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 path path; 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 path.mnt = fhp->fh_export->ex_path.mnt; 766 path.dentry = fhp->fh_dentry; 767 inode = path.dentry->d_inode; 768 769 /* Disallow write access to files with the append-only bit set 770 * or any access when mandatory locking enabled 771 */ 772 err = nfserr_perm; 773 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE)) 774 goto out; 775 /* 776 * We must ignore files (but only files) which might have mandatory 777 * locks on them because there is no way to know if the accesser has 778 * the lock. 779 */ 780 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) 781 goto out; 782 783 if (!inode->i_fop) 784 goto out; 785 786 host_err = nfsd_open_break_lease(inode, may_flags); 787 if (host_err) /* NOMEM or WOULDBLOCK */ 788 goto out_nfserr; 789 790 if (may_flags & NFSD_MAY_WRITE) { 791 if (may_flags & NFSD_MAY_READ) 792 flags = O_RDWR|O_LARGEFILE; 793 else 794 flags = O_WRONLY|O_LARGEFILE; 795 } 796 *filp = dentry_open(&path, 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 if (may_flags & NFSD_MAY_64BIT_COOKIE) 803 (*filp)->f_mode |= FMODE_64BITHASH; 804 else 805 (*filp)->f_mode |= FMODE_32BITHASH; 806 } 807 808 out_nfserr: 809 err = nfserrno(host_err); 810 out: 811 validate_process_creds(); 812 return err; 813 } 814 815 /* 816 * Close a file. 817 */ 818 void 819 nfsd_close(struct file *filp) 820 { 821 fput(filp); 822 } 823 824 /* 825 * Obtain the readahead parameters for the file 826 * specified by (dev, ino). 827 */ 828 829 static inline struct raparms * 830 nfsd_get_raparms(dev_t dev, ino_t ino) 831 { 832 struct raparms *ra, **rap, **frap = NULL; 833 int depth = 0; 834 unsigned int hash; 835 struct raparm_hbucket *rab; 836 837 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 838 rab = &raparm_hash[hash]; 839 840 spin_lock(&rab->pb_lock); 841 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 842 if (ra->p_ino == ino && ra->p_dev == dev) 843 goto found; 844 depth++; 845 if (ra->p_count == 0) 846 frap = rap; 847 } 848 depth = nfsdstats.ra_size; 849 if (!frap) { 850 spin_unlock(&rab->pb_lock); 851 return NULL; 852 } 853 rap = frap; 854 ra = *frap; 855 ra->p_dev = dev; 856 ra->p_ino = ino; 857 ra->p_set = 0; 858 ra->p_hindex = hash; 859 found: 860 if (rap != &rab->pb_head) { 861 *rap = ra->p_next; 862 ra->p_next = rab->pb_head; 863 rab->pb_head = ra; 864 } 865 ra->p_count++; 866 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 867 spin_unlock(&rab->pb_lock); 868 return ra; 869 } 870 871 /* 872 * Grab and keep cached pages associated with a file in the svc_rqst 873 * so that they can be passed to the network sendmsg/sendpage routines 874 * directly. They will be released after the sending has completed. 875 */ 876 static int 877 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 878 struct splice_desc *sd) 879 { 880 struct svc_rqst *rqstp = sd->u.data; 881 struct page **pp = rqstp->rq_respages + rqstp->rq_resused; 882 struct page *page = buf->page; 883 size_t size; 884 885 size = sd->len; 886 887 if (rqstp->rq_res.page_len == 0) { 888 get_page(page); 889 put_page(*pp); 890 *pp = page; 891 rqstp->rq_resused++; 892 rqstp->rq_res.page_base = buf->offset; 893 rqstp->rq_res.page_len = size; 894 } else if (page != pp[-1]) { 895 get_page(page); 896 if (*pp) 897 put_page(*pp); 898 *pp = page; 899 rqstp->rq_resused++; 900 rqstp->rq_res.page_len += size; 901 } else 902 rqstp->rq_res.page_len += size; 903 904 return size; 905 } 906 907 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 908 struct splice_desc *sd) 909 { 910 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 911 } 912 913 static __be32 914 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 915 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 916 { 917 mm_segment_t oldfs; 918 __be32 err; 919 int host_err; 920 921 err = nfserr_perm; 922 923 if (file->f_op->splice_read && rqstp->rq_splice_ok) { 924 struct splice_desc sd = { 925 .len = 0, 926 .total_len = *count, 927 .pos = offset, 928 .u.data = rqstp, 929 }; 930 931 rqstp->rq_resused = 1; 932 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 933 } else { 934 oldfs = get_fs(); 935 set_fs(KERNEL_DS); 936 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 937 set_fs(oldfs); 938 } 939 940 if (host_err >= 0) { 941 nfsdstats.io_read += host_err; 942 *count = host_err; 943 err = 0; 944 fsnotify_access(file); 945 } else 946 err = nfserrno(host_err); 947 return err; 948 } 949 950 static void kill_suid(struct dentry *dentry) 951 { 952 struct iattr ia; 953 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 954 955 mutex_lock(&dentry->d_inode->i_mutex); 956 notify_change(dentry, &ia); 957 mutex_unlock(&dentry->d_inode->i_mutex); 958 } 959 960 /* 961 * Gathered writes: If another process is currently writing to the file, 962 * there's a high chance this is another nfsd (triggered by a bulk write 963 * from a client's biod). Rather than syncing the file with each write 964 * request, we sleep for 10 msec. 965 * 966 * I don't know if this roughly approximates C. Juszak's idea of 967 * gathered writes, but it's a nice and simple solution (IMHO), and it 968 * seems to work:-) 969 * 970 * Note: we do this only in the NFSv2 case, since v3 and higher have a 971 * better tool (separate unstable writes and commits) for solving this 972 * problem. 973 */ 974 static int wait_for_concurrent_writes(struct file *file) 975 { 976 struct inode *inode = file->f_path.dentry->d_inode; 977 static ino_t last_ino; 978 static dev_t last_dev; 979 int err = 0; 980 981 if (atomic_read(&inode->i_writecount) > 1 982 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 983 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 984 msleep(10); 985 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 986 } 987 988 if (inode->i_state & I_DIRTY) { 989 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 990 err = vfs_fsync(file, 0); 991 } 992 last_ino = inode->i_ino; 993 last_dev = inode->i_sb->s_dev; 994 return err; 995 } 996 997 static __be32 998 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 999 loff_t offset, struct kvec *vec, int vlen, 1000 unsigned long *cnt, int *stablep) 1001 { 1002 struct svc_export *exp; 1003 struct dentry *dentry; 1004 struct inode *inode; 1005 mm_segment_t oldfs; 1006 __be32 err = 0; 1007 int host_err; 1008 int stable = *stablep; 1009 int use_wgather; 1010 1011 dentry = file->f_path.dentry; 1012 inode = dentry->d_inode; 1013 exp = fhp->fh_export; 1014 1015 /* 1016 * Request sync writes if 1017 * - the sync export option has been set, or 1018 * - the client requested O_SYNC behavior (NFSv3 feature). 1019 * - The file system doesn't support fsync(). 1020 * When NFSv2 gathered writes have been configured for this volume, 1021 * flushing the data to disk is handled separately below. 1022 */ 1023 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 1024 1025 if (!file->f_op->fsync) {/* COMMIT3 cannot work */ 1026 stable = 2; 1027 *stablep = 2; /* FILE_SYNC */ 1028 } 1029 1030 if (!EX_ISSYNC(exp)) 1031 stable = 0; 1032 if (stable && !use_wgather) { 1033 spin_lock(&file->f_lock); 1034 file->f_flags |= O_SYNC; 1035 spin_unlock(&file->f_lock); 1036 } 1037 1038 /* Write the data. */ 1039 oldfs = get_fs(); set_fs(KERNEL_DS); 1040 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); 1041 set_fs(oldfs); 1042 if (host_err < 0) 1043 goto out_nfserr; 1044 *cnt = host_err; 1045 nfsdstats.io_write += host_err; 1046 fsnotify_modify(file); 1047 1048 /* clear setuid/setgid flag after write */ 1049 if (inode->i_mode & (S_ISUID | S_ISGID)) 1050 kill_suid(dentry); 1051 1052 if (stable && use_wgather) 1053 host_err = wait_for_concurrent_writes(file); 1054 1055 out_nfserr: 1056 dprintk("nfsd: write complete host_err=%d\n", host_err); 1057 if (host_err >= 0) 1058 err = 0; 1059 else 1060 err = nfserrno(host_err); 1061 return err; 1062 } 1063 1064 /* 1065 * Read data from a file. count must contain the requested read count 1066 * on entry. On return, *count contains the number of bytes actually read. 1067 * N.B. After this call fhp needs an fh_put 1068 */ 1069 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, 1070 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 1071 { 1072 struct file *file; 1073 struct inode *inode; 1074 struct raparms *ra; 1075 __be32 err; 1076 1077 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); 1078 if (err) 1079 return err; 1080 1081 inode = file->f_path.dentry->d_inode; 1082 1083 /* Get readahead parameters */ 1084 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 1085 1086 if (ra && ra->p_set) 1087 file->f_ra = ra->p_ra; 1088 1089 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1090 1091 /* Write back readahead params */ 1092 if (ra) { 1093 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 1094 spin_lock(&rab->pb_lock); 1095 ra->p_ra = file->f_ra; 1096 ra->p_set = 1; 1097 ra->p_count--; 1098 spin_unlock(&rab->pb_lock); 1099 } 1100 1101 nfsd_close(file); 1102 return err; 1103 } 1104 1105 /* As above, but use the provided file descriptor. */ 1106 __be32 1107 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1108 loff_t offset, struct kvec *vec, int vlen, 1109 unsigned long *count) 1110 { 1111 __be32 err; 1112 1113 if (file) { 1114 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1115 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); 1116 if (err) 1117 goto out; 1118 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1119 } else /* Note file may still be NULL in NFSv4 special stateid case: */ 1120 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count); 1121 out: 1122 return err; 1123 } 1124 1125 /* 1126 * Write data to a file. 1127 * The stable flag requests synchronous writes. 1128 * N.B. After this call fhp needs an fh_put 1129 */ 1130 __be32 1131 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1132 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1133 int *stablep) 1134 { 1135 __be32 err = 0; 1136 1137 if (file) { 1138 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1139 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1140 if (err) 1141 goto out; 1142 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1143 stablep); 1144 } else { 1145 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1146 if (err) 1147 goto out; 1148 1149 if (cnt) 1150 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1151 cnt, stablep); 1152 nfsd_close(file); 1153 } 1154 out: 1155 return err; 1156 } 1157 1158 #ifdef CONFIG_NFSD_V3 1159 /* 1160 * Commit all pending writes to stable storage. 1161 * 1162 * Note: we only guarantee that data that lies within the range specified 1163 * by the 'offset' and 'count' parameters will be synced. 1164 * 1165 * Unfortunately we cannot lock the file to make sure we return full WCC 1166 * data to the client, as locking happens lower down in the filesystem. 1167 */ 1168 __be32 1169 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1170 loff_t offset, unsigned long count) 1171 { 1172 struct file *file; 1173 loff_t end = LLONG_MAX; 1174 __be32 err = nfserr_inval; 1175 1176 if (offset < 0) 1177 goto out; 1178 if (count != 0) { 1179 end = offset + (loff_t)count - 1; 1180 if (end < offset) 1181 goto out; 1182 } 1183 1184 err = nfsd_open(rqstp, fhp, S_IFREG, 1185 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file); 1186 if (err) 1187 goto out; 1188 if (EX_ISSYNC(fhp->fh_export)) { 1189 int err2 = vfs_fsync_range(file, offset, end, 0); 1190 1191 if (err2 != -EINVAL) 1192 err = nfserrno(err2); 1193 else 1194 err = nfserr_notsupp; 1195 } 1196 1197 nfsd_close(file); 1198 out: 1199 return err; 1200 } 1201 #endif /* CONFIG_NFSD_V3 */ 1202 1203 static __be32 1204 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1205 struct iattr *iap) 1206 { 1207 /* 1208 * Mode has already been set earlier in create: 1209 */ 1210 iap->ia_valid &= ~ATTR_MODE; 1211 /* 1212 * Setting uid/gid works only for root. Irix appears to 1213 * send along the gid on create when it tries to implement 1214 * setgid directories via NFS: 1215 */ 1216 if (current_fsuid() != 0) 1217 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1218 if (iap->ia_valid) 1219 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1220 return 0; 1221 } 1222 1223 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1224 * setting size to 0 may fail for some specific file systems by the permission 1225 * checking which requires WRITE permission but the mode is 000. 1226 * we ignore the resizing(to 0) on the just new created file, since the size is 1227 * 0 after file created. 1228 * 1229 * call this only after vfs_create() is called. 1230 * */ 1231 static void 1232 nfsd_check_ignore_resizing(struct iattr *iap) 1233 { 1234 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1235 iap->ia_valid &= ~ATTR_SIZE; 1236 } 1237 1238 /* 1239 * Create a file (regular, directory, device, fifo); UNIX sockets 1240 * not yet implemented. 1241 * If the response fh has been verified, the parent directory should 1242 * already be locked. Note that the parent directory is left locked. 1243 * 1244 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1245 */ 1246 __be32 1247 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1248 char *fname, int flen, struct iattr *iap, 1249 int type, dev_t rdev, struct svc_fh *resfhp) 1250 { 1251 struct dentry *dentry, *dchild = NULL; 1252 struct inode *dirp; 1253 __be32 err; 1254 __be32 err2; 1255 int host_err; 1256 1257 err = nfserr_perm; 1258 if (!flen) 1259 goto out; 1260 err = nfserr_exist; 1261 if (isdotent(fname, flen)) 1262 goto out; 1263 1264 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1265 if (err) 1266 goto out; 1267 1268 dentry = fhp->fh_dentry; 1269 dirp = dentry->d_inode; 1270 1271 err = nfserr_notdir; 1272 if (!dirp->i_op->lookup) 1273 goto out; 1274 /* 1275 * Check whether the response file handle has been verified yet. 1276 * If it has, the parent directory should already be locked. 1277 */ 1278 if (!resfhp->fh_dentry) { 1279 host_err = fh_want_write(fhp); 1280 if (host_err) 1281 goto out_nfserr; 1282 1283 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1284 fh_lock_nested(fhp, I_MUTEX_PARENT); 1285 dchild = lookup_one_len(fname, dentry, flen); 1286 host_err = PTR_ERR(dchild); 1287 if (IS_ERR(dchild)) 1288 goto out_nfserr; 1289 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1290 if (err) 1291 goto out; 1292 } else { 1293 /* called from nfsd_proc_create */ 1294 dchild = dget(resfhp->fh_dentry); 1295 if (!fhp->fh_locked) { 1296 /* not actually possible */ 1297 printk(KERN_ERR 1298 "nfsd_create: parent %s/%s not locked!\n", 1299 dentry->d_parent->d_name.name, 1300 dentry->d_name.name); 1301 err = nfserr_io; 1302 goto out; 1303 } 1304 } 1305 /* 1306 * Make sure the child dentry is still negative ... 1307 */ 1308 err = nfserr_exist; 1309 if (dchild->d_inode) { 1310 dprintk("nfsd_create: dentry %s/%s not negative!\n", 1311 dentry->d_name.name, dchild->d_name.name); 1312 goto out; 1313 } 1314 1315 if (!(iap->ia_valid & ATTR_MODE)) 1316 iap->ia_mode = 0; 1317 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1318 1319 err = nfserr_inval; 1320 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1321 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1322 type); 1323 goto out; 1324 } 1325 1326 /* 1327 * Get the dir op function pointer. 1328 */ 1329 err = 0; 1330 host_err = 0; 1331 switch (type) { 1332 case S_IFREG: 1333 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1334 if (!host_err) 1335 nfsd_check_ignore_resizing(iap); 1336 break; 1337 case S_IFDIR: 1338 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1339 break; 1340 case S_IFCHR: 1341 case S_IFBLK: 1342 case S_IFIFO: 1343 case S_IFSOCK: 1344 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1345 break; 1346 } 1347 if (host_err < 0) 1348 goto out_nfserr; 1349 1350 err = nfsd_create_setattr(rqstp, resfhp, iap); 1351 1352 /* 1353 * nfsd_setattr already committed the child. Transactional filesystems 1354 * had a chance to commit changes for both parent and child 1355 * simultaneously making the following commit_metadata a noop. 1356 */ 1357 err2 = nfserrno(commit_metadata(fhp)); 1358 if (err2) 1359 err = err2; 1360 /* 1361 * Update the file handle to get the new inode info. 1362 */ 1363 if (!err) 1364 err = fh_update(resfhp); 1365 out: 1366 if (dchild && !IS_ERR(dchild)) 1367 dput(dchild); 1368 return err; 1369 1370 out_nfserr: 1371 err = nfserrno(host_err); 1372 goto out; 1373 } 1374 1375 #ifdef CONFIG_NFSD_V3 1376 1377 static inline int nfsd_create_is_exclusive(int createmode) 1378 { 1379 return createmode == NFS3_CREATE_EXCLUSIVE 1380 || createmode == NFS4_CREATE_EXCLUSIVE4_1; 1381 } 1382 1383 /* 1384 * NFSv3 and NFSv4 version of nfsd_create 1385 */ 1386 __be32 1387 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1388 char *fname, int flen, struct iattr *iap, 1389 struct svc_fh *resfhp, int createmode, u32 *verifier, 1390 bool *truncp, bool *created) 1391 { 1392 struct dentry *dentry, *dchild = NULL; 1393 struct inode *dirp; 1394 __be32 err; 1395 int host_err; 1396 __u32 v_mtime=0, v_atime=0; 1397 1398 err = nfserr_perm; 1399 if (!flen) 1400 goto out; 1401 err = nfserr_exist; 1402 if (isdotent(fname, flen)) 1403 goto out; 1404 if (!(iap->ia_valid & ATTR_MODE)) 1405 iap->ia_mode = 0; 1406 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 1407 if (err) 1408 goto out; 1409 1410 dentry = fhp->fh_dentry; 1411 dirp = dentry->d_inode; 1412 1413 /* Get all the sanity checks out of the way before 1414 * we lock the parent. */ 1415 err = nfserr_notdir; 1416 if (!dirp->i_op->lookup) 1417 goto out; 1418 1419 host_err = fh_want_write(fhp); 1420 if (host_err) 1421 goto out_nfserr; 1422 1423 fh_lock_nested(fhp, I_MUTEX_PARENT); 1424 1425 /* 1426 * Compose the response file handle. 1427 */ 1428 dchild = lookup_one_len(fname, dentry, flen); 1429 host_err = PTR_ERR(dchild); 1430 if (IS_ERR(dchild)) 1431 goto out_nfserr; 1432 1433 /* If file doesn't exist, check for permissions to create one */ 1434 if (!dchild->d_inode) { 1435 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1436 if (err) 1437 goto out; 1438 } 1439 1440 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1441 if (err) 1442 goto out; 1443 1444 if (nfsd_create_is_exclusive(createmode)) { 1445 /* solaris7 gets confused (bugid 4218508) if these have 1446 * the high bit set, so just clear the high bits. If this is 1447 * ever changed to use different attrs for storing the 1448 * verifier, then do_open_lookup() will also need to be fixed 1449 * accordingly. 1450 */ 1451 v_mtime = verifier[0]&0x7fffffff; 1452 v_atime = verifier[1]&0x7fffffff; 1453 } 1454 1455 if (dchild->d_inode) { 1456 err = 0; 1457 1458 switch (createmode) { 1459 case NFS3_CREATE_UNCHECKED: 1460 if (! S_ISREG(dchild->d_inode->i_mode)) 1461 goto out; 1462 else if (truncp) { 1463 /* in nfsv4, we need to treat this case a little 1464 * differently. we don't want to truncate the 1465 * file now; this would be wrong if the OPEN 1466 * fails for some other reason. furthermore, 1467 * if the size is nonzero, we should ignore it 1468 * according to spec! 1469 */ 1470 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1471 } 1472 else { 1473 iap->ia_valid &= ATTR_SIZE; 1474 goto set_attr; 1475 } 1476 break; 1477 case NFS3_CREATE_EXCLUSIVE: 1478 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1479 && dchild->d_inode->i_atime.tv_sec == v_atime 1480 && dchild->d_inode->i_size == 0 ) 1481 break; 1482 case NFS4_CREATE_EXCLUSIVE4_1: 1483 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1484 && dchild->d_inode->i_atime.tv_sec == v_atime 1485 && dchild->d_inode->i_size == 0 ) 1486 goto set_attr; 1487 /* fallthru */ 1488 case NFS3_CREATE_GUARDED: 1489 err = nfserr_exist; 1490 } 1491 fh_drop_write(fhp); 1492 goto out; 1493 } 1494 1495 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1496 if (host_err < 0) { 1497 fh_drop_write(fhp); 1498 goto out_nfserr; 1499 } 1500 if (created) 1501 *created = 1; 1502 1503 nfsd_check_ignore_resizing(iap); 1504 1505 if (nfsd_create_is_exclusive(createmode)) { 1506 /* Cram the verifier into atime/mtime */ 1507 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1508 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1509 /* XXX someone who knows this better please fix it for nsec */ 1510 iap->ia_mtime.tv_sec = v_mtime; 1511 iap->ia_atime.tv_sec = v_atime; 1512 iap->ia_mtime.tv_nsec = 0; 1513 iap->ia_atime.tv_nsec = 0; 1514 } 1515 1516 set_attr: 1517 err = nfsd_create_setattr(rqstp, resfhp, iap); 1518 1519 /* 1520 * nfsd_setattr already committed the child (and possibly also the parent). 1521 */ 1522 if (!err) 1523 err = nfserrno(commit_metadata(fhp)); 1524 1525 /* 1526 * Update the filehandle to get the new inode info. 1527 */ 1528 if (!err) 1529 err = fh_update(resfhp); 1530 1531 out: 1532 fh_unlock(fhp); 1533 if (dchild && !IS_ERR(dchild)) 1534 dput(dchild); 1535 fh_drop_write(fhp); 1536 return err; 1537 1538 out_nfserr: 1539 err = nfserrno(host_err); 1540 goto out; 1541 } 1542 #endif /* CONFIG_NFSD_V3 */ 1543 1544 /* 1545 * Read a symlink. On entry, *lenp must contain the maximum path length that 1546 * fits into the buffer. On return, it contains the true length. 1547 * N.B. After this call fhp needs an fh_put 1548 */ 1549 __be32 1550 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1551 { 1552 struct inode *inode; 1553 mm_segment_t oldfs; 1554 __be32 err; 1555 int host_err; 1556 struct path path; 1557 1558 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1559 if (err) 1560 goto out; 1561 1562 path.mnt = fhp->fh_export->ex_path.mnt; 1563 path.dentry = fhp->fh_dentry; 1564 inode = path.dentry->d_inode; 1565 1566 err = nfserr_inval; 1567 if (!inode->i_op->readlink) 1568 goto out; 1569 1570 touch_atime(&path); 1571 /* N.B. Why does this call need a get_fs()?? 1572 * Remove the set_fs and watch the fireworks:-) --okir 1573 */ 1574 1575 oldfs = get_fs(); set_fs(KERNEL_DS); 1576 host_err = inode->i_op->readlink(path.dentry, buf, *lenp); 1577 set_fs(oldfs); 1578 1579 if (host_err < 0) 1580 goto out_nfserr; 1581 *lenp = host_err; 1582 err = 0; 1583 out: 1584 return err; 1585 1586 out_nfserr: 1587 err = nfserrno(host_err); 1588 goto out; 1589 } 1590 1591 /* 1592 * Create a symlink and look up its inode 1593 * N.B. After this call _both_ fhp and resfhp need an fh_put 1594 */ 1595 __be32 1596 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1597 char *fname, int flen, 1598 char *path, int plen, 1599 struct svc_fh *resfhp, 1600 struct iattr *iap) 1601 { 1602 struct dentry *dentry, *dnew; 1603 __be32 err, cerr; 1604 int host_err; 1605 1606 err = nfserr_noent; 1607 if (!flen || !plen) 1608 goto out; 1609 err = nfserr_exist; 1610 if (isdotent(fname, flen)) 1611 goto out; 1612 1613 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1614 if (err) 1615 goto out; 1616 1617 host_err = fh_want_write(fhp); 1618 if (host_err) 1619 goto out_nfserr; 1620 1621 fh_lock(fhp); 1622 dentry = fhp->fh_dentry; 1623 dnew = lookup_one_len(fname, dentry, flen); 1624 host_err = PTR_ERR(dnew); 1625 if (IS_ERR(dnew)) 1626 goto out_nfserr; 1627 1628 if (unlikely(path[plen] != 0)) { 1629 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1630 if (path_alloced == NULL) 1631 host_err = -ENOMEM; 1632 else { 1633 strncpy(path_alloced, path, plen); 1634 path_alloced[plen] = 0; 1635 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); 1636 kfree(path_alloced); 1637 } 1638 } else 1639 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1640 err = nfserrno(host_err); 1641 if (!err) 1642 err = nfserrno(commit_metadata(fhp)); 1643 fh_unlock(fhp); 1644 1645 fh_drop_write(fhp); 1646 1647 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1648 dput(dnew); 1649 if (err==0) err = cerr; 1650 out: 1651 return err; 1652 1653 out_nfserr: 1654 err = nfserrno(host_err); 1655 goto out; 1656 } 1657 1658 /* 1659 * Create a hardlink 1660 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1661 */ 1662 __be32 1663 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1664 char *name, int len, struct svc_fh *tfhp) 1665 { 1666 struct dentry *ddir, *dnew, *dold; 1667 struct inode *dirp; 1668 __be32 err; 1669 int host_err; 1670 1671 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1672 if (err) 1673 goto out; 1674 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP); 1675 if (err) 1676 goto out; 1677 err = nfserr_isdir; 1678 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode)) 1679 goto out; 1680 err = nfserr_perm; 1681 if (!len) 1682 goto out; 1683 err = nfserr_exist; 1684 if (isdotent(name, len)) 1685 goto out; 1686 1687 host_err = fh_want_write(tfhp); 1688 if (host_err) { 1689 err = nfserrno(host_err); 1690 goto out; 1691 } 1692 1693 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1694 ddir = ffhp->fh_dentry; 1695 dirp = ddir->d_inode; 1696 1697 dnew = lookup_one_len(name, ddir, len); 1698 host_err = PTR_ERR(dnew); 1699 if (IS_ERR(dnew)) 1700 goto out_nfserr; 1701 1702 dold = tfhp->fh_dentry; 1703 1704 err = nfserr_noent; 1705 if (!dold->d_inode) 1706 goto out_dput; 1707 host_err = nfsd_break_lease(dold->d_inode); 1708 if (host_err) { 1709 err = nfserrno(host_err); 1710 goto out_dput; 1711 } 1712 host_err = vfs_link(dold, dirp, dnew); 1713 if (!host_err) { 1714 err = nfserrno(commit_metadata(ffhp)); 1715 if (!err) 1716 err = nfserrno(commit_metadata(tfhp)); 1717 } else { 1718 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1719 err = nfserr_acces; 1720 else 1721 err = nfserrno(host_err); 1722 } 1723 out_dput: 1724 dput(dnew); 1725 out_unlock: 1726 fh_unlock(ffhp); 1727 fh_drop_write(tfhp); 1728 out: 1729 return err; 1730 1731 out_nfserr: 1732 err = nfserrno(host_err); 1733 goto out_unlock; 1734 } 1735 1736 /* 1737 * Rename a file 1738 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1739 */ 1740 __be32 1741 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1742 struct svc_fh *tfhp, char *tname, int tlen) 1743 { 1744 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1745 struct inode *fdir, *tdir; 1746 __be32 err; 1747 int host_err; 1748 1749 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1750 if (err) 1751 goto out; 1752 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1753 if (err) 1754 goto out; 1755 1756 fdentry = ffhp->fh_dentry; 1757 fdir = fdentry->d_inode; 1758 1759 tdentry = tfhp->fh_dentry; 1760 tdir = tdentry->d_inode; 1761 1762 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; 1763 if (ffhp->fh_export != tfhp->fh_export) 1764 goto out; 1765 1766 err = nfserr_perm; 1767 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1768 goto out; 1769 1770 host_err = fh_want_write(ffhp); 1771 if (host_err) { 1772 err = nfserrno(host_err); 1773 goto out; 1774 } 1775 1776 /* cannot use fh_lock as we need deadlock protective ordering 1777 * so do it by hand */ 1778 trap = lock_rename(tdentry, fdentry); 1779 ffhp->fh_locked = tfhp->fh_locked = 1; 1780 fill_pre_wcc(ffhp); 1781 fill_pre_wcc(tfhp); 1782 1783 odentry = lookup_one_len(fname, fdentry, flen); 1784 host_err = PTR_ERR(odentry); 1785 if (IS_ERR(odentry)) 1786 goto out_nfserr; 1787 1788 host_err = -ENOENT; 1789 if (!odentry->d_inode) 1790 goto out_dput_old; 1791 host_err = -EINVAL; 1792 if (odentry == trap) 1793 goto out_dput_old; 1794 1795 ndentry = lookup_one_len(tname, tdentry, tlen); 1796 host_err = PTR_ERR(ndentry); 1797 if (IS_ERR(ndentry)) 1798 goto out_dput_old; 1799 host_err = -ENOTEMPTY; 1800 if (ndentry == trap) 1801 goto out_dput_new; 1802 1803 host_err = -EXDEV; 1804 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1805 goto out_dput_new; 1806 1807 host_err = nfsd_break_lease(odentry->d_inode); 1808 if (host_err) 1809 goto out_dput_new; 1810 if (ndentry->d_inode) { 1811 host_err = nfsd_break_lease(ndentry->d_inode); 1812 if (host_err) 1813 goto out_dput_new; 1814 } 1815 host_err = vfs_rename(fdir, odentry, tdir, ndentry); 1816 if (!host_err) { 1817 host_err = commit_metadata(tfhp); 1818 if (!host_err) 1819 host_err = commit_metadata(ffhp); 1820 } 1821 out_dput_new: 1822 dput(ndentry); 1823 out_dput_old: 1824 dput(odentry); 1825 out_nfserr: 1826 err = nfserrno(host_err); 1827 1828 /* we cannot reply on fh_unlock on the two filehandles, 1829 * as that would do the wrong thing if the two directories 1830 * were the same, so again we do it by hand 1831 */ 1832 fill_post_wcc(ffhp); 1833 fill_post_wcc(tfhp); 1834 unlock_rename(tdentry, fdentry); 1835 ffhp->fh_locked = tfhp->fh_locked = 0; 1836 fh_drop_write(ffhp); 1837 1838 out: 1839 return err; 1840 } 1841 1842 /* 1843 * Unlink a file or directory 1844 * N.B. After this call fhp needs an fh_put 1845 */ 1846 __be32 1847 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1848 char *fname, int flen) 1849 { 1850 struct dentry *dentry, *rdentry; 1851 struct inode *dirp; 1852 __be32 err; 1853 int host_err; 1854 1855 err = nfserr_acces; 1856 if (!flen || isdotent(fname, flen)) 1857 goto out; 1858 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1859 if (err) 1860 goto out; 1861 1862 host_err = fh_want_write(fhp); 1863 if (host_err) 1864 goto out_nfserr; 1865 1866 fh_lock_nested(fhp, I_MUTEX_PARENT); 1867 dentry = fhp->fh_dentry; 1868 dirp = dentry->d_inode; 1869 1870 rdentry = lookup_one_len(fname, dentry, flen); 1871 host_err = PTR_ERR(rdentry); 1872 if (IS_ERR(rdentry)) 1873 goto out_nfserr; 1874 1875 if (!rdentry->d_inode) { 1876 dput(rdentry); 1877 err = nfserr_noent; 1878 goto out; 1879 } 1880 1881 if (!type) 1882 type = rdentry->d_inode->i_mode & S_IFMT; 1883 1884 host_err = nfsd_break_lease(rdentry->d_inode); 1885 if (host_err) 1886 goto out_put; 1887 if (type != S_IFDIR) 1888 host_err = vfs_unlink(dirp, rdentry); 1889 else 1890 host_err = vfs_rmdir(dirp, rdentry); 1891 if (!host_err) 1892 host_err = commit_metadata(fhp); 1893 out_put: 1894 dput(rdentry); 1895 1896 out_nfserr: 1897 err = nfserrno(host_err); 1898 out: 1899 return err; 1900 } 1901 1902 /* 1903 * We do this buffering because we must not call back into the file 1904 * system's ->lookup() method from the filldir callback. That may well 1905 * deadlock a number of file systems. 1906 * 1907 * This is based heavily on the implementation of same in XFS. 1908 */ 1909 struct buffered_dirent { 1910 u64 ino; 1911 loff_t offset; 1912 int namlen; 1913 unsigned int d_type; 1914 char name[]; 1915 }; 1916 1917 struct readdir_data { 1918 char *dirent; 1919 size_t used; 1920 int full; 1921 }; 1922 1923 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1924 loff_t offset, u64 ino, unsigned int d_type) 1925 { 1926 struct readdir_data *buf = __buf; 1927 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1928 unsigned int reclen; 1929 1930 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1931 if (buf->used + reclen > PAGE_SIZE) { 1932 buf->full = 1; 1933 return -EINVAL; 1934 } 1935 1936 de->namlen = namlen; 1937 de->offset = offset; 1938 de->ino = ino; 1939 de->d_type = d_type; 1940 memcpy(de->name, name, namlen); 1941 buf->used += reclen; 1942 1943 return 0; 1944 } 1945 1946 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1947 struct readdir_cd *cdp, loff_t *offsetp) 1948 { 1949 struct readdir_data buf; 1950 struct buffered_dirent *de; 1951 int host_err; 1952 int size; 1953 loff_t offset; 1954 1955 buf.dirent = (void *)__get_free_page(GFP_KERNEL); 1956 if (!buf.dirent) 1957 return nfserrno(-ENOMEM); 1958 1959 offset = *offsetp; 1960 1961 while (1) { 1962 struct inode *dir_inode = file->f_path.dentry->d_inode; 1963 unsigned int reclen; 1964 1965 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1966 buf.used = 0; 1967 buf.full = 0; 1968 1969 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf); 1970 if (buf.full) 1971 host_err = 0; 1972 1973 if (host_err < 0) 1974 break; 1975 1976 size = buf.used; 1977 1978 if (!size) 1979 break; 1980 1981 /* 1982 * Various filldir functions may end up calling back into 1983 * lookup_one_len() and the file system's ->lookup() method. 1984 * These expect i_mutex to be held, as it would within readdir. 1985 */ 1986 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1987 if (host_err) 1988 break; 1989 1990 de = (struct buffered_dirent *)buf.dirent; 1991 while (size > 0) { 1992 offset = de->offset; 1993 1994 if (func(cdp, de->name, de->namlen, de->offset, 1995 de->ino, de->d_type)) 1996 break; 1997 1998 if (cdp->err != nfs_ok) 1999 break; 2000 2001 reclen = ALIGN(sizeof(*de) + de->namlen, 2002 sizeof(u64)); 2003 size -= reclen; 2004 de = (struct buffered_dirent *)((char *)de + reclen); 2005 } 2006 mutex_unlock(&dir_inode->i_mutex); 2007 if (size > 0) /* We bailed out early */ 2008 break; 2009 2010 offset = vfs_llseek(file, 0, SEEK_CUR); 2011 } 2012 2013 free_page((unsigned long)(buf.dirent)); 2014 2015 if (host_err) 2016 return nfserrno(host_err); 2017 2018 *offsetp = offset; 2019 return cdp->err; 2020 } 2021 2022 /* 2023 * Read entries from a directory. 2024 * The NFSv3/4 verifier we ignore for now. 2025 */ 2026 __be32 2027 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 2028 struct readdir_cd *cdp, filldir_t func) 2029 { 2030 __be32 err; 2031 struct file *file; 2032 loff_t offset = *offsetp; 2033 int may_flags = NFSD_MAY_READ; 2034 2035 /* NFSv2 only supports 32 bit cookies */ 2036 if (rqstp->rq_vers > 2) 2037 may_flags |= NFSD_MAY_64BIT_COOKIE; 2038 2039 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file); 2040 if (err) 2041 goto out; 2042 2043 offset = vfs_llseek(file, offset, SEEK_SET); 2044 if (offset < 0) { 2045 err = nfserrno((int)offset); 2046 goto out_close; 2047 } 2048 2049 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 2050 2051 if (err == nfserr_eof || err == nfserr_toosmall) 2052 err = nfs_ok; /* can still be found in ->err */ 2053 out_close: 2054 nfsd_close(file); 2055 out: 2056 return err; 2057 } 2058 2059 /* 2060 * Get file system stats 2061 * N.B. After this call fhp needs an fh_put 2062 */ 2063 __be32 2064 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 2065 { 2066 __be32 err; 2067 2068 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 2069 if (!err) { 2070 struct path path = { 2071 .mnt = fhp->fh_export->ex_path.mnt, 2072 .dentry = fhp->fh_dentry, 2073 }; 2074 if (vfs_statfs(&path, stat)) 2075 err = nfserr_io; 2076 } 2077 return err; 2078 } 2079 2080 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 2081 { 2082 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 2083 } 2084 2085 /* 2086 * Check for a user's access permissions to this inode. 2087 */ 2088 __be32 2089 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 2090 struct dentry *dentry, int acc) 2091 { 2092 struct inode *inode = dentry->d_inode; 2093 int err; 2094 2095 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP) 2096 return 0; 2097 #if 0 2098 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 2099 acc, 2100 (acc & NFSD_MAY_READ)? " read" : "", 2101 (acc & NFSD_MAY_WRITE)? " write" : "", 2102 (acc & NFSD_MAY_EXEC)? " exec" : "", 2103 (acc & NFSD_MAY_SATTR)? " sattr" : "", 2104 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 2105 (acc & NFSD_MAY_LOCK)? " lock" : "", 2106 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 2107 inode->i_mode, 2108 IS_IMMUTABLE(inode)? " immut" : "", 2109 IS_APPEND(inode)? " append" : "", 2110 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2111 dprintk(" owner %d/%d user %d/%d\n", 2112 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2113 #endif 2114 2115 /* Normally we reject any write/sattr etc access on a read-only file 2116 * system. But if it is IRIX doing check on write-access for a 2117 * device special file, we ignore rofs. 2118 */ 2119 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2120 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2121 if (exp_rdonly(rqstp, exp) || 2122 __mnt_is_readonly(exp->ex_path.mnt)) 2123 return nfserr_rofs; 2124 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2125 return nfserr_perm; 2126 } 2127 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2128 return nfserr_perm; 2129 2130 if (acc & NFSD_MAY_LOCK) { 2131 /* If we cannot rely on authentication in NLM requests, 2132 * just allow locks, otherwise require read permission, or 2133 * ownership 2134 */ 2135 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2136 return 0; 2137 else 2138 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2139 } 2140 /* 2141 * The file owner always gets access permission for accesses that 2142 * would normally be checked at open time. This is to make 2143 * file access work even when the client has done a fchmod(fd, 0). 2144 * 2145 * However, `cp foo bar' should fail nevertheless when bar is 2146 * readonly. A sensible way to do this might be to reject all 2147 * attempts to truncate a read-only file, because a creat() call 2148 * always implies file truncation. 2149 * ... but this isn't really fair. A process may reasonably call 2150 * ftruncate on an open file descriptor on a file with perm 000. 2151 * We must trust the client to do permission checking - using "ACCESS" 2152 * with NFSv3. 2153 */ 2154 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2155 inode->i_uid == current_fsuid()) 2156 return 0; 2157 2158 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2159 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2160 2161 /* Allow read access to binaries even when mode 111 */ 2162 if (err == -EACCES && S_ISREG(inode->i_mode) && 2163 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) || 2164 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC))) 2165 err = inode_permission(inode, MAY_EXEC); 2166 2167 return err? nfserrno(err) : 0; 2168 } 2169 2170 void 2171 nfsd_racache_shutdown(void) 2172 { 2173 struct raparms *raparm, *last_raparm; 2174 unsigned int i; 2175 2176 dprintk("nfsd: freeing readahead buffers.\n"); 2177 2178 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2179 raparm = raparm_hash[i].pb_head; 2180 while(raparm) { 2181 last_raparm = raparm; 2182 raparm = raparm->p_next; 2183 kfree(last_raparm); 2184 } 2185 raparm_hash[i].pb_head = NULL; 2186 } 2187 } 2188 /* 2189 * Initialize readahead param cache 2190 */ 2191 int 2192 nfsd_racache_init(int cache_size) 2193 { 2194 int i; 2195 int j = 0; 2196 int nperbucket; 2197 struct raparms **raparm = NULL; 2198 2199 2200 if (raparm_hash[0].pb_head) 2201 return 0; 2202 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2203 if (nperbucket < 2) 2204 nperbucket = 2; 2205 cache_size = nperbucket * RAPARM_HASH_SIZE; 2206 2207 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2208 2209 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2210 spin_lock_init(&raparm_hash[i].pb_lock); 2211 2212 raparm = &raparm_hash[i].pb_head; 2213 for (j = 0; j < nperbucket; j++) { 2214 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2215 if (!*raparm) 2216 goto out_nomem; 2217 raparm = &(*raparm)->p_next; 2218 } 2219 *raparm = NULL; 2220 } 2221 2222 nfsdstats.ra_size = cache_size; 2223 return 0; 2224 2225 out_nomem: 2226 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2227 nfsd_racache_shutdown(); 2228 return -ENOMEM; 2229 } 2230 2231 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 2232 struct posix_acl * 2233 nfsd_get_posix_acl(struct svc_fh *fhp, int type) 2234 { 2235 struct inode *inode = fhp->fh_dentry->d_inode; 2236 char *name; 2237 void *value = NULL; 2238 ssize_t size; 2239 struct posix_acl *acl; 2240 2241 if (!IS_POSIXACL(inode)) 2242 return ERR_PTR(-EOPNOTSUPP); 2243 2244 switch (type) { 2245 case ACL_TYPE_ACCESS: 2246 name = POSIX_ACL_XATTR_ACCESS; 2247 break; 2248 case ACL_TYPE_DEFAULT: 2249 name = POSIX_ACL_XATTR_DEFAULT; 2250 break; 2251 default: 2252 return ERR_PTR(-EOPNOTSUPP); 2253 } 2254 2255 size = nfsd_getxattr(fhp->fh_dentry, name, &value); 2256 if (size < 0) 2257 return ERR_PTR(size); 2258 2259 acl = posix_acl_from_xattr(value, size); 2260 kfree(value); 2261 return acl; 2262 } 2263 2264 int 2265 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) 2266 { 2267 struct inode *inode = fhp->fh_dentry->d_inode; 2268 char *name; 2269 void *value = NULL; 2270 size_t size; 2271 int error; 2272 2273 if (!IS_POSIXACL(inode) || 2274 !inode->i_op->setxattr || !inode->i_op->removexattr) 2275 return -EOPNOTSUPP; 2276 switch(type) { 2277 case ACL_TYPE_ACCESS: 2278 name = POSIX_ACL_XATTR_ACCESS; 2279 break; 2280 case ACL_TYPE_DEFAULT: 2281 name = POSIX_ACL_XATTR_DEFAULT; 2282 break; 2283 default: 2284 return -EOPNOTSUPP; 2285 } 2286 2287 if (acl && acl->a_count) { 2288 size = posix_acl_xattr_size(acl->a_count); 2289 value = kmalloc(size, GFP_KERNEL); 2290 if (!value) 2291 return -ENOMEM; 2292 error = posix_acl_to_xattr(acl, value, size); 2293 if (error < 0) 2294 goto getout; 2295 size = error; 2296 } else 2297 size = 0; 2298 2299 error = fh_want_write(fhp); 2300 if (error) 2301 goto getout; 2302 if (size) 2303 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); 2304 else { 2305 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) 2306 error = 0; 2307 else { 2308 error = vfs_removexattr(fhp->fh_dentry, name); 2309 if (error == -ENODATA) 2310 error = 0; 2311 } 2312 } 2313 fh_drop_write(fhp); 2314 2315 getout: 2316 kfree(value); 2317 return error; 2318 } 2319 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 2320