xref: /openbmc/linux/fs/nfsd/nfsfh.c (revision a332018a)
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 svc_rqst * rqstp,struct dentry * dentry,umode_t requested)65 nfsd_mode_check(struct svc_rqst *rqstp, struct dentry *dentry,
66 		umode_t requested)
67 {
68 	umode_t mode = d_inode(dentry)->i_mode & S_IFMT;
69 
70 	if (requested == 0) /* the caller doesn't care */
71 		return nfs_ok;
72 	if (mode == requested) {
73 		if (mode == S_IFDIR && !d_can_lookup(dentry)) {
74 			WARN_ON_ONCE(1);
75 			return nfserr_notdir;
76 		}
77 		return nfs_ok;
78 	}
79 	/*
80 	 * v4 has an error more specific than err_notdir which we should
81 	 * return in preference to err_notdir:
82 	 */
83 	if (rqstp->rq_vers == 4 && mode == S_IFLNK)
84 		return nfserr_symlink;
85 	if (requested == S_IFDIR)
86 		return nfserr_notdir;
87 	if (mode == S_IFDIR)
88 		return nfserr_isdir;
89 	return nfserr_inval;
90 }
91 
nfsd_originating_port_ok(struct svc_rqst * rqstp,int flags)92 static bool nfsd_originating_port_ok(struct svc_rqst *rqstp, int flags)
93 {
94 	if (flags & NFSEXP_INSECURE_PORT)
95 		return true;
96 	/* We don't require gss requests to use low ports: */
97 	if (rqstp->rq_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_export * exp)102 static __be32 nfsd_setuser_and_check_port(struct svc_rqst *rqstp,
103 					  struct svc_export *exp)
104 {
105 	int flags = nfsexp_flags(rqstp, exp);
106 
107 	/* Check if the request originated from a secure port. */
108 	if (!nfsd_originating_port_ok(rqstp, flags)) {
109 		RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
110 		dprintk("nfsd: request from insecure port %s!\n",
111 		        svc_print_addr(rqstp, buf, sizeof(buf)));
112 		return nfserr_perm;
113 	}
114 
115 	/* Set user creds for this exportpoint */
116 	return nfserrno(nfsd_setuser(rqstp, exp));
117 }
118 
check_pseudo_root(struct svc_rqst * rqstp,struct dentry * dentry,struct svc_export * exp)119 static inline __be32 check_pseudo_root(struct svc_rqst *rqstp,
120 	struct dentry *dentry, struct svc_export *exp)
121 {
122 	if (!(exp->ex_flags & NFSEXP_V4ROOT))
123 		return nfs_ok;
124 	/*
125 	 * v2/v3 clients have no need for the V4ROOT export--they use
126 	 * the mount protocl instead; also, further V4ROOT checks may be
127 	 * in v4-specific code, in which case v2/v3 clients could bypass
128 	 * them.
129 	 */
130 	if (!nfsd_v4client(rqstp))
131 		return nfserr_stale;
132 	/*
133 	 * We're exposing only the directories and symlinks that have to be
134 	 * traversed on the way to real exports:
135 	 */
136 	if (unlikely(!d_is_dir(dentry) &&
137 		     !d_is_symlink(dentry)))
138 		return nfserr_stale;
139 	/*
140 	 * A pseudoroot export gives permission to access only one
141 	 * single directory; the kernel has to make another upcall
142 	 * before granting access to anything else under it:
143 	 */
144 	if (unlikely(dentry != exp->ex_path.dentry))
145 		return nfserr_stale;
146 	return nfs_ok;
147 }
148 
149 /*
150  * Use the given filehandle to look up the corresponding export and
151  * dentry.  On success, the results are used to set fh_export and
152  * fh_dentry.
153  */
nfsd_set_fh_dentry(struct svc_rqst * rqstp,struct svc_fh * fhp)154 static __be32 nfsd_set_fh_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp)
155 {
156 	struct knfsd_fh	*fh = &fhp->fh_handle;
157 	struct fid *fid = NULL;
158 	struct svc_export *exp;
159 	struct dentry *dentry;
160 	int fileid_type;
161 	int data_left = fh->fh_size/4;
162 	int len;
163 	__be32 error;
164 
165 	error = nfserr_stale;
166 	if (rqstp->rq_vers > 2)
167 		error = nfserr_badhandle;
168 	if (rqstp->rq_vers == 4 && fh->fh_size == 0)
169 		return nfserr_nofilehandle;
170 
171 	if (fh->fh_version != 1)
172 		return error;
173 
174 	if (--data_left < 0)
175 		return error;
176 	if (fh->fh_auth_type != 0)
177 		return error;
178 	len = key_len(fh->fh_fsid_type) / 4;
179 	if (len == 0)
180 		return error;
181 	if (fh->fh_fsid_type == FSID_MAJOR_MINOR) {
182 		/* deprecated, convert to type 3 */
183 		len = key_len(FSID_ENCODE_DEV)/4;
184 		fh->fh_fsid_type = FSID_ENCODE_DEV;
185 		/*
186 		 * struct knfsd_fh uses host-endian fields, which are
187 		 * sometimes used to hold net-endian values. This
188 		 * confuses sparse, so we must use __force here to
189 		 * keep it from complaining.
190 		 */
191 		fh->fh_fsid[0] = new_encode_dev(MKDEV(ntohl((__force __be32)fh->fh_fsid[0]),
192 						      ntohl((__force __be32)fh->fh_fsid[1])));
193 		fh->fh_fsid[1] = fh->fh_fsid[2];
194 	}
195 	data_left -= len;
196 	if (data_left < 0)
197 		return error;
198 	exp = rqst_exp_find(rqstp, fh->fh_fsid_type, fh->fh_fsid);
199 	fid = (struct fid *)(fh->fh_fsid + len);
200 
201 	error = nfserr_stale;
202 	if (IS_ERR(exp)) {
203 		trace_nfsd_set_fh_dentry_badexport(rqstp, fhp, PTR_ERR(exp));
204 
205 		if (PTR_ERR(exp) == -ENOENT)
206 			return error;
207 
208 		return nfserrno(PTR_ERR(exp));
209 	}
210 
211 	if (exp->ex_flags & NFSEXP_NOSUBTREECHECK) {
212 		/* Elevate privileges so that the lack of 'r' or 'x'
213 		 * permission on some parent directory will
214 		 * not stop exportfs_decode_fh from being able
215 		 * to reconnect a directory into the dentry cache.
216 		 * The same problem can affect "SUBTREECHECK" exports,
217 		 * but as nfsd_acceptable depends on correct
218 		 * access control settings being in effect, we cannot
219 		 * fix that case easily.
220 		 */
221 		struct cred *new = prepare_creds();
222 		if (!new) {
223 			error =  nfserrno(-ENOMEM);
224 			goto out;
225 		}
226 		new->cap_effective =
227 			cap_raise_nfsd_set(new->cap_effective,
228 					   new->cap_permitted);
229 		put_cred(override_creds(new));
230 		put_cred(new);
231 	} else {
232 		error = nfsd_setuser_and_check_port(rqstp, exp);
233 		if (error)
234 			goto out;
235 	}
236 
237 	/*
238 	 * Look up the dentry using the NFS file handle.
239 	 */
240 	error = nfserr_stale;
241 	if (rqstp->rq_vers > 2)
242 		error = nfserr_badhandle;
243 
244 	fileid_type = fh->fh_fileid_type;
245 
246 	if (fileid_type == FILEID_ROOT)
247 		dentry = dget(exp->ex_path.dentry);
248 	else {
249 		dentry = exportfs_decode_fh_raw(exp->ex_path.mnt, fid,
250 						data_left, fileid_type,
251 						nfsd_acceptable, exp);
252 		if (IS_ERR_OR_NULL(dentry)) {
253 			trace_nfsd_set_fh_dentry_badhandle(rqstp, fhp,
254 					dentry ?  PTR_ERR(dentry) : -ESTALE);
255 			switch (PTR_ERR(dentry)) {
256 			case -ENOMEM:
257 			case -ETIMEDOUT:
258 				break;
259 			default:
260 				dentry = ERR_PTR(-ESTALE);
261 			}
262 		}
263 	}
264 	if (dentry == NULL)
265 		goto out;
266 	if (IS_ERR(dentry)) {
267 		if (PTR_ERR(dentry) != -EINVAL)
268 			error = nfserrno(PTR_ERR(dentry));
269 		goto out;
270 	}
271 
272 	if (d_is_dir(dentry) &&
273 			(dentry->d_flags & DCACHE_DISCONNECTED)) {
274 		printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %pd2\n",
275 				dentry);
276 	}
277 
278 	fhp->fh_dentry = dentry;
279 	fhp->fh_export = exp;
280 
281 	switch (rqstp->rq_vers) {
282 	case 4:
283 		if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOATOMIC_ATTR)
284 			fhp->fh_no_atomic_attr = true;
285 		break;
286 	case 3:
287 		if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOWCC)
288 			fhp->fh_no_wcc = true;
289 		break;
290 	case 2:
291 		fhp->fh_no_wcc = true;
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: pointer to current rpc request
303  * @fhp: filehandle to be verified
304  * @type: expected type of object pointed to by filehandle
305  * @access: type of access needed to object
306  *
307  * Look up a dentry from the on-the-wire filehandle, check the client's
308  * access to the export, and set the current task's credentials.
309  *
310  * Regardless of success or failure of fh_verify(), fh_put() should be
311  * called on @fhp when the caller is finished with the filehandle.
312  *
313  * fh_verify() may be called multiple times on a given filehandle, for
314  * example, when processing an NFSv4 compound.  The first call will look
315  * up a dentry using the on-the-wire filehandle.  Subsequent calls will
316  * skip the lookup and just perform the other checks and possibly change
317  * the current task's credentials.
318  *
319  * @type specifies the type of object expected using one of the S_IF*
320  * constants defined in include/linux/stat.h.  The caller may use zero
321  * to indicate that it doesn't care, or a negative integer to indicate
322  * that it expects something not of the given type.
323  *
324  * @access is formed from the NFSD_MAY_* constants defined in
325  * fs/nfsd/vfs.h.
326  */
327 __be32
fh_verify(struct svc_rqst * rqstp,struct svc_fh * fhp,umode_t type,int access)328 fh_verify(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, int access)
329 {
330 	struct svc_export *exp = NULL;
331 	struct dentry	*dentry;
332 	__be32		error;
333 
334 	if (!fhp->fh_dentry) {
335 		error = nfsd_set_fh_dentry(rqstp, fhp);
336 		if (error)
337 			goto out;
338 	}
339 	dentry = fhp->fh_dentry;
340 	exp = fhp->fh_export;
341 
342 	trace_nfsd_fh_verify(rqstp, fhp, type, access);
343 
344 	/*
345 	 * We still have to do all these permission checks, even when
346 	 * fh_dentry is already set:
347 	 * 	- fh_verify may be called multiple times with different
348 	 * 	  "access" arguments (e.g. nfsd_proc_create calls
349 	 * 	  fh_verify(...,NFSD_MAY_EXEC) first, then later (in
350 	 * 	  nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
351 	 *	- in the NFSv4 case, the filehandle may have been filled
352 	 *	  in by fh_compose, and given a dentry, but further
353 	 *	  compound operations performed with that filehandle
354 	 *	  still need permissions checks.  In the worst case, a
355 	 *	  mountpoint crossing may have changed the export
356 	 *	  options, and we may now need to use a different uid
357 	 *	  (for example, if different id-squashing options are in
358 	 *	  effect on the new filesystem).
359 	 */
360 	error = check_pseudo_root(rqstp, dentry, exp);
361 	if (error)
362 		goto out;
363 
364 	error = nfsd_setuser_and_check_port(rqstp, exp);
365 	if (error)
366 		goto out;
367 
368 	error = nfsd_mode_check(rqstp, dentry, type);
369 	if (error)
370 		goto out;
371 
372 	/*
373 	 * pseudoflavor restrictions are not enforced on NLM,
374 	 * which clients virtually always use auth_sys for,
375 	 * even while using RPCSEC_GSS for NFS.
376 	 */
377 	if (access & NFSD_MAY_LOCK || access & NFSD_MAY_BYPASS_GSS)
378 		goto skip_pseudoflavor_check;
379 	/*
380 	 * Clients may expect to be able to use auth_sys during mount,
381 	 * even if they use gss for everything else; see section 2.3.2
382 	 * of rfc 2623.
383 	 */
384 	if (access & NFSD_MAY_BYPASS_GSS_ON_ROOT
385 			&& exp->ex_path.dentry == dentry)
386 		goto skip_pseudoflavor_check;
387 
388 	error = check_nfsd_access(exp, rqstp);
389 	if (error)
390 		goto out;
391 
392 skip_pseudoflavor_check:
393 	/* Finally, check access permissions. */
394 	error = nfsd_permission(rqstp, exp, dentry, access);
395 out:
396 	trace_nfsd_fh_verify_err(rqstp, fhp, type, access, error);
397 	if (error == nfserr_stale)
398 		nfsd_stats_fh_stale_inc(exp);
399 	return error;
400 }
401 
402 
403 /*
404  * Compose a file handle for an NFS reply.
405  *
406  * Note that when first composed, the dentry may not yet have
407  * an inode.  In this case a call to fh_update should be made
408  * before the fh goes out on the wire ...
409  */
_fh_update(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry)410 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp,
411 		struct dentry *dentry)
412 {
413 	if (dentry != exp->ex_path.dentry) {
414 		struct fid *fid = (struct fid *)
415 			(fhp->fh_handle.fh_fsid + fhp->fh_handle.fh_size/4 - 1);
416 		int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4;
417 		int fh_flags = (exp->ex_flags & NFSEXP_NOSUBTREECHECK) ? 0 :
418 				EXPORT_FH_CONNECTABLE;
419 		int fileid_type =
420 			exportfs_encode_fh(dentry, fid, &maxsize, fh_flags);
421 
422 		fhp->fh_handle.fh_fileid_type =
423 			fileid_type > 0 ? fileid_type : FILEID_INVALID;
424 		fhp->fh_handle.fh_size += maxsize * 4;
425 	} else {
426 		fhp->fh_handle.fh_fileid_type = FILEID_ROOT;
427 	}
428 }
429 
is_root_export(struct svc_export * exp)430 static bool is_root_export(struct svc_export *exp)
431 {
432 	return exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root;
433 }
434 
exp_sb(struct svc_export * exp)435 static struct super_block *exp_sb(struct svc_export *exp)
436 {
437 	return exp->ex_path.dentry->d_sb;
438 }
439 
fsid_type_ok_for_exp(u8 fsid_type,struct svc_export * exp)440 static bool fsid_type_ok_for_exp(u8 fsid_type, struct svc_export *exp)
441 {
442 	switch (fsid_type) {
443 	case FSID_DEV:
444 		if (!old_valid_dev(exp_sb(exp)->s_dev))
445 			return false;
446 		fallthrough;
447 	case FSID_MAJOR_MINOR:
448 	case FSID_ENCODE_DEV:
449 		return exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV;
450 	case FSID_NUM:
451 		return exp->ex_flags & NFSEXP_FSID;
452 	case FSID_UUID8:
453 	case FSID_UUID16:
454 		if (!is_root_export(exp))
455 			return false;
456 		fallthrough;
457 	case FSID_UUID4_INUM:
458 	case FSID_UUID16_INUM:
459 		return exp->ex_uuid != NULL;
460 	}
461 	return true;
462 }
463 
464 
set_version_and_fsid_type(struct svc_fh * fhp,struct svc_export * exp,struct svc_fh * ref_fh)465 static void set_version_and_fsid_type(struct svc_fh *fhp, struct svc_export *exp, struct svc_fh *ref_fh)
466 {
467 	u8 version;
468 	u8 fsid_type;
469 retry:
470 	version = 1;
471 	if (ref_fh && ref_fh->fh_export == exp) {
472 		version = ref_fh->fh_handle.fh_version;
473 		fsid_type = ref_fh->fh_handle.fh_fsid_type;
474 
475 		ref_fh = NULL;
476 
477 		switch (version) {
478 		case 0xca:
479 			fsid_type = FSID_DEV;
480 			break;
481 		case 1:
482 			break;
483 		default:
484 			goto retry;
485 		}
486 
487 		/*
488 		 * As the fsid -> filesystem mapping was guided by
489 		 * user-space, there is no guarantee that the filesystem
490 		 * actually supports that fsid type. If it doesn't we
491 		 * loop around again without ref_fh set.
492 		 */
493 		if (!fsid_type_ok_for_exp(fsid_type, exp))
494 			goto retry;
495 	} else if (exp->ex_flags & NFSEXP_FSID) {
496 		fsid_type = FSID_NUM;
497 	} else if (exp->ex_uuid) {
498 		if (fhp->fh_maxsize >= 64) {
499 			if (is_root_export(exp))
500 				fsid_type = FSID_UUID16;
501 			else
502 				fsid_type = FSID_UUID16_INUM;
503 		} else {
504 			if (is_root_export(exp))
505 				fsid_type = FSID_UUID8;
506 			else
507 				fsid_type = FSID_UUID4_INUM;
508 		}
509 	} else if (!old_valid_dev(exp_sb(exp)->s_dev))
510 		/* for newer device numbers, we must use a newer fsid format */
511 		fsid_type = FSID_ENCODE_DEV;
512 	else
513 		fsid_type = FSID_DEV;
514 	fhp->fh_handle.fh_version = version;
515 	if (version)
516 		fhp->fh_handle.fh_fsid_type = fsid_type;
517 }
518 
519 __be32
fh_compose(struct svc_fh * fhp,struct svc_export * exp,struct dentry * dentry,struct svc_fh * ref_fh)520 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry,
521 	   struct svc_fh *ref_fh)
522 {
523 	/* ref_fh is a reference file handle.
524 	 * if it is non-null and for the same filesystem, then we should compose
525 	 * a filehandle which is of the same version, where possible.
526 	 */
527 
528 	struct inode * inode = d_inode(dentry);
529 	dev_t ex_dev = exp_sb(exp)->s_dev;
530 
531 	dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n",
532 		MAJOR(ex_dev), MINOR(ex_dev),
533 		(long) d_inode(exp->ex_path.dentry)->i_ino,
534 		dentry,
535 		(inode ? inode->i_ino : 0));
536 
537 	/* Choose filehandle version and fsid type based on
538 	 * the reference filehandle (if it is in the same export)
539 	 * or the export options.
540 	 */
541 	set_version_and_fsid_type(fhp, exp, ref_fh);
542 
543 	/* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
544 	fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false;
545 
546 	if (ref_fh == fhp)
547 		fh_put(ref_fh);
548 
549 	if (fhp->fh_dentry) {
550 		printk(KERN_ERR "fh_compose: fh %pd2 not initialized!\n",
551 		       dentry);
552 	}
553 	if (fhp->fh_maxsize < NFS_FHSIZE)
554 		printk(KERN_ERR "fh_compose: called with maxsize %d! %pd2\n",
555 		       fhp->fh_maxsize,
556 		       dentry);
557 
558 	fhp->fh_dentry = dget(dentry); /* our internal copy */
559 	fhp->fh_export = exp_get(exp);
560 
561 	fhp->fh_handle.fh_size =
562 		key_len(fhp->fh_handle.fh_fsid_type) + 4;
563 	fhp->fh_handle.fh_auth_type = 0;
564 
565 	mk_fsid(fhp->fh_handle.fh_fsid_type,
566 		fhp->fh_handle.fh_fsid,
567 		ex_dev,
568 		d_inode(exp->ex_path.dentry)->i_ino,
569 		exp->ex_fsid, exp->ex_uuid);
570 
571 	if (inode)
572 		_fh_update(fhp, exp, dentry);
573 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) {
574 		fh_put(fhp);
575 		return nfserr_opnotsupp;
576 	}
577 
578 	return 0;
579 }
580 
581 /*
582  * Update file handle information after changing a dentry.
583  * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
584  */
585 __be32
fh_update(struct svc_fh * fhp)586 fh_update(struct svc_fh *fhp)
587 {
588 	struct dentry *dentry;
589 
590 	if (!fhp->fh_dentry)
591 		goto out_bad;
592 
593 	dentry = fhp->fh_dentry;
594 	if (d_really_is_negative(dentry))
595 		goto out_negative;
596 	if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT)
597 		return 0;
598 
599 	_fh_update(fhp, fhp->fh_export, dentry);
600 	if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID)
601 		return nfserr_opnotsupp;
602 	return 0;
603 out_bad:
604 	printk(KERN_ERR "fh_update: fh not verified!\n");
605 	return nfserr_serverfault;
606 out_negative:
607 	printk(KERN_ERR "fh_update: %pd2 still negative!\n",
608 		dentry);
609 	return nfserr_serverfault;
610 }
611 
612 /**
613  * fh_fill_pre_attrs - Fill in pre-op attributes
614  * @fhp: file handle to be updated
615  *
616  */
fh_fill_pre_attrs(struct svc_fh * fhp)617 __be32 __must_check fh_fill_pre_attrs(struct svc_fh *fhp)
618 {
619 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
620 	struct inode *inode;
621 	struct kstat stat;
622 	__be32 err;
623 
624 	if (fhp->fh_no_wcc || fhp->fh_pre_saved)
625 		return nfs_ok;
626 
627 	inode = d_inode(fhp->fh_dentry);
628 	err = fh_getattr(fhp, &stat);
629 	if (err)
630 		return err;
631 
632 	if (v4)
633 		fhp->fh_pre_change = nfsd4_change_attribute(&stat, inode);
634 
635 	fhp->fh_pre_mtime = stat.mtime;
636 	fhp->fh_pre_ctime = stat.ctime;
637 	fhp->fh_pre_size  = stat.size;
638 	fhp->fh_pre_saved = true;
639 	return nfs_ok;
640 }
641 
642 /**
643  * fh_fill_post_attrs - Fill in post-op attributes
644  * @fhp: file handle to be updated
645  *
646  */
fh_fill_post_attrs(struct svc_fh * fhp)647 __be32 fh_fill_post_attrs(struct svc_fh *fhp)
648 {
649 	bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
650 	struct inode *inode = d_inode(fhp->fh_dentry);
651 	__be32 err;
652 
653 	if (fhp->fh_no_wcc)
654 		return nfs_ok;
655 
656 	if (fhp->fh_post_saved)
657 		printk("nfsd: inode locked twice during operation.\n");
658 
659 	err = fh_getattr(fhp, &fhp->fh_post_attr);
660 	if (err)
661 		return err;
662 
663 	fhp->fh_post_saved = true;
664 	if (v4)
665 		fhp->fh_post_change =
666 			nfsd4_change_attribute(&fhp->fh_post_attr, inode);
667 	return nfs_ok;
668 }
669 
670 /**
671  * fh_fill_both_attrs - Fill pre-op and post-op attributes
672  * @fhp: file handle to be updated
673  *
674  * This is used when the directory wasn't changed, but wcc attributes
675  * are needed anyway.
676  */
fh_fill_both_attrs(struct svc_fh * fhp)677 __be32 __must_check fh_fill_both_attrs(struct svc_fh *fhp)
678 {
679 	__be32 err;
680 
681 	err = fh_fill_post_attrs(fhp);
682 	if (err)
683 		return err;
684 
685 	fhp->fh_pre_change = fhp->fh_post_change;
686 	fhp->fh_pre_mtime = fhp->fh_post_attr.mtime;
687 	fhp->fh_pre_ctime = fhp->fh_post_attr.ctime;
688 	fhp->fh_pre_size = fhp->fh_post_attr.size;
689 	fhp->fh_pre_saved = true;
690 	return nfs_ok;
691 }
692 
693 /*
694  * Release a file handle.
695  */
696 void
fh_put(struct svc_fh * fhp)697 fh_put(struct svc_fh *fhp)
698 {
699 	struct dentry * dentry = fhp->fh_dentry;
700 	struct svc_export * exp = fhp->fh_export;
701 	if (dentry) {
702 		fhp->fh_dentry = NULL;
703 		dput(dentry);
704 		fh_clear_pre_post_attrs(fhp);
705 	}
706 	fh_drop_write(fhp);
707 	if (exp) {
708 		exp_put(exp);
709 		fhp->fh_export = NULL;
710 	}
711 	fhp->fh_no_wcc = false;
712 	return;
713 }
714 
715 /*
716  * Shorthand for dprintk()'s
717  */
SVCFH_fmt(struct svc_fh * fhp)718 char * SVCFH_fmt(struct svc_fh *fhp)
719 {
720 	struct knfsd_fh *fh = &fhp->fh_handle;
721 	static char buf[2+1+1+64*3+1];
722 
723 	if (fh->fh_size < 0 || fh->fh_size> 64)
724 		return "bad-fh";
725 	sprintf(buf, "%d: %*ph", fh->fh_size, fh->fh_size, fh->fh_raw);
726 	return buf;
727 }
728 
fsid_source(const struct svc_fh * fhp)729 enum fsid_source fsid_source(const struct svc_fh *fhp)
730 {
731 	if (fhp->fh_handle.fh_version != 1)
732 		return FSIDSOURCE_DEV;
733 	switch(fhp->fh_handle.fh_fsid_type) {
734 	case FSID_DEV:
735 	case FSID_ENCODE_DEV:
736 	case FSID_MAJOR_MINOR:
737 		if (exp_sb(fhp->fh_export)->s_type->fs_flags & FS_REQUIRES_DEV)
738 			return FSIDSOURCE_DEV;
739 		break;
740 	case FSID_NUM:
741 		if (fhp->fh_export->ex_flags & NFSEXP_FSID)
742 			return FSIDSOURCE_FSID;
743 		break;
744 	default:
745 		break;
746 	}
747 	/* either a UUID type filehandle, or the filehandle doesn't
748 	 * match the export.
749 	 */
750 	if (fhp->fh_export->ex_flags & NFSEXP_FSID)
751 		return FSIDSOURCE_FSID;
752 	if (fhp->fh_export->ex_uuid)
753 		return FSIDSOURCE_UUID;
754 	return FSIDSOURCE_DEV;
755 }
756 
757 /*
758  * We could use i_version alone as the change attribute.  However, i_version
759  * can go backwards on a regular file after an unclean shutdown.  On its own
760  * that doesn't necessarily cause a problem, but if i_version goes backwards
761  * and then is incremented again it could reuse a value that was previously
762  * used before boot, and a client who queried the two values might incorrectly
763  * assume nothing changed.
764  *
765  * By using both ctime and the i_version counter we guarantee that as long as
766  * time doesn't go backwards we never reuse an old value. If the filesystem
767  * advertises STATX_ATTR_CHANGE_MONOTONIC, then this mitigation is not
768  * needed.
769  *
770  * We only need to do this for regular files as well. For directories, we
771  * assume that the new change attr is always logged to stable storage in some
772  * fashion before the results can be seen.
773  */
nfsd4_change_attribute(struct kstat * stat,struct inode * inode)774 u64 nfsd4_change_attribute(struct kstat *stat, struct inode *inode)
775 {
776 	u64 chattr;
777 
778 	if (stat->result_mask & STATX_CHANGE_COOKIE) {
779 		chattr = stat->change_cookie;
780 		if (S_ISREG(inode->i_mode) &&
781 		    !(stat->attributes & STATX_ATTR_CHANGE_MONOTONIC)) {
782 			chattr += (u64)stat->ctime.tv_sec << 30;
783 			chattr += stat->ctime.tv_nsec;
784 		}
785 	} else {
786 		chattr = time_to_chattr(&stat->ctime);
787 	}
788 	return chattr;
789 }
790