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, FMODE_WRITE | 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) ? FMODE_WRITE : 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 out_nfserr: 756 err = nfserrno(host_err); 757 out: 758 validate_process_creds(); 759 return err; 760 } 761 762 /* 763 * Close a file. 764 */ 765 void 766 nfsd_close(struct file *filp) 767 { 768 fput(filp); 769 } 770 771 /* 772 * Sync a directory to disk. 773 * 774 * We can't just call vfs_fsync because our requirements are slightly odd: 775 * 776 * a) we do not have a file struct available 777 * b) we expect to have i_mutex already held by the caller 778 */ 779 int 780 nfsd_sync_dir(struct dentry *dentry) 781 { 782 struct inode *inode = dentry->d_inode; 783 int error; 784 785 WARN_ON(!mutex_is_locked(&inode->i_mutex)); 786 787 error = filemap_write_and_wait(inode->i_mapping); 788 if (!error && inode->i_fop->fsync) 789 error = inode->i_fop->fsync(NULL, dentry, 0); 790 return error; 791 } 792 793 /* 794 * Obtain the readahead parameters for the file 795 * specified by (dev, ino). 796 */ 797 798 static inline struct raparms * 799 nfsd_get_raparms(dev_t dev, ino_t ino) 800 { 801 struct raparms *ra, **rap, **frap = NULL; 802 int depth = 0; 803 unsigned int hash; 804 struct raparm_hbucket *rab; 805 806 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 807 rab = &raparm_hash[hash]; 808 809 spin_lock(&rab->pb_lock); 810 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 811 if (ra->p_ino == ino && ra->p_dev == dev) 812 goto found; 813 depth++; 814 if (ra->p_count == 0) 815 frap = rap; 816 } 817 depth = nfsdstats.ra_size*11/10; 818 if (!frap) { 819 spin_unlock(&rab->pb_lock); 820 return NULL; 821 } 822 rap = frap; 823 ra = *frap; 824 ra->p_dev = dev; 825 ra->p_ino = ino; 826 ra->p_set = 0; 827 ra->p_hindex = hash; 828 found: 829 if (rap != &rab->pb_head) { 830 *rap = ra->p_next; 831 ra->p_next = rab->pb_head; 832 rab->pb_head = ra; 833 } 834 ra->p_count++; 835 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 836 spin_unlock(&rab->pb_lock); 837 return ra; 838 } 839 840 /* 841 * Grab and keep cached pages associated with a file in the svc_rqst 842 * so that they can be passed to the network sendmsg/sendpage routines 843 * directly. They will be released after the sending has completed. 844 */ 845 static int 846 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 847 struct splice_desc *sd) 848 { 849 struct svc_rqst *rqstp = sd->u.data; 850 struct page **pp = rqstp->rq_respages + rqstp->rq_resused; 851 struct page *page = buf->page; 852 size_t size; 853 int ret; 854 855 ret = buf->ops->confirm(pipe, buf); 856 if (unlikely(ret)) 857 return ret; 858 859 size = sd->len; 860 861 if (rqstp->rq_res.page_len == 0) { 862 get_page(page); 863 put_page(*pp); 864 *pp = page; 865 rqstp->rq_resused++; 866 rqstp->rq_res.page_base = buf->offset; 867 rqstp->rq_res.page_len = size; 868 } else if (page != pp[-1]) { 869 get_page(page); 870 if (*pp) 871 put_page(*pp); 872 *pp = page; 873 rqstp->rq_resused++; 874 rqstp->rq_res.page_len += size; 875 } else 876 rqstp->rq_res.page_len += size; 877 878 return size; 879 } 880 881 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 882 struct splice_desc *sd) 883 { 884 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 885 } 886 887 static inline int svc_msnfs(struct svc_fh *ffhp) 888 { 889 #ifdef MSNFS 890 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS); 891 #else 892 return 0; 893 #endif 894 } 895 896 static __be32 897 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 898 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 899 { 900 struct inode *inode; 901 struct raparms *ra; 902 mm_segment_t oldfs; 903 __be32 err; 904 int host_err; 905 906 err = nfserr_perm; 907 inode = file->f_path.dentry->d_inode; 908 909 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count)) 910 goto out; 911 912 /* Get readahead parameters */ 913 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 914 915 if (ra && ra->p_set) 916 file->f_ra = ra->p_ra; 917 918 if (file->f_op->splice_read && rqstp->rq_splice_ok) { 919 struct splice_desc sd = { 920 .len = 0, 921 .total_len = *count, 922 .pos = offset, 923 .u.data = rqstp, 924 }; 925 926 rqstp->rq_resused = 1; 927 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 928 } else { 929 oldfs = get_fs(); 930 set_fs(KERNEL_DS); 931 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 932 set_fs(oldfs); 933 } 934 935 /* Write back readahead params */ 936 if (ra) { 937 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 938 spin_lock(&rab->pb_lock); 939 ra->p_ra = file->f_ra; 940 ra->p_set = 1; 941 ra->p_count--; 942 spin_unlock(&rab->pb_lock); 943 } 944 945 if (host_err >= 0) { 946 nfsdstats.io_read += host_err; 947 *count = host_err; 948 err = 0; 949 fsnotify_access(file->f_path.dentry); 950 } else 951 err = nfserrno(host_err); 952 out: 953 return err; 954 } 955 956 static void kill_suid(struct dentry *dentry) 957 { 958 struct iattr ia; 959 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 960 961 mutex_lock(&dentry->d_inode->i_mutex); 962 notify_change(dentry, &ia); 963 mutex_unlock(&dentry->d_inode->i_mutex); 964 } 965 966 /* 967 * Gathered writes: If another process is currently writing to the file, 968 * there's a high chance this is another nfsd (triggered by a bulk write 969 * from a client's biod). Rather than syncing the file with each write 970 * request, we sleep for 10 msec. 971 * 972 * I don't know if this roughly approximates C. Juszak's idea of 973 * gathered writes, but it's a nice and simple solution (IMHO), and it 974 * seems to work:-) 975 * 976 * Note: we do this only in the NFSv2 case, since v3 and higher have a 977 * better tool (separate unstable writes and commits) for solving this 978 * problem. 979 */ 980 static int wait_for_concurrent_writes(struct file *file) 981 { 982 struct inode *inode = file->f_path.dentry->d_inode; 983 static ino_t last_ino; 984 static dev_t last_dev; 985 int err = 0; 986 987 if (atomic_read(&inode->i_writecount) > 1 988 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 989 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 990 msleep(10); 991 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 992 } 993 994 if (inode->i_state & I_DIRTY) { 995 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 996 err = vfs_fsync(file, file->f_path.dentry, 0); 997 } 998 last_ino = inode->i_ino; 999 last_dev = inode->i_sb->s_dev; 1000 return err; 1001 } 1002 1003 static __be32 1004 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1005 loff_t offset, struct kvec *vec, int vlen, 1006 unsigned long *cnt, int *stablep) 1007 { 1008 struct svc_export *exp; 1009 struct dentry *dentry; 1010 struct inode *inode; 1011 mm_segment_t oldfs; 1012 __be32 err = 0; 1013 int host_err; 1014 int stable = *stablep; 1015 int use_wgather; 1016 1017 #ifdef MSNFS 1018 err = nfserr_perm; 1019 1020 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1021 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt))) 1022 goto out; 1023 #endif 1024 1025 dentry = file->f_path.dentry; 1026 inode = dentry->d_inode; 1027 exp = fhp->fh_export; 1028 1029 /* 1030 * Request sync writes if 1031 * - the sync export option has been set, or 1032 * - the client requested O_SYNC behavior (NFSv3 feature). 1033 * - The file system doesn't support fsync(). 1034 * When NFSv2 gathered writes have been configured for this volume, 1035 * flushing the data to disk is handled separately below. 1036 */ 1037 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 1038 1039 if (!file->f_op->fsync) {/* COMMIT3 cannot work */ 1040 stable = 2; 1041 *stablep = 2; /* FILE_SYNC */ 1042 } 1043 1044 if (!EX_ISSYNC(exp)) 1045 stable = 0; 1046 if (stable && !use_wgather) { 1047 spin_lock(&file->f_lock); 1048 file->f_flags |= O_SYNC; 1049 spin_unlock(&file->f_lock); 1050 } 1051 1052 /* Write the data. */ 1053 oldfs = get_fs(); set_fs(KERNEL_DS); 1054 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); 1055 set_fs(oldfs); 1056 if (host_err < 0) 1057 goto out_nfserr; 1058 *cnt = host_err; 1059 nfsdstats.io_write += host_err; 1060 fsnotify_modify(file->f_path.dentry); 1061 1062 /* clear setuid/setgid flag after write */ 1063 if (inode->i_mode & (S_ISUID | S_ISGID)) 1064 kill_suid(dentry); 1065 1066 if (stable && use_wgather) 1067 host_err = wait_for_concurrent_writes(file); 1068 1069 out_nfserr: 1070 dprintk("nfsd: write complete host_err=%d\n", host_err); 1071 if (host_err >= 0) 1072 err = 0; 1073 else 1074 err = nfserrno(host_err); 1075 out: 1076 return err; 1077 } 1078 1079 /* 1080 * Read data from a file. count must contain the requested read count 1081 * on entry. On return, *count contains the number of bytes actually read. 1082 * N.B. After this call fhp needs an fh_put 1083 */ 1084 __be32 1085 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1086 loff_t offset, struct kvec *vec, int vlen, 1087 unsigned long *count) 1088 { 1089 __be32 err; 1090 1091 if (file) { 1092 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1093 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); 1094 if (err) 1095 goto out; 1096 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1097 } else { 1098 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); 1099 if (err) 1100 goto out; 1101 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1102 nfsd_close(file); 1103 } 1104 out: 1105 return err; 1106 } 1107 1108 /* 1109 * Write data to a file. 1110 * The stable flag requests synchronous writes. 1111 * N.B. After this call fhp needs an fh_put 1112 */ 1113 __be32 1114 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1115 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1116 int *stablep) 1117 { 1118 __be32 err = 0; 1119 1120 if (file) { 1121 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1122 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1123 if (err) 1124 goto out; 1125 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1126 stablep); 1127 } else { 1128 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1129 if (err) 1130 goto out; 1131 1132 if (cnt) 1133 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1134 cnt, stablep); 1135 nfsd_close(file); 1136 } 1137 out: 1138 return err; 1139 } 1140 1141 #ifdef CONFIG_NFSD_V3 1142 /* 1143 * Commit all pending writes to stable storage. 1144 * 1145 * Note: we only guarantee that data that lies within the range specified 1146 * by the 'offset' and 'count' parameters will be synced. 1147 * 1148 * Unfortunately we cannot lock the file to make sure we return full WCC 1149 * data to the client, as locking happens lower down in the filesystem. 1150 */ 1151 __be32 1152 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1153 loff_t offset, unsigned long count) 1154 { 1155 struct file *file; 1156 loff_t end = LLONG_MAX; 1157 __be32 err = nfserr_inval; 1158 1159 if (offset < 0) 1160 goto out; 1161 if (count != 0) { 1162 end = offset + (loff_t)count - 1; 1163 if (end < offset) 1164 goto out; 1165 } 1166 1167 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1168 if (err) 1169 goto out; 1170 if (EX_ISSYNC(fhp->fh_export)) { 1171 int err2 = vfs_fsync_range(file, file->f_path.dentry, 1172 offset, end, 0); 1173 1174 if (err2 != -EINVAL) 1175 err = nfserrno(err2); 1176 else 1177 err = nfserr_notsupp; 1178 } 1179 1180 nfsd_close(file); 1181 out: 1182 return err; 1183 } 1184 #endif /* CONFIG_NFSD_V3 */ 1185 1186 static __be32 1187 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1188 struct iattr *iap) 1189 { 1190 /* 1191 * Mode has already been set earlier in create: 1192 */ 1193 iap->ia_valid &= ~ATTR_MODE; 1194 /* 1195 * Setting uid/gid works only for root. Irix appears to 1196 * send along the gid on create when it tries to implement 1197 * setgid directories via NFS: 1198 */ 1199 if (current_fsuid() != 0) 1200 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1201 if (iap->ia_valid) 1202 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1203 return 0; 1204 } 1205 1206 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1207 * setting size to 0 may fail for some specific file systems by the permission 1208 * checking which requires WRITE permission but the mode is 000. 1209 * we ignore the resizing(to 0) on the just new created file, since the size is 1210 * 0 after file created. 1211 * 1212 * call this only after vfs_create() is called. 1213 * */ 1214 static void 1215 nfsd_check_ignore_resizing(struct iattr *iap) 1216 { 1217 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1218 iap->ia_valid &= ~ATTR_SIZE; 1219 } 1220 1221 /* 1222 * Create a file (regular, directory, device, fifo); UNIX sockets 1223 * not yet implemented. 1224 * If the response fh has been verified, the parent directory should 1225 * already be locked. Note that the parent directory is left locked. 1226 * 1227 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1228 */ 1229 __be32 1230 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1231 char *fname, int flen, struct iattr *iap, 1232 int type, dev_t rdev, struct svc_fh *resfhp) 1233 { 1234 struct dentry *dentry, *dchild = NULL; 1235 struct inode *dirp; 1236 __be32 err; 1237 __be32 err2; 1238 int host_err; 1239 1240 err = nfserr_perm; 1241 if (!flen) 1242 goto out; 1243 err = nfserr_exist; 1244 if (isdotent(fname, flen)) 1245 goto out; 1246 1247 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1248 if (err) 1249 goto out; 1250 1251 dentry = fhp->fh_dentry; 1252 dirp = dentry->d_inode; 1253 1254 err = nfserr_notdir; 1255 if (!dirp->i_op->lookup) 1256 goto out; 1257 /* 1258 * Check whether the response file handle has been verified yet. 1259 * If it has, the parent directory should already be locked. 1260 */ 1261 if (!resfhp->fh_dentry) { 1262 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1263 fh_lock_nested(fhp, I_MUTEX_PARENT); 1264 dchild = lookup_one_len(fname, dentry, flen); 1265 host_err = PTR_ERR(dchild); 1266 if (IS_ERR(dchild)) 1267 goto out_nfserr; 1268 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1269 if (err) 1270 goto out; 1271 } else { 1272 /* called from nfsd_proc_create */ 1273 dchild = dget(resfhp->fh_dentry); 1274 if (!fhp->fh_locked) { 1275 /* not actually possible */ 1276 printk(KERN_ERR 1277 "nfsd_create: parent %s/%s not locked!\n", 1278 dentry->d_parent->d_name.name, 1279 dentry->d_name.name); 1280 err = nfserr_io; 1281 goto out; 1282 } 1283 } 1284 /* 1285 * Make sure the child dentry is still negative ... 1286 */ 1287 err = nfserr_exist; 1288 if (dchild->d_inode) { 1289 dprintk("nfsd_create: dentry %s/%s not negative!\n", 1290 dentry->d_name.name, dchild->d_name.name); 1291 goto out; 1292 } 1293 1294 if (!(iap->ia_valid & ATTR_MODE)) 1295 iap->ia_mode = 0; 1296 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1297 1298 err = nfserr_inval; 1299 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1300 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1301 type); 1302 goto out; 1303 } 1304 1305 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1306 if (host_err) 1307 goto out_nfserr; 1308 1309 /* 1310 * Get the dir op function pointer. 1311 */ 1312 err = 0; 1313 switch (type) { 1314 case S_IFREG: 1315 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1316 if (!host_err) 1317 nfsd_check_ignore_resizing(iap); 1318 break; 1319 case S_IFDIR: 1320 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1321 break; 1322 case S_IFCHR: 1323 case S_IFBLK: 1324 case S_IFIFO: 1325 case S_IFSOCK: 1326 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1327 break; 1328 } 1329 if (host_err < 0) { 1330 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1331 goto out_nfserr; 1332 } 1333 1334 if (EX_ISSYNC(fhp->fh_export)) { 1335 err = nfserrno(nfsd_sync_dir(dentry)); 1336 write_inode_now(dchild->d_inode, 1); 1337 } 1338 1339 err2 = nfsd_create_setattr(rqstp, resfhp, iap); 1340 if (err2) 1341 err = err2; 1342 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1343 /* 1344 * Update the file handle to get the new inode info. 1345 */ 1346 if (!err) 1347 err = fh_update(resfhp); 1348 out: 1349 if (dchild && !IS_ERR(dchild)) 1350 dput(dchild); 1351 return err; 1352 1353 out_nfserr: 1354 err = nfserrno(host_err); 1355 goto out; 1356 } 1357 1358 #ifdef CONFIG_NFSD_V3 1359 /* 1360 * NFSv3 version of nfsd_create 1361 */ 1362 __be32 1363 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp, 1364 char *fname, int flen, struct iattr *iap, 1365 struct svc_fh *resfhp, int createmode, u32 *verifier, 1366 int *truncp, int *created) 1367 { 1368 struct dentry *dentry, *dchild = NULL; 1369 struct inode *dirp; 1370 __be32 err; 1371 __be32 err2; 1372 int host_err; 1373 __u32 v_mtime=0, v_atime=0; 1374 1375 err = nfserr_perm; 1376 if (!flen) 1377 goto out; 1378 err = nfserr_exist; 1379 if (isdotent(fname, flen)) 1380 goto out; 1381 if (!(iap->ia_valid & ATTR_MODE)) 1382 iap->ia_mode = 0; 1383 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1384 if (err) 1385 goto out; 1386 1387 dentry = fhp->fh_dentry; 1388 dirp = dentry->d_inode; 1389 1390 /* Get all the sanity checks out of the way before 1391 * we lock the parent. */ 1392 err = nfserr_notdir; 1393 if (!dirp->i_op->lookup) 1394 goto out; 1395 fh_lock_nested(fhp, I_MUTEX_PARENT); 1396 1397 /* 1398 * Compose the response file handle. 1399 */ 1400 dchild = lookup_one_len(fname, dentry, flen); 1401 host_err = PTR_ERR(dchild); 1402 if (IS_ERR(dchild)) 1403 goto out_nfserr; 1404 1405 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1406 if (err) 1407 goto out; 1408 1409 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1410 /* solaris7 gets confused (bugid 4218508) if these have 1411 * the high bit set, so just clear the high bits. If this is 1412 * ever changed to use different attrs for storing the 1413 * verifier, then do_open_lookup() will also need to be fixed 1414 * accordingly. 1415 */ 1416 v_mtime = verifier[0]&0x7fffffff; 1417 v_atime = verifier[1]&0x7fffffff; 1418 } 1419 1420 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1421 if (host_err) 1422 goto out_nfserr; 1423 if (dchild->d_inode) { 1424 err = 0; 1425 1426 switch (createmode) { 1427 case NFS3_CREATE_UNCHECKED: 1428 if (! S_ISREG(dchild->d_inode->i_mode)) 1429 err = nfserr_exist; 1430 else if (truncp) { 1431 /* in nfsv4, we need to treat this case a little 1432 * differently. we don't want to truncate the 1433 * file now; this would be wrong if the OPEN 1434 * fails for some other reason. furthermore, 1435 * if the size is nonzero, we should ignore it 1436 * according to spec! 1437 */ 1438 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1439 } 1440 else { 1441 iap->ia_valid &= ATTR_SIZE; 1442 goto set_attr; 1443 } 1444 break; 1445 case NFS3_CREATE_EXCLUSIVE: 1446 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1447 && dchild->d_inode->i_atime.tv_sec == v_atime 1448 && dchild->d_inode->i_size == 0 ) 1449 break; 1450 /* fallthru */ 1451 case NFS3_CREATE_GUARDED: 1452 err = nfserr_exist; 1453 } 1454 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1455 goto out; 1456 } 1457 1458 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1459 if (host_err < 0) { 1460 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1461 goto out_nfserr; 1462 } 1463 if (created) 1464 *created = 1; 1465 1466 if (EX_ISSYNC(fhp->fh_export)) { 1467 err = nfserrno(nfsd_sync_dir(dentry)); 1468 /* setattr will sync the child (or not) */ 1469 } 1470 1471 nfsd_check_ignore_resizing(iap); 1472 1473 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1474 /* Cram the verifier into atime/mtime */ 1475 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1476 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1477 /* XXX someone who knows this better please fix it for nsec */ 1478 iap->ia_mtime.tv_sec = v_mtime; 1479 iap->ia_atime.tv_sec = v_atime; 1480 iap->ia_mtime.tv_nsec = 0; 1481 iap->ia_atime.tv_nsec = 0; 1482 } 1483 1484 set_attr: 1485 err2 = nfsd_create_setattr(rqstp, resfhp, iap); 1486 if (err2) 1487 err = err2; 1488 1489 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1490 /* 1491 * Update the filehandle to get the new inode info. 1492 */ 1493 if (!err) 1494 err = fh_update(resfhp); 1495 1496 out: 1497 fh_unlock(fhp); 1498 if (dchild && !IS_ERR(dchild)) 1499 dput(dchild); 1500 return err; 1501 1502 out_nfserr: 1503 err = nfserrno(host_err); 1504 goto out; 1505 } 1506 #endif /* CONFIG_NFSD_V3 */ 1507 1508 /* 1509 * Read a symlink. On entry, *lenp must contain the maximum path length that 1510 * fits into the buffer. On return, it contains the true length. 1511 * N.B. After this call fhp needs an fh_put 1512 */ 1513 __be32 1514 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1515 { 1516 struct dentry *dentry; 1517 struct inode *inode; 1518 mm_segment_t oldfs; 1519 __be32 err; 1520 int host_err; 1521 1522 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1523 if (err) 1524 goto out; 1525 1526 dentry = fhp->fh_dentry; 1527 inode = dentry->d_inode; 1528 1529 err = nfserr_inval; 1530 if (!inode->i_op->readlink) 1531 goto out; 1532 1533 touch_atime(fhp->fh_export->ex_path.mnt, dentry); 1534 /* N.B. Why does this call need a get_fs()?? 1535 * Remove the set_fs and watch the fireworks:-) --okir 1536 */ 1537 1538 oldfs = get_fs(); set_fs(KERNEL_DS); 1539 host_err = inode->i_op->readlink(dentry, buf, *lenp); 1540 set_fs(oldfs); 1541 1542 if (host_err < 0) 1543 goto out_nfserr; 1544 *lenp = host_err; 1545 err = 0; 1546 out: 1547 return err; 1548 1549 out_nfserr: 1550 err = nfserrno(host_err); 1551 goto out; 1552 } 1553 1554 /* 1555 * Create a symlink and look up its inode 1556 * N.B. After this call _both_ fhp and resfhp need an fh_put 1557 */ 1558 __be32 1559 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1560 char *fname, int flen, 1561 char *path, int plen, 1562 struct svc_fh *resfhp, 1563 struct iattr *iap) 1564 { 1565 struct dentry *dentry, *dnew; 1566 __be32 err, cerr; 1567 int host_err; 1568 1569 err = nfserr_noent; 1570 if (!flen || !plen) 1571 goto out; 1572 err = nfserr_exist; 1573 if (isdotent(fname, flen)) 1574 goto out; 1575 1576 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1577 if (err) 1578 goto out; 1579 fh_lock(fhp); 1580 dentry = fhp->fh_dentry; 1581 dnew = lookup_one_len(fname, dentry, flen); 1582 host_err = PTR_ERR(dnew); 1583 if (IS_ERR(dnew)) 1584 goto out_nfserr; 1585 1586 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1587 if (host_err) 1588 goto out_nfserr; 1589 1590 if (unlikely(path[plen] != 0)) { 1591 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1592 if (path_alloced == NULL) 1593 host_err = -ENOMEM; 1594 else { 1595 strncpy(path_alloced, path, plen); 1596 path_alloced[plen] = 0; 1597 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); 1598 kfree(path_alloced); 1599 } 1600 } else 1601 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1602 1603 if (!host_err) { 1604 if (EX_ISSYNC(fhp->fh_export)) 1605 host_err = nfsd_sync_dir(dentry); 1606 } 1607 err = nfserrno(host_err); 1608 fh_unlock(fhp); 1609 1610 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1611 1612 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1613 dput(dnew); 1614 if (err==0) err = cerr; 1615 out: 1616 return err; 1617 1618 out_nfserr: 1619 err = nfserrno(host_err); 1620 goto out; 1621 } 1622 1623 /* 1624 * Create a hardlink 1625 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1626 */ 1627 __be32 1628 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1629 char *name, int len, struct svc_fh *tfhp) 1630 { 1631 struct dentry *ddir, *dnew, *dold; 1632 struct inode *dirp, *dest; 1633 __be32 err; 1634 int host_err; 1635 1636 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1637 if (err) 1638 goto out; 1639 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP); 1640 if (err) 1641 goto out; 1642 1643 err = nfserr_perm; 1644 if (!len) 1645 goto out; 1646 err = nfserr_exist; 1647 if (isdotent(name, len)) 1648 goto out; 1649 1650 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1651 ddir = ffhp->fh_dentry; 1652 dirp = ddir->d_inode; 1653 1654 dnew = lookup_one_len(name, ddir, len); 1655 host_err = PTR_ERR(dnew); 1656 if (IS_ERR(dnew)) 1657 goto out_nfserr; 1658 1659 dold = tfhp->fh_dentry; 1660 dest = dold->d_inode; 1661 1662 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt); 1663 if (host_err) { 1664 err = nfserrno(host_err); 1665 goto out_dput; 1666 } 1667 host_err = vfs_link(dold, dirp, dnew); 1668 if (!host_err) { 1669 if (EX_ISSYNC(ffhp->fh_export)) { 1670 err = nfserrno(nfsd_sync_dir(ddir)); 1671 write_inode_now(dest, 1); 1672 } 1673 err = 0; 1674 } else { 1675 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1676 err = nfserr_acces; 1677 else 1678 err = nfserrno(host_err); 1679 } 1680 mnt_drop_write(tfhp->fh_export->ex_path.mnt); 1681 out_dput: 1682 dput(dnew); 1683 out_unlock: 1684 fh_unlock(ffhp); 1685 out: 1686 return err; 1687 1688 out_nfserr: 1689 err = nfserrno(host_err); 1690 goto out_unlock; 1691 } 1692 1693 /* 1694 * Rename a file 1695 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1696 */ 1697 __be32 1698 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1699 struct svc_fh *tfhp, char *tname, int tlen) 1700 { 1701 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1702 struct inode *fdir, *tdir; 1703 __be32 err; 1704 int host_err; 1705 1706 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1707 if (err) 1708 goto out; 1709 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1710 if (err) 1711 goto out; 1712 1713 fdentry = ffhp->fh_dentry; 1714 fdir = fdentry->d_inode; 1715 1716 tdentry = tfhp->fh_dentry; 1717 tdir = tdentry->d_inode; 1718 1719 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; 1720 if (ffhp->fh_export != tfhp->fh_export) 1721 goto out; 1722 1723 err = nfserr_perm; 1724 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1725 goto out; 1726 1727 /* cannot use fh_lock as we need deadlock protective ordering 1728 * so do it by hand */ 1729 trap = lock_rename(tdentry, fdentry); 1730 ffhp->fh_locked = tfhp->fh_locked = 1; 1731 fill_pre_wcc(ffhp); 1732 fill_pre_wcc(tfhp); 1733 1734 odentry = lookup_one_len(fname, fdentry, flen); 1735 host_err = PTR_ERR(odentry); 1736 if (IS_ERR(odentry)) 1737 goto out_nfserr; 1738 1739 host_err = -ENOENT; 1740 if (!odentry->d_inode) 1741 goto out_dput_old; 1742 host_err = -EINVAL; 1743 if (odentry == trap) 1744 goto out_dput_old; 1745 1746 ndentry = lookup_one_len(tname, tdentry, tlen); 1747 host_err = PTR_ERR(ndentry); 1748 if (IS_ERR(ndentry)) 1749 goto out_dput_old; 1750 host_err = -ENOTEMPTY; 1751 if (ndentry == trap) 1752 goto out_dput_new; 1753 1754 if (svc_msnfs(ffhp) && 1755 ((atomic_read(&odentry->d_count) > 1) 1756 || (atomic_read(&ndentry->d_count) > 1))) { 1757 host_err = -EPERM; 1758 goto out_dput_new; 1759 } 1760 1761 host_err = -EXDEV; 1762 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1763 goto out_dput_new; 1764 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt); 1765 if (host_err) 1766 goto out_dput_new; 1767 1768 host_err = vfs_rename(fdir, odentry, tdir, ndentry); 1769 if (!host_err && EX_ISSYNC(tfhp->fh_export)) { 1770 host_err = nfsd_sync_dir(tdentry); 1771 if (!host_err) 1772 host_err = nfsd_sync_dir(fdentry); 1773 } 1774 1775 mnt_drop_write(ffhp->fh_export->ex_path.mnt); 1776 1777 out_dput_new: 1778 dput(ndentry); 1779 out_dput_old: 1780 dput(odentry); 1781 out_nfserr: 1782 err = nfserrno(host_err); 1783 1784 /* we cannot reply on fh_unlock on the two filehandles, 1785 * as that would do the wrong thing if the two directories 1786 * were the same, so again we do it by hand 1787 */ 1788 fill_post_wcc(ffhp); 1789 fill_post_wcc(tfhp); 1790 unlock_rename(tdentry, fdentry); 1791 ffhp->fh_locked = tfhp->fh_locked = 0; 1792 1793 out: 1794 return err; 1795 } 1796 1797 /* 1798 * Unlink a file or directory 1799 * N.B. After this call fhp needs an fh_put 1800 */ 1801 __be32 1802 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1803 char *fname, int flen) 1804 { 1805 struct dentry *dentry, *rdentry; 1806 struct inode *dirp; 1807 __be32 err; 1808 int host_err; 1809 1810 err = nfserr_acces; 1811 if (!flen || isdotent(fname, flen)) 1812 goto out; 1813 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1814 if (err) 1815 goto out; 1816 1817 fh_lock_nested(fhp, I_MUTEX_PARENT); 1818 dentry = fhp->fh_dentry; 1819 dirp = dentry->d_inode; 1820 1821 rdentry = lookup_one_len(fname, dentry, flen); 1822 host_err = PTR_ERR(rdentry); 1823 if (IS_ERR(rdentry)) 1824 goto out_nfserr; 1825 1826 if (!rdentry->d_inode) { 1827 dput(rdentry); 1828 err = nfserr_noent; 1829 goto out; 1830 } 1831 1832 if (!type) 1833 type = rdentry->d_inode->i_mode & S_IFMT; 1834 1835 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1836 if (host_err) 1837 goto out_nfserr; 1838 1839 if (type != S_IFDIR) { /* It's UNLINK */ 1840 #ifdef MSNFS 1841 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1842 (atomic_read(&rdentry->d_count) > 1)) { 1843 host_err = -EPERM; 1844 } else 1845 #endif 1846 host_err = vfs_unlink(dirp, rdentry); 1847 } else { /* It's RMDIR */ 1848 host_err = vfs_rmdir(dirp, rdentry); 1849 } 1850 1851 dput(rdentry); 1852 1853 if (host_err) 1854 goto out_drop; 1855 if (EX_ISSYNC(fhp->fh_export)) 1856 host_err = nfsd_sync_dir(dentry); 1857 1858 out_drop: 1859 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1860 out_nfserr: 1861 err = nfserrno(host_err); 1862 out: 1863 return err; 1864 } 1865 1866 /* 1867 * We do this buffering because we must not call back into the file 1868 * system's ->lookup() method from the filldir callback. That may well 1869 * deadlock a number of file systems. 1870 * 1871 * This is based heavily on the implementation of same in XFS. 1872 */ 1873 struct buffered_dirent { 1874 u64 ino; 1875 loff_t offset; 1876 int namlen; 1877 unsigned int d_type; 1878 char name[]; 1879 }; 1880 1881 struct readdir_data { 1882 char *dirent; 1883 size_t used; 1884 int full; 1885 }; 1886 1887 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1888 loff_t offset, u64 ino, unsigned int d_type) 1889 { 1890 struct readdir_data *buf = __buf; 1891 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1892 unsigned int reclen; 1893 1894 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1895 if (buf->used + reclen > PAGE_SIZE) { 1896 buf->full = 1; 1897 return -EINVAL; 1898 } 1899 1900 de->namlen = namlen; 1901 de->offset = offset; 1902 de->ino = ino; 1903 de->d_type = d_type; 1904 memcpy(de->name, name, namlen); 1905 buf->used += reclen; 1906 1907 return 0; 1908 } 1909 1910 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1911 struct readdir_cd *cdp, loff_t *offsetp) 1912 { 1913 struct readdir_data buf; 1914 struct buffered_dirent *de; 1915 int host_err; 1916 int size; 1917 loff_t offset; 1918 1919 buf.dirent = (void *)__get_free_page(GFP_KERNEL); 1920 if (!buf.dirent) 1921 return nfserrno(-ENOMEM); 1922 1923 offset = *offsetp; 1924 1925 while (1) { 1926 struct inode *dir_inode = file->f_path.dentry->d_inode; 1927 unsigned int reclen; 1928 1929 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1930 buf.used = 0; 1931 buf.full = 0; 1932 1933 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf); 1934 if (buf.full) 1935 host_err = 0; 1936 1937 if (host_err < 0) 1938 break; 1939 1940 size = buf.used; 1941 1942 if (!size) 1943 break; 1944 1945 /* 1946 * Various filldir functions may end up calling back into 1947 * lookup_one_len() and the file system's ->lookup() method. 1948 * These expect i_mutex to be held, as it would within readdir. 1949 */ 1950 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1951 if (host_err) 1952 break; 1953 1954 de = (struct buffered_dirent *)buf.dirent; 1955 while (size > 0) { 1956 offset = de->offset; 1957 1958 if (func(cdp, de->name, de->namlen, de->offset, 1959 de->ino, de->d_type)) 1960 break; 1961 1962 if (cdp->err != nfs_ok) 1963 break; 1964 1965 reclen = ALIGN(sizeof(*de) + de->namlen, 1966 sizeof(u64)); 1967 size -= reclen; 1968 de = (struct buffered_dirent *)((char *)de + reclen); 1969 } 1970 mutex_unlock(&dir_inode->i_mutex); 1971 if (size > 0) /* We bailed out early */ 1972 break; 1973 1974 offset = vfs_llseek(file, 0, SEEK_CUR); 1975 } 1976 1977 free_page((unsigned long)(buf.dirent)); 1978 1979 if (host_err) 1980 return nfserrno(host_err); 1981 1982 *offsetp = offset; 1983 return cdp->err; 1984 } 1985 1986 /* 1987 * Read entries from a directory. 1988 * The NFSv3/4 verifier we ignore for now. 1989 */ 1990 __be32 1991 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 1992 struct readdir_cd *cdp, filldir_t func) 1993 { 1994 __be32 err; 1995 struct file *file; 1996 loff_t offset = *offsetp; 1997 1998 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file); 1999 if (err) 2000 goto out; 2001 2002 offset = vfs_llseek(file, offset, 0); 2003 if (offset < 0) { 2004 err = nfserrno((int)offset); 2005 goto out_close; 2006 } 2007 2008 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 2009 2010 if (err == nfserr_eof || err == nfserr_toosmall) 2011 err = nfs_ok; /* can still be found in ->err */ 2012 out_close: 2013 nfsd_close(file); 2014 out: 2015 return err; 2016 } 2017 2018 /* 2019 * Get file system stats 2020 * N.B. After this call fhp needs an fh_put 2021 */ 2022 __be32 2023 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 2024 { 2025 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 2026 if (!err && vfs_statfs(fhp->fh_dentry,stat)) 2027 err = nfserr_io; 2028 return err; 2029 } 2030 2031 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 2032 { 2033 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 2034 } 2035 2036 /* 2037 * Check for a user's access permissions to this inode. 2038 */ 2039 __be32 2040 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 2041 struct dentry *dentry, int acc) 2042 { 2043 struct inode *inode = dentry->d_inode; 2044 struct path path; 2045 int err; 2046 2047 if (acc == NFSD_MAY_NOP) 2048 return 0; 2049 #if 0 2050 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 2051 acc, 2052 (acc & NFSD_MAY_READ)? " read" : "", 2053 (acc & NFSD_MAY_WRITE)? " write" : "", 2054 (acc & NFSD_MAY_EXEC)? " exec" : "", 2055 (acc & NFSD_MAY_SATTR)? " sattr" : "", 2056 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 2057 (acc & NFSD_MAY_LOCK)? " lock" : "", 2058 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 2059 inode->i_mode, 2060 IS_IMMUTABLE(inode)? " immut" : "", 2061 IS_APPEND(inode)? " append" : "", 2062 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2063 dprintk(" owner %d/%d user %d/%d\n", 2064 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2065 #endif 2066 2067 /* Normally we reject any write/sattr etc access on a read-only file 2068 * system. But if it is IRIX doing check on write-access for a 2069 * device special file, we ignore rofs. 2070 */ 2071 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2072 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2073 if (exp_rdonly(rqstp, exp) || 2074 __mnt_is_readonly(exp->ex_path.mnt)) 2075 return nfserr_rofs; 2076 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2077 return nfserr_perm; 2078 } 2079 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2080 return nfserr_perm; 2081 2082 if (acc & NFSD_MAY_LOCK) { 2083 /* If we cannot rely on authentication in NLM requests, 2084 * just allow locks, otherwise require read permission, or 2085 * ownership 2086 */ 2087 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2088 return 0; 2089 else 2090 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2091 } 2092 /* 2093 * The file owner always gets access permission for accesses that 2094 * would normally be checked at open time. This is to make 2095 * file access work even when the client has done a fchmod(fd, 0). 2096 * 2097 * However, `cp foo bar' should fail nevertheless when bar is 2098 * readonly. A sensible way to do this might be to reject all 2099 * attempts to truncate a read-only file, because a creat() call 2100 * always implies file truncation. 2101 * ... but this isn't really fair. A process may reasonably call 2102 * ftruncate on an open file descriptor on a file with perm 000. 2103 * We must trust the client to do permission checking - using "ACCESS" 2104 * with NFSv3. 2105 */ 2106 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2107 inode->i_uid == current_fsuid()) 2108 return 0; 2109 2110 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2111 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2112 2113 /* Allow read access to binaries even when mode 111 */ 2114 if (err == -EACCES && S_ISREG(inode->i_mode) && 2115 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE)) 2116 err = inode_permission(inode, MAY_EXEC); 2117 if (err) 2118 goto nfsd_out; 2119 2120 /* Do integrity (permission) checking now, but defer incrementing 2121 * IMA counts to the actual file open. 2122 */ 2123 path.mnt = exp->ex_path.mnt; 2124 path.dentry = dentry; 2125 err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC)); 2126 nfsd_out: 2127 return err? nfserrno(err) : 0; 2128 } 2129 2130 void 2131 nfsd_racache_shutdown(void) 2132 { 2133 struct raparms *raparm, *last_raparm; 2134 unsigned int i; 2135 2136 dprintk("nfsd: freeing readahead buffers.\n"); 2137 2138 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2139 raparm = raparm_hash[i].pb_head; 2140 while(raparm) { 2141 last_raparm = raparm; 2142 raparm = raparm->p_next; 2143 kfree(last_raparm); 2144 } 2145 raparm_hash[i].pb_head = NULL; 2146 } 2147 } 2148 /* 2149 * Initialize readahead param cache 2150 */ 2151 int 2152 nfsd_racache_init(int cache_size) 2153 { 2154 int i; 2155 int j = 0; 2156 int nperbucket; 2157 struct raparms **raparm = NULL; 2158 2159 2160 if (raparm_hash[0].pb_head) 2161 return 0; 2162 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2163 if (nperbucket < 2) 2164 nperbucket = 2; 2165 cache_size = nperbucket * RAPARM_HASH_SIZE; 2166 2167 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2168 2169 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2170 spin_lock_init(&raparm_hash[i].pb_lock); 2171 2172 raparm = &raparm_hash[i].pb_head; 2173 for (j = 0; j < nperbucket; j++) { 2174 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2175 if (!*raparm) 2176 goto out_nomem; 2177 raparm = &(*raparm)->p_next; 2178 } 2179 *raparm = NULL; 2180 } 2181 2182 nfsdstats.ra_size = cache_size; 2183 return 0; 2184 2185 out_nomem: 2186 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2187 nfsd_racache_shutdown(); 2188 return -ENOMEM; 2189 } 2190 2191 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 2192 struct posix_acl * 2193 nfsd_get_posix_acl(struct svc_fh *fhp, int type) 2194 { 2195 struct inode *inode = fhp->fh_dentry->d_inode; 2196 char *name; 2197 void *value = NULL; 2198 ssize_t size; 2199 struct posix_acl *acl; 2200 2201 if (!IS_POSIXACL(inode)) 2202 return ERR_PTR(-EOPNOTSUPP); 2203 2204 switch (type) { 2205 case ACL_TYPE_ACCESS: 2206 name = POSIX_ACL_XATTR_ACCESS; 2207 break; 2208 case ACL_TYPE_DEFAULT: 2209 name = POSIX_ACL_XATTR_DEFAULT; 2210 break; 2211 default: 2212 return ERR_PTR(-EOPNOTSUPP); 2213 } 2214 2215 size = nfsd_getxattr(fhp->fh_dentry, name, &value); 2216 if (size < 0) 2217 return ERR_PTR(size); 2218 2219 acl = posix_acl_from_xattr(value, size); 2220 kfree(value); 2221 return acl; 2222 } 2223 2224 int 2225 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) 2226 { 2227 struct inode *inode = fhp->fh_dentry->d_inode; 2228 char *name; 2229 void *value = NULL; 2230 size_t size; 2231 int error; 2232 2233 if (!IS_POSIXACL(inode) || 2234 !inode->i_op->setxattr || !inode->i_op->removexattr) 2235 return -EOPNOTSUPP; 2236 switch(type) { 2237 case ACL_TYPE_ACCESS: 2238 name = POSIX_ACL_XATTR_ACCESS; 2239 break; 2240 case ACL_TYPE_DEFAULT: 2241 name = POSIX_ACL_XATTR_DEFAULT; 2242 break; 2243 default: 2244 return -EOPNOTSUPP; 2245 } 2246 2247 if (acl && acl->a_count) { 2248 size = posix_acl_xattr_size(acl->a_count); 2249 value = kmalloc(size, GFP_KERNEL); 2250 if (!value) 2251 return -ENOMEM; 2252 error = posix_acl_to_xattr(acl, value, size); 2253 if (error < 0) 2254 goto getout; 2255 size = error; 2256 } else 2257 size = 0; 2258 2259 error = mnt_want_write(fhp->fh_export->ex_path.mnt); 2260 if (error) 2261 goto getout; 2262 if (size) 2263 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); 2264 else { 2265 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) 2266 error = 0; 2267 else { 2268 error = vfs_removexattr(fhp->fh_dentry, name); 2269 if (error == -ENODATA) 2270 error = 0; 2271 } 2272 } 2273 mnt_drop_write(fhp->fh_export->ex_path.mnt); 2274 2275 getout: 2276 kfree(value); 2277 return error; 2278 } 2279 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 2280