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 umode_t 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 may_flags argument indicates the type of open (read/write/lock) 741 * and additional flags. 742 * N.B. After this call fhp needs an fh_put 743 */ 744 __be32 745 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, 746 int may_flags, struct file **filp) 747 { 748 struct dentry *dentry; 749 struct inode *inode; 750 int flags = O_RDONLY|O_LARGEFILE; 751 __be32 err; 752 int host_err = 0; 753 754 validate_process_creds(); 755 756 /* 757 * If we get here, then the client has already done an "open", 758 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 759 * in case a chmod has now revoked permission. 760 */ 761 err = fh_verify(rqstp, fhp, type, may_flags | NFSD_MAY_OWNER_OVERRIDE); 762 if (err) 763 goto out; 764 765 dentry = fhp->fh_dentry; 766 inode = dentry->d_inode; 767 768 /* Disallow write access to files with the append-only bit set 769 * or any access when mandatory locking enabled 770 */ 771 err = nfserr_perm; 772 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE)) 773 goto out; 774 /* 775 * We must ignore files (but only files) which might have mandatory 776 * locks on them because there is no way to know if the accesser has 777 * the lock. 778 */ 779 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) 780 goto out; 781 782 if (!inode->i_fop) 783 goto out; 784 785 host_err = nfsd_open_break_lease(inode, may_flags); 786 if (host_err) /* NOMEM or WOULDBLOCK */ 787 goto out_nfserr; 788 789 if (may_flags & NFSD_MAY_WRITE) { 790 if (may_flags & NFSD_MAY_READ) 791 flags = O_RDWR|O_LARGEFILE; 792 else 793 flags = O_WRONLY|O_LARGEFILE; 794 } 795 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt), 796 flags, current_cred()); 797 if (IS_ERR(*filp)) 798 host_err = PTR_ERR(*filp); 799 else 800 host_err = ima_file_check(*filp, may_flags); 801 802 out_nfserr: 803 err = nfserrno(host_err); 804 out: 805 validate_process_creds(); 806 return err; 807 } 808 809 /* 810 * Close a file. 811 */ 812 void 813 nfsd_close(struct file *filp) 814 { 815 fput(filp); 816 } 817 818 /* 819 * Obtain the readahead parameters for the file 820 * specified by (dev, ino). 821 */ 822 823 static inline struct raparms * 824 nfsd_get_raparms(dev_t dev, ino_t ino) 825 { 826 struct raparms *ra, **rap, **frap = NULL; 827 int depth = 0; 828 unsigned int hash; 829 struct raparm_hbucket *rab; 830 831 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 832 rab = &raparm_hash[hash]; 833 834 spin_lock(&rab->pb_lock); 835 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 836 if (ra->p_ino == ino && ra->p_dev == dev) 837 goto found; 838 depth++; 839 if (ra->p_count == 0) 840 frap = rap; 841 } 842 depth = nfsdstats.ra_size; 843 if (!frap) { 844 spin_unlock(&rab->pb_lock); 845 return NULL; 846 } 847 rap = frap; 848 ra = *frap; 849 ra->p_dev = dev; 850 ra->p_ino = ino; 851 ra->p_set = 0; 852 ra->p_hindex = hash; 853 found: 854 if (rap != &rab->pb_head) { 855 *rap = ra->p_next; 856 ra->p_next = rab->pb_head; 857 rab->pb_head = ra; 858 } 859 ra->p_count++; 860 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 861 spin_unlock(&rab->pb_lock); 862 return ra; 863 } 864 865 /* 866 * Grab and keep cached pages associated with a file in the svc_rqst 867 * so that they can be passed to the network sendmsg/sendpage routines 868 * directly. They will be released after the sending has completed. 869 */ 870 static int 871 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 872 struct splice_desc *sd) 873 { 874 struct svc_rqst *rqstp = sd->u.data; 875 struct page **pp = rqstp->rq_respages + rqstp->rq_resused; 876 struct page *page = buf->page; 877 size_t size; 878 879 size = sd->len; 880 881 if (rqstp->rq_res.page_len == 0) { 882 get_page(page); 883 put_page(*pp); 884 *pp = page; 885 rqstp->rq_resused++; 886 rqstp->rq_res.page_base = buf->offset; 887 rqstp->rq_res.page_len = size; 888 } else if (page != pp[-1]) { 889 get_page(page); 890 if (*pp) 891 put_page(*pp); 892 *pp = page; 893 rqstp->rq_resused++; 894 rqstp->rq_res.page_len += size; 895 } else 896 rqstp->rq_res.page_len += size; 897 898 return size; 899 } 900 901 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 902 struct splice_desc *sd) 903 { 904 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 905 } 906 907 static __be32 908 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 909 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 910 { 911 mm_segment_t oldfs; 912 __be32 err; 913 int host_err; 914 915 err = nfserr_perm; 916 917 if (file->f_op->splice_read && rqstp->rq_splice_ok) { 918 struct splice_desc sd = { 919 .len = 0, 920 .total_len = *count, 921 .pos = offset, 922 .u.data = rqstp, 923 }; 924 925 rqstp->rq_resused = 1; 926 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 927 } else { 928 oldfs = get_fs(); 929 set_fs(KERNEL_DS); 930 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 931 set_fs(oldfs); 932 } 933 934 if (host_err >= 0) { 935 nfsdstats.io_read += host_err; 936 *count = host_err; 937 err = 0; 938 fsnotify_access(file); 939 } else 940 err = nfserrno(host_err); 941 return err; 942 } 943 944 static void kill_suid(struct dentry *dentry) 945 { 946 struct iattr ia; 947 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 948 949 mutex_lock(&dentry->d_inode->i_mutex); 950 notify_change(dentry, &ia); 951 mutex_unlock(&dentry->d_inode->i_mutex); 952 } 953 954 /* 955 * Gathered writes: If another process is currently writing to the file, 956 * there's a high chance this is another nfsd (triggered by a bulk write 957 * from a client's biod). Rather than syncing the file with each write 958 * request, we sleep for 10 msec. 959 * 960 * I don't know if this roughly approximates C. Juszak's idea of 961 * gathered writes, but it's a nice and simple solution (IMHO), and it 962 * seems to work:-) 963 * 964 * Note: we do this only in the NFSv2 case, since v3 and higher have a 965 * better tool (separate unstable writes and commits) for solving this 966 * problem. 967 */ 968 static int wait_for_concurrent_writes(struct file *file) 969 { 970 struct inode *inode = file->f_path.dentry->d_inode; 971 static ino_t last_ino; 972 static dev_t last_dev; 973 int err = 0; 974 975 if (atomic_read(&inode->i_writecount) > 1 976 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 977 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 978 msleep(10); 979 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 980 } 981 982 if (inode->i_state & I_DIRTY) { 983 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 984 err = vfs_fsync(file, 0); 985 } 986 last_ino = inode->i_ino; 987 last_dev = inode->i_sb->s_dev; 988 return err; 989 } 990 991 static __be32 992 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 993 loff_t offset, struct kvec *vec, int vlen, 994 unsigned long *cnt, int *stablep) 995 { 996 struct svc_export *exp; 997 struct dentry *dentry; 998 struct inode *inode; 999 mm_segment_t oldfs; 1000 __be32 err = 0; 1001 int host_err; 1002 int stable = *stablep; 1003 int use_wgather; 1004 1005 dentry = file->f_path.dentry; 1006 inode = dentry->d_inode; 1007 exp = fhp->fh_export; 1008 1009 /* 1010 * Request sync writes if 1011 * - the sync export option has been set, or 1012 * - the client requested O_SYNC behavior (NFSv3 feature). 1013 * - The file system doesn't support fsync(). 1014 * When NFSv2 gathered writes have been configured for this volume, 1015 * flushing the data to disk is handled separately below. 1016 */ 1017 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 1018 1019 if (!file->f_op->fsync) {/* COMMIT3 cannot work */ 1020 stable = 2; 1021 *stablep = 2; /* FILE_SYNC */ 1022 } 1023 1024 if (!EX_ISSYNC(exp)) 1025 stable = 0; 1026 if (stable && !use_wgather) { 1027 spin_lock(&file->f_lock); 1028 file->f_flags |= O_SYNC; 1029 spin_unlock(&file->f_lock); 1030 } 1031 1032 /* Write the data. */ 1033 oldfs = get_fs(); set_fs(KERNEL_DS); 1034 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); 1035 set_fs(oldfs); 1036 if (host_err < 0) 1037 goto out_nfserr; 1038 *cnt = host_err; 1039 nfsdstats.io_write += host_err; 1040 fsnotify_modify(file); 1041 1042 /* clear setuid/setgid flag after write */ 1043 if (inode->i_mode & (S_ISUID | S_ISGID)) 1044 kill_suid(dentry); 1045 1046 if (stable && use_wgather) 1047 host_err = wait_for_concurrent_writes(file); 1048 1049 out_nfserr: 1050 dprintk("nfsd: write complete host_err=%d\n", host_err); 1051 if (host_err >= 0) 1052 err = 0; 1053 else 1054 err = nfserrno(host_err); 1055 return err; 1056 } 1057 1058 /* 1059 * Read data from a file. count must contain the requested read count 1060 * on entry. On return, *count contains the number of bytes actually read. 1061 * N.B. After this call fhp needs an fh_put 1062 */ 1063 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, 1064 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 1065 { 1066 struct file *file; 1067 struct inode *inode; 1068 struct raparms *ra; 1069 __be32 err; 1070 1071 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); 1072 if (err) 1073 return err; 1074 1075 inode = file->f_path.dentry->d_inode; 1076 1077 /* Get readahead parameters */ 1078 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 1079 1080 if (ra && ra->p_set) 1081 file->f_ra = ra->p_ra; 1082 1083 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1084 1085 /* Write back readahead params */ 1086 if (ra) { 1087 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 1088 spin_lock(&rab->pb_lock); 1089 ra->p_ra = file->f_ra; 1090 ra->p_set = 1; 1091 ra->p_count--; 1092 spin_unlock(&rab->pb_lock); 1093 } 1094 1095 nfsd_close(file); 1096 return err; 1097 } 1098 1099 /* As above, but use the provided file descriptor. */ 1100 __be32 1101 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1102 loff_t offset, struct kvec *vec, int vlen, 1103 unsigned long *count) 1104 { 1105 __be32 err; 1106 1107 if (file) { 1108 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1109 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); 1110 if (err) 1111 goto out; 1112 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1113 } else /* Note file may still be NULL in NFSv4 special stateid case: */ 1114 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count); 1115 out: 1116 return err; 1117 } 1118 1119 /* 1120 * Write data to a file. 1121 * The stable flag requests synchronous writes. 1122 * N.B. After this call fhp needs an fh_put 1123 */ 1124 __be32 1125 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1126 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1127 int *stablep) 1128 { 1129 __be32 err = 0; 1130 1131 if (file) { 1132 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1133 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1134 if (err) 1135 goto out; 1136 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1137 stablep); 1138 } else { 1139 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1140 if (err) 1141 goto out; 1142 1143 if (cnt) 1144 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1145 cnt, stablep); 1146 nfsd_close(file); 1147 } 1148 out: 1149 return err; 1150 } 1151 1152 #ifdef CONFIG_NFSD_V3 1153 /* 1154 * Commit all pending writes to stable storage. 1155 * 1156 * Note: we only guarantee that data that lies within the range specified 1157 * by the 'offset' and 'count' parameters will be synced. 1158 * 1159 * Unfortunately we cannot lock the file to make sure we return full WCC 1160 * data to the client, as locking happens lower down in the filesystem. 1161 */ 1162 __be32 1163 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1164 loff_t offset, unsigned long count) 1165 { 1166 struct file *file; 1167 loff_t end = LLONG_MAX; 1168 __be32 err = nfserr_inval; 1169 1170 if (offset < 0) 1171 goto out; 1172 if (count != 0) { 1173 end = offset + (loff_t)count - 1; 1174 if (end < offset) 1175 goto out; 1176 } 1177 1178 err = nfsd_open(rqstp, fhp, S_IFREG, 1179 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file); 1180 if (err) 1181 goto out; 1182 if (EX_ISSYNC(fhp->fh_export)) { 1183 int err2 = vfs_fsync_range(file, offset, end, 0); 1184 1185 if (err2 != -EINVAL) 1186 err = nfserrno(err2); 1187 else 1188 err = nfserr_notsupp; 1189 } 1190 1191 nfsd_close(file); 1192 out: 1193 return err; 1194 } 1195 #endif /* CONFIG_NFSD_V3 */ 1196 1197 static __be32 1198 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1199 struct iattr *iap) 1200 { 1201 /* 1202 * Mode has already been set earlier in create: 1203 */ 1204 iap->ia_valid &= ~ATTR_MODE; 1205 /* 1206 * Setting uid/gid works only for root. Irix appears to 1207 * send along the gid on create when it tries to implement 1208 * setgid directories via NFS: 1209 */ 1210 if (current_fsuid() != 0) 1211 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1212 if (iap->ia_valid) 1213 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1214 return 0; 1215 } 1216 1217 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1218 * setting size to 0 may fail for some specific file systems by the permission 1219 * checking which requires WRITE permission but the mode is 000. 1220 * we ignore the resizing(to 0) on the just new created file, since the size is 1221 * 0 after file created. 1222 * 1223 * call this only after vfs_create() is called. 1224 * */ 1225 static void 1226 nfsd_check_ignore_resizing(struct iattr *iap) 1227 { 1228 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1229 iap->ia_valid &= ~ATTR_SIZE; 1230 } 1231 1232 /* 1233 * Create a file (regular, directory, device, fifo); UNIX sockets 1234 * not yet implemented. 1235 * If the response fh has been verified, the parent directory should 1236 * already be locked. Note that the parent directory is left locked. 1237 * 1238 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1239 */ 1240 __be32 1241 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1242 char *fname, int flen, struct iattr *iap, 1243 int type, dev_t rdev, struct svc_fh *resfhp) 1244 { 1245 struct dentry *dentry, *dchild = NULL; 1246 struct inode *dirp; 1247 __be32 err; 1248 __be32 err2; 1249 int host_err; 1250 1251 err = nfserr_perm; 1252 if (!flen) 1253 goto out; 1254 err = nfserr_exist; 1255 if (isdotent(fname, flen)) 1256 goto out; 1257 1258 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1259 if (err) 1260 goto out; 1261 1262 dentry = fhp->fh_dentry; 1263 dirp = dentry->d_inode; 1264 1265 err = nfserr_notdir; 1266 if (!dirp->i_op->lookup) 1267 goto out; 1268 /* 1269 * Check whether the response file handle has been verified yet. 1270 * If it has, the parent directory should already be locked. 1271 */ 1272 if (!resfhp->fh_dentry) { 1273 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1274 fh_lock_nested(fhp, I_MUTEX_PARENT); 1275 dchild = lookup_one_len(fname, dentry, flen); 1276 host_err = PTR_ERR(dchild); 1277 if (IS_ERR(dchild)) 1278 goto out_nfserr; 1279 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1280 if (err) 1281 goto out; 1282 } else { 1283 /* called from nfsd_proc_create */ 1284 dchild = dget(resfhp->fh_dentry); 1285 if (!fhp->fh_locked) { 1286 /* not actually possible */ 1287 printk(KERN_ERR 1288 "nfsd_create: parent %s/%s not locked!\n", 1289 dentry->d_parent->d_name.name, 1290 dentry->d_name.name); 1291 err = nfserr_io; 1292 goto out; 1293 } 1294 } 1295 /* 1296 * Make sure the child dentry is still negative ... 1297 */ 1298 err = nfserr_exist; 1299 if (dchild->d_inode) { 1300 dprintk("nfsd_create: dentry %s/%s not negative!\n", 1301 dentry->d_name.name, dchild->d_name.name); 1302 goto out; 1303 } 1304 1305 if (!(iap->ia_valid & ATTR_MODE)) 1306 iap->ia_mode = 0; 1307 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1308 1309 err = nfserr_inval; 1310 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1311 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1312 type); 1313 goto out; 1314 } 1315 1316 host_err = fh_want_write(fhp); 1317 if (host_err) 1318 goto out_nfserr; 1319 1320 /* 1321 * Get the dir op function pointer. 1322 */ 1323 err = 0; 1324 switch (type) { 1325 case S_IFREG: 1326 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1327 if (!host_err) 1328 nfsd_check_ignore_resizing(iap); 1329 break; 1330 case S_IFDIR: 1331 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1332 break; 1333 case S_IFCHR: 1334 case S_IFBLK: 1335 case S_IFIFO: 1336 case S_IFSOCK: 1337 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1338 break; 1339 } 1340 if (host_err < 0) { 1341 fh_drop_write(fhp); 1342 goto out_nfserr; 1343 } 1344 1345 err = nfsd_create_setattr(rqstp, resfhp, iap); 1346 1347 /* 1348 * nfsd_setattr already committed the child. Transactional filesystems 1349 * had a chance to commit changes for both parent and child 1350 * simultaneously making the following commit_metadata a noop. 1351 */ 1352 err2 = nfserrno(commit_metadata(fhp)); 1353 if (err2) 1354 err = err2; 1355 fh_drop_write(fhp); 1356 /* 1357 * Update the file handle to get the new inode info. 1358 */ 1359 if (!err) 1360 err = fh_update(resfhp); 1361 out: 1362 if (dchild && !IS_ERR(dchild)) 1363 dput(dchild); 1364 return err; 1365 1366 out_nfserr: 1367 err = nfserrno(host_err); 1368 goto out; 1369 } 1370 1371 #ifdef CONFIG_NFSD_V3 1372 1373 static inline int nfsd_create_is_exclusive(int createmode) 1374 { 1375 return createmode == NFS3_CREATE_EXCLUSIVE 1376 || createmode == NFS4_CREATE_EXCLUSIVE4_1; 1377 } 1378 1379 /* 1380 * NFSv3 and NFSv4 version of nfsd_create 1381 */ 1382 __be32 1383 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1384 char *fname, int flen, struct iattr *iap, 1385 struct svc_fh *resfhp, int createmode, u32 *verifier, 1386 bool *truncp, bool *created) 1387 { 1388 struct dentry *dentry, *dchild = NULL; 1389 struct inode *dirp; 1390 __be32 err; 1391 int host_err; 1392 __u32 v_mtime=0, v_atime=0; 1393 1394 err = nfserr_perm; 1395 if (!flen) 1396 goto out; 1397 err = nfserr_exist; 1398 if (isdotent(fname, flen)) 1399 goto out; 1400 if (!(iap->ia_valid & ATTR_MODE)) 1401 iap->ia_mode = 0; 1402 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 1403 if (err) 1404 goto out; 1405 1406 dentry = fhp->fh_dentry; 1407 dirp = dentry->d_inode; 1408 1409 /* Get all the sanity checks out of the way before 1410 * we lock the parent. */ 1411 err = nfserr_notdir; 1412 if (!dirp->i_op->lookup) 1413 goto out; 1414 fh_lock_nested(fhp, I_MUTEX_PARENT); 1415 1416 /* 1417 * Compose the response file handle. 1418 */ 1419 dchild = lookup_one_len(fname, dentry, flen); 1420 host_err = PTR_ERR(dchild); 1421 if (IS_ERR(dchild)) 1422 goto out_nfserr; 1423 1424 /* If file doesn't exist, check for permissions to create one */ 1425 if (!dchild->d_inode) { 1426 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1427 if (err) 1428 goto out; 1429 } 1430 1431 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1432 if (err) 1433 goto out; 1434 1435 if (nfsd_create_is_exclusive(createmode)) { 1436 /* solaris7 gets confused (bugid 4218508) if these have 1437 * the high bit set, so just clear the high bits. If this is 1438 * ever changed to use different attrs for storing the 1439 * verifier, then do_open_lookup() will also need to be fixed 1440 * accordingly. 1441 */ 1442 v_mtime = verifier[0]&0x7fffffff; 1443 v_atime = verifier[1]&0x7fffffff; 1444 } 1445 1446 host_err = fh_want_write(fhp); 1447 if (host_err) 1448 goto out_nfserr; 1449 if (dchild->d_inode) { 1450 err = 0; 1451 1452 switch (createmode) { 1453 case NFS3_CREATE_UNCHECKED: 1454 if (! S_ISREG(dchild->d_inode->i_mode)) 1455 err = nfserr_exist; 1456 else if (truncp) { 1457 /* in nfsv4, we need to treat this case a little 1458 * differently. we don't want to truncate the 1459 * file now; this would be wrong if the OPEN 1460 * fails for some other reason. furthermore, 1461 * if the size is nonzero, we should ignore it 1462 * according to spec! 1463 */ 1464 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1465 } 1466 else { 1467 iap->ia_valid &= ATTR_SIZE; 1468 goto set_attr; 1469 } 1470 break; 1471 case NFS3_CREATE_EXCLUSIVE: 1472 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1473 && dchild->d_inode->i_atime.tv_sec == v_atime 1474 && dchild->d_inode->i_size == 0 ) 1475 break; 1476 case NFS4_CREATE_EXCLUSIVE4_1: 1477 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1478 && dchild->d_inode->i_atime.tv_sec == v_atime 1479 && dchild->d_inode->i_size == 0 ) 1480 goto set_attr; 1481 /* fallthru */ 1482 case NFS3_CREATE_GUARDED: 1483 err = nfserr_exist; 1484 } 1485 fh_drop_write(fhp); 1486 goto out; 1487 } 1488 1489 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1490 if (host_err < 0) { 1491 fh_drop_write(fhp); 1492 goto out_nfserr; 1493 } 1494 if (created) 1495 *created = 1; 1496 1497 nfsd_check_ignore_resizing(iap); 1498 1499 if (nfsd_create_is_exclusive(createmode)) { 1500 /* Cram the verifier into atime/mtime */ 1501 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1502 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1503 /* XXX someone who knows this better please fix it for nsec */ 1504 iap->ia_mtime.tv_sec = v_mtime; 1505 iap->ia_atime.tv_sec = v_atime; 1506 iap->ia_mtime.tv_nsec = 0; 1507 iap->ia_atime.tv_nsec = 0; 1508 } 1509 1510 set_attr: 1511 err = nfsd_create_setattr(rqstp, resfhp, iap); 1512 1513 /* 1514 * nfsd_setattr already committed the child (and possibly also the parent). 1515 */ 1516 if (!err) 1517 err = nfserrno(commit_metadata(fhp)); 1518 1519 fh_drop_write(fhp); 1520 /* 1521 * Update the filehandle to get the new inode info. 1522 */ 1523 if (!err) 1524 err = fh_update(resfhp); 1525 1526 out: 1527 fh_unlock(fhp); 1528 if (dchild && !IS_ERR(dchild)) 1529 dput(dchild); 1530 return err; 1531 1532 out_nfserr: 1533 err = nfserrno(host_err); 1534 goto out; 1535 } 1536 #endif /* CONFIG_NFSD_V3 */ 1537 1538 /* 1539 * Read a symlink. On entry, *lenp must contain the maximum path length that 1540 * fits into the buffer. On return, it contains the true length. 1541 * N.B. After this call fhp needs an fh_put 1542 */ 1543 __be32 1544 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1545 { 1546 struct dentry *dentry; 1547 struct inode *inode; 1548 mm_segment_t oldfs; 1549 __be32 err; 1550 int host_err; 1551 1552 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1553 if (err) 1554 goto out; 1555 1556 dentry = fhp->fh_dentry; 1557 inode = dentry->d_inode; 1558 1559 err = nfserr_inval; 1560 if (!inode->i_op->readlink) 1561 goto out; 1562 1563 touch_atime(fhp->fh_export->ex_path.mnt, dentry); 1564 /* N.B. Why does this call need a get_fs()?? 1565 * Remove the set_fs and watch the fireworks:-) --okir 1566 */ 1567 1568 oldfs = get_fs(); set_fs(KERNEL_DS); 1569 host_err = inode->i_op->readlink(dentry, buf, *lenp); 1570 set_fs(oldfs); 1571 1572 if (host_err < 0) 1573 goto out_nfserr; 1574 *lenp = host_err; 1575 err = 0; 1576 out: 1577 return err; 1578 1579 out_nfserr: 1580 err = nfserrno(host_err); 1581 goto out; 1582 } 1583 1584 /* 1585 * Create a symlink and look up its inode 1586 * N.B. After this call _both_ fhp and resfhp need an fh_put 1587 */ 1588 __be32 1589 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1590 char *fname, int flen, 1591 char *path, int plen, 1592 struct svc_fh *resfhp, 1593 struct iattr *iap) 1594 { 1595 struct dentry *dentry, *dnew; 1596 __be32 err, cerr; 1597 int host_err; 1598 1599 err = nfserr_noent; 1600 if (!flen || !plen) 1601 goto out; 1602 err = nfserr_exist; 1603 if (isdotent(fname, flen)) 1604 goto out; 1605 1606 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1607 if (err) 1608 goto out; 1609 fh_lock(fhp); 1610 dentry = fhp->fh_dentry; 1611 dnew = lookup_one_len(fname, dentry, flen); 1612 host_err = PTR_ERR(dnew); 1613 if (IS_ERR(dnew)) 1614 goto out_nfserr; 1615 1616 host_err = fh_want_write(fhp); 1617 if (host_err) 1618 goto out_nfserr; 1619 1620 if (unlikely(path[plen] != 0)) { 1621 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1622 if (path_alloced == NULL) 1623 host_err = -ENOMEM; 1624 else { 1625 strncpy(path_alloced, path, plen); 1626 path_alloced[plen] = 0; 1627 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); 1628 kfree(path_alloced); 1629 } 1630 } else 1631 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1632 err = nfserrno(host_err); 1633 if (!err) 1634 err = nfserrno(commit_metadata(fhp)); 1635 fh_unlock(fhp); 1636 1637 fh_drop_write(fhp); 1638 1639 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1640 dput(dnew); 1641 if (err==0) err = cerr; 1642 out: 1643 return err; 1644 1645 out_nfserr: 1646 err = nfserrno(host_err); 1647 goto out; 1648 } 1649 1650 /* 1651 * Create a hardlink 1652 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1653 */ 1654 __be32 1655 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1656 char *name, int len, struct svc_fh *tfhp) 1657 { 1658 struct dentry *ddir, *dnew, *dold; 1659 struct inode *dirp; 1660 __be32 err; 1661 int host_err; 1662 1663 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1664 if (err) 1665 goto out; 1666 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP); 1667 if (err) 1668 goto out; 1669 err = nfserr_isdir; 1670 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode)) 1671 goto out; 1672 err = nfserr_perm; 1673 if (!len) 1674 goto out; 1675 err = nfserr_exist; 1676 if (isdotent(name, len)) 1677 goto out; 1678 1679 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1680 ddir = ffhp->fh_dentry; 1681 dirp = ddir->d_inode; 1682 1683 dnew = lookup_one_len(name, ddir, len); 1684 host_err = PTR_ERR(dnew); 1685 if (IS_ERR(dnew)) 1686 goto out_nfserr; 1687 1688 dold = tfhp->fh_dentry; 1689 1690 host_err = fh_want_write(tfhp); 1691 if (host_err) { 1692 err = nfserrno(host_err); 1693 goto out_dput; 1694 } 1695 err = nfserr_noent; 1696 if (!dold->d_inode) 1697 goto out_drop_write; 1698 host_err = nfsd_break_lease(dold->d_inode); 1699 if (host_err) { 1700 err = nfserrno(host_err); 1701 goto out_drop_write; 1702 } 1703 host_err = vfs_link(dold, dirp, dnew); 1704 if (!host_err) { 1705 err = nfserrno(commit_metadata(ffhp)); 1706 if (!err) 1707 err = nfserrno(commit_metadata(tfhp)); 1708 } else { 1709 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1710 err = nfserr_acces; 1711 else 1712 err = nfserrno(host_err); 1713 } 1714 out_drop_write: 1715 fh_drop_write(tfhp); 1716 out_dput: 1717 dput(dnew); 1718 out_unlock: 1719 fh_unlock(ffhp); 1720 out: 1721 return err; 1722 1723 out_nfserr: 1724 err = nfserrno(host_err); 1725 goto out_unlock; 1726 } 1727 1728 /* 1729 * Rename a file 1730 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1731 */ 1732 __be32 1733 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1734 struct svc_fh *tfhp, char *tname, int tlen) 1735 { 1736 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1737 struct inode *fdir, *tdir; 1738 __be32 err; 1739 int host_err; 1740 1741 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1742 if (err) 1743 goto out; 1744 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1745 if (err) 1746 goto out; 1747 1748 fdentry = ffhp->fh_dentry; 1749 fdir = fdentry->d_inode; 1750 1751 tdentry = tfhp->fh_dentry; 1752 tdir = tdentry->d_inode; 1753 1754 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; 1755 if (ffhp->fh_export != tfhp->fh_export) 1756 goto out; 1757 1758 err = nfserr_perm; 1759 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1760 goto out; 1761 1762 /* cannot use fh_lock as we need deadlock protective ordering 1763 * so do it by hand */ 1764 trap = lock_rename(tdentry, fdentry); 1765 ffhp->fh_locked = tfhp->fh_locked = 1; 1766 fill_pre_wcc(ffhp); 1767 fill_pre_wcc(tfhp); 1768 1769 odentry = lookup_one_len(fname, fdentry, flen); 1770 host_err = PTR_ERR(odentry); 1771 if (IS_ERR(odentry)) 1772 goto out_nfserr; 1773 1774 host_err = -ENOENT; 1775 if (!odentry->d_inode) 1776 goto out_dput_old; 1777 host_err = -EINVAL; 1778 if (odentry == trap) 1779 goto out_dput_old; 1780 1781 ndentry = lookup_one_len(tname, tdentry, tlen); 1782 host_err = PTR_ERR(ndentry); 1783 if (IS_ERR(ndentry)) 1784 goto out_dput_old; 1785 host_err = -ENOTEMPTY; 1786 if (ndentry == trap) 1787 goto out_dput_new; 1788 1789 host_err = -EXDEV; 1790 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1791 goto out_dput_new; 1792 host_err = fh_want_write(ffhp); 1793 if (host_err) 1794 goto out_dput_new; 1795 1796 host_err = nfsd_break_lease(odentry->d_inode); 1797 if (host_err) 1798 goto out_drop_write; 1799 if (ndentry->d_inode) { 1800 host_err = nfsd_break_lease(ndentry->d_inode); 1801 if (host_err) 1802 goto out_drop_write; 1803 } 1804 host_err = vfs_rename(fdir, odentry, tdir, ndentry); 1805 if (!host_err) { 1806 host_err = commit_metadata(tfhp); 1807 if (!host_err) 1808 host_err = commit_metadata(ffhp); 1809 } 1810 out_drop_write: 1811 fh_drop_write(ffhp); 1812 out_dput_new: 1813 dput(ndentry); 1814 out_dput_old: 1815 dput(odentry); 1816 out_nfserr: 1817 err = nfserrno(host_err); 1818 1819 /* we cannot reply on fh_unlock on the two filehandles, 1820 * as that would do the wrong thing if the two directories 1821 * were the same, so again we do it by hand 1822 */ 1823 fill_post_wcc(ffhp); 1824 fill_post_wcc(tfhp); 1825 unlock_rename(tdentry, fdentry); 1826 ffhp->fh_locked = tfhp->fh_locked = 0; 1827 1828 out: 1829 return err; 1830 } 1831 1832 /* 1833 * Unlink a file or directory 1834 * N.B. After this call fhp needs an fh_put 1835 */ 1836 __be32 1837 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1838 char *fname, int flen) 1839 { 1840 struct dentry *dentry, *rdentry; 1841 struct inode *dirp; 1842 __be32 err; 1843 int host_err; 1844 1845 err = nfserr_acces; 1846 if (!flen || isdotent(fname, flen)) 1847 goto out; 1848 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1849 if (err) 1850 goto out; 1851 1852 fh_lock_nested(fhp, I_MUTEX_PARENT); 1853 dentry = fhp->fh_dentry; 1854 dirp = dentry->d_inode; 1855 1856 rdentry = lookup_one_len(fname, dentry, flen); 1857 host_err = PTR_ERR(rdentry); 1858 if (IS_ERR(rdentry)) 1859 goto out_nfserr; 1860 1861 if (!rdentry->d_inode) { 1862 dput(rdentry); 1863 err = nfserr_noent; 1864 goto out; 1865 } 1866 1867 if (!type) 1868 type = rdentry->d_inode->i_mode & S_IFMT; 1869 1870 host_err = fh_want_write(fhp); 1871 if (host_err) 1872 goto out_put; 1873 1874 host_err = nfsd_break_lease(rdentry->d_inode); 1875 if (host_err) 1876 goto out_drop_write; 1877 if (type != S_IFDIR) 1878 host_err = vfs_unlink(dirp, rdentry); 1879 else 1880 host_err = vfs_rmdir(dirp, rdentry); 1881 if (!host_err) 1882 host_err = commit_metadata(fhp); 1883 out_drop_write: 1884 fh_drop_write(fhp); 1885 out_put: 1886 dput(rdentry); 1887 1888 out_nfserr: 1889 err = nfserrno(host_err); 1890 out: 1891 return err; 1892 } 1893 1894 /* 1895 * We do this buffering because we must not call back into the file 1896 * system's ->lookup() method from the filldir callback. That may well 1897 * deadlock a number of file systems. 1898 * 1899 * This is based heavily on the implementation of same in XFS. 1900 */ 1901 struct buffered_dirent { 1902 u64 ino; 1903 loff_t offset; 1904 int namlen; 1905 unsigned int d_type; 1906 char name[]; 1907 }; 1908 1909 struct readdir_data { 1910 char *dirent; 1911 size_t used; 1912 int full; 1913 }; 1914 1915 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1916 loff_t offset, u64 ino, unsigned int d_type) 1917 { 1918 struct readdir_data *buf = __buf; 1919 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1920 unsigned int reclen; 1921 1922 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1923 if (buf->used + reclen > PAGE_SIZE) { 1924 buf->full = 1; 1925 return -EINVAL; 1926 } 1927 1928 de->namlen = namlen; 1929 de->offset = offset; 1930 de->ino = ino; 1931 de->d_type = d_type; 1932 memcpy(de->name, name, namlen); 1933 buf->used += reclen; 1934 1935 return 0; 1936 } 1937 1938 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1939 struct readdir_cd *cdp, loff_t *offsetp) 1940 { 1941 struct readdir_data buf; 1942 struct buffered_dirent *de; 1943 int host_err; 1944 int size; 1945 loff_t offset; 1946 1947 buf.dirent = (void *)__get_free_page(GFP_KERNEL); 1948 if (!buf.dirent) 1949 return nfserrno(-ENOMEM); 1950 1951 offset = *offsetp; 1952 1953 while (1) { 1954 struct inode *dir_inode = file->f_path.dentry->d_inode; 1955 unsigned int reclen; 1956 1957 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1958 buf.used = 0; 1959 buf.full = 0; 1960 1961 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf); 1962 if (buf.full) 1963 host_err = 0; 1964 1965 if (host_err < 0) 1966 break; 1967 1968 size = buf.used; 1969 1970 if (!size) 1971 break; 1972 1973 /* 1974 * Various filldir functions may end up calling back into 1975 * lookup_one_len() and the file system's ->lookup() method. 1976 * These expect i_mutex to be held, as it would within readdir. 1977 */ 1978 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1979 if (host_err) 1980 break; 1981 1982 de = (struct buffered_dirent *)buf.dirent; 1983 while (size > 0) { 1984 offset = de->offset; 1985 1986 if (func(cdp, de->name, de->namlen, de->offset, 1987 de->ino, de->d_type)) 1988 break; 1989 1990 if (cdp->err != nfs_ok) 1991 break; 1992 1993 reclen = ALIGN(sizeof(*de) + de->namlen, 1994 sizeof(u64)); 1995 size -= reclen; 1996 de = (struct buffered_dirent *)((char *)de + reclen); 1997 } 1998 mutex_unlock(&dir_inode->i_mutex); 1999 if (size > 0) /* We bailed out early */ 2000 break; 2001 2002 offset = vfs_llseek(file, 0, SEEK_CUR); 2003 } 2004 2005 free_page((unsigned long)(buf.dirent)); 2006 2007 if (host_err) 2008 return nfserrno(host_err); 2009 2010 *offsetp = offset; 2011 return cdp->err; 2012 } 2013 2014 /* 2015 * Read entries from a directory. 2016 * The NFSv3/4 verifier we ignore for now. 2017 */ 2018 __be32 2019 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 2020 struct readdir_cd *cdp, filldir_t func) 2021 { 2022 __be32 err; 2023 struct file *file; 2024 loff_t offset = *offsetp; 2025 2026 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file); 2027 if (err) 2028 goto out; 2029 2030 offset = vfs_llseek(file, offset, 0); 2031 if (offset < 0) { 2032 err = nfserrno((int)offset); 2033 goto out_close; 2034 } 2035 2036 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 2037 2038 if (err == nfserr_eof || err == nfserr_toosmall) 2039 err = nfs_ok; /* can still be found in ->err */ 2040 out_close: 2041 nfsd_close(file); 2042 out: 2043 return err; 2044 } 2045 2046 /* 2047 * Get file system stats 2048 * N.B. After this call fhp needs an fh_put 2049 */ 2050 __be32 2051 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 2052 { 2053 __be32 err; 2054 2055 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 2056 if (!err) { 2057 struct path path = { 2058 .mnt = fhp->fh_export->ex_path.mnt, 2059 .dentry = fhp->fh_dentry, 2060 }; 2061 if (vfs_statfs(&path, stat)) 2062 err = nfserr_io; 2063 } 2064 return err; 2065 } 2066 2067 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 2068 { 2069 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 2070 } 2071 2072 /* 2073 * Check for a user's access permissions to this inode. 2074 */ 2075 __be32 2076 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 2077 struct dentry *dentry, int acc) 2078 { 2079 struct inode *inode = dentry->d_inode; 2080 int err; 2081 2082 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP) 2083 return 0; 2084 #if 0 2085 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 2086 acc, 2087 (acc & NFSD_MAY_READ)? " read" : "", 2088 (acc & NFSD_MAY_WRITE)? " write" : "", 2089 (acc & NFSD_MAY_EXEC)? " exec" : "", 2090 (acc & NFSD_MAY_SATTR)? " sattr" : "", 2091 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 2092 (acc & NFSD_MAY_LOCK)? " lock" : "", 2093 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 2094 inode->i_mode, 2095 IS_IMMUTABLE(inode)? " immut" : "", 2096 IS_APPEND(inode)? " append" : "", 2097 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2098 dprintk(" owner %d/%d user %d/%d\n", 2099 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2100 #endif 2101 2102 /* Normally we reject any write/sattr etc access on a read-only file 2103 * system. But if it is IRIX doing check on write-access for a 2104 * device special file, we ignore rofs. 2105 */ 2106 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2107 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2108 if (exp_rdonly(rqstp, exp) || 2109 __mnt_is_readonly(exp->ex_path.mnt)) 2110 return nfserr_rofs; 2111 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2112 return nfserr_perm; 2113 } 2114 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2115 return nfserr_perm; 2116 2117 if (acc & NFSD_MAY_LOCK) { 2118 /* If we cannot rely on authentication in NLM requests, 2119 * just allow locks, otherwise require read permission, or 2120 * ownership 2121 */ 2122 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2123 return 0; 2124 else 2125 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2126 } 2127 /* 2128 * The file owner always gets access permission for accesses that 2129 * would normally be checked at open time. This is to make 2130 * file access work even when the client has done a fchmod(fd, 0). 2131 * 2132 * However, `cp foo bar' should fail nevertheless when bar is 2133 * readonly. A sensible way to do this might be to reject all 2134 * attempts to truncate a read-only file, because a creat() call 2135 * always implies file truncation. 2136 * ... but this isn't really fair. A process may reasonably call 2137 * ftruncate on an open file descriptor on a file with perm 000. 2138 * We must trust the client to do permission checking - using "ACCESS" 2139 * with NFSv3. 2140 */ 2141 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2142 inode->i_uid == current_fsuid()) 2143 return 0; 2144 2145 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2146 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2147 2148 /* Allow read access to binaries even when mode 111 */ 2149 if (err == -EACCES && S_ISREG(inode->i_mode) && 2150 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) || 2151 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC))) 2152 err = inode_permission(inode, MAY_EXEC); 2153 2154 return err? nfserrno(err) : 0; 2155 } 2156 2157 void 2158 nfsd_racache_shutdown(void) 2159 { 2160 struct raparms *raparm, *last_raparm; 2161 unsigned int i; 2162 2163 dprintk("nfsd: freeing readahead buffers.\n"); 2164 2165 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2166 raparm = raparm_hash[i].pb_head; 2167 while(raparm) { 2168 last_raparm = raparm; 2169 raparm = raparm->p_next; 2170 kfree(last_raparm); 2171 } 2172 raparm_hash[i].pb_head = NULL; 2173 } 2174 } 2175 /* 2176 * Initialize readahead param cache 2177 */ 2178 int 2179 nfsd_racache_init(int cache_size) 2180 { 2181 int i; 2182 int j = 0; 2183 int nperbucket; 2184 struct raparms **raparm = NULL; 2185 2186 2187 if (raparm_hash[0].pb_head) 2188 return 0; 2189 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2190 if (nperbucket < 2) 2191 nperbucket = 2; 2192 cache_size = nperbucket * RAPARM_HASH_SIZE; 2193 2194 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2195 2196 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2197 spin_lock_init(&raparm_hash[i].pb_lock); 2198 2199 raparm = &raparm_hash[i].pb_head; 2200 for (j = 0; j < nperbucket; j++) { 2201 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2202 if (!*raparm) 2203 goto out_nomem; 2204 raparm = &(*raparm)->p_next; 2205 } 2206 *raparm = NULL; 2207 } 2208 2209 nfsdstats.ra_size = cache_size; 2210 return 0; 2211 2212 out_nomem: 2213 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2214 nfsd_racache_shutdown(); 2215 return -ENOMEM; 2216 } 2217 2218 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 2219 struct posix_acl * 2220 nfsd_get_posix_acl(struct svc_fh *fhp, int type) 2221 { 2222 struct inode *inode = fhp->fh_dentry->d_inode; 2223 char *name; 2224 void *value = NULL; 2225 ssize_t size; 2226 struct posix_acl *acl; 2227 2228 if (!IS_POSIXACL(inode)) 2229 return ERR_PTR(-EOPNOTSUPP); 2230 2231 switch (type) { 2232 case ACL_TYPE_ACCESS: 2233 name = POSIX_ACL_XATTR_ACCESS; 2234 break; 2235 case ACL_TYPE_DEFAULT: 2236 name = POSIX_ACL_XATTR_DEFAULT; 2237 break; 2238 default: 2239 return ERR_PTR(-EOPNOTSUPP); 2240 } 2241 2242 size = nfsd_getxattr(fhp->fh_dentry, name, &value); 2243 if (size < 0) 2244 return ERR_PTR(size); 2245 2246 acl = posix_acl_from_xattr(value, size); 2247 kfree(value); 2248 return acl; 2249 } 2250 2251 int 2252 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) 2253 { 2254 struct inode *inode = fhp->fh_dentry->d_inode; 2255 char *name; 2256 void *value = NULL; 2257 size_t size; 2258 int error; 2259 2260 if (!IS_POSIXACL(inode) || 2261 !inode->i_op->setxattr || !inode->i_op->removexattr) 2262 return -EOPNOTSUPP; 2263 switch(type) { 2264 case ACL_TYPE_ACCESS: 2265 name = POSIX_ACL_XATTR_ACCESS; 2266 break; 2267 case ACL_TYPE_DEFAULT: 2268 name = POSIX_ACL_XATTR_DEFAULT; 2269 break; 2270 default: 2271 return -EOPNOTSUPP; 2272 } 2273 2274 if (acl && acl->a_count) { 2275 size = posix_acl_xattr_size(acl->a_count); 2276 value = kmalloc(size, GFP_KERNEL); 2277 if (!value) 2278 return -ENOMEM; 2279 error = posix_acl_to_xattr(acl, value, size); 2280 if (error < 0) 2281 goto getout; 2282 size = error; 2283 } else 2284 size = 0; 2285 2286 error = fh_want_write(fhp); 2287 if (error) 2288 goto getout; 2289 if (size) 2290 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); 2291 else { 2292 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) 2293 error = 0; 2294 else { 2295 error = vfs_removexattr(fhp->fh_dentry, name); 2296 if (error == -ENODATA) 2297 error = 0; 2298 } 2299 } 2300 fh_drop_write(fhp); 2301 2302 getout: 2303 kfree(value); 2304 return error; 2305 } 2306 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 2307