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