1 #define MSNFS /* HACK HACK */ 2 /* 3 * File operations used by nfsd. Some of these have been ripped from 4 * other parts of the kernel because they weren't exported, others 5 * are partial duplicates with added or changed functionality. 6 * 7 * Note that several functions dget() the dentry upon which they want 8 * to act, most notably those that create directory entries. Response 9 * dentry's are dput()'d if necessary in the release callback. 10 * So if you notice code paths that apparently fail to dput() the 11 * dentry, don't worry--they have been taken care of. 12 * 13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de> 14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp> 15 */ 16 17 #include <linux/fs.h> 18 #include <linux/file.h> 19 #include <linux/splice.h> 20 #include <linux/fcntl.h> 21 #include <linux/namei.h> 22 #include <linux/delay.h> 23 #include <linux/quotaops.h> 24 #include <linux/fsnotify.h> 25 #include <linux/posix_acl_xattr.h> 26 #include <linux/xattr.h> 27 #include <linux/jhash.h> 28 #include <linux/ima.h> 29 #include <asm/uaccess.h> 30 31 #ifdef CONFIG_NFSD_V3 32 #include "xdr3.h" 33 #endif /* CONFIG_NFSD_V3 */ 34 35 #ifdef CONFIG_NFSD_V4 36 #include <linux/nfs4_acl.h> 37 #include <linux/nfsd_idmap.h> 38 #endif /* CONFIG_NFSD_V4 */ 39 40 #include "nfsd.h" 41 #include "vfs.h" 42 43 #define NFSDDBG_FACILITY NFSDDBG_FILEOP 44 45 46 /* 47 * This is a cache of readahead params that help us choose the proper 48 * readahead strategy. Initially, we set all readahead parameters to 0 49 * and let the VFS handle things. 50 * If you increase the number of cached files very much, you'll need to 51 * add a hash table here. 52 */ 53 struct raparms { 54 struct raparms *p_next; 55 unsigned int p_count; 56 ino_t p_ino; 57 dev_t p_dev; 58 int p_set; 59 struct file_ra_state p_ra; 60 unsigned int p_hindex; 61 }; 62 63 struct raparm_hbucket { 64 struct raparms *pb_head; 65 spinlock_t pb_lock; 66 } ____cacheline_aligned_in_smp; 67 68 #define RAPARM_HASH_BITS 4 69 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) 70 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) 71 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; 72 73 /* 74 * Called from nfsd_lookup and encode_dirent. Check if we have crossed 75 * a mount point. 76 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged, 77 * or nfs_ok having possibly changed *dpp and *expp 78 */ 79 int 80 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 81 struct svc_export **expp) 82 { 83 struct svc_export *exp = *expp, *exp2 = NULL; 84 struct dentry *dentry = *dpp; 85 struct path path = {.mnt = mntget(exp->ex_path.mnt), 86 .dentry = dget(dentry)}; 87 int err = 0; 88 89 while (d_mountpoint(path.dentry) && follow_down(&path)) 90 ; 91 92 exp2 = rqst_exp_get_by_name(rqstp, &path); 93 if (IS_ERR(exp2)) { 94 err = PTR_ERR(exp2); 95 /* 96 * We normally allow NFS clients to continue 97 * "underneath" a mountpoint that is not exported. 98 * The exception is V4ROOT, where no traversal is ever 99 * allowed without an explicit export of the new 100 * directory. 101 */ 102 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT)) 103 err = 0; 104 path_put(&path); 105 goto out; 106 } 107 if (nfsd_v4client(rqstp) || 108 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) { 109 /* successfully crossed mount point */ 110 /* 111 * This is subtle: path.dentry is *not* on path.mnt 112 * at this point. The only reason we are safe is that 113 * original mnt is pinned down by exp, so we should 114 * put path *before* putting exp 115 */ 116 *dpp = path.dentry; 117 path.dentry = dentry; 118 *expp = exp2; 119 exp2 = exp; 120 } 121 path_put(&path); 122 exp_put(exp2); 123 out: 124 return err; 125 } 126 127 static void follow_to_parent(struct path *path) 128 { 129 struct dentry *dp; 130 131 while (path->dentry == path->mnt->mnt_root && follow_up(path)) 132 ; 133 dp = dget_parent(path->dentry); 134 dput(path->dentry); 135 path->dentry = dp; 136 } 137 138 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp) 139 { 140 struct svc_export *exp2; 141 struct path path = {.mnt = mntget((*exp)->ex_path.mnt), 142 .dentry = dget(dparent)}; 143 144 follow_to_parent(&path); 145 146 exp2 = rqst_exp_parent(rqstp, &path); 147 if (PTR_ERR(exp2) == -ENOENT) { 148 *dentryp = dget(dparent); 149 } else if (IS_ERR(exp2)) { 150 path_put(&path); 151 return PTR_ERR(exp2); 152 } else { 153 *dentryp = dget(path.dentry); 154 exp_put(*exp); 155 *exp = exp2; 156 } 157 path_put(&path); 158 return 0; 159 } 160 161 /* 162 * For nfsd purposes, we treat V4ROOT exports as though there was an 163 * export at *every* directory. 164 */ 165 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp) 166 { 167 if (d_mountpoint(dentry)) 168 return 1; 169 if (!(exp->ex_flags & NFSEXP_V4ROOT)) 170 return 0; 171 return dentry->d_inode != NULL; 172 } 173 174 __be32 175 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp, 176 const char *name, unsigned int len, 177 struct svc_export **exp_ret, struct dentry **dentry_ret) 178 { 179 struct svc_export *exp; 180 struct dentry *dparent; 181 struct dentry *dentry; 182 __be32 err; 183 int host_err; 184 185 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); 186 187 /* Obtain dentry and export. */ 188 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 189 if (err) 190 return err; 191 192 dparent = fhp->fh_dentry; 193 exp = fhp->fh_export; 194 exp_get(exp); 195 196 /* Lookup the name, but don't follow links */ 197 if (isdotent(name, len)) { 198 if (len==1) 199 dentry = dget(dparent); 200 else if (dparent != exp->ex_path.dentry) 201 dentry = dget_parent(dparent); 202 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp)) 203 dentry = dget(dparent); /* .. == . just like at / */ 204 else { 205 /* checking mountpoint crossing is very different when stepping up */ 206 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry); 207 if (host_err) 208 goto out_nfserr; 209 } 210 } else { 211 fh_lock(fhp); 212 dentry = lookup_one_len(name, dparent, len); 213 host_err = PTR_ERR(dentry); 214 if (IS_ERR(dentry)) 215 goto out_nfserr; 216 /* 217 * check if we have crossed a mount point ... 218 */ 219 if (nfsd_mountpoint(dentry, exp)) { 220 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { 221 dput(dentry); 222 goto out_nfserr; 223 } 224 } 225 } 226 *dentry_ret = dentry; 227 *exp_ret = exp; 228 return 0; 229 230 out_nfserr: 231 exp_put(exp); 232 return nfserrno(host_err); 233 } 234 235 /* 236 * Look up one component of a pathname. 237 * N.B. After this call _both_ fhp and resfh need an fh_put 238 * 239 * If the lookup would cross a mountpoint, and the mounted filesystem 240 * is exported to the client with NFSEXP_NOHIDE, then the lookup is 241 * accepted as it stands and the mounted directory is 242 * returned. Otherwise the covered directory is returned. 243 * NOTE: this mountpoint crossing is not supported properly by all 244 * clients and is explicitly disallowed for NFSv3 245 * NeilBrown <neilb@cse.unsw.edu.au> 246 */ 247 __be32 248 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, 249 unsigned int len, struct svc_fh *resfh) 250 { 251 struct svc_export *exp; 252 struct dentry *dentry; 253 __be32 err; 254 255 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry); 256 if (err) 257 return err; 258 err = check_nfsd_access(exp, rqstp); 259 if (err) 260 goto out; 261 /* 262 * Note: we compose the file handle now, but as the 263 * dentry may be negative, it may need to be updated. 264 */ 265 err = fh_compose(resfh, exp, dentry, fhp); 266 if (!err && !dentry->d_inode) 267 err = nfserr_noent; 268 out: 269 dput(dentry); 270 exp_put(exp); 271 return err; 272 } 273 274 275 /* 276 * Set various file attributes. 277 * N.B. After this call fhp needs an fh_put 278 */ 279 __be32 280 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, 281 int check_guard, time_t guardtime) 282 { 283 struct dentry *dentry; 284 struct inode *inode; 285 int accmode = NFSD_MAY_SATTR; 286 int ftype = 0; 287 __be32 err; 288 int host_err; 289 int size_change = 0; 290 291 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) 292 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE; 293 if (iap->ia_valid & ATTR_SIZE) 294 ftype = S_IFREG; 295 296 /* Get inode */ 297 err = fh_verify(rqstp, fhp, ftype, accmode); 298 if (err) 299 goto out; 300 301 dentry = fhp->fh_dentry; 302 inode = dentry->d_inode; 303 304 /* Ignore any mode updates on symlinks */ 305 if (S_ISLNK(inode->i_mode)) 306 iap->ia_valid &= ~ATTR_MODE; 307 308 if (!iap->ia_valid) 309 goto out; 310 311 /* 312 * NFSv2 does not differentiate between "set-[ac]time-to-now" 313 * which only requires access, and "set-[ac]time-to-X" which 314 * requires ownership. 315 * So if it looks like it might be "set both to the same time which 316 * is close to now", and if inode_change_ok fails, then we 317 * convert to "set to now" instead of "set to explicit time" 318 * 319 * We only call inode_change_ok as the last test as technically 320 * it is not an interface that we should be using. It is only 321 * valid if the filesystem does not define it's own i_op->setattr. 322 */ 323 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) 324 #define MAX_TOUCH_TIME_ERROR (30*60) 325 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET && 326 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) { 327 /* 328 * Looks probable. 329 * 330 * Now just make sure time is in the right ballpark. 331 * Solaris, at least, doesn't seem to care what the time 332 * request is. We require it be within 30 minutes of now. 333 */ 334 time_t delta = iap->ia_atime.tv_sec - get_seconds(); 335 if (delta < 0) 336 delta = -delta; 337 if (delta < MAX_TOUCH_TIME_ERROR && 338 inode_change_ok(inode, iap) != 0) { 339 /* 340 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME. 341 * This will cause notify_change to set these times 342 * to "now" 343 */ 344 iap->ia_valid &= ~BOTH_TIME_SET; 345 } 346 } 347 348 /* 349 * The size case is special. 350 * It changes the file as well as the attributes. 351 */ 352 if (iap->ia_valid & ATTR_SIZE) { 353 if (iap->ia_size < inode->i_size) { 354 err = nfsd_permission(rqstp, fhp->fh_export, dentry, 355 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE); 356 if (err) 357 goto out; 358 } 359 360 /* 361 * If we are changing the size of the file, then 362 * we need to break all leases. 363 */ 364 host_err = break_lease(inode, O_WRONLY | O_NONBLOCK); 365 if (host_err == -EWOULDBLOCK) 366 host_err = -ETIMEDOUT; 367 if (host_err) /* ENOMEM or EWOULDBLOCK */ 368 goto out_nfserr; 369 370 host_err = get_write_access(inode); 371 if (host_err) 372 goto out_nfserr; 373 374 size_change = 1; 375 host_err = locks_verify_truncate(inode, NULL, iap->ia_size); 376 if (host_err) { 377 put_write_access(inode); 378 goto out_nfserr; 379 } 380 vfs_dq_init(inode); 381 } 382 383 /* sanitize the mode change */ 384 if (iap->ia_valid & ATTR_MODE) { 385 iap->ia_mode &= S_IALLUGO; 386 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO); 387 } 388 389 /* Revoke setuid/setgid on chown */ 390 if (!S_ISDIR(inode->i_mode) && 391 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) || 392 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) { 393 iap->ia_valid |= ATTR_KILL_PRIV; 394 if (iap->ia_valid & ATTR_MODE) { 395 /* we're setting mode too, just clear the s*id bits */ 396 iap->ia_mode &= ~S_ISUID; 397 if (iap->ia_mode & S_IXGRP) 398 iap->ia_mode &= ~S_ISGID; 399 } else { 400 /* set ATTR_KILL_* bits and let VFS handle it */ 401 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID); 402 } 403 } 404 405 /* Change the attributes. */ 406 407 iap->ia_valid |= ATTR_CTIME; 408 409 err = nfserr_notsync; 410 if (!check_guard || guardtime == inode->i_ctime.tv_sec) { 411 fh_lock(fhp); 412 host_err = notify_change(dentry, iap); 413 err = nfserrno(host_err); 414 fh_unlock(fhp); 415 } 416 if (size_change) 417 put_write_access(inode); 418 if (!err) 419 if (EX_ISSYNC(fhp->fh_export)) 420 write_inode_now(inode, 1); 421 out: 422 return err; 423 424 out_nfserr: 425 err = nfserrno(host_err); 426 goto out; 427 } 428 429 #if defined(CONFIG_NFSD_V2_ACL) || \ 430 defined(CONFIG_NFSD_V3_ACL) || \ 431 defined(CONFIG_NFSD_V4) 432 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf) 433 { 434 ssize_t buflen; 435 ssize_t ret; 436 437 buflen = vfs_getxattr(dentry, key, NULL, 0); 438 if (buflen <= 0) 439 return buflen; 440 441 *buf = kmalloc(buflen, GFP_KERNEL); 442 if (!*buf) 443 return -ENOMEM; 444 445 ret = vfs_getxattr(dentry, key, *buf, buflen); 446 if (ret < 0) 447 kfree(*buf); 448 return ret; 449 } 450 #endif 451 452 #if defined(CONFIG_NFSD_V4) 453 static int 454 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) 455 { 456 int len; 457 size_t buflen; 458 char *buf = NULL; 459 int error = 0; 460 461 buflen = posix_acl_xattr_size(pacl->a_count); 462 buf = kmalloc(buflen, GFP_KERNEL); 463 error = -ENOMEM; 464 if (buf == NULL) 465 goto out; 466 467 len = posix_acl_to_xattr(pacl, buf, buflen); 468 if (len < 0) { 469 error = len; 470 goto out; 471 } 472 473 error = vfs_setxattr(dentry, key, buf, len, 0); 474 out: 475 kfree(buf); 476 return error; 477 } 478 479 __be32 480 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, 481 struct nfs4_acl *acl) 482 { 483 __be32 error; 484 int host_error; 485 struct dentry *dentry; 486 struct inode *inode; 487 struct posix_acl *pacl = NULL, *dpacl = NULL; 488 unsigned int flags = 0; 489 490 /* Get inode */ 491 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR); 492 if (error) 493 return error; 494 495 dentry = fhp->fh_dentry; 496 inode = dentry->d_inode; 497 if (S_ISDIR(inode->i_mode)) 498 flags = NFS4_ACL_DIR; 499 500 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); 501 if (host_error == -EINVAL) { 502 return nfserr_attrnotsupp; 503 } else if (host_error < 0) 504 goto out_nfserr; 505 506 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); 507 if (host_error < 0) 508 goto out_release; 509 510 if (S_ISDIR(inode->i_mode)) 511 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); 512 513 out_release: 514 posix_acl_release(pacl); 515 posix_acl_release(dpacl); 516 out_nfserr: 517 if (host_error == -EOPNOTSUPP) 518 return nfserr_attrnotsupp; 519 else 520 return nfserrno(host_error); 521 } 522 523 static struct posix_acl * 524 _get_posix_acl(struct dentry *dentry, char *key) 525 { 526 void *buf = NULL; 527 struct posix_acl *pacl = NULL; 528 int buflen; 529 530 buflen = nfsd_getxattr(dentry, key, &buf); 531 if (!buflen) 532 buflen = -ENODATA; 533 if (buflen <= 0) 534 return ERR_PTR(buflen); 535 536 pacl = posix_acl_from_xattr(buf, buflen); 537 kfree(buf); 538 return pacl; 539 } 540 541 int 542 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) 543 { 544 struct inode *inode = dentry->d_inode; 545 int error = 0; 546 struct posix_acl *pacl = NULL, *dpacl = NULL; 547 unsigned int flags = 0; 548 549 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); 550 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) 551 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); 552 if (IS_ERR(pacl)) { 553 error = PTR_ERR(pacl); 554 pacl = NULL; 555 goto out; 556 } 557 558 if (S_ISDIR(inode->i_mode)) { 559 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); 560 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) 561 dpacl = NULL; 562 else if (IS_ERR(dpacl)) { 563 error = PTR_ERR(dpacl); 564 dpacl = NULL; 565 goto out; 566 } 567 flags = NFS4_ACL_DIR; 568 } 569 570 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); 571 if (IS_ERR(*acl)) { 572 error = PTR_ERR(*acl); 573 *acl = NULL; 574 } 575 out: 576 posix_acl_release(pacl); 577 posix_acl_release(dpacl); 578 return error; 579 } 580 581 #endif /* defined(CONFIG_NFS_V4) */ 582 583 #ifdef CONFIG_NFSD_V3 584 /* 585 * Check server access rights to a file system object 586 */ 587 struct accessmap { 588 u32 access; 589 int how; 590 }; 591 static struct accessmap nfs3_regaccess[] = { 592 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 593 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 594 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, 595 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, 596 597 { 0, 0 } 598 }; 599 600 static struct accessmap nfs3_diraccess[] = { 601 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 602 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, 603 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, 604 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, 605 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, 606 607 { 0, 0 } 608 }; 609 610 static struct accessmap nfs3_anyaccess[] = { 611 /* Some clients - Solaris 2.6 at least, make an access call 612 * to the server to check for access for things like /dev/null 613 * (which really, the server doesn't care about). So 614 * We provide simple access checking for them, looking 615 * mainly at mode bits, and we make sure to ignore read-only 616 * filesystem checks 617 */ 618 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 619 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 620 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 621 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 622 623 { 0, 0 } 624 }; 625 626 __be32 627 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) 628 { 629 struct accessmap *map; 630 struct svc_export *export; 631 struct dentry *dentry; 632 u32 query, result = 0, sresult = 0; 633 __be32 error; 634 635 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); 636 if (error) 637 goto out; 638 639 export = fhp->fh_export; 640 dentry = fhp->fh_dentry; 641 642 if (S_ISREG(dentry->d_inode->i_mode)) 643 map = nfs3_regaccess; 644 else if (S_ISDIR(dentry->d_inode->i_mode)) 645 map = nfs3_diraccess; 646 else 647 map = nfs3_anyaccess; 648 649 650 query = *access; 651 for (; map->access; map++) { 652 if (map->access & query) { 653 __be32 err2; 654 655 sresult |= map->access; 656 657 err2 = nfsd_permission(rqstp, export, dentry, map->how); 658 switch (err2) { 659 case nfs_ok: 660 result |= map->access; 661 break; 662 663 /* the following error codes just mean the access was not allowed, 664 * rather than an error occurred */ 665 case nfserr_rofs: 666 case nfserr_acces: 667 case nfserr_perm: 668 /* simply don't "or" in the access bit. */ 669 break; 670 default: 671 error = err2; 672 goto out; 673 } 674 } 675 } 676 *access = result; 677 if (supported) 678 *supported = sresult; 679 680 out: 681 return error; 682 } 683 #endif /* CONFIG_NFSD_V3 */ 684 685 686 687 /* 688 * Open an existing file or directory. 689 * The access argument indicates the type of open (read/write/lock) 690 * N.B. After this call fhp needs an fh_put 691 */ 692 __be32 693 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 694 int access, struct file **filp) 695 { 696 struct dentry *dentry; 697 struct inode *inode; 698 int flags = O_RDONLY|O_LARGEFILE; 699 __be32 err; 700 int host_err; 701 702 validate_process_creds(); 703 704 /* 705 * If we get here, then the client has already done an "open", 706 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 707 * in case a chmod has now revoked permission. 708 */ 709 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE); 710 if (err) 711 goto out; 712 713 dentry = fhp->fh_dentry; 714 inode = dentry->d_inode; 715 716 /* Disallow write access to files with the append-only bit set 717 * or any access when mandatory locking enabled 718 */ 719 err = nfserr_perm; 720 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE)) 721 goto out; 722 /* 723 * We must ignore files (but only files) which might have mandatory 724 * locks on them because there is no way to know if the accesser has 725 * the lock. 726 */ 727 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) 728 goto out; 729 730 if (!inode->i_fop) 731 goto out; 732 733 /* 734 * Check to see if there are any leases on this file. 735 * This may block while leases are broken. 736 */ 737 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0)); 738 if (host_err == -EWOULDBLOCK) 739 host_err = -ETIMEDOUT; 740 if (host_err) /* NOMEM or WOULDBLOCK */ 741 goto out_nfserr; 742 743 if (access & NFSD_MAY_WRITE) { 744 if (access & NFSD_MAY_READ) 745 flags = O_RDWR|O_LARGEFILE; 746 else 747 flags = O_WRONLY|O_LARGEFILE; 748 749 vfs_dq_init(inode); 750 } 751 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt), 752 flags, current_cred()); 753 if (IS_ERR(*filp)) 754 host_err = PTR_ERR(*filp); 755 else 756 host_err = ima_file_check(*filp, access); 757 out_nfserr: 758 err = nfserrno(host_err); 759 out: 760 validate_process_creds(); 761 return err; 762 } 763 764 /* 765 * Close a file. 766 */ 767 void 768 nfsd_close(struct file *filp) 769 { 770 fput(filp); 771 } 772 773 /* 774 * Sync a file 775 * As this calls fsync (not fdatasync) there is no need for a write_inode 776 * after it. 777 */ 778 static inline int nfsd_dosync(struct file *filp, struct dentry *dp, 779 const struct file_operations *fop) 780 { 781 struct inode *inode = dp->d_inode; 782 int (*fsync) (struct file *, struct dentry *, int); 783 int err; 784 785 err = filemap_write_and_wait(inode->i_mapping); 786 if (err == 0 && fop && (fsync = fop->fsync)) 787 err = fsync(filp, dp, 0); 788 return err; 789 } 790 791 static int 792 nfsd_sync(struct file *filp) 793 { 794 int err; 795 struct inode *inode = filp->f_path.dentry->d_inode; 796 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name); 797 mutex_lock(&inode->i_mutex); 798 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op); 799 mutex_unlock(&inode->i_mutex); 800 801 return err; 802 } 803 804 int 805 nfsd_sync_dir(struct dentry *dp) 806 { 807 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop); 808 } 809 810 /* 811 * Obtain the readahead parameters for the file 812 * specified by (dev, ino). 813 */ 814 815 static inline struct raparms * 816 nfsd_get_raparms(dev_t dev, ino_t ino) 817 { 818 struct raparms *ra, **rap, **frap = NULL; 819 int depth = 0; 820 unsigned int hash; 821 struct raparm_hbucket *rab; 822 823 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 824 rab = &raparm_hash[hash]; 825 826 spin_lock(&rab->pb_lock); 827 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 828 if (ra->p_ino == ino && ra->p_dev == dev) 829 goto found; 830 depth++; 831 if (ra->p_count == 0) 832 frap = rap; 833 } 834 depth = nfsdstats.ra_size*11/10; 835 if (!frap) { 836 spin_unlock(&rab->pb_lock); 837 return NULL; 838 } 839 rap = frap; 840 ra = *frap; 841 ra->p_dev = dev; 842 ra->p_ino = ino; 843 ra->p_set = 0; 844 ra->p_hindex = hash; 845 found: 846 if (rap != &rab->pb_head) { 847 *rap = ra->p_next; 848 ra->p_next = rab->pb_head; 849 rab->pb_head = ra; 850 } 851 ra->p_count++; 852 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 853 spin_unlock(&rab->pb_lock); 854 return ra; 855 } 856 857 /* 858 * Grab and keep cached pages associated with a file in the svc_rqst 859 * so that they can be passed to the network sendmsg/sendpage routines 860 * directly. They will be released after the sending has completed. 861 */ 862 static int 863 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 864 struct splice_desc *sd) 865 { 866 struct svc_rqst *rqstp = sd->u.data; 867 struct page **pp = rqstp->rq_respages + rqstp->rq_resused; 868 struct page *page = buf->page; 869 size_t size; 870 int ret; 871 872 ret = buf->ops->confirm(pipe, buf); 873 if (unlikely(ret)) 874 return ret; 875 876 size = sd->len; 877 878 if (rqstp->rq_res.page_len == 0) { 879 get_page(page); 880 put_page(*pp); 881 *pp = page; 882 rqstp->rq_resused++; 883 rqstp->rq_res.page_base = buf->offset; 884 rqstp->rq_res.page_len = size; 885 } else if (page != pp[-1]) { 886 get_page(page); 887 if (*pp) 888 put_page(*pp); 889 *pp = page; 890 rqstp->rq_resused++; 891 rqstp->rq_res.page_len += size; 892 } else 893 rqstp->rq_res.page_len += size; 894 895 return size; 896 } 897 898 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 899 struct splice_desc *sd) 900 { 901 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 902 } 903 904 static inline int svc_msnfs(struct svc_fh *ffhp) 905 { 906 #ifdef MSNFS 907 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS); 908 #else 909 return 0; 910 #endif 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 struct inode *inode; 918 struct raparms *ra; 919 mm_segment_t oldfs; 920 __be32 err; 921 int host_err; 922 923 err = nfserr_perm; 924 inode = file->f_path.dentry->d_inode; 925 926 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count)) 927 goto out; 928 929 /* Get readahead parameters */ 930 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 931 932 if (ra && ra->p_set) 933 file->f_ra = ra->p_ra; 934 935 if (file->f_op->splice_read && rqstp->rq_splice_ok) { 936 struct splice_desc sd = { 937 .len = 0, 938 .total_len = *count, 939 .pos = offset, 940 .u.data = rqstp, 941 }; 942 943 rqstp->rq_resused = 1; 944 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 945 } else { 946 oldfs = get_fs(); 947 set_fs(KERNEL_DS); 948 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 949 set_fs(oldfs); 950 } 951 952 /* Write back readahead params */ 953 if (ra) { 954 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 955 spin_lock(&rab->pb_lock); 956 ra->p_ra = file->f_ra; 957 ra->p_set = 1; 958 ra->p_count--; 959 spin_unlock(&rab->pb_lock); 960 } 961 962 if (host_err >= 0) { 963 nfsdstats.io_read += host_err; 964 *count = host_err; 965 err = 0; 966 fsnotify_access(file->f_path.dentry); 967 } else 968 err = nfserrno(host_err); 969 out: 970 return err; 971 } 972 973 static void kill_suid(struct dentry *dentry) 974 { 975 struct iattr ia; 976 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 977 978 mutex_lock(&dentry->d_inode->i_mutex); 979 notify_change(dentry, &ia); 980 mutex_unlock(&dentry->d_inode->i_mutex); 981 } 982 983 /* 984 * Gathered writes: If another process is currently writing to the file, 985 * there's a high chance this is another nfsd (triggered by a bulk write 986 * from a client's biod). Rather than syncing the file with each write 987 * request, we sleep for 10 msec. 988 * 989 * I don't know if this roughly approximates C. Juszak's idea of 990 * gathered writes, but it's a nice and simple solution (IMHO), and it 991 * seems to work:-) 992 * 993 * Note: we do this only in the NFSv2 case, since v3 and higher have a 994 * better tool (separate unstable writes and commits) for solving this 995 * problem. 996 */ 997 static int wait_for_concurrent_writes(struct file *file) 998 { 999 struct inode *inode = file->f_path.dentry->d_inode; 1000 static ino_t last_ino; 1001 static dev_t last_dev; 1002 int err = 0; 1003 1004 if (atomic_read(&inode->i_writecount) > 1 1005 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 1006 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 1007 msleep(10); 1008 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 1009 } 1010 1011 if (inode->i_state & I_DIRTY) { 1012 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 1013 err = nfsd_sync(file); 1014 } 1015 last_ino = inode->i_ino; 1016 last_dev = inode->i_sb->s_dev; 1017 return err; 1018 } 1019 1020 static __be32 1021 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1022 loff_t offset, struct kvec *vec, int vlen, 1023 unsigned long *cnt, int *stablep) 1024 { 1025 struct svc_export *exp; 1026 struct dentry *dentry; 1027 struct inode *inode; 1028 mm_segment_t oldfs; 1029 __be32 err = 0; 1030 int host_err; 1031 int stable = *stablep; 1032 int use_wgather; 1033 1034 #ifdef MSNFS 1035 err = nfserr_perm; 1036 1037 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1038 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt))) 1039 goto out; 1040 #endif 1041 1042 dentry = file->f_path.dentry; 1043 inode = dentry->d_inode; 1044 exp = fhp->fh_export; 1045 1046 /* 1047 * Request sync writes if 1048 * - the sync export option has been set, or 1049 * - the client requested O_SYNC behavior (NFSv3 feature). 1050 * - The file system doesn't support fsync(). 1051 * When NFSv2 gathered writes have been configured for this volume, 1052 * flushing the data to disk is handled separately below. 1053 */ 1054 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 1055 1056 if (!file->f_op->fsync) {/* COMMIT3 cannot work */ 1057 stable = 2; 1058 *stablep = 2; /* FILE_SYNC */ 1059 } 1060 1061 if (!EX_ISSYNC(exp)) 1062 stable = 0; 1063 if (stable && !use_wgather) { 1064 spin_lock(&file->f_lock); 1065 file->f_flags |= O_SYNC; 1066 spin_unlock(&file->f_lock); 1067 } 1068 1069 /* Write the data. */ 1070 oldfs = get_fs(); set_fs(KERNEL_DS); 1071 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); 1072 set_fs(oldfs); 1073 if (host_err < 0) 1074 goto out_nfserr; 1075 *cnt = host_err; 1076 nfsdstats.io_write += host_err; 1077 fsnotify_modify(file->f_path.dentry); 1078 1079 /* clear setuid/setgid flag after write */ 1080 if (inode->i_mode & (S_ISUID | S_ISGID)) 1081 kill_suid(dentry); 1082 1083 if (stable && use_wgather) 1084 host_err = wait_for_concurrent_writes(file); 1085 1086 out_nfserr: 1087 dprintk("nfsd: write complete host_err=%d\n", host_err); 1088 if (host_err >= 0) 1089 err = 0; 1090 else 1091 err = nfserrno(host_err); 1092 out: 1093 return err; 1094 } 1095 1096 /* 1097 * Read data from a file. count must contain the requested read count 1098 * on entry. On return, *count contains the number of bytes actually read. 1099 * N.B. After this call fhp needs an fh_put 1100 */ 1101 __be32 1102 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1103 loff_t offset, struct kvec *vec, int vlen, 1104 unsigned long *count) 1105 { 1106 __be32 err; 1107 1108 if (file) { 1109 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1110 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); 1111 if (err) 1112 goto out; 1113 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1114 } else { 1115 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); 1116 if (err) 1117 goto out; 1118 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1119 nfsd_close(file); 1120 } 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 * Strictly speaking, we could sync just the indicated file region here, 1162 * but there's currently no way we can ask the VFS to do so. 1163 * 1164 * Unfortunately we cannot lock the file to make sure we return full WCC 1165 * data to the client, as locking happens lower down in the filesystem. 1166 */ 1167 __be32 1168 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1169 loff_t offset, unsigned long count) 1170 { 1171 struct file *file; 1172 __be32 err; 1173 1174 if ((u64)count > ~(u64)offset) 1175 return nfserr_inval; 1176 1177 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1178 if (err) 1179 return err; 1180 if (EX_ISSYNC(fhp->fh_export)) { 1181 if (file->f_op && file->f_op->fsync) { 1182 err = nfserrno(nfsd_sync(file)); 1183 } else { 1184 err = nfserr_notsupp; 1185 } 1186 } 1187 1188 nfsd_close(file); 1189 return err; 1190 } 1191 #endif /* CONFIG_NFSD_V3 */ 1192 1193 static __be32 1194 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1195 struct iattr *iap) 1196 { 1197 /* 1198 * Mode has already been set earlier in create: 1199 */ 1200 iap->ia_valid &= ~ATTR_MODE; 1201 /* 1202 * Setting uid/gid works only for root. Irix appears to 1203 * send along the gid on create when it tries to implement 1204 * setgid directories via NFS: 1205 */ 1206 if (current_fsuid() != 0) 1207 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1208 if (iap->ia_valid) 1209 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1210 return 0; 1211 } 1212 1213 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1214 * setting size to 0 may fail for some specific file systems by the permission 1215 * checking which requires WRITE permission but the mode is 000. 1216 * we ignore the resizing(to 0) on the just new created file, since the size is 1217 * 0 after file created. 1218 * 1219 * call this only after vfs_create() is called. 1220 * */ 1221 static void 1222 nfsd_check_ignore_resizing(struct iattr *iap) 1223 { 1224 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1225 iap->ia_valid &= ~ATTR_SIZE; 1226 } 1227 1228 /* 1229 * Create a file (regular, directory, device, fifo); UNIX sockets 1230 * not yet implemented. 1231 * If the response fh has been verified, the parent directory should 1232 * already be locked. Note that the parent directory is left locked. 1233 * 1234 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1235 */ 1236 __be32 1237 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1238 char *fname, int flen, struct iattr *iap, 1239 int type, dev_t rdev, struct svc_fh *resfhp) 1240 { 1241 struct dentry *dentry, *dchild = NULL; 1242 struct inode *dirp; 1243 __be32 err; 1244 __be32 err2; 1245 int host_err; 1246 1247 err = nfserr_perm; 1248 if (!flen) 1249 goto out; 1250 err = nfserr_exist; 1251 if (isdotent(fname, flen)) 1252 goto out; 1253 1254 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1255 if (err) 1256 goto out; 1257 1258 dentry = fhp->fh_dentry; 1259 dirp = dentry->d_inode; 1260 1261 err = nfserr_notdir; 1262 if (!dirp->i_op->lookup) 1263 goto out; 1264 /* 1265 * Check whether the response file handle has been verified yet. 1266 * If it has, the parent directory should already be locked. 1267 */ 1268 if (!resfhp->fh_dentry) { 1269 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1270 fh_lock_nested(fhp, I_MUTEX_PARENT); 1271 dchild = lookup_one_len(fname, dentry, flen); 1272 host_err = PTR_ERR(dchild); 1273 if (IS_ERR(dchild)) 1274 goto out_nfserr; 1275 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1276 if (err) 1277 goto out; 1278 } else { 1279 /* called from nfsd_proc_create */ 1280 dchild = dget(resfhp->fh_dentry); 1281 if (!fhp->fh_locked) { 1282 /* not actually possible */ 1283 printk(KERN_ERR 1284 "nfsd_create: parent %s/%s not locked!\n", 1285 dentry->d_parent->d_name.name, 1286 dentry->d_name.name); 1287 err = nfserr_io; 1288 goto out; 1289 } 1290 } 1291 /* 1292 * Make sure the child dentry is still negative ... 1293 */ 1294 err = nfserr_exist; 1295 if (dchild->d_inode) { 1296 dprintk("nfsd_create: dentry %s/%s not negative!\n", 1297 dentry->d_name.name, dchild->d_name.name); 1298 goto out; 1299 } 1300 1301 if (!(iap->ia_valid & ATTR_MODE)) 1302 iap->ia_mode = 0; 1303 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1304 1305 err = nfserr_inval; 1306 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1307 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1308 type); 1309 goto out; 1310 } 1311 1312 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1313 if (host_err) 1314 goto out_nfserr; 1315 1316 /* 1317 * Get the dir op function pointer. 1318 */ 1319 err = 0; 1320 switch (type) { 1321 case S_IFREG: 1322 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1323 if (!host_err) 1324 nfsd_check_ignore_resizing(iap); 1325 break; 1326 case S_IFDIR: 1327 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1328 break; 1329 case S_IFCHR: 1330 case S_IFBLK: 1331 case S_IFIFO: 1332 case S_IFSOCK: 1333 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1334 break; 1335 } 1336 if (host_err < 0) { 1337 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1338 goto out_nfserr; 1339 } 1340 1341 if (EX_ISSYNC(fhp->fh_export)) { 1342 err = nfserrno(nfsd_sync_dir(dentry)); 1343 write_inode_now(dchild->d_inode, 1); 1344 } 1345 1346 err2 = nfsd_create_setattr(rqstp, resfhp, iap); 1347 if (err2) 1348 err = err2; 1349 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1350 /* 1351 * Update the file handle to get the new inode info. 1352 */ 1353 if (!err) 1354 err = fh_update(resfhp); 1355 out: 1356 if (dchild && !IS_ERR(dchild)) 1357 dput(dchild); 1358 return err; 1359 1360 out_nfserr: 1361 err = nfserrno(host_err); 1362 goto out; 1363 } 1364 1365 #ifdef CONFIG_NFSD_V3 1366 /* 1367 * NFSv3 version of nfsd_create 1368 */ 1369 __be32 1370 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp, 1371 char *fname, int flen, struct iattr *iap, 1372 struct svc_fh *resfhp, int createmode, u32 *verifier, 1373 int *truncp, int *created) 1374 { 1375 struct dentry *dentry, *dchild = NULL; 1376 struct inode *dirp; 1377 __be32 err; 1378 __be32 err2; 1379 int host_err; 1380 __u32 v_mtime=0, v_atime=0; 1381 1382 err = nfserr_perm; 1383 if (!flen) 1384 goto out; 1385 err = nfserr_exist; 1386 if (isdotent(fname, flen)) 1387 goto out; 1388 if (!(iap->ia_valid & ATTR_MODE)) 1389 iap->ia_mode = 0; 1390 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1391 if (err) 1392 goto out; 1393 1394 dentry = fhp->fh_dentry; 1395 dirp = dentry->d_inode; 1396 1397 /* Get all the sanity checks out of the way before 1398 * we lock the parent. */ 1399 err = nfserr_notdir; 1400 if (!dirp->i_op->lookup) 1401 goto out; 1402 fh_lock_nested(fhp, I_MUTEX_PARENT); 1403 1404 /* 1405 * Compose the response file handle. 1406 */ 1407 dchild = lookup_one_len(fname, dentry, flen); 1408 host_err = PTR_ERR(dchild); 1409 if (IS_ERR(dchild)) 1410 goto out_nfserr; 1411 1412 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1413 if (err) 1414 goto out; 1415 1416 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1417 /* solaris7 gets confused (bugid 4218508) if these have 1418 * the high bit set, so just clear the high bits. If this is 1419 * ever changed to use different attrs for storing the 1420 * verifier, then do_open_lookup() will also need to be fixed 1421 * accordingly. 1422 */ 1423 v_mtime = verifier[0]&0x7fffffff; 1424 v_atime = verifier[1]&0x7fffffff; 1425 } 1426 1427 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1428 if (host_err) 1429 goto out_nfserr; 1430 if (dchild->d_inode) { 1431 err = 0; 1432 1433 switch (createmode) { 1434 case NFS3_CREATE_UNCHECKED: 1435 if (! S_ISREG(dchild->d_inode->i_mode)) 1436 err = nfserr_exist; 1437 else if (truncp) { 1438 /* in nfsv4, we need to treat this case a little 1439 * differently. we don't want to truncate the 1440 * file now; this would be wrong if the OPEN 1441 * fails for some other reason. furthermore, 1442 * if the size is nonzero, we should ignore it 1443 * according to spec! 1444 */ 1445 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1446 } 1447 else { 1448 iap->ia_valid &= ATTR_SIZE; 1449 goto set_attr; 1450 } 1451 break; 1452 case NFS3_CREATE_EXCLUSIVE: 1453 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1454 && dchild->d_inode->i_atime.tv_sec == v_atime 1455 && dchild->d_inode->i_size == 0 ) 1456 break; 1457 /* fallthru */ 1458 case NFS3_CREATE_GUARDED: 1459 err = nfserr_exist; 1460 } 1461 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1462 goto out; 1463 } 1464 1465 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1466 if (host_err < 0) { 1467 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1468 goto out_nfserr; 1469 } 1470 if (created) 1471 *created = 1; 1472 1473 if (EX_ISSYNC(fhp->fh_export)) { 1474 err = nfserrno(nfsd_sync_dir(dentry)); 1475 /* setattr will sync the child (or not) */ 1476 } 1477 1478 nfsd_check_ignore_resizing(iap); 1479 1480 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1481 /* Cram the verifier into atime/mtime */ 1482 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1483 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1484 /* XXX someone who knows this better please fix it for nsec */ 1485 iap->ia_mtime.tv_sec = v_mtime; 1486 iap->ia_atime.tv_sec = v_atime; 1487 iap->ia_mtime.tv_nsec = 0; 1488 iap->ia_atime.tv_nsec = 0; 1489 } 1490 1491 set_attr: 1492 err2 = nfsd_create_setattr(rqstp, resfhp, iap); 1493 if (err2) 1494 err = err2; 1495 1496 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1497 /* 1498 * Update the filehandle to get the new inode info. 1499 */ 1500 if (!err) 1501 err = fh_update(resfhp); 1502 1503 out: 1504 fh_unlock(fhp); 1505 if (dchild && !IS_ERR(dchild)) 1506 dput(dchild); 1507 return err; 1508 1509 out_nfserr: 1510 err = nfserrno(host_err); 1511 goto out; 1512 } 1513 #endif /* CONFIG_NFSD_V3 */ 1514 1515 /* 1516 * Read a symlink. On entry, *lenp must contain the maximum path length that 1517 * fits into the buffer. On return, it contains the true length. 1518 * N.B. After this call fhp needs an fh_put 1519 */ 1520 __be32 1521 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1522 { 1523 struct dentry *dentry; 1524 struct inode *inode; 1525 mm_segment_t oldfs; 1526 __be32 err; 1527 int host_err; 1528 1529 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1530 if (err) 1531 goto out; 1532 1533 dentry = fhp->fh_dentry; 1534 inode = dentry->d_inode; 1535 1536 err = nfserr_inval; 1537 if (!inode->i_op->readlink) 1538 goto out; 1539 1540 touch_atime(fhp->fh_export->ex_path.mnt, dentry); 1541 /* N.B. Why does this call need a get_fs()?? 1542 * Remove the set_fs and watch the fireworks:-) --okir 1543 */ 1544 1545 oldfs = get_fs(); set_fs(KERNEL_DS); 1546 host_err = inode->i_op->readlink(dentry, buf, *lenp); 1547 set_fs(oldfs); 1548 1549 if (host_err < 0) 1550 goto out_nfserr; 1551 *lenp = host_err; 1552 err = 0; 1553 out: 1554 return err; 1555 1556 out_nfserr: 1557 err = nfserrno(host_err); 1558 goto out; 1559 } 1560 1561 /* 1562 * Create a symlink and look up its inode 1563 * N.B. After this call _both_ fhp and resfhp need an fh_put 1564 */ 1565 __be32 1566 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1567 char *fname, int flen, 1568 char *path, int plen, 1569 struct svc_fh *resfhp, 1570 struct iattr *iap) 1571 { 1572 struct dentry *dentry, *dnew; 1573 __be32 err, cerr; 1574 int host_err; 1575 1576 err = nfserr_noent; 1577 if (!flen || !plen) 1578 goto out; 1579 err = nfserr_exist; 1580 if (isdotent(fname, flen)) 1581 goto out; 1582 1583 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1584 if (err) 1585 goto out; 1586 fh_lock(fhp); 1587 dentry = fhp->fh_dentry; 1588 dnew = lookup_one_len(fname, dentry, flen); 1589 host_err = PTR_ERR(dnew); 1590 if (IS_ERR(dnew)) 1591 goto out_nfserr; 1592 1593 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1594 if (host_err) 1595 goto out_nfserr; 1596 1597 if (unlikely(path[plen] != 0)) { 1598 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1599 if (path_alloced == NULL) 1600 host_err = -ENOMEM; 1601 else { 1602 strncpy(path_alloced, path, plen); 1603 path_alloced[plen] = 0; 1604 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); 1605 kfree(path_alloced); 1606 } 1607 } else 1608 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1609 1610 if (!host_err) { 1611 if (EX_ISSYNC(fhp->fh_export)) 1612 host_err = nfsd_sync_dir(dentry); 1613 } 1614 err = nfserrno(host_err); 1615 fh_unlock(fhp); 1616 1617 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1618 1619 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1620 dput(dnew); 1621 if (err==0) err = cerr; 1622 out: 1623 return err; 1624 1625 out_nfserr: 1626 err = nfserrno(host_err); 1627 goto out; 1628 } 1629 1630 /* 1631 * Create a hardlink 1632 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1633 */ 1634 __be32 1635 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1636 char *name, int len, struct svc_fh *tfhp) 1637 { 1638 struct dentry *ddir, *dnew, *dold; 1639 struct inode *dirp, *dest; 1640 __be32 err; 1641 int host_err; 1642 1643 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1644 if (err) 1645 goto out; 1646 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP); 1647 if (err) 1648 goto out; 1649 1650 err = nfserr_perm; 1651 if (!len) 1652 goto out; 1653 err = nfserr_exist; 1654 if (isdotent(name, len)) 1655 goto out; 1656 1657 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1658 ddir = ffhp->fh_dentry; 1659 dirp = ddir->d_inode; 1660 1661 dnew = lookup_one_len(name, ddir, len); 1662 host_err = PTR_ERR(dnew); 1663 if (IS_ERR(dnew)) 1664 goto out_nfserr; 1665 1666 dold = tfhp->fh_dentry; 1667 dest = dold->d_inode; 1668 1669 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt); 1670 if (host_err) { 1671 err = nfserrno(host_err); 1672 goto out_dput; 1673 } 1674 host_err = vfs_link(dold, dirp, dnew); 1675 if (!host_err) { 1676 if (EX_ISSYNC(ffhp->fh_export)) { 1677 err = nfserrno(nfsd_sync_dir(ddir)); 1678 write_inode_now(dest, 1); 1679 } 1680 err = 0; 1681 } else { 1682 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1683 err = nfserr_acces; 1684 else 1685 err = nfserrno(host_err); 1686 } 1687 mnt_drop_write(tfhp->fh_export->ex_path.mnt); 1688 out_dput: 1689 dput(dnew); 1690 out_unlock: 1691 fh_unlock(ffhp); 1692 out: 1693 return err; 1694 1695 out_nfserr: 1696 err = nfserrno(host_err); 1697 goto out_unlock; 1698 } 1699 1700 /* 1701 * Rename a file 1702 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1703 */ 1704 __be32 1705 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1706 struct svc_fh *tfhp, char *tname, int tlen) 1707 { 1708 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1709 struct inode *fdir, *tdir; 1710 __be32 err; 1711 int host_err; 1712 1713 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1714 if (err) 1715 goto out; 1716 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1717 if (err) 1718 goto out; 1719 1720 fdentry = ffhp->fh_dentry; 1721 fdir = fdentry->d_inode; 1722 1723 tdentry = tfhp->fh_dentry; 1724 tdir = tdentry->d_inode; 1725 1726 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; 1727 if (ffhp->fh_export != tfhp->fh_export) 1728 goto out; 1729 1730 err = nfserr_perm; 1731 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1732 goto out; 1733 1734 /* cannot use fh_lock as we need deadlock protective ordering 1735 * so do it by hand */ 1736 trap = lock_rename(tdentry, fdentry); 1737 ffhp->fh_locked = tfhp->fh_locked = 1; 1738 fill_pre_wcc(ffhp); 1739 fill_pre_wcc(tfhp); 1740 1741 odentry = lookup_one_len(fname, fdentry, flen); 1742 host_err = PTR_ERR(odentry); 1743 if (IS_ERR(odentry)) 1744 goto out_nfserr; 1745 1746 host_err = -ENOENT; 1747 if (!odentry->d_inode) 1748 goto out_dput_old; 1749 host_err = -EINVAL; 1750 if (odentry == trap) 1751 goto out_dput_old; 1752 1753 ndentry = lookup_one_len(tname, tdentry, tlen); 1754 host_err = PTR_ERR(ndentry); 1755 if (IS_ERR(ndentry)) 1756 goto out_dput_old; 1757 host_err = -ENOTEMPTY; 1758 if (ndentry == trap) 1759 goto out_dput_new; 1760 1761 if (svc_msnfs(ffhp) && 1762 ((atomic_read(&odentry->d_count) > 1) 1763 || (atomic_read(&ndentry->d_count) > 1))) { 1764 host_err = -EPERM; 1765 goto out_dput_new; 1766 } 1767 1768 host_err = -EXDEV; 1769 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1770 goto out_dput_new; 1771 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt); 1772 if (host_err) 1773 goto out_dput_new; 1774 1775 host_err = vfs_rename(fdir, odentry, tdir, ndentry); 1776 if (!host_err && EX_ISSYNC(tfhp->fh_export)) { 1777 host_err = nfsd_sync_dir(tdentry); 1778 if (!host_err) 1779 host_err = nfsd_sync_dir(fdentry); 1780 } 1781 1782 mnt_drop_write(ffhp->fh_export->ex_path.mnt); 1783 1784 out_dput_new: 1785 dput(ndentry); 1786 out_dput_old: 1787 dput(odentry); 1788 out_nfserr: 1789 err = nfserrno(host_err); 1790 1791 /* we cannot reply on fh_unlock on the two filehandles, 1792 * as that would do the wrong thing if the two directories 1793 * were the same, so again we do it by hand 1794 */ 1795 fill_post_wcc(ffhp); 1796 fill_post_wcc(tfhp); 1797 unlock_rename(tdentry, fdentry); 1798 ffhp->fh_locked = tfhp->fh_locked = 0; 1799 1800 out: 1801 return err; 1802 } 1803 1804 /* 1805 * Unlink a file or directory 1806 * N.B. After this call fhp needs an fh_put 1807 */ 1808 __be32 1809 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1810 char *fname, int flen) 1811 { 1812 struct dentry *dentry, *rdentry; 1813 struct inode *dirp; 1814 __be32 err; 1815 int host_err; 1816 1817 err = nfserr_acces; 1818 if (!flen || isdotent(fname, flen)) 1819 goto out; 1820 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1821 if (err) 1822 goto out; 1823 1824 fh_lock_nested(fhp, I_MUTEX_PARENT); 1825 dentry = fhp->fh_dentry; 1826 dirp = dentry->d_inode; 1827 1828 rdentry = lookup_one_len(fname, dentry, flen); 1829 host_err = PTR_ERR(rdentry); 1830 if (IS_ERR(rdentry)) 1831 goto out_nfserr; 1832 1833 if (!rdentry->d_inode) { 1834 dput(rdentry); 1835 err = nfserr_noent; 1836 goto out; 1837 } 1838 1839 if (!type) 1840 type = rdentry->d_inode->i_mode & S_IFMT; 1841 1842 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1843 if (host_err) 1844 goto out_nfserr; 1845 1846 if (type != S_IFDIR) { /* It's UNLINK */ 1847 #ifdef MSNFS 1848 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1849 (atomic_read(&rdentry->d_count) > 1)) { 1850 host_err = -EPERM; 1851 } else 1852 #endif 1853 host_err = vfs_unlink(dirp, rdentry); 1854 } else { /* It's RMDIR */ 1855 host_err = vfs_rmdir(dirp, rdentry); 1856 } 1857 1858 dput(rdentry); 1859 1860 if (host_err) 1861 goto out_drop; 1862 if (EX_ISSYNC(fhp->fh_export)) 1863 host_err = nfsd_sync_dir(dentry); 1864 1865 out_drop: 1866 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1867 out_nfserr: 1868 err = nfserrno(host_err); 1869 out: 1870 return err; 1871 } 1872 1873 /* 1874 * We do this buffering because we must not call back into the file 1875 * system's ->lookup() method from the filldir callback. That may well 1876 * deadlock a number of file systems. 1877 * 1878 * This is based heavily on the implementation of same in XFS. 1879 */ 1880 struct buffered_dirent { 1881 u64 ino; 1882 loff_t offset; 1883 int namlen; 1884 unsigned int d_type; 1885 char name[]; 1886 }; 1887 1888 struct readdir_data { 1889 char *dirent; 1890 size_t used; 1891 int full; 1892 }; 1893 1894 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1895 loff_t offset, u64 ino, unsigned int d_type) 1896 { 1897 struct readdir_data *buf = __buf; 1898 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1899 unsigned int reclen; 1900 1901 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1902 if (buf->used + reclen > PAGE_SIZE) { 1903 buf->full = 1; 1904 return -EINVAL; 1905 } 1906 1907 de->namlen = namlen; 1908 de->offset = offset; 1909 de->ino = ino; 1910 de->d_type = d_type; 1911 memcpy(de->name, name, namlen); 1912 buf->used += reclen; 1913 1914 return 0; 1915 } 1916 1917 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1918 struct readdir_cd *cdp, loff_t *offsetp) 1919 { 1920 struct readdir_data buf; 1921 struct buffered_dirent *de; 1922 int host_err; 1923 int size; 1924 loff_t offset; 1925 1926 buf.dirent = (void *)__get_free_page(GFP_KERNEL); 1927 if (!buf.dirent) 1928 return nfserrno(-ENOMEM); 1929 1930 offset = *offsetp; 1931 1932 while (1) { 1933 struct inode *dir_inode = file->f_path.dentry->d_inode; 1934 unsigned int reclen; 1935 1936 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1937 buf.used = 0; 1938 buf.full = 0; 1939 1940 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf); 1941 if (buf.full) 1942 host_err = 0; 1943 1944 if (host_err < 0) 1945 break; 1946 1947 size = buf.used; 1948 1949 if (!size) 1950 break; 1951 1952 /* 1953 * Various filldir functions may end up calling back into 1954 * lookup_one_len() and the file system's ->lookup() method. 1955 * These expect i_mutex to be held, as it would within readdir. 1956 */ 1957 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1958 if (host_err) 1959 break; 1960 1961 de = (struct buffered_dirent *)buf.dirent; 1962 while (size > 0) { 1963 offset = de->offset; 1964 1965 if (func(cdp, de->name, de->namlen, de->offset, 1966 de->ino, de->d_type)) 1967 break; 1968 1969 if (cdp->err != nfs_ok) 1970 break; 1971 1972 reclen = ALIGN(sizeof(*de) + de->namlen, 1973 sizeof(u64)); 1974 size -= reclen; 1975 de = (struct buffered_dirent *)((char *)de + reclen); 1976 } 1977 mutex_unlock(&dir_inode->i_mutex); 1978 if (size > 0) /* We bailed out early */ 1979 break; 1980 1981 offset = vfs_llseek(file, 0, SEEK_CUR); 1982 } 1983 1984 free_page((unsigned long)(buf.dirent)); 1985 1986 if (host_err) 1987 return nfserrno(host_err); 1988 1989 *offsetp = offset; 1990 return cdp->err; 1991 } 1992 1993 /* 1994 * Read entries from a directory. 1995 * The NFSv3/4 verifier we ignore for now. 1996 */ 1997 __be32 1998 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 1999 struct readdir_cd *cdp, filldir_t func) 2000 { 2001 __be32 err; 2002 struct file *file; 2003 loff_t offset = *offsetp; 2004 2005 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file); 2006 if (err) 2007 goto out; 2008 2009 offset = vfs_llseek(file, offset, 0); 2010 if (offset < 0) { 2011 err = nfserrno((int)offset); 2012 goto out_close; 2013 } 2014 2015 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 2016 2017 if (err == nfserr_eof || err == nfserr_toosmall) 2018 err = nfs_ok; /* can still be found in ->err */ 2019 out_close: 2020 nfsd_close(file); 2021 out: 2022 return err; 2023 } 2024 2025 /* 2026 * Get file system stats 2027 * N.B. After this call fhp needs an fh_put 2028 */ 2029 __be32 2030 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 2031 { 2032 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 2033 if (!err && vfs_statfs(fhp->fh_dentry,stat)) 2034 err = nfserr_io; 2035 return err; 2036 } 2037 2038 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 2039 { 2040 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 2041 } 2042 2043 /* 2044 * Check for a user's access permissions to this inode. 2045 */ 2046 __be32 2047 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 2048 struct dentry *dentry, int acc) 2049 { 2050 struct inode *inode = dentry->d_inode; 2051 struct path path; 2052 int err; 2053 2054 if (acc == NFSD_MAY_NOP) 2055 return 0; 2056 #if 0 2057 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 2058 acc, 2059 (acc & NFSD_MAY_READ)? " read" : "", 2060 (acc & NFSD_MAY_WRITE)? " write" : "", 2061 (acc & NFSD_MAY_EXEC)? " exec" : "", 2062 (acc & NFSD_MAY_SATTR)? " sattr" : "", 2063 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 2064 (acc & NFSD_MAY_LOCK)? " lock" : "", 2065 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 2066 inode->i_mode, 2067 IS_IMMUTABLE(inode)? " immut" : "", 2068 IS_APPEND(inode)? " append" : "", 2069 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2070 dprintk(" owner %d/%d user %d/%d\n", 2071 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2072 #endif 2073 2074 /* Normally we reject any write/sattr etc access on a read-only file 2075 * system. But if it is IRIX doing check on write-access for a 2076 * device special file, we ignore rofs. 2077 */ 2078 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2079 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2080 if (exp_rdonly(rqstp, exp) || 2081 __mnt_is_readonly(exp->ex_path.mnt)) 2082 return nfserr_rofs; 2083 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2084 return nfserr_perm; 2085 } 2086 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2087 return nfserr_perm; 2088 2089 if (acc & NFSD_MAY_LOCK) { 2090 /* If we cannot rely on authentication in NLM requests, 2091 * just allow locks, otherwise require read permission, or 2092 * ownership 2093 */ 2094 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2095 return 0; 2096 else 2097 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2098 } 2099 /* 2100 * The file owner always gets access permission for accesses that 2101 * would normally be checked at open time. This is to make 2102 * file access work even when the client has done a fchmod(fd, 0). 2103 * 2104 * However, `cp foo bar' should fail nevertheless when bar is 2105 * readonly. A sensible way to do this might be to reject all 2106 * attempts to truncate a read-only file, because a creat() call 2107 * always implies file truncation. 2108 * ... but this isn't really fair. A process may reasonably call 2109 * ftruncate on an open file descriptor on a file with perm 000. 2110 * We must trust the client to do permission checking - using "ACCESS" 2111 * with NFSv3. 2112 */ 2113 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2114 inode->i_uid == current_fsuid()) 2115 return 0; 2116 2117 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2118 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2119 2120 /* Allow read access to binaries even when mode 111 */ 2121 if (err == -EACCES && S_ISREG(inode->i_mode) && 2122 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE)) 2123 err = inode_permission(inode, MAY_EXEC); 2124 if (err) 2125 goto nfsd_out; 2126 2127 /* Do integrity (permission) checking now, but defer incrementing 2128 * IMA counts to the actual file open. 2129 */ 2130 path.mnt = exp->ex_path.mnt; 2131 path.dentry = dentry; 2132 nfsd_out: 2133 return err? nfserrno(err) : 0; 2134 } 2135 2136 void 2137 nfsd_racache_shutdown(void) 2138 { 2139 struct raparms *raparm, *last_raparm; 2140 unsigned int i; 2141 2142 dprintk("nfsd: freeing readahead buffers.\n"); 2143 2144 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2145 raparm = raparm_hash[i].pb_head; 2146 while(raparm) { 2147 last_raparm = raparm; 2148 raparm = raparm->p_next; 2149 kfree(last_raparm); 2150 } 2151 raparm_hash[i].pb_head = NULL; 2152 } 2153 } 2154 /* 2155 * Initialize readahead param cache 2156 */ 2157 int 2158 nfsd_racache_init(int cache_size) 2159 { 2160 int i; 2161 int j = 0; 2162 int nperbucket; 2163 struct raparms **raparm = NULL; 2164 2165 2166 if (raparm_hash[0].pb_head) 2167 return 0; 2168 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2169 if (nperbucket < 2) 2170 nperbucket = 2; 2171 cache_size = nperbucket * RAPARM_HASH_SIZE; 2172 2173 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2174 2175 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2176 spin_lock_init(&raparm_hash[i].pb_lock); 2177 2178 raparm = &raparm_hash[i].pb_head; 2179 for (j = 0; j < nperbucket; j++) { 2180 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2181 if (!*raparm) 2182 goto out_nomem; 2183 raparm = &(*raparm)->p_next; 2184 } 2185 *raparm = NULL; 2186 } 2187 2188 nfsdstats.ra_size = cache_size; 2189 return 0; 2190 2191 out_nomem: 2192 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2193 nfsd_racache_shutdown(); 2194 return -ENOMEM; 2195 } 2196 2197 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 2198 struct posix_acl * 2199 nfsd_get_posix_acl(struct svc_fh *fhp, int type) 2200 { 2201 struct inode *inode = fhp->fh_dentry->d_inode; 2202 char *name; 2203 void *value = NULL; 2204 ssize_t size; 2205 struct posix_acl *acl; 2206 2207 if (!IS_POSIXACL(inode)) 2208 return ERR_PTR(-EOPNOTSUPP); 2209 2210 switch (type) { 2211 case ACL_TYPE_ACCESS: 2212 name = POSIX_ACL_XATTR_ACCESS; 2213 break; 2214 case ACL_TYPE_DEFAULT: 2215 name = POSIX_ACL_XATTR_DEFAULT; 2216 break; 2217 default: 2218 return ERR_PTR(-EOPNOTSUPP); 2219 } 2220 2221 size = nfsd_getxattr(fhp->fh_dentry, name, &value); 2222 if (size < 0) 2223 return ERR_PTR(size); 2224 2225 acl = posix_acl_from_xattr(value, size); 2226 kfree(value); 2227 return acl; 2228 } 2229 2230 int 2231 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) 2232 { 2233 struct inode *inode = fhp->fh_dentry->d_inode; 2234 char *name; 2235 void *value = NULL; 2236 size_t size; 2237 int error; 2238 2239 if (!IS_POSIXACL(inode) || 2240 !inode->i_op->setxattr || !inode->i_op->removexattr) 2241 return -EOPNOTSUPP; 2242 switch(type) { 2243 case ACL_TYPE_ACCESS: 2244 name = POSIX_ACL_XATTR_ACCESS; 2245 break; 2246 case ACL_TYPE_DEFAULT: 2247 name = POSIX_ACL_XATTR_DEFAULT; 2248 break; 2249 default: 2250 return -EOPNOTSUPP; 2251 } 2252 2253 if (acl && acl->a_count) { 2254 size = posix_acl_xattr_size(acl->a_count); 2255 value = kmalloc(size, GFP_KERNEL); 2256 if (!value) 2257 return -ENOMEM; 2258 error = posix_acl_to_xattr(acl, value, size); 2259 if (error < 0) 2260 goto getout; 2261 size = error; 2262 } else 2263 size = 0; 2264 2265 error = mnt_want_write(fhp->fh_export->ex_path.mnt); 2266 if (error) 2267 goto getout; 2268 if (size) 2269 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); 2270 else { 2271 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) 2272 error = 0; 2273 else { 2274 error = vfs_removexattr(fhp->fh_dentry, name); 2275 if (error == -ENODATA) 2276 error = 0; 2277 } 2278 } 2279 mnt_drop_write(fhp->fh_export->ex_path.mnt); 2280 2281 getout: 2282 kfree(value); 2283 return error; 2284 } 2285 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 2286