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 */ 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(&init_user_ns, 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 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 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 } 101 102 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 } 118 119 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 */ 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 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 dprintk("nfsd: fh_verify(%s)\n", SVCFH_fmt(fhp)); 335 336 if (!fhp->fh_dentry) { 337 error = nfsd_set_fh_dentry(rqstp, fhp); 338 if (error) 339 goto out; 340 } 341 dentry = fhp->fh_dentry; 342 exp = fhp->fh_export; 343 /* 344 * We still have to do all these permission checks, even when 345 * fh_dentry is already set: 346 * - fh_verify may be called multiple times with different 347 * "access" arguments (e.g. nfsd_proc_create calls 348 * fh_verify(...,NFSD_MAY_EXEC) first, then later (in 349 * nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE). 350 * - in the NFSv4 case, the filehandle may have been filled 351 * in by fh_compose, and given a dentry, but further 352 * compound operations performed with that filehandle 353 * still need permissions checks. In the worst case, a 354 * mountpoint crossing may have changed the export 355 * options, and we may now need to use a different uid 356 * (for example, if different id-squashing options are in 357 * effect on the new filesystem). 358 */ 359 error = check_pseudo_root(rqstp, dentry, exp); 360 if (error) 361 goto out; 362 363 error = nfsd_setuser_and_check_port(rqstp, exp); 364 if (error) 365 goto out; 366 367 error = nfsd_mode_check(rqstp, dentry, type); 368 if (error) 369 goto out; 370 371 /* 372 * pseudoflavor restrictions are not enforced on NLM, 373 * which clients virtually always use auth_sys for, 374 * even while using RPCSEC_GSS for NFS. 375 */ 376 if (access & NFSD_MAY_LOCK || access & NFSD_MAY_BYPASS_GSS) 377 goto skip_pseudoflavor_check; 378 /* 379 * Clients may expect to be able to use auth_sys during mount, 380 * even if they use gss for everything else; see section 2.3.2 381 * of rfc 2623. 382 */ 383 if (access & NFSD_MAY_BYPASS_GSS_ON_ROOT 384 && exp->ex_path.dentry == dentry) 385 goto skip_pseudoflavor_check; 386 387 error = check_nfsd_access(exp, rqstp); 388 if (error) 389 goto out; 390 391 skip_pseudoflavor_check: 392 /* Finally, check access permissions. */ 393 error = nfsd_permission(rqstp, exp, dentry, access); 394 395 if (error) { 396 dprintk("fh_verify: %pd2 permission failure, " 397 "acc=%x, error=%d\n", 398 dentry, 399 access, ntohl(error)); 400 } 401 out: 402 if (error == nfserr_stale) 403 nfsd_stats_fh_stale_inc(exp); 404 return error; 405 } 406 407 408 /* 409 * Compose a file handle for an NFS reply. 410 * 411 * Note that when first composed, the dentry may not yet have 412 * an inode. In this case a call to fh_update should be made 413 * before the fh goes out on the wire ... 414 */ 415 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp, 416 struct dentry *dentry) 417 { 418 if (dentry != exp->ex_path.dentry) { 419 struct fid *fid = (struct fid *) 420 (fhp->fh_handle.fh_fsid + fhp->fh_handle.fh_size/4 - 1); 421 int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4; 422 int subtreecheck = !(exp->ex_flags & NFSEXP_NOSUBTREECHECK); 423 424 fhp->fh_handle.fh_fileid_type = 425 exportfs_encode_fh(dentry, fid, &maxsize, subtreecheck); 426 fhp->fh_handle.fh_size += maxsize * 4; 427 } else { 428 fhp->fh_handle.fh_fileid_type = FILEID_ROOT; 429 } 430 } 431 432 static bool is_root_export(struct svc_export *exp) 433 { 434 return exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root; 435 } 436 437 static struct super_block *exp_sb(struct svc_export *exp) 438 { 439 return exp->ex_path.dentry->d_sb; 440 } 441 442 static bool fsid_type_ok_for_exp(u8 fsid_type, struct svc_export *exp) 443 { 444 switch (fsid_type) { 445 case FSID_DEV: 446 if (!old_valid_dev(exp_sb(exp)->s_dev)) 447 return false; 448 fallthrough; 449 case FSID_MAJOR_MINOR: 450 case FSID_ENCODE_DEV: 451 return exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV; 452 case FSID_NUM: 453 return exp->ex_flags & NFSEXP_FSID; 454 case FSID_UUID8: 455 case FSID_UUID16: 456 if (!is_root_export(exp)) 457 return false; 458 fallthrough; 459 case FSID_UUID4_INUM: 460 case FSID_UUID16_INUM: 461 return exp->ex_uuid != NULL; 462 } 463 return true; 464 } 465 466 467 static void set_version_and_fsid_type(struct svc_fh *fhp, struct svc_export *exp, struct svc_fh *ref_fh) 468 { 469 u8 version; 470 u8 fsid_type; 471 retry: 472 version = 1; 473 if (ref_fh && ref_fh->fh_export == exp) { 474 version = ref_fh->fh_handle.fh_version; 475 fsid_type = ref_fh->fh_handle.fh_fsid_type; 476 477 ref_fh = NULL; 478 479 switch (version) { 480 case 0xca: 481 fsid_type = FSID_DEV; 482 break; 483 case 1: 484 break; 485 default: 486 goto retry; 487 } 488 489 /* 490 * As the fsid -> filesystem mapping was guided by 491 * user-space, there is no guarantee that the filesystem 492 * actually supports that fsid type. If it doesn't we 493 * loop around again without ref_fh set. 494 */ 495 if (!fsid_type_ok_for_exp(fsid_type, exp)) 496 goto retry; 497 } else if (exp->ex_flags & NFSEXP_FSID) { 498 fsid_type = FSID_NUM; 499 } else if (exp->ex_uuid) { 500 if (fhp->fh_maxsize >= 64) { 501 if (is_root_export(exp)) 502 fsid_type = FSID_UUID16; 503 else 504 fsid_type = FSID_UUID16_INUM; 505 } else { 506 if (is_root_export(exp)) 507 fsid_type = FSID_UUID8; 508 else 509 fsid_type = FSID_UUID4_INUM; 510 } 511 } else if (!old_valid_dev(exp_sb(exp)->s_dev)) 512 /* for newer device numbers, we must use a newer fsid format */ 513 fsid_type = FSID_ENCODE_DEV; 514 else 515 fsid_type = FSID_DEV; 516 fhp->fh_handle.fh_version = version; 517 if (version) 518 fhp->fh_handle.fh_fsid_type = fsid_type; 519 } 520 521 __be32 522 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry, 523 struct svc_fh *ref_fh) 524 { 525 /* ref_fh is a reference file handle. 526 * if it is non-null and for the same filesystem, then we should compose 527 * a filehandle which is of the same version, where possible. 528 */ 529 530 struct inode * inode = d_inode(dentry); 531 dev_t ex_dev = exp_sb(exp)->s_dev; 532 533 dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n", 534 MAJOR(ex_dev), MINOR(ex_dev), 535 (long) d_inode(exp->ex_path.dentry)->i_ino, 536 dentry, 537 (inode ? inode->i_ino : 0)); 538 539 /* Choose filehandle version and fsid type based on 540 * the reference filehandle (if it is in the same export) 541 * or the export options. 542 */ 543 set_version_and_fsid_type(fhp, exp, ref_fh); 544 545 /* If we have a ref_fh, then copy the fh_no_wcc setting from it. */ 546 fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false; 547 548 if (ref_fh == fhp) 549 fh_put(ref_fh); 550 551 if (fhp->fh_locked || fhp->fh_dentry) { 552 printk(KERN_ERR "fh_compose: fh %pd2 not initialized!\n", 553 dentry); 554 } 555 if (fhp->fh_maxsize < NFS_FHSIZE) 556 printk(KERN_ERR "fh_compose: called with maxsize %d! %pd2\n", 557 fhp->fh_maxsize, 558 dentry); 559 560 fhp->fh_dentry = dget(dentry); /* our internal copy */ 561 fhp->fh_export = exp_get(exp); 562 563 fhp->fh_handle.fh_size = 564 key_len(fhp->fh_handle.fh_fsid_type) + 4; 565 fhp->fh_handle.fh_auth_type = 0; 566 567 mk_fsid(fhp->fh_handle.fh_fsid_type, 568 fhp->fh_handle.fh_fsid, 569 ex_dev, 570 d_inode(exp->ex_path.dentry)->i_ino, 571 exp->ex_fsid, exp->ex_uuid); 572 573 if (inode) 574 _fh_update(fhp, exp, dentry); 575 if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) { 576 fh_put(fhp); 577 return nfserr_opnotsupp; 578 } 579 580 return 0; 581 } 582 583 /* 584 * Update file handle information after changing a dentry. 585 * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create 586 */ 587 __be32 588 fh_update(struct svc_fh *fhp) 589 { 590 struct dentry *dentry; 591 592 if (!fhp->fh_dentry) 593 goto out_bad; 594 595 dentry = fhp->fh_dentry; 596 if (d_really_is_negative(dentry)) 597 goto out_negative; 598 if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT) 599 return 0; 600 601 _fh_update(fhp, fhp->fh_export, dentry); 602 if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) 603 return nfserr_opnotsupp; 604 return 0; 605 out_bad: 606 printk(KERN_ERR "fh_update: fh not verified!\n"); 607 return nfserr_serverfault; 608 out_negative: 609 printk(KERN_ERR "fh_update: %pd2 still negative!\n", 610 dentry); 611 return nfserr_serverfault; 612 } 613 614 /* 615 * Release a file handle. 616 */ 617 void 618 fh_put(struct svc_fh *fhp) 619 { 620 struct dentry * dentry = fhp->fh_dentry; 621 struct svc_export * exp = fhp->fh_export; 622 if (dentry) { 623 fh_unlock(fhp); 624 fhp->fh_dentry = NULL; 625 dput(dentry); 626 fh_clear_wcc(fhp); 627 } 628 fh_drop_write(fhp); 629 if (exp) { 630 exp_put(exp); 631 fhp->fh_export = NULL; 632 } 633 fhp->fh_no_wcc = false; 634 return; 635 } 636 637 /* 638 * Shorthand for dprintk()'s 639 */ 640 char * SVCFH_fmt(struct svc_fh *fhp) 641 { 642 struct knfsd_fh *fh = &fhp->fh_handle; 643 static char buf[2+1+1+64*3+1]; 644 645 if (fh->fh_size < 0 || fh->fh_size> 64) 646 return "bad-fh"; 647 sprintf(buf, "%d: %*ph", fh->fh_size, fh->fh_size, fh->fh_raw); 648 return buf; 649 } 650 651 enum fsid_source fsid_source(const struct svc_fh *fhp) 652 { 653 if (fhp->fh_handle.fh_version != 1) 654 return FSIDSOURCE_DEV; 655 switch(fhp->fh_handle.fh_fsid_type) { 656 case FSID_DEV: 657 case FSID_ENCODE_DEV: 658 case FSID_MAJOR_MINOR: 659 if (exp_sb(fhp->fh_export)->s_type->fs_flags & FS_REQUIRES_DEV) 660 return FSIDSOURCE_DEV; 661 break; 662 case FSID_NUM: 663 if (fhp->fh_export->ex_flags & NFSEXP_FSID) 664 return FSIDSOURCE_FSID; 665 break; 666 default: 667 break; 668 } 669 /* either a UUID type filehandle, or the filehandle doesn't 670 * match the export. 671 */ 672 if (fhp->fh_export->ex_flags & NFSEXP_FSID) 673 return FSIDSOURCE_FSID; 674 if (fhp->fh_export->ex_uuid) 675 return FSIDSOURCE_UUID; 676 return FSIDSOURCE_DEV; 677 } 678