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