xref: /linux/fs/nfsd/nfsfh.c (revision ce3f5bb7504ca802efa710280a4601a06545bd2e)
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 		u32 *fsid = fh_fsid(fh);
176 
177 		/* deprecated, convert to type 3 */
178 		len = key_len(FSID_ENCODE_DEV)/4;
179 		fh->fh_fsid_type = FSID_ENCODE_DEV;
180 		/*
181 		 * struct knfsd_fh uses host-endian fields, which are
182 		 * sometimes used to hold net-endian values. This
183 		 * confuses sparse, so we must use __force here to
184 		 * keep it from complaining.
185 		 */
186 		fsid[0] = new_encode_dev(MKDEV(ntohl((__force __be32)fsid[0]),
187 					       ntohl((__force __be32)fsid[1])));
188 		fsid[1] = fsid[2];
189 	}
190 	data_left -= len;
191 	if (data_left < 0)
192 		return error;
193 	exp = rqst_exp_find(rqstp ? &rqstp->rq_chandle : NULL,
194 			    net, client, gssclient,
195 			    fh->fh_fsid_type, fh_fsid(fh));
196 	fid = (struct fid *)(fh_fsid(fh) + len);
197 
198 	error = nfserr_stale;
199 	if (IS_ERR(exp)) {
200 		trace_nfsd_set_fh_dentry_badexport(rqstp, fhp, PTR_ERR(exp));
201 
202 		if (PTR_ERR(exp) == -ENOENT)
203 			return error;
204 
205 		return nfserrno(PTR_ERR(exp));
206 	}
207 
208 	if (exp->ex_flags & NFSEXP_NOSUBTREECHECK) {
209 		/* Elevate privileges so that the lack of 'r' or 'x'
210 		 * permission on some parent directory will
211 		 * not stop exportfs_decode_fh from being able
212 		 * to reconnect a directory into the dentry cache.
213 		 * The same problem can affect "SUBTREECHECK" exports,
214 		 * but as nfsd_acceptable depends on correct
215 		 * access control settings being in effect, we cannot
216 		 * fix that case easily.
217 		 */
218 		struct cred *new = prepare_creds();
219 		if (!new) {
220 			error =  nfserrno(-ENOMEM);
221 			goto out;
222 		}
223 		new->cap_effective =
224 			cap_raise_nfsd_set(new->cap_effective,
225 					   new->cap_permitted);
226 		put_cred(override_creds(new));
227 	} else {
228 		error = nfsd_setuser_and_check_port(rqstp, cred, exp);
229 		if (error)
230 			goto out;
231 	}
232 
233 	/*
234 	 * Look up the dentry using the NFS file handle.
235 	 */
236 	error = nfserr_badhandle;
237 
238 	fileid_type = fh->fh_fileid_type;
239 
240 	if (fileid_type == FILEID_ROOT)
241 		dentry = dget(exp->ex_path.dentry);
242 	else {
243 		dentry = exportfs_decode_fh_raw(exp->ex_path.mnt, fid,
244 						data_left, fileid_type, 0,
245 						nfsd_acceptable, exp);
246 		if (IS_ERR_OR_NULL(dentry)) {
247 			trace_nfsd_set_fh_dentry_badhandle(rqstp, fhp,
248 					dentry ?  PTR_ERR(dentry) : -ESTALE);
249 			switch (PTR_ERR(dentry)) {
250 			case -ENOMEM:
251 			case -ETIMEDOUT:
252 				break;
253 			default:
254 				dentry = ERR_PTR(-ESTALE);
255 			}
256 		}
257 	}
258 	if (dentry == NULL)
259 		goto out;
260 	if (IS_ERR(dentry)) {
261 		if (PTR_ERR(dentry) != -EINVAL)
262 			error = nfserrno(PTR_ERR(dentry));
263 		goto out;
264 	}
265 
266 	if (d_is_dir(dentry) &&
267 			(dentry->d_flags & DCACHE_DISCONNECTED)) {
268 		printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %pd2\n",
269 				dentry);
270 	}
271 
272 	fhp->fh_dentry = dentry;
273 	fhp->fh_export = exp;
274 
275 	switch (fhp->fh_maxsize) {
276 	case NFS4_FHSIZE:
277 		if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOATOMIC_ATTR)
278 			fhp->fh_no_atomic_attr = true;
279 		fhp->fh_64bit_cookies = true;
280 		break;
281 	case NFS3_FHSIZE:
282 		if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOWCC)
283 			fhp->fh_no_wcc = true;
284 		fhp->fh_64bit_cookies = true;
285 		if (exp->ex_flags & NFSEXP_V4ROOT)
286 			goto out;
287 		break;
288 	case NFS_FHSIZE:
289 		fhp->fh_no_wcc = true;
290 		if (EX_WGATHER(exp))
291 			fhp->fh_use_wgather = true;
292 		if (exp->ex_flags & NFSEXP_V4ROOT)
293 			goto out;
294 	}
295 
296 	return 0;
297 out:
298 	exp_put(exp);
299 	return error;
300 }
301 
302 /**
303  * __fh_verify - filehandle lookup and access checking
304  * @rqstp: RPC transaction context, or NULL
305  * @net: net namespace in which to perform the export lookup
306  * @cred: RPC user credential
307  * @client: RPC auth domain
308  * @gssclient: RPC GSS auth domain, or NULL
309  * @fhp: filehandle to be verified
310  * @type: expected type of object pointed to by filehandle
311  * @access: type of access needed to object
312  *
313  * See fh_verify() for further descriptions of @fhp, @type, and @access.
314  */
315 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)316 __fh_verify(struct svc_rqst *rqstp,
317 	    struct net *net, struct svc_cred *cred,
318 	    struct auth_domain *client,
319 	    struct auth_domain *gssclient,
320 	    struct svc_fh *fhp, umode_t type, int access)
321 {
322 	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
323 	struct svc_export *exp = NULL;
324 	bool may_bypass_gss = false;
325 	struct dentry	*dentry;
326 	__be32		error;
327 
328 	if (!fhp->fh_dentry) {
329 		error = nfsd_set_fh_dentry(rqstp, net, cred, client,
330 					   gssclient, fhp);
331 		if (error)
332 			goto out;
333 	}
334 	dentry = fhp->fh_dentry;
335 	exp = fhp->fh_export;
336 
337 	trace_nfsd_fh_verify(rqstp, fhp, type, access);
338 
339 	/*
340 	 * We still have to do all these permission checks, even when
341 	 * fh_dentry is already set:
342 	 * 	- fh_verify may be called multiple times with different
343 	 * 	  "access" arguments (e.g. nfsd_proc_create calls
344 	 * 	  fh_verify(...,NFSD_MAY_EXEC) first, then later (in
345 	 * 	  nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
346 	 *	- in the NFSv4 case, the filehandle may have been filled
347 	 *	  in by fh_compose, and given a dentry, but further
348 	 *	  compound operations performed with that filehandle
349 	 *	  still need permissions checks.  In the worst case, a
350 	 *	  mountpoint crossing may have changed the export
351 	 *	  options, and we may now need to use a different uid
352 	 *	  (for example, if different id-squashing options are in
353 	 *	  effect on the new filesystem).
354 	 */
355 	error = check_pseudo_root(dentry, exp);
356 	if (error)
357 		goto out;
358 
359 	error = nfsd_setuser_and_check_port(rqstp, cred, exp);
360 	if (error)
361 		goto out;
362 
363 	error = nfsd_mode_check(dentry, type);
364 	if (error)
365 		goto out;
366 
367 	if ((access & NFSD_MAY_NLM) && (exp->ex_flags & NFSEXP_NOAUTHNLM))
368 		/* NLM is allowed to fully bypass authentication */
369 		goto out;
370 
371 	if (access & NFSD_MAY_BYPASS_GSS)
372 		may_bypass_gss = true;
373 	/*
374 	 * Clients may expect to be able to use auth_sys during mount,
375 	 * even if they use gss for everything else; see section 2.3.2
376 	 * of rfc 2623.
377 	 */
378 	if (access & NFSD_MAY_BYPASS_GSS_ON_ROOT
379 			&& exp->ex_path.dentry == dentry)
380 		may_bypass_gss = true;
381 
382 	error = check_nfsd_access(exp, rqstp, may_bypass_gss);
383 	if (error)
384 		goto out;
385 	/* During LOCALIO call to fh_verify will be called with a NULL rqstp */
386 	if (rqstp)
387 		svc_xprt_set_valid(rqstp->rq_xprt);
388 
389 	/* Finally, check access permissions. */
390 	error = nfsd_permission(cred, exp, dentry, access);
391 out:
392 	trace_nfsd_fh_verify_err(rqstp, fhp, type, access, error);
393 	if (error == nfserr_stale)
394 		nfsd_stats_fh_stale_inc(nn, exp);
395 	return error;
396 }
397 
398 /**
399  * fh_verify_local - filehandle lookup and access checking
400  * @net: net namespace in which to perform the export lookup
401  * @cred: RPC user credential
402  * @client: RPC auth domain
403  * @fhp: filehandle to be verified
404  * @type: expected type of object pointed to by filehandle
405  * @access: type of access needed to object
406  *
407  * This API can be used by callers who do not have an RPC
408  * transaction context (ie are not running in an nfsd thread).
409  *
410  * See fh_verify() for further descriptions of @fhp, @type, and @access.
411  */
412 __be32
fh_verify_local(struct net * net,struct svc_cred * cred,struct auth_domain * client,struct svc_fh * fhp,umode_t type,int access)413 fh_verify_local(struct net *net, struct svc_cred *cred,
414 		struct auth_domain *client, struct svc_fh *fhp,
415 		umode_t type, int access)
416 {
417 	return __fh_verify(NULL, net, cred, client, NULL,
418 			   fhp, type, access);
419 }
420 
421 /**
422  * fh_verify - filehandle lookup and access checking
423  * @rqstp: pointer to current rpc request
424  * @fhp: filehandle to be verified
425  * @type: expected type of object pointed to by filehandle
426  * @access: type of access needed to object
427  *
428  * Look up a dentry from the on-the-wire filehandle, check the client's
429  * access to the export, and set the current task's credentials.
430  *
431  * Regardless of success or failure of fh_verify(), fh_put() should be
432  * called on @fhp when the caller is finished with the filehandle.
433  *
434  * fh_verify() may be called multiple times on a given filehandle, for
435  * example, when processing an NFSv4 compound.  The first call will look
436  * up a dentry using the on-the-wire filehandle.  Subsequent calls will
437  * skip the lookup and just perform the other checks and possibly change
438  * the current task's credentials.
439  *
440  * @type specifies the type of object expected using one of the S_IF*
441  * constants defined in include/linux/stat.h.  The caller may use zero
442  * to indicate that it doesn't care, or a negative integer to indicate
443  * that it expects something not of the given type.
444  *
445  * @access is formed from the NFSD_MAY_* constants defined in
446  * fs/nfsd/vfs.h.
447  */
448 __be32
fh_verify(struct svc_rqst * rqstp,struct svc_fh * fhp,umode_t type,int access)449 fh_verify(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, int access)
450 {
451 	return __fh_verify(rqstp, SVC_NET(rqstp), &rqstp->rq_cred,
452 			   rqstp->rq_client, rqstp->rq_gssclient,
453 			   fhp, type, access);
454 }
455 
456 /*
457  * Compose a file handle for an NFS reply.
458  *
459  * Note that when first composed, the dentry may not yet have
460  * an inode.  In this case a call to fh_update should be made
461  * before the fh goes out on the wire ...
462  */
_fh_update(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry)463 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp,
464 		struct dentry *dentry)
465 {
466 	if (dentry != exp->ex_path.dentry) {
467 		struct fid *fid = (struct fid *)
468 			(fh_fsid(&fhp->fh_handle) + fhp->fh_handle.fh_size/4 - 1);
469 		int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4;
470 		int fh_flags = (exp->ex_flags & NFSEXP_NOSUBTREECHECK) ? 0 :
471 				EXPORT_FH_CONNECTABLE;
472 		int fileid_type =
473 			exportfs_encode_fh(dentry, fid, &maxsize, fh_flags);
474 
475 		fhp->fh_handle.fh_fileid_type =
476 			fileid_type > 0 ? fileid_type : FILEID_INVALID;
477 		fhp->fh_handle.fh_size += maxsize * 4;
478 	} else {
479 		fhp->fh_handle.fh_fileid_type = FILEID_ROOT;
480 	}
481 }
482 
is_root_export(struct svc_export * exp)483 static bool is_root_export(struct svc_export *exp)
484 {
485 	return exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root;
486 }
487 
exp_sb(struct svc_export * exp)488 static struct super_block *exp_sb(struct svc_export *exp)
489 {
490 	return exp->ex_path.dentry->d_sb;
491 }
492 
fsid_type_ok_for_exp(u8 fsid_type,struct svc_export * exp)493 static bool fsid_type_ok_for_exp(u8 fsid_type, struct svc_export *exp)
494 {
495 	switch (fsid_type) {
496 	case FSID_DEV:
497 		if (!old_valid_dev(exp_sb(exp)->s_dev))
498 			return false;
499 		fallthrough;
500 	case FSID_MAJOR_MINOR:
501 	case FSID_ENCODE_DEV:
502 		return exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV;
503 	case FSID_NUM:
504 		return exp->ex_flags & NFSEXP_FSID;
505 	case FSID_UUID8:
506 	case FSID_UUID16:
507 		if (!is_root_export(exp))
508 			return false;
509 		fallthrough;
510 	case FSID_UUID4_INUM:
511 	case FSID_UUID16_INUM:
512 		return exp->ex_uuid != NULL;
513 	}
514 	return true;
515 }
516 
517 
set_version_and_fsid_type(struct svc_fh * fhp,struct svc_export * exp,struct svc_fh * ref_fh)518 static void set_version_and_fsid_type(struct svc_fh *fhp, struct svc_export *exp, struct svc_fh *ref_fh)
519 {
520 	u8 version;
521 	u8 fsid_type;
522 retry:
523 	version = 1;
524 	if (ref_fh && ref_fh->fh_export == exp) {
525 		version = ref_fh->fh_handle.fh_version;
526 		fsid_type = ref_fh->fh_handle.fh_fsid_type;
527 
528 		ref_fh = NULL;
529 
530 		switch (version) {
531 		case 0xca:
532 			fsid_type = FSID_DEV;
533 			break;
534 		case 1:
535 			break;
536 		default:
537 			goto retry;
538 		}
539 
540 		/*
541 		 * As the fsid -> filesystem mapping was guided by
542 		 * user-space, there is no guarantee that the filesystem
543 		 * actually supports that fsid type. If it doesn't we
544 		 * loop around again without ref_fh set.
545 		 */
546 		if (!fsid_type_ok_for_exp(fsid_type, exp))
547 			goto retry;
548 	} else if (exp->ex_flags & NFSEXP_FSID) {
549 		fsid_type = FSID_NUM;
550 	} else if (exp->ex_uuid) {
551 		if (fhp->fh_maxsize >= 64) {
552 			if (is_root_export(exp))
553 				fsid_type = FSID_UUID16;
554 			else
555 				fsid_type = FSID_UUID16_INUM;
556 		} else {
557 			if (is_root_export(exp))
558 				fsid_type = FSID_UUID8;
559 			else
560 				fsid_type = FSID_UUID4_INUM;
561 		}
562 	} else if (!old_valid_dev(exp_sb(exp)->s_dev))
563 		/* for newer device numbers, we must use a newer fsid format */
564 		fsid_type = FSID_ENCODE_DEV;
565 	else
566 		fsid_type = FSID_DEV;
567 	fhp->fh_handle.fh_version = version;
568 	if (version)
569 		fhp->fh_handle.fh_fsid_type = fsid_type;
570 }
571 
572 __be32
fh_compose(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry,struct svc_fh * ref_fh)573 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry,
574 	   struct svc_fh *ref_fh)
575 {
576 	/* ref_fh is a reference file handle.
577 	 * if it is non-null and for the same filesystem, then we should compose
578 	 * a filehandle which is of the same version, where possible.
579 	 */
580 
581 	struct inode * inode = d_inode(dentry);
582 	dev_t ex_dev = exp_sb(exp)->s_dev;
583 
584 	dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n",
585 		MAJOR(ex_dev), MINOR(ex_dev),
586 		(long) d_inode(exp->ex_path.dentry)->i_ino,
587 		dentry,
588 		(inode ? inode->i_ino : 0));
589 
590 	/* Choose filehandle version and fsid type based on
591 	 * the reference filehandle (if it is in the same export)
592 	 * or the export options.
593 	 */
594 	set_version_and_fsid_type(fhp, exp, ref_fh);
595 
596 	/* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
597 	fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false;
598 
599 	if (ref_fh == fhp)
600 		fh_put(ref_fh);
601 
602 	if (fhp->fh_dentry) {
603 		printk(KERN_ERR "fh_compose: fh %pd2 not initialized!\n",
604 		       dentry);
605 	}
606 	if (fhp->fh_maxsize < NFS_FHSIZE)
607 		printk(KERN_ERR "fh_compose: called with maxsize %d! %pd2\n",
608 		       fhp->fh_maxsize,
609 		       dentry);
610 
611 	fhp->fh_dentry = dget(dentry); /* our internal copy */
612 	fhp->fh_export = exp_get(exp);
613 
614 	fhp->fh_handle.fh_size =
615 		key_len(fhp->fh_handle.fh_fsid_type) + 4;
616 	fhp->fh_handle.fh_auth_type = 0;
617 
618 	mk_fsid(fhp->fh_handle.fh_fsid_type,
619 		fh_fsid(&fhp->fh_handle),
620 		ex_dev,
621 		d_inode(exp->ex_path.dentry)->i_ino,
622 		exp->ex_fsid, exp->ex_uuid);
623 
624 	if (inode)
625 		_fh_update(fhp, exp, dentry);
626 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) {
627 		fh_put(fhp);
628 		return nfserr_stale;
629 	}
630 
631 	return 0;
632 }
633 
634 /*
635  * Update file handle information after changing a dentry.
636  * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
637  */
638 __be32
fh_update(struct svc_fh * fhp)639 fh_update(struct svc_fh *fhp)
640 {
641 	struct dentry *dentry;
642 
643 	if (!fhp->fh_dentry)
644 		goto out_bad;
645 
646 	dentry = fhp->fh_dentry;
647 	if (d_really_is_negative(dentry))
648 		goto out_negative;
649 	if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT)
650 		return 0;
651 
652 	_fh_update(fhp, fhp->fh_export, dentry);
653 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID)
654 		return nfserr_stale;
655 	return 0;
656 out_bad:
657 	printk(KERN_ERR "fh_update: fh not verified!\n");
658 	return nfserr_serverfault;
659 out_negative:
660 	printk(KERN_ERR "fh_update: %pd2 still negative!\n",
661 		dentry);
662 	return nfserr_serverfault;
663 }
664 
665 /**
666  * fh_fill_pre_attrs - Fill in pre-op attributes
667  * @fhp: file handle to be updated
668  *
669  */
fh_fill_pre_attrs(struct svc_fh * fhp)670 __be32 __must_check fh_fill_pre_attrs(struct svc_fh *fhp)
671 {
672 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
673 	struct kstat stat;
674 	__be32 err;
675 
676 	if (fhp->fh_no_wcc || fhp->fh_pre_saved)
677 		return nfs_ok;
678 
679 	err = fh_getattr(fhp, &stat);
680 	if (err)
681 		return err;
682 
683 	if (v4)
684 		fhp->fh_pre_change = nfsd4_change_attribute(&stat);
685 
686 	fhp->fh_pre_mtime = stat.mtime;
687 	fhp->fh_pre_ctime = stat.ctime;
688 	fhp->fh_pre_size  = stat.size;
689 	fhp->fh_pre_saved = true;
690 	return nfs_ok;
691 }
692 
693 /**
694  * fh_fill_post_attrs - Fill in post-op attributes
695  * @fhp: file handle to be updated
696  *
697  */
fh_fill_post_attrs(struct svc_fh * fhp)698 __be32 fh_fill_post_attrs(struct svc_fh *fhp)
699 {
700 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
701 	__be32 err;
702 
703 	if (fhp->fh_no_wcc)
704 		return nfs_ok;
705 
706 	if (fhp->fh_post_saved)
707 		printk("nfsd: inode locked twice during operation.\n");
708 
709 	err = fh_getattr(fhp, &fhp->fh_post_attr);
710 	if (err)
711 		return err;
712 
713 	fhp->fh_post_saved = true;
714 	if (v4)
715 		fhp->fh_post_change =
716 			nfsd4_change_attribute(&fhp->fh_post_attr);
717 	return nfs_ok;
718 }
719 
720 /**
721  * fh_fill_both_attrs - Fill pre-op and post-op attributes
722  * @fhp: file handle to be updated
723  *
724  * This is used when the directory wasn't changed, but wcc attributes
725  * are needed anyway.
726  */
fh_fill_both_attrs(struct svc_fh * fhp)727 __be32 __must_check fh_fill_both_attrs(struct svc_fh *fhp)
728 {
729 	__be32 err;
730 
731 	err = fh_fill_post_attrs(fhp);
732 	if (err)
733 		return err;
734 
735 	fhp->fh_pre_change = fhp->fh_post_change;
736 	fhp->fh_pre_mtime = fhp->fh_post_attr.mtime;
737 	fhp->fh_pre_ctime = fhp->fh_post_attr.ctime;
738 	fhp->fh_pre_size = fhp->fh_post_attr.size;
739 	fhp->fh_pre_saved = true;
740 	return nfs_ok;
741 }
742 
743 /*
744  * Release a file handle.
745  */
746 void
fh_put(struct svc_fh * fhp)747 fh_put(struct svc_fh *fhp)
748 {
749 	struct dentry * dentry = fhp->fh_dentry;
750 	struct svc_export * exp = fhp->fh_export;
751 	if (dentry) {
752 		fhp->fh_dentry = NULL;
753 		dput(dentry);
754 		fh_clear_pre_post_attrs(fhp);
755 	}
756 	fh_drop_write(fhp);
757 	if (exp) {
758 		exp_put(exp);
759 		fhp->fh_export = NULL;
760 	}
761 	fhp->fh_no_wcc = false;
762 	return;
763 }
764 
765 /*
766  * Shorthand for dprintk()'s
767  */
SVCFH_fmt(struct svc_fh * fhp)768 char * SVCFH_fmt(struct svc_fh *fhp)
769 {
770 	struct knfsd_fh *fh = &fhp->fh_handle;
771 	static char buf[2+1+1+64*3+1];
772 
773 	if (fh->fh_size > 64)
774 		return "bad-fh";
775 	sprintf(buf, "%d: %*ph", fh->fh_size, fh->fh_size, fh->fh_raw);
776 	return buf;
777 }
778 
fsid_source(const struct svc_fh * fhp)779 enum fsid_source fsid_source(const struct svc_fh *fhp)
780 {
781 	if (fhp->fh_handle.fh_version != 1)
782 		return FSIDSOURCE_DEV;
783 	switch(fhp->fh_handle.fh_fsid_type) {
784 	case FSID_DEV:
785 	case FSID_ENCODE_DEV:
786 	case FSID_MAJOR_MINOR:
787 		if (exp_sb(fhp->fh_export)->s_type->fs_flags & FS_REQUIRES_DEV)
788 			return FSIDSOURCE_DEV;
789 		break;
790 	case FSID_NUM:
791 		if (fhp->fh_export->ex_flags & NFSEXP_FSID)
792 			return FSIDSOURCE_FSID;
793 		break;
794 	default:
795 		break;
796 	}
797 	/* either a UUID type filehandle, or the filehandle doesn't
798 	 * match the export.
799 	 */
800 	if (fhp->fh_export->ex_flags & NFSEXP_FSID)
801 		return FSIDSOURCE_FSID;
802 	if (fhp->fh_export->ex_uuid)
803 		return FSIDSOURCE_UUID;
804 	return FSIDSOURCE_DEV;
805 }
806 
807 /**
808  * nfsd4_change_attribute - Generate an NFSv4 change_attribute value
809  * @stat: inode attributes
810  *
811  * Caller must fill in @stat before calling, typically by invoking
812  * vfs_getattr() with STATX_MODE, STATX_CTIME, and STATX_CHANGE_COOKIE.
813  * Returns an unsigned 64-bit changeid4 value (RFC 8881 Section 3.2).
814  *
815  * We could use i_version alone as the change attribute.  However, i_version
816  * can go backwards on a regular file after an unclean shutdown.  On its own
817  * that doesn't necessarily cause a problem, but if i_version goes backwards
818  * and then is incremented again it could reuse a value that was previously
819  * used before boot, and a client who queried the two values might incorrectly
820  * assume nothing changed.
821  *
822  * By using both ctime and the i_version counter we guarantee that as long as
823  * time doesn't go backwards we never reuse an old value. If the filesystem
824  * advertises STATX_ATTR_CHANGE_MONOTONIC, then this mitigation is not
825  * needed.
826  *
827  * We only need to do this for regular files as well. For directories, we
828  * assume that the new change attr is always logged to stable storage in some
829  * fashion before the results can be seen.
830  */
nfsd4_change_attribute(const struct kstat * stat)831 u64 nfsd4_change_attribute(const struct kstat *stat)
832 {
833 	u64 chattr;
834 
835 	if (stat->result_mask & STATX_CHANGE_COOKIE) {
836 		chattr = stat->change_cookie;
837 		if (S_ISREG(stat->mode) &&
838 		    !(stat->attributes & STATX_ATTR_CHANGE_MONOTONIC)) {
839 			chattr += (u64)stat->ctime.tv_sec << 30;
840 			chattr += stat->ctime.tv_nsec;
841 		}
842 	} else {
843 		chattr = time_to_chattr(&stat->ctime);
844 	}
845 	return chattr;
846 }
847