1 #define MSNFS /* HACK HACK */ 2 /* 3 * linux/fs/nfsd/vfs.c 4 * 5 * File operations used by nfsd. Some of these have been ripped from 6 * other parts of the kernel because they weren't exported, others 7 * are partial duplicates with added or changed functionality. 8 * 9 * Note that several functions dget() the dentry upon which they want 10 * to act, most notably those that create directory entries. Response 11 * dentry's are dput()'d if necessary in the release callback. 12 * So if you notice code paths that apparently fail to dput() the 13 * dentry, don't worry--they have been taken care of. 14 * 15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de> 16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp> 17 */ 18 19 #include <linux/string.h> 20 #include <linux/time.h> 21 #include <linux/errno.h> 22 #include <linux/fs.h> 23 #include <linux/file.h> 24 #include <linux/mount.h> 25 #include <linux/major.h> 26 #include <linux/splice.h> 27 #include <linux/proc_fs.h> 28 #include <linux/stat.h> 29 #include <linux/fcntl.h> 30 #include <linux/net.h> 31 #include <linux/unistd.h> 32 #include <linux/slab.h> 33 #include <linux/pagemap.h> 34 #include <linux/in.h> 35 #include <linux/module.h> 36 #include <linux/namei.h> 37 #include <linux/vfs.h> 38 #include <linux/delay.h> 39 #include <linux/sunrpc/svc.h> 40 #include <linux/nfsd/nfsd.h> 41 #ifdef CONFIG_NFSD_V3 42 #include <linux/nfs3.h> 43 #include <linux/nfsd/xdr3.h> 44 #endif /* CONFIG_NFSD_V3 */ 45 #include <linux/nfsd/nfsfh.h> 46 #include <linux/quotaops.h> 47 #include <linux/fsnotify.h> 48 #include <linux/posix_acl.h> 49 #include <linux/posix_acl_xattr.h> 50 #include <linux/xattr.h> 51 #ifdef CONFIG_NFSD_V4 52 #include <linux/nfs4.h> 53 #include <linux/nfs4_acl.h> 54 #include <linux/nfsd_idmap.h> 55 #include <linux/security.h> 56 #endif /* CONFIG_NFSD_V4 */ 57 #include <linux/jhash.h> 58 #include <linux/ima.h> 59 60 #include <asm/uaccess.h> 61 62 #define NFSDDBG_FACILITY NFSDDBG_FILEOP 63 64 65 /* 66 * This is a cache of readahead params that help us choose the proper 67 * readahead strategy. Initially, we set all readahead parameters to 0 68 * and let the VFS handle things. 69 * If you increase the number of cached files very much, you'll need to 70 * add a hash table here. 71 */ 72 struct raparms { 73 struct raparms *p_next; 74 unsigned int p_count; 75 ino_t p_ino; 76 dev_t p_dev; 77 int p_set; 78 struct file_ra_state p_ra; 79 unsigned int p_hindex; 80 }; 81 82 struct raparm_hbucket { 83 struct raparms *pb_head; 84 spinlock_t pb_lock; 85 } ____cacheline_aligned_in_smp; 86 87 #define RAPARM_HASH_BITS 4 88 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) 89 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) 90 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; 91 92 static inline int 93 nfsd_v4client(struct svc_rqst *rq) 94 { 95 return rq->rq_prog == NFS_PROGRAM && rq->rq_vers == 4; 96 } 97 98 /* 99 * Called from nfsd_lookup and encode_dirent. Check if we have crossed 100 * a mount point. 101 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged, 102 * or nfs_ok having possibly changed *dpp and *expp 103 */ 104 int 105 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 106 struct svc_export **expp) 107 { 108 struct svc_export *exp = *expp, *exp2 = NULL; 109 struct dentry *dentry = *dpp; 110 struct path path = {.mnt = mntget(exp->ex_path.mnt), 111 .dentry = dget(dentry)}; 112 int err = 0; 113 114 while (d_mountpoint(path.dentry) && follow_down(&path)) 115 ; 116 117 exp2 = rqst_exp_get_by_name(rqstp, &path); 118 if (IS_ERR(exp2)) { 119 if (PTR_ERR(exp2) != -ENOENT) 120 err = PTR_ERR(exp2); 121 path_put(&path); 122 goto out; 123 } 124 if (nfsd_v4client(rqstp) || 125 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) { 126 /* successfully crossed mount point */ 127 /* 128 * This is subtle: path.dentry is *not* on path.mnt 129 * at this point. The only reason we are safe is that 130 * original mnt is pinned down by exp, so we should 131 * put path *before* putting exp 132 */ 133 *dpp = path.dentry; 134 path.dentry = dentry; 135 *expp = exp2; 136 exp2 = exp; 137 } 138 path_put(&path); 139 exp_put(exp2); 140 out: 141 return err; 142 } 143 144 __be32 145 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp, 146 const char *name, unsigned int len, 147 struct svc_export **exp_ret, struct dentry **dentry_ret) 148 { 149 struct svc_export *exp; 150 struct dentry *dparent; 151 struct dentry *dentry; 152 __be32 err; 153 int host_err; 154 155 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); 156 157 /* Obtain dentry and export. */ 158 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 159 if (err) 160 return err; 161 162 dparent = fhp->fh_dentry; 163 exp = fhp->fh_export; 164 exp_get(exp); 165 166 /* Lookup the name, but don't follow links */ 167 if (isdotent(name, len)) { 168 if (len==1) 169 dentry = dget(dparent); 170 else if (dparent != exp->ex_path.dentry) 171 dentry = dget_parent(dparent); 172 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp)) 173 dentry = dget(dparent); /* .. == . just like at / */ 174 else { 175 /* checking mountpoint crossing is very different when stepping up */ 176 struct svc_export *exp2 = NULL; 177 struct dentry *dp; 178 struct path path = {.mnt = mntget(exp->ex_path.mnt), 179 .dentry = dget(dparent)}; 180 181 while (path.dentry == path.mnt->mnt_root && 182 follow_up(&path)) 183 ; 184 dp = dget_parent(path.dentry); 185 dput(path.dentry); 186 path.dentry = dp; 187 188 exp2 = rqst_exp_parent(rqstp, &path); 189 if (PTR_ERR(exp2) == -ENOENT) { 190 dentry = dget(dparent); 191 } else if (IS_ERR(exp2)) { 192 host_err = PTR_ERR(exp2); 193 path_put(&path); 194 goto out_nfserr; 195 } else { 196 dentry = dget(path.dentry); 197 exp_put(exp); 198 exp = exp2; 199 } 200 path_put(&path); 201 } 202 } else { 203 fh_lock(fhp); 204 dentry = lookup_one_len(name, dparent, len); 205 host_err = PTR_ERR(dentry); 206 if (IS_ERR(dentry)) 207 goto out_nfserr; 208 /* 209 * check if we have crossed a mount point ... 210 */ 211 if (d_mountpoint(dentry)) { 212 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { 213 dput(dentry); 214 goto out_nfserr; 215 } 216 } 217 } 218 *dentry_ret = dentry; 219 *exp_ret = exp; 220 return 0; 221 222 out_nfserr: 223 exp_put(exp); 224 return nfserrno(host_err); 225 } 226 227 /* 228 * Look up one component of a pathname. 229 * N.B. After this call _both_ fhp and resfh need an fh_put 230 * 231 * If the lookup would cross a mountpoint, and the mounted filesystem 232 * is exported to the client with NFSEXP_NOHIDE, then the lookup is 233 * accepted as it stands and the mounted directory is 234 * returned. Otherwise the covered directory is returned. 235 * NOTE: this mountpoint crossing is not supported properly by all 236 * clients and is explicitly disallowed for NFSv3 237 * NeilBrown <neilb@cse.unsw.edu.au> 238 */ 239 __be32 240 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, 241 unsigned int len, struct svc_fh *resfh) 242 { 243 struct svc_export *exp; 244 struct dentry *dentry; 245 __be32 err; 246 247 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry); 248 if (err) 249 return err; 250 err = check_nfsd_access(exp, rqstp); 251 if (err) 252 goto out; 253 /* 254 * Note: we compose the file handle now, but as the 255 * dentry may be negative, it may need to be updated. 256 */ 257 err = fh_compose(resfh, exp, dentry, fhp); 258 if (!err && !dentry->d_inode) 259 err = nfserr_noent; 260 out: 261 dput(dentry); 262 exp_put(exp); 263 return err; 264 } 265 266 267 /* 268 * Set various file attributes. 269 * N.B. After this call fhp needs an fh_put 270 */ 271 __be32 272 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, 273 int check_guard, time_t guardtime) 274 { 275 struct dentry *dentry; 276 struct inode *inode; 277 int accmode = NFSD_MAY_SATTR; 278 int ftype = 0; 279 __be32 err; 280 int host_err; 281 int size_change = 0; 282 283 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) 284 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE; 285 if (iap->ia_valid & ATTR_SIZE) 286 ftype = S_IFREG; 287 288 /* Get inode */ 289 err = fh_verify(rqstp, fhp, ftype, accmode); 290 if (err) 291 goto out; 292 293 dentry = fhp->fh_dentry; 294 inode = dentry->d_inode; 295 296 /* Ignore any mode updates on symlinks */ 297 if (S_ISLNK(inode->i_mode)) 298 iap->ia_valid &= ~ATTR_MODE; 299 300 if (!iap->ia_valid) 301 goto out; 302 303 /* 304 * NFSv2 does not differentiate between "set-[ac]time-to-now" 305 * which only requires access, and "set-[ac]time-to-X" which 306 * requires ownership. 307 * So if it looks like it might be "set both to the same time which 308 * is close to now", and if inode_change_ok fails, then we 309 * convert to "set to now" instead of "set to explicit time" 310 * 311 * We only call inode_change_ok as the last test as technically 312 * it is not an interface that we should be using. It is only 313 * valid if the filesystem does not define it's own i_op->setattr. 314 */ 315 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) 316 #define MAX_TOUCH_TIME_ERROR (30*60) 317 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET && 318 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) { 319 /* 320 * Looks probable. 321 * 322 * Now just make sure time is in the right ballpark. 323 * Solaris, at least, doesn't seem to care what the time 324 * request is. We require it be within 30 minutes of now. 325 */ 326 time_t delta = iap->ia_atime.tv_sec - get_seconds(); 327 if (delta < 0) 328 delta = -delta; 329 if (delta < MAX_TOUCH_TIME_ERROR && 330 inode_change_ok(inode, iap) != 0) { 331 /* 332 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME. 333 * This will cause notify_change to set these times 334 * to "now" 335 */ 336 iap->ia_valid &= ~BOTH_TIME_SET; 337 } 338 } 339 340 /* 341 * The size case is special. 342 * It changes the file as well as the attributes. 343 */ 344 if (iap->ia_valid & ATTR_SIZE) { 345 if (iap->ia_size < inode->i_size) { 346 err = nfsd_permission(rqstp, fhp->fh_export, dentry, 347 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE); 348 if (err) 349 goto out; 350 } 351 352 /* 353 * If we are changing the size of the file, then 354 * we need to break all leases. 355 */ 356 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK); 357 if (host_err == -EWOULDBLOCK) 358 host_err = -ETIMEDOUT; 359 if (host_err) /* ENOMEM or EWOULDBLOCK */ 360 goto out_nfserr; 361 362 host_err = get_write_access(inode); 363 if (host_err) 364 goto out_nfserr; 365 366 size_change = 1; 367 host_err = locks_verify_truncate(inode, NULL, iap->ia_size); 368 if (host_err) { 369 put_write_access(inode); 370 goto out_nfserr; 371 } 372 vfs_dq_init(inode); 373 } 374 375 /* sanitize the mode change */ 376 if (iap->ia_valid & ATTR_MODE) { 377 iap->ia_mode &= S_IALLUGO; 378 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO); 379 } 380 381 /* Revoke setuid/setgid on chown */ 382 if (!S_ISDIR(inode->i_mode) && 383 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) || 384 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) { 385 iap->ia_valid |= ATTR_KILL_PRIV; 386 if (iap->ia_valid & ATTR_MODE) { 387 /* we're setting mode too, just clear the s*id bits */ 388 iap->ia_mode &= ~S_ISUID; 389 if (iap->ia_mode & S_IXGRP) 390 iap->ia_mode &= ~S_ISGID; 391 } else { 392 /* set ATTR_KILL_* bits and let VFS handle it */ 393 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID); 394 } 395 } 396 397 /* Change the attributes. */ 398 399 iap->ia_valid |= ATTR_CTIME; 400 401 err = nfserr_notsync; 402 if (!check_guard || guardtime == inode->i_ctime.tv_sec) { 403 fh_lock(fhp); 404 host_err = notify_change(dentry, iap); 405 err = nfserrno(host_err); 406 fh_unlock(fhp); 407 } 408 if (size_change) 409 put_write_access(inode); 410 if (!err) 411 if (EX_ISSYNC(fhp->fh_export)) 412 write_inode_now(inode, 1); 413 out: 414 return err; 415 416 out_nfserr: 417 err = nfserrno(host_err); 418 goto out; 419 } 420 421 #if defined(CONFIG_NFSD_V2_ACL) || \ 422 defined(CONFIG_NFSD_V3_ACL) || \ 423 defined(CONFIG_NFSD_V4) 424 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf) 425 { 426 ssize_t buflen; 427 ssize_t ret; 428 429 buflen = vfs_getxattr(dentry, key, NULL, 0); 430 if (buflen <= 0) 431 return buflen; 432 433 *buf = kmalloc(buflen, GFP_KERNEL); 434 if (!*buf) 435 return -ENOMEM; 436 437 ret = vfs_getxattr(dentry, key, *buf, buflen); 438 if (ret < 0) 439 kfree(*buf); 440 return ret; 441 } 442 #endif 443 444 #if defined(CONFIG_NFSD_V4) 445 static int 446 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) 447 { 448 int len; 449 size_t buflen; 450 char *buf = NULL; 451 int error = 0; 452 453 buflen = posix_acl_xattr_size(pacl->a_count); 454 buf = kmalloc(buflen, GFP_KERNEL); 455 error = -ENOMEM; 456 if (buf == NULL) 457 goto out; 458 459 len = posix_acl_to_xattr(pacl, buf, buflen); 460 if (len < 0) { 461 error = len; 462 goto out; 463 } 464 465 error = vfs_setxattr(dentry, key, buf, len, 0); 466 out: 467 kfree(buf); 468 return error; 469 } 470 471 __be32 472 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, 473 struct nfs4_acl *acl) 474 { 475 __be32 error; 476 int host_error; 477 struct dentry *dentry; 478 struct inode *inode; 479 struct posix_acl *pacl = NULL, *dpacl = NULL; 480 unsigned int flags = 0; 481 482 /* Get inode */ 483 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR); 484 if (error) 485 return error; 486 487 dentry = fhp->fh_dentry; 488 inode = dentry->d_inode; 489 if (S_ISDIR(inode->i_mode)) 490 flags = NFS4_ACL_DIR; 491 492 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); 493 if (host_error == -EINVAL) { 494 return nfserr_attrnotsupp; 495 } else if (host_error < 0) 496 goto out_nfserr; 497 498 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); 499 if (host_error < 0) 500 goto out_release; 501 502 if (S_ISDIR(inode->i_mode)) 503 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); 504 505 out_release: 506 posix_acl_release(pacl); 507 posix_acl_release(dpacl); 508 out_nfserr: 509 if (host_error == -EOPNOTSUPP) 510 return nfserr_attrnotsupp; 511 else 512 return nfserrno(host_error); 513 } 514 515 static struct posix_acl * 516 _get_posix_acl(struct dentry *dentry, char *key) 517 { 518 void *buf = NULL; 519 struct posix_acl *pacl = NULL; 520 int buflen; 521 522 buflen = nfsd_getxattr(dentry, key, &buf); 523 if (!buflen) 524 buflen = -ENODATA; 525 if (buflen <= 0) 526 return ERR_PTR(buflen); 527 528 pacl = posix_acl_from_xattr(buf, buflen); 529 kfree(buf); 530 return pacl; 531 } 532 533 int 534 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) 535 { 536 struct inode *inode = dentry->d_inode; 537 int error = 0; 538 struct posix_acl *pacl = NULL, *dpacl = NULL; 539 unsigned int flags = 0; 540 541 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); 542 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) 543 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); 544 if (IS_ERR(pacl)) { 545 error = PTR_ERR(pacl); 546 pacl = NULL; 547 goto out; 548 } 549 550 if (S_ISDIR(inode->i_mode)) { 551 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); 552 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) 553 dpacl = NULL; 554 else if (IS_ERR(dpacl)) { 555 error = PTR_ERR(dpacl); 556 dpacl = NULL; 557 goto out; 558 } 559 flags = NFS4_ACL_DIR; 560 } 561 562 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); 563 if (IS_ERR(*acl)) { 564 error = PTR_ERR(*acl); 565 *acl = NULL; 566 } 567 out: 568 posix_acl_release(pacl); 569 posix_acl_release(dpacl); 570 return error; 571 } 572 573 #endif /* defined(CONFIG_NFS_V4) */ 574 575 #ifdef CONFIG_NFSD_V3 576 /* 577 * Check server access rights to a file system object 578 */ 579 struct accessmap { 580 u32 access; 581 int how; 582 }; 583 static struct accessmap nfs3_regaccess[] = { 584 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 585 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 586 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, 587 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, 588 589 { 0, 0 } 590 }; 591 592 static struct accessmap nfs3_diraccess[] = { 593 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 594 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, 595 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, 596 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, 597 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, 598 599 { 0, 0 } 600 }; 601 602 static struct accessmap nfs3_anyaccess[] = { 603 /* Some clients - Solaris 2.6 at least, make an access call 604 * to the server to check for access for things like /dev/null 605 * (which really, the server doesn't care about). So 606 * We provide simple access checking for them, looking 607 * mainly at mode bits, and we make sure to ignore read-only 608 * filesystem checks 609 */ 610 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 611 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 612 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 613 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 614 615 { 0, 0 } 616 }; 617 618 __be32 619 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) 620 { 621 struct accessmap *map; 622 struct svc_export *export; 623 struct dentry *dentry; 624 u32 query, result = 0, sresult = 0; 625 __be32 error; 626 627 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); 628 if (error) 629 goto out; 630 631 export = fhp->fh_export; 632 dentry = fhp->fh_dentry; 633 634 if (S_ISREG(dentry->d_inode->i_mode)) 635 map = nfs3_regaccess; 636 else if (S_ISDIR(dentry->d_inode->i_mode)) 637 map = nfs3_diraccess; 638 else 639 map = nfs3_anyaccess; 640 641 642 query = *access; 643 for (; map->access; map++) { 644 if (map->access & query) { 645 __be32 err2; 646 647 sresult |= map->access; 648 649 err2 = nfsd_permission(rqstp, export, dentry, map->how); 650 switch (err2) { 651 case nfs_ok: 652 result |= map->access; 653 break; 654 655 /* the following error codes just mean the access was not allowed, 656 * rather than an error occurred */ 657 case nfserr_rofs: 658 case nfserr_acces: 659 case nfserr_perm: 660 /* simply don't "or" in the access bit. */ 661 break; 662 default: 663 error = err2; 664 goto out; 665 } 666 } 667 } 668 *access = result; 669 if (supported) 670 *supported = sresult; 671 672 out: 673 return error; 674 } 675 #endif /* CONFIG_NFSD_V3 */ 676 677 678 679 /* 680 * Open an existing file or directory. 681 * The access argument indicates the type of open (read/write/lock) 682 * N.B. After this call fhp needs an fh_put 683 */ 684 __be32 685 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 686 int access, struct file **filp) 687 { 688 struct dentry *dentry; 689 struct inode *inode; 690 int flags = O_RDONLY|O_LARGEFILE; 691 __be32 err; 692 int host_err; 693 694 validate_process_creds(); 695 696 /* 697 * If we get here, then the client has already done an "open", 698 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 699 * in case a chmod has now revoked permission. 700 */ 701 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE); 702 if (err) 703 goto out; 704 705 dentry = fhp->fh_dentry; 706 inode = dentry->d_inode; 707 708 /* Disallow write access to files with the append-only bit set 709 * or any access when mandatory locking enabled 710 */ 711 err = nfserr_perm; 712 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE)) 713 goto out; 714 /* 715 * We must ignore files (but only files) which might have mandatory 716 * locks on them because there is no way to know if the accesser has 717 * the lock. 718 */ 719 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) 720 goto out; 721 722 if (!inode->i_fop) 723 goto out; 724 725 /* 726 * Check to see if there are any leases on this file. 727 * This may block while leases are broken. 728 */ 729 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0)); 730 if (host_err == -EWOULDBLOCK) 731 host_err = -ETIMEDOUT; 732 if (host_err) /* NOMEM or WOULDBLOCK */ 733 goto out_nfserr; 734 735 if (access & NFSD_MAY_WRITE) { 736 if (access & NFSD_MAY_READ) 737 flags = O_RDWR|O_LARGEFILE; 738 else 739 flags = O_WRONLY|O_LARGEFILE; 740 741 vfs_dq_init(inode); 742 } 743 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt), 744 flags, current_cred()); 745 if (IS_ERR(*filp)) 746 host_err = PTR_ERR(*filp); 747 else 748 ima_counts_get(*filp); 749 out_nfserr: 750 err = nfserrno(host_err); 751 out: 752 validate_process_creds(); 753 return err; 754 } 755 756 /* 757 * Close a file. 758 */ 759 void 760 nfsd_close(struct file *filp) 761 { 762 fput(filp); 763 } 764 765 /* 766 * Sync a file 767 * As this calls fsync (not fdatasync) there is no need for a write_inode 768 * after it. 769 */ 770 static inline int nfsd_dosync(struct file *filp, struct dentry *dp, 771 const struct file_operations *fop) 772 { 773 struct inode *inode = dp->d_inode; 774 int (*fsync) (struct file *, struct dentry *, int); 775 int err; 776 777 err = filemap_fdatawrite(inode->i_mapping); 778 if (err == 0 && fop && (fsync = fop->fsync)) 779 err = fsync(filp, dp, 0); 780 if (err == 0) 781 err = filemap_fdatawait(inode->i_mapping); 782 783 return err; 784 } 785 786 static int 787 nfsd_sync(struct file *filp) 788 { 789 int err; 790 struct inode *inode = filp->f_path.dentry->d_inode; 791 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name); 792 mutex_lock(&inode->i_mutex); 793 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op); 794 mutex_unlock(&inode->i_mutex); 795 796 return err; 797 } 798 799 int 800 nfsd_sync_dir(struct dentry *dp) 801 { 802 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop); 803 } 804 805 /* 806 * Obtain the readahead parameters for the file 807 * specified by (dev, ino). 808 */ 809 810 static inline struct raparms * 811 nfsd_get_raparms(dev_t dev, ino_t ino) 812 { 813 struct raparms *ra, **rap, **frap = NULL; 814 int depth = 0; 815 unsigned int hash; 816 struct raparm_hbucket *rab; 817 818 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 819 rab = &raparm_hash[hash]; 820 821 spin_lock(&rab->pb_lock); 822 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 823 if (ra->p_ino == ino && ra->p_dev == dev) 824 goto found; 825 depth++; 826 if (ra->p_count == 0) 827 frap = rap; 828 } 829 depth = nfsdstats.ra_size*11/10; 830 if (!frap) { 831 spin_unlock(&rab->pb_lock); 832 return NULL; 833 } 834 rap = frap; 835 ra = *frap; 836 ra->p_dev = dev; 837 ra->p_ino = ino; 838 ra->p_set = 0; 839 ra->p_hindex = hash; 840 found: 841 if (rap != &rab->pb_head) { 842 *rap = ra->p_next; 843 ra->p_next = rab->pb_head; 844 rab->pb_head = ra; 845 } 846 ra->p_count++; 847 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 848 spin_unlock(&rab->pb_lock); 849 return ra; 850 } 851 852 /* 853 * Grab and keep cached pages associated with a file in the svc_rqst 854 * so that they can be passed to the network sendmsg/sendpage routines 855 * directly. They will be released after the sending has completed. 856 */ 857 static int 858 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 859 struct splice_desc *sd) 860 { 861 struct svc_rqst *rqstp = sd->u.data; 862 struct page **pp = rqstp->rq_respages + rqstp->rq_resused; 863 struct page *page = buf->page; 864 size_t size; 865 int ret; 866 867 ret = buf->ops->confirm(pipe, buf); 868 if (unlikely(ret)) 869 return ret; 870 871 size = sd->len; 872 873 if (rqstp->rq_res.page_len == 0) { 874 get_page(page); 875 put_page(*pp); 876 *pp = page; 877 rqstp->rq_resused++; 878 rqstp->rq_res.page_base = buf->offset; 879 rqstp->rq_res.page_len = size; 880 } else if (page != pp[-1]) { 881 get_page(page); 882 if (*pp) 883 put_page(*pp); 884 *pp = page; 885 rqstp->rq_resused++; 886 rqstp->rq_res.page_len += size; 887 } else 888 rqstp->rq_res.page_len += size; 889 890 return size; 891 } 892 893 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 894 struct splice_desc *sd) 895 { 896 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 897 } 898 899 static inline int svc_msnfs(struct svc_fh *ffhp) 900 { 901 #ifdef MSNFS 902 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS); 903 #else 904 return 0; 905 #endif 906 } 907 908 static __be32 909 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 910 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 911 { 912 struct inode *inode; 913 struct raparms *ra; 914 mm_segment_t oldfs; 915 __be32 err; 916 int host_err; 917 918 err = nfserr_perm; 919 inode = file->f_path.dentry->d_inode; 920 921 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count)) 922 goto out; 923 924 /* Get readahead parameters */ 925 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 926 927 if (ra && ra->p_set) 928 file->f_ra = ra->p_ra; 929 930 if (file->f_op->splice_read && rqstp->rq_splice_ok) { 931 struct splice_desc sd = { 932 .len = 0, 933 .total_len = *count, 934 .pos = offset, 935 .u.data = rqstp, 936 }; 937 938 rqstp->rq_resused = 1; 939 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 940 } else { 941 oldfs = get_fs(); 942 set_fs(KERNEL_DS); 943 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 944 set_fs(oldfs); 945 } 946 947 /* Write back readahead params */ 948 if (ra) { 949 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 950 spin_lock(&rab->pb_lock); 951 ra->p_ra = file->f_ra; 952 ra->p_set = 1; 953 ra->p_count--; 954 spin_unlock(&rab->pb_lock); 955 } 956 957 if (host_err >= 0) { 958 nfsdstats.io_read += host_err; 959 *count = host_err; 960 err = 0; 961 fsnotify_access(file->f_path.dentry); 962 } else 963 err = nfserrno(host_err); 964 out: 965 return err; 966 } 967 968 static void kill_suid(struct dentry *dentry) 969 { 970 struct iattr ia; 971 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 972 973 mutex_lock(&dentry->d_inode->i_mutex); 974 notify_change(dentry, &ia); 975 mutex_unlock(&dentry->d_inode->i_mutex); 976 } 977 978 /* 979 * Gathered writes: If another process is currently writing to the file, 980 * there's a high chance this is another nfsd (triggered by a bulk write 981 * from a client's biod). Rather than syncing the file with each write 982 * request, we sleep for 10 msec. 983 * 984 * I don't know if this roughly approximates C. Juszak's idea of 985 * gathered writes, but it's a nice and simple solution (IMHO), and it 986 * seems to work:-) 987 * 988 * Note: we do this only in the NFSv2 case, since v3 and higher have a 989 * better tool (separate unstable writes and commits) for solving this 990 * problem. 991 */ 992 static int wait_for_concurrent_writes(struct file *file) 993 { 994 struct inode *inode = file->f_path.dentry->d_inode; 995 static ino_t last_ino; 996 static dev_t last_dev; 997 int err = 0; 998 999 if (atomic_read(&inode->i_writecount) > 1 1000 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 1001 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 1002 msleep(10); 1003 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 1004 } 1005 1006 if (inode->i_state & I_DIRTY) { 1007 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 1008 err = nfsd_sync(file); 1009 } 1010 last_ino = inode->i_ino; 1011 last_dev = inode->i_sb->s_dev; 1012 return err; 1013 } 1014 1015 static __be32 1016 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1017 loff_t offset, struct kvec *vec, int vlen, 1018 unsigned long *cnt, int *stablep) 1019 { 1020 struct svc_export *exp; 1021 struct dentry *dentry; 1022 struct inode *inode; 1023 mm_segment_t oldfs; 1024 __be32 err = 0; 1025 int host_err; 1026 int stable = *stablep; 1027 int use_wgather; 1028 1029 #ifdef MSNFS 1030 err = nfserr_perm; 1031 1032 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1033 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt))) 1034 goto out; 1035 #endif 1036 1037 dentry = file->f_path.dentry; 1038 inode = dentry->d_inode; 1039 exp = fhp->fh_export; 1040 1041 /* 1042 * Request sync writes if 1043 * - the sync export option has been set, or 1044 * - the client requested O_SYNC behavior (NFSv3 feature). 1045 * - The file system doesn't support fsync(). 1046 * When NFSv2 gathered writes have been configured for this volume, 1047 * flushing the data to disk is handled separately below. 1048 */ 1049 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 1050 1051 if (!file->f_op->fsync) {/* COMMIT3 cannot work */ 1052 stable = 2; 1053 *stablep = 2; /* FILE_SYNC */ 1054 } 1055 1056 if (!EX_ISSYNC(exp)) 1057 stable = 0; 1058 if (stable && !use_wgather) { 1059 spin_lock(&file->f_lock); 1060 file->f_flags |= O_SYNC; 1061 spin_unlock(&file->f_lock); 1062 } 1063 1064 /* Write the data. */ 1065 oldfs = get_fs(); set_fs(KERNEL_DS); 1066 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); 1067 set_fs(oldfs); 1068 if (host_err < 0) 1069 goto out_nfserr; 1070 *cnt = host_err; 1071 nfsdstats.io_write += host_err; 1072 fsnotify_modify(file->f_path.dentry); 1073 1074 /* clear setuid/setgid flag after write */ 1075 if (inode->i_mode & (S_ISUID | S_ISGID)) 1076 kill_suid(dentry); 1077 1078 if (stable && use_wgather) 1079 host_err = wait_for_concurrent_writes(file); 1080 1081 out_nfserr: 1082 dprintk("nfsd: write complete host_err=%d\n", host_err); 1083 if (host_err >= 0) 1084 err = 0; 1085 else 1086 err = nfserrno(host_err); 1087 out: 1088 return err; 1089 } 1090 1091 /* 1092 * Read data from a file. count must contain the requested read count 1093 * on entry. On return, *count contains the number of bytes actually read. 1094 * N.B. After this call fhp needs an fh_put 1095 */ 1096 __be32 1097 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1098 loff_t offset, struct kvec *vec, int vlen, 1099 unsigned long *count) 1100 { 1101 __be32 err; 1102 1103 if (file) { 1104 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1105 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); 1106 if (err) 1107 goto out; 1108 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1109 } else { 1110 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); 1111 if (err) 1112 goto out; 1113 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1114 nfsd_close(file); 1115 } 1116 out: 1117 return err; 1118 } 1119 1120 /* 1121 * Write data to a file. 1122 * The stable flag requests synchronous writes. 1123 * N.B. After this call fhp needs an fh_put 1124 */ 1125 __be32 1126 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1127 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1128 int *stablep) 1129 { 1130 __be32 err = 0; 1131 1132 if (file) { 1133 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1134 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1135 if (err) 1136 goto out; 1137 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1138 stablep); 1139 } else { 1140 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1141 if (err) 1142 goto out; 1143 1144 if (cnt) 1145 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1146 cnt, stablep); 1147 nfsd_close(file); 1148 } 1149 out: 1150 return err; 1151 } 1152 1153 #ifdef CONFIG_NFSD_V3 1154 /* 1155 * Commit all pending writes to stable storage. 1156 * Strictly speaking, we could sync just the indicated file region here, 1157 * but there's currently no way we can ask the VFS to do so. 1158 * 1159 * Unfortunately we cannot lock the file to make sure we return full WCC 1160 * data to the client, as locking happens lower down in the filesystem. 1161 */ 1162 __be32 1163 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1164 loff_t offset, unsigned long count) 1165 { 1166 struct file *file; 1167 __be32 err; 1168 1169 if ((u64)count > ~(u64)offset) 1170 return nfserr_inval; 1171 1172 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1173 if (err) 1174 return err; 1175 if (EX_ISSYNC(fhp->fh_export)) { 1176 if (file->f_op && file->f_op->fsync) { 1177 err = nfserrno(nfsd_sync(file)); 1178 } else { 1179 err = nfserr_notsupp; 1180 } 1181 } 1182 1183 nfsd_close(file); 1184 return err; 1185 } 1186 #endif /* CONFIG_NFSD_V3 */ 1187 1188 static __be32 1189 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1190 struct iattr *iap) 1191 { 1192 /* 1193 * Mode has already been set earlier in create: 1194 */ 1195 iap->ia_valid &= ~ATTR_MODE; 1196 /* 1197 * Setting uid/gid works only for root. Irix appears to 1198 * send along the gid on create when it tries to implement 1199 * setgid directories via NFS: 1200 */ 1201 if (current_fsuid() != 0) 1202 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1203 if (iap->ia_valid) 1204 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1205 return 0; 1206 } 1207 1208 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1209 * setting size to 0 may fail for some specific file systems by the permission 1210 * checking which requires WRITE permission but the mode is 000. 1211 * we ignore the resizing(to 0) on the just new created file, since the size is 1212 * 0 after file created. 1213 * 1214 * call this only after vfs_create() is called. 1215 * */ 1216 static void 1217 nfsd_check_ignore_resizing(struct iattr *iap) 1218 { 1219 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1220 iap->ia_valid &= ~ATTR_SIZE; 1221 } 1222 1223 /* 1224 * Create a file (regular, directory, device, fifo); UNIX sockets 1225 * not yet implemented. 1226 * If the response fh has been verified, the parent directory should 1227 * already be locked. Note that the parent directory is left locked. 1228 * 1229 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1230 */ 1231 __be32 1232 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1233 char *fname, int flen, struct iattr *iap, 1234 int type, dev_t rdev, struct svc_fh *resfhp) 1235 { 1236 struct dentry *dentry, *dchild = NULL; 1237 struct inode *dirp; 1238 __be32 err; 1239 __be32 err2; 1240 int host_err; 1241 1242 err = nfserr_perm; 1243 if (!flen) 1244 goto out; 1245 err = nfserr_exist; 1246 if (isdotent(fname, flen)) 1247 goto out; 1248 1249 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1250 if (err) 1251 goto out; 1252 1253 dentry = fhp->fh_dentry; 1254 dirp = dentry->d_inode; 1255 1256 err = nfserr_notdir; 1257 if (!dirp->i_op->lookup) 1258 goto out; 1259 /* 1260 * Check whether the response file handle has been verified yet. 1261 * If it has, the parent directory should already be locked. 1262 */ 1263 if (!resfhp->fh_dentry) { 1264 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1265 fh_lock_nested(fhp, I_MUTEX_PARENT); 1266 dchild = lookup_one_len(fname, dentry, flen); 1267 host_err = PTR_ERR(dchild); 1268 if (IS_ERR(dchild)) 1269 goto out_nfserr; 1270 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1271 if (err) 1272 goto out; 1273 } else { 1274 /* called from nfsd_proc_create */ 1275 dchild = dget(resfhp->fh_dentry); 1276 if (!fhp->fh_locked) { 1277 /* not actually possible */ 1278 printk(KERN_ERR 1279 "nfsd_create: parent %s/%s not locked!\n", 1280 dentry->d_parent->d_name.name, 1281 dentry->d_name.name); 1282 err = nfserr_io; 1283 goto out; 1284 } 1285 } 1286 /* 1287 * Make sure the child dentry is still negative ... 1288 */ 1289 err = nfserr_exist; 1290 if (dchild->d_inode) { 1291 dprintk("nfsd_create: dentry %s/%s not negative!\n", 1292 dentry->d_name.name, dchild->d_name.name); 1293 goto out; 1294 } 1295 1296 if (!(iap->ia_valid & ATTR_MODE)) 1297 iap->ia_mode = 0; 1298 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1299 1300 err = nfserr_inval; 1301 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1302 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1303 type); 1304 goto out; 1305 } 1306 1307 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1308 if (host_err) 1309 goto out_nfserr; 1310 1311 /* 1312 * Get the dir op function pointer. 1313 */ 1314 err = 0; 1315 switch (type) { 1316 case S_IFREG: 1317 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1318 if (!host_err) 1319 nfsd_check_ignore_resizing(iap); 1320 break; 1321 case S_IFDIR: 1322 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1323 break; 1324 case S_IFCHR: 1325 case S_IFBLK: 1326 case S_IFIFO: 1327 case S_IFSOCK: 1328 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1329 break; 1330 } 1331 if (host_err < 0) { 1332 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1333 goto out_nfserr; 1334 } 1335 1336 if (EX_ISSYNC(fhp->fh_export)) { 1337 err = nfserrno(nfsd_sync_dir(dentry)); 1338 write_inode_now(dchild->d_inode, 1); 1339 } 1340 1341 err2 = nfsd_create_setattr(rqstp, resfhp, iap); 1342 if (err2) 1343 err = err2; 1344 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1345 /* 1346 * Update the file handle to get the new inode info. 1347 */ 1348 if (!err) 1349 err = fh_update(resfhp); 1350 out: 1351 if (dchild && !IS_ERR(dchild)) 1352 dput(dchild); 1353 return err; 1354 1355 out_nfserr: 1356 err = nfserrno(host_err); 1357 goto out; 1358 } 1359 1360 #ifdef CONFIG_NFSD_V3 1361 /* 1362 * NFSv3 version of nfsd_create 1363 */ 1364 __be32 1365 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp, 1366 char *fname, int flen, struct iattr *iap, 1367 struct svc_fh *resfhp, int createmode, u32 *verifier, 1368 int *truncp, int *created) 1369 { 1370 struct dentry *dentry, *dchild = NULL; 1371 struct inode *dirp; 1372 __be32 err; 1373 __be32 err2; 1374 int host_err; 1375 __u32 v_mtime=0, v_atime=0; 1376 1377 err = nfserr_perm; 1378 if (!flen) 1379 goto out; 1380 err = nfserr_exist; 1381 if (isdotent(fname, flen)) 1382 goto out; 1383 if (!(iap->ia_valid & ATTR_MODE)) 1384 iap->ia_mode = 0; 1385 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1386 if (err) 1387 goto out; 1388 1389 dentry = fhp->fh_dentry; 1390 dirp = dentry->d_inode; 1391 1392 /* Get all the sanity checks out of the way before 1393 * we lock the parent. */ 1394 err = nfserr_notdir; 1395 if (!dirp->i_op->lookup) 1396 goto out; 1397 fh_lock_nested(fhp, I_MUTEX_PARENT); 1398 1399 /* 1400 * Compose the response file handle. 1401 */ 1402 dchild = lookup_one_len(fname, dentry, flen); 1403 host_err = PTR_ERR(dchild); 1404 if (IS_ERR(dchild)) 1405 goto out_nfserr; 1406 1407 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1408 if (err) 1409 goto out; 1410 1411 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1412 /* solaris7 gets confused (bugid 4218508) if these have 1413 * the high bit set, so just clear the high bits. If this is 1414 * ever changed to use different attrs for storing the 1415 * verifier, then do_open_lookup() will also need to be fixed 1416 * accordingly. 1417 */ 1418 v_mtime = verifier[0]&0x7fffffff; 1419 v_atime = verifier[1]&0x7fffffff; 1420 } 1421 1422 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1423 if (host_err) 1424 goto out_nfserr; 1425 if (dchild->d_inode) { 1426 err = 0; 1427 1428 switch (createmode) { 1429 case NFS3_CREATE_UNCHECKED: 1430 if (! S_ISREG(dchild->d_inode->i_mode)) 1431 err = nfserr_exist; 1432 else if (truncp) { 1433 /* in nfsv4, we need to treat this case a little 1434 * differently. we don't want to truncate the 1435 * file now; this would be wrong if the OPEN 1436 * fails for some other reason. furthermore, 1437 * if the size is nonzero, we should ignore it 1438 * according to spec! 1439 */ 1440 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1441 } 1442 else { 1443 iap->ia_valid &= ATTR_SIZE; 1444 goto set_attr; 1445 } 1446 break; 1447 case NFS3_CREATE_EXCLUSIVE: 1448 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1449 && dchild->d_inode->i_atime.tv_sec == v_atime 1450 && dchild->d_inode->i_size == 0 ) 1451 break; 1452 /* fallthru */ 1453 case NFS3_CREATE_GUARDED: 1454 err = nfserr_exist; 1455 } 1456 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1457 goto out; 1458 } 1459 1460 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1461 if (host_err < 0) { 1462 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1463 goto out_nfserr; 1464 } 1465 if (created) 1466 *created = 1; 1467 1468 if (EX_ISSYNC(fhp->fh_export)) { 1469 err = nfserrno(nfsd_sync_dir(dentry)); 1470 /* setattr will sync the child (or not) */ 1471 } 1472 1473 nfsd_check_ignore_resizing(iap); 1474 1475 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1476 /* Cram the verifier into atime/mtime */ 1477 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1478 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1479 /* XXX someone who knows this better please fix it for nsec */ 1480 iap->ia_mtime.tv_sec = v_mtime; 1481 iap->ia_atime.tv_sec = v_atime; 1482 iap->ia_mtime.tv_nsec = 0; 1483 iap->ia_atime.tv_nsec = 0; 1484 } 1485 1486 set_attr: 1487 err2 = nfsd_create_setattr(rqstp, resfhp, iap); 1488 if (err2) 1489 err = err2; 1490 1491 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1492 /* 1493 * Update the filehandle to get the new inode info. 1494 */ 1495 if (!err) 1496 err = fh_update(resfhp); 1497 1498 out: 1499 fh_unlock(fhp); 1500 if (dchild && !IS_ERR(dchild)) 1501 dput(dchild); 1502 return err; 1503 1504 out_nfserr: 1505 err = nfserrno(host_err); 1506 goto out; 1507 } 1508 #endif /* CONFIG_NFSD_V3 */ 1509 1510 /* 1511 * Read a symlink. On entry, *lenp must contain the maximum path length that 1512 * fits into the buffer. On return, it contains the true length. 1513 * N.B. After this call fhp needs an fh_put 1514 */ 1515 __be32 1516 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1517 { 1518 struct dentry *dentry; 1519 struct inode *inode; 1520 mm_segment_t oldfs; 1521 __be32 err; 1522 int host_err; 1523 1524 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1525 if (err) 1526 goto out; 1527 1528 dentry = fhp->fh_dentry; 1529 inode = dentry->d_inode; 1530 1531 err = nfserr_inval; 1532 if (!inode->i_op->readlink) 1533 goto out; 1534 1535 touch_atime(fhp->fh_export->ex_path.mnt, dentry); 1536 /* N.B. Why does this call need a get_fs()?? 1537 * Remove the set_fs and watch the fireworks:-) --okir 1538 */ 1539 1540 oldfs = get_fs(); set_fs(KERNEL_DS); 1541 host_err = inode->i_op->readlink(dentry, buf, *lenp); 1542 set_fs(oldfs); 1543 1544 if (host_err < 0) 1545 goto out_nfserr; 1546 *lenp = host_err; 1547 err = 0; 1548 out: 1549 return err; 1550 1551 out_nfserr: 1552 err = nfserrno(host_err); 1553 goto out; 1554 } 1555 1556 /* 1557 * Create a symlink and look up its inode 1558 * N.B. After this call _both_ fhp and resfhp need an fh_put 1559 */ 1560 __be32 1561 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1562 char *fname, int flen, 1563 char *path, int plen, 1564 struct svc_fh *resfhp, 1565 struct iattr *iap) 1566 { 1567 struct dentry *dentry, *dnew; 1568 __be32 err, cerr; 1569 int host_err; 1570 1571 err = nfserr_noent; 1572 if (!flen || !plen) 1573 goto out; 1574 err = nfserr_exist; 1575 if (isdotent(fname, flen)) 1576 goto out; 1577 1578 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1579 if (err) 1580 goto out; 1581 fh_lock(fhp); 1582 dentry = fhp->fh_dentry; 1583 dnew = lookup_one_len(fname, dentry, flen); 1584 host_err = PTR_ERR(dnew); 1585 if (IS_ERR(dnew)) 1586 goto out_nfserr; 1587 1588 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1589 if (host_err) 1590 goto out_nfserr; 1591 1592 if (unlikely(path[plen] != 0)) { 1593 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1594 if (path_alloced == NULL) 1595 host_err = -ENOMEM; 1596 else { 1597 strncpy(path_alloced, path, plen); 1598 path_alloced[plen] = 0; 1599 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); 1600 kfree(path_alloced); 1601 } 1602 } else 1603 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1604 1605 if (!host_err) { 1606 if (EX_ISSYNC(fhp->fh_export)) 1607 host_err = nfsd_sync_dir(dentry); 1608 } 1609 err = nfserrno(host_err); 1610 fh_unlock(fhp); 1611 1612 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1613 1614 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1615 dput(dnew); 1616 if (err==0) err = cerr; 1617 out: 1618 return err; 1619 1620 out_nfserr: 1621 err = nfserrno(host_err); 1622 goto out; 1623 } 1624 1625 /* 1626 * Create a hardlink 1627 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1628 */ 1629 __be32 1630 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1631 char *name, int len, struct svc_fh *tfhp) 1632 { 1633 struct dentry *ddir, *dnew, *dold; 1634 struct inode *dirp, *dest; 1635 __be32 err; 1636 int host_err; 1637 1638 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1639 if (err) 1640 goto out; 1641 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP); 1642 if (err) 1643 goto out; 1644 1645 err = nfserr_perm; 1646 if (!len) 1647 goto out; 1648 err = nfserr_exist; 1649 if (isdotent(name, len)) 1650 goto out; 1651 1652 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1653 ddir = ffhp->fh_dentry; 1654 dirp = ddir->d_inode; 1655 1656 dnew = lookup_one_len(name, ddir, len); 1657 host_err = PTR_ERR(dnew); 1658 if (IS_ERR(dnew)) 1659 goto out_nfserr; 1660 1661 dold = tfhp->fh_dentry; 1662 dest = dold->d_inode; 1663 1664 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt); 1665 if (host_err) { 1666 err = nfserrno(host_err); 1667 goto out_dput; 1668 } 1669 host_err = vfs_link(dold, dirp, dnew); 1670 if (!host_err) { 1671 if (EX_ISSYNC(ffhp->fh_export)) { 1672 err = nfserrno(nfsd_sync_dir(ddir)); 1673 write_inode_now(dest, 1); 1674 } 1675 err = 0; 1676 } else { 1677 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1678 err = nfserr_acces; 1679 else 1680 err = nfserrno(host_err); 1681 } 1682 mnt_drop_write(tfhp->fh_export->ex_path.mnt); 1683 out_dput: 1684 dput(dnew); 1685 out_unlock: 1686 fh_unlock(ffhp); 1687 out: 1688 return err; 1689 1690 out_nfserr: 1691 err = nfserrno(host_err); 1692 goto out_unlock; 1693 } 1694 1695 /* 1696 * Rename a file 1697 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1698 */ 1699 __be32 1700 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1701 struct svc_fh *tfhp, char *tname, int tlen) 1702 { 1703 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1704 struct inode *fdir, *tdir; 1705 __be32 err; 1706 int host_err; 1707 1708 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1709 if (err) 1710 goto out; 1711 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1712 if (err) 1713 goto out; 1714 1715 fdentry = ffhp->fh_dentry; 1716 fdir = fdentry->d_inode; 1717 1718 tdentry = tfhp->fh_dentry; 1719 tdir = tdentry->d_inode; 1720 1721 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; 1722 if (ffhp->fh_export != tfhp->fh_export) 1723 goto out; 1724 1725 err = nfserr_perm; 1726 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1727 goto out; 1728 1729 /* cannot use fh_lock as we need deadlock protective ordering 1730 * so do it by hand */ 1731 trap = lock_rename(tdentry, fdentry); 1732 ffhp->fh_locked = tfhp->fh_locked = 1; 1733 fill_pre_wcc(ffhp); 1734 fill_pre_wcc(tfhp); 1735 1736 odentry = lookup_one_len(fname, fdentry, flen); 1737 host_err = PTR_ERR(odentry); 1738 if (IS_ERR(odentry)) 1739 goto out_nfserr; 1740 1741 host_err = -ENOENT; 1742 if (!odentry->d_inode) 1743 goto out_dput_old; 1744 host_err = -EINVAL; 1745 if (odentry == trap) 1746 goto out_dput_old; 1747 1748 ndentry = lookup_one_len(tname, tdentry, tlen); 1749 host_err = PTR_ERR(ndentry); 1750 if (IS_ERR(ndentry)) 1751 goto out_dput_old; 1752 host_err = -ENOTEMPTY; 1753 if (ndentry == trap) 1754 goto out_dput_new; 1755 1756 if (svc_msnfs(ffhp) && 1757 ((atomic_read(&odentry->d_count) > 1) 1758 || (atomic_read(&ndentry->d_count) > 1))) { 1759 host_err = -EPERM; 1760 goto out_dput_new; 1761 } 1762 1763 host_err = -EXDEV; 1764 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1765 goto out_dput_new; 1766 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt); 1767 if (host_err) 1768 goto out_dput_new; 1769 1770 host_err = vfs_rename(fdir, odentry, tdir, ndentry); 1771 if (!host_err && EX_ISSYNC(tfhp->fh_export)) { 1772 host_err = nfsd_sync_dir(tdentry); 1773 if (!host_err) 1774 host_err = nfsd_sync_dir(fdentry); 1775 } 1776 1777 mnt_drop_write(ffhp->fh_export->ex_path.mnt); 1778 1779 out_dput_new: 1780 dput(ndentry); 1781 out_dput_old: 1782 dput(odentry); 1783 out_nfserr: 1784 err = nfserrno(host_err); 1785 1786 /* we cannot reply on fh_unlock on the two filehandles, 1787 * as that would do the wrong thing if the two directories 1788 * were the same, so again we do it by hand 1789 */ 1790 fill_post_wcc(ffhp); 1791 fill_post_wcc(tfhp); 1792 unlock_rename(tdentry, fdentry); 1793 ffhp->fh_locked = tfhp->fh_locked = 0; 1794 1795 out: 1796 return err; 1797 } 1798 1799 /* 1800 * Unlink a file or directory 1801 * N.B. After this call fhp needs an fh_put 1802 */ 1803 __be32 1804 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1805 char *fname, int flen) 1806 { 1807 struct dentry *dentry, *rdentry; 1808 struct inode *dirp; 1809 __be32 err; 1810 int host_err; 1811 1812 err = nfserr_acces; 1813 if (!flen || isdotent(fname, flen)) 1814 goto out; 1815 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1816 if (err) 1817 goto out; 1818 1819 fh_lock_nested(fhp, I_MUTEX_PARENT); 1820 dentry = fhp->fh_dentry; 1821 dirp = dentry->d_inode; 1822 1823 rdentry = lookup_one_len(fname, dentry, flen); 1824 host_err = PTR_ERR(rdentry); 1825 if (IS_ERR(rdentry)) 1826 goto out_nfserr; 1827 1828 if (!rdentry->d_inode) { 1829 dput(rdentry); 1830 err = nfserr_noent; 1831 goto out; 1832 } 1833 1834 if (!type) 1835 type = rdentry->d_inode->i_mode & S_IFMT; 1836 1837 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1838 if (host_err) 1839 goto out_nfserr; 1840 1841 if (type != S_IFDIR) { /* It's UNLINK */ 1842 #ifdef MSNFS 1843 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1844 (atomic_read(&rdentry->d_count) > 1)) { 1845 host_err = -EPERM; 1846 } else 1847 #endif 1848 host_err = vfs_unlink(dirp, rdentry); 1849 } else { /* It's RMDIR */ 1850 host_err = vfs_rmdir(dirp, rdentry); 1851 } 1852 1853 dput(rdentry); 1854 1855 if (host_err) 1856 goto out_drop; 1857 if (EX_ISSYNC(fhp->fh_export)) 1858 host_err = nfsd_sync_dir(dentry); 1859 1860 out_drop: 1861 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1862 out_nfserr: 1863 err = nfserrno(host_err); 1864 out: 1865 return err; 1866 } 1867 1868 /* 1869 * We do this buffering because we must not call back into the file 1870 * system's ->lookup() method from the filldir callback. That may well 1871 * deadlock a number of file systems. 1872 * 1873 * This is based heavily on the implementation of same in XFS. 1874 */ 1875 struct buffered_dirent { 1876 u64 ino; 1877 loff_t offset; 1878 int namlen; 1879 unsigned int d_type; 1880 char name[]; 1881 }; 1882 1883 struct readdir_data { 1884 char *dirent; 1885 size_t used; 1886 int full; 1887 }; 1888 1889 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1890 loff_t offset, u64 ino, unsigned int d_type) 1891 { 1892 struct readdir_data *buf = __buf; 1893 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1894 unsigned int reclen; 1895 1896 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1897 if (buf->used + reclen > PAGE_SIZE) { 1898 buf->full = 1; 1899 return -EINVAL; 1900 } 1901 1902 de->namlen = namlen; 1903 de->offset = offset; 1904 de->ino = ino; 1905 de->d_type = d_type; 1906 memcpy(de->name, name, namlen); 1907 buf->used += reclen; 1908 1909 return 0; 1910 } 1911 1912 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1913 struct readdir_cd *cdp, loff_t *offsetp) 1914 { 1915 struct readdir_data buf; 1916 struct buffered_dirent *de; 1917 int host_err; 1918 int size; 1919 loff_t offset; 1920 1921 buf.dirent = (void *)__get_free_page(GFP_KERNEL); 1922 if (!buf.dirent) 1923 return nfserrno(-ENOMEM); 1924 1925 offset = *offsetp; 1926 1927 while (1) { 1928 struct inode *dir_inode = file->f_path.dentry->d_inode; 1929 unsigned int reclen; 1930 1931 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1932 buf.used = 0; 1933 buf.full = 0; 1934 1935 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf); 1936 if (buf.full) 1937 host_err = 0; 1938 1939 if (host_err < 0) 1940 break; 1941 1942 size = buf.used; 1943 1944 if (!size) 1945 break; 1946 1947 /* 1948 * Various filldir functions may end up calling back into 1949 * lookup_one_len() and the file system's ->lookup() method. 1950 * These expect i_mutex to be held, as it would within readdir. 1951 */ 1952 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1953 if (host_err) 1954 break; 1955 1956 de = (struct buffered_dirent *)buf.dirent; 1957 while (size > 0) { 1958 offset = de->offset; 1959 1960 if (func(cdp, de->name, de->namlen, de->offset, 1961 de->ino, de->d_type)) 1962 break; 1963 1964 if (cdp->err != nfs_ok) 1965 break; 1966 1967 reclen = ALIGN(sizeof(*de) + de->namlen, 1968 sizeof(u64)); 1969 size -= reclen; 1970 de = (struct buffered_dirent *)((char *)de + reclen); 1971 } 1972 mutex_unlock(&dir_inode->i_mutex); 1973 if (size > 0) /* We bailed out early */ 1974 break; 1975 1976 offset = vfs_llseek(file, 0, SEEK_CUR); 1977 } 1978 1979 free_page((unsigned long)(buf.dirent)); 1980 1981 if (host_err) 1982 return nfserrno(host_err); 1983 1984 *offsetp = offset; 1985 return cdp->err; 1986 } 1987 1988 /* 1989 * Read entries from a directory. 1990 * The NFSv3/4 verifier we ignore for now. 1991 */ 1992 __be32 1993 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 1994 struct readdir_cd *cdp, filldir_t func) 1995 { 1996 __be32 err; 1997 struct file *file; 1998 loff_t offset = *offsetp; 1999 2000 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file); 2001 if (err) 2002 goto out; 2003 2004 offset = vfs_llseek(file, offset, 0); 2005 if (offset < 0) { 2006 err = nfserrno((int)offset); 2007 goto out_close; 2008 } 2009 2010 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 2011 2012 if (err == nfserr_eof || err == nfserr_toosmall) 2013 err = nfs_ok; /* can still be found in ->err */ 2014 out_close: 2015 nfsd_close(file); 2016 out: 2017 return err; 2018 } 2019 2020 /* 2021 * Get file system stats 2022 * N.B. After this call fhp needs an fh_put 2023 */ 2024 __be32 2025 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 2026 { 2027 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 2028 if (!err && vfs_statfs(fhp->fh_dentry,stat)) 2029 err = nfserr_io; 2030 return err; 2031 } 2032 2033 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 2034 { 2035 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 2036 } 2037 2038 /* 2039 * Check for a user's access permissions to this inode. 2040 */ 2041 __be32 2042 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 2043 struct dentry *dentry, int acc) 2044 { 2045 struct inode *inode = dentry->d_inode; 2046 struct path path; 2047 int err; 2048 2049 if (acc == NFSD_MAY_NOP) 2050 return 0; 2051 #if 0 2052 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 2053 acc, 2054 (acc & NFSD_MAY_READ)? " read" : "", 2055 (acc & NFSD_MAY_WRITE)? " write" : "", 2056 (acc & NFSD_MAY_EXEC)? " exec" : "", 2057 (acc & NFSD_MAY_SATTR)? " sattr" : "", 2058 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 2059 (acc & NFSD_MAY_LOCK)? " lock" : "", 2060 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 2061 inode->i_mode, 2062 IS_IMMUTABLE(inode)? " immut" : "", 2063 IS_APPEND(inode)? " append" : "", 2064 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2065 dprintk(" owner %d/%d user %d/%d\n", 2066 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2067 #endif 2068 2069 /* Normally we reject any write/sattr etc access on a read-only file 2070 * system. But if it is IRIX doing check on write-access for a 2071 * device special file, we ignore rofs. 2072 */ 2073 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2074 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2075 if (exp_rdonly(rqstp, exp) || 2076 __mnt_is_readonly(exp->ex_path.mnt)) 2077 return nfserr_rofs; 2078 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2079 return nfserr_perm; 2080 } 2081 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2082 return nfserr_perm; 2083 2084 if (acc & NFSD_MAY_LOCK) { 2085 /* If we cannot rely on authentication in NLM requests, 2086 * just allow locks, otherwise require read permission, or 2087 * ownership 2088 */ 2089 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2090 return 0; 2091 else 2092 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2093 } 2094 /* 2095 * The file owner always gets access permission for accesses that 2096 * would normally be checked at open time. This is to make 2097 * file access work even when the client has done a fchmod(fd, 0). 2098 * 2099 * However, `cp foo bar' should fail nevertheless when bar is 2100 * readonly. A sensible way to do this might be to reject all 2101 * attempts to truncate a read-only file, because a creat() call 2102 * always implies file truncation. 2103 * ... but this isn't really fair. A process may reasonably call 2104 * ftruncate on an open file descriptor on a file with perm 000. 2105 * We must trust the client to do permission checking - using "ACCESS" 2106 * with NFSv3. 2107 */ 2108 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2109 inode->i_uid == current_fsuid()) 2110 return 0; 2111 2112 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2113 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2114 2115 /* Allow read access to binaries even when mode 111 */ 2116 if (err == -EACCES && S_ISREG(inode->i_mode) && 2117 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE)) 2118 err = inode_permission(inode, MAY_EXEC); 2119 if (err) 2120 goto nfsd_out; 2121 2122 /* Do integrity (permission) checking now, but defer incrementing 2123 * IMA counts to the actual file open. 2124 */ 2125 path.mnt = exp->ex_path.mnt; 2126 path.dentry = dentry; 2127 err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC), 2128 IMA_COUNT_LEAVE); 2129 nfsd_out: 2130 return err? nfserrno(err) : 0; 2131 } 2132 2133 void 2134 nfsd_racache_shutdown(void) 2135 { 2136 struct raparms *raparm, *last_raparm; 2137 unsigned int i; 2138 2139 dprintk("nfsd: freeing readahead buffers.\n"); 2140 2141 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2142 raparm = raparm_hash[i].pb_head; 2143 while(raparm) { 2144 last_raparm = raparm; 2145 raparm = raparm->p_next; 2146 kfree(last_raparm); 2147 } 2148 raparm_hash[i].pb_head = NULL; 2149 } 2150 } 2151 /* 2152 * Initialize readahead param cache 2153 */ 2154 int 2155 nfsd_racache_init(int cache_size) 2156 { 2157 int i; 2158 int j = 0; 2159 int nperbucket; 2160 struct raparms **raparm = NULL; 2161 2162 2163 if (raparm_hash[0].pb_head) 2164 return 0; 2165 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2166 if (nperbucket < 2) 2167 nperbucket = 2; 2168 cache_size = nperbucket * RAPARM_HASH_SIZE; 2169 2170 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2171 2172 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2173 spin_lock_init(&raparm_hash[i].pb_lock); 2174 2175 raparm = &raparm_hash[i].pb_head; 2176 for (j = 0; j < nperbucket; j++) { 2177 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2178 if (!*raparm) 2179 goto out_nomem; 2180 raparm = &(*raparm)->p_next; 2181 } 2182 *raparm = NULL; 2183 } 2184 2185 nfsdstats.ra_size = cache_size; 2186 return 0; 2187 2188 out_nomem: 2189 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2190 nfsd_racache_shutdown(); 2191 return -ENOMEM; 2192 } 2193 2194 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 2195 struct posix_acl * 2196 nfsd_get_posix_acl(struct svc_fh *fhp, int type) 2197 { 2198 struct inode *inode = fhp->fh_dentry->d_inode; 2199 char *name; 2200 void *value = NULL; 2201 ssize_t size; 2202 struct posix_acl *acl; 2203 2204 if (!IS_POSIXACL(inode)) 2205 return ERR_PTR(-EOPNOTSUPP); 2206 2207 switch (type) { 2208 case ACL_TYPE_ACCESS: 2209 name = POSIX_ACL_XATTR_ACCESS; 2210 break; 2211 case ACL_TYPE_DEFAULT: 2212 name = POSIX_ACL_XATTR_DEFAULT; 2213 break; 2214 default: 2215 return ERR_PTR(-EOPNOTSUPP); 2216 } 2217 2218 size = nfsd_getxattr(fhp->fh_dentry, name, &value); 2219 if (size < 0) 2220 return ERR_PTR(size); 2221 2222 acl = posix_acl_from_xattr(value, size); 2223 kfree(value); 2224 return acl; 2225 } 2226 2227 int 2228 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) 2229 { 2230 struct inode *inode = fhp->fh_dentry->d_inode; 2231 char *name; 2232 void *value = NULL; 2233 size_t size; 2234 int error; 2235 2236 if (!IS_POSIXACL(inode) || 2237 !inode->i_op->setxattr || !inode->i_op->removexattr) 2238 return -EOPNOTSUPP; 2239 switch(type) { 2240 case ACL_TYPE_ACCESS: 2241 name = POSIX_ACL_XATTR_ACCESS; 2242 break; 2243 case ACL_TYPE_DEFAULT: 2244 name = POSIX_ACL_XATTR_DEFAULT; 2245 break; 2246 default: 2247 return -EOPNOTSUPP; 2248 } 2249 2250 if (acl && acl->a_count) { 2251 size = posix_acl_xattr_size(acl->a_count); 2252 value = kmalloc(size, GFP_KERNEL); 2253 if (!value) 2254 return -ENOMEM; 2255 error = posix_acl_to_xattr(acl, value, size); 2256 if (error < 0) 2257 goto getout; 2258 size = error; 2259 } else 2260 size = 0; 2261 2262 error = mnt_want_write(fhp->fh_export->ex_path.mnt); 2263 if (error) 2264 goto getout; 2265 if (size) 2266 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); 2267 else { 2268 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) 2269 error = 0; 2270 else { 2271 error = vfs_removexattr(fhp->fh_dentry, name); 2272 if (error == -ENODATA) 2273 error = 0; 2274 } 2275 } 2276 mnt_drop_write(fhp->fh_export->ex_path.mnt); 2277 2278 getout: 2279 kfree(value); 2280 return error; 2281 } 2282 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 2283