xref: /linux/fs/nfsd/nfsfh.c (revision 81538c8e42806eed71ce125723877a7c2307370c)
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 	/*
368 	 * If rqstp is NULL, this is a LOCALIO request which will only
369 	 * ever use a filehandle/credential pair for which access has
370 	 * been affirmed (by ACCESS or OPEN NFS requests) over the
371 	 * wire.  Skip both the xprtsec policy and the security flavor
372 	 * checks.
373 	 */
374 	if (!rqstp)
375 		goto check_permissions;
376 
377 	if ((access & NFSD_MAY_NLM) && (exp->ex_flags & NFSEXP_NOAUTHNLM))
378 		/* NLM is allowed to fully bypass authentication */
379 		goto out;
380 
381 	/*
382 	 * NLM is allowed to bypass the xprtsec policy check because lockd
383 	 * doesn't support xprtsec.
384 	 */
385 	if (!(access & NFSD_MAY_NLM)) {
386 		error = check_xprtsec_policy(exp, rqstp);
387 		if (error)
388 			goto out;
389 	}
390 
391 	if (access & NFSD_MAY_BYPASS_GSS)
392 		may_bypass_gss = true;
393 	/*
394 	 * Clients may expect to be able to use auth_sys during mount,
395 	 * even if they use gss for everything else; see section 2.3.2
396 	 * of rfc 2623.
397 	 */
398 	if (access & NFSD_MAY_BYPASS_GSS_ON_ROOT
399 			&& exp->ex_path.dentry == dentry)
400 		may_bypass_gss = true;
401 
402 	error = check_security_flavor(exp, rqstp, may_bypass_gss);
403 	if (error)
404 		goto out;
405 
406 	svc_xprt_set_valid(rqstp->rq_xprt);
407 
408 check_permissions:
409 	/* Finally, check access permissions. */
410 	error = nfsd_permission(cred, exp, dentry, access);
411 out:
412 	trace_nfsd_fh_verify_err(rqstp, fhp, type, access, error);
413 	if (error == nfserr_stale)
414 		nfsd_stats_fh_stale_inc(nn, exp);
415 	return error;
416 }
417 
418 /**
419  * fh_verify_local - filehandle lookup and access checking
420  * @net: net namespace in which to perform the export lookup
421  * @cred: RPC user credential
422  * @client: RPC auth domain
423  * @fhp: filehandle to be verified
424  * @type: expected type of object pointed to by filehandle
425  * @access: type of access needed to object
426  *
427  * This API can be used by callers who do not have an RPC
428  * transaction context (ie are not running in an nfsd thread).
429  *
430  * See fh_verify() for further descriptions of @fhp, @type, and @access.
431  */
432 __be32
fh_verify_local(struct net * net,struct svc_cred * cred,struct auth_domain * client,struct svc_fh * fhp,umode_t type,int access)433 fh_verify_local(struct net *net, struct svc_cred *cred,
434 		struct auth_domain *client, struct svc_fh *fhp,
435 		umode_t type, int access)
436 {
437 	return __fh_verify(NULL, net, cred, client, NULL,
438 			   fhp, type, access);
439 }
440 
441 /**
442  * fh_verify - filehandle lookup and access checking
443  * @rqstp: pointer to current rpc request
444  * @fhp: filehandle to be verified
445  * @type: expected type of object pointed to by filehandle
446  * @access: type of access needed to object
447  *
448  * Look up a dentry from the on-the-wire filehandle, check the client's
449  * access to the export, and set the current task's credentials.
450  *
451  * Regardless of success or failure of fh_verify(), fh_put() should be
452  * called on @fhp when the caller is finished with the filehandle.
453  *
454  * fh_verify() may be called multiple times on a given filehandle, for
455  * example, when processing an NFSv4 compound.  The first call will look
456  * up a dentry using the on-the-wire filehandle.  Subsequent calls will
457  * skip the lookup and just perform the other checks and possibly change
458  * the current task's credentials.
459  *
460  * @type specifies the type of object expected using one of the S_IF*
461  * constants defined in include/linux/stat.h.  The caller may use zero
462  * to indicate that it doesn't care, or a negative integer to indicate
463  * that it expects something not of the given type.
464  *
465  * @access is formed from the NFSD_MAY_* constants defined in
466  * fs/nfsd/vfs.h.
467  */
468 __be32
fh_verify(struct svc_rqst * rqstp,struct svc_fh * fhp,umode_t type,int access)469 fh_verify(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, int access)
470 {
471 	return __fh_verify(rqstp, SVC_NET(rqstp), &rqstp->rq_cred,
472 			   rqstp->rq_client, rqstp->rq_gssclient,
473 			   fhp, type, access);
474 }
475 
476 /*
477  * Compose a file handle for an NFS reply.
478  *
479  * Note that when first composed, the dentry may not yet have
480  * an inode.  In this case a call to fh_update should be made
481  * before the fh goes out on the wire ...
482  */
_fh_update(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry)483 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp,
484 		struct dentry *dentry)
485 {
486 	if (dentry != exp->ex_path.dentry) {
487 		struct fid *fid = (struct fid *)
488 			(fh_fsid(&fhp->fh_handle) + fhp->fh_handle.fh_size/4 - 1);
489 		int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4;
490 		int fh_flags = (exp->ex_flags & NFSEXP_NOSUBTREECHECK) ? 0 :
491 				EXPORT_FH_CONNECTABLE;
492 		int fileid_type =
493 			exportfs_encode_fh(dentry, fid, &maxsize, fh_flags);
494 
495 		fhp->fh_handle.fh_fileid_type =
496 			fileid_type > 0 ? fileid_type : FILEID_INVALID;
497 		fhp->fh_handle.fh_size += maxsize * 4;
498 	} else {
499 		fhp->fh_handle.fh_fileid_type = FILEID_ROOT;
500 	}
501 }
502 
is_root_export(struct svc_export * exp)503 static bool is_root_export(struct svc_export *exp)
504 {
505 	return exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root;
506 }
507 
exp_sb(struct svc_export * exp)508 static struct super_block *exp_sb(struct svc_export *exp)
509 {
510 	return exp->ex_path.dentry->d_sb;
511 }
512 
fsid_type_ok_for_exp(u8 fsid_type,struct svc_export * exp)513 static bool fsid_type_ok_for_exp(u8 fsid_type, struct svc_export *exp)
514 {
515 	switch (fsid_type) {
516 	case FSID_DEV:
517 		if (!old_valid_dev(exp_sb(exp)->s_dev))
518 			return false;
519 		fallthrough;
520 	case FSID_MAJOR_MINOR:
521 	case FSID_ENCODE_DEV:
522 		return exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV;
523 	case FSID_NUM:
524 		return exp->ex_flags & NFSEXP_FSID;
525 	case FSID_UUID8:
526 	case FSID_UUID16:
527 		if (!is_root_export(exp))
528 			return false;
529 		fallthrough;
530 	case FSID_UUID4_INUM:
531 	case FSID_UUID16_INUM:
532 		return exp->ex_uuid != NULL;
533 	}
534 	return true;
535 }
536 
537 
set_version_and_fsid_type(struct svc_fh * fhp,struct svc_export * exp,struct svc_fh * ref_fh)538 static void set_version_and_fsid_type(struct svc_fh *fhp, struct svc_export *exp, struct svc_fh *ref_fh)
539 {
540 	u8 version;
541 	u8 fsid_type;
542 retry:
543 	version = 1;
544 	if (ref_fh && ref_fh->fh_export == exp) {
545 		version = ref_fh->fh_handle.fh_version;
546 		fsid_type = ref_fh->fh_handle.fh_fsid_type;
547 
548 		ref_fh = NULL;
549 
550 		switch (version) {
551 		case 0xca:
552 			fsid_type = FSID_DEV;
553 			break;
554 		case 1:
555 			break;
556 		default:
557 			goto retry;
558 		}
559 
560 		/*
561 		 * As the fsid -> filesystem mapping was guided by
562 		 * user-space, there is no guarantee that the filesystem
563 		 * actually supports that fsid type. If it doesn't we
564 		 * loop around again without ref_fh set.
565 		 */
566 		if (!fsid_type_ok_for_exp(fsid_type, exp))
567 			goto retry;
568 	} else if (exp->ex_flags & NFSEXP_FSID) {
569 		fsid_type = FSID_NUM;
570 	} else if (exp->ex_uuid) {
571 		if (fhp->fh_maxsize >= 64) {
572 			if (is_root_export(exp))
573 				fsid_type = FSID_UUID16;
574 			else
575 				fsid_type = FSID_UUID16_INUM;
576 		} else {
577 			if (is_root_export(exp))
578 				fsid_type = FSID_UUID8;
579 			else
580 				fsid_type = FSID_UUID4_INUM;
581 		}
582 	} else if (!old_valid_dev(exp_sb(exp)->s_dev))
583 		/* for newer device numbers, we must use a newer fsid format */
584 		fsid_type = FSID_ENCODE_DEV;
585 	else
586 		fsid_type = FSID_DEV;
587 	fhp->fh_handle.fh_version = version;
588 	if (version)
589 		fhp->fh_handle.fh_fsid_type = fsid_type;
590 }
591 
592 __be32
fh_compose(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry,struct svc_fh * ref_fh)593 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry,
594 	   struct svc_fh *ref_fh)
595 {
596 	/* ref_fh is a reference file handle.
597 	 * if it is non-null and for the same filesystem, then we should compose
598 	 * a filehandle which is of the same version, where possible.
599 	 */
600 
601 	struct inode * inode = d_inode(dentry);
602 	dev_t ex_dev = exp_sb(exp)->s_dev;
603 
604 	dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n",
605 		MAJOR(ex_dev), MINOR(ex_dev),
606 		(long) d_inode(exp->ex_path.dentry)->i_ino,
607 		dentry,
608 		(inode ? inode->i_ino : 0));
609 
610 	/* Choose filehandle version and fsid type based on
611 	 * the reference filehandle (if it is in the same export)
612 	 * or the export options.
613 	 */
614 	set_version_and_fsid_type(fhp, exp, ref_fh);
615 
616 	/* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
617 	fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false;
618 
619 	if (ref_fh == fhp)
620 		fh_put(ref_fh);
621 
622 	if (fhp->fh_dentry) {
623 		printk(KERN_ERR "fh_compose: fh %pd2 not initialized!\n",
624 		       dentry);
625 	}
626 	if (fhp->fh_maxsize < NFS_FHSIZE)
627 		printk(KERN_ERR "fh_compose: called with maxsize %d! %pd2\n",
628 		       fhp->fh_maxsize,
629 		       dentry);
630 
631 	fhp->fh_dentry = dget(dentry); /* our internal copy */
632 	fhp->fh_export = exp_get(exp);
633 
634 	fhp->fh_handle.fh_size =
635 		key_len(fhp->fh_handle.fh_fsid_type) + 4;
636 	fhp->fh_handle.fh_auth_type = 0;
637 
638 	mk_fsid(fhp->fh_handle.fh_fsid_type,
639 		fh_fsid(&fhp->fh_handle),
640 		ex_dev,
641 		d_inode(exp->ex_path.dentry)->i_ino,
642 		exp->ex_fsid, exp->ex_uuid);
643 
644 	if (inode)
645 		_fh_update(fhp, exp, dentry);
646 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) {
647 		fh_put(fhp);
648 		return nfserr_stale;
649 	}
650 
651 	return 0;
652 }
653 
654 /*
655  * Update file handle information after changing a dentry.
656  * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
657  */
658 __be32
fh_update(struct svc_fh * fhp)659 fh_update(struct svc_fh *fhp)
660 {
661 	struct dentry *dentry;
662 
663 	if (!fhp->fh_dentry)
664 		goto out_bad;
665 
666 	dentry = fhp->fh_dentry;
667 	if (d_really_is_negative(dentry))
668 		goto out_negative;
669 	if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT)
670 		return 0;
671 
672 	_fh_update(fhp, fhp->fh_export, dentry);
673 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID)
674 		return nfserr_stale;
675 	return 0;
676 out_bad:
677 	printk(KERN_ERR "fh_update: fh not verified!\n");
678 	return nfserr_serverfault;
679 out_negative:
680 	printk(KERN_ERR "fh_update: %pd2 still negative!\n",
681 		dentry);
682 	return nfserr_serverfault;
683 }
684 
685 /**
686  * fh_getattr - Retrieve attributes on a local file
687  * @fhp: File handle of target file
688  * @stat: Caller-supplied kstat buffer to be filled in
689  *
690  * Returns nfs_ok on success, otherwise an NFS status code is
691  * returned.
692  */
fh_getattr(const struct svc_fh * fhp,struct kstat * stat)693 __be32 fh_getattr(const struct svc_fh *fhp, struct kstat *stat)
694 {
695 	struct path p = {
696 		.mnt		= fhp->fh_export->ex_path.mnt,
697 		.dentry		= fhp->fh_dentry,
698 	};
699 	struct inode *inode = d_inode(p.dentry);
700 	u32 request_mask = STATX_BASIC_STATS;
701 
702 	if (S_ISREG(inode->i_mode))
703 		request_mask |= (STATX_DIOALIGN | STATX_DIO_READ_ALIGN);
704 
705 	if (fhp->fh_maxsize == NFS4_FHSIZE)
706 		request_mask |= (STATX_BTIME | STATX_CHANGE_COOKIE);
707 
708 	return nfserrno(vfs_getattr(&p, stat, request_mask,
709 				    AT_STATX_SYNC_AS_STAT));
710 }
711 
712 /**
713  * fh_fill_pre_attrs - Fill in pre-op attributes
714  * @fhp: file handle to be updated
715  *
716  */
fh_fill_pre_attrs(struct svc_fh * fhp)717 __be32 __must_check fh_fill_pre_attrs(struct svc_fh *fhp)
718 {
719 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
720 	struct kstat stat;
721 	__be32 err;
722 
723 	if (fhp->fh_no_wcc || fhp->fh_pre_saved)
724 		return nfs_ok;
725 
726 	err = fh_getattr(fhp, &stat);
727 	if (err)
728 		return err;
729 
730 	if (v4)
731 		fhp->fh_pre_change = nfsd4_change_attribute(&stat);
732 
733 	fhp->fh_pre_mtime = stat.mtime;
734 	fhp->fh_pre_ctime = stat.ctime;
735 	fhp->fh_pre_size  = stat.size;
736 	fhp->fh_pre_saved = true;
737 	return nfs_ok;
738 }
739 
740 /**
741  * fh_fill_post_attrs - Fill in post-op attributes
742  * @fhp: file handle to be updated
743  *
744  */
fh_fill_post_attrs(struct svc_fh * fhp)745 __be32 fh_fill_post_attrs(struct svc_fh *fhp)
746 {
747 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
748 	__be32 err;
749 
750 	if (fhp->fh_no_wcc)
751 		return nfs_ok;
752 
753 	if (fhp->fh_post_saved)
754 		printk("nfsd: inode locked twice during operation.\n");
755 
756 	err = fh_getattr(fhp, &fhp->fh_post_attr);
757 	if (err)
758 		return err;
759 
760 	fhp->fh_post_saved = true;
761 	if (v4)
762 		fhp->fh_post_change =
763 			nfsd4_change_attribute(&fhp->fh_post_attr);
764 	return nfs_ok;
765 }
766 
767 /**
768  * fh_fill_both_attrs - Fill pre-op and post-op attributes
769  * @fhp: file handle to be updated
770  *
771  * This is used when the directory wasn't changed, but wcc attributes
772  * are needed anyway.
773  */
fh_fill_both_attrs(struct svc_fh * fhp)774 __be32 __must_check fh_fill_both_attrs(struct svc_fh *fhp)
775 {
776 	__be32 err;
777 
778 	err = fh_fill_post_attrs(fhp);
779 	if (err)
780 		return err;
781 
782 	fhp->fh_pre_change = fhp->fh_post_change;
783 	fhp->fh_pre_mtime = fhp->fh_post_attr.mtime;
784 	fhp->fh_pre_ctime = fhp->fh_post_attr.ctime;
785 	fhp->fh_pre_size = fhp->fh_post_attr.size;
786 	fhp->fh_pre_saved = true;
787 	return nfs_ok;
788 }
789 
790 /*
791  * Release a file handle.
792  */
793 void
fh_put(struct svc_fh * fhp)794 fh_put(struct svc_fh *fhp)
795 {
796 	struct dentry * dentry = fhp->fh_dentry;
797 	struct svc_export * exp = fhp->fh_export;
798 	if (dentry) {
799 		fhp->fh_dentry = NULL;
800 		dput(dentry);
801 		fh_clear_pre_post_attrs(fhp);
802 	}
803 	fh_drop_write(fhp);
804 	if (exp) {
805 		exp_put(exp);
806 		fhp->fh_export = NULL;
807 	}
808 	fhp->fh_no_wcc = false;
809 	return;
810 }
811 
812 /*
813  * Shorthand for dprintk()'s
814  */
SVCFH_fmt(struct svc_fh * fhp)815 char * SVCFH_fmt(struct svc_fh *fhp)
816 {
817 	struct knfsd_fh *fh = &fhp->fh_handle;
818 	static char buf[2+1+1+64*3+1];
819 
820 	if (fh->fh_size > 64)
821 		return "bad-fh";
822 	sprintf(buf, "%d: %*ph", fh->fh_size, fh->fh_size, fh->fh_raw);
823 	return buf;
824 }
825 
fsid_source(const struct svc_fh * fhp)826 enum fsid_source fsid_source(const struct svc_fh *fhp)
827 {
828 	if (fhp->fh_handle.fh_version != 1)
829 		return FSIDSOURCE_DEV;
830 	switch(fhp->fh_handle.fh_fsid_type) {
831 	case FSID_DEV:
832 	case FSID_ENCODE_DEV:
833 	case FSID_MAJOR_MINOR:
834 		if (exp_sb(fhp->fh_export)->s_type->fs_flags & FS_REQUIRES_DEV)
835 			return FSIDSOURCE_DEV;
836 		break;
837 	case FSID_NUM:
838 		if (fhp->fh_export->ex_flags & NFSEXP_FSID)
839 			return FSIDSOURCE_FSID;
840 		break;
841 	default:
842 		break;
843 	}
844 	/* either a UUID type filehandle, or the filehandle doesn't
845 	 * match the export.
846 	 */
847 	if (fhp->fh_export->ex_flags & NFSEXP_FSID)
848 		return FSIDSOURCE_FSID;
849 	if (fhp->fh_export->ex_uuid)
850 		return FSIDSOURCE_UUID;
851 	return FSIDSOURCE_DEV;
852 }
853 
854 /**
855  * nfsd4_change_attribute - Generate an NFSv4 change_attribute value
856  * @stat: inode attributes
857  *
858  * Caller must fill in @stat before calling, typically by invoking
859  * vfs_getattr() with STATX_MODE, STATX_CTIME, and STATX_CHANGE_COOKIE.
860  * Returns an unsigned 64-bit changeid4 value (RFC 8881 Section 3.2).
861  *
862  * We could use i_version alone as the change attribute.  However, i_version
863  * can go backwards on a regular file after an unclean shutdown.  On its own
864  * that doesn't necessarily cause a problem, but if i_version goes backwards
865  * and then is incremented again it could reuse a value that was previously
866  * used before boot, and a client who queried the two values might incorrectly
867  * assume nothing changed.
868  *
869  * By using both ctime and the i_version counter we guarantee that as long as
870  * time doesn't go backwards we never reuse an old value. If the filesystem
871  * advertises STATX_ATTR_CHANGE_MONOTONIC, then this mitigation is not
872  * needed.
873  *
874  * We only need to do this for regular files as well. For directories, we
875  * assume that the new change attr is always logged to stable storage in some
876  * fashion before the results can be seen.
877  */
nfsd4_change_attribute(const struct kstat * stat)878 u64 nfsd4_change_attribute(const struct kstat *stat)
879 {
880 	u64 chattr;
881 
882 	if (stat->result_mask & STATX_CHANGE_COOKIE) {
883 		chattr = stat->change_cookie;
884 		if (S_ISREG(stat->mode) &&
885 		    !(stat->attributes & STATX_ATTR_CHANGE_MONOTONIC)) {
886 			chattr += (u64)stat->ctime.tv_sec << 30;
887 			chattr += stat->ctime.tv_nsec;
888 		}
889 	} else {
890 		chattr = time_to_chattr(&stat->ctime);
891 	}
892 	return chattr;
893 }
894