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