xref: /linux/fs/nfsd/vfs.c (revision 040932cdcfca9b0ac55a4f74f194c2e2c8a2527b)
1 #define MSNFS	/* HACK HACK */
2 /*
3  * linux/fs/nfsd/vfs.c
4  *
5  * File operations used by nfsd. Some of these have been ripped from
6  * other parts of the kernel because they weren't exported, others
7  * are partial duplicates with added or changed functionality.
8  *
9  * Note that several functions dget() the dentry upon which they want
10  * to act, most notably those that create directory entries. Response
11  * dentry's are dput()'d if necessary in the release callback.
12  * So if you notice code paths that apparently fail to dput() the
13  * dentry, don't worry--they have been taken care of.
14  *
15  * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16  * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17  */
18 
19 #include <linux/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/fs.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/splice.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
34 #include <linux/in.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
41 #ifdef CONFIG_NFSD_V3
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
51 #ifdef CONFIG_NFSD_V4
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
57 #include <linux/jhash.h>
58 #include <linux/ima.h>
59 
60 #include <asm/uaccess.h>
61 
62 #define NFSDDBG_FACILITY		NFSDDBG_FILEOP
63 
64 
65 /*
66  * This is a cache of readahead params that help us choose the proper
67  * readahead strategy. Initially, we set all readahead parameters to 0
68  * and let the VFS handle things.
69  * If you increase the number of cached files very much, you'll need to
70  * add a hash table here.
71  */
72 struct raparms {
73 	struct raparms		*p_next;
74 	unsigned int		p_count;
75 	ino_t			p_ino;
76 	dev_t			p_dev;
77 	int			p_set;
78 	struct file_ra_state	p_ra;
79 	unsigned int		p_hindex;
80 };
81 
82 struct raparm_hbucket {
83 	struct raparms		*pb_head;
84 	spinlock_t		pb_lock;
85 } ____cacheline_aligned_in_smp;
86 
87 #define RAPARM_HASH_BITS	4
88 #define RAPARM_HASH_SIZE	(1<<RAPARM_HASH_BITS)
89 #define RAPARM_HASH_MASK	(RAPARM_HASH_SIZE-1)
90 static struct raparm_hbucket	raparm_hash[RAPARM_HASH_SIZE];
91 
92 /*
93  * Called from nfsd_lookup and encode_dirent. Check if we have crossed
94  * a mount point.
95  * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
96  *  or nfs_ok having possibly changed *dpp and *expp
97  */
98 int
99 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
100 		        struct svc_export **expp)
101 {
102 	struct svc_export *exp = *expp, *exp2 = NULL;
103 	struct dentry *dentry = *dpp;
104 	struct path path = {.mnt = mntget(exp->ex_path.mnt),
105 			    .dentry = dget(dentry)};
106 	int err = 0;
107 
108 	while (d_mountpoint(path.dentry) && follow_down(&path))
109 		;
110 
111 	exp2 = rqst_exp_get_by_name(rqstp, &path);
112 	if (IS_ERR(exp2)) {
113 		if (PTR_ERR(exp2) != -ENOENT)
114 			err = PTR_ERR(exp2);
115 		path_put(&path);
116 		goto out;
117 	}
118 	if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
119 		/* successfully crossed mount point */
120 		/*
121 		 * This is subtle: path.dentry is *not* on path.mnt
122 		 * at this point.  The only reason we are safe is that
123 		 * original mnt is pinned down by exp, so we should
124 		 * put path *before* putting exp
125 		 */
126 		*dpp = path.dentry;
127 		path.dentry = dentry;
128 		*expp = exp2;
129 		exp2 = exp;
130 	}
131 	path_put(&path);
132 	exp_put(exp2);
133 out:
134 	return err;
135 }
136 
137 __be32
138 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
139 		   const char *name, unsigned int len,
140 		   struct svc_export **exp_ret, struct dentry **dentry_ret)
141 {
142 	struct svc_export	*exp;
143 	struct dentry		*dparent;
144 	struct dentry		*dentry;
145 	__be32			err;
146 	int			host_err;
147 
148 	dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
149 
150 	/* Obtain dentry and export. */
151 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
152 	if (err)
153 		return err;
154 
155 	dparent = fhp->fh_dentry;
156 	exp  = fhp->fh_export;
157 	exp_get(exp);
158 
159 	/* Lookup the name, but don't follow links */
160 	if (isdotent(name, len)) {
161 		if (len==1)
162 			dentry = dget(dparent);
163 		else if (dparent != exp->ex_path.dentry)
164 			dentry = dget_parent(dparent);
165 		else if (!EX_NOHIDE(exp))
166 			dentry = dget(dparent); /* .. == . just like at / */
167 		else {
168 			/* checking mountpoint crossing is very different when stepping up */
169 			struct svc_export *exp2 = NULL;
170 			struct dentry *dp;
171 			struct path path = {.mnt = mntget(exp->ex_path.mnt),
172 					    .dentry = dget(dparent)};
173 
174 			while (path.dentry == path.mnt->mnt_root &&
175 			       follow_up(&path))
176 				;
177 			dp = dget_parent(path.dentry);
178 			dput(path.dentry);
179 			path.dentry = dp;
180 
181 			exp2 = rqst_exp_parent(rqstp, &path);
182 			if (PTR_ERR(exp2) == -ENOENT) {
183 				dentry = dget(dparent);
184 			} else if (IS_ERR(exp2)) {
185 				host_err = PTR_ERR(exp2);
186 				path_put(&path);
187 				goto out_nfserr;
188 			} else {
189 				dentry = dget(path.dentry);
190 				exp_put(exp);
191 				exp = exp2;
192 			}
193 			path_put(&path);
194 		}
195 	} else {
196 		fh_lock(fhp);
197 		dentry = lookup_one_len(name, dparent, len);
198 		host_err = PTR_ERR(dentry);
199 		if (IS_ERR(dentry))
200 			goto out_nfserr;
201 		/*
202 		 * check if we have crossed a mount point ...
203 		 */
204 		if (d_mountpoint(dentry)) {
205 			if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
206 				dput(dentry);
207 				goto out_nfserr;
208 			}
209 		}
210 	}
211 	*dentry_ret = dentry;
212 	*exp_ret = exp;
213 	return 0;
214 
215 out_nfserr:
216 	exp_put(exp);
217 	return nfserrno(host_err);
218 }
219 
220 /*
221  * Look up one component of a pathname.
222  * N.B. After this call _both_ fhp and resfh need an fh_put
223  *
224  * If the lookup would cross a mountpoint, and the mounted filesystem
225  * is exported to the client with NFSEXP_NOHIDE, then the lookup is
226  * accepted as it stands and the mounted directory is
227  * returned. Otherwise the covered directory is returned.
228  * NOTE: this mountpoint crossing is not supported properly by all
229  *   clients and is explicitly disallowed for NFSv3
230  *      NeilBrown <neilb@cse.unsw.edu.au>
231  */
232 __be32
233 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
234 				unsigned int len, struct svc_fh *resfh)
235 {
236 	struct svc_export	*exp;
237 	struct dentry		*dentry;
238 	__be32 err;
239 
240 	err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
241 	if (err)
242 		return err;
243 	err = check_nfsd_access(exp, rqstp);
244 	if (err)
245 		goto out;
246 	/*
247 	 * Note: we compose the file handle now, but as the
248 	 * dentry may be negative, it may need to be updated.
249 	 */
250 	err = fh_compose(resfh, exp, dentry, fhp);
251 	if (!err && !dentry->d_inode)
252 		err = nfserr_noent;
253 out:
254 	dput(dentry);
255 	exp_put(exp);
256 	return err;
257 }
258 
259 
260 /*
261  * Set various file attributes.
262  * N.B. After this call fhp needs an fh_put
263  */
264 __be32
265 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
266 	     int check_guard, time_t guardtime)
267 {
268 	struct dentry	*dentry;
269 	struct inode	*inode;
270 	int		accmode = NFSD_MAY_SATTR;
271 	int		ftype = 0;
272 	__be32		err;
273 	int		host_err;
274 	int		size_change = 0;
275 
276 	if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
277 		accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
278 	if (iap->ia_valid & ATTR_SIZE)
279 		ftype = S_IFREG;
280 
281 	/* Get inode */
282 	err = fh_verify(rqstp, fhp, ftype, accmode);
283 	if (err)
284 		goto out;
285 
286 	dentry = fhp->fh_dentry;
287 	inode = dentry->d_inode;
288 
289 	/* Ignore any mode updates on symlinks */
290 	if (S_ISLNK(inode->i_mode))
291 		iap->ia_valid &= ~ATTR_MODE;
292 
293 	if (!iap->ia_valid)
294 		goto out;
295 
296 	/*
297 	 * NFSv2 does not differentiate between "set-[ac]time-to-now"
298 	 * which only requires access, and "set-[ac]time-to-X" which
299 	 * requires ownership.
300 	 * So if it looks like it might be "set both to the same time which
301 	 * is close to now", and if inode_change_ok fails, then we
302 	 * convert to "set to now" instead of "set to explicit time"
303 	 *
304 	 * We only call inode_change_ok as the last test as technically
305 	 * it is not an interface that we should be using.  It is only
306 	 * valid if the filesystem does not define it's own i_op->setattr.
307 	 */
308 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
309 #define	MAX_TOUCH_TIME_ERROR (30*60)
310 	if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
311 	    iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
312 		/*
313 		 * Looks probable.
314 		 *
315 		 * Now just make sure time is in the right ballpark.
316 		 * Solaris, at least, doesn't seem to care what the time
317 		 * request is.  We require it be within 30 minutes of now.
318 		 */
319 		time_t delta = iap->ia_atime.tv_sec - get_seconds();
320 		if (delta < 0)
321 			delta = -delta;
322 		if (delta < MAX_TOUCH_TIME_ERROR &&
323 		    inode_change_ok(inode, iap) != 0) {
324 			/*
325 			 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
326 			 * This will cause notify_change to set these times
327 			 * to "now"
328 			 */
329 			iap->ia_valid &= ~BOTH_TIME_SET;
330 		}
331 	}
332 
333 	/*
334 	 * The size case is special.
335 	 * It changes the file as well as the attributes.
336 	 */
337 	if (iap->ia_valid & ATTR_SIZE) {
338 		if (iap->ia_size < inode->i_size) {
339 			err = nfsd_permission(rqstp, fhp->fh_export, dentry,
340 					NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
341 			if (err)
342 				goto out;
343 		}
344 
345 		/*
346 		 * If we are changing the size of the file, then
347 		 * we need to break all leases.
348 		 */
349 		host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
350 		if (host_err == -EWOULDBLOCK)
351 			host_err = -ETIMEDOUT;
352 		if (host_err) /* ENOMEM or EWOULDBLOCK */
353 			goto out_nfserr;
354 
355 		host_err = get_write_access(inode);
356 		if (host_err)
357 			goto out_nfserr;
358 
359 		size_change = 1;
360 		host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
361 		if (host_err) {
362 			put_write_access(inode);
363 			goto out_nfserr;
364 		}
365 		vfs_dq_init(inode);
366 	}
367 
368 	/* sanitize the mode change */
369 	if (iap->ia_valid & ATTR_MODE) {
370 		iap->ia_mode &= S_IALLUGO;
371 		iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
372 	}
373 
374 	/* Revoke setuid/setgid on chown */
375 	if (!S_ISDIR(inode->i_mode) &&
376 	    (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
377 	     ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
378 		iap->ia_valid |= ATTR_KILL_PRIV;
379 		if (iap->ia_valid & ATTR_MODE) {
380 			/* we're setting mode too, just clear the s*id bits */
381 			iap->ia_mode &= ~S_ISUID;
382 			if (iap->ia_mode & S_IXGRP)
383 				iap->ia_mode &= ~S_ISGID;
384 		} else {
385 			/* set ATTR_KILL_* bits and let VFS handle it */
386 			iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
387 		}
388 	}
389 
390 	/* Change the attributes. */
391 
392 	iap->ia_valid |= ATTR_CTIME;
393 
394 	err = nfserr_notsync;
395 	if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
396 		fh_lock(fhp);
397 		host_err = notify_change(dentry, iap);
398 		err = nfserrno(host_err);
399 		fh_unlock(fhp);
400 	}
401 	if (size_change)
402 		put_write_access(inode);
403 	if (!err)
404 		if (EX_ISSYNC(fhp->fh_export))
405 			write_inode_now(inode, 1);
406 out:
407 	return err;
408 
409 out_nfserr:
410 	err = nfserrno(host_err);
411 	goto out;
412 }
413 
414 #if defined(CONFIG_NFSD_V2_ACL) || \
415     defined(CONFIG_NFSD_V3_ACL) || \
416     defined(CONFIG_NFSD_V4)
417 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
418 {
419 	ssize_t buflen;
420 	ssize_t ret;
421 
422 	buflen = vfs_getxattr(dentry, key, NULL, 0);
423 	if (buflen <= 0)
424 		return buflen;
425 
426 	*buf = kmalloc(buflen, GFP_KERNEL);
427 	if (!*buf)
428 		return -ENOMEM;
429 
430 	ret = vfs_getxattr(dentry, key, *buf, buflen);
431 	if (ret < 0)
432 		kfree(*buf);
433 	return ret;
434 }
435 #endif
436 
437 #if defined(CONFIG_NFSD_V4)
438 static int
439 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
440 {
441 	int len;
442 	size_t buflen;
443 	char *buf = NULL;
444 	int error = 0;
445 
446 	buflen = posix_acl_xattr_size(pacl->a_count);
447 	buf = kmalloc(buflen, GFP_KERNEL);
448 	error = -ENOMEM;
449 	if (buf == NULL)
450 		goto out;
451 
452 	len = posix_acl_to_xattr(pacl, buf, buflen);
453 	if (len < 0) {
454 		error = len;
455 		goto out;
456 	}
457 
458 	error = vfs_setxattr(dentry, key, buf, len, 0);
459 out:
460 	kfree(buf);
461 	return error;
462 }
463 
464 __be32
465 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
466     struct nfs4_acl *acl)
467 {
468 	__be32 error;
469 	int host_error;
470 	struct dentry *dentry;
471 	struct inode *inode;
472 	struct posix_acl *pacl = NULL, *dpacl = NULL;
473 	unsigned int flags = 0;
474 
475 	/* Get inode */
476 	error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
477 	if (error)
478 		return error;
479 
480 	dentry = fhp->fh_dentry;
481 	inode = dentry->d_inode;
482 	if (S_ISDIR(inode->i_mode))
483 		flags = NFS4_ACL_DIR;
484 
485 	host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
486 	if (host_error == -EINVAL) {
487 		return nfserr_attrnotsupp;
488 	} else if (host_error < 0)
489 		goto out_nfserr;
490 
491 	host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
492 	if (host_error < 0)
493 		goto out_release;
494 
495 	if (S_ISDIR(inode->i_mode))
496 		host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
497 
498 out_release:
499 	posix_acl_release(pacl);
500 	posix_acl_release(dpacl);
501 out_nfserr:
502 	if (host_error == -EOPNOTSUPP)
503 		return nfserr_attrnotsupp;
504 	else
505 		return nfserrno(host_error);
506 }
507 
508 static struct posix_acl *
509 _get_posix_acl(struct dentry *dentry, char *key)
510 {
511 	void *buf = NULL;
512 	struct posix_acl *pacl = NULL;
513 	int buflen;
514 
515 	buflen = nfsd_getxattr(dentry, key, &buf);
516 	if (!buflen)
517 		buflen = -ENODATA;
518 	if (buflen <= 0)
519 		return ERR_PTR(buflen);
520 
521 	pacl = posix_acl_from_xattr(buf, buflen);
522 	kfree(buf);
523 	return pacl;
524 }
525 
526 int
527 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
528 {
529 	struct inode *inode = dentry->d_inode;
530 	int error = 0;
531 	struct posix_acl *pacl = NULL, *dpacl = NULL;
532 	unsigned int flags = 0;
533 
534 	pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
535 	if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
536 		pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
537 	if (IS_ERR(pacl)) {
538 		error = PTR_ERR(pacl);
539 		pacl = NULL;
540 		goto out;
541 	}
542 
543 	if (S_ISDIR(inode->i_mode)) {
544 		dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
545 		if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
546 			dpacl = NULL;
547 		else if (IS_ERR(dpacl)) {
548 			error = PTR_ERR(dpacl);
549 			dpacl = NULL;
550 			goto out;
551 		}
552 		flags = NFS4_ACL_DIR;
553 	}
554 
555 	*acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
556 	if (IS_ERR(*acl)) {
557 		error = PTR_ERR(*acl);
558 		*acl = NULL;
559 	}
560  out:
561 	posix_acl_release(pacl);
562 	posix_acl_release(dpacl);
563 	return error;
564 }
565 
566 #endif /* defined(CONFIG_NFS_V4) */
567 
568 #ifdef CONFIG_NFSD_V3
569 /*
570  * Check server access rights to a file system object
571  */
572 struct accessmap {
573 	u32		access;
574 	int		how;
575 };
576 static struct accessmap	nfs3_regaccess[] = {
577     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
578     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
579     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_TRUNC	},
580     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE			},
581 
582     {	0,			0				}
583 };
584 
585 static struct accessmap	nfs3_diraccess[] = {
586     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
587     {	NFS3_ACCESS_LOOKUP,	NFSD_MAY_EXEC			},
588     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
589     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_EXEC|NFSD_MAY_WRITE	},
590     {	NFS3_ACCESS_DELETE,	NFSD_MAY_REMOVE			},
591 
592     {	0,			0				}
593 };
594 
595 static struct accessmap	nfs3_anyaccess[] = {
596 	/* Some clients - Solaris 2.6 at least, make an access call
597 	 * to the server to check for access for things like /dev/null
598 	 * (which really, the server doesn't care about).  So
599 	 * We provide simple access checking for them, looking
600 	 * mainly at mode bits, and we make sure to ignore read-only
601 	 * filesystem checks
602 	 */
603     {	NFS3_ACCESS_READ,	NFSD_MAY_READ			},
604     {	NFS3_ACCESS_EXECUTE,	NFSD_MAY_EXEC			},
605     {	NFS3_ACCESS_MODIFY,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
606     {	NFS3_ACCESS_EXTEND,	NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS	},
607 
608     {	0,			0				}
609 };
610 
611 __be32
612 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
613 {
614 	struct accessmap	*map;
615 	struct svc_export	*export;
616 	struct dentry		*dentry;
617 	u32			query, result = 0, sresult = 0;
618 	__be32			error;
619 
620 	error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
621 	if (error)
622 		goto out;
623 
624 	export = fhp->fh_export;
625 	dentry = fhp->fh_dentry;
626 
627 	if (S_ISREG(dentry->d_inode->i_mode))
628 		map = nfs3_regaccess;
629 	else if (S_ISDIR(dentry->d_inode->i_mode))
630 		map = nfs3_diraccess;
631 	else
632 		map = nfs3_anyaccess;
633 
634 
635 	query = *access;
636 	for  (; map->access; map++) {
637 		if (map->access & query) {
638 			__be32 err2;
639 
640 			sresult |= map->access;
641 
642 			err2 = nfsd_permission(rqstp, export, dentry, map->how);
643 			switch (err2) {
644 			case nfs_ok:
645 				result |= map->access;
646 				break;
647 
648 			/* the following error codes just mean the access was not allowed,
649 			 * rather than an error occurred */
650 			case nfserr_rofs:
651 			case nfserr_acces:
652 			case nfserr_perm:
653 				/* simply don't "or" in the access bit. */
654 				break;
655 			default:
656 				error = err2;
657 				goto out;
658 			}
659 		}
660 	}
661 	*access = result;
662 	if (supported)
663 		*supported = sresult;
664 
665  out:
666 	return error;
667 }
668 #endif /* CONFIG_NFSD_V3 */
669 
670 
671 
672 /*
673  * Open an existing file or directory.
674  * The access argument indicates the type of open (read/write/lock)
675  * N.B. After this call fhp needs an fh_put
676  */
677 __be32
678 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
679 			int access, struct file **filp)
680 {
681 	struct dentry	*dentry;
682 	struct inode	*inode;
683 	int		flags = O_RDONLY|O_LARGEFILE;
684 	__be32		err;
685 	int		host_err;
686 
687 	validate_process_creds();
688 
689 	/*
690 	 * If we get here, then the client has already done an "open",
691 	 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
692 	 * in case a chmod has now revoked permission.
693 	 */
694 	err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
695 	if (err)
696 		goto out;
697 
698 	dentry = fhp->fh_dentry;
699 	inode = dentry->d_inode;
700 
701 	/* Disallow write access to files with the append-only bit set
702 	 * or any access when mandatory locking enabled
703 	 */
704 	err = nfserr_perm;
705 	if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
706 		goto out;
707 	/*
708 	 * We must ignore files (but only files) which might have mandatory
709 	 * locks on them because there is no way to know if the accesser has
710 	 * the lock.
711 	 */
712 	if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
713 		goto out;
714 
715 	if (!inode->i_fop)
716 		goto out;
717 
718 	/*
719 	 * Check to see if there are any leases on this file.
720 	 * This may block while leases are broken.
721 	 */
722 	host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
723 	if (host_err == -EWOULDBLOCK)
724 		host_err = -ETIMEDOUT;
725 	if (host_err) /* NOMEM or WOULDBLOCK */
726 		goto out_nfserr;
727 
728 	if (access & NFSD_MAY_WRITE) {
729 		if (access & NFSD_MAY_READ)
730 			flags = O_RDWR|O_LARGEFILE;
731 		else
732 			flags = O_WRONLY|O_LARGEFILE;
733 
734 		vfs_dq_init(inode);
735 	}
736 	*filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
737 			    flags, current_cred());
738 	if (IS_ERR(*filp))
739 		host_err = PTR_ERR(*filp);
740 	else
741 		ima_counts_get(*filp);
742 out_nfserr:
743 	err = nfserrno(host_err);
744 out:
745 	validate_process_creds();
746 	return err;
747 }
748 
749 /*
750  * Close a file.
751  */
752 void
753 nfsd_close(struct file *filp)
754 {
755 	fput(filp);
756 }
757 
758 /*
759  * Sync a file
760  * As this calls fsync (not fdatasync) there is no need for a write_inode
761  * after it.
762  */
763 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
764 			      const struct file_operations *fop)
765 {
766 	struct inode *inode = dp->d_inode;
767 	int (*fsync) (struct file *, struct dentry *, int);
768 	int err;
769 
770 	err = filemap_fdatawrite(inode->i_mapping);
771 	if (err == 0 && fop && (fsync = fop->fsync))
772 		err = fsync(filp, dp, 0);
773 	if (err == 0)
774 		err = filemap_fdatawait(inode->i_mapping);
775 
776 	return err;
777 }
778 
779 static int
780 nfsd_sync(struct file *filp)
781 {
782         int err;
783 	struct inode *inode = filp->f_path.dentry->d_inode;
784 	dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
785 	mutex_lock(&inode->i_mutex);
786 	err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
787 	mutex_unlock(&inode->i_mutex);
788 
789 	return err;
790 }
791 
792 int
793 nfsd_sync_dir(struct dentry *dp)
794 {
795 	return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
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 	struct raparms	*ra;
907 	mm_segment_t	oldfs;
908 	__be32		err;
909 	int		host_err;
910 
911 	err = nfserr_perm;
912 	inode = file->f_path.dentry->d_inode;
913 
914 	if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
915 		goto out;
916 
917 	/* Get readahead parameters */
918 	ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
919 
920 	if (ra && ra->p_set)
921 		file->f_ra = ra->p_ra;
922 
923 	if (file->f_op->splice_read && rqstp->rq_splice_ok) {
924 		struct splice_desc sd = {
925 			.len		= 0,
926 			.total_len	= *count,
927 			.pos		= offset,
928 			.u.data		= rqstp,
929 		};
930 
931 		rqstp->rq_resused = 1;
932 		host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
933 	} else {
934 		oldfs = get_fs();
935 		set_fs(KERNEL_DS);
936 		host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
937 		set_fs(oldfs);
938 	}
939 
940 	/* Write back readahead params */
941 	if (ra) {
942 		struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
943 		spin_lock(&rab->pb_lock);
944 		ra->p_ra = file->f_ra;
945 		ra->p_set = 1;
946 		ra->p_count--;
947 		spin_unlock(&rab->pb_lock);
948 	}
949 
950 	if (host_err >= 0) {
951 		nfsdstats.io_read += host_err;
952 		*count = host_err;
953 		err = 0;
954 		fsnotify_access(file->f_path.dentry);
955 	} else
956 		err = nfserrno(host_err);
957 out:
958 	return err;
959 }
960 
961 static void kill_suid(struct dentry *dentry)
962 {
963 	struct iattr	ia;
964 	ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
965 
966 	mutex_lock(&dentry->d_inode->i_mutex);
967 	notify_change(dentry, &ia);
968 	mutex_unlock(&dentry->d_inode->i_mutex);
969 }
970 
971 /*
972  * Gathered writes: If another process is currently writing to the file,
973  * there's a high chance this is another nfsd (triggered by a bulk write
974  * from a client's biod). Rather than syncing the file with each write
975  * request, we sleep for 10 msec.
976  *
977  * I don't know if this roughly approximates C. Juszak's idea of
978  * gathered writes, but it's a nice and simple solution (IMHO), and it
979  * seems to work:-)
980  *
981  * Note: we do this only in the NFSv2 case, since v3 and higher have a
982  * better tool (separate unstable writes and commits) for solving this
983  * problem.
984  */
985 static int wait_for_concurrent_writes(struct file *file)
986 {
987 	struct inode *inode = file->f_path.dentry->d_inode;
988 	static ino_t last_ino;
989 	static dev_t last_dev;
990 	int err = 0;
991 
992 	if (atomic_read(&inode->i_writecount) > 1
993 	    || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
994 		dprintk("nfsd: write defer %d\n", task_pid_nr(current));
995 		msleep(10);
996 		dprintk("nfsd: write resume %d\n", task_pid_nr(current));
997 	}
998 
999 	if (inode->i_state & I_DIRTY) {
1000 		dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1001 		err = nfsd_sync(file);
1002 	}
1003 	last_ino = inode->i_ino;
1004 	last_dev = inode->i_sb->s_dev;
1005 	return err;
1006 }
1007 
1008 static __be32
1009 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1010 				loff_t offset, struct kvec *vec, int vlen,
1011 				unsigned long *cnt, int *stablep)
1012 {
1013 	struct svc_export	*exp;
1014 	struct dentry		*dentry;
1015 	struct inode		*inode;
1016 	mm_segment_t		oldfs;
1017 	__be32			err = 0;
1018 	int			host_err;
1019 	int			stable = *stablep;
1020 	int			use_wgather;
1021 
1022 #ifdef MSNFS
1023 	err = nfserr_perm;
1024 
1025 	if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1026 		(!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1027 		goto out;
1028 #endif
1029 
1030 	dentry = file->f_path.dentry;
1031 	inode = dentry->d_inode;
1032 	exp   = fhp->fh_export;
1033 
1034 	/*
1035 	 * Request sync writes if
1036 	 *  -	the sync export option has been set, or
1037 	 *  -	the client requested O_SYNC behavior (NFSv3 feature).
1038 	 *  -   The file system doesn't support fsync().
1039 	 * When NFSv2 gathered writes have been configured for this volume,
1040 	 * flushing the data to disk is handled separately below.
1041 	 */
1042 	use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1043 
1044 	if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1045 	       stable = 2;
1046 	       *stablep = 2; /* FILE_SYNC */
1047 	}
1048 
1049 	if (!EX_ISSYNC(exp))
1050 		stable = 0;
1051 	if (stable && !use_wgather) {
1052 		spin_lock(&file->f_lock);
1053 		file->f_flags |= O_SYNC;
1054 		spin_unlock(&file->f_lock);
1055 	}
1056 
1057 	/* Write the data. */
1058 	oldfs = get_fs(); set_fs(KERNEL_DS);
1059 	host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1060 	set_fs(oldfs);
1061 	if (host_err < 0)
1062 		goto out_nfserr;
1063 	*cnt = host_err;
1064 	nfsdstats.io_write += host_err;
1065 	fsnotify_modify(file->f_path.dentry);
1066 
1067 	/* clear setuid/setgid flag after write */
1068 	if (inode->i_mode & (S_ISUID | S_ISGID))
1069 		kill_suid(dentry);
1070 
1071 	if (stable && use_wgather)
1072 		host_err = wait_for_concurrent_writes(file);
1073 
1074 out_nfserr:
1075 	dprintk("nfsd: write complete host_err=%d\n", host_err);
1076 	if (host_err >= 0)
1077 		err = 0;
1078 	else
1079 		err = nfserrno(host_err);
1080 out:
1081 	return err;
1082 }
1083 
1084 /*
1085  * Read data from a file. count must contain the requested read count
1086  * on entry. On return, *count contains the number of bytes actually read.
1087  * N.B. After this call fhp needs an fh_put
1088  */
1089 __be32
1090 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1091 		loff_t offset, struct kvec *vec, int vlen,
1092 		unsigned long *count)
1093 {
1094 	__be32		err;
1095 
1096 	if (file) {
1097 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1098 				NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1099 		if (err)
1100 			goto out;
1101 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1102 	} else {
1103 		err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1104 		if (err)
1105 			goto out;
1106 		err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1107 		nfsd_close(file);
1108 	}
1109 out:
1110 	return err;
1111 }
1112 
1113 /*
1114  * Write data to a file.
1115  * The stable flag requests synchronous writes.
1116  * N.B. After this call fhp needs an fh_put
1117  */
1118 __be32
1119 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1120 		loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1121 		int *stablep)
1122 {
1123 	__be32			err = 0;
1124 
1125 	if (file) {
1126 		err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1127 				NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1128 		if (err)
1129 			goto out;
1130 		err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1131 				stablep);
1132 	} else {
1133 		err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1134 		if (err)
1135 			goto out;
1136 
1137 		if (cnt)
1138 			err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1139 					     cnt, stablep);
1140 		nfsd_close(file);
1141 	}
1142 out:
1143 	return err;
1144 }
1145 
1146 #ifdef CONFIG_NFSD_V3
1147 /*
1148  * Commit all pending writes to stable storage.
1149  * Strictly speaking, we could sync just the indicated file region here,
1150  * but there's currently no way we can ask the VFS to do so.
1151  *
1152  * Unfortunately we cannot lock the file to make sure we return full WCC
1153  * data to the client, as locking happens lower down in the filesystem.
1154  */
1155 __be32
1156 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1157                loff_t offset, unsigned long count)
1158 {
1159 	struct file	*file;
1160 	__be32		err;
1161 
1162 	if ((u64)count > ~(u64)offset)
1163 		return nfserr_inval;
1164 
1165 	err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1166 	if (err)
1167 		return err;
1168 	if (EX_ISSYNC(fhp->fh_export)) {
1169 		if (file->f_op && file->f_op->fsync) {
1170 			err = nfserrno(nfsd_sync(file));
1171 		} else {
1172 			err = nfserr_notsupp;
1173 		}
1174 	}
1175 
1176 	nfsd_close(file);
1177 	return err;
1178 }
1179 #endif /* CONFIG_NFSD_V3 */
1180 
1181 static __be32
1182 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1183 			struct iattr *iap)
1184 {
1185 	/*
1186 	 * Mode has already been set earlier in create:
1187 	 */
1188 	iap->ia_valid &= ~ATTR_MODE;
1189 	/*
1190 	 * Setting uid/gid works only for root.  Irix appears to
1191 	 * send along the gid on create when it tries to implement
1192 	 * setgid directories via NFS:
1193 	 */
1194 	if (current_fsuid() != 0)
1195 		iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1196 	if (iap->ia_valid)
1197 		return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1198 	return 0;
1199 }
1200 
1201 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1202  * setting size to 0 may fail for some specific file systems by the permission
1203  * checking which requires WRITE permission but the mode is 000.
1204  * we ignore the resizing(to 0) on the just new created file, since the size is
1205  * 0 after file created.
1206  *
1207  * call this only after vfs_create() is called.
1208  * */
1209 static void
1210 nfsd_check_ignore_resizing(struct iattr *iap)
1211 {
1212 	if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1213 		iap->ia_valid &= ~ATTR_SIZE;
1214 }
1215 
1216 /*
1217  * Create a file (regular, directory, device, fifo); UNIX sockets
1218  * not yet implemented.
1219  * If the response fh has been verified, the parent directory should
1220  * already be locked. Note that the parent directory is left locked.
1221  *
1222  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1223  */
1224 __be32
1225 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1226 		char *fname, int flen, struct iattr *iap,
1227 		int type, dev_t rdev, struct svc_fh *resfhp)
1228 {
1229 	struct dentry	*dentry, *dchild = NULL;
1230 	struct inode	*dirp;
1231 	__be32		err;
1232 	__be32		err2;
1233 	int		host_err;
1234 
1235 	err = nfserr_perm;
1236 	if (!flen)
1237 		goto out;
1238 	err = nfserr_exist;
1239 	if (isdotent(fname, flen))
1240 		goto out;
1241 
1242 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1243 	if (err)
1244 		goto out;
1245 
1246 	dentry = fhp->fh_dentry;
1247 	dirp = dentry->d_inode;
1248 
1249 	err = nfserr_notdir;
1250 	if (!dirp->i_op->lookup)
1251 		goto out;
1252 	/*
1253 	 * Check whether the response file handle has been verified yet.
1254 	 * If it has, the parent directory should already be locked.
1255 	 */
1256 	if (!resfhp->fh_dentry) {
1257 		/* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1258 		fh_lock_nested(fhp, I_MUTEX_PARENT);
1259 		dchild = lookup_one_len(fname, dentry, flen);
1260 		host_err = PTR_ERR(dchild);
1261 		if (IS_ERR(dchild))
1262 			goto out_nfserr;
1263 		err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1264 		if (err)
1265 			goto out;
1266 	} else {
1267 		/* called from nfsd_proc_create */
1268 		dchild = dget(resfhp->fh_dentry);
1269 		if (!fhp->fh_locked) {
1270 			/* not actually possible */
1271 			printk(KERN_ERR
1272 				"nfsd_create: parent %s/%s not locked!\n",
1273 				dentry->d_parent->d_name.name,
1274 				dentry->d_name.name);
1275 			err = nfserr_io;
1276 			goto out;
1277 		}
1278 	}
1279 	/*
1280 	 * Make sure the child dentry is still negative ...
1281 	 */
1282 	err = nfserr_exist;
1283 	if (dchild->d_inode) {
1284 		dprintk("nfsd_create: dentry %s/%s not negative!\n",
1285 			dentry->d_name.name, dchild->d_name.name);
1286 		goto out;
1287 	}
1288 
1289 	if (!(iap->ia_valid & ATTR_MODE))
1290 		iap->ia_mode = 0;
1291 	iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1292 
1293 	err = nfserr_inval;
1294 	if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1295 		printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1296 		       type);
1297 		goto out;
1298 	}
1299 
1300 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1301 	if (host_err)
1302 		goto out_nfserr;
1303 
1304 	/*
1305 	 * Get the dir op function pointer.
1306 	 */
1307 	err = 0;
1308 	switch (type) {
1309 	case S_IFREG:
1310 		host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1311 		if (!host_err)
1312 			nfsd_check_ignore_resizing(iap);
1313 		break;
1314 	case S_IFDIR:
1315 		host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1316 		break;
1317 	case S_IFCHR:
1318 	case S_IFBLK:
1319 	case S_IFIFO:
1320 	case S_IFSOCK:
1321 		host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1322 		break;
1323 	}
1324 	if (host_err < 0) {
1325 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1326 		goto out_nfserr;
1327 	}
1328 
1329 	if (EX_ISSYNC(fhp->fh_export)) {
1330 		err = nfserrno(nfsd_sync_dir(dentry));
1331 		write_inode_now(dchild->d_inode, 1);
1332 	}
1333 
1334 	err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1335 	if (err2)
1336 		err = err2;
1337 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1338 	/*
1339 	 * Update the file handle to get the new inode info.
1340 	 */
1341 	if (!err)
1342 		err = fh_update(resfhp);
1343 out:
1344 	if (dchild && !IS_ERR(dchild))
1345 		dput(dchild);
1346 	return err;
1347 
1348 out_nfserr:
1349 	err = nfserrno(host_err);
1350 	goto out;
1351 }
1352 
1353 #ifdef CONFIG_NFSD_V3
1354 /*
1355  * NFSv3 version of nfsd_create
1356  */
1357 __be32
1358 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1359 		char *fname, int flen, struct iattr *iap,
1360 		struct svc_fh *resfhp, int createmode, u32 *verifier,
1361 	        int *truncp, int *created)
1362 {
1363 	struct dentry	*dentry, *dchild = NULL;
1364 	struct inode	*dirp;
1365 	__be32		err;
1366 	__be32		err2;
1367 	int		host_err;
1368 	__u32		v_mtime=0, v_atime=0;
1369 
1370 	err = nfserr_perm;
1371 	if (!flen)
1372 		goto out;
1373 	err = nfserr_exist;
1374 	if (isdotent(fname, flen))
1375 		goto out;
1376 	if (!(iap->ia_valid & ATTR_MODE))
1377 		iap->ia_mode = 0;
1378 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1379 	if (err)
1380 		goto out;
1381 
1382 	dentry = fhp->fh_dentry;
1383 	dirp = dentry->d_inode;
1384 
1385 	/* Get all the sanity checks out of the way before
1386 	 * we lock the parent. */
1387 	err = nfserr_notdir;
1388 	if (!dirp->i_op->lookup)
1389 		goto out;
1390 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1391 
1392 	/*
1393 	 * Compose the response file handle.
1394 	 */
1395 	dchild = lookup_one_len(fname, dentry, flen);
1396 	host_err = PTR_ERR(dchild);
1397 	if (IS_ERR(dchild))
1398 		goto out_nfserr;
1399 
1400 	err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1401 	if (err)
1402 		goto out;
1403 
1404 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1405 		/* solaris7 gets confused (bugid 4218508) if these have
1406 		 * the high bit set, so just clear the high bits. If this is
1407 		 * ever changed to use different attrs for storing the
1408 		 * verifier, then do_open_lookup() will also need to be fixed
1409 		 * accordingly.
1410 		 */
1411 		v_mtime = verifier[0]&0x7fffffff;
1412 		v_atime = verifier[1]&0x7fffffff;
1413 	}
1414 
1415 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1416 	if (host_err)
1417 		goto out_nfserr;
1418 	if (dchild->d_inode) {
1419 		err = 0;
1420 
1421 		switch (createmode) {
1422 		case NFS3_CREATE_UNCHECKED:
1423 			if (! S_ISREG(dchild->d_inode->i_mode))
1424 				err = nfserr_exist;
1425 			else if (truncp) {
1426 				/* in nfsv4, we need to treat this case a little
1427 				 * differently.  we don't want to truncate the
1428 				 * file now; this would be wrong if the OPEN
1429 				 * fails for some other reason.  furthermore,
1430 				 * if the size is nonzero, we should ignore it
1431 				 * according to spec!
1432 				 */
1433 				*truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1434 			}
1435 			else {
1436 				iap->ia_valid &= ATTR_SIZE;
1437 				goto set_attr;
1438 			}
1439 			break;
1440 		case NFS3_CREATE_EXCLUSIVE:
1441 			if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1442 			    && dchild->d_inode->i_atime.tv_sec == v_atime
1443 			    && dchild->d_inode->i_size  == 0 )
1444 				break;
1445 			 /* fallthru */
1446 		case NFS3_CREATE_GUARDED:
1447 			err = nfserr_exist;
1448 		}
1449 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1450 		goto out;
1451 	}
1452 
1453 	host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1454 	if (host_err < 0) {
1455 		mnt_drop_write(fhp->fh_export->ex_path.mnt);
1456 		goto out_nfserr;
1457 	}
1458 	if (created)
1459 		*created = 1;
1460 
1461 	if (EX_ISSYNC(fhp->fh_export)) {
1462 		err = nfserrno(nfsd_sync_dir(dentry));
1463 		/* setattr will sync the child (or not) */
1464 	}
1465 
1466 	nfsd_check_ignore_resizing(iap);
1467 
1468 	if (createmode == NFS3_CREATE_EXCLUSIVE) {
1469 		/* Cram the verifier into atime/mtime */
1470 		iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1471 			| ATTR_MTIME_SET|ATTR_ATIME_SET;
1472 		/* XXX someone who knows this better please fix it for nsec */
1473 		iap->ia_mtime.tv_sec = v_mtime;
1474 		iap->ia_atime.tv_sec = v_atime;
1475 		iap->ia_mtime.tv_nsec = 0;
1476 		iap->ia_atime.tv_nsec = 0;
1477 	}
1478 
1479  set_attr:
1480 	err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1481 	if (err2)
1482 		err = err2;
1483 
1484 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1485 	/*
1486 	 * Update the filehandle to get the new inode info.
1487 	 */
1488 	if (!err)
1489 		err = fh_update(resfhp);
1490 
1491  out:
1492 	fh_unlock(fhp);
1493 	if (dchild && !IS_ERR(dchild))
1494 		dput(dchild);
1495  	return err;
1496 
1497  out_nfserr:
1498 	err = nfserrno(host_err);
1499 	goto out;
1500 }
1501 #endif /* CONFIG_NFSD_V3 */
1502 
1503 /*
1504  * Read a symlink. On entry, *lenp must contain the maximum path length that
1505  * fits into the buffer. On return, it contains the true length.
1506  * N.B. After this call fhp needs an fh_put
1507  */
1508 __be32
1509 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1510 {
1511 	struct dentry	*dentry;
1512 	struct inode	*inode;
1513 	mm_segment_t	oldfs;
1514 	__be32		err;
1515 	int		host_err;
1516 
1517 	err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1518 	if (err)
1519 		goto out;
1520 
1521 	dentry = fhp->fh_dentry;
1522 	inode = dentry->d_inode;
1523 
1524 	err = nfserr_inval;
1525 	if (!inode->i_op->readlink)
1526 		goto out;
1527 
1528 	touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1529 	/* N.B. Why does this call need a get_fs()??
1530 	 * Remove the set_fs and watch the fireworks:-) --okir
1531 	 */
1532 
1533 	oldfs = get_fs(); set_fs(KERNEL_DS);
1534 	host_err = inode->i_op->readlink(dentry, buf, *lenp);
1535 	set_fs(oldfs);
1536 
1537 	if (host_err < 0)
1538 		goto out_nfserr;
1539 	*lenp = host_err;
1540 	err = 0;
1541 out:
1542 	return err;
1543 
1544 out_nfserr:
1545 	err = nfserrno(host_err);
1546 	goto out;
1547 }
1548 
1549 /*
1550  * Create a symlink and look up its inode
1551  * N.B. After this call _both_ fhp and resfhp need an fh_put
1552  */
1553 __be32
1554 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1555 				char *fname, int flen,
1556 				char *path,  int plen,
1557 				struct svc_fh *resfhp,
1558 				struct iattr *iap)
1559 {
1560 	struct dentry	*dentry, *dnew;
1561 	__be32		err, cerr;
1562 	int		host_err;
1563 
1564 	err = nfserr_noent;
1565 	if (!flen || !plen)
1566 		goto out;
1567 	err = nfserr_exist;
1568 	if (isdotent(fname, flen))
1569 		goto out;
1570 
1571 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1572 	if (err)
1573 		goto out;
1574 	fh_lock(fhp);
1575 	dentry = fhp->fh_dentry;
1576 	dnew = lookup_one_len(fname, dentry, flen);
1577 	host_err = PTR_ERR(dnew);
1578 	if (IS_ERR(dnew))
1579 		goto out_nfserr;
1580 
1581 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1582 	if (host_err)
1583 		goto out_nfserr;
1584 
1585 	if (unlikely(path[plen] != 0)) {
1586 		char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1587 		if (path_alloced == NULL)
1588 			host_err = -ENOMEM;
1589 		else {
1590 			strncpy(path_alloced, path, plen);
1591 			path_alloced[plen] = 0;
1592 			host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1593 			kfree(path_alloced);
1594 		}
1595 	} else
1596 		host_err = vfs_symlink(dentry->d_inode, dnew, path);
1597 
1598 	if (!host_err) {
1599 		if (EX_ISSYNC(fhp->fh_export))
1600 			host_err = nfsd_sync_dir(dentry);
1601 	}
1602 	err = nfserrno(host_err);
1603 	fh_unlock(fhp);
1604 
1605 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1606 
1607 	cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1608 	dput(dnew);
1609 	if (err==0) err = cerr;
1610 out:
1611 	return err;
1612 
1613 out_nfserr:
1614 	err = nfserrno(host_err);
1615 	goto out;
1616 }
1617 
1618 /*
1619  * Create a hardlink
1620  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1621  */
1622 __be32
1623 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1624 				char *name, int len, struct svc_fh *tfhp)
1625 {
1626 	struct dentry	*ddir, *dnew, *dold;
1627 	struct inode	*dirp, *dest;
1628 	__be32		err;
1629 	int		host_err;
1630 
1631 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1632 	if (err)
1633 		goto out;
1634 	err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1635 	if (err)
1636 		goto out;
1637 
1638 	err = nfserr_perm;
1639 	if (!len)
1640 		goto out;
1641 	err = nfserr_exist;
1642 	if (isdotent(name, len))
1643 		goto out;
1644 
1645 	fh_lock_nested(ffhp, I_MUTEX_PARENT);
1646 	ddir = ffhp->fh_dentry;
1647 	dirp = ddir->d_inode;
1648 
1649 	dnew = lookup_one_len(name, ddir, len);
1650 	host_err = PTR_ERR(dnew);
1651 	if (IS_ERR(dnew))
1652 		goto out_nfserr;
1653 
1654 	dold = tfhp->fh_dentry;
1655 	dest = dold->d_inode;
1656 
1657 	host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1658 	if (host_err) {
1659 		err = nfserrno(host_err);
1660 		goto out_dput;
1661 	}
1662 	host_err = vfs_link(dold, dirp, dnew);
1663 	if (!host_err) {
1664 		if (EX_ISSYNC(ffhp->fh_export)) {
1665 			err = nfserrno(nfsd_sync_dir(ddir));
1666 			write_inode_now(dest, 1);
1667 		}
1668 		err = 0;
1669 	} else {
1670 		if (host_err == -EXDEV && rqstp->rq_vers == 2)
1671 			err = nfserr_acces;
1672 		else
1673 			err = nfserrno(host_err);
1674 	}
1675 	mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1676 out_dput:
1677 	dput(dnew);
1678 out_unlock:
1679 	fh_unlock(ffhp);
1680 out:
1681 	return err;
1682 
1683 out_nfserr:
1684 	err = nfserrno(host_err);
1685 	goto out_unlock;
1686 }
1687 
1688 /*
1689  * Rename a file
1690  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1691  */
1692 __be32
1693 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1694 			    struct svc_fh *tfhp, char *tname, int tlen)
1695 {
1696 	struct dentry	*fdentry, *tdentry, *odentry, *ndentry, *trap;
1697 	struct inode	*fdir, *tdir;
1698 	__be32		err;
1699 	int		host_err;
1700 
1701 	err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1702 	if (err)
1703 		goto out;
1704 	err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1705 	if (err)
1706 		goto out;
1707 
1708 	fdentry = ffhp->fh_dentry;
1709 	fdir = fdentry->d_inode;
1710 
1711 	tdentry = tfhp->fh_dentry;
1712 	tdir = tdentry->d_inode;
1713 
1714 	err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1715 	if (ffhp->fh_export != tfhp->fh_export)
1716 		goto out;
1717 
1718 	err = nfserr_perm;
1719 	if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1720 		goto out;
1721 
1722 	/* cannot use fh_lock as we need deadlock protective ordering
1723 	 * so do it by hand */
1724 	trap = lock_rename(tdentry, fdentry);
1725 	ffhp->fh_locked = tfhp->fh_locked = 1;
1726 	fill_pre_wcc(ffhp);
1727 	fill_pre_wcc(tfhp);
1728 
1729 	odentry = lookup_one_len(fname, fdentry, flen);
1730 	host_err = PTR_ERR(odentry);
1731 	if (IS_ERR(odentry))
1732 		goto out_nfserr;
1733 
1734 	host_err = -ENOENT;
1735 	if (!odentry->d_inode)
1736 		goto out_dput_old;
1737 	host_err = -EINVAL;
1738 	if (odentry == trap)
1739 		goto out_dput_old;
1740 
1741 	ndentry = lookup_one_len(tname, tdentry, tlen);
1742 	host_err = PTR_ERR(ndentry);
1743 	if (IS_ERR(ndentry))
1744 		goto out_dput_old;
1745 	host_err = -ENOTEMPTY;
1746 	if (ndentry == trap)
1747 		goto out_dput_new;
1748 
1749 	if (svc_msnfs(ffhp) &&
1750 		((atomic_read(&odentry->d_count) > 1)
1751 		 || (atomic_read(&ndentry->d_count) > 1))) {
1752 			host_err = -EPERM;
1753 			goto out_dput_new;
1754 	}
1755 
1756 	host_err = -EXDEV;
1757 	if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1758 		goto out_dput_new;
1759 	host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1760 	if (host_err)
1761 		goto out_dput_new;
1762 
1763 	host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1764 	if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1765 		host_err = nfsd_sync_dir(tdentry);
1766 		if (!host_err)
1767 			host_err = nfsd_sync_dir(fdentry);
1768 	}
1769 
1770 	mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1771 
1772  out_dput_new:
1773 	dput(ndentry);
1774  out_dput_old:
1775 	dput(odentry);
1776  out_nfserr:
1777 	err = nfserrno(host_err);
1778 
1779 	/* we cannot reply on fh_unlock on the two filehandles,
1780 	 * as that would do the wrong thing if the two directories
1781 	 * were the same, so again we do it by hand
1782 	 */
1783 	fill_post_wcc(ffhp);
1784 	fill_post_wcc(tfhp);
1785 	unlock_rename(tdentry, fdentry);
1786 	ffhp->fh_locked = tfhp->fh_locked = 0;
1787 
1788 out:
1789 	return err;
1790 }
1791 
1792 /*
1793  * Unlink a file or directory
1794  * N.B. After this call fhp needs an fh_put
1795  */
1796 __be32
1797 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1798 				char *fname, int flen)
1799 {
1800 	struct dentry	*dentry, *rdentry;
1801 	struct inode	*dirp;
1802 	__be32		err;
1803 	int		host_err;
1804 
1805 	err = nfserr_acces;
1806 	if (!flen || isdotent(fname, flen))
1807 		goto out;
1808 	err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1809 	if (err)
1810 		goto out;
1811 
1812 	fh_lock_nested(fhp, I_MUTEX_PARENT);
1813 	dentry = fhp->fh_dentry;
1814 	dirp = dentry->d_inode;
1815 
1816 	rdentry = lookup_one_len(fname, dentry, flen);
1817 	host_err = PTR_ERR(rdentry);
1818 	if (IS_ERR(rdentry))
1819 		goto out_nfserr;
1820 
1821 	if (!rdentry->d_inode) {
1822 		dput(rdentry);
1823 		err = nfserr_noent;
1824 		goto out;
1825 	}
1826 
1827 	if (!type)
1828 		type = rdentry->d_inode->i_mode & S_IFMT;
1829 
1830 	host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1831 	if (host_err)
1832 		goto out_nfserr;
1833 
1834 	if (type != S_IFDIR) { /* It's UNLINK */
1835 #ifdef MSNFS
1836 		if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1837 			(atomic_read(&rdentry->d_count) > 1)) {
1838 			host_err = -EPERM;
1839 		} else
1840 #endif
1841 		host_err = vfs_unlink(dirp, rdentry);
1842 	} else { /* It's RMDIR */
1843 		host_err = vfs_rmdir(dirp, rdentry);
1844 	}
1845 
1846 	dput(rdentry);
1847 
1848 	if (host_err)
1849 		goto out_drop;
1850 	if (EX_ISSYNC(fhp->fh_export))
1851 		host_err = nfsd_sync_dir(dentry);
1852 
1853 out_drop:
1854 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
1855 out_nfserr:
1856 	err = nfserrno(host_err);
1857 out:
1858 	return err;
1859 }
1860 
1861 /*
1862  * We do this buffering because we must not call back into the file
1863  * system's ->lookup() method from the filldir callback. That may well
1864  * deadlock a number of file systems.
1865  *
1866  * This is based heavily on the implementation of same in XFS.
1867  */
1868 struct buffered_dirent {
1869 	u64		ino;
1870 	loff_t		offset;
1871 	int		namlen;
1872 	unsigned int	d_type;
1873 	char		name[];
1874 };
1875 
1876 struct readdir_data {
1877 	char		*dirent;
1878 	size_t		used;
1879 	int		full;
1880 };
1881 
1882 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1883 				 loff_t offset, u64 ino, unsigned int d_type)
1884 {
1885 	struct readdir_data *buf = __buf;
1886 	struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1887 	unsigned int reclen;
1888 
1889 	reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1890 	if (buf->used + reclen > PAGE_SIZE) {
1891 		buf->full = 1;
1892 		return -EINVAL;
1893 	}
1894 
1895 	de->namlen = namlen;
1896 	de->offset = offset;
1897 	de->ino = ino;
1898 	de->d_type = d_type;
1899 	memcpy(de->name, name, namlen);
1900 	buf->used += reclen;
1901 
1902 	return 0;
1903 }
1904 
1905 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1906 				    struct readdir_cd *cdp, loff_t *offsetp)
1907 {
1908 	struct readdir_data buf;
1909 	struct buffered_dirent *de;
1910 	int host_err;
1911 	int size;
1912 	loff_t offset;
1913 
1914 	buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1915 	if (!buf.dirent)
1916 		return nfserrno(-ENOMEM);
1917 
1918 	offset = *offsetp;
1919 
1920 	while (1) {
1921 		struct inode *dir_inode = file->f_path.dentry->d_inode;
1922 		unsigned int reclen;
1923 
1924 		cdp->err = nfserr_eof; /* will be cleared on successful read */
1925 		buf.used = 0;
1926 		buf.full = 0;
1927 
1928 		host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1929 		if (buf.full)
1930 			host_err = 0;
1931 
1932 		if (host_err < 0)
1933 			break;
1934 
1935 		size = buf.used;
1936 
1937 		if (!size)
1938 			break;
1939 
1940 		/*
1941 		 * Various filldir functions may end up calling back into
1942 		 * lookup_one_len() and the file system's ->lookup() method.
1943 		 * These expect i_mutex to be held, as it would within readdir.
1944 		 */
1945 		host_err = mutex_lock_killable(&dir_inode->i_mutex);
1946 		if (host_err)
1947 			break;
1948 
1949 		de = (struct buffered_dirent *)buf.dirent;
1950 		while (size > 0) {
1951 			offset = de->offset;
1952 
1953 			if (func(cdp, de->name, de->namlen, de->offset,
1954 				 de->ino, de->d_type))
1955 				break;
1956 
1957 			if (cdp->err != nfs_ok)
1958 				break;
1959 
1960 			reclen = ALIGN(sizeof(*de) + de->namlen,
1961 				       sizeof(u64));
1962 			size -= reclen;
1963 			de = (struct buffered_dirent *)((char *)de + reclen);
1964 		}
1965 		mutex_unlock(&dir_inode->i_mutex);
1966 		if (size > 0) /* We bailed out early */
1967 			break;
1968 
1969 		offset = vfs_llseek(file, 0, SEEK_CUR);
1970 	}
1971 
1972 	free_page((unsigned long)(buf.dirent));
1973 
1974 	if (host_err)
1975 		return nfserrno(host_err);
1976 
1977 	*offsetp = offset;
1978 	return cdp->err;
1979 }
1980 
1981 /*
1982  * Read entries from a directory.
1983  * The  NFSv3/4 verifier we ignore for now.
1984  */
1985 __be32
1986 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1987 	     struct readdir_cd *cdp, filldir_t func)
1988 {
1989 	__be32		err;
1990 	struct file	*file;
1991 	loff_t		offset = *offsetp;
1992 
1993 	err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1994 	if (err)
1995 		goto out;
1996 
1997 	offset = vfs_llseek(file, offset, 0);
1998 	if (offset < 0) {
1999 		err = nfserrno((int)offset);
2000 		goto out_close;
2001 	}
2002 
2003 	err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2004 
2005 	if (err == nfserr_eof || err == nfserr_toosmall)
2006 		err = nfs_ok; /* can still be found in ->err */
2007 out_close:
2008 	nfsd_close(file);
2009 out:
2010 	return err;
2011 }
2012 
2013 /*
2014  * Get file system stats
2015  * N.B. After this call fhp needs an fh_put
2016  */
2017 __be32
2018 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2019 {
2020 	__be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2021 	if (!err && vfs_statfs(fhp->fh_dentry,stat))
2022 		err = nfserr_io;
2023 	return err;
2024 }
2025 
2026 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2027 {
2028 	return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2029 }
2030 
2031 /*
2032  * Check for a user's access permissions to this inode.
2033  */
2034 __be32
2035 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2036 					struct dentry *dentry, int acc)
2037 {
2038 	struct inode	*inode = dentry->d_inode;
2039 	struct path	path;
2040 	int		err;
2041 
2042 	if (acc == NFSD_MAY_NOP)
2043 		return 0;
2044 #if 0
2045 	dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2046 		acc,
2047 		(acc & NFSD_MAY_READ)?	" read"  : "",
2048 		(acc & NFSD_MAY_WRITE)?	" write" : "",
2049 		(acc & NFSD_MAY_EXEC)?	" exec"  : "",
2050 		(acc & NFSD_MAY_SATTR)?	" sattr" : "",
2051 		(acc & NFSD_MAY_TRUNC)?	" trunc" : "",
2052 		(acc & NFSD_MAY_LOCK)?	" lock"  : "",
2053 		(acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2054 		inode->i_mode,
2055 		IS_IMMUTABLE(inode)?	" immut" : "",
2056 		IS_APPEND(inode)?	" append" : "",
2057 		__mnt_is_readonly(exp->ex_path.mnt)?	" ro" : "");
2058 	dprintk("      owner %d/%d user %d/%d\n",
2059 		inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2060 #endif
2061 
2062 	/* Normally we reject any write/sattr etc access on a read-only file
2063 	 * system.  But if it is IRIX doing check on write-access for a
2064 	 * device special file, we ignore rofs.
2065 	 */
2066 	if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2067 		if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2068 			if (exp_rdonly(rqstp, exp) ||
2069 			    __mnt_is_readonly(exp->ex_path.mnt))
2070 				return nfserr_rofs;
2071 			if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2072 				return nfserr_perm;
2073 		}
2074 	if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2075 		return nfserr_perm;
2076 
2077 	if (acc & NFSD_MAY_LOCK) {
2078 		/* If we cannot rely on authentication in NLM requests,
2079 		 * just allow locks, otherwise require read permission, or
2080 		 * ownership
2081 		 */
2082 		if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2083 			return 0;
2084 		else
2085 			acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2086 	}
2087 	/*
2088 	 * The file owner always gets access permission for accesses that
2089 	 * would normally be checked at open time. This is to make
2090 	 * file access work even when the client has done a fchmod(fd, 0).
2091 	 *
2092 	 * However, `cp foo bar' should fail nevertheless when bar is
2093 	 * readonly. A sensible way to do this might be to reject all
2094 	 * attempts to truncate a read-only file, because a creat() call
2095 	 * always implies file truncation.
2096 	 * ... but this isn't really fair.  A process may reasonably call
2097 	 * ftruncate on an open file descriptor on a file with perm 000.
2098 	 * We must trust the client to do permission checking - using "ACCESS"
2099 	 * with NFSv3.
2100 	 */
2101 	if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2102 	    inode->i_uid == current_fsuid())
2103 		return 0;
2104 
2105 	/* This assumes  NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2106 	err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2107 
2108 	/* Allow read access to binaries even when mode 111 */
2109 	if (err == -EACCES && S_ISREG(inode->i_mode) &&
2110 	    acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2111 		err = inode_permission(inode, MAY_EXEC);
2112 	if (err)
2113 		goto nfsd_out;
2114 
2115 	/* Do integrity (permission) checking now, but defer incrementing
2116 	 * IMA counts to the actual file open.
2117 	 */
2118 	path.mnt = exp->ex_path.mnt;
2119 	path.dentry = dentry;
2120 	err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC),
2121 			     IMA_COUNT_LEAVE);
2122 nfsd_out:
2123 	return err? nfserrno(err) : 0;
2124 }
2125 
2126 void
2127 nfsd_racache_shutdown(void)
2128 {
2129 	struct raparms *raparm, *last_raparm;
2130 	unsigned int i;
2131 
2132 	dprintk("nfsd: freeing readahead buffers.\n");
2133 
2134 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2135 		raparm = raparm_hash[i].pb_head;
2136 		while(raparm) {
2137 			last_raparm = raparm;
2138 			raparm = raparm->p_next;
2139 			kfree(last_raparm);
2140 		}
2141 		raparm_hash[i].pb_head = NULL;
2142 	}
2143 }
2144 /*
2145  * Initialize readahead param cache
2146  */
2147 int
2148 nfsd_racache_init(int cache_size)
2149 {
2150 	int	i;
2151 	int	j = 0;
2152 	int	nperbucket;
2153 	struct raparms **raparm = NULL;
2154 
2155 
2156 	if (raparm_hash[0].pb_head)
2157 		return 0;
2158 	nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2159 	if (nperbucket < 2)
2160 		nperbucket = 2;
2161 	cache_size = nperbucket * RAPARM_HASH_SIZE;
2162 
2163 	dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2164 
2165 	for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2166 		spin_lock_init(&raparm_hash[i].pb_lock);
2167 
2168 		raparm = &raparm_hash[i].pb_head;
2169 		for (j = 0; j < nperbucket; j++) {
2170 			*raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2171 			if (!*raparm)
2172 				goto out_nomem;
2173 			raparm = &(*raparm)->p_next;
2174 		}
2175 		*raparm = NULL;
2176 	}
2177 
2178 	nfsdstats.ra_size = cache_size;
2179 	return 0;
2180 
2181 out_nomem:
2182 	dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2183 	nfsd_racache_shutdown();
2184 	return -ENOMEM;
2185 }
2186 
2187 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2188 struct posix_acl *
2189 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2190 {
2191 	struct inode *inode = fhp->fh_dentry->d_inode;
2192 	char *name;
2193 	void *value = NULL;
2194 	ssize_t size;
2195 	struct posix_acl *acl;
2196 
2197 	if (!IS_POSIXACL(inode))
2198 		return ERR_PTR(-EOPNOTSUPP);
2199 
2200 	switch (type) {
2201 	case ACL_TYPE_ACCESS:
2202 		name = POSIX_ACL_XATTR_ACCESS;
2203 		break;
2204 	case ACL_TYPE_DEFAULT:
2205 		name = POSIX_ACL_XATTR_DEFAULT;
2206 		break;
2207 	default:
2208 		return ERR_PTR(-EOPNOTSUPP);
2209 	}
2210 
2211 	size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2212 	if (size < 0)
2213 		return ERR_PTR(size);
2214 
2215 	acl = posix_acl_from_xattr(value, size);
2216 	kfree(value);
2217 	return acl;
2218 }
2219 
2220 int
2221 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2222 {
2223 	struct inode *inode = fhp->fh_dentry->d_inode;
2224 	char *name;
2225 	void *value = NULL;
2226 	size_t size;
2227 	int error;
2228 
2229 	if (!IS_POSIXACL(inode) ||
2230 	    !inode->i_op->setxattr || !inode->i_op->removexattr)
2231 		return -EOPNOTSUPP;
2232 	switch(type) {
2233 		case ACL_TYPE_ACCESS:
2234 			name = POSIX_ACL_XATTR_ACCESS;
2235 			break;
2236 		case ACL_TYPE_DEFAULT:
2237 			name = POSIX_ACL_XATTR_DEFAULT;
2238 			break;
2239 		default:
2240 			return -EOPNOTSUPP;
2241 	}
2242 
2243 	if (acl && acl->a_count) {
2244 		size = posix_acl_xattr_size(acl->a_count);
2245 		value = kmalloc(size, GFP_KERNEL);
2246 		if (!value)
2247 			return -ENOMEM;
2248 		error = posix_acl_to_xattr(acl, value, size);
2249 		if (error < 0)
2250 			goto getout;
2251 		size = error;
2252 	} else
2253 		size = 0;
2254 
2255 	error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2256 	if (error)
2257 		goto getout;
2258 	if (size)
2259 		error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2260 	else {
2261 		if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2262 			error = 0;
2263 		else {
2264 			error = vfs_removexattr(fhp->fh_dentry, name);
2265 			if (error == -ENODATA)
2266 				error = 0;
2267 		}
2268 	}
2269 	mnt_drop_write(fhp->fh_export->ex_path.mnt);
2270 
2271 getout:
2272 	kfree(value);
2273 	return error;
2274 }
2275 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
2276