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/ext2_fs.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 /* We must ignore files (but only files) which might have mandatory 65 * locks on them because there is no way to know if the accesser has 66 * the lock. 67 */ 68 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i)) 69 70 /* 71 * This is a cache of readahead params that help us choose the proper 72 * readahead strategy. Initially, we set all readahead parameters to 0 73 * and let the VFS handle things. 74 * If you increase the number of cached files very much, you'll need to 75 * add a hash table here. 76 */ 77 struct raparms { 78 struct raparms *p_next; 79 unsigned int p_count; 80 ino_t p_ino; 81 dev_t p_dev; 82 int p_set; 83 struct file_ra_state p_ra; 84 unsigned int p_hindex; 85 }; 86 87 struct raparm_hbucket { 88 struct raparms *pb_head; 89 spinlock_t pb_lock; 90 } ____cacheline_aligned_in_smp; 91 92 static struct raparms * raparml; 93 #define RAPARM_HASH_BITS 4 94 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) 95 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) 96 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; 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 vfsmount *mnt = mntget(exp->ex_mnt); 111 struct dentry *mounts = dget(dentry); 112 int err = 0; 113 114 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts)); 115 116 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle); 117 if (IS_ERR(exp2)) { 118 err = PTR_ERR(exp2); 119 dput(mounts); 120 mntput(mnt); 121 goto out; 122 } 123 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) { 124 /* successfully crossed mount point */ 125 exp_put(exp); 126 *expp = exp2; 127 dput(dentry); 128 *dpp = mounts; 129 } else { 130 if (exp2) exp_put(exp2); 131 dput(mounts); 132 } 133 mntput(mnt); 134 out: 135 return err; 136 } 137 138 /* 139 * Look up one component of a pathname. 140 * N.B. After this call _both_ fhp and resfh need an fh_put 141 * 142 * If the lookup would cross a mountpoint, and the mounted filesystem 143 * is exported to the client with NFSEXP_NOHIDE, then the lookup is 144 * accepted as it stands and the mounted directory is 145 * returned. Otherwise the covered directory is returned. 146 * NOTE: this mountpoint crossing is not supported properly by all 147 * clients and is explicitly disallowed for NFSv3 148 * NeilBrown <neilb@cse.unsw.edu.au> 149 */ 150 __be32 151 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, 152 int len, struct svc_fh *resfh) 153 { 154 struct svc_export *exp; 155 struct dentry *dparent; 156 struct dentry *dentry; 157 __be32 err; 158 int host_err; 159 160 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); 161 162 /* Obtain dentry and export. */ 163 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC); 164 if (err) 165 return err; 166 167 dparent = fhp->fh_dentry; 168 exp = fhp->fh_export; 169 exp_get(exp); 170 171 err = nfserr_acces; 172 173 /* Lookup the name, but don't follow links */ 174 if (isdotent(name, len)) { 175 if (len==1) 176 dentry = dget(dparent); 177 else if (dparent != exp->ex_dentry) { 178 dentry = dget_parent(dparent); 179 } else if (!EX_NOHIDE(exp)) 180 dentry = dget(dparent); /* .. == . just like at / */ 181 else { 182 /* checking mountpoint crossing is very different when stepping up */ 183 struct svc_export *exp2 = NULL; 184 struct dentry *dp; 185 struct vfsmount *mnt = mntget(exp->ex_mnt); 186 dentry = dget(dparent); 187 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry)) 188 ; 189 dp = dget_parent(dentry); 190 dput(dentry); 191 dentry = dp; 192 193 exp2 = exp_parent(exp->ex_client, mnt, dentry, 194 &rqstp->rq_chandle); 195 if (IS_ERR(exp2)) { 196 host_err = PTR_ERR(exp2); 197 dput(dentry); 198 mntput(mnt); 199 goto out_nfserr; 200 } 201 if (!exp2) { 202 dput(dentry); 203 dentry = dget(dparent); 204 } else { 205 exp_put(exp); 206 exp = exp2; 207 } 208 mntput(mnt); 209 } 210 } else { 211 fh_lock(fhp); 212 dentry = lookup_one_len(name, dparent, len); 213 host_err = PTR_ERR(dentry); 214 if (IS_ERR(dentry)) 215 goto out_nfserr; 216 /* 217 * check if we have crossed a mount point ... 218 */ 219 if (d_mountpoint(dentry)) { 220 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { 221 dput(dentry); 222 goto out_nfserr; 223 } 224 } 225 } 226 /* 227 * Note: we compose the file handle now, but as the 228 * dentry may be negative, it may need to be updated. 229 */ 230 err = fh_compose(resfh, exp, dentry, fhp); 231 if (!err && !dentry->d_inode) 232 err = nfserr_noent; 233 dput(dentry); 234 out: 235 exp_put(exp); 236 return err; 237 238 out_nfserr: 239 err = nfserrno(host_err); 240 goto out; 241 } 242 243 /* 244 * Set various file attributes. 245 * N.B. After this call fhp needs an fh_put 246 */ 247 __be32 248 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, 249 int check_guard, time_t guardtime) 250 { 251 struct dentry *dentry; 252 struct inode *inode; 253 int accmode = MAY_SATTR; 254 int ftype = 0; 255 int imode; 256 __be32 err; 257 int host_err; 258 int size_change = 0; 259 260 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) 261 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE; 262 if (iap->ia_valid & ATTR_SIZE) 263 ftype = S_IFREG; 264 265 /* Get inode */ 266 err = fh_verify(rqstp, fhp, ftype, accmode); 267 if (err) 268 goto out; 269 270 dentry = fhp->fh_dentry; 271 inode = dentry->d_inode; 272 273 /* Ignore any mode updates on symlinks */ 274 if (S_ISLNK(inode->i_mode)) 275 iap->ia_valid &= ~ATTR_MODE; 276 277 if (!iap->ia_valid) 278 goto out; 279 280 /* NFSv2 does not differentiate between "set-[ac]time-to-now" 281 * which only requires access, and "set-[ac]time-to-X" which 282 * requires ownership. 283 * So if it looks like it might be "set both to the same time which 284 * is close to now", and if inode_change_ok fails, then we 285 * convert to "set to now" instead of "set to explicit time" 286 * 287 * We only call inode_change_ok as the last test as technically 288 * it is not an interface that we should be using. It is only 289 * valid if the filesystem does not define it's own i_op->setattr. 290 */ 291 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) 292 #define MAX_TOUCH_TIME_ERROR (30*60) 293 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET 294 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec 295 ) { 296 /* Looks probable. Now just make sure time is in the right ballpark. 297 * Solaris, at least, doesn't seem to care what the time request is. 298 * We require it be within 30 minutes of now. 299 */ 300 time_t delta = iap->ia_atime.tv_sec - get_seconds(); 301 if (delta<0) delta = -delta; 302 if (delta < MAX_TOUCH_TIME_ERROR && 303 inode_change_ok(inode, iap) != 0) { 304 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME 305 * this will cause notify_change to set these times to "now" 306 */ 307 iap->ia_valid &= ~BOTH_TIME_SET; 308 } 309 } 310 311 /* The size case is special. It changes the file as well as the attributes. */ 312 if (iap->ia_valid & ATTR_SIZE) { 313 if (iap->ia_size < inode->i_size) { 314 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE); 315 if (err) 316 goto out; 317 } 318 319 /* 320 * If we are changing the size of the file, then 321 * we need to break all leases. 322 */ 323 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK); 324 if (host_err == -EWOULDBLOCK) 325 host_err = -ETIMEDOUT; 326 if (host_err) /* ENOMEM or EWOULDBLOCK */ 327 goto out_nfserr; 328 329 host_err = get_write_access(inode); 330 if (host_err) 331 goto out_nfserr; 332 333 size_change = 1; 334 host_err = locks_verify_truncate(inode, NULL, iap->ia_size); 335 if (host_err) { 336 put_write_access(inode); 337 goto out_nfserr; 338 } 339 DQUOT_INIT(inode); 340 } 341 342 imode = inode->i_mode; 343 if (iap->ia_valid & ATTR_MODE) { 344 iap->ia_mode &= S_IALLUGO; 345 imode = iap->ia_mode |= (imode & ~S_IALLUGO); 346 } 347 348 /* Revoke setuid/setgid bit on chown/chgrp */ 349 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) 350 iap->ia_valid |= ATTR_KILL_SUID; 351 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid) 352 iap->ia_valid |= ATTR_KILL_SGID; 353 354 /* Change the attributes. */ 355 356 iap->ia_valid |= ATTR_CTIME; 357 358 err = nfserr_notsync; 359 if (!check_guard || guardtime == inode->i_ctime.tv_sec) { 360 fh_lock(fhp); 361 host_err = notify_change(dentry, iap); 362 err = nfserrno(host_err); 363 fh_unlock(fhp); 364 } 365 if (size_change) 366 put_write_access(inode); 367 if (!err) 368 if (EX_ISSYNC(fhp->fh_export)) 369 write_inode_now(inode, 1); 370 out: 371 return err; 372 373 out_nfserr: 374 err = nfserrno(host_err); 375 goto out; 376 } 377 378 #if defined(CONFIG_NFSD_V2_ACL) || \ 379 defined(CONFIG_NFSD_V3_ACL) || \ 380 defined(CONFIG_NFSD_V4) 381 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf) 382 { 383 ssize_t buflen; 384 385 buflen = vfs_getxattr(dentry, key, NULL, 0); 386 if (buflen <= 0) 387 return buflen; 388 389 *buf = kmalloc(buflen, GFP_KERNEL); 390 if (!*buf) 391 return -ENOMEM; 392 393 return vfs_getxattr(dentry, key, *buf, buflen); 394 } 395 #endif 396 397 #if defined(CONFIG_NFSD_V4) 398 static int 399 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) 400 { 401 int len; 402 size_t buflen; 403 char *buf = NULL; 404 int error = 0; 405 406 buflen = posix_acl_xattr_size(pacl->a_count); 407 buf = kmalloc(buflen, GFP_KERNEL); 408 error = -ENOMEM; 409 if (buf == NULL) 410 goto out; 411 412 len = posix_acl_to_xattr(pacl, buf, buflen); 413 if (len < 0) { 414 error = len; 415 goto out; 416 } 417 418 error = vfs_setxattr(dentry, key, buf, len, 0); 419 out: 420 kfree(buf); 421 return error; 422 } 423 424 __be32 425 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, 426 struct nfs4_acl *acl) 427 { 428 __be32 error; 429 int host_error; 430 struct dentry *dentry; 431 struct inode *inode; 432 struct posix_acl *pacl = NULL, *dpacl = NULL; 433 unsigned int flags = 0; 434 435 /* Get inode */ 436 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR); 437 if (error) 438 goto out; 439 440 dentry = fhp->fh_dentry; 441 inode = dentry->d_inode; 442 if (S_ISDIR(inode->i_mode)) 443 flags = NFS4_ACL_DIR; 444 445 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); 446 if (host_error == -EINVAL) { 447 error = nfserr_attrnotsupp; 448 goto out; 449 } else if (host_error < 0) 450 goto out_nfserr; 451 452 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); 453 if (host_error < 0) 454 goto out_nfserr; 455 456 if (S_ISDIR(inode->i_mode)) { 457 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); 458 if (host_error < 0) 459 goto out_nfserr; 460 } 461 462 error = nfs_ok; 463 464 out: 465 posix_acl_release(pacl); 466 posix_acl_release(dpacl); 467 return (error); 468 out_nfserr: 469 if (host_error == -EOPNOTSUPP) 470 error = nfserr_attrnotsupp; 471 else 472 error = nfserrno(host_error); 473 goto out; 474 } 475 476 static struct posix_acl * 477 _get_posix_acl(struct dentry *dentry, char *key) 478 { 479 void *buf = NULL; 480 struct posix_acl *pacl = NULL; 481 int buflen; 482 483 buflen = nfsd_getxattr(dentry, key, &buf); 484 if (!buflen) 485 buflen = -ENODATA; 486 if (buflen <= 0) 487 return ERR_PTR(buflen); 488 489 pacl = posix_acl_from_xattr(buf, buflen); 490 kfree(buf); 491 return pacl; 492 } 493 494 int 495 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) 496 { 497 struct inode *inode = dentry->d_inode; 498 int error = 0; 499 struct posix_acl *pacl = NULL, *dpacl = NULL; 500 unsigned int flags = 0; 501 502 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); 503 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) 504 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); 505 if (IS_ERR(pacl)) { 506 error = PTR_ERR(pacl); 507 pacl = NULL; 508 goto out; 509 } 510 511 if (S_ISDIR(inode->i_mode)) { 512 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); 513 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) 514 dpacl = NULL; 515 else if (IS_ERR(dpacl)) { 516 error = PTR_ERR(dpacl); 517 dpacl = NULL; 518 goto out; 519 } 520 flags = NFS4_ACL_DIR; 521 } 522 523 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); 524 if (IS_ERR(*acl)) { 525 error = PTR_ERR(*acl); 526 *acl = NULL; 527 } 528 out: 529 posix_acl_release(pacl); 530 posix_acl_release(dpacl); 531 return error; 532 } 533 534 #endif /* defined(CONFIG_NFS_V4) */ 535 536 #ifdef CONFIG_NFSD_V3 537 /* 538 * Check server access rights to a file system object 539 */ 540 struct accessmap { 541 u32 access; 542 int how; 543 }; 544 static struct accessmap nfs3_regaccess[] = { 545 { NFS3_ACCESS_READ, MAY_READ }, 546 { NFS3_ACCESS_EXECUTE, MAY_EXEC }, 547 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC }, 548 { NFS3_ACCESS_EXTEND, MAY_WRITE }, 549 550 { 0, 0 } 551 }; 552 553 static struct accessmap nfs3_diraccess[] = { 554 { NFS3_ACCESS_READ, MAY_READ }, 555 { NFS3_ACCESS_LOOKUP, MAY_EXEC }, 556 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC }, 557 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE }, 558 { NFS3_ACCESS_DELETE, MAY_REMOVE }, 559 560 { 0, 0 } 561 }; 562 563 static struct accessmap nfs3_anyaccess[] = { 564 /* Some clients - Solaris 2.6 at least, make an access call 565 * to the server to check for access for things like /dev/null 566 * (which really, the server doesn't care about). So 567 * We provide simple access checking for them, looking 568 * mainly at mode bits, and we make sure to ignore read-only 569 * filesystem checks 570 */ 571 { NFS3_ACCESS_READ, MAY_READ }, 572 { NFS3_ACCESS_EXECUTE, MAY_EXEC }, 573 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS }, 574 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS }, 575 576 { 0, 0 } 577 }; 578 579 __be32 580 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) 581 { 582 struct accessmap *map; 583 struct svc_export *export; 584 struct dentry *dentry; 585 u32 query, result = 0, sresult = 0; 586 __be32 error; 587 588 error = fh_verify(rqstp, fhp, 0, MAY_NOP); 589 if (error) 590 goto out; 591 592 export = fhp->fh_export; 593 dentry = fhp->fh_dentry; 594 595 if (S_ISREG(dentry->d_inode->i_mode)) 596 map = nfs3_regaccess; 597 else if (S_ISDIR(dentry->d_inode->i_mode)) 598 map = nfs3_diraccess; 599 else 600 map = nfs3_anyaccess; 601 602 603 query = *access; 604 for (; map->access; map++) { 605 if (map->access & query) { 606 __be32 err2; 607 608 sresult |= map->access; 609 610 err2 = nfsd_permission(export, dentry, map->how); 611 switch (err2) { 612 case nfs_ok: 613 result |= map->access; 614 break; 615 616 /* the following error codes just mean the access was not allowed, 617 * rather than an error occurred */ 618 case nfserr_rofs: 619 case nfserr_acces: 620 case nfserr_perm: 621 /* simply don't "or" in the access bit. */ 622 break; 623 default: 624 error = err2; 625 goto out; 626 } 627 } 628 } 629 *access = result; 630 if (supported) 631 *supported = sresult; 632 633 out: 634 return error; 635 } 636 #endif /* CONFIG_NFSD_V3 */ 637 638 639 640 /* 641 * Open an existing file or directory. 642 * The access argument indicates the type of open (read/write/lock) 643 * N.B. After this call fhp needs an fh_put 644 */ 645 __be32 646 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 647 int access, struct file **filp) 648 { 649 struct dentry *dentry; 650 struct inode *inode; 651 int flags = O_RDONLY|O_LARGEFILE; 652 __be32 err; 653 int host_err; 654 655 /* 656 * If we get here, then the client has already done an "open", 657 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 658 * in case a chmod has now revoked permission. 659 */ 660 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE); 661 if (err) 662 goto out; 663 664 dentry = fhp->fh_dentry; 665 inode = dentry->d_inode; 666 667 /* Disallow write access to files with the append-only bit set 668 * or any access when mandatory locking enabled 669 */ 670 err = nfserr_perm; 671 if (IS_APPEND(inode) && (access & MAY_WRITE)) 672 goto out; 673 if (IS_ISMNDLK(inode)) 674 goto out; 675 676 if (!inode->i_fop) 677 goto out; 678 679 /* 680 * Check to see if there are any leases on this file. 681 * This may block while leases are broken. 682 */ 683 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0)); 684 if (host_err == -EWOULDBLOCK) 685 host_err = -ETIMEDOUT; 686 if (host_err) /* NOMEM or WOULDBLOCK */ 687 goto out_nfserr; 688 689 if (access & MAY_WRITE) { 690 if (access & MAY_READ) 691 flags = O_RDWR|O_LARGEFILE; 692 else 693 flags = O_WRONLY|O_LARGEFILE; 694 695 DQUOT_INIT(inode); 696 } 697 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags); 698 if (IS_ERR(*filp)) 699 host_err = PTR_ERR(*filp); 700 out_nfserr: 701 err = nfserrno(host_err); 702 out: 703 return err; 704 } 705 706 /* 707 * Close a file. 708 */ 709 void 710 nfsd_close(struct file *filp) 711 { 712 fput(filp); 713 } 714 715 /* 716 * Sync a file 717 * As this calls fsync (not fdatasync) there is no need for a write_inode 718 * after it. 719 */ 720 static inline int nfsd_dosync(struct file *filp, struct dentry *dp, 721 const struct file_operations *fop) 722 { 723 struct inode *inode = dp->d_inode; 724 int (*fsync) (struct file *, struct dentry *, int); 725 int err; 726 727 err = filemap_fdatawrite(inode->i_mapping); 728 if (err == 0 && fop && (fsync = fop->fsync)) 729 err = fsync(filp, dp, 0); 730 if (err == 0) 731 err = filemap_fdatawait(inode->i_mapping); 732 733 return err; 734 } 735 736 737 static int 738 nfsd_sync(struct file *filp) 739 { 740 int err; 741 struct inode *inode = filp->f_path.dentry->d_inode; 742 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name); 743 mutex_lock(&inode->i_mutex); 744 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op); 745 mutex_unlock(&inode->i_mutex); 746 747 return err; 748 } 749 750 int 751 nfsd_sync_dir(struct dentry *dp) 752 { 753 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop); 754 } 755 756 /* 757 * Obtain the readahead parameters for the file 758 * specified by (dev, ino). 759 */ 760 761 static inline struct raparms * 762 nfsd_get_raparms(dev_t dev, ino_t ino) 763 { 764 struct raparms *ra, **rap, **frap = NULL; 765 int depth = 0; 766 unsigned int hash; 767 struct raparm_hbucket *rab; 768 769 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 770 rab = &raparm_hash[hash]; 771 772 spin_lock(&rab->pb_lock); 773 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 774 if (ra->p_ino == ino && ra->p_dev == dev) 775 goto found; 776 depth++; 777 if (ra->p_count == 0) 778 frap = rap; 779 } 780 depth = nfsdstats.ra_size*11/10; 781 if (!frap) { 782 spin_unlock(&rab->pb_lock); 783 return NULL; 784 } 785 rap = frap; 786 ra = *frap; 787 ra->p_dev = dev; 788 ra->p_ino = ino; 789 ra->p_set = 0; 790 ra->p_hindex = hash; 791 found: 792 if (rap != &rab->pb_head) { 793 *rap = ra->p_next; 794 ra->p_next = rab->pb_head; 795 rab->pb_head = ra; 796 } 797 ra->p_count++; 798 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 799 spin_unlock(&rab->pb_lock); 800 return ra; 801 } 802 803 /* 804 * Grab and keep cached pages assosiated with a file in the svc_rqst 805 * so that they can be passed to the netowork sendmsg/sendpage routines 806 * directrly. They will be released after the sending has completed. 807 */ 808 static int 809 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size) 810 { 811 unsigned long count = desc->count; 812 struct svc_rqst *rqstp = desc->arg.data; 813 struct page **pp = rqstp->rq_respages + rqstp->rq_resused; 814 815 if (size > count) 816 size = count; 817 818 if (rqstp->rq_res.page_len == 0) { 819 get_page(page); 820 put_page(*pp); 821 *pp = page; 822 rqstp->rq_resused++; 823 rqstp->rq_res.page_base = offset; 824 rqstp->rq_res.page_len = size; 825 } else if (page != pp[-1]) { 826 get_page(page); 827 if (*pp) 828 put_page(*pp); 829 *pp = page; 830 rqstp->rq_resused++; 831 rqstp->rq_res.page_len += size; 832 } else 833 rqstp->rq_res.page_len += size; 834 835 desc->count = count - size; 836 desc->written += size; 837 return size; 838 } 839 840 static __be32 841 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 842 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 843 { 844 struct inode *inode; 845 struct raparms *ra; 846 mm_segment_t oldfs; 847 __be32 err; 848 int host_err; 849 850 err = nfserr_perm; 851 inode = file->f_path.dentry->d_inode; 852 #ifdef MSNFS 853 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 854 (!lock_may_read(inode, offset, *count))) 855 goto out; 856 #endif 857 858 /* Get readahead parameters */ 859 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 860 861 if (ra && ra->p_set) 862 file->f_ra = ra->p_ra; 863 864 if (file->f_op->sendfile && rqstp->rq_sendfile_ok) { 865 rqstp->rq_resused = 1; 866 host_err = file->f_op->sendfile(file, &offset, *count, 867 nfsd_read_actor, rqstp); 868 } else { 869 oldfs = get_fs(); 870 set_fs(KERNEL_DS); 871 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 872 set_fs(oldfs); 873 } 874 875 /* Write back readahead params */ 876 if (ra) { 877 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 878 spin_lock(&rab->pb_lock); 879 ra->p_ra = file->f_ra; 880 ra->p_set = 1; 881 ra->p_count--; 882 spin_unlock(&rab->pb_lock); 883 } 884 885 if (host_err >= 0) { 886 nfsdstats.io_read += host_err; 887 *count = host_err; 888 err = 0; 889 fsnotify_access(file->f_path.dentry); 890 } else 891 err = nfserrno(host_err); 892 out: 893 return err; 894 } 895 896 static void kill_suid(struct dentry *dentry) 897 { 898 struct iattr ia; 899 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID; 900 901 mutex_lock(&dentry->d_inode->i_mutex); 902 notify_change(dentry, &ia); 903 mutex_unlock(&dentry->d_inode->i_mutex); 904 } 905 906 static __be32 907 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 908 loff_t offset, struct kvec *vec, int vlen, 909 unsigned long cnt, int *stablep) 910 { 911 struct svc_export *exp; 912 struct dentry *dentry; 913 struct inode *inode; 914 mm_segment_t oldfs; 915 __be32 err = 0; 916 int host_err; 917 int stable = *stablep; 918 919 #ifdef MSNFS 920 err = nfserr_perm; 921 922 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 923 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt))) 924 goto out; 925 #endif 926 927 dentry = file->f_path.dentry; 928 inode = dentry->d_inode; 929 exp = fhp->fh_export; 930 931 /* 932 * Request sync writes if 933 * - the sync export option has been set, or 934 * - the client requested O_SYNC behavior (NFSv3 feature). 935 * - The file system doesn't support fsync(). 936 * When gathered writes have been configured for this volume, 937 * flushing the data to disk is handled separately below. 938 */ 939 940 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */ 941 stable = 2; 942 *stablep = 2; /* FILE_SYNC */ 943 } 944 945 if (!EX_ISSYNC(exp)) 946 stable = 0; 947 if (stable && !EX_WGATHER(exp)) 948 file->f_flags |= O_SYNC; 949 950 /* Write the data. */ 951 oldfs = get_fs(); set_fs(KERNEL_DS); 952 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); 953 set_fs(oldfs); 954 if (host_err >= 0) { 955 nfsdstats.io_write += cnt; 956 fsnotify_modify(file->f_path.dentry); 957 } 958 959 /* clear setuid/setgid flag after write */ 960 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) 961 kill_suid(dentry); 962 963 if (host_err >= 0 && stable) { 964 static ino_t last_ino; 965 static dev_t last_dev; 966 967 /* 968 * Gathered writes: If another process is currently 969 * writing to the file, there's a high chance 970 * this is another nfsd (triggered by a bulk write 971 * from a client's biod). Rather than syncing the 972 * file with each write request, we sleep for 10 msec. 973 * 974 * I don't know if this roughly approximates 975 * C. Juszak's idea of gathered writes, but it's a 976 * nice and simple solution (IMHO), and it seems to 977 * work:-) 978 */ 979 if (EX_WGATHER(exp)) { 980 if (atomic_read(&inode->i_writecount) > 1 981 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 982 dprintk("nfsd: write defer %d\n", current->pid); 983 msleep(10); 984 dprintk("nfsd: write resume %d\n", current->pid); 985 } 986 987 if (inode->i_state & I_DIRTY) { 988 dprintk("nfsd: write sync %d\n", current->pid); 989 host_err=nfsd_sync(file); 990 } 991 #if 0 992 wake_up(&inode->i_wait); 993 #endif 994 } 995 last_ino = inode->i_ino; 996 last_dev = inode->i_sb->s_dev; 997 } 998 999 dprintk("nfsd: write complete host_err=%d\n", host_err); 1000 if (host_err >= 0) 1001 err = 0; 1002 else 1003 err = nfserrno(host_err); 1004 out: 1005 return err; 1006 } 1007 1008 /* 1009 * Read data from a file. count must contain the requested read count 1010 * on entry. On return, *count contains the number of bytes actually read. 1011 * N.B. After this call fhp needs an fh_put 1012 */ 1013 __be32 1014 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1015 loff_t offset, struct kvec *vec, int vlen, 1016 unsigned long *count) 1017 { 1018 __be32 err; 1019 1020 if (file) { 1021 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry, 1022 MAY_READ|MAY_OWNER_OVERRIDE); 1023 if (err) 1024 goto out; 1025 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1026 } else { 1027 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file); 1028 if (err) 1029 goto out; 1030 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1031 nfsd_close(file); 1032 } 1033 out: 1034 return err; 1035 } 1036 1037 /* 1038 * Write data to a file. 1039 * The stable flag requests synchronous writes. 1040 * N.B. After this call fhp needs an fh_put 1041 */ 1042 __be32 1043 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1044 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt, 1045 int *stablep) 1046 { 1047 __be32 err = 0; 1048 1049 if (file) { 1050 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry, 1051 MAY_WRITE|MAY_OWNER_OVERRIDE); 1052 if (err) 1053 goto out; 1054 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1055 stablep); 1056 } else { 1057 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file); 1058 if (err) 1059 goto out; 1060 1061 if (cnt) 1062 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1063 cnt, stablep); 1064 nfsd_close(file); 1065 } 1066 out: 1067 return err; 1068 } 1069 1070 #ifdef CONFIG_NFSD_V3 1071 /* 1072 * Commit all pending writes to stable storage. 1073 * Strictly speaking, we could sync just the indicated file region here, 1074 * but there's currently no way we can ask the VFS to do so. 1075 * 1076 * Unfortunately we cannot lock the file to make sure we return full WCC 1077 * data to the client, as locking happens lower down in the filesystem. 1078 */ 1079 __be32 1080 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1081 loff_t offset, unsigned long count) 1082 { 1083 struct file *file; 1084 __be32 err; 1085 1086 if ((u64)count > ~(u64)offset) 1087 return nfserr_inval; 1088 1089 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0) 1090 return err; 1091 if (EX_ISSYNC(fhp->fh_export)) { 1092 if (file->f_op && file->f_op->fsync) { 1093 err = nfserrno(nfsd_sync(file)); 1094 } else { 1095 err = nfserr_notsupp; 1096 } 1097 } 1098 1099 nfsd_close(file); 1100 return err; 1101 } 1102 #endif /* CONFIG_NFSD_V3 */ 1103 1104 /* 1105 * Create a file (regular, directory, device, fifo); UNIX sockets 1106 * not yet implemented. 1107 * If the response fh has been verified, the parent directory should 1108 * already be locked. Note that the parent directory is left locked. 1109 * 1110 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1111 */ 1112 __be32 1113 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1114 char *fname, int flen, struct iattr *iap, 1115 int type, dev_t rdev, struct svc_fh *resfhp) 1116 { 1117 struct dentry *dentry, *dchild = NULL; 1118 struct inode *dirp; 1119 __be32 err; 1120 int host_err; 1121 1122 err = nfserr_perm; 1123 if (!flen) 1124 goto out; 1125 err = nfserr_exist; 1126 if (isdotent(fname, flen)) 1127 goto out; 1128 1129 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); 1130 if (err) 1131 goto out; 1132 1133 dentry = fhp->fh_dentry; 1134 dirp = dentry->d_inode; 1135 1136 err = nfserr_notdir; 1137 if(!dirp->i_op || !dirp->i_op->lookup) 1138 goto out; 1139 /* 1140 * Check whether the response file handle has been verified yet. 1141 * If it has, the parent directory should already be locked. 1142 */ 1143 if (!resfhp->fh_dentry) { 1144 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1145 fh_lock_nested(fhp, I_MUTEX_PARENT); 1146 dchild = lookup_one_len(fname, dentry, flen); 1147 host_err = PTR_ERR(dchild); 1148 if (IS_ERR(dchild)) 1149 goto out_nfserr; 1150 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1151 if (err) 1152 goto out; 1153 } else { 1154 /* called from nfsd_proc_create */ 1155 dchild = dget(resfhp->fh_dentry); 1156 if (!fhp->fh_locked) { 1157 /* not actually possible */ 1158 printk(KERN_ERR 1159 "nfsd_create: parent %s/%s not locked!\n", 1160 dentry->d_parent->d_name.name, 1161 dentry->d_name.name); 1162 err = nfserr_io; 1163 goto out; 1164 } 1165 } 1166 /* 1167 * Make sure the child dentry is still negative ... 1168 */ 1169 err = nfserr_exist; 1170 if (dchild->d_inode) { 1171 dprintk("nfsd_create: dentry %s/%s not negative!\n", 1172 dentry->d_name.name, dchild->d_name.name); 1173 goto out; 1174 } 1175 1176 if (!(iap->ia_valid & ATTR_MODE)) 1177 iap->ia_mode = 0; 1178 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1179 1180 /* 1181 * Get the dir op function pointer. 1182 */ 1183 err = 0; 1184 switch (type) { 1185 case S_IFREG: 1186 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1187 break; 1188 case S_IFDIR: 1189 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1190 break; 1191 case S_IFCHR: 1192 case S_IFBLK: 1193 case S_IFIFO: 1194 case S_IFSOCK: 1195 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1196 break; 1197 default: 1198 printk("nfsd: bad file type %o in nfsd_create\n", type); 1199 host_err = -EINVAL; 1200 } 1201 if (host_err < 0) 1202 goto out_nfserr; 1203 1204 if (EX_ISSYNC(fhp->fh_export)) { 1205 err = nfserrno(nfsd_sync_dir(dentry)); 1206 write_inode_now(dchild->d_inode, 1); 1207 } 1208 1209 1210 /* Set file attributes. Mode has already been set and 1211 * setting uid/gid works only for root. Irix appears to 1212 * send along the gid when it tries to implement setgid 1213 * directories via NFS. 1214 */ 1215 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) { 1216 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1217 if (err2) 1218 err = err2; 1219 } 1220 /* 1221 * Update the file handle to get the new inode info. 1222 */ 1223 if (!err) 1224 err = fh_update(resfhp); 1225 out: 1226 if (dchild && !IS_ERR(dchild)) 1227 dput(dchild); 1228 return err; 1229 1230 out_nfserr: 1231 err = nfserrno(host_err); 1232 goto out; 1233 } 1234 1235 #ifdef CONFIG_NFSD_V3 1236 /* 1237 * NFSv3 version of nfsd_create 1238 */ 1239 __be32 1240 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp, 1241 char *fname, int flen, struct iattr *iap, 1242 struct svc_fh *resfhp, int createmode, u32 *verifier, 1243 int *truncp, int *created) 1244 { 1245 struct dentry *dentry, *dchild = NULL; 1246 struct inode *dirp; 1247 __be32 err; 1248 int host_err; 1249 __u32 v_mtime=0, v_atime=0; 1250 1251 err = nfserr_perm; 1252 if (!flen) 1253 goto out; 1254 err = nfserr_exist; 1255 if (isdotent(fname, flen)) 1256 goto out; 1257 if (!(iap->ia_valid & ATTR_MODE)) 1258 iap->ia_mode = 0; 1259 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); 1260 if (err) 1261 goto out; 1262 1263 dentry = fhp->fh_dentry; 1264 dirp = dentry->d_inode; 1265 1266 /* Get all the sanity checks out of the way before 1267 * we lock the parent. */ 1268 err = nfserr_notdir; 1269 if(!dirp->i_op || !dirp->i_op->lookup) 1270 goto out; 1271 fh_lock_nested(fhp, I_MUTEX_PARENT); 1272 1273 /* 1274 * Compose the response file handle. 1275 */ 1276 dchild = lookup_one_len(fname, dentry, flen); 1277 host_err = PTR_ERR(dchild); 1278 if (IS_ERR(dchild)) 1279 goto out_nfserr; 1280 1281 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1282 if (err) 1283 goto out; 1284 1285 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1286 /* solaris7 gets confused (bugid 4218508) if these have 1287 * the high bit set, so just clear the high bits. 1288 */ 1289 v_mtime = verifier[0]&0x7fffffff; 1290 v_atime = verifier[1]&0x7fffffff; 1291 } 1292 1293 if (dchild->d_inode) { 1294 err = 0; 1295 1296 switch (createmode) { 1297 case NFS3_CREATE_UNCHECKED: 1298 if (! S_ISREG(dchild->d_inode->i_mode)) 1299 err = nfserr_exist; 1300 else if (truncp) { 1301 /* in nfsv4, we need to treat this case a little 1302 * differently. we don't want to truncate the 1303 * file now; this would be wrong if the OPEN 1304 * fails for some other reason. furthermore, 1305 * if the size is nonzero, we should ignore it 1306 * according to spec! 1307 */ 1308 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1309 } 1310 else { 1311 iap->ia_valid &= ATTR_SIZE; 1312 goto set_attr; 1313 } 1314 break; 1315 case NFS3_CREATE_EXCLUSIVE: 1316 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1317 && dchild->d_inode->i_atime.tv_sec == v_atime 1318 && dchild->d_inode->i_size == 0 ) 1319 break; 1320 /* fallthru */ 1321 case NFS3_CREATE_GUARDED: 1322 err = nfserr_exist; 1323 } 1324 goto out; 1325 } 1326 1327 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1328 if (host_err < 0) 1329 goto out_nfserr; 1330 if (created) 1331 *created = 1; 1332 1333 if (EX_ISSYNC(fhp->fh_export)) { 1334 err = nfserrno(nfsd_sync_dir(dentry)); 1335 /* setattr will sync the child (or not) */ 1336 } 1337 1338 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1339 /* Cram the verifier into atime/mtime */ 1340 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1341 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1342 /* XXX someone who knows this better please fix it for nsec */ 1343 iap->ia_mtime.tv_sec = v_mtime; 1344 iap->ia_atime.tv_sec = v_atime; 1345 iap->ia_mtime.tv_nsec = 0; 1346 iap->ia_atime.tv_nsec = 0; 1347 } 1348 1349 /* Set file attributes. 1350 * Irix appears to send along the gid when it tries to 1351 * implement setgid directories via NFS. Clear out all that cruft. 1352 */ 1353 set_attr: 1354 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) { 1355 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1356 if (err2) 1357 err = err2; 1358 } 1359 1360 /* 1361 * Update the filehandle to get the new inode info. 1362 */ 1363 if (!err) 1364 err = fh_update(resfhp); 1365 1366 out: 1367 fh_unlock(fhp); 1368 if (dchild && !IS_ERR(dchild)) 1369 dput(dchild); 1370 return err; 1371 1372 out_nfserr: 1373 err = nfserrno(host_err); 1374 goto out; 1375 } 1376 #endif /* CONFIG_NFSD_V3 */ 1377 1378 /* 1379 * Read a symlink. On entry, *lenp must contain the maximum path length that 1380 * fits into the buffer. On return, it contains the true length. 1381 * N.B. After this call fhp needs an fh_put 1382 */ 1383 __be32 1384 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1385 { 1386 struct dentry *dentry; 1387 struct inode *inode; 1388 mm_segment_t oldfs; 1389 __be32 err; 1390 int host_err; 1391 1392 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP); 1393 if (err) 1394 goto out; 1395 1396 dentry = fhp->fh_dentry; 1397 inode = dentry->d_inode; 1398 1399 err = nfserr_inval; 1400 if (!inode->i_op || !inode->i_op->readlink) 1401 goto out; 1402 1403 touch_atime(fhp->fh_export->ex_mnt, dentry); 1404 /* N.B. Why does this call need a get_fs()?? 1405 * Remove the set_fs and watch the fireworks:-) --okir 1406 */ 1407 1408 oldfs = get_fs(); set_fs(KERNEL_DS); 1409 host_err = inode->i_op->readlink(dentry, buf, *lenp); 1410 set_fs(oldfs); 1411 1412 if (host_err < 0) 1413 goto out_nfserr; 1414 *lenp = host_err; 1415 err = 0; 1416 out: 1417 return err; 1418 1419 out_nfserr: 1420 err = nfserrno(host_err); 1421 goto out; 1422 } 1423 1424 /* 1425 * Create a symlink and look up its inode 1426 * N.B. After this call _both_ fhp and resfhp need an fh_put 1427 */ 1428 __be32 1429 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1430 char *fname, int flen, 1431 char *path, int plen, 1432 struct svc_fh *resfhp, 1433 struct iattr *iap) 1434 { 1435 struct dentry *dentry, *dnew; 1436 __be32 err, cerr; 1437 int host_err; 1438 umode_t mode; 1439 1440 err = nfserr_noent; 1441 if (!flen || !plen) 1442 goto out; 1443 err = nfserr_exist; 1444 if (isdotent(fname, flen)) 1445 goto out; 1446 1447 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); 1448 if (err) 1449 goto out; 1450 fh_lock(fhp); 1451 dentry = fhp->fh_dentry; 1452 dnew = lookup_one_len(fname, dentry, flen); 1453 host_err = PTR_ERR(dnew); 1454 if (IS_ERR(dnew)) 1455 goto out_nfserr; 1456 1457 mode = S_IALLUGO; 1458 /* Only the MODE ATTRibute is even vaguely meaningful */ 1459 if (iap && (iap->ia_valid & ATTR_MODE)) 1460 mode = iap->ia_mode & S_IALLUGO; 1461 1462 if (unlikely(path[plen] != 0)) { 1463 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1464 if (path_alloced == NULL) 1465 host_err = -ENOMEM; 1466 else { 1467 strncpy(path_alloced, path, plen); 1468 path_alloced[plen] = 0; 1469 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode); 1470 kfree(path_alloced); 1471 } 1472 } else 1473 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode); 1474 1475 if (!host_err) { 1476 if (EX_ISSYNC(fhp->fh_export)) 1477 host_err = nfsd_sync_dir(dentry); 1478 } 1479 err = nfserrno(host_err); 1480 fh_unlock(fhp); 1481 1482 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1483 dput(dnew); 1484 if (err==0) err = cerr; 1485 out: 1486 return err; 1487 1488 out_nfserr: 1489 err = nfserrno(host_err); 1490 goto out; 1491 } 1492 1493 /* 1494 * Create a hardlink 1495 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1496 */ 1497 __be32 1498 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1499 char *name, int len, struct svc_fh *tfhp) 1500 { 1501 struct dentry *ddir, *dnew, *dold; 1502 struct inode *dirp, *dest; 1503 __be32 err; 1504 int host_err; 1505 1506 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE); 1507 if (err) 1508 goto out; 1509 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP); 1510 if (err) 1511 goto out; 1512 1513 err = nfserr_perm; 1514 if (!len) 1515 goto out; 1516 err = nfserr_exist; 1517 if (isdotent(name, len)) 1518 goto out; 1519 1520 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1521 ddir = ffhp->fh_dentry; 1522 dirp = ddir->d_inode; 1523 1524 dnew = lookup_one_len(name, ddir, len); 1525 host_err = PTR_ERR(dnew); 1526 if (IS_ERR(dnew)) 1527 goto out_nfserr; 1528 1529 dold = tfhp->fh_dentry; 1530 dest = dold->d_inode; 1531 1532 host_err = vfs_link(dold, dirp, dnew); 1533 if (!host_err) { 1534 if (EX_ISSYNC(ffhp->fh_export)) { 1535 err = nfserrno(nfsd_sync_dir(ddir)); 1536 write_inode_now(dest, 1); 1537 } 1538 err = 0; 1539 } else { 1540 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1541 err = nfserr_acces; 1542 else 1543 err = nfserrno(host_err); 1544 } 1545 1546 dput(dnew); 1547 out_unlock: 1548 fh_unlock(ffhp); 1549 out: 1550 return err; 1551 1552 out_nfserr: 1553 err = nfserrno(host_err); 1554 goto out_unlock; 1555 } 1556 1557 /* 1558 * Rename a file 1559 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1560 */ 1561 __be32 1562 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1563 struct svc_fh *tfhp, char *tname, int tlen) 1564 { 1565 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1566 struct inode *fdir, *tdir; 1567 __be32 err; 1568 int host_err; 1569 1570 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE); 1571 if (err) 1572 goto out; 1573 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE); 1574 if (err) 1575 goto out; 1576 1577 fdentry = ffhp->fh_dentry; 1578 fdir = fdentry->d_inode; 1579 1580 tdentry = tfhp->fh_dentry; 1581 tdir = tdentry->d_inode; 1582 1583 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; 1584 if (ffhp->fh_export != tfhp->fh_export) 1585 goto out; 1586 1587 err = nfserr_perm; 1588 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1589 goto out; 1590 1591 /* cannot use fh_lock as we need deadlock protective ordering 1592 * so do it by hand */ 1593 trap = lock_rename(tdentry, fdentry); 1594 ffhp->fh_locked = tfhp->fh_locked = 1; 1595 fill_pre_wcc(ffhp); 1596 fill_pre_wcc(tfhp); 1597 1598 odentry = lookup_one_len(fname, fdentry, flen); 1599 host_err = PTR_ERR(odentry); 1600 if (IS_ERR(odentry)) 1601 goto out_nfserr; 1602 1603 host_err = -ENOENT; 1604 if (!odentry->d_inode) 1605 goto out_dput_old; 1606 host_err = -EINVAL; 1607 if (odentry == trap) 1608 goto out_dput_old; 1609 1610 ndentry = lookup_one_len(tname, tdentry, tlen); 1611 host_err = PTR_ERR(ndentry); 1612 if (IS_ERR(ndentry)) 1613 goto out_dput_old; 1614 host_err = -ENOTEMPTY; 1615 if (ndentry == trap) 1616 goto out_dput_new; 1617 1618 #ifdef MSNFS 1619 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1620 ((atomic_read(&odentry->d_count) > 1) 1621 || (atomic_read(&ndentry->d_count) > 1))) { 1622 host_err = -EPERM; 1623 } else 1624 #endif 1625 host_err = vfs_rename(fdir, odentry, tdir, ndentry); 1626 if (!host_err && EX_ISSYNC(tfhp->fh_export)) { 1627 host_err = nfsd_sync_dir(tdentry); 1628 if (!host_err) 1629 host_err = nfsd_sync_dir(fdentry); 1630 } 1631 1632 out_dput_new: 1633 dput(ndentry); 1634 out_dput_old: 1635 dput(odentry); 1636 out_nfserr: 1637 err = nfserrno(host_err); 1638 1639 /* we cannot reply on fh_unlock on the two filehandles, 1640 * as that would do the wrong thing if the two directories 1641 * were the same, so again we do it by hand 1642 */ 1643 fill_post_wcc(ffhp); 1644 fill_post_wcc(tfhp); 1645 unlock_rename(tdentry, fdentry); 1646 ffhp->fh_locked = tfhp->fh_locked = 0; 1647 1648 out: 1649 return err; 1650 } 1651 1652 /* 1653 * Unlink a file or directory 1654 * N.B. After this call fhp needs an fh_put 1655 */ 1656 __be32 1657 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1658 char *fname, int flen) 1659 { 1660 struct dentry *dentry, *rdentry; 1661 struct inode *dirp; 1662 __be32 err; 1663 int host_err; 1664 1665 err = nfserr_acces; 1666 if (!flen || isdotent(fname, flen)) 1667 goto out; 1668 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE); 1669 if (err) 1670 goto out; 1671 1672 fh_lock_nested(fhp, I_MUTEX_PARENT); 1673 dentry = fhp->fh_dentry; 1674 dirp = dentry->d_inode; 1675 1676 rdentry = lookup_one_len(fname, dentry, flen); 1677 host_err = PTR_ERR(rdentry); 1678 if (IS_ERR(rdentry)) 1679 goto out_nfserr; 1680 1681 if (!rdentry->d_inode) { 1682 dput(rdentry); 1683 err = nfserr_noent; 1684 goto out; 1685 } 1686 1687 if (!type) 1688 type = rdentry->d_inode->i_mode & S_IFMT; 1689 1690 if (type != S_IFDIR) { /* It's UNLINK */ 1691 #ifdef MSNFS 1692 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1693 (atomic_read(&rdentry->d_count) > 1)) { 1694 host_err = -EPERM; 1695 } else 1696 #endif 1697 host_err = vfs_unlink(dirp, rdentry); 1698 } else { /* It's RMDIR */ 1699 host_err = vfs_rmdir(dirp, rdentry); 1700 } 1701 1702 dput(rdentry); 1703 1704 if (host_err) 1705 goto out_nfserr; 1706 if (EX_ISSYNC(fhp->fh_export)) 1707 host_err = nfsd_sync_dir(dentry); 1708 1709 out_nfserr: 1710 err = nfserrno(host_err); 1711 out: 1712 return err; 1713 } 1714 1715 /* 1716 * Read entries from a directory. 1717 * The NFSv3/4 verifier we ignore for now. 1718 */ 1719 __be32 1720 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 1721 struct readdir_cd *cdp, filldir_t func) 1722 { 1723 __be32 err; 1724 int host_err; 1725 struct file *file; 1726 loff_t offset = *offsetp; 1727 1728 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file); 1729 if (err) 1730 goto out; 1731 1732 offset = vfs_llseek(file, offset, 0); 1733 if (offset < 0) { 1734 err = nfserrno((int)offset); 1735 goto out_close; 1736 } 1737 1738 /* 1739 * Read the directory entries. This silly loop is necessary because 1740 * readdir() is not guaranteed to fill up the entire buffer, but 1741 * may choose to do less. 1742 */ 1743 1744 do { 1745 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1746 host_err = vfs_readdir(file, func, cdp); 1747 } while (host_err >=0 && cdp->err == nfs_ok); 1748 if (host_err) 1749 err = nfserrno(host_err); 1750 else 1751 err = cdp->err; 1752 *offsetp = vfs_llseek(file, 0, 1); 1753 1754 if (err == nfserr_eof || err == nfserr_toosmall) 1755 err = nfs_ok; /* can still be found in ->err */ 1756 out_close: 1757 nfsd_close(file); 1758 out: 1759 return err; 1760 } 1761 1762 /* 1763 * Get file system stats 1764 * N.B. After this call fhp needs an fh_put 1765 */ 1766 __be32 1767 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat) 1768 { 1769 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP); 1770 if (!err && vfs_statfs(fhp->fh_dentry,stat)) 1771 err = nfserr_io; 1772 return err; 1773 } 1774 1775 /* 1776 * Check for a user's access permissions to this inode. 1777 */ 1778 __be32 1779 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc) 1780 { 1781 struct inode *inode = dentry->d_inode; 1782 int err; 1783 1784 if (acc == MAY_NOP) 1785 return 0; 1786 #if 0 1787 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 1788 acc, 1789 (acc & MAY_READ)? " read" : "", 1790 (acc & MAY_WRITE)? " write" : "", 1791 (acc & MAY_EXEC)? " exec" : "", 1792 (acc & MAY_SATTR)? " sattr" : "", 1793 (acc & MAY_TRUNC)? " trunc" : "", 1794 (acc & MAY_LOCK)? " lock" : "", 1795 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "", 1796 inode->i_mode, 1797 IS_IMMUTABLE(inode)? " immut" : "", 1798 IS_APPEND(inode)? " append" : "", 1799 IS_RDONLY(inode)? " ro" : ""); 1800 dprintk(" owner %d/%d user %d/%d\n", 1801 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid); 1802 #endif 1803 1804 /* Normally we reject any write/sattr etc access on a read-only file 1805 * system. But if it is IRIX doing check on write-access for a 1806 * device special file, we ignore rofs. 1807 */ 1808 if (!(acc & MAY_LOCAL_ACCESS)) 1809 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) { 1810 if (EX_RDONLY(exp) || IS_RDONLY(inode)) 1811 return nfserr_rofs; 1812 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode)) 1813 return nfserr_perm; 1814 } 1815 if ((acc & MAY_TRUNC) && IS_APPEND(inode)) 1816 return nfserr_perm; 1817 1818 if (acc & MAY_LOCK) { 1819 /* If we cannot rely on authentication in NLM requests, 1820 * just allow locks, otherwise require read permission, or 1821 * ownership 1822 */ 1823 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 1824 return 0; 1825 else 1826 acc = MAY_READ | MAY_OWNER_OVERRIDE; 1827 } 1828 /* 1829 * The file owner always gets access permission for accesses that 1830 * would normally be checked at open time. This is to make 1831 * file access work even when the client has done a fchmod(fd, 0). 1832 * 1833 * However, `cp foo bar' should fail nevertheless when bar is 1834 * readonly. A sensible way to do this might be to reject all 1835 * attempts to truncate a read-only file, because a creat() call 1836 * always implies file truncation. 1837 * ... but this isn't really fair. A process may reasonably call 1838 * ftruncate on an open file descriptor on a file with perm 000. 1839 * We must trust the client to do permission checking - using "ACCESS" 1840 * with NFSv3. 1841 */ 1842 if ((acc & MAY_OWNER_OVERRIDE) && 1843 inode->i_uid == current->fsuid) 1844 return 0; 1845 1846 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL); 1847 1848 /* Allow read access to binaries even when mode 111 */ 1849 if (err == -EACCES && S_ISREG(inode->i_mode) && 1850 acc == (MAY_READ | MAY_OWNER_OVERRIDE)) 1851 err = permission(inode, MAY_EXEC, NULL); 1852 1853 return err? nfserrno(err) : 0; 1854 } 1855 1856 void 1857 nfsd_racache_shutdown(void) 1858 { 1859 if (!raparml) 1860 return; 1861 dprintk("nfsd: freeing readahead buffers.\n"); 1862 kfree(raparml); 1863 raparml = NULL; 1864 } 1865 /* 1866 * Initialize readahead param cache 1867 */ 1868 int 1869 nfsd_racache_init(int cache_size) 1870 { 1871 int i; 1872 int j = 0; 1873 int nperbucket; 1874 1875 1876 if (raparml) 1877 return 0; 1878 if (cache_size < 2*RAPARM_HASH_SIZE) 1879 cache_size = 2*RAPARM_HASH_SIZE; 1880 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL); 1881 1882 if (!raparml) { 1883 printk(KERN_WARNING 1884 "nfsd: Could not allocate memory read-ahead cache.\n"); 1885 return -ENOMEM; 1886 } 1887 1888 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 1889 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) { 1890 raparm_hash[i].pb_head = NULL; 1891 spin_lock_init(&raparm_hash[i].pb_lock); 1892 } 1893 nperbucket = cache_size >> RAPARM_HASH_BITS; 1894 for (i = 0; i < cache_size - 1; i++) { 1895 if (i % nperbucket == 0) 1896 raparm_hash[j++].pb_head = raparml + i; 1897 if (i % nperbucket < nperbucket-1) 1898 raparml[i].p_next = raparml + i + 1; 1899 } 1900 1901 nfsdstats.ra_size = cache_size; 1902 return 0; 1903 } 1904 1905 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 1906 struct posix_acl * 1907 nfsd_get_posix_acl(struct svc_fh *fhp, int type) 1908 { 1909 struct inode *inode = fhp->fh_dentry->d_inode; 1910 char *name; 1911 void *value = NULL; 1912 ssize_t size; 1913 struct posix_acl *acl; 1914 1915 if (!IS_POSIXACL(inode)) 1916 return ERR_PTR(-EOPNOTSUPP); 1917 1918 switch (type) { 1919 case ACL_TYPE_ACCESS: 1920 name = POSIX_ACL_XATTR_ACCESS; 1921 break; 1922 case ACL_TYPE_DEFAULT: 1923 name = POSIX_ACL_XATTR_DEFAULT; 1924 break; 1925 default: 1926 return ERR_PTR(-EOPNOTSUPP); 1927 } 1928 1929 size = nfsd_getxattr(fhp->fh_dentry, name, &value); 1930 if (size < 0) 1931 return ERR_PTR(size); 1932 1933 acl = posix_acl_from_xattr(value, size); 1934 kfree(value); 1935 return acl; 1936 } 1937 1938 int 1939 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) 1940 { 1941 struct inode *inode = fhp->fh_dentry->d_inode; 1942 char *name; 1943 void *value = NULL; 1944 size_t size; 1945 int error; 1946 1947 if (!IS_POSIXACL(inode) || !inode->i_op || 1948 !inode->i_op->setxattr || !inode->i_op->removexattr) 1949 return -EOPNOTSUPP; 1950 switch(type) { 1951 case ACL_TYPE_ACCESS: 1952 name = POSIX_ACL_XATTR_ACCESS; 1953 break; 1954 case ACL_TYPE_DEFAULT: 1955 name = POSIX_ACL_XATTR_DEFAULT; 1956 break; 1957 default: 1958 return -EOPNOTSUPP; 1959 } 1960 1961 if (acl && acl->a_count) { 1962 size = posix_acl_xattr_size(acl->a_count); 1963 value = kmalloc(size, GFP_KERNEL); 1964 if (!value) 1965 return -ENOMEM; 1966 error = posix_acl_to_xattr(acl, value, size); 1967 if (error < 0) 1968 goto getout; 1969 size = error; 1970 } else 1971 size = 0; 1972 1973 if (size) 1974 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); 1975 else { 1976 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) 1977 error = 0; 1978 else { 1979 error = vfs_removexattr(fhp->fh_dentry, name); 1980 if (error == -ENODATA) 1981 error = 0; 1982 } 1983 } 1984 1985 getout: 1986 kfree(value); 1987 return error; 1988 } 1989 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 1990