376 	err = nfserrno(host_err);
1377 	goto out;
1378 }
1379 
1380 #ifdef CONFIG_NFSD_V3
1381 /*
1382  * NFSv3 version of nfsd_create
1383  */
1384 __be32
1385 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1386 		char *fname, int flen, struct iattr *iap,
1387 		struct svc_fh *resfhp, int createmode, u32 *verifier,
1388 	        int *truncp, int *created)
1389 {
1390 	struct dentry	*dentry, *dchild = NULL;
1391 	struct inode	*dirp;
1392 	__be32		err;
1393 	int		host_err;
1394 	__u32		v_mtime=0, v_atime=0;
1395 
1396 	err = nfserr_perm;
1397 	if (!flen)
1398 		goto out;
1399 	err = nfserr_exist;
1400 	if (isdotent(fname, flen))
1401 		goto out;
1402 	if (!(iap->ia_valid & ATTR_MODE))
1403 		iap->ia_mode = 0;
1404 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1405 	if (err)
1406 		goto out;
1407 
1408 	dentry = fhp->fh_dentry;
1409 	dirp = dentry->d_inode;
1410 
1411 	/* Get all the sanity checks out of the way before
1412 	 * we lock the parent. */
1413 	err = nfserr_notdir;
1414 	if (!dirp->i_op->lookup)
1415 		goto out;
1416 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1417 
1418 	/*
1419 	 * Compose the response file handle.
1420 	 */
1421 	dchild = lookup_one_len(fname, dentry, flen);
1422 	host_err = PTR_ERR(dchild);
1423 	if (IS_ERR(dchild))
1424 		goto out_nfserr;
1425 
1426 	err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1427 	if (err)
1428 		goto out;
1429 
1430 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1431 		/* solaris7 gets confused (bugid 4218508) if these have
1432 		 * the high bit set, so just clear the high bits. If this is
1433 		 * ever changed to use different attrs for storing the
1434 		 * verifier, then do_open_lookup() will also need to be fixed
1435 		 * accordingly.
1436 		 */
1437 		v_mtime = verifier[0]&0x7fffffff;
1438 		v_atime = verifier[1]&0x7fffffff;
1439 	}
1440 
1441 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1442 	if (host_err)
1443 		goto out_nfserr;
1444 	if (dchild->d_inode) {
1445 		err = 0;
1446 
1447 		switch (createmode) {
1448 		case NFS3_CREATE_UNCHECKED:
1449 			if (! S_ISREG(dchild->d_inode->i_mode))
1450 				err = nfserr_exist;
1451 			else if (truncp) {
1452 				/* in nfsv4, we need to treat this case a little
1453 				 * differently.  we don't want to truncate the
1454 				 * file now; this would be wrong if the OPEN
1455 				 * fails for some other reason.  furthermore,
1456 				 * if the size is nonzero, we should ignore it
1457 				 * according to spec!
1458 				 */
1459 				*truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1460 			}
1461 			else {
1462 				iap->ia_valid &= ATTR_SIZE;
1463 				goto set_attr;
1464 			}
1465 			break;
1466 		case NFS3_CREATE_EXCLUSIVE:
1467 			if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1468 			    && dchild->d_inode->i_atime.tv_sec == v_atime
1469 			    && dchild->d_inode->i_size  == 0 )
1470 				break;
1471 			 /* fallthru */
1472 		case NFS3_CREATE_GUARDED:
1473 			err = nfserr_exist;
1474 		}
1475 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1476 		goto out;
1477 	}
1478 
1479 	host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1480 	if (host_err < 0) {
1481 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1482 		goto out_nfserr;
1483 	}
1484 	if (created)
1485 		*created = 1;
1486 
1487 	nfsd_check_ignore_resizing(iap);
1488 
1489 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1490 		/* Cram the verifier into atime/mtime */
1491 		iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1492 			| ATTR_MTIME_SET|ATTR_ATIME_SET;
1493 		/* XXX someone who knows this better please fix it for nsec */
1494 		iap->ia_mtime.tv_sec = v_mtime;
1495 		iap->ia_atime.tv_sec = v_atime;
1496 		iap->ia_mtime.tv_nsec = 0;
1497 		iap->ia_atime.tv_nsec = 0;
1498 	}
1499 
1500  set_attr:
1501 	err = nfsd_create_setattr(rqstp, resfhp, iap);
1502 
1503 	/*
1504 	 * nfsd_setattr already committed the child (and possibly also the parent).
1505 	 */
1506 	if (!err)
1507 		err = nfserrno(commit_metadata(fhp));
1508 
1509 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1510 	/*
1511 	 * Update the filehandle to get the new inode info.
1512 	 */
1513 	if (!err)
1514 		err = fh_update(resfhp);
1515 
1516  out:
1517 	fh_unlock(fhp);
1518 	if (dchild && !IS_ERR(dchild))
1519 		dput(dchild);
1520  	return err;
1521 
1522  out_nfserr:
1523 	err = nfserrno(host_err);
1524 	goto out;
1525 }
1526 #endif /* CONFIG_NFSD_V3 */
1527 
1528 /*
1529  * Read a symlink. On entry, *lenp must contain the maximum path length that
1530  * fits into the buffer. On return, it contains the true length.
1531  * N.B. After this call fhp needs an fh_put
1532  */
1533 __be32
1534 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1535 {
1536 	struct dentry	*dentry;
1537 	struct inode	*inode;
1538 	mm_segment_t	oldfs;
1539 	__be32		err;
1540 	int		host_err;
1541 
1542 	err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1543 	if (err)
1544 		goto out;
1545 
1546 	dentry = fhp->fh_dentry;
1547 	inode = dentry->d_inode;
1548 
1549 	err = nfserr_inval;
1550 	if (!inode->i_op->readlink)
1551 		goto out;
1552 
1553 	touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1554 	/* N.B. Why does this call need a get_fs()??
1555 	 * Remove the set_fs and watch the fireworks:-) --okir
1556 	 */
1557 
1558 	oldfs = get_fs(); set_fs(KERNEL_DS);
1559 	host_err = inode->i_op->readlink(dentry, buf, *lenp);
1560 	set_fs(oldfs);
1561 
1562 	if (host_err < 0)
1563 		goto out_nfserr;
1564 	*lenp = host_err;
1565 	err = 0;
1566 out:
1567 	return err;
1568 
1569 out_nfserr:
1570 	err = nfserrno(host_err);
1571 	goto out;
1572 }
1573 
1574 /*
1575  * Create a symlink and look up its inode
1576  * N.B. After this call _both_ fhp and resfhp need an fh_put
1577  */
1578 __be32
1579 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1580 				char *fname, int flen,
1581 				char *path,  int plen,
1582 				struct svc_fh *resfhp,
1583 				struct iattr *iap)
1584 {
1585 	struct dentry	*dentry, *dnew;
1586 	__be32		err, cerr;
1587 	int		host_err;
1588 
1589 	err = nfserr_noent;
1590 	if (!flen || !plen)
1591 		goto out;
1592 	err = nfserr_exist;
1593 	if (isdotent(fname, flen))
1594 		goto out;
1595 
1596 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1597 	if (err)
1598 		goto out;
1599 	fh_lock(fhp);
1600 	dentry = fhp->fh_dentry;
1601 	dnew = lookup_one_len(fname, dentry, flen);
1602 	host_err = PTR_ERR(dnew);
1603 	if (IS_ERR(dnew))
1604 		goto out_nfserr;
1605 
1606 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1607 	if (host_err)
1608 		goto out_nfserr;
1609 
1610 	if (unlikely(path[plen] != 0)) {
1611 		char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1612 		if (path_alloced == NULL)
1613 			host_err = -ENOMEM;
1614 		else {
1615 			strncpy(path_alloced, path, plen);
1616 			path_alloced[plen] = 0;
1617 			host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1618 			kfree(path_alloced);
1619 		}
1620 	} else
1621 		host_err = vfs_symlink(dentry->d_inode, dnew, path);
1622 	err = nfserrno(host_err);
1623 	if (!err)
1624 		err = nfserrno(commit_metadata(fhp));
1625 	fh_unlock(fhp);
1626 
1627 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1628 
1629 	cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1630 	dput(dnew);
1631 	if (err==0) err = cerr;
1632 out:
1633 	return err;
1634 
1635 out_nfserr:
1636 	err = nfserrno(host_err);
1637 	goto out;
1638 }
1639 
1640 /*
1641  * Create a hardlink
1642  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1643  */
1644 __be32
1645 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1646 				char *name, int len, struct svc_fh *tfhp)
1647 {
1648 	struct dentry	*ddir, *dnew, *dold;
1649 	struct inode	*dirp;
1650 	__be32		err;
1651 	int		host_err;
1652 
1653 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1654 	if (err)
1655 		goto out;
1656 	err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1657 	if (err)
1658 		goto out;
1659 
1660 	err = nfserr_perm;
1661 	if (!len)
1662 		goto out;
1663 	err = nfserr_exist;
1664 	if (isdotent(name, len))
1665 		goto out;
1666 
1667 	fh_lock_nested(ffhp, I_MUTEX_PARENT);
1668 	ddir = ffhp->fh_dentry;
1669 	dirp = ddir->d_inode;
1670 
1671 	dnew = lookup_one_len(name, ddir, len);
1672 	host_err = PTR_ERR(dnew);
1673 	if (IS_ERR(dnew))
1674 		goto out_nfserr;
1675 
1676 	dold = tfhp->fh_dentry;
1677 
1678 	host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1679 	if (host_err) {
1680 		err = nfserrno(host_err);
1681 		goto out_dput;
1682 	}
1683 	host_err = vfs_link(dold, dirp, dnew);
1684 	if (!host_err) {
1685 		err = nfserrno(commit_metadata(ffhp));
1686 		if (!err)
1687 			err = nfserrno(commit_metadata(tfhp));
1688 	} else {
1689 		if (host_err == -EXDEV && rqstp->rq_vers == 2)
1690 			err = nfserr_acces;
1691 		else
1692 			err = nfserrno(host_err);
1693 	}
1694 	mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1695 out_dput:
1696 	dput(dnew);
1697 out_unlock:
1698 	fh_unlock(ffhp);
1699 out:
1700 	return err;
1701 
1702 out_nfserr:
1703 	err = nfserrno(host_err);
1704 	goto out_unlock;
1705 }
1706 
1707 /*
1708  * Rename a file
1709  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1710  */
1711 __be32
1712 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1713 			    struct svc_fh *tfhp, char *tname, int tlen)
1714 {
1715 	struct dentry	*fdentry, *tdentry, *odentry, *ndentry, *trap;
1716 	struct inode	*fdir, *tdir;
1717 	__be32		err;
1718 	int		host_err;
1719 
1720 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1721 	if (err)
1722 		goto out;
1723 	err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1724 	if (err)
1725 		goto out;
1726 
1727 	fdentry = ffhp->fh_dentry;
1728 	fdir = fdentry->d_inode;
1729 
1730 	tdentry = tfhp->fh_dentry;
1731 	tdir = tdentry->d_inode;
1732 
1733 	err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1734 	if (ffhp->fh_export != tfhp->fh_export)
1735 		goto out;
1736 
1737 	err = nfserr_perm;
1738 	if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1739 		goto out;
1740 
1741 	/* cannot use fh_lock as we need deadlock protective ordering
1742 	 * so do it by hand */
1743 	trap = lock_rename(tdentry, fdentry);
1744 	ffhp->fh_locked = tfhp->fh_locked = 1;
1745 	fill_pre_wcc(ffhp);
1746 	fill_pre_wcc(tfhp);
1747 
1748 	odentry = lookup_one_len(fname, fdentry, flen);
1749 	host_err = PTR_ERR(odentry);
1750 	if (IS_ERR(odentry))
1751 		goto out_nfserr;
1752 
1753 	host_err = -ENOENT;
1754 	if (!odentry->d_inode)
1755 		goto out_dput_old;
1756 	host_err = -EINVAL;
1757 	if (odentry == trap)
1758 		goto out_dput_old;
1759 
1760 	ndentry = lookup_one_len(tname, tdentry, tlen);
1761 	host_err = PTR_ERR(ndentry);
1762 	if (IS_ERR(ndentry))
1763 		goto out_dput_old;
1764 	host_err = -ENOTEMPTY;
1765 	if (ndentry == trap)
1766 		goto out_dput_new;
1767 
1768 	if (svc_msnfs(ffhp) &&
1769 		((atomic_read(&odentry->d_count) > 1)
1770 		 || (atomic_read(&ndentry->d_count) > 1))) {
1771 			host_err = -EPERM;
1772 			goto out_dput_new;
1773 	}
1774 
1775 	host_err = -EXDEV;
1776 	if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1777 		goto out_dput_new;
1778 	host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1779 	if (host_err)
1780 		goto out_dput_new;
1781 
1782 	host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1783 	if (!host_err) {
1784 		host_err = commit_metadata(tfhp);
1785 		if (!host_err)
1786 			host_err = commit_metadata(ffhp);
1787 	}
1788 
1789 	mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1790 
1791  out_dput_new:
1792 	dput(ndentry);
1793  out_dput_old:
1794 	dput(odentry);
1795  out_nfserr:
1796 	err = nfserrno(host_err);
1797 
1798 	/* we cannot reply on fh_unlock on the two filehandles,
1799 	 * as that would do the wrong thing if the two directories
1800 	 * were the same, so again we do it by hand
1801 	 */
1802 	fill_post_wcc(ffhp);
1803 	fill_post_wcc(tfhp);
1804 	unlock_rename(tdentry, fdentry);
1805 	ffhp->fh_locked = tfhp->fh_locked = 0;
1806 
1807 out:
1808 	return err;
1809 }
1810 
1811 /*
1812  * Unlink a file or directory
1813  * N.B. After this call fhp needs an fh_put
1814  */
1815 __be32
1816 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1817 				char *fname, int flen)
1818 {
1819 	struct dentry	*dentry, *rdentry;
1820 	struct inode	*dirp;
1821 	__be32		err;
1822 	int		host_err;
1823 
1824 	err = nfserr_acces;
1825 	if (!flen || isdotent(fname, flen))
1826 		goto out;
1827 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1828 	if (err)
1829 		goto out;
1830 
1831 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1832 	dentry = fhp->fh_dentry;
1833 	dirp = dentry->d_inode;
1834 
1835 	rdentry = lookup_one_len(fname, dentry, flen);
1836 	host_err = PTR_ERR(rdentry);
1837 	if (IS_ERR(rdentry))
1838 		goto out_nfserr;
1839 
1840 	if (!rdentry->d_inode) {
1841 		dput(rdentry);
1842 		err = nfserr_noent;
1843 		goto out;
1844 	}
1845 
1846 	if (!type)
1847 		type = rdentry->d_inode->i_mode & S_IFMT;
1848 
1849 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1850 	if (host_err)
1851 		goto out_nfserr;
1852 
1853 	if (type != S_IFDIR) { /* It's UNLINK */
1854 #ifdef MSNFS
1855 		if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1856 			(atomic_read(&rdentry->d_count) > 1)) {
1857 			host_err = -EPERM;
1858 		} else
1859 #endif
1860 		host_err = vfs_unlink(dirp, rdentry);
1861 	} else { /* It's RMDIR */
1862 		host_err = vfs_rmdir(dirp, rdentry);
1863 	}
1864 
1865 	dput(rdentry);
1866 
1867 	if (!host_err)
1868 		host_err = commit_metadata(fhp);
1869 
1870 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1871 out_nfserr:
1872 	err = nfserrno(host_err);
1873 out:
1874 	return err;
1875 }
1876 
1877 /*
1878  * We do this buffering because we must not call back into the file
1879  * system's ->lookup() method from the filldir callback. That may well
1880  * deadlock a number of file systems.
1881  *
1882  * This is based heavily on the implementation of same in XFS.
1883  */
1884 struct buffered_dirent {
1885 	u64		ino;
1886 	loff_t		offset;
1887 	int		namlen;
1888 	unsigned int	d_type;
1889 	char		name[];
1890 };
1891 
1892 struct readdir_data {
1893 	char		*dirent;
1894 	size_t		used;
1895 	int		full;
1896 };
1897 
1898 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1899 				 loff_t offset, u64 ino, unsigned int d_type)
1900 {
1901 	struct readdir_data *buf = __buf;
1902 	struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1903 	unsigned int reclen;
1904 
1905 	reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1906 	if (buf->used + reclen > PAGE_SIZE) {
1907 		buf->full = 1;
1908 		return -EINVAL;
1909 	}
1910 
1911 	de->namlen = namlen;
1912 	de->offset = offset;
1913 	de->ino = ino;
1914 	de->d_type = d_type;
1915 	memcpy(de->name, name, namlen);
1916 	buf->used += reclen;
1917 
1918 	return 0;
1919 }
1920 
1921 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1922 				    struct readdir_cd *cdp, loff_t *offsetp)
1923 {
1924 	struct readdir_data buf;
1925 	struct buffered_dirent *de;
1926 	int host_err;
1927 	int size;
1928 	loff_t offset;
1929 
1930 	buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1931 	if (!buf.dirent)
1932 		return nfserrno(-ENOMEM);
1933 
1934 	offset = *offsetp;
1935 
1936 	while (1) {
1937 		struct inode *dir_inode = file->f_path.dentry->d_inode;
1938 		unsigned int reclen;
1939 
1940 		cdp->err = nfserr_eof; /* will be cleared on successful read */
1941 		buf.used = 0;
1942 		buf.full = 0;
1943 
1944 		host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1945 		if (buf.full)
1946 			host_err = 0;
1947 
1948 		if (host_err < 0)
1949 			break;
1950 
1951 		size = buf.used;
1952 
1953 		if (!size)
1954 			break;
1955 
1956 		/*
1957 		 * Various filldir functions may end up calling back into
1958 		 * lookup_one_len() and the file system's ->lookup() method.
1959 		 * These expect i_mutex to be held, as it would within readdir.
1960 		 */
1961 		host_err = mutex_lock_killable(&dir_inode->i_mutex);
1962 		if (host_err)
1963 			break;
1964 
1965 		de = (struct buffered_dirent *)buf.dirent;
1966 		while (size > 0) {
1967 			offset = de->offset;
1968 
1969 			if (func(cdp, de->name, de->namlen, de->offset,
1970 				 de->ino, de->d_type))
1971 				break;
1972 
1973 			if (cdp->err != nfs_ok)
1974 				break;
1975 
1976 			reclen = ALIGN(sizeof(*de) + de->namlen,
1977 				       sizeof(u64));
1978 			size -= reclen;
1979 			de = (struct buffered_dirent *)((char *)de + reclen);
1980 		}
1981 		mutex_unlock(&dir_inode->i_mutex);
1982 		if (size > 0) /* We bailed out early */
1983 			break;
1984 
1985 		offset = vfs_llseek(file, 0, SEEK_CUR);
1986 	}
1987 
1988 	free_page((unsigned long)(buf.dirent));
1989 
1990 	if (host_err)
1991 		return nfserrno(host_err);
1992 
1993 	*offsetp = offset;
1994 	return cdp->err;
1995 }
1996 
1997 /*
1998  * Read entries from a directory.
1999  * The  NFSv3/4 verifier we ignore for now.
2000  */
2001 __be32
2002 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2003 	     struct readdir_cd *cdp, filldir_t func)
2004 {
2005 	__be32		err;
2006 	struct file	*file;
2007 	loff_t		offset = *offsetp;
2008 
2009 	err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2010 	if (err)
2011 		goto out;
2012 
2013 	offset = vfs_llseek(file, offset, 0);
2014 	if (offset < 0) {
2015 		err = nfserrno((int)offset);
2016 		goto out_close;
2017 	}
2018 
2019 	err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2020 
2021 	if (err == nfserr_eof || err == nfserr_toosmall)
2022 		err = nfs_ok; /* can still be found in ->err */
2023 out_close:
2024 	nfsd_close(file);
2025 out:
2026 	return err;
2027 }
2028 
2029 /*
2030  * Get file system stats
2031  * N.B. After this call fhp needs an fh_put
2032  */
2033 __be32
2034 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2035 {
2036 	struct path path = {
2037 		.mnt	= fhp->fh_export->ex_path.mnt,
2038 		.dentry	= fhp->fh_dentry,
2039 	};
2040 	__be32 err;
2041 
2042 	err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2043 	if (!err && vfs_statfs(&path, stat))
2044 		err = nfserr_io;
2045 	return err;
2046 }
2047 
2048 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2049 {
2050 	return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2051 }
2052 
2053 /*
2054  * Check for a user's access permissions to this inode.
2055  */
2056 __be32
2057 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2058 					struct dentry *dentry, int acc)
2059 {
2060 	struct inode	*inode = dentry->d_inode;
2061 	int		err;
2062 
2063 	if (acc == NFSD_MAY_NOP)
2064 		return 0;
2065 #if 0
2066 	dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2067 		acc,
2068 		(acc & NFSD_MAY_READ)?	" read"  : "",
2069 		(acc & NFSD_MAY_WRITE)?	" write" : "",
2070 		(acc & NFSD_MAY_EXEC)?	" exec"  : "",
2071 		(acc & NFSD_MAY_SATTR)?	" sattr" : "",
2072 		(acc & NFSD_MAY_TRUNC)?	" trunc" : "",
2073 		(acc & NFSD_MAY_LOCK)?	" lock"  : "",
2074 		(acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2075 		inode->i_mode,
2076 		IS_IMMUTABLE(inode)?	" immut" : "",
2077 		IS_APPEND(inode)?	" append" : "",
2078 		__mnt_is_readonly(exp->ex_path.mnt)?	" ro" : "");
2079 	dprintk("      owner %d/%d user %d/%d\n",
2080 		inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2081 #endif
2082 
2083 	/* Normally we reject any write/sattr etc access on a read-only file
2084 	 * system.  But if it is IRIX doing check on write-access for a
2085 	 * device special file, we ignore rofs.
2086 	 */
2087 	if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2088 		if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2089 			if (exp_rdonly(rqstp, exp) ||
2090 			    __mnt_is_readonly(exp->ex_path.mnt))
2091 				return nfserr_rofs;
2092 			if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2093 				return nfserr_perm;
2094 		}
2095 	if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2096 		return nfserr_perm;
2097 
2098 	if (acc & NFSD_MAY_LOCK) {
2099 		/* If we cannot rely on authentication in NLM requests,
2100 		 * just allow locks, otherwise require read permission, or
2101 		 * ownership
2102 		 */
2103 		if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2104 			return 0;
2105 		else
2106 			acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2107 	}
2108 	/*
2109 	 * The file owner always gets access permission for accesses that
2110 	 * would normally be checked at open time. This is to make
2111 	 * file access work even when the client has done a fchmod(fd, 0).
2112 	 *
2113 	 * However, `cp foo bar' should fail nevertheless when bar is
2114 	 * readonly. A sensible way to do this might be to reject all
2115 	 * attempts to truncate a read-only file, because a creat() call
2116 	 * always implies file truncation.
2117 	 * ... but this isn't really fair.  A process may reasonably call
2118 	 * ftruncate on an open file descriptor on a file with perm 000.
2119 	 * We must trust the client to do permission checking - using "ACCESS"
2120 	 * with NFSv3.
2121 	 */
2122 	if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2123 	    inode->i_uid == current_fsuid())
2124 		return 0;
2125 
2126 	/* This assumes  NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2127 	err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2128 
2129 	/* Allow read access to binaries even when mode 111 */
2130 	if (err == -EACCES && S_ISREG(inode->i_mode) &&
2131 	    acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2132 		err = inode_permission(inode, MAY_EXEC);
2133 
2134 	return err? nfserrno(err) : 0;
2135 }
2136 
2137 void
2138 nfsd_racache_shutdown(void)
2139 {
2140 	struct raparms *raparm, *last_raparm;
2141 	unsigned int i;
2142 
2143 	dprintk("nfsd: freeing readahead buffers.\n");
2144 
2145 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2146 		raparm = raparm_hash[i].pb_head;
2147 		while(raparm) {
2148 			last_raparm = raparm;
2149 			raparm = raparm->p_next;
2150 			kfree(last_raparm);
2151 		}
2152 		raparm_hash[i].pb_head = NULL;
2153 	}
2154 }
2155 /*
2156  * Initialize readahead param cache
2157  */
2158 int
2159 nfsd_racache_init(int cache_size)
2160 {
2161 	int	i;
2162 	int	j = 0;
2163 	int	nperbucket;
2164 	struct raparms **raparm = NULL;
2165 
2166 
2167 	if (raparm_hash[0].pb_head)
2168 		return 0;
2169 	nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2170 	if (nperbucket < 2)
2171 		nperbucket = 2;
2172 	cache_size = nperbucket * RAPARM_HASH_SIZE;
2173 
2174 	dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2175 
2176 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2177 		spin_lock_init(&raparm_hash[i].pb_lock);
2178 
2179 		raparm = &raparm_hash[i].pb_head;
2180 		for (j = 0; j < nperbucket; j++) {
2181 			*raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2182 			if (!*raparm)
2183 				goto out_nomem;
2184 			raparm = &(*raparm)->p_next;
2185 		}
2186 		*raparm = NULL;
2187 	}
2188 
2189 	nfsdstats.ra_size = cache_size;
2190 	return 0;
2191 
2192 out_nomem:
2193 	dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2194 	nfsd_racache_shutdown();
2195 	return -ENOMEM;
2196 }
2197 
2198 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2199 struct posix_acl *
2200 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2201 {
2202 	struct inode *inode = fhp->fh_dentry->d_inode;
2203 	char *name;
2204 	void *value = NULL;
2205 	ssize_t size;
2206 	struct posix_acl *acl;
2207 
2208 	if (!IS_POSIXACL(inode))
2209 		return ERR_PTR(-EOPNOTSUPP);
2210 
2211 	switch (type) {
2212 	case ACL_TYPE_ACCESS:
2213 		name = POSIX_ACL_XATTR_ACCESS;
2214 		break;
2215 	case ACL_TYPE_DEFAULT:
2216 		name = POSIX_ACL_XATTR_DEFAULT;
2217 		break;
2218 	default:
2219 		return ERR_PTR(-EOPNOTSUPP);
2220 	}
2221 
2222 	size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2223 	if (size < 0)
2224 		return ERR_PTR(size);
2225 
2226 	acl = posix_acl_from_xattr(value, size);
2227 	kfree(value);
2228 	return acl;
2229 }
2230 
2231 int
2232 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2233 {
2234 	struct inode *inode = fhp->fh_dentry->d_inode;
2235 	char *name;
2236 	void *value = NULL;
2237 	size_t size;
2238 	int error;
2239 
2240 	if (!IS_POSIXACL(inode) ||
2241 	    !inode->i_op->setxattr || !inode->i_op->removexattr)
2242 		return -EOPNOTSUPP;
2243 	switch(type) {
2244 		case ACL_TYPE_ACCESS:
2245 			name = POSIX_ACL_XATTR_ACCESS;
2246 			break;
2247 		case ACL_TYPE_DEFAULT:
2248 			name = POSIX_ACL_XATTR_DEFAULT;
2249 			break;
2250 		default:
2251 			return -EOPNOTSUPP;
2252 	}
2253 
2254 	if (acl && acl->a_count) {
2255 		size = posix_acl_xattr_size(acl->a_count);
2256 		value = kmalloc(size, GFP_KERNEL);
2257 		if (!value)
2258 			return -ENOMEM;
2259 		error = posix_acl_to_xattr(acl, value, size);
2260 		if (error < 0)
2261 			goto getout;
2262 		size = error;
2263 	} else
2264 		size = 0;
2265 
2266 	error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2267 	if (error)
2268 		goto getout;
2269 	if (size)
2270 		error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2271 	else {
2272 		if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2273 			error = 0;
2274 		else {
2275 			error = vfs_removexattr(fhp->fh_dentry, name);
2276 			if (error == -ENODATA)
2277 				error = 0;
2278 		}
2279 	}
2280 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
2281 
2282 getout:
2283 	kfree(value);
2284 	return error;
2285 }
2286 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
2287