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/config.h> 20 #include <linux/string.h> 21 #include <linux/time.h> 22 #include <linux/errno.h> 23 #include <linux/fs.h> 24 #include <linux/file.h> 25 #include <linux/mount.h> 26 #include <linux/major.h> 27 #include <linux/ext2_fs.h> 28 #include <linux/proc_fs.h> 29 #include <linux/stat.h> 30 #include <linux/fcntl.h> 31 #include <linux/net.h> 32 #include <linux/unistd.h> 33 #include <linux/slab.h> 34 #include <linux/pagemap.h> 35 #include <linux/in.h> 36 #include <linux/module.h> 37 #include <linux/namei.h> 38 #include <linux/vfs.h> 39 #include <linux/delay.h> 40 #include <linux/sunrpc/svc.h> 41 #include <linux/nfsd/nfsd.h> 42 #ifdef CONFIG_NFSD_V3 43 #include <linux/nfs3.h> 44 #include <linux/nfsd/xdr3.h> 45 #endif /* CONFIG_NFSD_V3 */ 46 #include <linux/nfsd/nfsfh.h> 47 #include <linux/quotaops.h> 48 #include <linux/fsnotify.h> 49 #include <linux/posix_acl.h> 50 #include <linux/posix_acl_xattr.h> 51 #ifdef CONFIG_NFSD_V4 52 #include <linux/xattr.h> 53 #include <linux/nfs4.h> 54 #include <linux/nfs4_acl.h> 55 #include <linux/nfsd_idmap.h> 56 #include <linux/security.h> 57 #endif /* CONFIG_NFSD_V4 */ 58 59 #include <asm/uaccess.h> 60 61 #define NFSDDBG_FACILITY NFSDDBG_FILEOP 62 #define NFSD_PARANOIA 63 64 65 /* We must ignore files (but only files) which might have mandatory 66 * locks on them because there is no way to know if the accesser has 67 * the lock. 68 */ 69 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i)) 70 71 /* 72 * This is a cache of readahead params that help us choose the proper 73 * readahead strategy. Initially, we set all readahead parameters to 0 74 * and let the VFS handle things. 75 * If you increase the number of cached files very much, you'll need to 76 * add a hash table here. 77 */ 78 struct raparms { 79 struct raparms *p_next; 80 unsigned int p_count; 81 ino_t p_ino; 82 dev_t p_dev; 83 int p_set; 84 struct file_ra_state p_ra; 85 }; 86 87 static struct raparms * raparml; 88 static struct raparms * raparm_cache; 89 90 /* 91 * Called from nfsd_lookup and encode_dirent. Check if we have crossed 92 * a mount point. 93 * Returns -EAGAIN leaving *dpp and *expp unchanged, 94 * or nfs_ok having possibly changed *dpp and *expp 95 */ 96 int 97 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 98 struct svc_export **expp) 99 { 100 struct svc_export *exp = *expp, *exp2 = NULL; 101 struct dentry *dentry = *dpp; 102 struct vfsmount *mnt = mntget(exp->ex_mnt); 103 struct dentry *mounts = dget(dentry); 104 int err = nfs_ok; 105 106 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts)); 107 108 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle); 109 if (IS_ERR(exp2)) { 110 err = PTR_ERR(exp2); 111 dput(mounts); 112 mntput(mnt); 113 goto out; 114 } 115 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) { 116 /* successfully crossed mount point */ 117 exp_put(exp); 118 *expp = exp2; 119 dput(dentry); 120 *dpp = mounts; 121 } else { 122 if (exp2) exp_put(exp2); 123 dput(mounts); 124 } 125 mntput(mnt); 126 out: 127 return err; 128 } 129 130 /* 131 * Look up one component of a pathname. 132 * N.B. After this call _both_ fhp and resfh need an fh_put 133 * 134 * If the lookup would cross a mountpoint, and the mounted filesystem 135 * is exported to the client with NFSEXP_NOHIDE, then the lookup is 136 * accepted as it stands and the mounted directory is 137 * returned. Otherwise the covered directory is returned. 138 * NOTE: this mountpoint crossing is not supported properly by all 139 * clients and is explicitly disallowed for NFSv3 140 * NeilBrown <neilb@cse.unsw.edu.au> 141 */ 142 int 143 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, 144 int len, struct svc_fh *resfh) 145 { 146 struct svc_export *exp; 147 struct dentry *dparent; 148 struct dentry *dentry; 149 int err; 150 151 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); 152 153 /* Obtain dentry and export. */ 154 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC); 155 if (err) 156 return err; 157 158 dparent = fhp->fh_dentry; 159 exp = fhp->fh_export; 160 exp_get(exp); 161 162 err = nfserr_acces; 163 164 /* Lookup the name, but don't follow links */ 165 if (isdotent(name, len)) { 166 if (len==1) 167 dentry = dget(dparent); 168 else if (dparent != exp->ex_dentry) { 169 dentry = dget_parent(dparent); 170 } else if (!EX_NOHIDE(exp)) 171 dentry = dget(dparent); /* .. == . just like at / */ 172 else { 173 /* checking mountpoint crossing is very different when stepping up */ 174 struct svc_export *exp2 = NULL; 175 struct dentry *dp; 176 struct vfsmount *mnt = mntget(exp->ex_mnt); 177 dentry = dget(dparent); 178 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry)) 179 ; 180 dp = dget_parent(dentry); 181 dput(dentry); 182 dentry = dp; 183 184 exp2 = exp_parent(exp->ex_client, mnt, dentry, 185 &rqstp->rq_chandle); 186 if (IS_ERR(exp2)) { 187 err = PTR_ERR(exp2); 188 dput(dentry); 189 mntput(mnt); 190 goto out_nfserr; 191 } 192 if (!exp2) { 193 dput(dentry); 194 dentry = dget(dparent); 195 } else { 196 exp_put(exp); 197 exp = exp2; 198 } 199 mntput(mnt); 200 } 201 } else { 202 fh_lock(fhp); 203 dentry = lookup_one_len(name, dparent, len); 204 err = PTR_ERR(dentry); 205 if (IS_ERR(dentry)) 206 goto out_nfserr; 207 /* 208 * check if we have crossed a mount point ... 209 */ 210 if (d_mountpoint(dentry)) { 211 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { 212 dput(dentry); 213 goto out_nfserr; 214 } 215 } 216 } 217 /* 218 * Note: we compose the file handle now, but as the 219 * dentry may be negative, it may need to be updated. 220 */ 221 err = fh_compose(resfh, exp, dentry, fhp); 222 if (!err && !dentry->d_inode) 223 err = nfserr_noent; 224 dput(dentry); 225 out: 226 exp_put(exp); 227 return err; 228 229 out_nfserr: 230 err = nfserrno(err); 231 goto out; 232 } 233 234 /* 235 * Set various file attributes. 236 * N.B. After this call fhp needs an fh_put 237 */ 238 int 239 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, 240 int check_guard, time_t guardtime) 241 { 242 struct dentry *dentry; 243 struct inode *inode; 244 int accmode = MAY_SATTR; 245 int ftype = 0; 246 int imode; 247 int err; 248 int size_change = 0; 249 250 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) 251 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE; 252 if (iap->ia_valid & ATTR_SIZE) 253 ftype = S_IFREG; 254 255 /* Get inode */ 256 err = fh_verify(rqstp, fhp, ftype, accmode); 257 if (err) 258 goto out; 259 260 dentry = fhp->fh_dentry; 261 inode = dentry->d_inode; 262 263 /* Ignore any mode updates on symlinks */ 264 if (S_ISLNK(inode->i_mode)) 265 iap->ia_valid &= ~ATTR_MODE; 266 267 if (!iap->ia_valid) 268 goto out; 269 270 /* NFSv2 does not differentiate between "set-[ac]time-to-now" 271 * which only requires access, and "set-[ac]time-to-X" which 272 * requires ownership. 273 * So if it looks like it might be "set both to the same time which 274 * is close to now", and if inode_change_ok fails, then we 275 * convert to "set to now" instead of "set to explicit time" 276 * 277 * We only call inode_change_ok as the last test as technically 278 * it is not an interface that we should be using. It is only 279 * valid if the filesystem does not define it's own i_op->setattr. 280 */ 281 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) 282 #define MAX_TOUCH_TIME_ERROR (30*60) 283 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET 284 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec 285 ) { 286 /* Looks probable. Now just make sure time is in the right ballpark. 287 * Solaris, at least, doesn't seem to care what the time request is. 288 * We require it be within 30 minutes of now. 289 */ 290 time_t delta = iap->ia_atime.tv_sec - get_seconds(); 291 if (delta<0) delta = -delta; 292 if (delta < MAX_TOUCH_TIME_ERROR && 293 inode_change_ok(inode, iap) != 0) { 294 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME 295 * this will cause notify_change to set these times to "now" 296 */ 297 iap->ia_valid &= ~BOTH_TIME_SET; 298 } 299 } 300 301 /* The size case is special. It changes the file as well as the attributes. */ 302 if (iap->ia_valid & ATTR_SIZE) { 303 if (iap->ia_size < inode->i_size) { 304 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE); 305 if (err) 306 goto out; 307 } 308 309 /* 310 * If we are changing the size of the file, then 311 * we need to break all leases. 312 */ 313 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK); 314 if (err == -EWOULDBLOCK) 315 err = -ETIMEDOUT; 316 if (err) /* ENOMEM or EWOULDBLOCK */ 317 goto out_nfserr; 318 319 err = get_write_access(inode); 320 if (err) 321 goto out_nfserr; 322 323 size_change = 1; 324 err = locks_verify_truncate(inode, NULL, iap->ia_size); 325 if (err) { 326 put_write_access(inode); 327 goto out_nfserr; 328 } 329 DQUOT_INIT(inode); 330 } 331 332 imode = inode->i_mode; 333 if (iap->ia_valid & ATTR_MODE) { 334 iap->ia_mode &= S_IALLUGO; 335 imode = iap->ia_mode |= (imode & ~S_IALLUGO); 336 } 337 338 /* Revoke setuid/setgid bit on chown/chgrp */ 339 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) 340 iap->ia_valid |= ATTR_KILL_SUID; 341 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid) 342 iap->ia_valid |= ATTR_KILL_SGID; 343 344 /* Change the attributes. */ 345 346 iap->ia_valid |= ATTR_CTIME; 347 348 err = nfserr_notsync; 349 if (!check_guard || guardtime == inode->i_ctime.tv_sec) { 350 fh_lock(fhp); 351 err = notify_change(dentry, iap); 352 err = nfserrno(err); 353 fh_unlock(fhp); 354 } 355 if (size_change) 356 put_write_access(inode); 357 if (!err) 358 if (EX_ISSYNC(fhp->fh_export)) 359 write_inode_now(inode, 1); 360 out: 361 return err; 362 363 out_nfserr: 364 err = nfserrno(err); 365 goto out; 366 } 367 368 #if defined(CONFIG_NFSD_V4) 369 370 static int 371 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) 372 { 373 int len; 374 size_t buflen; 375 char *buf = NULL; 376 int error = 0; 377 struct inode *inode = dentry->d_inode; 378 379 buflen = posix_acl_xattr_size(pacl->a_count); 380 buf = kmalloc(buflen, GFP_KERNEL); 381 error = -ENOMEM; 382 if (buf == NULL) 383 goto out; 384 385 len = posix_acl_to_xattr(pacl, buf, buflen); 386 if (len < 0) { 387 error = len; 388 goto out; 389 } 390 391 error = -EOPNOTSUPP; 392 if (inode->i_op && inode->i_op->setxattr) { 393 down(&inode->i_sem); 394 security_inode_setxattr(dentry, key, buf, len, 0); 395 error = inode->i_op->setxattr(dentry, key, buf, len, 0); 396 if (!error) 397 security_inode_post_setxattr(dentry, key, buf, len, 0); 398 up(&inode->i_sem); 399 } 400 out: 401 kfree(buf); 402 return error; 403 } 404 405 int 406 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, 407 struct nfs4_acl *acl) 408 { 409 int error; 410 struct dentry *dentry; 411 struct inode *inode; 412 struct posix_acl *pacl = NULL, *dpacl = NULL; 413 unsigned int flags = 0; 414 415 /* Get inode */ 416 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR); 417 if (error) 418 goto out; 419 420 dentry = fhp->fh_dentry; 421 inode = dentry->d_inode; 422 if (S_ISDIR(inode->i_mode)) 423 flags = NFS4_ACL_DIR; 424 425 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); 426 if (error == -EINVAL) { 427 error = nfserr_attrnotsupp; 428 goto out; 429 } else if (error < 0) 430 goto out_nfserr; 431 432 if (pacl) { 433 error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); 434 if (error < 0) 435 goto out_nfserr; 436 } 437 438 if (dpacl) { 439 error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); 440 if (error < 0) 441 goto out_nfserr; 442 } 443 444 error = nfs_ok; 445 446 out: 447 posix_acl_release(pacl); 448 posix_acl_release(dpacl); 449 return (error); 450 out_nfserr: 451 error = nfserrno(error); 452 goto out; 453 } 454 455 static struct posix_acl * 456 _get_posix_acl(struct dentry *dentry, char *key) 457 { 458 struct inode *inode = dentry->d_inode; 459 char *buf = NULL; 460 int buflen, error = 0; 461 struct posix_acl *pacl = NULL; 462 463 error = -EOPNOTSUPP; 464 if (inode->i_op == NULL) 465 goto out_err; 466 if (inode->i_op->getxattr == NULL) 467 goto out_err; 468 469 error = security_inode_getxattr(dentry, key); 470 if (error) 471 goto out_err; 472 473 buflen = inode->i_op->getxattr(dentry, key, NULL, 0); 474 if (buflen <= 0) { 475 error = buflen < 0 ? buflen : -ENODATA; 476 goto out_err; 477 } 478 479 buf = kmalloc(buflen, GFP_KERNEL); 480 if (buf == NULL) { 481 error = -ENOMEM; 482 goto out_err; 483 } 484 485 error = inode->i_op->getxattr(dentry, key, buf, buflen); 486 if (error < 0) 487 goto out_err; 488 489 pacl = posix_acl_from_xattr(buf, buflen); 490 out: 491 kfree(buf); 492 return pacl; 493 out_err: 494 pacl = ERR_PTR(error); 495 goto out; 496 } 497 498 int 499 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) 500 { 501 struct inode *inode = dentry->d_inode; 502 int error = 0; 503 struct posix_acl *pacl = NULL, *dpacl = NULL; 504 unsigned int flags = 0; 505 506 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); 507 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) 508 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); 509 if (IS_ERR(pacl)) { 510 error = PTR_ERR(pacl); 511 pacl = NULL; 512 goto out; 513 } 514 515 if (S_ISDIR(inode->i_mode)) { 516 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); 517 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) 518 dpacl = NULL; 519 else if (IS_ERR(dpacl)) { 520 error = PTR_ERR(dpacl); 521 dpacl = NULL; 522 goto out; 523 } 524 flags = NFS4_ACL_DIR; 525 } 526 527 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); 528 if (IS_ERR(*acl)) { 529 error = PTR_ERR(*acl); 530 *acl = NULL; 531 } 532 out: 533 posix_acl_release(pacl); 534 posix_acl_release(dpacl); 535 return error; 536 } 537 538 #endif /* defined(CONFIG_NFS_V4) */ 539 540 #ifdef CONFIG_NFSD_V3 541 /* 542 * Check server access rights to a file system object 543 */ 544 struct accessmap { 545 u32 access; 546 int how; 547 }; 548 static struct accessmap nfs3_regaccess[] = { 549 { NFS3_ACCESS_READ, MAY_READ }, 550 { NFS3_ACCESS_EXECUTE, MAY_EXEC }, 551 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC }, 552 { NFS3_ACCESS_EXTEND, MAY_WRITE }, 553 554 { 0, 0 } 555 }; 556 557 static struct accessmap nfs3_diraccess[] = { 558 { NFS3_ACCESS_READ, MAY_READ }, 559 { NFS3_ACCESS_LOOKUP, MAY_EXEC }, 560 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC }, 561 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE }, 562 { NFS3_ACCESS_DELETE, MAY_REMOVE }, 563 564 { 0, 0 } 565 }; 566 567 static struct accessmap nfs3_anyaccess[] = { 568 /* Some clients - Solaris 2.6 at least, make an access call 569 * to the server to check for access for things like /dev/null 570 * (which really, the server doesn't care about). So 571 * We provide simple access checking for them, looking 572 * mainly at mode bits, and we make sure to ignore read-only 573 * filesystem checks 574 */ 575 { NFS3_ACCESS_READ, MAY_READ }, 576 { NFS3_ACCESS_EXECUTE, MAY_EXEC }, 577 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS }, 578 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS }, 579 580 { 0, 0 } 581 }; 582 583 int 584 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) 585 { 586 struct accessmap *map; 587 struct svc_export *export; 588 struct dentry *dentry; 589 u32 query, result = 0, sresult = 0; 590 unsigned int error; 591 592 error = fh_verify(rqstp, fhp, 0, MAY_NOP); 593 if (error) 594 goto out; 595 596 export = fhp->fh_export; 597 dentry = fhp->fh_dentry; 598 599 if (S_ISREG(dentry->d_inode->i_mode)) 600 map = nfs3_regaccess; 601 else if (S_ISDIR(dentry->d_inode->i_mode)) 602 map = nfs3_diraccess; 603 else 604 map = nfs3_anyaccess; 605 606 607 query = *access; 608 for (; map->access; map++) { 609 if (map->access & query) { 610 unsigned int err2; 611 612 sresult |= map->access; 613 614 err2 = nfsd_permission(export, dentry, map->how); 615 switch (err2) { 616 case nfs_ok: 617 result |= map->access; 618 break; 619 620 /* the following error codes just mean the access was not allowed, 621 * rather than an error occurred */ 622 case nfserr_rofs: 623 case nfserr_acces: 624 case nfserr_perm: 625 /* simply don't "or" in the access bit. */ 626 break; 627 default: 628 error = err2; 629 goto out; 630 } 631 } 632 } 633 *access = result; 634 if (supported) 635 *supported = sresult; 636 637 out: 638 return error; 639 } 640 #endif /* CONFIG_NFSD_V3 */ 641 642 643 644 /* 645 * Open an existing file or directory. 646 * The access argument indicates the type of open (read/write/lock) 647 * N.B. After this call fhp needs an fh_put 648 */ 649 int 650 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 651 int access, struct file **filp) 652 { 653 struct dentry *dentry; 654 struct inode *inode; 655 int flags = O_RDONLY|O_LARGEFILE, err; 656 657 /* 658 * If we get here, then the client has already done an "open", 659 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 660 * in case a chmod has now revoked permission. 661 */ 662 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE); 663 if (err) 664 goto out; 665 666 dentry = fhp->fh_dentry; 667 inode = dentry->d_inode; 668 669 /* Disallow write access to files with the append-only bit set 670 * or any access when mandatory locking enabled 671 */ 672 err = nfserr_perm; 673 if (IS_APPEND(inode) && (access & MAY_WRITE)) 674 goto out; 675 if (IS_ISMNDLK(inode)) 676 goto out; 677 678 if (!inode->i_fop) 679 goto out; 680 681 /* 682 * Check to see if there are any leases on this file. 683 * This may block while leases are broken. 684 */ 685 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0)); 686 if (err == -EWOULDBLOCK) 687 err = -ETIMEDOUT; 688 if (err) /* NOMEM or WOULDBLOCK */ 689 goto out_nfserr; 690 691 if (access & MAY_WRITE) { 692 flags = O_WRONLY|O_LARGEFILE; 693 694 DQUOT_INIT(inode); 695 } 696 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags); 697 if (IS_ERR(*filp)) 698 err = PTR_ERR(*filp); 699 out_nfserr: 700 if (err) 701 err = nfserrno(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 struct file_operations *fop) 722 { 723 struct inode *inode = dp->d_inode; 724 int (*fsync) (struct file *, struct dentry *, int); 725 int err = nfs_ok; 726 727 filemap_fdatawrite(inode->i_mapping); 728 if (fop && (fsync = fop->fsync)) 729 err=fsync(filp, dp, 0); 730 filemap_fdatawait(inode->i_mapping); 731 732 return nfserrno(err); 733 } 734 735 736 static int 737 nfsd_sync(struct file *filp) 738 { 739 int err; 740 struct inode *inode = filp->f_dentry->d_inode; 741 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name); 742 down(&inode->i_sem); 743 err=nfsd_dosync(filp, filp->f_dentry, filp->f_op); 744 up(&inode->i_sem); 745 746 return err; 747 } 748 749 void 750 nfsd_sync_dir(struct dentry *dp) 751 { 752 nfsd_dosync(NULL, dp, dp->d_inode->i_fop); 753 } 754 755 /* 756 * Obtain the readahead parameters for the file 757 * specified by (dev, ino). 758 */ 759 static DEFINE_SPINLOCK(ra_lock); 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; 766 767 spin_lock(&ra_lock); 768 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) { 769 if (ra->p_ino == ino && ra->p_dev == dev) 770 goto found; 771 depth++; 772 if (ra->p_count == 0) 773 frap = rap; 774 } 775 depth = nfsdstats.ra_size*11/10; 776 if (!frap) { 777 spin_unlock(&ra_lock); 778 return NULL; 779 } 780 rap = frap; 781 ra = *frap; 782 ra->p_dev = dev; 783 ra->p_ino = ino; 784 ra->p_set = 0; 785 found: 786 if (rap != &raparm_cache) { 787 *rap = ra->p_next; 788 ra->p_next = raparm_cache; 789 raparm_cache = ra; 790 } 791 ra->p_count++; 792 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 793 spin_unlock(&ra_lock); 794 return ra; 795 } 796 797 /* 798 * Grab and keep cached pages assosiated with a file in the svc_rqst 799 * so that they can be passed to the netowork sendmsg/sendpage routines 800 * directrly. They will be released after the sending has completed. 801 */ 802 static int 803 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size) 804 { 805 unsigned long count = desc->count; 806 struct svc_rqst *rqstp = desc->arg.data; 807 808 if (size > count) 809 size = count; 810 811 if (rqstp->rq_res.page_len == 0) { 812 get_page(page); 813 rqstp->rq_respages[rqstp->rq_resused++] = page; 814 rqstp->rq_res.page_base = offset; 815 rqstp->rq_res.page_len = size; 816 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) { 817 get_page(page); 818 rqstp->rq_respages[rqstp->rq_resused++] = page; 819 rqstp->rq_res.page_len += size; 820 } else { 821 rqstp->rq_res.page_len += size; 822 } 823 824 desc->count = count - size; 825 desc->written += size; 826 return size; 827 } 828 829 static inline int 830 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 831 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 832 { 833 struct inode *inode; 834 struct raparms *ra; 835 mm_segment_t oldfs; 836 int err; 837 838 err = nfserr_perm; 839 inode = file->f_dentry->d_inode; 840 #ifdef MSNFS 841 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 842 (!lock_may_read(inode, offset, *count))) 843 goto out; 844 #endif 845 846 /* Get readahead parameters */ 847 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 848 849 if (ra && ra->p_set) 850 file->f_ra = ra->p_ra; 851 852 if (file->f_op->sendfile) { 853 svc_pushback_unused_pages(rqstp); 854 err = file->f_op->sendfile(file, &offset, *count, 855 nfsd_read_actor, rqstp); 856 } else { 857 oldfs = get_fs(); 858 set_fs(KERNEL_DS); 859 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 860 set_fs(oldfs); 861 } 862 863 /* Write back readahead params */ 864 if (ra) { 865 spin_lock(&ra_lock); 866 ra->p_ra = file->f_ra; 867 ra->p_set = 1; 868 ra->p_count--; 869 spin_unlock(&ra_lock); 870 } 871 872 if (err >= 0) { 873 nfsdstats.io_read += err; 874 *count = err; 875 err = 0; 876 fsnotify_access(file->f_dentry); 877 } else 878 err = nfserrno(err); 879 out: 880 return err; 881 } 882 883 static void kill_suid(struct dentry *dentry) 884 { 885 struct iattr ia; 886 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID; 887 888 down(&dentry->d_inode->i_sem); 889 notify_change(dentry, &ia); 890 up(&dentry->d_inode->i_sem); 891 } 892 893 static inline int 894 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 895 loff_t offset, struct kvec *vec, int vlen, 896 unsigned long cnt, int *stablep) 897 { 898 struct svc_export *exp; 899 struct dentry *dentry; 900 struct inode *inode; 901 mm_segment_t oldfs; 902 int err = 0; 903 int stable = *stablep; 904 905 err = nfserr_perm; 906 907 #ifdef MSNFS 908 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 909 (!lock_may_write(file->f_dentry->d_inode, offset, cnt))) 910 goto out; 911 #endif 912 913 dentry = file->f_dentry; 914 inode = dentry->d_inode; 915 exp = fhp->fh_export; 916 917 /* 918 * Request sync writes if 919 * - the sync export option has been set, or 920 * - the client requested O_SYNC behavior (NFSv3 feature). 921 * - The file system doesn't support fsync(). 922 * When gathered writes have been configured for this volume, 923 * flushing the data to disk is handled separately below. 924 */ 925 926 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */ 927 stable = 2; 928 *stablep = 2; /* FILE_SYNC */ 929 } 930 931 if (!EX_ISSYNC(exp)) 932 stable = 0; 933 if (stable && !EX_WGATHER(exp)) 934 file->f_flags |= O_SYNC; 935 936 /* Write the data. */ 937 oldfs = get_fs(); set_fs(KERNEL_DS); 938 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); 939 set_fs(oldfs); 940 if (err >= 0) { 941 nfsdstats.io_write += cnt; 942 fsnotify_modify(file->f_dentry); 943 } 944 945 /* clear setuid/setgid flag after write */ 946 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) 947 kill_suid(dentry); 948 949 if (err >= 0 && stable) { 950 static ino_t last_ino; 951 static dev_t last_dev; 952 953 /* 954 * Gathered writes: If another process is currently 955 * writing to the file, there's a high chance 956 * this is another nfsd (triggered by a bulk write 957 * from a client's biod). Rather than syncing the 958 * file with each write request, we sleep for 10 msec. 959 * 960 * I don't know if this roughly approximates 961 * C. Juszak's idea of gathered writes, but it's a 962 * nice and simple solution (IMHO), and it seems to 963 * work:-) 964 */ 965 if (EX_WGATHER(exp)) { 966 if (atomic_read(&inode->i_writecount) > 1 967 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 968 dprintk("nfsd: write defer %d\n", current->pid); 969 msleep(10); 970 dprintk("nfsd: write resume %d\n", current->pid); 971 } 972 973 if (inode->i_state & I_DIRTY) { 974 dprintk("nfsd: write sync %d\n", current->pid); 975 err=nfsd_sync(file); 976 } 977 #if 0 978 wake_up(&inode->i_wait); 979 #endif 980 } 981 last_ino = inode->i_ino; 982 last_dev = inode->i_sb->s_dev; 983 } 984 985 dprintk("nfsd: write complete err=%d\n", err); 986 if (err >= 0) 987 err = 0; 988 else 989 err = nfserrno(err); 990 out: 991 return err; 992 } 993 994 /* 995 * Read data from a file. count must contain the requested read count 996 * on entry. On return, *count contains the number of bytes actually read. 997 * N.B. After this call fhp needs an fh_put 998 */ 999 int 1000 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1001 loff_t offset, struct kvec *vec, int vlen, 1002 unsigned long *count) 1003 { 1004 int err; 1005 1006 if (file) { 1007 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry, 1008 MAY_READ|MAY_OWNER_OVERRIDE); 1009 if (err) 1010 goto out; 1011 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1012 } else { 1013 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file); 1014 if (err) 1015 goto out; 1016 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1017 nfsd_close(file); 1018 } 1019 out: 1020 return err; 1021 } 1022 1023 /* 1024 * Write data to a file. 1025 * The stable flag requests synchronous writes. 1026 * N.B. After this call fhp needs an fh_put 1027 */ 1028 int 1029 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1030 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt, 1031 int *stablep) 1032 { 1033 int err = 0; 1034 1035 if (file) { 1036 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry, 1037 MAY_WRITE|MAY_OWNER_OVERRIDE); 1038 if (err) 1039 goto out; 1040 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1041 stablep); 1042 } else { 1043 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file); 1044 if (err) 1045 goto out; 1046 1047 if (cnt) 1048 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1049 cnt, stablep); 1050 nfsd_close(file); 1051 } 1052 out: 1053 return err; 1054 } 1055 1056 #ifdef CONFIG_NFSD_V3 1057 /* 1058 * Commit all pending writes to stable storage. 1059 * Strictly speaking, we could sync just the indicated file region here, 1060 * but there's currently no way we can ask the VFS to do so. 1061 * 1062 * Unfortunately we cannot lock the file to make sure we return full WCC 1063 * data to the client, as locking happens lower down in the filesystem. 1064 */ 1065 int 1066 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1067 loff_t offset, unsigned long count) 1068 { 1069 struct file *file; 1070 int err; 1071 1072 if ((u64)count > ~(u64)offset) 1073 return nfserr_inval; 1074 1075 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0) 1076 return err; 1077 if (EX_ISSYNC(fhp->fh_export)) { 1078 if (file->f_op && file->f_op->fsync) { 1079 err = nfsd_sync(file); 1080 } else { 1081 err = nfserr_notsupp; 1082 } 1083 } 1084 1085 nfsd_close(file); 1086 return err; 1087 } 1088 #endif /* CONFIG_NFSD_V3 */ 1089 1090 /* 1091 * Create a file (regular, directory, device, fifo); UNIX sockets 1092 * not yet implemented. 1093 * If the response fh has been verified, the parent directory should 1094 * already be locked. Note that the parent directory is left locked. 1095 * 1096 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1097 */ 1098 int 1099 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1100 char *fname, int flen, struct iattr *iap, 1101 int type, dev_t rdev, struct svc_fh *resfhp) 1102 { 1103 struct dentry *dentry, *dchild = NULL; 1104 struct inode *dirp; 1105 int err; 1106 1107 err = nfserr_perm; 1108 if (!flen) 1109 goto out; 1110 err = nfserr_exist; 1111 if (isdotent(fname, flen)) 1112 goto out; 1113 1114 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); 1115 if (err) 1116 goto out; 1117 1118 dentry = fhp->fh_dentry; 1119 dirp = dentry->d_inode; 1120 1121 err = nfserr_notdir; 1122 if(!dirp->i_op || !dirp->i_op->lookup) 1123 goto out; 1124 /* 1125 * Check whether the response file handle has been verified yet. 1126 * If it has, the parent directory should already be locked. 1127 */ 1128 if (!resfhp->fh_dentry) { 1129 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1130 fh_lock(fhp); 1131 dchild = lookup_one_len(fname, dentry, flen); 1132 err = PTR_ERR(dchild); 1133 if (IS_ERR(dchild)) 1134 goto out_nfserr; 1135 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1136 if (err) 1137 goto out; 1138 } else { 1139 /* called from nfsd_proc_create */ 1140 dchild = dget(resfhp->fh_dentry); 1141 if (!fhp->fh_locked) { 1142 /* not actually possible */ 1143 printk(KERN_ERR 1144 "nfsd_create: parent %s/%s not locked!\n", 1145 dentry->d_parent->d_name.name, 1146 dentry->d_name.name); 1147 err = -EIO; 1148 goto out; 1149 } 1150 } 1151 /* 1152 * Make sure the child dentry is still negative ... 1153 */ 1154 err = nfserr_exist; 1155 if (dchild->d_inode) { 1156 dprintk("nfsd_create: dentry %s/%s not negative!\n", 1157 dentry->d_name.name, dchild->d_name.name); 1158 goto out; 1159 } 1160 1161 if (!(iap->ia_valid & ATTR_MODE)) 1162 iap->ia_mode = 0; 1163 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1164 1165 /* 1166 * Get the dir op function pointer. 1167 */ 1168 err = nfserr_perm; 1169 switch (type) { 1170 case S_IFREG: 1171 err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1172 break; 1173 case S_IFDIR: 1174 err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1175 break; 1176 case S_IFCHR: 1177 case S_IFBLK: 1178 case S_IFIFO: 1179 case S_IFSOCK: 1180 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1181 break; 1182 default: 1183 printk("nfsd: bad file type %o in nfsd_create\n", type); 1184 err = -EINVAL; 1185 } 1186 if (err < 0) 1187 goto out_nfserr; 1188 1189 if (EX_ISSYNC(fhp->fh_export)) { 1190 nfsd_sync_dir(dentry); 1191 write_inode_now(dchild->d_inode, 1); 1192 } 1193 1194 1195 /* Set file attributes. Mode has already been set and 1196 * setting uid/gid works only for root. Irix appears to 1197 * send along the gid when it tries to implement setgid 1198 * directories via NFS. 1199 */ 1200 err = 0; 1201 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) 1202 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1203 /* 1204 * Update the file handle to get the new inode info. 1205 */ 1206 if (!err) 1207 err = fh_update(resfhp); 1208 out: 1209 if (dchild && !IS_ERR(dchild)) 1210 dput(dchild); 1211 return err; 1212 1213 out_nfserr: 1214 err = nfserrno(err); 1215 goto out; 1216 } 1217 1218 #ifdef CONFIG_NFSD_V3 1219 /* 1220 * NFSv3 version of nfsd_create 1221 */ 1222 int 1223 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp, 1224 char *fname, int flen, struct iattr *iap, 1225 struct svc_fh *resfhp, int createmode, u32 *verifier, 1226 int *truncp) 1227 { 1228 struct dentry *dentry, *dchild = NULL; 1229 struct inode *dirp; 1230 int err; 1231 __u32 v_mtime=0, v_atime=0; 1232 int v_mode=0; 1233 1234 err = nfserr_perm; 1235 if (!flen) 1236 goto out; 1237 err = nfserr_exist; 1238 if (isdotent(fname, flen)) 1239 goto out; 1240 if (!(iap->ia_valid & ATTR_MODE)) 1241 iap->ia_mode = 0; 1242 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); 1243 if (err) 1244 goto out; 1245 1246 dentry = fhp->fh_dentry; 1247 dirp = dentry->d_inode; 1248 1249 /* Get all the sanity checks out of the way before 1250 * we lock the parent. */ 1251 err = nfserr_notdir; 1252 if(!dirp->i_op || !dirp->i_op->lookup) 1253 goto out; 1254 fh_lock(fhp); 1255 1256 /* 1257 * Compose the response file handle. 1258 */ 1259 dchild = lookup_one_len(fname, dentry, flen); 1260 err = PTR_ERR(dchild); 1261 if (IS_ERR(dchild)) 1262 goto out_nfserr; 1263 1264 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1265 if (err) 1266 goto out; 1267 1268 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1269 /* while the verifier would fit in mtime+atime, 1270 * solaris7 gets confused (bugid 4218508) if these have 1271 * the high bit set, so we use the mode as well 1272 */ 1273 v_mtime = verifier[0]&0x7fffffff; 1274 v_atime = verifier[1]&0x7fffffff; 1275 v_mode = S_IFREG 1276 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */ 1277 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */ 1278 ; 1279 } 1280 1281 if (dchild->d_inode) { 1282 err = 0; 1283 1284 switch (createmode) { 1285 case NFS3_CREATE_UNCHECKED: 1286 if (! S_ISREG(dchild->d_inode->i_mode)) 1287 err = nfserr_exist; 1288 else if (truncp) { 1289 /* in nfsv4, we need to treat this case a little 1290 * differently. we don't want to truncate the 1291 * file now; this would be wrong if the OPEN 1292 * fails for some other reason. furthermore, 1293 * if the size is nonzero, we should ignore it 1294 * according to spec! 1295 */ 1296 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1297 } 1298 else { 1299 iap->ia_valid &= ATTR_SIZE; 1300 goto set_attr; 1301 } 1302 break; 1303 case NFS3_CREATE_EXCLUSIVE: 1304 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1305 && dchild->d_inode->i_atime.tv_sec == v_atime 1306 && dchild->d_inode->i_mode == v_mode 1307 && dchild->d_inode->i_size == 0 ) 1308 break; 1309 /* fallthru */ 1310 case NFS3_CREATE_GUARDED: 1311 err = nfserr_exist; 1312 } 1313 goto out; 1314 } 1315 1316 err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1317 if (err < 0) 1318 goto out_nfserr; 1319 1320 if (EX_ISSYNC(fhp->fh_export)) { 1321 nfsd_sync_dir(dentry); 1322 /* setattr will sync the child (or not) */ 1323 } 1324 1325 /* 1326 * Update the filehandle to get the new inode info. 1327 */ 1328 err = fh_update(resfhp); 1329 if (err) 1330 goto out; 1331 1332 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1333 /* Cram the verifier into atime/mtime/mode */ 1334 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1335 | ATTR_MTIME_SET|ATTR_ATIME_SET 1336 | ATTR_MODE; 1337 /* XXX someone who knows this better please fix it for nsec */ 1338 iap->ia_mtime.tv_sec = v_mtime; 1339 iap->ia_atime.tv_sec = v_atime; 1340 iap->ia_mtime.tv_nsec = 0; 1341 iap->ia_atime.tv_nsec = 0; 1342 iap->ia_mode = v_mode; 1343 } 1344 1345 /* Set file attributes. 1346 * Mode has already been set but we might need to reset it 1347 * for CREATE_EXCLUSIVE 1348 * Irix appears to send along the gid when it tries to 1349 * implement setgid directories via NFS. Clear out all that cruft. 1350 */ 1351 set_attr: 1352 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) 1353 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1354 1355 out: 1356 fh_unlock(fhp); 1357 if (dchild && !IS_ERR(dchild)) 1358 dput(dchild); 1359 return err; 1360 1361 out_nfserr: 1362 err = nfserrno(err); 1363 goto out; 1364 } 1365 #endif /* CONFIG_NFSD_V3 */ 1366 1367 /* 1368 * Read a symlink. On entry, *lenp must contain the maximum path length that 1369 * fits into the buffer. On return, it contains the true length. 1370 * N.B. After this call fhp needs an fh_put 1371 */ 1372 int 1373 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1374 { 1375 struct dentry *dentry; 1376 struct inode *inode; 1377 mm_segment_t oldfs; 1378 int err; 1379 1380 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP); 1381 if (err) 1382 goto out; 1383 1384 dentry = fhp->fh_dentry; 1385 inode = dentry->d_inode; 1386 1387 err = nfserr_inval; 1388 if (!inode->i_op || !inode->i_op->readlink) 1389 goto out; 1390 1391 touch_atime(fhp->fh_export->ex_mnt, dentry); 1392 /* N.B. Why does this call need a get_fs()?? 1393 * Remove the set_fs and watch the fireworks:-) --okir 1394 */ 1395 1396 oldfs = get_fs(); set_fs(KERNEL_DS); 1397 err = inode->i_op->readlink(dentry, buf, *lenp); 1398 set_fs(oldfs); 1399 1400 if (err < 0) 1401 goto out_nfserr; 1402 *lenp = err; 1403 err = 0; 1404 out: 1405 return err; 1406 1407 out_nfserr: 1408 err = nfserrno(err); 1409 goto out; 1410 } 1411 1412 /* 1413 * Create a symlink and look up its inode 1414 * N.B. After this call _both_ fhp and resfhp need an fh_put 1415 */ 1416 int 1417 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1418 char *fname, int flen, 1419 char *path, int plen, 1420 struct svc_fh *resfhp, 1421 struct iattr *iap) 1422 { 1423 struct dentry *dentry, *dnew; 1424 int err, cerr; 1425 umode_t mode; 1426 1427 err = nfserr_noent; 1428 if (!flen || !plen) 1429 goto out; 1430 err = nfserr_exist; 1431 if (isdotent(fname, flen)) 1432 goto out; 1433 1434 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE); 1435 if (err) 1436 goto out; 1437 fh_lock(fhp); 1438 dentry = fhp->fh_dentry; 1439 dnew = lookup_one_len(fname, dentry, flen); 1440 err = PTR_ERR(dnew); 1441 if (IS_ERR(dnew)) 1442 goto out_nfserr; 1443 1444 mode = S_IALLUGO; 1445 /* Only the MODE ATTRibute is even vaguely meaningful */ 1446 if (iap && (iap->ia_valid & ATTR_MODE)) 1447 mode = iap->ia_mode & S_IALLUGO; 1448 1449 if (unlikely(path[plen] != 0)) { 1450 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1451 if (path_alloced == NULL) 1452 err = -ENOMEM; 1453 else { 1454 strncpy(path_alloced, path, plen); 1455 path_alloced[plen] = 0; 1456 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode); 1457 kfree(path_alloced); 1458 } 1459 } else 1460 err = vfs_symlink(dentry->d_inode, dnew, path, mode); 1461 1462 if (!err) { 1463 if (EX_ISSYNC(fhp->fh_export)) 1464 nfsd_sync_dir(dentry); 1465 } else 1466 err = nfserrno(err); 1467 fh_unlock(fhp); 1468 1469 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1470 dput(dnew); 1471 if (err==0) err = cerr; 1472 out: 1473 return err; 1474 1475 out_nfserr: 1476 err = nfserrno(err); 1477 goto out; 1478 } 1479 1480 /* 1481 * Create a hardlink 1482 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1483 */ 1484 int 1485 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1486 char *name, int len, struct svc_fh *tfhp) 1487 { 1488 struct dentry *ddir, *dnew, *dold; 1489 struct inode *dirp, *dest; 1490 int err; 1491 1492 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE); 1493 if (err) 1494 goto out; 1495 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP); 1496 if (err) 1497 goto out; 1498 1499 err = nfserr_perm; 1500 if (!len) 1501 goto out; 1502 err = nfserr_exist; 1503 if (isdotent(name, len)) 1504 goto out; 1505 1506 fh_lock(ffhp); 1507 ddir = ffhp->fh_dentry; 1508 dirp = ddir->d_inode; 1509 1510 dnew = lookup_one_len(name, ddir, len); 1511 err = PTR_ERR(dnew); 1512 if (IS_ERR(dnew)) 1513 goto out_nfserr; 1514 1515 dold = tfhp->fh_dentry; 1516 dest = dold->d_inode; 1517 1518 err = vfs_link(dold, dirp, dnew); 1519 if (!err) { 1520 if (EX_ISSYNC(ffhp->fh_export)) { 1521 nfsd_sync_dir(ddir); 1522 write_inode_now(dest, 1); 1523 } 1524 } else { 1525 if (err == -EXDEV && rqstp->rq_vers == 2) 1526 err = nfserr_acces; 1527 else 1528 err = nfserrno(err); 1529 } 1530 1531 fh_unlock(ffhp); 1532 dput(dnew); 1533 out: 1534 return err; 1535 1536 out_nfserr: 1537 err = nfserrno(err); 1538 goto out; 1539 } 1540 1541 /* 1542 * Rename a file 1543 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1544 */ 1545 int 1546 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1547 struct svc_fh *tfhp, char *tname, int tlen) 1548 { 1549 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1550 struct inode *fdir, *tdir; 1551 int err; 1552 1553 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE); 1554 if (err) 1555 goto out; 1556 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE); 1557 if (err) 1558 goto out; 1559 1560 fdentry = ffhp->fh_dentry; 1561 fdir = fdentry->d_inode; 1562 1563 tdentry = tfhp->fh_dentry; 1564 tdir = tdentry->d_inode; 1565 1566 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; 1567 if (fdir->i_sb != tdir->i_sb) 1568 goto out; 1569 1570 err = nfserr_perm; 1571 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1572 goto out; 1573 1574 /* cannot use fh_lock as we need deadlock protective ordering 1575 * so do it by hand */ 1576 trap = lock_rename(tdentry, fdentry); 1577 ffhp->fh_locked = tfhp->fh_locked = 1; 1578 fill_pre_wcc(ffhp); 1579 fill_pre_wcc(tfhp); 1580 1581 odentry = lookup_one_len(fname, fdentry, flen); 1582 err = PTR_ERR(odentry); 1583 if (IS_ERR(odentry)) 1584 goto out_nfserr; 1585 1586 err = -ENOENT; 1587 if (!odentry->d_inode) 1588 goto out_dput_old; 1589 err = -EINVAL; 1590 if (odentry == trap) 1591 goto out_dput_old; 1592 1593 ndentry = lookup_one_len(tname, tdentry, tlen); 1594 err = PTR_ERR(ndentry); 1595 if (IS_ERR(ndentry)) 1596 goto out_dput_old; 1597 err = -ENOTEMPTY; 1598 if (ndentry == trap) 1599 goto out_dput_new; 1600 1601 #ifdef MSNFS 1602 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1603 ((atomic_read(&odentry->d_count) > 1) 1604 || (atomic_read(&ndentry->d_count) > 1))) { 1605 err = nfserr_perm; 1606 } else 1607 #endif 1608 err = vfs_rename(fdir, odentry, tdir, ndentry); 1609 if (!err && EX_ISSYNC(tfhp->fh_export)) { 1610 nfsd_sync_dir(tdentry); 1611 nfsd_sync_dir(fdentry); 1612 } 1613 1614 out_dput_new: 1615 dput(ndentry); 1616 out_dput_old: 1617 dput(odentry); 1618 out_nfserr: 1619 if (err) 1620 err = nfserrno(err); 1621 1622 /* we cannot reply on fh_unlock on the two filehandles, 1623 * as that would do the wrong thing if the two directories 1624 * were the same, so again we do it by hand 1625 */ 1626 fill_post_wcc(ffhp); 1627 fill_post_wcc(tfhp); 1628 unlock_rename(tdentry, fdentry); 1629 ffhp->fh_locked = tfhp->fh_locked = 0; 1630 1631 out: 1632 return err; 1633 } 1634 1635 /* 1636 * Unlink a file or directory 1637 * N.B. After this call fhp needs an fh_put 1638 */ 1639 int 1640 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1641 char *fname, int flen) 1642 { 1643 struct dentry *dentry, *rdentry; 1644 struct inode *dirp; 1645 int err; 1646 1647 err = nfserr_acces; 1648 if (!flen || isdotent(fname, flen)) 1649 goto out; 1650 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE); 1651 if (err) 1652 goto out; 1653 1654 fh_lock(fhp); 1655 dentry = fhp->fh_dentry; 1656 dirp = dentry->d_inode; 1657 1658 rdentry = lookup_one_len(fname, dentry, flen); 1659 err = PTR_ERR(rdentry); 1660 if (IS_ERR(rdentry)) 1661 goto out_nfserr; 1662 1663 if (!rdentry->d_inode) { 1664 dput(rdentry); 1665 err = nfserr_noent; 1666 goto out; 1667 } 1668 1669 if (!type) 1670 type = rdentry->d_inode->i_mode & S_IFMT; 1671 1672 if (type != S_IFDIR) { /* It's UNLINK */ 1673 #ifdef MSNFS 1674 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1675 (atomic_read(&rdentry->d_count) > 1)) { 1676 err = nfserr_perm; 1677 } else 1678 #endif 1679 err = vfs_unlink(dirp, rdentry); 1680 } else { /* It's RMDIR */ 1681 err = vfs_rmdir(dirp, rdentry); 1682 } 1683 1684 dput(rdentry); 1685 1686 if (err) 1687 goto out_nfserr; 1688 if (EX_ISSYNC(fhp->fh_export)) 1689 nfsd_sync_dir(dentry); 1690 1691 out: 1692 return err; 1693 1694 out_nfserr: 1695 err = nfserrno(err); 1696 goto out; 1697 } 1698 1699 /* 1700 * Read entries from a directory. 1701 * The NFSv3/4 verifier we ignore for now. 1702 */ 1703 int 1704 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 1705 struct readdir_cd *cdp, encode_dent_fn func) 1706 { 1707 int err; 1708 struct file *file; 1709 loff_t offset = *offsetp; 1710 1711 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file); 1712 if (err) 1713 goto out; 1714 1715 offset = vfs_llseek(file, offset, 0); 1716 if (offset < 0) { 1717 err = nfserrno((int)offset); 1718 goto out_close; 1719 } 1720 1721 /* 1722 * Read the directory entries. This silly loop is necessary because 1723 * readdir() is not guaranteed to fill up the entire buffer, but 1724 * may choose to do less. 1725 */ 1726 1727 do { 1728 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1729 err = vfs_readdir(file, (filldir_t) func, cdp); 1730 } while (err >=0 && cdp->err == nfs_ok); 1731 if (err) 1732 err = nfserrno(err); 1733 else 1734 err = cdp->err; 1735 *offsetp = vfs_llseek(file, 0, 1); 1736 1737 if (err == nfserr_eof || err == nfserr_toosmall) 1738 err = nfs_ok; /* can still be found in ->err */ 1739 out_close: 1740 nfsd_close(file); 1741 out: 1742 return err; 1743 } 1744 1745 /* 1746 * Get file system stats 1747 * N.B. After this call fhp needs an fh_put 1748 */ 1749 int 1750 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat) 1751 { 1752 int err = fh_verify(rqstp, fhp, 0, MAY_NOP); 1753 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat)) 1754 err = nfserr_io; 1755 return err; 1756 } 1757 1758 /* 1759 * Check for a user's access permissions to this inode. 1760 */ 1761 int 1762 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc) 1763 { 1764 struct inode *inode = dentry->d_inode; 1765 int err; 1766 1767 if (acc == MAY_NOP) 1768 return 0; 1769 #if 0 1770 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 1771 acc, 1772 (acc & MAY_READ)? " read" : "", 1773 (acc & MAY_WRITE)? " write" : "", 1774 (acc & MAY_EXEC)? " exec" : "", 1775 (acc & MAY_SATTR)? " sattr" : "", 1776 (acc & MAY_TRUNC)? " trunc" : "", 1777 (acc & MAY_LOCK)? " lock" : "", 1778 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "", 1779 inode->i_mode, 1780 IS_IMMUTABLE(inode)? " immut" : "", 1781 IS_APPEND(inode)? " append" : "", 1782 IS_RDONLY(inode)? " ro" : ""); 1783 dprintk(" owner %d/%d user %d/%d\n", 1784 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid); 1785 #endif 1786 1787 /* Normally we reject any write/sattr etc access on a read-only file 1788 * system. But if it is IRIX doing check on write-access for a 1789 * device special file, we ignore rofs. 1790 */ 1791 if (!(acc & MAY_LOCAL_ACCESS)) 1792 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) { 1793 if (EX_RDONLY(exp) || IS_RDONLY(inode)) 1794 return nfserr_rofs; 1795 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode)) 1796 return nfserr_perm; 1797 } 1798 if ((acc & MAY_TRUNC) && IS_APPEND(inode)) 1799 return nfserr_perm; 1800 1801 if (acc & MAY_LOCK) { 1802 /* If we cannot rely on authentication in NLM requests, 1803 * just allow locks, otherwise require read permission, or 1804 * ownership 1805 */ 1806 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 1807 return 0; 1808 else 1809 acc = MAY_READ | MAY_OWNER_OVERRIDE; 1810 } 1811 /* 1812 * The file owner always gets access permission for accesses that 1813 * would normally be checked at open time. This is to make 1814 * file access work even when the client has done a fchmod(fd, 0). 1815 * 1816 * However, `cp foo bar' should fail nevertheless when bar is 1817 * readonly. A sensible way to do this might be to reject all 1818 * attempts to truncate a read-only file, because a creat() call 1819 * always implies file truncation. 1820 * ... but this isn't really fair. A process may reasonably call 1821 * ftruncate on an open file descriptor on a file with perm 000. 1822 * We must trust the client to do permission checking - using "ACCESS" 1823 * with NFSv3. 1824 */ 1825 if ((acc & MAY_OWNER_OVERRIDE) && 1826 inode->i_uid == current->fsuid) 1827 return 0; 1828 1829 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL); 1830 1831 /* Allow read access to binaries even when mode 111 */ 1832 if (err == -EACCES && S_ISREG(inode->i_mode) && 1833 acc == (MAY_READ | MAY_OWNER_OVERRIDE)) 1834 err = permission(inode, MAY_EXEC, NULL); 1835 1836 return err? nfserrno(err) : 0; 1837 } 1838 1839 void 1840 nfsd_racache_shutdown(void) 1841 { 1842 if (!raparm_cache) 1843 return; 1844 dprintk("nfsd: freeing readahead buffers.\n"); 1845 kfree(raparml); 1846 raparm_cache = raparml = NULL; 1847 } 1848 /* 1849 * Initialize readahead param cache 1850 */ 1851 int 1852 nfsd_racache_init(int cache_size) 1853 { 1854 int i; 1855 1856 if (raparm_cache) 1857 return 0; 1858 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL); 1859 1860 if (raparml != NULL) { 1861 dprintk("nfsd: allocating %d readahead buffers.\n", 1862 cache_size); 1863 memset(raparml, 0, sizeof(struct raparms) * cache_size); 1864 for (i = 0; i < cache_size - 1; i++) { 1865 raparml[i].p_next = raparml + i + 1; 1866 } 1867 raparm_cache = raparml; 1868 } else { 1869 printk(KERN_WARNING 1870 "nfsd: Could not allocate memory read-ahead cache.\n"); 1871 return -ENOMEM; 1872 } 1873 nfsdstats.ra_size = cache_size; 1874 return 0; 1875 } 1876 1877 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 1878 struct posix_acl * 1879 nfsd_get_posix_acl(struct svc_fh *fhp, int type) 1880 { 1881 struct inode *inode = fhp->fh_dentry->d_inode; 1882 char *name; 1883 void *value = NULL; 1884 ssize_t size; 1885 struct posix_acl *acl; 1886 1887 if (!IS_POSIXACL(inode) || !inode->i_op || !inode->i_op->getxattr) 1888 return ERR_PTR(-EOPNOTSUPP); 1889 switch(type) { 1890 case ACL_TYPE_ACCESS: 1891 name = POSIX_ACL_XATTR_ACCESS; 1892 break; 1893 case ACL_TYPE_DEFAULT: 1894 name = POSIX_ACL_XATTR_DEFAULT; 1895 break; 1896 default: 1897 return ERR_PTR(-EOPNOTSUPP); 1898 } 1899 1900 size = inode->i_op->getxattr(fhp->fh_dentry, name, NULL, 0); 1901 1902 if (size < 0) { 1903 acl = ERR_PTR(size); 1904 goto getout; 1905 } else if (size > 0) { 1906 value = kmalloc(size, GFP_KERNEL); 1907 if (!value) { 1908 acl = ERR_PTR(-ENOMEM); 1909 goto getout; 1910 } 1911 size = inode->i_op->getxattr(fhp->fh_dentry, name, value, size); 1912 if (size < 0) { 1913 acl = ERR_PTR(size); 1914 goto getout; 1915 } 1916 } 1917 acl = posix_acl_from_xattr(value, size); 1918 1919 getout: 1920 kfree(value); 1921 return acl; 1922 } 1923 1924 int 1925 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) 1926 { 1927 struct inode *inode = fhp->fh_dentry->d_inode; 1928 char *name; 1929 void *value = NULL; 1930 size_t size; 1931 int error; 1932 1933 if (!IS_POSIXACL(inode) || !inode->i_op || 1934 !inode->i_op->setxattr || !inode->i_op->removexattr) 1935 return -EOPNOTSUPP; 1936 switch(type) { 1937 case ACL_TYPE_ACCESS: 1938 name = POSIX_ACL_XATTR_ACCESS; 1939 break; 1940 case ACL_TYPE_DEFAULT: 1941 name = POSIX_ACL_XATTR_DEFAULT; 1942 break; 1943 default: 1944 return -EOPNOTSUPP; 1945 } 1946 1947 if (acl && acl->a_count) { 1948 size = posix_acl_xattr_size(acl->a_count); 1949 value = kmalloc(size, GFP_KERNEL); 1950 if (!value) 1951 return -ENOMEM; 1952 size = posix_acl_to_xattr(acl, value, size); 1953 if (size < 0) { 1954 error = size; 1955 goto getout; 1956 } 1957 } else 1958 size = 0; 1959 1960 if (!fhp->fh_locked) 1961 fh_lock(fhp); /* unlocking is done automatically */ 1962 if (size) 1963 error = inode->i_op->setxattr(fhp->fh_dentry, name, 1964 value, size, 0); 1965 else { 1966 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) 1967 error = 0; 1968 else { 1969 error = inode->i_op->removexattr(fhp->fh_dentry, name); 1970 if (error == -ENODATA) 1971 error = 0; 1972 } 1973 } 1974 1975 getout: 1976 kfree(value); 1977 return error; 1978 } 1979 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 1980