1 /* 2 * File operations used by nfsd. Some of these have been ripped from 3 * other parts of the kernel because they weren't exported, others 4 * are partial duplicates with added or changed functionality. 5 * 6 * Note that several functions dget() the dentry upon which they want 7 * to act, most notably those that create directory entries. Response 8 * dentry's are dput()'d if necessary in the release callback. 9 * So if you notice code paths that apparently fail to dput() the 10 * dentry, don't worry--they have been taken care of. 11 * 12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de> 13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp> 14 */ 15 16 #include <linux/fs.h> 17 #include <linux/file.h> 18 #include <linux/splice.h> 19 #include <linux/fcntl.h> 20 #include <linux/namei.h> 21 #include <linux/delay.h> 22 #include <linux/fsnotify.h> 23 #include <linux/posix_acl_xattr.h> 24 #include <linux/xattr.h> 25 #include <linux/jhash.h> 26 #include <linux/ima.h> 27 #include <linux/slab.h> 28 #include <asm/uaccess.h> 29 #include <linux/exportfs.h> 30 #include <linux/writeback.h> 31 32 #ifdef CONFIG_NFSD_V3 33 #include "xdr3.h" 34 #endif /* CONFIG_NFSD_V3 */ 35 36 #ifdef CONFIG_NFSD_V4 37 #include "acl.h" 38 #include "idmap.h" 39 #endif /* CONFIG_NFSD_V4 */ 40 41 #include "nfsd.h" 42 #include "vfs.h" 43 44 #define NFSDDBG_FACILITY NFSDDBG_FILEOP 45 46 47 /* 48 * This is a cache of readahead params that help us choose the proper 49 * readahead strategy. Initially, we set all readahead parameters to 0 50 * and let the VFS handle things. 51 * If you increase the number of cached files very much, you'll need to 52 * add a hash table here. 53 */ 54 struct raparms { 55 struct raparms *p_next; 56 unsigned int p_count; 57 ino_t p_ino; 58 dev_t p_dev; 59 int p_set; 60 struct file_ra_state p_ra; 61 unsigned int p_hindex; 62 }; 63 64 struct raparm_hbucket { 65 struct raparms *pb_head; 66 spinlock_t pb_lock; 67 } ____cacheline_aligned_in_smp; 68 69 #define RAPARM_HASH_BITS 4 70 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) 71 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) 72 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; 73 74 /* 75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed 76 * a mount point. 77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged, 78 * or nfs_ok having possibly changed *dpp and *expp 79 */ 80 int 81 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 82 struct svc_export **expp) 83 { 84 struct svc_export *exp = *expp, *exp2 = NULL; 85 struct dentry *dentry = *dpp; 86 struct path path = {.mnt = mntget(exp->ex_path.mnt), 87 .dentry = dget(dentry)}; 88 int err = 0; 89 90 err = follow_down(&path); 91 if (err < 0) 92 goto out; 93 94 exp2 = rqst_exp_get_by_name(rqstp, &path); 95 if (IS_ERR(exp2)) { 96 err = PTR_ERR(exp2); 97 /* 98 * We normally allow NFS clients to continue 99 * "underneath" a mountpoint that is not exported. 100 * The exception is V4ROOT, where no traversal is ever 101 * allowed without an explicit export of the new 102 * directory. 103 */ 104 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT)) 105 err = 0; 106 path_put(&path); 107 goto out; 108 } 109 if (nfsd_v4client(rqstp) || 110 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) { 111 /* successfully crossed mount point */ 112 /* 113 * This is subtle: path.dentry is *not* on path.mnt 114 * at this point. The only reason we are safe is that 115 * original mnt is pinned down by exp, so we should 116 * put path *before* putting exp 117 */ 118 *dpp = path.dentry; 119 path.dentry = dentry; 120 *expp = exp2; 121 exp2 = exp; 122 } 123 path_put(&path); 124 exp_put(exp2); 125 out: 126 return err; 127 } 128 129 static void follow_to_parent(struct path *path) 130 { 131 struct dentry *dp; 132 133 while (path->dentry == path->mnt->mnt_root && follow_up(path)) 134 ; 135 dp = dget_parent(path->dentry); 136 dput(path->dentry); 137 path->dentry = dp; 138 } 139 140 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp) 141 { 142 struct svc_export *exp2; 143 struct path path = {.mnt = mntget((*exp)->ex_path.mnt), 144 .dentry = dget(dparent)}; 145 146 follow_to_parent(&path); 147 148 exp2 = rqst_exp_parent(rqstp, &path); 149 if (PTR_ERR(exp2) == -ENOENT) { 150 *dentryp = dget(dparent); 151 } else if (IS_ERR(exp2)) { 152 path_put(&path); 153 return PTR_ERR(exp2); 154 } else { 155 *dentryp = dget(path.dentry); 156 exp_put(*exp); 157 *exp = exp2; 158 } 159 path_put(&path); 160 return 0; 161 } 162 163 /* 164 * For nfsd purposes, we treat V4ROOT exports as though there was an 165 * export at *every* directory. 166 */ 167 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp) 168 { 169 if (d_mountpoint(dentry)) 170 return 1; 171 if (nfsd4_is_junction(dentry)) 172 return 1; 173 if (!(exp->ex_flags & NFSEXP_V4ROOT)) 174 return 0; 175 return dentry->d_inode != NULL; 176 } 177 178 __be32 179 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp, 180 const char *name, unsigned int len, 181 struct svc_export **exp_ret, struct dentry **dentry_ret) 182 { 183 struct svc_export *exp; 184 struct dentry *dparent; 185 struct dentry *dentry; 186 int host_err; 187 188 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); 189 190 dparent = fhp->fh_dentry; 191 exp = fhp->fh_export; 192 exp_get(exp); 193 194 /* Lookup the name, but don't follow links */ 195 if (isdotent(name, len)) { 196 if (len==1) 197 dentry = dget(dparent); 198 else if (dparent != exp->ex_path.dentry) 199 dentry = dget_parent(dparent); 200 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp)) 201 dentry = dget(dparent); /* .. == . just like at / */ 202 else { 203 /* checking mountpoint crossing is very different when stepping up */ 204 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry); 205 if (host_err) 206 goto out_nfserr; 207 } 208 } else { 209 fh_lock(fhp); 210 dentry = lookup_one_len(name, dparent, len); 211 host_err = PTR_ERR(dentry); 212 if (IS_ERR(dentry)) 213 goto out_nfserr; 214 /* 215 * check if we have crossed a mount point ... 216 */ 217 if (nfsd_mountpoint(dentry, exp)) { 218 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { 219 dput(dentry); 220 goto out_nfserr; 221 } 222 } 223 } 224 *dentry_ret = dentry; 225 *exp_ret = exp; 226 return 0; 227 228 out_nfserr: 229 exp_put(exp); 230 return nfserrno(host_err); 231 } 232 233 /* 234 * Look up one component of a pathname. 235 * N.B. After this call _both_ fhp and resfh need an fh_put 236 * 237 * If the lookup would cross a mountpoint, and the mounted filesystem 238 * is exported to the client with NFSEXP_NOHIDE, then the lookup is 239 * accepted as it stands and the mounted directory is 240 * returned. Otherwise the covered directory is returned. 241 * NOTE: this mountpoint crossing is not supported properly by all 242 * clients and is explicitly disallowed for NFSv3 243 * NeilBrown <neilb@cse.unsw.edu.au> 244 */ 245 __be32 246 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, 247 unsigned int len, struct svc_fh *resfh) 248 { 249 struct svc_export *exp; 250 struct dentry *dentry; 251 __be32 err; 252 253 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 254 if (err) 255 return err; 256 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry); 257 if (err) 258 return err; 259 err = check_nfsd_access(exp, rqstp); 260 if (err) 261 goto out; 262 /* 263 * Note: we compose the file handle now, but as the 264 * dentry may be negative, it may need to be updated. 265 */ 266 err = fh_compose(resfh, exp, dentry, fhp); 267 if (!err && !dentry->d_inode) 268 err = nfserr_noent; 269 out: 270 dput(dentry); 271 exp_put(exp); 272 return err; 273 } 274 275 static int nfsd_break_lease(struct inode *inode) 276 { 277 if (!S_ISREG(inode->i_mode)) 278 return 0; 279 return break_lease(inode, O_WRONLY | O_NONBLOCK); 280 } 281 282 /* 283 * Commit metadata changes to stable storage. 284 */ 285 static int 286 commit_metadata(struct svc_fh *fhp) 287 { 288 struct inode *inode = fhp->fh_dentry->d_inode; 289 const struct export_operations *export_ops = inode->i_sb->s_export_op; 290 291 if (!EX_ISSYNC(fhp->fh_export)) 292 return 0; 293 294 if (export_ops->commit_metadata) 295 return export_ops->commit_metadata(inode); 296 return sync_inode_metadata(inode, 1); 297 } 298 299 /* 300 * Set various file attributes. 301 * N.B. After this call fhp needs an fh_put 302 */ 303 __be32 304 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, 305 int check_guard, time_t guardtime) 306 { 307 struct dentry *dentry; 308 struct inode *inode; 309 int accmode = NFSD_MAY_SATTR; 310 int ftype = 0; 311 __be32 err; 312 int host_err; 313 int size_change = 0; 314 315 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) 316 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE; 317 if (iap->ia_valid & ATTR_SIZE) 318 ftype = S_IFREG; 319 320 /* Get inode */ 321 err = fh_verify(rqstp, fhp, ftype, accmode); 322 if (err) 323 goto out; 324 325 dentry = fhp->fh_dentry; 326 inode = dentry->d_inode; 327 328 /* Ignore any mode updates on symlinks */ 329 if (S_ISLNK(inode->i_mode)) 330 iap->ia_valid &= ~ATTR_MODE; 331 332 if (!iap->ia_valid) 333 goto out; 334 335 /* 336 * NFSv2 does not differentiate between "set-[ac]time-to-now" 337 * which only requires access, and "set-[ac]time-to-X" which 338 * requires ownership. 339 * So if it looks like it might be "set both to the same time which 340 * is close to now", and if inode_change_ok fails, then we 341 * convert to "set to now" instead of "set to explicit time" 342 * 343 * We only call inode_change_ok as the last test as technically 344 * it is not an interface that we should be using. It is only 345 * valid if the filesystem does not define it's own i_op->setattr. 346 */ 347 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) 348 #define MAX_TOUCH_TIME_ERROR (30*60) 349 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET && 350 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) { 351 /* 352 * Looks probable. 353 * 354 * Now just make sure time is in the right ballpark. 355 * Solaris, at least, doesn't seem to care what the time 356 * request is. We require it be within 30 minutes of now. 357 */ 358 time_t delta = iap->ia_atime.tv_sec - get_seconds(); 359 if (delta < 0) 360 delta = -delta; 361 if (delta < MAX_TOUCH_TIME_ERROR && 362 inode_change_ok(inode, iap) != 0) { 363 /* 364 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME. 365 * This will cause notify_change to set these times 366 * to "now" 367 */ 368 iap->ia_valid &= ~BOTH_TIME_SET; 369 } 370 } 371 372 /* 373 * The size case is special. 374 * It changes the file as well as the attributes. 375 */ 376 if (iap->ia_valid & ATTR_SIZE) { 377 if (iap->ia_size < inode->i_size) { 378 err = nfsd_permission(rqstp, fhp->fh_export, dentry, 379 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE); 380 if (err) 381 goto out; 382 } 383 384 host_err = get_write_access(inode); 385 if (host_err) 386 goto out_nfserr; 387 388 size_change = 1; 389 host_err = locks_verify_truncate(inode, NULL, iap->ia_size); 390 if (host_err) { 391 put_write_access(inode); 392 goto out_nfserr; 393 } 394 } 395 396 /* sanitize the mode change */ 397 if (iap->ia_valid & ATTR_MODE) { 398 iap->ia_mode &= S_IALLUGO; 399 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO); 400 } 401 402 /* Revoke setuid/setgid on chown */ 403 if (!S_ISDIR(inode->i_mode) && 404 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) || 405 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) { 406 iap->ia_valid |= ATTR_KILL_PRIV; 407 if (iap->ia_valid & ATTR_MODE) { 408 /* we're setting mode too, just clear the s*id bits */ 409 iap->ia_mode &= ~S_ISUID; 410 if (iap->ia_mode & S_IXGRP) 411 iap->ia_mode &= ~S_ISGID; 412 } else { 413 /* set ATTR_KILL_* bits and let VFS handle it */ 414 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID); 415 } 416 } 417 418 /* Change the attributes. */ 419 420 iap->ia_valid |= ATTR_CTIME; 421 422 err = nfserr_notsync; 423 if (!check_guard || guardtime == inode->i_ctime.tv_sec) { 424 host_err = nfsd_break_lease(inode); 425 if (host_err) 426 goto out_nfserr; 427 fh_lock(fhp); 428 429 host_err = notify_change(dentry, iap); 430 err = nfserrno(host_err); 431 fh_unlock(fhp); 432 } 433 if (size_change) 434 put_write_access(inode); 435 if (!err) 436 commit_metadata(fhp); 437 out: 438 return err; 439 440 out_nfserr: 441 err = nfserrno(host_err); 442 goto out; 443 } 444 445 #if defined(CONFIG_NFSD_V2_ACL) || \ 446 defined(CONFIG_NFSD_V3_ACL) || \ 447 defined(CONFIG_NFSD_V4) 448 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf) 449 { 450 ssize_t buflen; 451 ssize_t ret; 452 453 buflen = vfs_getxattr(dentry, key, NULL, 0); 454 if (buflen <= 0) 455 return buflen; 456 457 *buf = kmalloc(buflen, GFP_KERNEL); 458 if (!*buf) 459 return -ENOMEM; 460 461 ret = vfs_getxattr(dentry, key, *buf, buflen); 462 if (ret < 0) 463 kfree(*buf); 464 return ret; 465 } 466 #endif 467 468 #if defined(CONFIG_NFSD_V4) 469 static int 470 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) 471 { 472 int len; 473 size_t buflen; 474 char *buf = NULL; 475 int error = 0; 476 477 buflen = posix_acl_xattr_size(pacl->a_count); 478 buf = kmalloc(buflen, GFP_KERNEL); 479 error = -ENOMEM; 480 if (buf == NULL) 481 goto out; 482 483 len = posix_acl_to_xattr(pacl, buf, buflen); 484 if (len < 0) { 485 error = len; 486 goto out; 487 } 488 489 error = vfs_setxattr(dentry, key, buf, len, 0); 490 out: 491 kfree(buf); 492 return error; 493 } 494 495 __be32 496 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, 497 struct nfs4_acl *acl) 498 { 499 __be32 error; 500 int host_error; 501 struct dentry *dentry; 502 struct inode *inode; 503 struct posix_acl *pacl = NULL, *dpacl = NULL; 504 unsigned int flags = 0; 505 506 /* Get inode */ 507 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR); 508 if (error) 509 return error; 510 511 dentry = fhp->fh_dentry; 512 inode = dentry->d_inode; 513 if (S_ISDIR(inode->i_mode)) 514 flags = NFS4_ACL_DIR; 515 516 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); 517 if (host_error == -EINVAL) { 518 return nfserr_attrnotsupp; 519 } else if (host_error < 0) 520 goto out_nfserr; 521 522 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); 523 if (host_error < 0) 524 goto out_release; 525 526 if (S_ISDIR(inode->i_mode)) 527 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); 528 529 out_release: 530 posix_acl_release(pacl); 531 posix_acl_release(dpacl); 532 out_nfserr: 533 if (host_error == -EOPNOTSUPP) 534 return nfserr_attrnotsupp; 535 else 536 return nfserrno(host_error); 537 } 538 539 static struct posix_acl * 540 _get_posix_acl(struct dentry *dentry, char *key) 541 { 542 void *buf = NULL; 543 struct posix_acl *pacl = NULL; 544 int buflen; 545 546 buflen = nfsd_getxattr(dentry, key, &buf); 547 if (!buflen) 548 buflen = -ENODATA; 549 if (buflen <= 0) 550 return ERR_PTR(buflen); 551 552 pacl = posix_acl_from_xattr(buf, buflen); 553 kfree(buf); 554 return pacl; 555 } 556 557 int 558 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) 559 { 560 struct inode *inode = dentry->d_inode; 561 int error = 0; 562 struct posix_acl *pacl = NULL, *dpacl = NULL; 563 unsigned int flags = 0; 564 565 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); 566 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) 567 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); 568 if (IS_ERR(pacl)) { 569 error = PTR_ERR(pacl); 570 pacl = NULL; 571 goto out; 572 } 573 574 if (S_ISDIR(inode->i_mode)) { 575 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); 576 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) 577 dpacl = NULL; 578 else if (IS_ERR(dpacl)) { 579 error = PTR_ERR(dpacl); 580 dpacl = NULL; 581 goto out; 582 } 583 flags = NFS4_ACL_DIR; 584 } 585 586 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); 587 if (IS_ERR(*acl)) { 588 error = PTR_ERR(*acl); 589 *acl = NULL; 590 } 591 out: 592 posix_acl_release(pacl); 593 posix_acl_release(dpacl); 594 return error; 595 } 596 597 /* 598 * NFS junction information is stored in an extended attribute. 599 */ 600 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs" 601 602 /** 603 * nfsd4_is_junction - Test if an object could be an NFS junction 604 * 605 * @dentry: object to test 606 * 607 * Returns 1 if "dentry" appears to contain NFS junction information. 608 * Otherwise 0 is returned. 609 */ 610 int nfsd4_is_junction(struct dentry *dentry) 611 { 612 struct inode *inode = dentry->d_inode; 613 614 if (inode == NULL) 615 return 0; 616 if (inode->i_mode & S_IXUGO) 617 return 0; 618 if (!(inode->i_mode & S_ISVTX)) 619 return 0; 620 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0) 621 return 0; 622 return 1; 623 } 624 #endif /* defined(CONFIG_NFSD_V4) */ 625 626 #ifdef CONFIG_NFSD_V3 627 /* 628 * Check server access rights to a file system object 629 */ 630 struct accessmap { 631 u32 access; 632 int how; 633 }; 634 static struct accessmap nfs3_regaccess[] = { 635 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 636 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 637 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, 638 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, 639 640 { 0, 0 } 641 }; 642 643 static struct accessmap nfs3_diraccess[] = { 644 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 645 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, 646 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, 647 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, 648 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, 649 650 { 0, 0 } 651 }; 652 653 static struct accessmap nfs3_anyaccess[] = { 654 /* Some clients - Solaris 2.6 at least, make an access call 655 * to the server to check for access for things like /dev/null 656 * (which really, the server doesn't care about). So 657 * We provide simple access checking for them, looking 658 * mainly at mode bits, and we make sure to ignore read-only 659 * filesystem checks 660 */ 661 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 662 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 663 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 664 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 665 666 { 0, 0 } 667 }; 668 669 __be32 670 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) 671 { 672 struct accessmap *map; 673 struct svc_export *export; 674 struct dentry *dentry; 675 u32 query, result = 0, sresult = 0; 676 __be32 error; 677 678 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); 679 if (error) 680 goto out; 681 682 export = fhp->fh_export; 683 dentry = fhp->fh_dentry; 684 685 if (S_ISREG(dentry->d_inode->i_mode)) 686 map = nfs3_regaccess; 687 else if (S_ISDIR(dentry->d_inode->i_mode)) 688 map = nfs3_diraccess; 689 else 690 map = nfs3_anyaccess; 691 692 693 query = *access; 694 for (; map->access; map++) { 695 if (map->access & query) { 696 __be32 err2; 697 698 sresult |= map->access; 699 700 err2 = nfsd_permission(rqstp, export, dentry, map->how); 701 switch (err2) { 702 case nfs_ok: 703 result |= map->access; 704 break; 705 706 /* the following error codes just mean the access was not allowed, 707 * rather than an error occurred */ 708 case nfserr_rofs: 709 case nfserr_acces: 710 case nfserr_perm: 711 /* simply don't "or" in the access bit. */ 712 break; 713 default: 714 error = err2; 715 goto out; 716 } 717 } 718 } 719 *access = result; 720 if (supported) 721 *supported = sresult; 722 723 out: 724 return error; 725 } 726 #endif /* CONFIG_NFSD_V3 */ 727 728 static int nfsd_open_break_lease(struct inode *inode, int access) 729 { 730 unsigned int mode; 731 732 if (access & NFSD_MAY_NOT_BREAK_LEASE) 733 return 0; 734 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY; 735 return break_lease(inode, mode | O_NONBLOCK); 736 } 737 738 /* 739 * Open an existing file or directory. 740 * The access argument indicates the type of open (read/write/lock) 741 * N.B. After this call fhp needs an fh_put 742 */ 743 __be32 744 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 745 int access, struct file **filp) 746 { 747 struct dentry *dentry; 748 struct inode *inode; 749 int flags = O_RDONLY|O_LARGEFILE; 750 __be32 err; 751 int host_err = 0; 752 753 validate_process_creds(); 754 755 /* 756 * If we get here, then the client has already done an "open", 757 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 758 * in case a chmod has now revoked permission. 759 */ 760 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE); 761 if (err) 762 goto out; 763 764 dentry = fhp->fh_dentry; 765 inode = dentry->d_inode; 766 767 /* Disallow write access to files with the append-only bit set 768 * or any access when mandatory locking enabled 769 */ 770 err = nfserr_perm; 771 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE)) 772 goto out; 773 /* 774 * We must ignore files (but only files) which might have mandatory 775 * locks on them because there is no way to know if the accesser has 776 * the lock. 777 */ 778 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) 779 goto out; 780 781 if (!inode->i_fop) 782 goto out; 783 784 host_err = nfsd_open_break_lease(inode, access); 785 if (host_err) /* NOMEM or WOULDBLOCK */ 786 goto out_nfserr; 787 788 if (access & NFSD_MAY_WRITE) { 789 if (access & NFSD_MAY_READ) 790 flags = O_RDWR|O_LARGEFILE; 791 else 792 flags = O_WRONLY|O_LARGEFILE; 793 } 794 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt), 795 flags, current_cred()); 796 if (IS_ERR(*filp)) 797 host_err = PTR_ERR(*filp); 798 else 799 host_err = ima_file_check(*filp, access); 800 out_nfserr: 801 err = nfserrno(host_err); 802 out: 803 validate_process_creds(); 804 return err; 805 } 806 807 /* 808 * Close a file. 809 */ 810 void 811 nfsd_close(struct file *filp) 812 { 813 fput(filp); 814 } 815 816 /* 817 * Obtain the readahead parameters for the file 818 * specified by (dev, ino). 819 */ 820 821 static inline struct raparms * 822 nfsd_get_raparms(dev_t dev, ino_t ino) 823 { 824 struct raparms *ra, **rap, **frap = NULL; 825 int depth = 0; 826 unsigned int hash; 827 struct raparm_hbucket *rab; 828 829 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 830 rab = &raparm_hash[hash]; 831 832 spin_lock(&rab->pb_lock); 833 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 834 if (ra->p_ino == ino && ra->p_dev == dev) 835 goto found; 836 depth++; 837 if (ra->p_count == 0) 838 frap = rap; 839 } 840 depth = nfsdstats.ra_size; 841 if (!frap) { 842 spin_unlock(&rab->pb_lock); 843 return NULL; 844 } 845 rap = frap; 846 ra = *frap; 847 ra->p_dev = dev; 848 ra->p_ino = ino; 849 ra->p_set = 0; 850 ra->p_hindex = hash; 851 found: 852 if (rap != &rab->pb_head) { 853 *rap = ra->p_next; 854 ra->p_next = rab->pb_head; 855 rab->pb_head = ra; 856 } 857 ra->p_count++; 858 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 859 spin_unlock(&rab->pb_lock); 860 return ra; 861 } 862 863 /* 864 * Grab and keep cached pages associated with a file in the svc_rqst 865 * so that they can be passed to the network sendmsg/sendpage routines 866 * directly. They will be released after the sending has completed. 867 */ 868 static int 869 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 870 struct splice_desc *sd) 871 { 872 struct svc_rqst *rqstp = sd->u.data; 873 struct page **pp = rqstp->rq_respages + rqstp->rq_resused; 874 struct page *page = buf->page; 875 size_t size; 876 877 size = sd->len; 878 879 if (rqstp->rq_res.page_len == 0) { 880 get_page(page); 881 put_page(*pp); 882 *pp = page; 883 rqstp->rq_resused++; 884 rqstp->rq_res.page_base = buf->offset; 885 rqstp->rq_res.page_len = size; 886 } else if (page != pp[-1]) { 887 get_page(page); 888 if (*pp) 889 put_page(*pp); 890 *pp = page; 891 rqstp->rq_resused++; 892 rqstp->rq_res.page_len += size; 893 } else 894 rqstp->rq_res.page_len += size; 895 896 return size; 897 } 898 899 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 900 struct splice_desc *sd) 901 { 902 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 903 } 904 905 static __be32 906 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 907 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 908 { 909 mm_segment_t oldfs; 910 __be32 err; 911 int host_err; 912 913 err = nfserr_perm; 914 915 if (file->f_op->splice_read && rqstp->rq_splice_ok) { 916 struct splice_desc sd = { 917 .len = 0, 918 .total_len = *count, 919 .pos = offset, 920 .u.data = rqstp, 921 }; 922 923 rqstp->rq_resused = 1; 924 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 925 } else { 926 oldfs = get_fs(); 927 set_fs(KERNEL_DS); 928 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 929 set_fs(oldfs); 930 } 931 932 if (host_err >= 0) { 933 nfsdstats.io_read += host_err; 934 *count = host_err; 935 err = 0; 936 fsnotify_access(file); 937 } else 938 err = nfserrno(host_err); 939 return err; 940 } 941 942 static void kill_suid(struct dentry *dentry) 943 { 944 struct iattr ia; 945 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 946 947 mutex_lock(&dentry->d_inode->i_mutex); 948 notify_change(dentry, &ia); 949 mutex_unlock(&dentry->d_inode->i_mutex); 950 } 951 952 /* 953 * Gathered writes: If another process is currently writing to the file, 954 * there's a high chance this is another nfsd (triggered by a bulk write 955 * from a client's biod). Rather than syncing the file with each write 956 * request, we sleep for 10 msec. 957 * 958 * I don't know if this roughly approximates C. Juszak's idea of 959 * gathered writes, but it's a nice and simple solution (IMHO), and it 960 * seems to work:-) 961 * 962 * Note: we do this only in the NFSv2 case, since v3 and higher have a 963 * better tool (separate unstable writes and commits) for solving this 964 * problem. 965 */ 966 static int wait_for_concurrent_writes(struct file *file) 967 { 968 struct inode *inode = file->f_path.dentry->d_inode; 969 static ino_t last_ino; 970 static dev_t last_dev; 971 int err = 0; 972 973 if (atomic_read(&inode->i_writecount) > 1 974 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 975 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 976 msleep(10); 977 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 978 } 979 980 if (inode->i_state & I_DIRTY) { 981 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 982 err = vfs_fsync(file, 0); 983 } 984 last_ino = inode->i_ino; 985 last_dev = inode->i_sb->s_dev; 986 return err; 987 } 988 989 static __be32 990 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 991 loff_t offset, struct kvec *vec, int vlen, 992 unsigned long *cnt, int *stablep) 993 { 994 struct svc_export *exp; 995 struct dentry *dentry; 996 struct inode *inode; 997 mm_segment_t oldfs; 998 __be32 err = 0; 999 int host_err; 1000 int stable = *stablep; 1001 int use_wgather; 1002 1003 dentry = file->f_path.dentry; 1004 inode = dentry->d_inode; 1005 exp = fhp->fh_export; 1006 1007 /* 1008 * Request sync writes if 1009 * - the sync export option has been set, or 1010 * - the client requested O_SYNC behavior (NFSv3 feature). 1011 * - The file system doesn't support fsync(). 1012 * When NFSv2 gathered writes have been configured for this volume, 1013 * flushing the data to disk is handled separately below. 1014 */ 1015 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 1016 1017 if (!file->f_op->fsync) {/* COMMIT3 cannot work */ 1018 stable = 2; 1019 *stablep = 2; /* FILE_SYNC */ 1020 } 1021 1022 if (!EX_ISSYNC(exp)) 1023 stable = 0; 1024 if (stable && !use_wgather) { 1025 spin_lock(&file->f_lock); 1026 file->f_flags |= O_SYNC; 1027 spin_unlock(&file->f_lock); 1028 } 1029 1030 /* Write the data. */ 1031 oldfs = get_fs(); set_fs(KERNEL_DS); 1032 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); 1033 set_fs(oldfs); 1034 if (host_err < 0) 1035 goto out_nfserr; 1036 *cnt = host_err; 1037 nfsdstats.io_write += host_err; 1038 fsnotify_modify(file); 1039 1040 /* clear setuid/setgid flag after write */ 1041 if (inode->i_mode & (S_ISUID | S_ISGID)) 1042 kill_suid(dentry); 1043 1044 if (stable && use_wgather) 1045 host_err = wait_for_concurrent_writes(file); 1046 1047 out_nfserr: 1048 dprintk("nfsd: write complete host_err=%d\n", host_err); 1049 if (host_err >= 0) 1050 err = 0; 1051 else 1052 err = nfserrno(host_err); 1053 return err; 1054 } 1055 1056 /* 1057 * Read data from a file. count must contain the requested read count 1058 * on entry. On return, *count contains the number of bytes actually read. 1059 * N.B. After this call fhp needs an fh_put 1060 */ 1061 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, 1062 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 1063 { 1064 struct file *file; 1065 struct inode *inode; 1066 struct raparms *ra; 1067 __be32 err; 1068 1069 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); 1070 if (err) 1071 return err; 1072 1073 inode = file->f_path.dentry->d_inode; 1074 1075 /* Get readahead parameters */ 1076 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 1077 1078 if (ra && ra->p_set) 1079 file->f_ra = ra->p_ra; 1080 1081 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1082 1083 /* Write back readahead params */ 1084 if (ra) { 1085 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 1086 spin_lock(&rab->pb_lock); 1087 ra->p_ra = file->f_ra; 1088 ra->p_set = 1; 1089 ra->p_count--; 1090 spin_unlock(&rab->pb_lock); 1091 } 1092 1093 nfsd_close(file); 1094 return err; 1095 } 1096 1097 /* As above, but use the provided file descriptor. */ 1098 __be32 1099 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1100 loff_t offset, struct kvec *vec, int vlen, 1101 unsigned long *count) 1102 { 1103 __be32 err; 1104 1105 if (file) { 1106 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1107 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); 1108 if (err) 1109 goto out; 1110 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1111 } else /* Note file may still be NULL in NFSv4 special stateid case: */ 1112 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count); 1113 out: 1114 return err; 1115 } 1116 1117 /* 1118 * Write data to a file. 1119 * The stable flag requests synchronous writes. 1120 * N.B. After this call fhp needs an fh_put 1121 */ 1122 __be32 1123 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1124 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1125 int *stablep) 1126 { 1127 __be32 err = 0; 1128 1129 if (file) { 1130 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1131 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1132 if (err) 1133 goto out; 1134 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1135 stablep); 1136 } else { 1137 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1138 if (err) 1139 goto out; 1140 1141 if (cnt) 1142 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1143 cnt, stablep); 1144 nfsd_close(file); 1145 } 1146 out: 1147 return err; 1148 } 1149 1150 #ifdef CONFIG_NFSD_V3 1151 /* 1152 * Commit all pending writes to stable storage. 1153 * 1154 * Note: we only guarantee that data that lies within the range specified 1155 * by the 'offset' and 'count' parameters will be synced. 1156 * 1157 * Unfortunately we cannot lock the file to make sure we return full WCC 1158 * data to the client, as locking happens lower down in the filesystem. 1159 */ 1160 __be32 1161 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1162 loff_t offset, unsigned long count) 1163 { 1164 struct file *file; 1165 loff_t end = LLONG_MAX; 1166 __be32 err = nfserr_inval; 1167 1168 if (offset < 0) 1169 goto out; 1170 if (count != 0) { 1171 end = offset + (loff_t)count - 1; 1172 if (end < offset) 1173 goto out; 1174 } 1175 1176 err = nfsd_open(rqstp, fhp, S_IFREG, 1177 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file); 1178 if (err) 1179 goto out; 1180 if (EX_ISSYNC(fhp->fh_export)) { 1181 int err2 = vfs_fsync_range(file, offset, end, 0); 1182 1183 if (err2 != -EINVAL) 1184 err = nfserrno(err2); 1185 else 1186 err = nfserr_notsupp; 1187 } 1188 1189 nfsd_close(file); 1190 out: 1191 return err; 1192 } 1193 #endif /* CONFIG_NFSD_V3 */ 1194 1195 static __be32 1196 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1197 struct iattr *iap) 1198 { 1199 /* 1200 * Mode has already been set earlier in create: 1201 */ 1202 iap->ia_valid &= ~ATTR_MODE; 1203 /* 1204 * Setting uid/gid works only for root. Irix appears to 1205 * send along the gid on create when it tries to implement 1206 * setgid directories via NFS: 1207 */ 1208 if (current_fsuid() != 0) 1209 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1210 if (iap->ia_valid) 1211 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1212 return 0; 1213 } 1214 1215 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1216 * setting size to 0 may fail for some specific file systems by the permission 1217 * checking which requires WRITE permission but the mode is 000. 1218 * we ignore the resizing(to 0) on the just new created file, since the size is 1219 * 0 after file created. 1220 * 1221 * call this only after vfs_create() is called. 1222 * */ 1223 static void 1224 nfsd_check_ignore_resizing(struct iattr *iap) 1225 { 1226 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1227 iap->ia_valid &= ~ATTR_SIZE; 1228 } 1229 1230 /* 1231 * Create a file (regular, directory, device, fifo); UNIX sockets 1232 * not yet implemented. 1233 * If the response fh has been verified, the parent directory should 1234 * already be locked. Note that the parent directory is left locked. 1235 * 1236 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1237 */ 1238 __be32 1239 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1240 char *fname, int flen, struct iattr *iap, 1241 int type, dev_t rdev, struct svc_fh *resfhp) 1242 { 1243 struct dentry *dentry, *dchild = NULL; 1244 struct inode *dirp; 1245 __be32 err; 1246 __be32 err2; 1247 int host_err; 1248 1249 err = nfserr_perm; 1250 if (!flen) 1251 goto out; 1252 err = nfserr_exist; 1253 if (isdotent(fname, flen)) 1254 goto out; 1255 1256 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1257 if (err) 1258 goto out; 1259 1260 dentry = fhp->fh_dentry; 1261 dirp = dentry->d_inode; 1262 1263 err = nfserr_notdir; 1264 if (!dirp->i_op->lookup) 1265 goto out; 1266 /* 1267 * Check whether the response file handle has been verified yet. 1268 * If it has, the parent directory should already be locked. 1269 */ 1270 if (!resfhp->fh_dentry) { 1271 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1272 fh_lock_nested(fhp, I_MUTEX_PARENT); 1273 dchild = lookup_one_len(fname, dentry, flen); 1274 host_err = PTR_ERR(dchild); 1275 if (IS_ERR(dchild)) 1276 goto out_nfserr; 1277 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1278 if (err) 1279 goto out; 1280 } else { 1281 /* called from nfsd_proc_create */ 1282 dchild = dget(resfhp->fh_dentry); 1283 if (!fhp->fh_locked) { 1284 /* not actually possible */ 1285 printk(KERN_ERR 1286 "nfsd_create: parent %s/%s not locked!\n", 1287 dentry->d_parent->d_name.name, 1288 dentry->d_name.name); 1289 err = nfserr_io; 1290 goto out; 1291 } 1292 } 1293 /* 1294 * Make sure the child dentry is still negative ... 1295 */ 1296 err = nfserr_exist; 1297 if (dchild->d_inode) { 1298 dprintk("nfsd_create: dentry %s/%s not negative!\n", 1299 dentry->d_name.name, dchild->d_name.name); 1300 goto out; 1301 } 1302 1303 if (!(iap->ia_valid & ATTR_MODE)) 1304 iap->ia_mode = 0; 1305 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1306 1307 err = nfserr_inval; 1308 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1309 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1310 type); 1311 goto out; 1312 } 1313 1314 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1315 if (host_err) 1316 goto out_nfserr; 1317 1318 /* 1319 * Get the dir op function pointer. 1320 */ 1321 err = 0; 1322 switch (type) { 1323 case S_IFREG: 1324 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1325 if (!host_err) 1326 nfsd_check_ignore_resizing(iap); 1327 break; 1328 case S_IFDIR: 1329 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1330 break; 1331 case S_IFCHR: 1332 case S_IFBLK: 1333 case S_IFIFO: 1334 case S_IFSOCK: 1335 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1336 break; 1337 } 1338 if (host_err < 0) { 1339 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1340 goto out_nfserr; 1341 } 1342 1343 err = nfsd_create_setattr(rqstp, resfhp, iap); 1344 1345 /* 1346 * nfsd_setattr already committed the child. Transactional filesystems 1347 * had a chance to commit changes for both parent and child 1348 * simultaneously making the following commit_metadata a noop. 1349 */ 1350 err2 = nfserrno(commit_metadata(fhp)); 1351 if (err2) 1352 err = err2; 1353 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1354 /* 1355 * Update the file handle to get the new inode info. 1356 */ 1357 if (!err) 1358 err = fh_update(resfhp); 1359 out: 1360 if (dchild && !IS_ERR(dchild)) 1361 dput(dchild); 1362 return err; 1363 1364 out_nfserr: 1365 err = nfserrno(host_err); 1366 goto out; 1367 } 1368 1369 #ifdef CONFIG_NFSD_V3 1370 1371 static inline int nfsd_create_is_exclusive(int createmode) 1372 { 1373 return createmode == NFS3_CREATE_EXCLUSIVE 1374 || createmode == NFS4_CREATE_EXCLUSIVE4_1; 1375 } 1376 1377 /* 1378 * NFSv3 and NFSv4 version of nfsd_create 1379 */ 1380 __be32 1381 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1382 char *fname, int flen, struct iattr *iap, 1383 struct svc_fh *resfhp, int createmode, u32 *verifier, 1384 bool *truncp, bool *created) 1385 { 1386 struct dentry *dentry, *dchild = NULL; 1387 struct inode *dirp; 1388 __be32 err; 1389 int host_err; 1390 __u32 v_mtime=0, v_atime=0; 1391 1392 err = nfserr_perm; 1393 if (!flen) 1394 goto out; 1395 err = nfserr_exist; 1396 if (isdotent(fname, flen)) 1397 goto out; 1398 if (!(iap->ia_valid & ATTR_MODE)) 1399 iap->ia_mode = 0; 1400 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 1401 if (err) 1402 goto out; 1403 1404 dentry = fhp->fh_dentry; 1405 dirp = dentry->d_inode; 1406 1407 /* Get all the sanity checks out of the way before 1408 * we lock the parent. */ 1409 err = nfserr_notdir; 1410 if (!dirp->i_op->lookup) 1411 goto out; 1412 fh_lock_nested(fhp, I_MUTEX_PARENT); 1413 1414 /* 1415 * Compose the response file handle. 1416 */ 1417 dchild = lookup_one_len(fname, dentry, flen); 1418 host_err = PTR_ERR(dchild); 1419 if (IS_ERR(dchild)) 1420 goto out_nfserr; 1421 1422 /* If file doesn't exist, check for permissions to create one */ 1423 if (!dchild->d_inode) { 1424 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1425 if (err) 1426 goto out; 1427 } 1428 1429 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1430 if (err) 1431 goto out; 1432 1433 if (nfsd_create_is_exclusive(createmode)) { 1434 /* solaris7 gets confused (bugid 4218508) if these have 1435 * the high bit set, so just clear the high bits. If this is 1436 * ever changed to use different attrs for storing the 1437 * verifier, then do_open_lookup() will also need to be fixed 1438 * accordingly. 1439 */ 1440 v_mtime = verifier[0]&0x7fffffff; 1441 v_atime = verifier[1]&0x7fffffff; 1442 } 1443 1444 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1445 if (host_err) 1446 goto out_nfserr; 1447 if (dchild->d_inode) { 1448 err = 0; 1449 1450 switch (createmode) { 1451 case NFS3_CREATE_UNCHECKED: 1452 if (! S_ISREG(dchild->d_inode->i_mode)) 1453 err = nfserr_exist; 1454 else if (truncp) { 1455 /* in nfsv4, we need to treat this case a little 1456 * differently. we don't want to truncate the 1457 * file now; this would be wrong if the OPEN 1458 * fails for some other reason. furthermore, 1459 * if the size is nonzero, we should ignore it 1460 * according to spec! 1461 */ 1462 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1463 } 1464 else { 1465 iap->ia_valid &= ATTR_SIZE; 1466 goto set_attr; 1467 } 1468 break; 1469 case NFS3_CREATE_EXCLUSIVE: 1470 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1471 && dchild->d_inode->i_atime.tv_sec == v_atime 1472 && dchild->d_inode->i_size == 0 ) 1473 break; 1474 case NFS4_CREATE_EXCLUSIVE4_1: 1475 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1476 && dchild->d_inode->i_atime.tv_sec == v_atime 1477 && dchild->d_inode->i_size == 0 ) 1478 goto set_attr; 1479 /* fallthru */ 1480 case NFS3_CREATE_GUARDED: 1481 err = nfserr_exist; 1482 } 1483 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1484 goto out; 1485 } 1486 1487 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1488 if (host_err < 0) { 1489 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1490 goto out_nfserr; 1491 } 1492 if (created) 1493 *created = 1; 1494 1495 nfsd_check_ignore_resizing(iap); 1496 1497 if (nfsd_create_is_exclusive(createmode)) { 1498 /* Cram the verifier into atime/mtime */ 1499 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1500 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1501 /* XXX someone who knows this better please fix it for nsec */ 1502 iap->ia_mtime.tv_sec = v_mtime; 1503 iap->ia_atime.tv_sec = v_atime; 1504 iap->ia_mtime.tv_nsec = 0; 1505 iap->ia_atime.tv_nsec = 0; 1506 } 1507 1508 set_attr: 1509 err = nfsd_create_setattr(rqstp, resfhp, iap); 1510 1511 /* 1512 * nfsd_setattr already committed the child (and possibly also the parent). 1513 */ 1514 if (!err) 1515 err = nfserrno(commit_metadata(fhp)); 1516 1517 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1518 /* 1519 * Update the filehandle to get the new inode info. 1520 */ 1521 if (!err) 1522 err = fh_update(resfhp); 1523 1524 out: 1525 fh_unlock(fhp); 1526 if (dchild && !IS_ERR(dchild)) 1527 dput(dchild); 1528 return err; 1529 1530 out_nfserr: 1531 err = nfserrno(host_err); 1532 goto out; 1533 } 1534 #endif /* CONFIG_NFSD_V3 */ 1535 1536 /* 1537 * Read a symlink. On entry, *lenp must contain the maximum path length that 1538 * fits into the buffer. On return, it contains the true length. 1539 * N.B. After this call fhp needs an fh_put 1540 */ 1541 __be32 1542 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1543 { 1544 struct dentry *dentry; 1545 struct inode *inode; 1546 mm_segment_t oldfs; 1547 __be32 err; 1548 int host_err; 1549 1550 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1551 if (err) 1552 goto out; 1553 1554 dentry = fhp->fh_dentry; 1555 inode = dentry->d_inode; 1556 1557 err = nfserr_inval; 1558 if (!inode->i_op->readlink) 1559 goto out; 1560 1561 touch_atime(fhp->fh_export->ex_path.mnt, dentry); 1562 /* N.B. Why does this call need a get_fs()?? 1563 * Remove the set_fs and watch the fireworks:-) --okir 1564 */ 1565 1566 oldfs = get_fs(); set_fs(KERNEL_DS); 1567 host_err = inode->i_op->readlink(dentry, buf, *lenp); 1568 set_fs(oldfs); 1569 1570 if (host_err < 0) 1571 goto out_nfserr; 1572 *lenp = host_err; 1573 err = 0; 1574 out: 1575 return err; 1576 1577 out_nfserr: 1578 err = nfserrno(host_err); 1579 goto out; 1580 } 1581 1582 /* 1583 * Create a symlink and look up its inode 1584 * N.B. After this call _both_ fhp and resfhp need an fh_put 1585 */ 1586 __be32 1587 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1588 char *fname, int flen, 1589 char *path, int plen, 1590 struct svc_fh *resfhp, 1591 struct iattr *iap) 1592 { 1593 struct dentry *dentry, *dnew; 1594 __be32 err, cerr; 1595 int host_err; 1596 1597 err = nfserr_noent; 1598 if (!flen || !plen) 1599 goto out; 1600 err = nfserr_exist; 1601 if (isdotent(fname, flen)) 1602 goto out; 1603 1604 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1605 if (err) 1606 goto out; 1607 fh_lock(fhp); 1608 dentry = fhp->fh_dentry; 1609 dnew = lookup_one_len(fname, dentry, flen); 1610 host_err = PTR_ERR(dnew); 1611 if (IS_ERR(dnew)) 1612 goto out_nfserr; 1613 1614 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1615 if (host_err) 1616 goto out_nfserr; 1617 1618 if (unlikely(path[plen] != 0)) { 1619 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1620 if (path_alloced == NULL) 1621 host_err = -ENOMEM; 1622 else { 1623 strncpy(path_alloced, path, plen); 1624 path_alloced[plen] = 0; 1625 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); 1626 kfree(path_alloced); 1627 } 1628 } else 1629 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1630 err = nfserrno(host_err); 1631 if (!err) 1632 err = nfserrno(commit_metadata(fhp)); 1633 fh_unlock(fhp); 1634 1635 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1636 1637 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1638 dput(dnew); 1639 if (err==0) err = cerr; 1640 out: 1641 return err; 1642 1643 out_nfserr: 1644 err = nfserrno(host_err); 1645 goto out; 1646 } 1647 1648 /* 1649 * Create a hardlink 1650 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1651 */ 1652 __be32 1653 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1654 char *name, int len, struct svc_fh *tfhp) 1655 { 1656 struct dentry *ddir, *dnew, *dold; 1657 struct inode *dirp; 1658 __be32 err; 1659 int host_err; 1660 1661 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1662 if (err) 1663 goto out; 1664 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP); 1665 if (err) 1666 goto out; 1667 err = nfserr_isdir; 1668 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode)) 1669 goto out; 1670 err = nfserr_perm; 1671 if (!len) 1672 goto out; 1673 err = nfserr_exist; 1674 if (isdotent(name, len)) 1675 goto out; 1676 1677 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1678 ddir = ffhp->fh_dentry; 1679 dirp = ddir->d_inode; 1680 1681 dnew = lookup_one_len(name, ddir, len); 1682 host_err = PTR_ERR(dnew); 1683 if (IS_ERR(dnew)) 1684 goto out_nfserr; 1685 1686 dold = tfhp->fh_dentry; 1687 1688 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt); 1689 if (host_err) { 1690 err = nfserrno(host_err); 1691 goto out_dput; 1692 } 1693 err = nfserr_noent; 1694 if (!dold->d_inode) 1695 goto out_drop_write; 1696 host_err = nfsd_break_lease(dold->d_inode); 1697 if (host_err) { 1698 err = nfserrno(host_err); 1699 goto out_drop_write; 1700 } 1701 host_err = vfs_link(dold, dirp, dnew); 1702 if (!host_err) { 1703 err = nfserrno(commit_metadata(ffhp)); 1704 if (!err) 1705 err = nfserrno(commit_metadata(tfhp)); 1706 } else { 1707 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1708 err = nfserr_acces; 1709 else 1710 err = nfserrno(host_err); 1711 } 1712 out_drop_write: 1713 mnt_drop_write(tfhp->fh_export->ex_path.mnt); 1714 out_dput: 1715 dput(dnew); 1716 out_unlock: 1717 fh_unlock(ffhp); 1718 out: 1719 return err; 1720 1721 out_nfserr: 1722 err = nfserrno(host_err); 1723 goto out_unlock; 1724 } 1725 1726 /* 1727 * Rename a file 1728 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1729 */ 1730 __be32 1731 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1732 struct svc_fh *tfhp, char *tname, int tlen) 1733 { 1734 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1735 struct inode *fdir, *tdir; 1736 __be32 err; 1737 int host_err; 1738 1739 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1740 if (err) 1741 goto out; 1742 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1743 if (err) 1744 goto out; 1745 1746 fdentry = ffhp->fh_dentry; 1747 fdir = fdentry->d_inode; 1748 1749 tdentry = tfhp->fh_dentry; 1750 tdir = tdentry->d_inode; 1751 1752 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; 1753 if (ffhp->fh_export != tfhp->fh_export) 1754 goto out; 1755 1756 err = nfserr_perm; 1757 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1758 goto out; 1759 1760 /* cannot use fh_lock as we need deadlock protective ordering 1761 * so do it by hand */ 1762 trap = lock_rename(tdentry, fdentry); 1763 ffhp->fh_locked = tfhp->fh_locked = 1; 1764 fill_pre_wcc(ffhp); 1765 fill_pre_wcc(tfhp); 1766 1767 odentry = lookup_one_len(fname, fdentry, flen); 1768 host_err = PTR_ERR(odentry); 1769 if (IS_ERR(odentry)) 1770 goto out_nfserr; 1771 1772 host_err = -ENOENT; 1773 if (!odentry->d_inode) 1774 goto out_dput_old; 1775 host_err = -EINVAL; 1776 if (odentry == trap) 1777 goto out_dput_old; 1778 1779 ndentry = lookup_one_len(tname, tdentry, tlen); 1780 host_err = PTR_ERR(ndentry); 1781 if (IS_ERR(ndentry)) 1782 goto out_dput_old; 1783 host_err = -ENOTEMPTY; 1784 if (ndentry == trap) 1785 goto out_dput_new; 1786 1787 host_err = -EXDEV; 1788 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1789 goto out_dput_new; 1790 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt); 1791 if (host_err) 1792 goto out_dput_new; 1793 1794 host_err = nfsd_break_lease(odentry->d_inode); 1795 if (host_err) 1796 goto out_drop_write; 1797 if (ndentry->d_inode) { 1798 host_err = nfsd_break_lease(ndentry->d_inode); 1799 if (host_err) 1800 goto out_drop_write; 1801 } 1802 host_err = vfs_rename(fdir, odentry, tdir, ndentry); 1803 if (!host_err) { 1804 host_err = commit_metadata(tfhp); 1805 if (!host_err) 1806 host_err = commit_metadata(ffhp); 1807 } 1808 out_drop_write: 1809 mnt_drop_write(ffhp->fh_export->ex_path.mnt); 1810 out_dput_new: 1811 dput(ndentry); 1812 out_dput_old: 1813 dput(odentry); 1814 out_nfserr: 1815 err = nfserrno(host_err); 1816 1817 /* we cannot reply on fh_unlock on the two filehandles, 1818 * as that would do the wrong thing if the two directories 1819 * were the same, so again we do it by hand 1820 */ 1821 fill_post_wcc(ffhp); 1822 fill_post_wcc(tfhp); 1823 unlock_rename(tdentry, fdentry); 1824 ffhp->fh_locked = tfhp->fh_locked = 0; 1825 1826 out: 1827 return err; 1828 } 1829 1830 /* 1831 * Unlink a file or directory 1832 * N.B. After this call fhp needs an fh_put 1833 */ 1834 __be32 1835 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1836 char *fname, int flen) 1837 { 1838 struct dentry *dentry, *rdentry; 1839 struct inode *dirp; 1840 __be32 err; 1841 int host_err; 1842 1843 err = nfserr_acces; 1844 if (!flen || isdotent(fname, flen)) 1845 goto out; 1846 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1847 if (err) 1848 goto out; 1849 1850 fh_lock_nested(fhp, I_MUTEX_PARENT); 1851 dentry = fhp->fh_dentry; 1852 dirp = dentry->d_inode; 1853 1854 rdentry = lookup_one_len(fname, dentry, flen); 1855 host_err = PTR_ERR(rdentry); 1856 if (IS_ERR(rdentry)) 1857 goto out_nfserr; 1858 1859 if (!rdentry->d_inode) { 1860 dput(rdentry); 1861 err = nfserr_noent; 1862 goto out; 1863 } 1864 1865 if (!type) 1866 type = rdentry->d_inode->i_mode & S_IFMT; 1867 1868 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1869 if (host_err) 1870 goto out_put; 1871 1872 host_err = nfsd_break_lease(rdentry->d_inode); 1873 if (host_err) 1874 goto out_drop_write; 1875 if (type != S_IFDIR) 1876 host_err = vfs_unlink(dirp, rdentry); 1877 else 1878 host_err = vfs_rmdir(dirp, rdentry); 1879 if (!host_err) 1880 host_err = commit_metadata(fhp); 1881 out_drop_write: 1882 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1883 out_put: 1884 dput(rdentry); 1885 1886 out_nfserr: 1887 err = nfserrno(host_err); 1888 out: 1889 return err; 1890 } 1891 1892 /* 1893 * We do this buffering because we must not call back into the file 1894 * system's ->lookup() method from the filldir callback. That may well 1895 * deadlock a number of file systems. 1896 * 1897 * This is based heavily on the implementation of same in XFS. 1898 */ 1899 struct buffered_dirent { 1900 u64 ino; 1901 loff_t offset; 1902 int namlen; 1903 unsigned int d_type; 1904 char name[]; 1905 }; 1906 1907 struct readdir_data { 1908 char *dirent; 1909 size_t used; 1910 int full; 1911 }; 1912 1913 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1914 loff_t offset, u64 ino, unsigned int d_type) 1915 { 1916 struct readdir_data *buf = __buf; 1917 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1918 unsigned int reclen; 1919 1920 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1921 if (buf->used + reclen > PAGE_SIZE) { 1922 buf->full = 1; 1923 return -EINVAL; 1924 } 1925 1926 de->namlen = namlen; 1927 de->offset = offset; 1928 de->ino = ino; 1929 de->d_type = d_type; 1930 memcpy(de->name, name, namlen); 1931 buf->used += reclen; 1932 1933 return 0; 1934 } 1935 1936 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1937 struct readdir_cd *cdp, loff_t *offsetp) 1938 { 1939 struct readdir_data buf; 1940 struct buffered_dirent *de; 1941 int host_err; 1942 int size; 1943 loff_t offset; 1944 1945 buf.dirent = (void *)__get_free_page(GFP_KERNEL); 1946 if (!buf.dirent) 1947 return nfserrno(-ENOMEM); 1948 1949 offset = *offsetp; 1950 1951 while (1) { 1952 struct inode *dir_inode = file->f_path.dentry->d_inode; 1953 unsigned int reclen; 1954 1955 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1956 buf.used = 0; 1957 buf.full = 0; 1958 1959 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf); 1960 if (buf.full) 1961 host_err = 0; 1962 1963 if (host_err < 0) 1964 break; 1965 1966 size = buf.used; 1967 1968 if (!size) 1969 break; 1970 1971 /* 1972 * Various filldir functions may end up calling back into 1973 * lookup_one_len() and the file system's ->lookup() method. 1974 * These expect i_mutex to be held, as it would within readdir. 1975 */ 1976 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1977 if (host_err) 1978 break; 1979 1980 de = (struct buffered_dirent *)buf.dirent; 1981 while (size > 0) { 1982 offset = de->offset; 1983 1984 if (func(cdp, de->name, de->namlen, de->offset, 1985 de->ino, de->d_type)) 1986 break; 1987 1988 if (cdp->err != nfs_ok) 1989 break; 1990 1991 reclen = ALIGN(sizeof(*de) + de->namlen, 1992 sizeof(u64)); 1993 size -= reclen; 1994 de = (struct buffered_dirent *)((char *)de + reclen); 1995 } 1996 mutex_unlock(&dir_inode->i_mutex); 1997 if (size > 0) /* We bailed out early */ 1998 break; 1999 2000 offset = vfs_llseek(file, 0, SEEK_CUR); 2001 } 2002 2003 free_page((unsigned long)(buf.dirent)); 2004 2005 if (host_err) 2006 return nfserrno(host_err); 2007 2008 *offsetp = offset; 2009 return cdp->err; 2010 } 2011 2012 /* 2013 * Read entries from a directory. 2014 * The NFSv3/4 verifier we ignore for now. 2015 */ 2016 __be32 2017 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 2018 struct readdir_cd *cdp, filldir_t func) 2019 { 2020 __be32 err; 2021 struct file *file; 2022 loff_t offset = *offsetp; 2023 2024 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file); 2025 if (err) 2026 goto out; 2027 2028 offset = vfs_llseek(file, offset, 0); 2029 if (offset < 0) { 2030 err = nfserrno((int)offset); 2031 goto out_close; 2032 } 2033 2034 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 2035 2036 if (err == nfserr_eof || err == nfserr_toosmall) 2037 err = nfs_ok; /* can still be found in ->err */ 2038 out_close: 2039 nfsd_close(file); 2040 out: 2041 return err; 2042 } 2043 2044 /* 2045 * Get file system stats 2046 * N.B. After this call fhp needs an fh_put 2047 */ 2048 __be32 2049 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 2050 { 2051 __be32 err; 2052 2053 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 2054 if (!err) { 2055 struct path path = { 2056 .mnt = fhp->fh_export->ex_path.mnt, 2057 .dentry = fhp->fh_dentry, 2058 }; 2059 if (vfs_statfs(&path, stat)) 2060 err = nfserr_io; 2061 } 2062 return err; 2063 } 2064 2065 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 2066 { 2067 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 2068 } 2069 2070 /* 2071 * Check for a user's access permissions to this inode. 2072 */ 2073 __be32 2074 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 2075 struct dentry *dentry, int acc) 2076 { 2077 struct inode *inode = dentry->d_inode; 2078 int err; 2079 2080 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP) 2081 return 0; 2082 #if 0 2083 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 2084 acc, 2085 (acc & NFSD_MAY_READ)? " read" : "", 2086 (acc & NFSD_MAY_WRITE)? " write" : "", 2087 (acc & NFSD_MAY_EXEC)? " exec" : "", 2088 (acc & NFSD_MAY_SATTR)? " sattr" : "", 2089 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 2090 (acc & NFSD_MAY_LOCK)? " lock" : "", 2091 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 2092 inode->i_mode, 2093 IS_IMMUTABLE(inode)? " immut" : "", 2094 IS_APPEND(inode)? " append" : "", 2095 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2096 dprintk(" owner %d/%d user %d/%d\n", 2097 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2098 #endif 2099 2100 /* Normally we reject any write/sattr etc access on a read-only file 2101 * system. But if it is IRIX doing check on write-access for a 2102 * device special file, we ignore rofs. 2103 */ 2104 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2105 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2106 if (exp_rdonly(rqstp, exp) || 2107 __mnt_is_readonly(exp->ex_path.mnt)) 2108 return nfserr_rofs; 2109 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2110 return nfserr_perm; 2111 } 2112 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2113 return nfserr_perm; 2114 2115 if (acc & NFSD_MAY_LOCK) { 2116 /* If we cannot rely on authentication in NLM requests, 2117 * just allow locks, otherwise require read permission, or 2118 * ownership 2119 */ 2120 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2121 return 0; 2122 else 2123 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2124 } 2125 /* 2126 * The file owner always gets access permission for accesses that 2127 * would normally be checked at open time. This is to make 2128 * file access work even when the client has done a fchmod(fd, 0). 2129 * 2130 * However, `cp foo bar' should fail nevertheless when bar is 2131 * readonly. A sensible way to do this might be to reject all 2132 * attempts to truncate a read-only file, because a creat() call 2133 * always implies file truncation. 2134 * ... but this isn't really fair. A process may reasonably call 2135 * ftruncate on an open file descriptor on a file with perm 000. 2136 * We must trust the client to do permission checking - using "ACCESS" 2137 * with NFSv3. 2138 */ 2139 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2140 inode->i_uid == current_fsuid()) 2141 return 0; 2142 2143 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2144 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2145 2146 /* Allow read access to binaries even when mode 111 */ 2147 if (err == -EACCES && S_ISREG(inode->i_mode) && 2148 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) || 2149 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC))) 2150 err = inode_permission(inode, MAY_EXEC); 2151 2152 return err? nfserrno(err) : 0; 2153 } 2154 2155 void 2156 nfsd_racache_shutdown(void) 2157 { 2158 struct raparms *raparm, *last_raparm; 2159 unsigned int i; 2160 2161 dprintk("nfsd: freeing readahead buffers.\n"); 2162 2163 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2164 raparm = raparm_hash[i].pb_head; 2165 while(raparm) { 2166 last_raparm = raparm; 2167 raparm = raparm->p_next; 2168 kfree(last_raparm); 2169 } 2170 raparm_hash[i].pb_head = NULL; 2171 } 2172 } 2173 /* 2174 * Initialize readahead param cache 2175 */ 2176 int 2177 nfsd_racache_init(int cache_size) 2178 { 2179 int i; 2180 int j = 0; 2181 int nperbucket; 2182 struct raparms **raparm = NULL; 2183 2184 2185 if (raparm_hash[0].pb_head) 2186 return 0; 2187 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2188 if (nperbucket < 2) 2189 nperbucket = 2; 2190 cache_size = nperbucket * RAPARM_HASH_SIZE; 2191 2192 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2193 2194 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2195 spin_lock_init(&raparm_hash[i].pb_lock); 2196 2197 raparm = &raparm_hash[i].pb_head; 2198 for (j = 0; j < nperbucket; j++) { 2199 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2200 if (!*raparm) 2201 goto out_nomem; 2202 raparm = &(*raparm)->p_next; 2203 } 2204 *raparm = NULL; 2205 } 2206 2207 nfsdstats.ra_size = cache_size; 2208 return 0; 2209 2210 out_nomem: 2211 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2212 nfsd_racache_shutdown(); 2213 return -ENOMEM; 2214 } 2215 2216 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 2217 struct posix_acl * 2218 nfsd_get_posix_acl(struct svc_fh *fhp, int type) 2219 { 2220 struct inode *inode = fhp->fh_dentry->d_inode; 2221 char *name; 2222 void *value = NULL; 2223 ssize_t size; 2224 struct posix_acl *acl; 2225 2226 if (!IS_POSIXACL(inode)) 2227 return ERR_PTR(-EOPNOTSUPP); 2228 2229 switch (type) { 2230 case ACL_TYPE_ACCESS: 2231 name = POSIX_ACL_XATTR_ACCESS; 2232 break; 2233 case ACL_TYPE_DEFAULT: 2234 name = POSIX_ACL_XATTR_DEFAULT; 2235 break; 2236 default: 2237 return ERR_PTR(-EOPNOTSUPP); 2238 } 2239 2240 size = nfsd_getxattr(fhp->fh_dentry, name, &value); 2241 if (size < 0) 2242 return ERR_PTR(size); 2243 2244 acl = posix_acl_from_xattr(value, size); 2245 kfree(value); 2246 return acl; 2247 } 2248 2249 int 2250 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) 2251 { 2252 struct inode *inode = fhp->fh_dentry->d_inode; 2253 char *name; 2254 void *value = NULL; 2255 size_t size; 2256 int error; 2257 2258 if (!IS_POSIXACL(inode) || 2259 !inode->i_op->setxattr || !inode->i_op->removexattr) 2260 return -EOPNOTSUPP; 2261 switch(type) { 2262 case ACL_TYPE_ACCESS: 2263 name = POSIX_ACL_XATTR_ACCESS; 2264 break; 2265 case ACL_TYPE_DEFAULT: 2266 name = POSIX_ACL_XATTR_DEFAULT; 2267 break; 2268 default: 2269 return -EOPNOTSUPP; 2270 } 2271 2272 if (acl && acl->a_count) { 2273 size = posix_acl_xattr_size(acl->a_count); 2274 value = kmalloc(size, GFP_KERNEL); 2275 if (!value) 2276 return -ENOMEM; 2277 error = posix_acl_to_xattr(acl, value, size); 2278 if (error < 0) 2279 goto getout; 2280 size = error; 2281 } else 2282 size = 0; 2283 2284 error = mnt_want_write(fhp->fh_export->ex_path.mnt); 2285 if (error) 2286 goto getout; 2287 if (size) 2288 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); 2289 else { 2290 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) 2291 error = 0; 2292 else { 2293 error = vfs_removexattr(fhp->fh_dentry, name); 2294 if (error == -ENODATA) 2295 error = 0; 2296 } 2297 } 2298 mnt_drop_write(fhp->fh_export->ex_path.mnt); 2299 2300 getout: 2301 kfree(value); 2302 return error; 2303 } 2304 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 2305