xref: /linux/fs/nfsd/nfsfh.c (revision febbc555cf0fff895546ddb8ba2c9a523692fb55)
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 "nfsd.h"
15 #include "vfs.h"
16 #include "auth.h"
17 #include "trace.h"
18 
19 #define NFSDDBG_FACILITY		NFSDDBG_FH
20 
21 
22 /*
23  * our acceptability function.
24  * if NOSUBTREECHECK, accept anything
25  * if not, require that we can walk up to exp->ex_dentry
26  * doing some checks on the 'x' bits
27  */
nfsd_acceptable(void * expv,struct dentry * dentry)28 static int nfsd_acceptable(void *expv, struct dentry *dentry)
29 {
30 	struct svc_export *exp = expv;
31 	int rv;
32 	struct dentry *tdentry;
33 	struct dentry *parent;
34 
35 	if (exp->ex_flags & NFSEXP_NOSUBTREECHECK)
36 		return 1;
37 
38 	tdentry = dget(dentry);
39 	while (tdentry != exp->ex_path.dentry && !IS_ROOT(tdentry)) {
40 		/* make sure parents give x permission to user */
41 		int err;
42 		parent = dget_parent(tdentry);
43 		err = inode_permission(&nop_mnt_idmap,
44 				       d_inode(parent), MAY_EXEC);
45 		if (err < 0) {
46 			dput(parent);
47 			break;
48 		}
49 		dput(tdentry);
50 		tdentry = parent;
51 	}
52 	if (tdentry != exp->ex_path.dentry)
53 		dprintk("nfsd_acceptable failed at %p %pd\n", tdentry, tdentry);
54 	rv = (tdentry == exp->ex_path.dentry);
55 	dput(tdentry);
56 	return rv;
57 }
58 
59 /* Type check. The correct error return for type mismatches does not seem to be
60  * generally agreed upon. SunOS seems to use EISDIR if file isn't S_IFREG; a
61  * comment in the NFSv3 spec says this is incorrect (implementation notes for
62  * the write call).
63  */
64 static inline __be32
nfsd_mode_check(struct dentry * dentry,umode_t requested)65 nfsd_mode_check(struct dentry *dentry, umode_t requested)
66 {
67 	umode_t mode = d_inode(dentry)->i_mode & S_IFMT;
68 
69 	if (requested == 0) /* the caller doesn't care */
70 		return nfs_ok;
71 	if (mode == requested) {
72 		if (mode == S_IFDIR && !d_can_lookup(dentry)) {
73 			WARN_ON_ONCE(1);
74 			return nfserr_notdir;
75 		}
76 		return nfs_ok;
77 	}
78 	if (mode == S_IFLNK) {
79 		if (requested == S_IFDIR)
80 			return nfserr_symlink_not_dir;
81 		return nfserr_symlink;
82 	}
83 	if (requested == S_IFDIR)
84 		return nfserr_notdir;
85 	if (mode == S_IFDIR)
86 		return nfserr_isdir;
87 	return nfserr_wrong_type;
88 }
89 
nfsd_originating_port_ok(struct svc_rqst * rqstp,struct svc_cred * cred,struct svc_export * exp)90 static bool nfsd_originating_port_ok(struct svc_rqst *rqstp,
91 				     struct svc_cred *cred,
92 				     struct svc_export *exp)
93 {
94 	if (nfsexp_flags(cred, exp) & NFSEXP_INSECURE_PORT)
95 		return true;
96 	/* We don't require gss requests to use low ports: */
97 	if (cred->cr_flavor >= RPC_AUTH_GSS)
98 		return true;
99 	return test_bit(RQ_SECURE, &rqstp->rq_flags);
100 }
101 
nfsd_setuser_and_check_port(struct svc_rqst * rqstp,struct svc_cred * cred,struct svc_export * exp)102 static __be32 nfsd_setuser_and_check_port(struct svc_rqst *rqstp,
103 					  struct svc_cred *cred,
104 					  struct svc_export *exp)
105 {
106 	/* Check if the request originated from a secure port. */
107 	if (rqstp && !nfsd_originating_port_ok(rqstp, cred, exp)) {
108 		RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
109 		dprintk("nfsd: request from insecure port %s!\n",
110 		        svc_print_addr(rqstp, buf, sizeof(buf)));
111 		return nfserr_perm;
112 	}
113 
114 	/* Set user creds for this exportpoint */
115 	return nfserrno(nfsd_setuser(cred, exp));
116 }
117 
check_pseudo_root(struct dentry * dentry,struct svc_export * exp)118 static inline __be32 check_pseudo_root(struct dentry *dentry,
119 				       struct svc_export *exp)
120 {
121 	if (!(exp->ex_flags & NFSEXP_V4ROOT))
122 		return nfs_ok;
123 	/*
124 	 * We're exposing only the directories and symlinks that have to be
125 	 * traversed on the way to real exports:
126 	 */
127 	if (unlikely(!d_is_dir(dentry) &&
128 		     !d_is_symlink(dentry)))
129 		return nfserr_stale;
130 	/*
131 	 * A pseudoroot export gives permission to access only one
132 	 * single directory; the kernel has to make another upcall
133 	 * before granting access to anything else under it:
134 	 */
135 	if (unlikely(dentry != exp->ex_path.dentry))
136 		return nfserr_stale;
137 	return nfs_ok;
138 }
139 
140 /*
141  * Use the given filehandle to look up the corresponding export and
142  * dentry.  On success, the results are used to set fh_export and
143  * fh_dentry.
144  */
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)145 static __be32 nfsd_set_fh_dentry(struct svc_rqst *rqstp, struct net *net,
146 				 struct svc_cred *cred,
147 				 struct auth_domain *client,
148 				 struct auth_domain *gssclient,
149 				 struct svc_fh *fhp)
150 {
151 	struct knfsd_fh	*fh = &fhp->fh_handle;
152 	struct fid *fid = NULL;
153 	struct svc_export *exp;
154 	struct dentry *dentry;
155 	int fileid_type;
156 	int data_left = fh->fh_size/4;
157 	int len;
158 	__be32 error;
159 
160 	error = nfserr_badhandle;
161 	if (fh->fh_size == 0)
162 		return nfserr_nofilehandle;
163 
164 	if (fh->fh_version != 1)
165 		return error;
166 
167 	if (--data_left < 0)
168 		return error;
169 	if (fh->fh_auth_type != 0)
170 		return error;
171 	len = key_len(fh->fh_fsid_type) / 4;
172 	if (len == 0)
173 		return error;
174 	if (fh->fh_fsid_type == FSID_MAJOR_MINOR) {
175 		/* deprecated, convert to type 3 */
176 		len = key_len(FSID_ENCODE_DEV)/4;
177 		fh->fh_fsid_type = FSID_ENCODE_DEV;
178 		/*
179 		 * struct knfsd_fh uses host-endian fields, which are
180 		 * sometimes used to hold net-endian values. This
181 		 * confuses sparse, so we must use __force here to
182 		 * keep it from complaining.
183 		 */
184 		fh->fh_fsid[0] = new_encode_dev(MKDEV(ntohl((__force __be32)fh->fh_fsid[0]),
185 						      ntohl((__force __be32)fh->fh_fsid[1])));
186 		fh->fh_fsid[1] = fh->fh_fsid[2];
187 	}
188 	data_left -= len;
189 	if (data_left < 0)
190 		return error;
191 	exp = rqst_exp_find(rqstp ? &rqstp->rq_chandle : NULL,
192 			    net, client, gssclient,
193 			    fh->fh_fsid_type, fh->fh_fsid);
194 	fid = (struct fid *)(fh->fh_fsid + len);
195 
196 	error = nfserr_stale;
197 	if (IS_ERR(exp)) {
198 		trace_nfsd_set_fh_dentry_badexport(rqstp, fhp, PTR_ERR(exp));
199 
200 		if (PTR_ERR(exp) == -ENOENT)
201 			return error;
202 
203 		return nfserrno(PTR_ERR(exp));
204 	}
205 
206 	if (exp->ex_flags & NFSEXP_NOSUBTREECHECK) {
207 		/* Elevate privileges so that the lack of 'r' or 'x'
208 		 * permission on some parent directory will
209 		 * not stop exportfs_decode_fh from being able
210 		 * to reconnect a directory into the dentry cache.
211 		 * The same problem can affect "SUBTREECHECK" exports,
212 		 * but as nfsd_acceptable depends on correct
213 		 * access control settings being in effect, we cannot
214 		 * fix that case easily.
215 		 */
216 		struct cred *new = prepare_creds();
217 		if (!new) {
218 			error =  nfserrno(-ENOMEM);
219 			goto out;
220 		}
221 		new->cap_effective =
222 			cap_raise_nfsd_set(new->cap_effective,
223 					   new->cap_permitted);
224 		put_cred(override_creds(new));
225 	} else {
226 		error = nfsd_setuser_and_check_port(rqstp, cred, exp);
227 		if (error)
228 			goto out;
229 	}
230 
231 	/*
232 	 * Look up the dentry using the NFS file handle.
233 	 */
234 	error = nfserr_badhandle;
235 
236 	fileid_type = fh->fh_fileid_type;
237 
238 	if (fileid_type == FILEID_ROOT)
239 		dentry = dget(exp->ex_path.dentry);
240 	else {
241 		dentry = exportfs_decode_fh_raw(exp->ex_path.mnt, fid,
242 						data_left, fileid_type, 0,
243 						nfsd_acceptable, exp);
244 		if (IS_ERR_OR_NULL(dentry)) {
245 			trace_nfsd_set_fh_dentry_badhandle(rqstp, fhp,
246 					dentry ?  PTR_ERR(dentry) : -ESTALE);
247 			switch (PTR_ERR(dentry)) {
248 			case -ENOMEM:
249 			case -ETIMEDOUT:
250 				break;
251 			default:
252 				dentry = ERR_PTR(-ESTALE);
253 			}
254 		}
255 	}
256 	if (dentry == NULL)
257 		goto out;
258 	if (IS_ERR(dentry)) {
259 		if (PTR_ERR(dentry) != -EINVAL)
260 			error = nfserrno(PTR_ERR(dentry));
261 		goto out;
262 	}
263 
264 	if (d_is_dir(dentry) &&
265 			(dentry->d_flags & DCACHE_DISCONNECTED)) {
266 		printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %pd2\n",
267 				dentry);
268 	}
269 
270 	fhp->fh_dentry = dentry;
271 	fhp->fh_export = exp;
272 
273 	switch (fhp->fh_maxsize) {
274 	case NFS4_FHSIZE:
275 		if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOATOMIC_ATTR)
276 			fhp->fh_no_atomic_attr = true;
277 		fhp->fh_64bit_cookies = true;
278 		break;
279 	case NFS3_FHSIZE:
280 		if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOWCC)
281 			fhp->fh_no_wcc = true;
282 		fhp->fh_64bit_cookies = true;
283 		if (exp->ex_flags & NFSEXP_V4ROOT)
284 			goto out;
285 		break;
286 	case NFS_FHSIZE:
287 		fhp->fh_no_wcc = true;
288 		if (EX_WGATHER(exp))
289 			fhp->fh_use_wgather = true;
290 		if (exp->ex_flags & NFSEXP_V4ROOT)
291 			goto out;
292 	}
293 
294 	return 0;
295 out:
296 	exp_put(exp);
297 	return error;
298 }
299 
300 /**
301  * __fh_verify - filehandle lookup and access checking
302  * @rqstp: RPC transaction context, or NULL
303  * @net: net namespace in which to perform the export lookup
304  * @cred: RPC user credential
305  * @client: RPC auth domain
306  * @gssclient: RPC GSS auth domain, or NULL
307  * @fhp: filehandle to be verified
308  * @type: expected type of object pointed to by filehandle
309  * @access: type of access needed to object
310  *
311  * See fh_verify() for further descriptions of @fhp, @type, and @access.
312  */
313 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)314 __fh_verify(struct svc_rqst *rqstp,
315 	    struct net *net, struct svc_cred *cred,
316 	    struct auth_domain *client,
317 	    struct auth_domain *gssclient,
318 	    struct svc_fh *fhp, umode_t type, int access)
319 {
320 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
321 	struct svc_export *exp = NULL;
322 	bool may_bypass_gss = false;
323 	struct dentry	*dentry;
324 	__be32		error;
325 
326 	if (!fhp->fh_dentry) {
327 		error = nfsd_set_fh_dentry(rqstp, net, cred, client,
328 					   gssclient, fhp);
329 		if (error)
330 			goto out;
331 	}
332 	dentry = fhp->fh_dentry;
333 	exp = fhp->fh_export;
334 
335 	trace_nfsd_fh_verify(rqstp, fhp, type, access);
336 
337 	/*
338 	 * We still have to do all these permission checks, even when
339 	 * fh_dentry is already set:
340 	 * 	- fh_verify may be called multiple times with different
341 	 * 	  "access" arguments (e.g. nfsd_proc_create calls
342 	 * 	  fh_verify(...,NFSD_MAY_EXEC) first, then later (in
343 	 * 	  nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
344 	 *	- in the NFSv4 case, the filehandle may have been filled
345 	 *	  in by fh_compose, and given a dentry, but further
346 	 *	  compound operations performed with that filehandle
347 	 *	  still need permissions checks.  In the worst case, a
348 	 *	  mountpoint crossing may have changed the export
349 	 *	  options, and we may now need to use a different uid
350 	 *	  (for example, if different id-squashing options are in
351 	 *	  effect on the new filesystem).
352 	 */
353 	error = check_pseudo_root(dentry, exp);
354 	if (error)
355 		goto out;
356 
357 	error = nfsd_setuser_and_check_port(rqstp, cred, exp);
358 	if (error)
359 		goto out;
360 
361 	error = nfsd_mode_check(dentry, type);
362 	if (error)
363 		goto out;
364 
365 	if ((access & NFSD_MAY_NLM) && (exp->ex_flags & NFSEXP_NOAUTHNLM))
366 		/* NLM is allowed to fully bypass authentication */
367 		goto out;
368 
369 	if (access & NFSD_MAY_BYPASS_GSS)
370 		may_bypass_gss = true;
371 	/*
372 	 * Clients may expect to be able to use auth_sys during mount,
373 	 * even if they use gss for everything else; see section 2.3.2
374 	 * of rfc 2623.
375 	 */
376 	if (access & NFSD_MAY_BYPASS_GSS_ON_ROOT
377 			&& exp->ex_path.dentry == dentry)
378 		may_bypass_gss = true;
379 
380 	error = check_nfsd_access(exp, rqstp, may_bypass_gss);
381 	if (error)
382 		goto out;
383 	/* During LOCALIO call to fh_verify will be called with a NULL rqstp */
384 	if (rqstp)
385 		svc_xprt_set_valid(rqstp->rq_xprt);
386 
387 	/* Finally, check access permissions. */
388 	error = nfsd_permission(cred, exp, dentry, access);
389 out:
390 	trace_nfsd_fh_verify_err(rqstp, fhp, type, access, error);
391 	if (error == nfserr_stale)
392 		nfsd_stats_fh_stale_inc(nn, exp);
393 	return error;
394 }
395 
396 /**
397  * fh_verify_local - filehandle lookup and access checking
398  * @net: net namespace in which to perform the export lookup
399  * @cred: RPC user credential
400  * @client: RPC auth domain
401  * @fhp: filehandle to be verified
402  * @type: expected type of object pointed to by filehandle
403  * @access: type of access needed to object
404  *
405  * This API can be used by callers who do not have an RPC
406  * transaction context (ie are not running in an nfsd thread).
407  *
408  * See fh_verify() for further descriptions of @fhp, @type, and @access.
409  */
410 __be32
fh_verify_local(struct net * net,struct svc_cred * cred,struct auth_domain * client,struct svc_fh * fhp,umode_t type,int access)411 fh_verify_local(struct net *net, struct svc_cred *cred,
412 		struct auth_domain *client, struct svc_fh *fhp,
413 		umode_t type, int access)
414 {
415 	return __fh_verify(NULL, net, cred, client, NULL,
416 			   fhp, type, access);
417 }
418 
419 /**
420  * fh_verify - filehandle lookup and access checking
421  * @rqstp: pointer to current rpc request
422  * @fhp: filehandle to be verified
423  * @type: expected type of object pointed to by filehandle
424  * @access: type of access needed to object
425  *
426  * Look up a dentry from the on-the-wire filehandle, check the client's
427  * access to the export, and set the current task's credentials.
428  *
429  * Regardless of success or failure of fh_verify(), fh_put() should be
430  * called on @fhp when the caller is finished with the filehandle.
431  *
432  * fh_verify() may be called multiple times on a given filehandle, for
433  * example, when processing an NFSv4 compound.  The first call will look
434  * up a dentry using the on-the-wire filehandle.  Subsequent calls will
435  * skip the lookup and just perform the other checks and possibly change
436  * the current task's credentials.
437  *
438  * @type specifies the type of object expected using one of the S_IF*
439  * constants defined in include/linux/stat.h.  The caller may use zero
440  * to indicate that it doesn't care, or a negative integer to indicate
441  * that it expects something not of the given type.
442  *
443  * @access is formed from the NFSD_MAY_* constants defined in
444  * fs/nfsd/vfs.h.
445  */
446 __be32
fh_verify(struct svc_rqst * rqstp,struct svc_fh * fhp,umode_t type,int access)447 fh_verify(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, int access)
448 {
449 	return __fh_verify(rqstp, SVC_NET(rqstp), &rqstp->rq_cred,
450 			   rqstp->rq_client, rqstp->rq_gssclient,
451 			   fhp, type, access);
452 }
453 
454 /*
455  * Compose a file handle for an NFS reply.
456  *
457  * Note that when first composed, the dentry may not yet have
458  * an inode.  In this case a call to fh_update should be made
459  * before the fh goes out on the wire ...
460  */
_fh_update(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry)461 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp,
462 		struct dentry *dentry)
463 {
464 	if (dentry != exp->ex_path.dentry) {
465 		struct fid *fid = (struct fid *)
466 			(fhp->fh_handle.fh_fsid + fhp->fh_handle.fh_size/4 - 1);
467 		int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4;
468 		int fh_flags = (exp->ex_flags & NFSEXP_NOSUBTREECHECK) ? 0 :
469 				EXPORT_FH_CONNECTABLE;
470 		int fileid_type =
471 			exportfs_encode_fh(dentry, fid, &maxsize, fh_flags);
472 
473 		fhp->fh_handle.fh_fileid_type =
474 			fileid_type > 0 ? fileid_type : FILEID_INVALID;
475 		fhp->fh_handle.fh_size += maxsize * 4;
476 	} else {
477 		fhp->fh_handle.fh_fileid_type = FILEID_ROOT;
478 	}
479 }
480 
is_root_export(struct svc_export * exp)481 static bool is_root_export(struct svc_export *exp)
482 {
483 	return exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root;
484 }
485 
exp_sb(struct svc_export * exp)486 static struct super_block *exp_sb(struct svc_export *exp)
487 {
488 	return exp->ex_path.dentry->d_sb;
489 }
490 
fsid_type_ok_for_exp(u8 fsid_type,struct svc_export * exp)491 static bool fsid_type_ok_for_exp(u8 fsid_type, struct svc_export *exp)
492 {
493 	switch (fsid_type) {
494 	case FSID_DEV:
495 		if (!old_valid_dev(exp_sb(exp)->s_dev))
496 			return false;
497 		fallthrough;
498 	case FSID_MAJOR_MINOR:
499 	case FSID_ENCODE_DEV:
500 		return exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV;
501 	case FSID_NUM:
502 		return exp->ex_flags & NFSEXP_FSID;
503 	case FSID_UUID8:
504 	case FSID_UUID16:
505 		if (!is_root_export(exp))
506 			return false;
507 		fallthrough;
508 	case FSID_UUID4_INUM:
509 	case FSID_UUID16_INUM:
510 		return exp->ex_uuid != NULL;
511 	}
512 	return true;
513 }
514 
515 
set_version_and_fsid_type(struct svc_fh * fhp,struct svc_export * exp,struct svc_fh * ref_fh)516 static void set_version_and_fsid_type(struct svc_fh *fhp, struct svc_export *exp, struct svc_fh *ref_fh)
517 {
518 	u8 version;
519 	u8 fsid_type;
520 retry:
521 	version = 1;
522 	if (ref_fh && ref_fh->fh_export == exp) {
523 		version = ref_fh->fh_handle.fh_version;
524 		fsid_type = ref_fh->fh_handle.fh_fsid_type;
525 
526 		ref_fh = NULL;
527 
528 		switch (version) {
529 		case 0xca:
530 			fsid_type = FSID_DEV;
531 			break;
532 		case 1:
533 			break;
534 		default:
535 			goto retry;
536 		}
537 
538 		/*
539 		 * As the fsid -> filesystem mapping was guided by
540 		 * user-space, there is no guarantee that the filesystem
541 		 * actually supports that fsid type. If it doesn't we
542 		 * loop around again without ref_fh set.
543 		 */
544 		if (!fsid_type_ok_for_exp(fsid_type, exp))
545 			goto retry;
546 	} else if (exp->ex_flags & NFSEXP_FSID) {
547 		fsid_type = FSID_NUM;
548 	} else if (exp->ex_uuid) {
549 		if (fhp->fh_maxsize >= 64) {
550 			if (is_root_export(exp))
551 				fsid_type = FSID_UUID16;
552 			else
553 				fsid_type = FSID_UUID16_INUM;
554 		} else {
555 			if (is_root_export(exp))
556 				fsid_type = FSID_UUID8;
557 			else
558 				fsid_type = FSID_UUID4_INUM;
559 		}
560 	} else if (!old_valid_dev(exp_sb(exp)->s_dev))
561 		/* for newer device numbers, we must use a newer fsid format */
562 		fsid_type = FSID_ENCODE_DEV;
563 	else
564 		fsid_type = FSID_DEV;
565 	fhp->fh_handle.fh_version = version;
566 	if (version)
567 		fhp->fh_handle.fh_fsid_type = fsid_type;
568 }
569 
570 __be32
fh_compose(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry,struct svc_fh * ref_fh)571 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry,
572 	   struct svc_fh *ref_fh)
573 {
574 	/* ref_fh is a reference file handle.
575 	 * if it is non-null and for the same filesystem, then we should compose
576 	 * a filehandle which is of the same version, where possible.
577 	 */
578 
579 	struct inode * inode = d_inode(dentry);
580 	dev_t ex_dev = exp_sb(exp)->s_dev;
581 
582 	dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n",
583 		MAJOR(ex_dev), MINOR(ex_dev),
584 		(long) d_inode(exp->ex_path.dentry)->i_ino,
585 		dentry,
586 		(inode ? inode->i_ino : 0));
587 
588 	/* Choose filehandle version and fsid type based on
589 	 * the reference filehandle (if it is in the same export)
590 	 * or the export options.
591 	 */
592 	set_version_and_fsid_type(fhp, exp, ref_fh);
593 
594 	/* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
595 	fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false;
596 
597 	if (ref_fh == fhp)
598 		fh_put(ref_fh);
599 
600 	if (fhp->fh_dentry) {
601 		printk(KERN_ERR "fh_compose: fh %pd2 not initialized!\n",
602 		       dentry);
603 	}
604 	if (fhp->fh_maxsize < NFS_FHSIZE)
605 		printk(KERN_ERR "fh_compose: called with maxsize %d! %pd2\n",
606 		       fhp->fh_maxsize,
607 		       dentry);
608 
609 	fhp->fh_dentry = dget(dentry); /* our internal copy */
610 	fhp->fh_export = exp_get(exp);
611 
612 	fhp->fh_handle.fh_size =
613 		key_len(fhp->fh_handle.fh_fsid_type) + 4;
614 	fhp->fh_handle.fh_auth_type = 0;
615 
616 	mk_fsid(fhp->fh_handle.fh_fsid_type,
617 		fhp->fh_handle.fh_fsid,
618 		ex_dev,
619 		d_inode(exp->ex_path.dentry)->i_ino,
620 		exp->ex_fsid, exp->ex_uuid);
621 
622 	if (inode)
623 		_fh_update(fhp, exp, dentry);
624 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) {
625 		fh_put(fhp);
626 		return nfserr_stale;
627 	}
628 
629 	return 0;
630 }
631 
632 /*
633  * Update file handle information after changing a dentry.
634  * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
635  */
636 __be32
fh_update(struct svc_fh * fhp)637 fh_update(struct svc_fh *fhp)
638 {
639 	struct dentry *dentry;
640 
641 	if (!fhp->fh_dentry)
642 		goto out_bad;
643 
644 	dentry = fhp->fh_dentry;
645 	if (d_really_is_negative(dentry))
646 		goto out_negative;
647 	if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT)
648 		return 0;
649 
650 	_fh_update(fhp, fhp->fh_export, dentry);
651 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID)
652 		return nfserr_stale;
653 	return 0;
654 out_bad:
655 	printk(KERN_ERR "fh_update: fh not verified!\n");
656 	return nfserr_serverfault;
657 out_negative:
658 	printk(KERN_ERR "fh_update: %pd2 still negative!\n",
659 		dentry);
660 	return nfserr_serverfault;
661 }
662 
663 /**
664  * fh_fill_pre_attrs - Fill in pre-op attributes
665  * @fhp: file handle to be updated
666  *
667  */
fh_fill_pre_attrs(struct svc_fh * fhp)668 __be32 __must_check fh_fill_pre_attrs(struct svc_fh *fhp)
669 {
670 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
671 	struct kstat stat;
672 	__be32 err;
673 
674 	if (fhp->fh_no_wcc || fhp->fh_pre_saved)
675 		return nfs_ok;
676 
677 	err = fh_getattr(fhp, &stat);
678 	if (err)
679 		return err;
680 
681 	if (v4)
682 		fhp->fh_pre_change = nfsd4_change_attribute(&stat);
683 
684 	fhp->fh_pre_mtime = stat.mtime;
685 	fhp->fh_pre_ctime = stat.ctime;
686 	fhp->fh_pre_size  = stat.size;
687 	fhp->fh_pre_saved = true;
688 	return nfs_ok;
689 }
690 
691 /**
692  * fh_fill_post_attrs - Fill in post-op attributes
693  * @fhp: file handle to be updated
694  *
695  */
fh_fill_post_attrs(struct svc_fh * fhp)696 __be32 fh_fill_post_attrs(struct svc_fh *fhp)
697 {
698 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
699 	__be32 err;
700 
701 	if (fhp->fh_no_wcc)
702 		return nfs_ok;
703 
704 	if (fhp->fh_post_saved)
705 		printk("nfsd: inode locked twice during operation.\n");
706 
707 	err = fh_getattr(fhp, &fhp->fh_post_attr);
708 	if (err)
709 		return err;
710 
711 	fhp->fh_post_saved = true;
712 	if (v4)
713 		fhp->fh_post_change =
714 			nfsd4_change_attribute(&fhp->fh_post_attr);
715 	return nfs_ok;
716 }
717 
718 /**
719  * fh_fill_both_attrs - Fill pre-op and post-op attributes
720  * @fhp: file handle to be updated
721  *
722  * This is used when the directory wasn't changed, but wcc attributes
723  * are needed anyway.
724  */
fh_fill_both_attrs(struct svc_fh * fhp)725 __be32 __must_check fh_fill_both_attrs(struct svc_fh *fhp)
726 {
727 	__be32 err;
728 
729 	err = fh_fill_post_attrs(fhp);
730 	if (err)
731 		return err;
732 
733 	fhp->fh_pre_change = fhp->fh_post_change;
734 	fhp->fh_pre_mtime = fhp->fh_post_attr.mtime;
735 	fhp->fh_pre_ctime = fhp->fh_post_attr.ctime;
736 	fhp->fh_pre_size = fhp->fh_post_attr.size;
737 	fhp->fh_pre_saved = true;
738 	return nfs_ok;
739 }
740 
741 /*
742  * Release a file handle.
743  */
744 void
fh_put(struct svc_fh * fhp)745 fh_put(struct svc_fh *fhp)
746 {
747 	struct dentry * dentry = fhp->fh_dentry;
748 	struct svc_export * exp = fhp->fh_export;
749 	if (dentry) {
750 		fhp->fh_dentry = NULL;
751 		dput(dentry);
752 		fh_clear_pre_post_attrs(fhp);
753 	}
754 	fh_drop_write(fhp);
755 	if (exp) {
756 		exp_put(exp);
757 		fhp->fh_export = NULL;
758 	}
759 	fhp->fh_no_wcc = false;
760 	return;
761 }
762 
763 /*
764  * Shorthand for dprintk()'s
765  */
SVCFH_fmt(struct svc_fh * fhp)766 char * SVCFH_fmt(struct svc_fh *fhp)
767 {
768 	struct knfsd_fh *fh = &fhp->fh_handle;
769 	static char buf[2+1+1+64*3+1];
770 
771 	if (fh->fh_size > 64)
772 		return "bad-fh";
773 	sprintf(buf, "%d: %*ph", fh->fh_size, fh->fh_size, fh->fh_raw);
774 	return buf;
775 }
776 
fsid_source(const struct svc_fh * fhp)777 enum fsid_source fsid_source(const struct svc_fh *fhp)
778 {
779 	if (fhp->fh_handle.fh_version != 1)
780 		return FSIDSOURCE_DEV;
781 	switch(fhp->fh_handle.fh_fsid_type) {
782 	case FSID_DEV:
783 	case FSID_ENCODE_DEV:
784 	case FSID_MAJOR_MINOR:
785 		if (exp_sb(fhp->fh_export)->s_type->fs_flags & FS_REQUIRES_DEV)
786 			return FSIDSOURCE_DEV;
787 		break;
788 	case FSID_NUM:
789 		if (fhp->fh_export->ex_flags & NFSEXP_FSID)
790 			return FSIDSOURCE_FSID;
791 		break;
792 	default:
793 		break;
794 	}
795 	/* either a UUID type filehandle, or the filehandle doesn't
796 	 * match the export.
797 	 */
798 	if (fhp->fh_export->ex_flags & NFSEXP_FSID)
799 		return FSIDSOURCE_FSID;
800 	if (fhp->fh_export->ex_uuid)
801 		return FSIDSOURCE_UUID;
802 	return FSIDSOURCE_DEV;
803 }
804 
805 /**
806  * nfsd4_change_attribute - Generate an NFSv4 change_attribute value
807  * @stat: inode attributes
808  *
809  * Caller must fill in @stat before calling, typically by invoking
810  * vfs_getattr() with STATX_MODE, STATX_CTIME, and STATX_CHANGE_COOKIE.
811  * Returns an unsigned 64-bit changeid4 value (RFC 8881 Section 3.2).
812  *
813  * We could use i_version alone as the change attribute.  However, i_version
814  * can go backwards on a regular file after an unclean shutdown.  On its own
815  * that doesn't necessarily cause a problem, but if i_version goes backwards
816  * and then is incremented again it could reuse a value that was previously
817  * used before boot, and a client who queried the two values might incorrectly
818  * assume nothing changed.
819  *
820  * By using both ctime and the i_version counter we guarantee that as long as
821  * time doesn't go backwards we never reuse an old value. If the filesystem
822  * advertises STATX_ATTR_CHANGE_MONOTONIC, then this mitigation is not
823  * needed.
824  *
825  * We only need to do this for regular files as well. For directories, we
826  * assume that the new change attr is always logged to stable storage in some
827  * fashion before the results can be seen.
828  */
nfsd4_change_attribute(const struct kstat * stat)829 u64 nfsd4_change_attribute(const struct kstat *stat)
830 {
831 	u64 chattr;
832 
833 	if (stat->result_mask & STATX_CHANGE_COOKIE) {
834 		chattr = stat->change_cookie;
835 		if (S_ISREG(stat->mode) &&
836 		    !(stat->attributes & STATX_ATTR_CHANGE_MONOTONIC)) {
837 			chattr += (u64)stat->ctime.tv_sec << 30;
838 			chattr += stat->ctime.tv_nsec;
839 		}
840 	} else {
841 		chattr = time_to_chattr(&stat->ctime);
842 	}
843 	return chattr;
844 }
845