xref: /linux/fs/nfsd/nfsfh.c (revision d141ec2825b4d3ec52f27c43bdd864090159273a)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * NFS server file handle treatment.
4  *
5  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
6  * Portions Copyright (C) 1999 G. Allen Morris III <gam3@acm.org>
7  * Extensive rewrite by Neil Brown <neilb@cse.unsw.edu.au> Southern-Spring 1999
8  * ... and again Southern-Winter 2001 to support export_operations
9  */
10 
11 #include <linux/exportfs.h>
12 
13 #include <linux/sunrpc/svcauth_gss.h>
14 #include <crypto/utils.h>
15 #include "nfsd.h"
16 #include "netns.h"
17 #include "stats.h"
18 #include "vfs.h"
19 #include "auth.h"
20 #include "trace.h"
21 
22 #define NFSDDBG_FACILITY		NFSDDBG_FH
23 
24 
25 /*
26  * our acceptability function.
27  * if NOSUBTREECHECK, accept anything
28  * if not, require that we can walk up to exp->ex_dentry
29  * doing some checks on the 'x' bits
30  */
nfsd_acceptable(void * expv,struct dentry * dentry)31 static int nfsd_acceptable(void *expv, struct dentry *dentry)
32 {
33 	struct svc_export *exp = expv;
34 	int rv;
35 	struct dentry *tdentry;
36 	struct dentry *parent;
37 
38 	if (exp->ex_flags & NFSEXP_NOSUBTREECHECK)
39 		return 1;
40 
41 	tdentry = dget(dentry);
42 	while (tdentry != exp->ex_path.dentry && !IS_ROOT(tdentry)) {
43 		/* make sure parents give x permission to user */
44 		int err;
45 		parent = dget_parent(tdentry);
46 		err = inode_permission(&nop_mnt_idmap,
47 				       d_inode(parent), MAY_EXEC);
48 		if (err < 0) {
49 			dput(parent);
50 			break;
51 		}
52 		dput(tdentry);
53 		tdentry = parent;
54 	}
55 	if (tdentry != exp->ex_path.dentry)
56 		dprintk("nfsd_acceptable failed at %p %pd\n", tdentry, tdentry);
57 	rv = (tdentry == exp->ex_path.dentry);
58 	dput(tdentry);
59 	return rv;
60 }
61 
62 /* Type check. The correct error return for type mismatches does not seem to be
63  * generally agreed upon. SunOS seems to use EISDIR if file isn't S_IFREG; a
64  * comment in the NFSv3 spec says this is incorrect (implementation notes for
65  * the write call).
66  */
67 static inline __be32
nfsd_mode_check(struct dentry * dentry,umode_t requested)68 nfsd_mode_check(struct dentry *dentry, umode_t requested)
69 {
70 	umode_t mode = d_inode(dentry)->i_mode & S_IFMT;
71 
72 	if (requested == 0) /* the caller doesn't care */
73 		return nfs_ok;
74 	if (mode == requested) {
75 		if (mode == S_IFDIR && !d_can_lookup(dentry))
76 			return nfserr_notdir;
77 		return nfs_ok;
78 	}
79 	if (mode == S_IFLNK) {
80 		if (requested == S_IFDIR)
81 			return nfserr_symlink_not_dir;
82 		return nfserr_symlink;
83 	}
84 	if (requested == S_IFDIR)
85 		return nfserr_notdir;
86 	if (mode == S_IFDIR)
87 		return nfserr_isdir;
88 	return nfserr_wrong_type;
89 }
90 
nfsd_originating_port_ok(struct svc_rqst * rqstp,struct svc_cred * cred,struct svc_export * exp)91 static bool nfsd_originating_port_ok(struct svc_rqst *rqstp,
92 				     struct svc_cred *cred,
93 				     struct svc_export *exp)
94 {
95 	if (nfsexp_flags(cred, exp) & NFSEXP_INSECURE_PORT)
96 		return true;
97 	/* We don't require gss requests to use low ports: */
98 	if (cred->cr_flavor >= RPC_AUTH_GSS)
99 		return true;
100 	return test_bit(RQ_SECURE, &rqstp->rq_flags);
101 }
102 
nfsd_setuser_and_check_port(struct svc_rqst * rqstp,struct svc_cred * cred,struct svc_export * exp)103 static __be32 nfsd_setuser_and_check_port(struct svc_rqst *rqstp,
104 					  struct svc_cred *cred,
105 					  struct svc_export *exp)
106 {
107 	/* Check if the request originated from a secure port. */
108 	if (rqstp && !nfsd_originating_port_ok(rqstp, cred, exp)) {
109 		if (IS_ENABLED(CONFIG_SUNRPC_DEBUG)) {
110 			char buf[RPC_MAX_ADDRBUFLEN];
111 
112 			dprintk("nfsd: request from insecure port %s!\n",
113 			        svc_print_addr(rqstp, buf, sizeof(buf)));
114 		}
115 		return nfserr_perm;
116 	}
117 
118 	/* Set user creds for this exportpoint */
119 	return nfserrno(nfsd_setuser(cred, exp));
120 }
121 
check_pseudo_root(struct dentry * dentry,struct svc_export * exp)122 static inline __be32 check_pseudo_root(struct dentry *dentry,
123 				       struct svc_export *exp)
124 {
125 	if (!(exp->ex_flags & NFSEXP_V4ROOT))
126 		return nfs_ok;
127 	/*
128 	 * We're exposing only the directories and symlinks that have to be
129 	 * traversed on the way to real exports:
130 	 */
131 	if (unlikely(!d_is_dir(dentry) &&
132 		     !d_is_symlink(dentry)))
133 		return nfserr_stale;
134 	/*
135 	 * A pseudoroot export gives permission to access only one
136 	 * single directory; the kernel has to make another upcall
137 	 * before granting access to anything else under it:
138 	 */
139 	if (unlikely(dentry != exp->ex_path.dentry))
140 		return nfserr_stale;
141 	return nfs_ok;
142 }
143 
144 /* Size of a file handle MAC, in 4-octet words */
145 #define FH_MAC_WORDS (sizeof(__le64) / 4)
146 
fh_append_mac(struct knfsd_fh * fh,int fh_maxsize,struct net * net)147 bool fh_append_mac(struct knfsd_fh *fh, int fh_maxsize, struct net *net)
148 {
149 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
150 	siphash_key_t *fh_key = nn->fh_key;
151 	__le64 hash;
152 
153 	if (!fh_key)
154 		goto out_no_key;
155 	if (fh->fh_size + sizeof(hash) > fh_maxsize)
156 		goto out_no_space;
157 
158 	hash = cpu_to_le64(siphash(&fh->fh_raw, fh->fh_size, fh_key));
159 	memcpy(&fh->fh_raw[fh->fh_size], &hash, sizeof(hash));
160 	fh->fh_size += sizeof(hash);
161 	return true;
162 
163 out_no_key:
164 	pr_warn_ratelimited("NFSD: unable to sign filehandles, fh_key not set.\n");
165 	return false;
166 
167 out_no_space:
168 	pr_warn_ratelimited("NFSD: unable to sign filehandles, fh_size %zu would be greater than fh_maxsize %d.\n",
169 			    fh->fh_size + sizeof(hash), fh_maxsize);
170 	return false;
171 }
172 
173 /*
174  * Verify that the filehandle's MAC was hashed from this filehandle
175  * given the server's fh_key:
176  */
fh_verify_mac(struct svc_fh * fhp,struct net * net)177 static bool fh_verify_mac(struct svc_fh *fhp, struct net *net)
178 {
179 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
180 	struct knfsd_fh *fh = &fhp->fh_handle;
181 	siphash_key_t *fh_key = nn->fh_key;
182 	__le64 hash;
183 
184 	if (!fh_key) {
185 		pr_warn_ratelimited("NFSD: unable to verify signed filehandles, fh_key not set.\n");
186 		return false;
187 	}
188 
189 	hash = cpu_to_le64(siphash(&fh->fh_raw, fh->fh_size - sizeof(hash),  fh_key));
190 	return crypto_memneq(&fh->fh_raw[fh->fh_size - sizeof(hash)],
191 					&hash, sizeof(hash)) == 0;
192 }
193 
194 /*
195  * Use the given filehandle to look up the corresponding export and
196  * dentry.  On success, the results are used to set fh_export and
197  * fh_dentry.
198  */
nfsd_set_fh_dentry(struct svc_rqst * rqstp,struct net * net,struct svc_cred * cred,struct auth_domain * client,struct auth_domain * gssclient,struct svc_fh * fhp)199 static __be32 nfsd_set_fh_dentry(struct svc_rqst *rqstp, struct net *net,
200 				 struct svc_cred *cred,
201 				 struct auth_domain *client,
202 				 struct auth_domain *gssclient,
203 				 struct svc_fh *fhp)
204 {
205 	struct knfsd_fh	*fh = &fhp->fh_handle;
206 	struct fid *fid = NULL;
207 	struct svc_export *exp;
208 	struct dentry *dentry;
209 	int fileid_type;
210 	int data_left = fh->fh_size/4;
211 	int len;
212 	__be32 error;
213 
214 	error = nfserr_badhandle;
215 	if (fh->fh_size == 0)
216 		return nfserr_nofilehandle;
217 
218 	if (fh->fh_version != 1)
219 		return error;
220 
221 	if (--data_left < 0)
222 		return error;
223 	if (fh->fh_auth_type != 0)
224 		return error;
225 	len = key_len(fh->fh_fsid_type) / 4;
226 	if (len == 0)
227 		return error;
228 	if (fh->fh_fsid_type == FSID_MAJOR_MINOR) {
229 		u32 *fsid = fh_fsid(fh);
230 
231 		/* deprecated, convert to type 3 */
232 		len = key_len(FSID_ENCODE_DEV)/4;
233 		fh->fh_fsid_type = FSID_ENCODE_DEV;
234 		/*
235 		 * struct knfsd_fh uses host-endian fields, which are
236 		 * sometimes used to hold net-endian values. This
237 		 * confuses sparse, so we must use __force here to
238 		 * keep it from complaining.
239 		 */
240 		fsid[0] = new_encode_dev(MKDEV(ntohl((__force __be32)fsid[0]),
241 					       ntohl((__force __be32)fsid[1])));
242 		fsid[1] = fsid[2];
243 	}
244 	data_left -= len;
245 	if (data_left < 0)
246 		return error;
247 	exp = rqst_exp_find(rqstp ? &rqstp->rq_chandle : NULL,
248 			    net, client, gssclient,
249 			    fh->fh_fsid_type, fh_fsid(fh));
250 	fid = (struct fid *)(fh_fsid(fh) + len);
251 
252 	error = nfserr_stale;
253 	if (IS_ERR(exp)) {
254 		trace_nfsd_set_fh_dentry_badexport(rqstp, fhp, PTR_ERR(exp));
255 
256 		if (PTR_ERR(exp) == -ENOENT)
257 			return error;
258 
259 		return nfserrno(PTR_ERR(exp));
260 	}
261 
262 	if (exp->ex_flags & NFSEXP_NOSUBTREECHECK) {
263 		/* Elevate privileges so that the lack of 'r' or 'x'
264 		 * permission on some parent directory will
265 		 * not stop exportfs_decode_fh from being able
266 		 * to reconnect a directory into the dentry cache.
267 		 * The same problem can affect "SUBTREECHECK" exports,
268 		 * but as nfsd_acceptable depends on correct
269 		 * access control settings being in effect, we cannot
270 		 * fix that case easily.
271 		 */
272 		struct cred *new = prepare_creds();
273 		if (!new) {
274 			error =  nfserrno(-ENOMEM);
275 			goto out;
276 		}
277 		new->cap_effective =
278 			cap_raise_nfsd_set(new->cap_effective,
279 					   new->cap_permitted);
280 		put_cred(override_creds(new));
281 	} else {
282 		error = nfsd_setuser_and_check_port(rqstp, cred, exp);
283 		if (error)
284 			goto out;
285 	}
286 
287 	/*
288 	 * Look up the dentry using the NFS file handle.
289 	 */
290 	fileid_type = fh->fh_fileid_type;
291 	error = nfserr_stale;
292 
293 	if (fileid_type == FILEID_ROOT) {
294 		/* We don't sign or verify the root, no per-file identity */
295 		dentry = dget(exp->ex_path.dentry);
296 	} else {
297 		if (exp->ex_flags & NFSEXP_SIGN_FH) {
298 			if (!fh_verify_mac(fhp, net)) {
299 				trace_nfsd_set_fh_dentry_badmac(rqstp, fhp, -ESTALE);
300 				goto out;
301 			}
302 			data_left -= FH_MAC_WORDS;
303 		}
304 
305 		dentry = exportfs_decode_fh_raw(exp->ex_path.mnt, fid,
306 						data_left, fileid_type, 0,
307 						nfsd_acceptable, exp);
308 		if (IS_ERR_OR_NULL(dentry)) {
309 			trace_nfsd_set_fh_dentry_badhandle(rqstp, fhp,
310 					dentry ?  PTR_ERR(dentry) : -ESTALE);
311 			switch (PTR_ERR(dentry)) {
312 			case -ENOMEM:
313 			case -ETIMEDOUT:
314 				break;
315 			default:
316 				dentry = ERR_PTR(-ESTALE);
317 			}
318 		}
319 	}
320 
321 	error = nfserr_badhandle;
322 	if (dentry == NULL)
323 		goto out;
324 	if (IS_ERR(dentry)) {
325 		if (PTR_ERR(dentry) != -EINVAL)
326 			error = nfserrno(PTR_ERR(dentry));
327 		goto out;
328 	}
329 
330 	if (d_is_dir(dentry) &&
331 			(dentry->d_flags & DCACHE_DISCONNECTED)) {
332 		printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %pd2\n",
333 				dentry);
334 	}
335 
336 	switch (fhp->fh_maxsize) {
337 	case NFS4_FHSIZE:
338 		if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOATOMIC_ATTR)
339 			fhp->fh_no_atomic_attr = true;
340 		fhp->fh_64bit_cookies = true;
341 		break;
342 	case NFS3_FHSIZE:
343 		if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOWCC)
344 			fhp->fh_no_wcc = true;
345 		fhp->fh_64bit_cookies = true;
346 		if (exp->ex_flags & NFSEXP_V4ROOT) {
347 			dput(dentry);
348 			goto out;
349 		}
350 		break;
351 	case NFS_FHSIZE:
352 		fhp->fh_no_wcc = true;
353 		if (EX_WGATHER(exp))
354 			fhp->fh_use_wgather = true;
355 		if (exp->ex_flags & NFSEXP_V4ROOT) {
356 			dput(dentry);
357 			goto out;
358 		}
359 	}
360 
361 	fhp->fh_dentry = dentry;
362 	fhp->fh_export = exp;
363 
364 	return 0;
365 out:
366 	exp_put(exp);
367 	return error;
368 }
369 
370 /**
371  * __fh_verify - filehandle lookup and access checking
372  * @rqstp: RPC transaction context, or NULL
373  * @net: net namespace in which to perform the export lookup
374  * @cred: RPC user credential
375  * @client: RPC auth domain
376  * @gssclient: RPC GSS auth domain, or NULL
377  * @fhp: filehandle to be verified
378  * @type: expected type of object pointed to by filehandle
379  * @access: type of access needed to object
380  *
381  * See fh_verify() for further descriptions of @fhp, @type, and @access.
382  */
383 static __be32
__fh_verify(struct svc_rqst * rqstp,struct net * net,struct svc_cred * cred,struct auth_domain * client,struct auth_domain * gssclient,struct svc_fh * fhp,umode_t type,int access)384 __fh_verify(struct svc_rqst *rqstp,
385 	    struct net *net, struct svc_cred *cred,
386 	    struct auth_domain *client,
387 	    struct auth_domain *gssclient,
388 	    struct svc_fh *fhp, umode_t type, int access)
389 {
390 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
391 	struct svc_export *exp = NULL;
392 	bool may_bypass_gss = false;
393 	struct dentry	*dentry;
394 	__be32		error;
395 
396 	if (!fhp->fh_dentry) {
397 		error = nfsd_set_fh_dentry(rqstp, net, cred, client,
398 					   gssclient, fhp);
399 		if (error)
400 			goto out;
401 	}
402 	dentry = fhp->fh_dentry;
403 	exp = fhp->fh_export;
404 
405 	trace_nfsd_fh_verify(rqstp, fhp, type, access);
406 
407 	/*
408 	 * We still have to do all these permission checks, even when
409 	 * fh_dentry is already set:
410 	 * 	- fh_verify may be called multiple times with different
411 	 * 	  "access" arguments (e.g. nfsd_proc_create calls
412 	 * 	  fh_verify(...,NFSD_MAY_EXEC) first, then later (in
413 	 * 	  nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
414 	 *	- in the NFSv4 case, the filehandle may have been filled
415 	 *	  in by fh_compose, and given a dentry, but further
416 	 *	  compound operations performed with that filehandle
417 	 *	  still need permissions checks.  In the worst case, a
418 	 *	  mountpoint crossing may have changed the export
419 	 *	  options, and we may now need to use a different uid
420 	 *	  (for example, if different id-squashing options are in
421 	 *	  effect on the new filesystem).
422 	 */
423 	error = check_pseudo_root(dentry, exp);
424 	if (error)
425 		goto out;
426 
427 	error = nfsd_setuser_and_check_port(rqstp, cred, exp);
428 	if (error)
429 		goto out;
430 
431 	error = nfsd_mode_check(dentry, type);
432 	if (error)
433 		goto out;
434 
435 	/*
436 	 * If rqstp is NULL, this is a LOCALIO request which will only
437 	 * ever use a filehandle/credential pair for which access has
438 	 * been affirmed (by ACCESS or OPEN NFS requests) over the
439 	 * wire.  Skip both the xprtsec policy and the security flavor
440 	 * checks.
441 	 */
442 	if (!rqstp)
443 		goto check_permissions;
444 
445 	if ((access & NFSD_MAY_NLM) && (exp->ex_flags & NFSEXP_NOAUTHNLM))
446 		/* NLM is allowed to fully bypass authentication */
447 		goto out;
448 
449 	/*
450 	 * NLM is allowed to bypass the xprtsec policy check because lockd
451 	 * doesn't support xprtsec.
452 	 */
453 	if (!(access & NFSD_MAY_NLM)) {
454 		error = check_xprtsec_policy(exp, rqstp);
455 		if (error)
456 			goto out;
457 	}
458 
459 	if (access & NFSD_MAY_BYPASS_GSS)
460 		may_bypass_gss = true;
461 	/*
462 	 * Clients may expect to be able to use auth_sys during mount,
463 	 * even if they use gss for everything else; see section 2.3.2
464 	 * of rfc 2623.
465 	 */
466 	if (access & NFSD_MAY_BYPASS_GSS_ON_ROOT
467 			&& exp->ex_path.dentry == dentry)
468 		may_bypass_gss = true;
469 
470 	error = check_security_flavor(exp, rqstp, may_bypass_gss);
471 	if (error)
472 		goto out;
473 
474 	svc_xprt_set_valid(rqstp->rq_xprt);
475 
476 check_permissions:
477 	/* Finally, check access permissions. */
478 	error = nfsd_permission(cred, exp, dentry, access);
479 out:
480 	trace_nfsd_fh_verify_err(rqstp, fhp, type, access, error);
481 	if (error == nfserr_stale)
482 		nfsd_stats_fh_stale_inc(nn, exp);
483 	return error;
484 }
485 
486 /**
487  * fh_verify_local - filehandle lookup and access checking
488  * @net: net namespace in which to perform the export lookup
489  * @cred: RPC user credential
490  * @client: RPC auth domain
491  * @fhp: filehandle to be verified
492  * @type: expected type of object pointed to by filehandle
493  * @access: type of access needed to object
494  *
495  * This API can be used by callers who do not have an RPC
496  * transaction context (ie are not running in an nfsd thread).
497  *
498  * See fh_verify() for further descriptions of @fhp, @type, and @access.
499  */
500 __be32
fh_verify_local(struct net * net,struct svc_cred * cred,struct auth_domain * client,struct svc_fh * fhp,umode_t type,int access)501 fh_verify_local(struct net *net, struct svc_cred *cred,
502 		struct auth_domain *client, struct svc_fh *fhp,
503 		umode_t type, int access)
504 {
505 	return __fh_verify(NULL, net, cred, client, NULL,
506 			   fhp, type, access);
507 }
508 
509 /**
510  * fh_verify - filehandle lookup and access checking
511  * @rqstp: pointer to current rpc request
512  * @fhp: filehandle to be verified
513  * @type: expected type of object pointed to by filehandle
514  * @access: type of access needed to object
515  *
516  * Look up a dentry from the on-the-wire filehandle, check the client's
517  * access to the export, and set the current task's credentials.
518  *
519  * Regardless of success or failure of fh_verify(), fh_put() should be
520  * called on @fhp when the caller is finished with the filehandle.
521  *
522  * fh_verify() may be called multiple times on a given filehandle, for
523  * example, when processing an NFSv4 compound.  The first call will look
524  * up a dentry using the on-the-wire filehandle.  Subsequent calls will
525  * skip the lookup and just perform the other checks and possibly change
526  * the current task's credentials.
527  *
528  * @type specifies the type of object expected using one of the S_IF*
529  * constants defined in include/linux/stat.h.  The caller may use zero
530  * to indicate that it doesn't care, or a negative integer to indicate
531  * that it expects something not of the given type.
532  *
533  * @access is formed from the NFSD_MAY_* constants defined in
534  * fs/nfsd/vfs.h.
535  */
536 __be32
fh_verify(struct svc_rqst * rqstp,struct svc_fh * fhp,umode_t type,int access)537 fh_verify(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, int access)
538 {
539 	return __fh_verify(rqstp, SVC_NET(rqstp), &rqstp->rq_cred,
540 			   rqstp->rq_client, rqstp->rq_gssclient,
541 			   fhp, type, access);
542 }
543 
544 /*
545  * Compose a file handle for an NFS reply.
546  *
547  * Note that when first composed, the dentry may not yet have
548  * an inode.  In this case a call to fh_update should be made
549  * before the fh goes out on the wire ...
550  */
_fh_update(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry)551 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp,
552 		struct dentry *dentry)
553 {
554 	if (dentry != exp->ex_path.dentry) {
555 		struct fid *fid = (struct fid *)
556 			(fh_fsid(&fhp->fh_handle) + fhp->fh_handle.fh_size/4 - 1);
557 		int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4;
558 		int fh_flags = (exp->ex_flags & NFSEXP_NOSUBTREECHECK) ? 0 :
559 				EXPORT_FH_CONNECTABLE;
560 		int fileid_type =
561 			exportfs_encode_fh(dentry, fid, &maxsize, fh_flags);
562 
563 		fhp->fh_handle.fh_fileid_type =
564 			fileid_type > 0 ? fileid_type : FILEID_INVALID;
565 		fhp->fh_handle.fh_size += maxsize * 4;
566 
567 		if (exp->ex_flags & NFSEXP_SIGN_FH)
568 			if (!fh_append_mac(&fhp->fh_handle, fhp->fh_maxsize,
569 					   exp->cd->net))
570 				fhp->fh_handle.fh_fileid_type = FILEID_INVALID;
571 	} else {
572 		fhp->fh_handle.fh_fileid_type = FILEID_ROOT;
573 	}
574 }
575 
is_root_export(struct svc_export * exp)576 static bool is_root_export(struct svc_export *exp)
577 {
578 	return exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root;
579 }
580 
exp_sb(struct svc_export * exp)581 static struct super_block *exp_sb(struct svc_export *exp)
582 {
583 	return exp->ex_path.dentry->d_sb;
584 }
585 
fsid_type_ok_for_exp(u8 fsid_type,struct svc_export * exp)586 static bool fsid_type_ok_for_exp(u8 fsid_type, struct svc_export *exp)
587 {
588 	switch (fsid_type) {
589 	case FSID_DEV:
590 		if (!old_valid_dev(exp_sb(exp)->s_dev))
591 			return false;
592 		fallthrough;
593 	case FSID_MAJOR_MINOR:
594 	case FSID_ENCODE_DEV:
595 		return exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV;
596 	case FSID_NUM:
597 		return exp->ex_flags & NFSEXP_FSID;
598 	case FSID_UUID8:
599 	case FSID_UUID16:
600 		if (!is_root_export(exp))
601 			return false;
602 		fallthrough;
603 	case FSID_UUID4_INUM:
604 	case FSID_UUID16_INUM:
605 		return exp->ex_uuid != NULL;
606 	}
607 	return true;
608 }
609 
610 
set_version_and_fsid_type(struct svc_fh * fhp,struct svc_export * exp,struct svc_fh * ref_fh)611 static void set_version_and_fsid_type(struct svc_fh *fhp, struct svc_export *exp, struct svc_fh *ref_fh)
612 {
613 	u8 version;
614 	u8 fsid_type;
615 retry:
616 	version = 1;
617 	if (ref_fh && ref_fh->fh_export == exp) {
618 		version = ref_fh->fh_handle.fh_version;
619 		fsid_type = ref_fh->fh_handle.fh_fsid_type;
620 
621 		ref_fh = NULL;
622 
623 		switch (version) {
624 		case 0xca:
625 			fsid_type = FSID_DEV;
626 			break;
627 		case 1:
628 			break;
629 		default:
630 			goto retry;
631 		}
632 
633 		/*
634 		 * As the fsid -> filesystem mapping was guided by
635 		 * user-space, there is no guarantee that the filesystem
636 		 * actually supports that fsid type. If it doesn't we
637 		 * loop around again without ref_fh set.
638 		 */
639 		if (!fsid_type_ok_for_exp(fsid_type, exp))
640 			goto retry;
641 	} else if (exp->ex_flags & NFSEXP_FSID) {
642 		fsid_type = FSID_NUM;
643 	} else if (exp->ex_uuid) {
644 		if (fhp->fh_maxsize >= 64) {
645 			if (is_root_export(exp))
646 				fsid_type = FSID_UUID16;
647 			else
648 				fsid_type = FSID_UUID16_INUM;
649 		} else {
650 			if (is_root_export(exp))
651 				fsid_type = FSID_UUID8;
652 			else
653 				fsid_type = FSID_UUID4_INUM;
654 		}
655 	} else if (!old_valid_dev(exp_sb(exp)->s_dev))
656 		/* for newer device numbers, we must use a newer fsid format */
657 		fsid_type = FSID_ENCODE_DEV;
658 	else
659 		fsid_type = FSID_DEV;
660 	fhp->fh_handle.fh_version = version;
661 	if (version)
662 		fhp->fh_handle.fh_fsid_type = fsid_type;
663 }
664 
665 __be32
fh_compose(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry,struct svc_fh * ref_fh)666 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry,
667 	   struct svc_fh *ref_fh)
668 {
669 	/* ref_fh is a reference file handle.
670 	 * if it is non-null and for the same filesystem, then we should compose
671 	 * a filehandle which is of the same version, where possible.
672 	 */
673 
674 	struct inode * inode = d_inode(dentry);
675 	dev_t ex_dev = exp_sb(exp)->s_dev;
676 
677 	dprintk("nfsd: fh_compose(exp %02x:%02x/%llu %pd2, ino=%llu)\n",
678 		MAJOR(ex_dev), MINOR(ex_dev),
679 		d_inode(exp->ex_path.dentry)->i_ino,
680 		dentry,
681 		(inode ? inode->i_ino : 0));
682 
683 	/* Choose filehandle version and fsid type based on
684 	 * the reference filehandle (if it is in the same export)
685 	 * or the export options.
686 	 */
687 	set_version_and_fsid_type(fhp, exp, ref_fh);
688 
689 	/* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
690 	fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false;
691 
692 	if (ref_fh == fhp)
693 		fh_put(ref_fh);
694 
695 	if (fhp->fh_dentry) {
696 		printk(KERN_ERR "fh_compose: fh %pd2 not initialized!\n",
697 		       dentry);
698 	}
699 	if (fhp->fh_maxsize < NFS_FHSIZE)
700 		printk(KERN_ERR "fh_compose: called with maxsize %d! %pd2\n",
701 		       fhp->fh_maxsize,
702 		       dentry);
703 
704 	fhp->fh_dentry = dget(dentry); /* our internal copy */
705 	fhp->fh_export = exp_get(exp);
706 
707 	fhp->fh_handle.fh_size =
708 		key_len(fhp->fh_handle.fh_fsid_type) + 4;
709 	fhp->fh_handle.fh_auth_type = 0;
710 
711 	mk_fsid(fhp->fh_handle.fh_fsid_type,
712 		fh_fsid(&fhp->fh_handle),
713 		ex_dev,
714 		d_inode(exp->ex_path.dentry)->i_ino,
715 		exp->ex_fsid, exp->ex_uuid);
716 
717 	if (inode)
718 		_fh_update(fhp, exp, dentry);
719 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) {
720 		fh_put(fhp);
721 		return nfserr_stale;
722 	}
723 
724 	return 0;
725 }
726 
727 /*
728  * Update file handle information after changing a dentry.
729  * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
730  */
731 __be32
fh_update(struct svc_fh * fhp)732 fh_update(struct svc_fh *fhp)
733 {
734 	struct dentry *dentry;
735 
736 	if (!fhp->fh_dentry)
737 		goto out_bad;
738 
739 	dentry = fhp->fh_dentry;
740 	if (d_really_is_negative(dentry))
741 		goto out_negative;
742 	if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT)
743 		return 0;
744 
745 	_fh_update(fhp, fhp->fh_export, dentry);
746 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID)
747 		return nfserr_stale;
748 	return 0;
749 out_bad:
750 	printk(KERN_ERR "fh_update: fh not verified!\n");
751 	return nfserr_serverfault;
752 out_negative:
753 	printk(KERN_ERR "fh_update: %pd2 still negative!\n",
754 		dentry);
755 	return nfserr_serverfault;
756 }
757 
758 /**
759  * fh_getattr - Retrieve attributes on a local file
760  * @fhp: File handle of target file
761  * @stat: Caller-supplied kstat buffer to be filled in
762  *
763  * Returns nfs_ok on success, otherwise an NFS status code is
764  * returned.
765  */
fh_getattr(const struct svc_fh * fhp,struct kstat * stat)766 __be32 fh_getattr(const struct svc_fh *fhp, struct kstat *stat)
767 {
768 	struct path p = {
769 		.mnt		= fhp->fh_export->ex_path.mnt,
770 		.dentry		= fhp->fh_dentry,
771 	};
772 	struct inode *inode = d_inode(p.dentry);
773 	u32 request_mask = STATX_BASIC_STATS;
774 
775 	if (S_ISREG(inode->i_mode))
776 		request_mask |= (STATX_DIOALIGN | STATX_DIO_READ_ALIGN);
777 
778 	if (fhp->fh_maxsize == NFS4_FHSIZE)
779 		request_mask |= (STATX_BTIME | STATX_CHANGE_COOKIE);
780 
781 	return nfserrno(vfs_getattr(&p, stat, request_mask,
782 				    AT_STATX_SYNC_AS_STAT));
783 }
784 
785 /**
786  * fh_fill_pre_attrs - Fill in pre-op attributes
787  * @fhp: file handle to be updated
788  *
789  */
fh_fill_pre_attrs(struct svc_fh * fhp)790 __be32 __must_check fh_fill_pre_attrs(struct svc_fh *fhp)
791 {
792 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
793 	struct kstat stat;
794 	__be32 err;
795 
796 	if (fhp->fh_no_wcc || fhp->fh_pre_saved)
797 		return nfs_ok;
798 
799 	err = fh_getattr(fhp, &stat);
800 	if (err)
801 		return err;
802 
803 	if (v4)
804 		fhp->fh_pre_change = nfsd4_change_attribute(&stat);
805 
806 	fhp->fh_pre_mtime = stat.mtime;
807 	fhp->fh_pre_ctime = stat.ctime;
808 	fhp->fh_pre_size  = stat.size;
809 	fhp->fh_pre_saved = true;
810 	return nfs_ok;
811 }
812 
813 /**
814  * fh_fill_post_attrs - Fill in post-op attributes
815  * @fhp: file handle to be updated
816  *
817  */
fh_fill_post_attrs(struct svc_fh * fhp)818 __be32 fh_fill_post_attrs(struct svc_fh *fhp)
819 {
820 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
821 	__be32 err;
822 
823 	if (fhp->fh_no_wcc)
824 		return nfs_ok;
825 
826 	if (fhp->fh_post_saved)
827 		printk("nfsd: inode locked twice during operation.\n");
828 
829 	err = fh_getattr(fhp, &fhp->fh_post_attr);
830 	if (err)
831 		return err;
832 
833 	fhp->fh_post_saved = true;
834 	if (v4)
835 		fhp->fh_post_change =
836 			nfsd4_change_attribute(&fhp->fh_post_attr);
837 	return nfs_ok;
838 }
839 
840 /**
841  * fh_fill_both_attrs - Fill pre-op and post-op attributes
842  * @fhp: file handle to be updated
843  *
844  * This is used when the directory wasn't changed, but wcc attributes
845  * are needed anyway.
846  */
fh_fill_both_attrs(struct svc_fh * fhp)847 __be32 __must_check fh_fill_both_attrs(struct svc_fh *fhp)
848 {
849 	__be32 err;
850 
851 	err = fh_fill_post_attrs(fhp);
852 	if (err)
853 		return err;
854 
855 	fhp->fh_pre_change = fhp->fh_post_change;
856 	fhp->fh_pre_mtime = fhp->fh_post_attr.mtime;
857 	fhp->fh_pre_ctime = fhp->fh_post_attr.ctime;
858 	fhp->fh_pre_size = fhp->fh_post_attr.size;
859 	fhp->fh_pre_saved = true;
860 	return nfs_ok;
861 }
862 
863 /*
864  * Release a file handle.
865  */
866 void
fh_put(struct svc_fh * fhp)867 fh_put(struct svc_fh *fhp)
868 {
869 	struct dentry * dentry = fhp->fh_dentry;
870 	struct svc_export * exp = fhp->fh_export;
871 	if (dentry) {
872 		fhp->fh_dentry = NULL;
873 		dput(dentry);
874 		fh_clear_pre_post_attrs(fhp);
875 	}
876 	fh_drop_write(fhp);
877 	if (exp) {
878 		exp_put(exp);
879 		fhp->fh_export = NULL;
880 	}
881 	fhp->fh_no_wcc = false;
882 	return;
883 }
884 
885 /*
886  * Shorthand for dprintk()'s
887  */
SVCFH_fmt(struct svc_fh * fhp)888 char * SVCFH_fmt(struct svc_fh *fhp)
889 {
890 	struct knfsd_fh *fh = &fhp->fh_handle;
891 	static char buf[2+1+1+64*3+1];
892 
893 	if (fh->fh_size > 64)
894 		return "bad-fh";
895 	sprintf(buf, "%d: %*ph", fh->fh_size, fh->fh_size, fh->fh_raw);
896 	return buf;
897 }
898 
fsid_source_fh(const struct knfsd_fh * fh,struct svc_export * exp)899 enum fsid_source fsid_source_fh(const struct knfsd_fh *fh,
900 				struct svc_export *exp)
901 {
902 	if (fh->fh_version != 1)
903 		return FSIDSOURCE_DEV;
904 	switch (fh->fh_fsid_type) {
905 	case FSID_DEV:
906 	case FSID_ENCODE_DEV:
907 	case FSID_MAJOR_MINOR:
908 		if (exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV)
909 			return FSIDSOURCE_DEV;
910 		break;
911 	case FSID_NUM:
912 		if (exp->ex_flags & NFSEXP_FSID)
913 			return FSIDSOURCE_FSID;
914 		break;
915 	default:
916 		break;
917 	}
918 	/* either a UUID type filehandle, or the filehandle doesn't
919 	 * match the export.
920 	 */
921 	if (exp->ex_flags & NFSEXP_FSID)
922 		return FSIDSOURCE_FSID;
923 	if (exp->ex_uuid)
924 		return FSIDSOURCE_UUID;
925 	return FSIDSOURCE_DEV;
926 }
927 
fsid_source(const struct svc_fh * fhp)928 enum fsid_source fsid_source(const struct svc_fh *fhp)
929 {
930 	return fsid_source_fh(&fhp->fh_handle, fhp->fh_export);
931 }
932 
933 /**
934  * nfsd4_change_attribute - Generate an NFSv4 change_attribute value
935  * @stat: inode attributes
936  *
937  * Caller must fill in @stat before calling, typically by invoking
938  * vfs_getattr() with STATX_MODE, STATX_CTIME, and STATX_CHANGE_COOKIE.
939  * Returns an unsigned 64-bit changeid4 value (RFC 8881 Section 3.2).
940  *
941  * We could use i_version alone as the change attribute.  However, i_version
942  * can go backwards on a regular file after an unclean shutdown.  On its own
943  * that doesn't necessarily cause a problem, but if i_version goes backwards
944  * and then is incremented again it could reuse a value that was previously
945  * used before boot, and a client who queried the two values might incorrectly
946  * assume nothing changed.
947  *
948  * By using both ctime and the i_version counter we guarantee that as long as
949  * time doesn't go backwards we never reuse an old value. If the filesystem
950  * advertises STATX_ATTR_CHANGE_MONOTONIC, then this mitigation is not
951  * needed.
952  *
953  * We only need to do this for regular files as well. For directories, we
954  * assume that the new change attr is always logged to stable storage in some
955  * fashion before the results can be seen.
956  */
nfsd4_change_attribute(const struct kstat * stat)957 u64 nfsd4_change_attribute(const struct kstat *stat)
958 {
959 	u64 chattr;
960 
961 	if (stat->result_mask & STATX_CHANGE_COOKIE) {
962 		chattr = stat->change_cookie;
963 		if (S_ISREG(stat->mode) &&
964 		    !(stat->attributes & STATX_ATTR_CHANGE_MONOTONIC)) {
965 			chattr += (u64)stat->ctime.tv_sec << 30;
966 			chattr += stat->ctime.tv_nsec;
967 		}
968 	} else {
969 		chattr = time_to_chattr(&stat->ctime);
970 	}
971 	return chattr;
972 }
973