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 #include <linux/ima.h> 59 #include "vfs.h" 60 61 #include <asm/uaccess.h> 62 63 #define NFSDDBG_FACILITY NFSDDBG_FILEOP 64 65 66 /* 67 * This is a cache of readahead params that help us choose the proper 68 * readahead strategy. Initially, we set all readahead parameters to 0 69 * and let the VFS handle things. 70 * If you increase the number of cached files very much, you'll need to 71 * add a hash table here. 72 */ 73 struct raparms { 74 struct raparms *p_next; 75 unsigned int p_count; 76 ino_t p_ino; 77 dev_t p_dev; 78 int p_set; 79 struct file_ra_state p_ra; 80 unsigned int p_hindex; 81 }; 82 83 struct raparm_hbucket { 84 struct raparms *pb_head; 85 spinlock_t pb_lock; 86 } ____cacheline_aligned_in_smp; 87 88 #define RAPARM_HASH_BITS 4 89 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) 90 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) 91 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; 92 93 static inline int 94 nfsd_v4client(struct svc_rqst *rq) 95 { 96 return rq->rq_prog == NFS_PROGRAM && rq->rq_vers == 4; 97 } 98 99 /* 100 * Called from nfsd_lookup and encode_dirent. Check if we have crossed 101 * a mount point. 102 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged, 103 * or nfs_ok having possibly changed *dpp and *expp 104 */ 105 int 106 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 107 struct svc_export **expp) 108 { 109 struct svc_export *exp = *expp, *exp2 = NULL; 110 struct dentry *dentry = *dpp; 111 struct path path = {.mnt = mntget(exp->ex_path.mnt), 112 .dentry = dget(dentry)}; 113 int err = 0; 114 115 while (d_mountpoint(path.dentry) && follow_down(&path)) 116 ; 117 118 exp2 = rqst_exp_get_by_name(rqstp, &path); 119 if (IS_ERR(exp2)) { 120 if (PTR_ERR(exp2) != -ENOENT) 121 err = PTR_ERR(exp2); 122 path_put(&path); 123 goto out; 124 } 125 if (nfsd_v4client(rqstp) || 126 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) { 127 /* successfully crossed mount point */ 128 /* 129 * This is subtle: path.dentry is *not* on path.mnt 130 * at this point. The only reason we are safe is that 131 * original mnt is pinned down by exp, so we should 132 * put path *before* putting exp 133 */ 134 *dpp = path.dentry; 135 path.dentry = dentry; 136 *expp = exp2; 137 exp2 = exp; 138 } 139 path_put(&path); 140 exp_put(exp2); 141 out: 142 return err; 143 } 144 145 static void follow_to_parent(struct path *path) 146 { 147 struct dentry *dp; 148 149 while (path->dentry == path->mnt->mnt_root && follow_up(path)) 150 ; 151 dp = dget_parent(path->dentry); 152 dput(path->dentry); 153 path->dentry = dp; 154 } 155 156 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp) 157 { 158 struct svc_export *exp2; 159 struct path path = {.mnt = mntget((*exp)->ex_path.mnt), 160 .dentry = dget(dparent)}; 161 162 follow_to_parent(&path); 163 164 exp2 = rqst_exp_parent(rqstp, &path); 165 if (PTR_ERR(exp2) == -ENOENT) { 166 *dentryp = dget(dparent); 167 } else if (IS_ERR(exp2)) { 168 path_put(&path); 169 return PTR_ERR(exp2); 170 } else { 171 *dentryp = dget(path.dentry); 172 exp_put(*exp); 173 *exp = exp2; 174 } 175 path_put(&path); 176 return 0; 177 } 178 179 __be32 180 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp, 181 const char *name, unsigned int len, 182 struct svc_export **exp_ret, struct dentry **dentry_ret) 183 { 184 struct svc_export *exp; 185 struct dentry *dparent; 186 struct dentry *dentry; 187 __be32 err; 188 int host_err; 189 190 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); 191 192 /* Obtain dentry and export. */ 193 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 194 if (err) 195 return err; 196 197 dparent = fhp->fh_dentry; 198 exp = fhp->fh_export; 199 exp_get(exp); 200 201 /* Lookup the name, but don't follow links */ 202 if (isdotent(name, len)) { 203 if (len==1) 204 dentry = dget(dparent); 205 else if (dparent != exp->ex_path.dentry) 206 dentry = dget_parent(dparent); 207 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp)) 208 dentry = dget(dparent); /* .. == . just like at / */ 209 else { 210 /* checking mountpoint crossing is very different when stepping up */ 211 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry); 212 if (host_err) 213 goto out_nfserr; 214 } 215 } else { 216 fh_lock(fhp); 217 dentry = lookup_one_len(name, dparent, len); 218 host_err = PTR_ERR(dentry); 219 if (IS_ERR(dentry)) 220 goto out_nfserr; 221 /* 222 * check if we have crossed a mount point ... 223 */ 224 if (d_mountpoint(dentry)) { 225 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { 226 dput(dentry); 227 goto out_nfserr; 228 } 229 } 230 } 231 *dentry_ret = dentry; 232 *exp_ret = exp; 233 return 0; 234 235 out_nfserr: 236 exp_put(exp); 237 return nfserrno(host_err); 238 } 239 240 /* 241 * Look up one component of a pathname. 242 * N.B. After this call _both_ fhp and resfh need an fh_put 243 * 244 * If the lookup would cross a mountpoint, and the mounted filesystem 245 * is exported to the client with NFSEXP_NOHIDE, then the lookup is 246 * accepted as it stands and the mounted directory is 247 * returned. Otherwise the covered directory is returned. 248 * NOTE: this mountpoint crossing is not supported properly by all 249 * clients and is explicitly disallowed for NFSv3 250 * NeilBrown <neilb@cse.unsw.edu.au> 251 */ 252 __be32 253 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, 254 unsigned int len, struct svc_fh *resfh) 255 { 256 struct svc_export *exp; 257 struct dentry *dentry; 258 __be32 err; 259 260 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry); 261 if (err) 262 return err; 263 err = check_nfsd_access(exp, rqstp); 264 if (err) 265 goto out; 266 /* 267 * Note: we compose the file handle now, but as the 268 * dentry may be negative, it may need to be updated. 269 */ 270 err = fh_compose(resfh, exp, dentry, fhp); 271 if (!err && !dentry->d_inode) 272 err = nfserr_noent; 273 out: 274 dput(dentry); 275 exp_put(exp); 276 return err; 277 } 278 279 280 /* 281 * Set various file attributes. 282 * N.B. After this call fhp needs an fh_put 283 */ 284 __be32 285 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, 286 int check_guard, time_t guardtime) 287 { 288 struct dentry *dentry; 289 struct inode *inode; 290 int accmode = NFSD_MAY_SATTR; 291 int ftype = 0; 292 __be32 err; 293 int host_err; 294 int size_change = 0; 295 296 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) 297 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE; 298 if (iap->ia_valid & ATTR_SIZE) 299 ftype = S_IFREG; 300 301 /* Get inode */ 302 err = fh_verify(rqstp, fhp, ftype, accmode); 303 if (err) 304 goto out; 305 306 dentry = fhp->fh_dentry; 307 inode = dentry->d_inode; 308 309 /* Ignore any mode updates on symlinks */ 310 if (S_ISLNK(inode->i_mode)) 311 iap->ia_valid &= ~ATTR_MODE; 312 313 if (!iap->ia_valid) 314 goto out; 315 316 /* 317 * NFSv2 does not differentiate between "set-[ac]time-to-now" 318 * which only requires access, and "set-[ac]time-to-X" which 319 * requires ownership. 320 * So if it looks like it might be "set both to the same time which 321 * is close to now", and if inode_change_ok fails, then we 322 * convert to "set to now" instead of "set to explicit time" 323 * 324 * We only call inode_change_ok as the last test as technically 325 * it is not an interface that we should be using. It is only 326 * valid if the filesystem does not define it's own i_op->setattr. 327 */ 328 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) 329 #define MAX_TOUCH_TIME_ERROR (30*60) 330 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET && 331 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) { 332 /* 333 * Looks probable. 334 * 335 * Now just make sure time is in the right ballpark. 336 * Solaris, at least, doesn't seem to care what the time 337 * request is. We require it be within 30 minutes of now. 338 */ 339 time_t delta = iap->ia_atime.tv_sec - get_seconds(); 340 if (delta < 0) 341 delta = -delta; 342 if (delta < MAX_TOUCH_TIME_ERROR && 343 inode_change_ok(inode, iap) != 0) { 344 /* 345 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME. 346 * This will cause notify_change to set these times 347 * to "now" 348 */ 349 iap->ia_valid &= ~BOTH_TIME_SET; 350 } 351 } 352 353 /* 354 * The size case is special. 355 * It changes the file as well as the attributes. 356 */ 357 if (iap->ia_valid & ATTR_SIZE) { 358 if (iap->ia_size < inode->i_size) { 359 err = nfsd_permission(rqstp, fhp->fh_export, dentry, 360 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE); 361 if (err) 362 goto out; 363 } 364 365 /* 366 * If we are changing the size of the file, then 367 * we need to break all leases. 368 */ 369 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK); 370 if (host_err == -EWOULDBLOCK) 371 host_err = -ETIMEDOUT; 372 if (host_err) /* ENOMEM or EWOULDBLOCK */ 373 goto out_nfserr; 374 375 host_err = get_write_access(inode); 376 if (host_err) 377 goto out_nfserr; 378 379 size_change = 1; 380 host_err = locks_verify_truncate(inode, NULL, iap->ia_size); 381 if (host_err) { 382 put_write_access(inode); 383 goto out_nfserr; 384 } 385 vfs_dq_init(inode); 386 } 387 388 /* sanitize the mode change */ 389 if (iap->ia_valid & ATTR_MODE) { 390 iap->ia_mode &= S_IALLUGO; 391 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO); 392 } 393 394 /* Revoke setuid/setgid on chown */ 395 if (!S_ISDIR(inode->i_mode) && 396 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) || 397 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) { 398 iap->ia_valid |= ATTR_KILL_PRIV; 399 if (iap->ia_valid & ATTR_MODE) { 400 /* we're setting mode too, just clear the s*id bits */ 401 iap->ia_mode &= ~S_ISUID; 402 if (iap->ia_mode & S_IXGRP) 403 iap->ia_mode &= ~S_ISGID; 404 } else { 405 /* set ATTR_KILL_* bits and let VFS handle it */ 406 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID); 407 } 408 } 409 410 /* Change the attributes. */ 411 412 iap->ia_valid |= ATTR_CTIME; 413 414 err = nfserr_notsync; 415 if (!check_guard || guardtime == inode->i_ctime.tv_sec) { 416 fh_lock(fhp); 417 host_err = notify_change(dentry, iap); 418 err = nfserrno(host_err); 419 fh_unlock(fhp); 420 } 421 if (size_change) 422 put_write_access(inode); 423 if (!err) 424 if (EX_ISSYNC(fhp->fh_export)) 425 write_inode_now(inode, 1); 426 out: 427 return err; 428 429 out_nfserr: 430 err = nfserrno(host_err); 431 goto out; 432 } 433 434 #if defined(CONFIG_NFSD_V2_ACL) || \ 435 defined(CONFIG_NFSD_V3_ACL) || \ 436 defined(CONFIG_NFSD_V4) 437 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf) 438 { 439 ssize_t buflen; 440 ssize_t ret; 441 442 buflen = vfs_getxattr(dentry, key, NULL, 0); 443 if (buflen <= 0) 444 return buflen; 445 446 *buf = kmalloc(buflen, GFP_KERNEL); 447 if (!*buf) 448 return -ENOMEM; 449 450 ret = vfs_getxattr(dentry, key, *buf, buflen); 451 if (ret < 0) 452 kfree(*buf); 453 return ret; 454 } 455 #endif 456 457 #if defined(CONFIG_NFSD_V4) 458 static int 459 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key) 460 { 461 int len; 462 size_t buflen; 463 char *buf = NULL; 464 int error = 0; 465 466 buflen = posix_acl_xattr_size(pacl->a_count); 467 buf = kmalloc(buflen, GFP_KERNEL); 468 error = -ENOMEM; 469 if (buf == NULL) 470 goto out; 471 472 len = posix_acl_to_xattr(pacl, buf, buflen); 473 if (len < 0) { 474 error = len; 475 goto out; 476 } 477 478 error = vfs_setxattr(dentry, key, buf, len, 0); 479 out: 480 kfree(buf); 481 return error; 482 } 483 484 __be32 485 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp, 486 struct nfs4_acl *acl) 487 { 488 __be32 error; 489 int host_error; 490 struct dentry *dentry; 491 struct inode *inode; 492 struct posix_acl *pacl = NULL, *dpacl = NULL; 493 unsigned int flags = 0; 494 495 /* Get inode */ 496 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR); 497 if (error) 498 return error; 499 500 dentry = fhp->fh_dentry; 501 inode = dentry->d_inode; 502 if (S_ISDIR(inode->i_mode)) 503 flags = NFS4_ACL_DIR; 504 505 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags); 506 if (host_error == -EINVAL) { 507 return nfserr_attrnotsupp; 508 } else if (host_error < 0) 509 goto out_nfserr; 510 511 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS); 512 if (host_error < 0) 513 goto out_release; 514 515 if (S_ISDIR(inode->i_mode)) 516 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT); 517 518 out_release: 519 posix_acl_release(pacl); 520 posix_acl_release(dpacl); 521 out_nfserr: 522 if (host_error == -EOPNOTSUPP) 523 return nfserr_attrnotsupp; 524 else 525 return nfserrno(host_error); 526 } 527 528 static struct posix_acl * 529 _get_posix_acl(struct dentry *dentry, char *key) 530 { 531 void *buf = NULL; 532 struct posix_acl *pacl = NULL; 533 int buflen; 534 535 buflen = nfsd_getxattr(dentry, key, &buf); 536 if (!buflen) 537 buflen = -ENODATA; 538 if (buflen <= 0) 539 return ERR_PTR(buflen); 540 541 pacl = posix_acl_from_xattr(buf, buflen); 542 kfree(buf); 543 return pacl; 544 } 545 546 int 547 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl) 548 { 549 struct inode *inode = dentry->d_inode; 550 int error = 0; 551 struct posix_acl *pacl = NULL, *dpacl = NULL; 552 unsigned int flags = 0; 553 554 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS); 555 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA) 556 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL); 557 if (IS_ERR(pacl)) { 558 error = PTR_ERR(pacl); 559 pacl = NULL; 560 goto out; 561 } 562 563 if (S_ISDIR(inode->i_mode)) { 564 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT); 565 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA) 566 dpacl = NULL; 567 else if (IS_ERR(dpacl)) { 568 error = PTR_ERR(dpacl); 569 dpacl = NULL; 570 goto out; 571 } 572 flags = NFS4_ACL_DIR; 573 } 574 575 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags); 576 if (IS_ERR(*acl)) { 577 error = PTR_ERR(*acl); 578 *acl = NULL; 579 } 580 out: 581 posix_acl_release(pacl); 582 posix_acl_release(dpacl); 583 return error; 584 } 585 586 #endif /* defined(CONFIG_NFS_V4) */ 587 588 #ifdef CONFIG_NFSD_V3 589 /* 590 * Check server access rights to a file system object 591 */ 592 struct accessmap { 593 u32 access; 594 int how; 595 }; 596 static struct accessmap nfs3_regaccess[] = { 597 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 598 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 599 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, 600 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, 601 602 { 0, 0 } 603 }; 604 605 static struct accessmap nfs3_diraccess[] = { 606 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 607 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, 608 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, 609 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, 610 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, 611 612 { 0, 0 } 613 }; 614 615 static struct accessmap nfs3_anyaccess[] = { 616 /* Some clients - Solaris 2.6 at least, make an access call 617 * to the server to check for access for things like /dev/null 618 * (which really, the server doesn't care about). So 619 * We provide simple access checking for them, looking 620 * mainly at mode bits, and we make sure to ignore read-only 621 * filesystem checks 622 */ 623 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 624 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 625 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 626 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 627 628 { 0, 0 } 629 }; 630 631 __be32 632 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) 633 { 634 struct accessmap *map; 635 struct svc_export *export; 636 struct dentry *dentry; 637 u32 query, result = 0, sresult = 0; 638 __be32 error; 639 640 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); 641 if (error) 642 goto out; 643 644 export = fhp->fh_export; 645 dentry = fhp->fh_dentry; 646 647 if (S_ISREG(dentry->d_inode->i_mode)) 648 map = nfs3_regaccess; 649 else if (S_ISDIR(dentry->d_inode->i_mode)) 650 map = nfs3_diraccess; 651 else 652 map = nfs3_anyaccess; 653 654 655 query = *access; 656 for (; map->access; map++) { 657 if (map->access & query) { 658 __be32 err2; 659 660 sresult |= map->access; 661 662 err2 = nfsd_permission(rqstp, export, dentry, map->how); 663 switch (err2) { 664 case nfs_ok: 665 result |= map->access; 666 break; 667 668 /* the following error codes just mean the access was not allowed, 669 * rather than an error occurred */ 670 case nfserr_rofs: 671 case nfserr_acces: 672 case nfserr_perm: 673 /* simply don't "or" in the access bit. */ 674 break; 675 default: 676 error = err2; 677 goto out; 678 } 679 } 680 } 681 *access = result; 682 if (supported) 683 *supported = sresult; 684 685 out: 686 return error; 687 } 688 #endif /* CONFIG_NFSD_V3 */ 689 690 691 692 /* 693 * Open an existing file or directory. 694 * The access argument indicates the type of open (read/write/lock) 695 * N.B. After this call fhp needs an fh_put 696 */ 697 __be32 698 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 699 int access, struct file **filp) 700 { 701 struct dentry *dentry; 702 struct inode *inode; 703 int flags = O_RDONLY|O_LARGEFILE; 704 __be32 err; 705 int host_err; 706 707 validate_process_creds(); 708 709 /* 710 * If we get here, then the client has already done an "open", 711 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 712 * in case a chmod has now revoked permission. 713 */ 714 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE); 715 if (err) 716 goto out; 717 718 dentry = fhp->fh_dentry; 719 inode = dentry->d_inode; 720 721 /* Disallow write access to files with the append-only bit set 722 * or any access when mandatory locking enabled 723 */ 724 err = nfserr_perm; 725 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE)) 726 goto out; 727 /* 728 * We must ignore files (but only files) which might have mandatory 729 * locks on them because there is no way to know if the accesser has 730 * the lock. 731 */ 732 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) 733 goto out; 734 735 if (!inode->i_fop) 736 goto out; 737 738 /* 739 * Check to see if there are any leases on this file. 740 * This may block while leases are broken. 741 */ 742 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0)); 743 if (host_err == -EWOULDBLOCK) 744 host_err = -ETIMEDOUT; 745 if (host_err) /* NOMEM or WOULDBLOCK */ 746 goto out_nfserr; 747 748 if (access & NFSD_MAY_WRITE) { 749 if (access & NFSD_MAY_READ) 750 flags = O_RDWR|O_LARGEFILE; 751 else 752 flags = O_WRONLY|O_LARGEFILE; 753 754 vfs_dq_init(inode); 755 } 756 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt), 757 flags, current_cred()); 758 if (IS_ERR(*filp)) 759 host_err = PTR_ERR(*filp); 760 else 761 ima_counts_get(*filp); 762 out_nfserr: 763 err = nfserrno(host_err); 764 out: 765 validate_process_creds(); 766 return err; 767 } 768 769 /* 770 * Close a file. 771 */ 772 void 773 nfsd_close(struct file *filp) 774 { 775 fput(filp); 776 } 777 778 /* 779 * Sync a file 780 * As this calls fsync (not fdatasync) there is no need for a write_inode 781 * after it. 782 */ 783 static inline int nfsd_dosync(struct file *filp, struct dentry *dp, 784 const struct file_operations *fop) 785 { 786 struct inode *inode = dp->d_inode; 787 int (*fsync) (struct file *, struct dentry *, int); 788 int err; 789 790 err = filemap_fdatawrite(inode->i_mapping); 791 if (err == 0 && fop && (fsync = fop->fsync)) 792 err = fsync(filp, dp, 0); 793 if (err == 0) 794 err = filemap_fdatawait(inode->i_mapping); 795 796 return err; 797 } 798 799 static int 800 nfsd_sync(struct file *filp) 801 { 802 int err; 803 struct inode *inode = filp->f_path.dentry->d_inode; 804 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name); 805 mutex_lock(&inode->i_mutex); 806 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op); 807 mutex_unlock(&inode->i_mutex); 808 809 return err; 810 } 811 812 int 813 nfsd_sync_dir(struct dentry *dp) 814 { 815 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop); 816 } 817 818 /* 819 * Obtain the readahead parameters for the file 820 * specified by (dev, ino). 821 */ 822 823 static inline struct raparms * 824 nfsd_get_raparms(dev_t dev, ino_t ino) 825 { 826 struct raparms *ra, **rap, **frap = NULL; 827 int depth = 0; 828 unsigned int hash; 829 struct raparm_hbucket *rab; 830 831 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 832 rab = &raparm_hash[hash]; 833 834 spin_lock(&rab->pb_lock); 835 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 836 if (ra->p_ino == ino && ra->p_dev == dev) 837 goto found; 838 depth++; 839 if (ra->p_count == 0) 840 frap = rap; 841 } 842 depth = nfsdstats.ra_size*11/10; 843 if (!frap) { 844 spin_unlock(&rab->pb_lock); 845 return NULL; 846 } 847 rap = frap; 848 ra = *frap; 849 ra->p_dev = dev; 850 ra->p_ino = ino; 851 ra->p_set = 0; 852 ra->p_hindex = hash; 853 found: 854 if (rap != &rab->pb_head) { 855 *rap = ra->p_next; 856 ra->p_next = rab->pb_head; 857 rab->pb_head = ra; 858 } 859 ra->p_count++; 860 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 861 spin_unlock(&rab->pb_lock); 862 return ra; 863 } 864 865 /* 866 * Grab and keep cached pages associated with a file in the svc_rqst 867 * so that they can be passed to the network sendmsg/sendpage routines 868 * directly. They will be released after the sending has completed. 869 */ 870 static int 871 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 872 struct splice_desc *sd) 873 { 874 struct svc_rqst *rqstp = sd->u.data; 875 struct page **pp = rqstp->rq_respages + rqstp->rq_resused; 876 struct page *page = buf->page; 877 size_t size; 878 int ret; 879 880 ret = buf->ops->confirm(pipe, buf); 881 if (unlikely(ret)) 882 return ret; 883 884 size = sd->len; 885 886 if (rqstp->rq_res.page_len == 0) { 887 get_page(page); 888 put_page(*pp); 889 *pp = page; 890 rqstp->rq_resused++; 891 rqstp->rq_res.page_base = buf->offset; 892 rqstp->rq_res.page_len = size; 893 } else if (page != pp[-1]) { 894 get_page(page); 895 if (*pp) 896 put_page(*pp); 897 *pp = page; 898 rqstp->rq_resused++; 899 rqstp->rq_res.page_len += size; 900 } else 901 rqstp->rq_res.page_len += size; 902 903 return size; 904 } 905 906 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 907 struct splice_desc *sd) 908 { 909 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 910 } 911 912 static inline int svc_msnfs(struct svc_fh *ffhp) 913 { 914 #ifdef MSNFS 915 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS); 916 #else 917 return 0; 918 #endif 919 } 920 921 static __be32 922 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 923 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 924 { 925 struct inode *inode; 926 struct raparms *ra; 927 mm_segment_t oldfs; 928 __be32 err; 929 int host_err; 930 931 err = nfserr_perm; 932 inode = file->f_path.dentry->d_inode; 933 934 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count)) 935 goto out; 936 937 /* Get readahead parameters */ 938 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 939 940 if (ra && ra->p_set) 941 file->f_ra = ra->p_ra; 942 943 if (file->f_op->splice_read && rqstp->rq_splice_ok) { 944 struct splice_desc sd = { 945 .len = 0, 946 .total_len = *count, 947 .pos = offset, 948 .u.data = rqstp, 949 }; 950 951 rqstp->rq_resused = 1; 952 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 953 } else { 954 oldfs = get_fs(); 955 set_fs(KERNEL_DS); 956 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 957 set_fs(oldfs); 958 } 959 960 /* Write back readahead params */ 961 if (ra) { 962 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 963 spin_lock(&rab->pb_lock); 964 ra->p_ra = file->f_ra; 965 ra->p_set = 1; 966 ra->p_count--; 967 spin_unlock(&rab->pb_lock); 968 } 969 970 if (host_err >= 0) { 971 nfsdstats.io_read += host_err; 972 *count = host_err; 973 err = 0; 974 fsnotify_access(file->f_path.dentry); 975 } else 976 err = nfserrno(host_err); 977 out: 978 return err; 979 } 980 981 static void kill_suid(struct dentry *dentry) 982 { 983 struct iattr ia; 984 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 985 986 mutex_lock(&dentry->d_inode->i_mutex); 987 notify_change(dentry, &ia); 988 mutex_unlock(&dentry->d_inode->i_mutex); 989 } 990 991 /* 992 * Gathered writes: If another process is currently writing to the file, 993 * there's a high chance this is another nfsd (triggered by a bulk write 994 * from a client's biod). Rather than syncing the file with each write 995 * request, we sleep for 10 msec. 996 * 997 * I don't know if this roughly approximates C. Juszak's idea of 998 * gathered writes, but it's a nice and simple solution (IMHO), and it 999 * seems to work:-) 1000 * 1001 * Note: we do this only in the NFSv2 case, since v3 and higher have a 1002 * better tool (separate unstable writes and commits) for solving this 1003 * problem. 1004 */ 1005 static int wait_for_concurrent_writes(struct file *file) 1006 { 1007 struct inode *inode = file->f_path.dentry->d_inode; 1008 static ino_t last_ino; 1009 static dev_t last_dev; 1010 int err = 0; 1011 1012 if (atomic_read(&inode->i_writecount) > 1 1013 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 1014 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 1015 msleep(10); 1016 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 1017 } 1018 1019 if (inode->i_state & I_DIRTY) { 1020 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 1021 err = nfsd_sync(file); 1022 } 1023 last_ino = inode->i_ino; 1024 last_dev = inode->i_sb->s_dev; 1025 return err; 1026 } 1027 1028 static __be32 1029 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1030 loff_t offset, struct kvec *vec, int vlen, 1031 unsigned long *cnt, int *stablep) 1032 { 1033 struct svc_export *exp; 1034 struct dentry *dentry; 1035 struct inode *inode; 1036 mm_segment_t oldfs; 1037 __be32 err = 0; 1038 int host_err; 1039 int stable = *stablep; 1040 int use_wgather; 1041 1042 #ifdef MSNFS 1043 err = nfserr_perm; 1044 1045 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1046 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt))) 1047 goto out; 1048 #endif 1049 1050 dentry = file->f_path.dentry; 1051 inode = dentry->d_inode; 1052 exp = fhp->fh_export; 1053 1054 /* 1055 * Request sync writes if 1056 * - the sync export option has been set, or 1057 * - the client requested O_SYNC behavior (NFSv3 feature). 1058 * - The file system doesn't support fsync(). 1059 * When NFSv2 gathered writes have been configured for this volume, 1060 * flushing the data to disk is handled separately below. 1061 */ 1062 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 1063 1064 if (!file->f_op->fsync) {/* COMMIT3 cannot work */ 1065 stable = 2; 1066 *stablep = 2; /* FILE_SYNC */ 1067 } 1068 1069 if (!EX_ISSYNC(exp)) 1070 stable = 0; 1071 if (stable && !use_wgather) { 1072 spin_lock(&file->f_lock); 1073 file->f_flags |= O_SYNC; 1074 spin_unlock(&file->f_lock); 1075 } 1076 1077 /* Write the data. */ 1078 oldfs = get_fs(); set_fs(KERNEL_DS); 1079 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset); 1080 set_fs(oldfs); 1081 if (host_err < 0) 1082 goto out_nfserr; 1083 *cnt = host_err; 1084 nfsdstats.io_write += host_err; 1085 fsnotify_modify(file->f_path.dentry); 1086 1087 /* clear setuid/setgid flag after write */ 1088 if (inode->i_mode & (S_ISUID | S_ISGID)) 1089 kill_suid(dentry); 1090 1091 if (stable && use_wgather) 1092 host_err = wait_for_concurrent_writes(file); 1093 1094 out_nfserr: 1095 dprintk("nfsd: write complete host_err=%d\n", host_err); 1096 if (host_err >= 0) 1097 err = 0; 1098 else 1099 err = nfserrno(host_err); 1100 out: 1101 return err; 1102 } 1103 1104 /* 1105 * Read data from a file. count must contain the requested read count 1106 * on entry. On return, *count contains the number of bytes actually read. 1107 * N.B. After this call fhp needs an fh_put 1108 */ 1109 __be32 1110 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1111 loff_t offset, struct kvec *vec, int vlen, 1112 unsigned long *count) 1113 { 1114 __be32 err; 1115 1116 if (file) { 1117 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1118 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); 1119 if (err) 1120 goto out; 1121 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1122 } else { 1123 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); 1124 if (err) 1125 goto out; 1126 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1127 nfsd_close(file); 1128 } 1129 out: 1130 return err; 1131 } 1132 1133 /* 1134 * Write data to a file. 1135 * The stable flag requests synchronous writes. 1136 * N.B. After this call fhp needs an fh_put 1137 */ 1138 __be32 1139 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1140 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1141 int *stablep) 1142 { 1143 __be32 err = 0; 1144 1145 if (file) { 1146 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1147 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1148 if (err) 1149 goto out; 1150 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1151 stablep); 1152 } else { 1153 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1154 if (err) 1155 goto out; 1156 1157 if (cnt) 1158 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1159 cnt, stablep); 1160 nfsd_close(file); 1161 } 1162 out: 1163 return err; 1164 } 1165 1166 #ifdef CONFIG_NFSD_V3 1167 /* 1168 * Commit all pending writes to stable storage. 1169 * Strictly speaking, we could sync just the indicated file region here, 1170 * but there's currently no way we can ask the VFS to do so. 1171 * 1172 * Unfortunately we cannot lock the file to make sure we return full WCC 1173 * data to the client, as locking happens lower down in the filesystem. 1174 */ 1175 __be32 1176 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1177 loff_t offset, unsigned long count) 1178 { 1179 struct file *file; 1180 __be32 err; 1181 1182 if ((u64)count > ~(u64)offset) 1183 return nfserr_inval; 1184 1185 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1186 if (err) 1187 return err; 1188 if (EX_ISSYNC(fhp->fh_export)) { 1189 if (file->f_op && file->f_op->fsync) { 1190 err = nfserrno(nfsd_sync(file)); 1191 } else { 1192 err = nfserr_notsupp; 1193 } 1194 } 1195 1196 nfsd_close(file); 1197 return err; 1198 } 1199 #endif /* CONFIG_NFSD_V3 */ 1200 1201 static __be32 1202 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1203 struct iattr *iap) 1204 { 1205 /* 1206 * Mode has already been set earlier in create: 1207 */ 1208 iap->ia_valid &= ~ATTR_MODE; 1209 /* 1210 * Setting uid/gid works only for root. Irix appears to 1211 * send along the gid on create when it tries to implement 1212 * setgid directories via NFS: 1213 */ 1214 if (current_fsuid() != 0) 1215 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1216 if (iap->ia_valid) 1217 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1218 return 0; 1219 } 1220 1221 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1222 * setting size to 0 may fail for some specific file systems by the permission 1223 * checking which requires WRITE permission but the mode is 000. 1224 * we ignore the resizing(to 0) on the just new created file, since the size is 1225 * 0 after file created. 1226 * 1227 * call this only after vfs_create() is called. 1228 * */ 1229 static void 1230 nfsd_check_ignore_resizing(struct iattr *iap) 1231 { 1232 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1233 iap->ia_valid &= ~ATTR_SIZE; 1234 } 1235 1236 /* 1237 * Create a file (regular, directory, device, fifo); UNIX sockets 1238 * not yet implemented. 1239 * If the response fh has been verified, the parent directory should 1240 * already be locked. Note that the parent directory is left locked. 1241 * 1242 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1243 */ 1244 __be32 1245 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1246 char *fname, int flen, struct iattr *iap, 1247 int type, dev_t rdev, struct svc_fh *resfhp) 1248 { 1249 struct dentry *dentry, *dchild = NULL; 1250 struct inode *dirp; 1251 __be32 err; 1252 __be32 err2; 1253 int host_err; 1254 1255 err = nfserr_perm; 1256 if (!flen) 1257 goto out; 1258 err = nfserr_exist; 1259 if (isdotent(fname, flen)) 1260 goto out; 1261 1262 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1263 if (err) 1264 goto out; 1265 1266 dentry = fhp->fh_dentry; 1267 dirp = dentry->d_inode; 1268 1269 err = nfserr_notdir; 1270 if (!dirp->i_op->lookup) 1271 goto out; 1272 /* 1273 * Check whether the response file handle has been verified yet. 1274 * If it has, the parent directory should already be locked. 1275 */ 1276 if (!resfhp->fh_dentry) { 1277 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1278 fh_lock_nested(fhp, I_MUTEX_PARENT); 1279 dchild = lookup_one_len(fname, dentry, flen); 1280 host_err = PTR_ERR(dchild); 1281 if (IS_ERR(dchild)) 1282 goto out_nfserr; 1283 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1284 if (err) 1285 goto out; 1286 } else { 1287 /* called from nfsd_proc_create */ 1288 dchild = dget(resfhp->fh_dentry); 1289 if (!fhp->fh_locked) { 1290 /* not actually possible */ 1291 printk(KERN_ERR 1292 "nfsd_create: parent %s/%s not locked!\n", 1293 dentry->d_parent->d_name.name, 1294 dentry->d_name.name); 1295 err = nfserr_io; 1296 goto out; 1297 } 1298 } 1299 /* 1300 * Make sure the child dentry is still negative ... 1301 */ 1302 err = nfserr_exist; 1303 if (dchild->d_inode) { 1304 dprintk("nfsd_create: dentry %s/%s not negative!\n", 1305 dentry->d_name.name, dchild->d_name.name); 1306 goto out; 1307 } 1308 1309 if (!(iap->ia_valid & ATTR_MODE)) 1310 iap->ia_mode = 0; 1311 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1312 1313 err = nfserr_inval; 1314 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1315 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1316 type); 1317 goto out; 1318 } 1319 1320 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1321 if (host_err) 1322 goto out_nfserr; 1323 1324 /* 1325 * Get the dir op function pointer. 1326 */ 1327 err = 0; 1328 switch (type) { 1329 case S_IFREG: 1330 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1331 if (!host_err) 1332 nfsd_check_ignore_resizing(iap); 1333 break; 1334 case S_IFDIR: 1335 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1336 break; 1337 case S_IFCHR: 1338 case S_IFBLK: 1339 case S_IFIFO: 1340 case S_IFSOCK: 1341 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1342 break; 1343 } 1344 if (host_err < 0) { 1345 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1346 goto out_nfserr; 1347 } 1348 1349 if (EX_ISSYNC(fhp->fh_export)) { 1350 err = nfserrno(nfsd_sync_dir(dentry)); 1351 write_inode_now(dchild->d_inode, 1); 1352 } 1353 1354 err2 = nfsd_create_setattr(rqstp, resfhp, iap); 1355 if (err2) 1356 err = err2; 1357 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1358 /* 1359 * Update the file handle to get the new inode info. 1360 */ 1361 if (!err) 1362 err = fh_update(resfhp); 1363 out: 1364 if (dchild && !IS_ERR(dchild)) 1365 dput(dchild); 1366 return err; 1367 1368 out_nfserr: 1369 err = nfserrno(host_err); 1370 goto out; 1371 } 1372 1373 #ifdef CONFIG_NFSD_V3 1374 /* 1375 * NFSv3 version of nfsd_create 1376 */ 1377 __be32 1378 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp, 1379 char *fname, int flen, struct iattr *iap, 1380 struct svc_fh *resfhp, int createmode, u32 *verifier, 1381 int *truncp, int *created) 1382 { 1383 struct dentry *dentry, *dchild = NULL; 1384 struct inode *dirp; 1385 __be32 err; 1386 __be32 err2; 1387 int host_err; 1388 __u32 v_mtime=0, v_atime=0; 1389 1390 err = nfserr_perm; 1391 if (!flen) 1392 goto out; 1393 err = nfserr_exist; 1394 if (isdotent(fname, flen)) 1395 goto out; 1396 if (!(iap->ia_valid & ATTR_MODE)) 1397 iap->ia_mode = 0; 1398 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1399 if (err) 1400 goto out; 1401 1402 dentry = fhp->fh_dentry; 1403 dirp = dentry->d_inode; 1404 1405 /* Get all the sanity checks out of the way before 1406 * we lock the parent. */ 1407 err = nfserr_notdir; 1408 if (!dirp->i_op->lookup) 1409 goto out; 1410 fh_lock_nested(fhp, I_MUTEX_PARENT); 1411 1412 /* 1413 * Compose the response file handle. 1414 */ 1415 dchild = lookup_one_len(fname, dentry, flen); 1416 host_err = PTR_ERR(dchild); 1417 if (IS_ERR(dchild)) 1418 goto out_nfserr; 1419 1420 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1421 if (err) 1422 goto out; 1423 1424 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1425 /* solaris7 gets confused (bugid 4218508) if these have 1426 * the high bit set, so just clear the high bits. If this is 1427 * ever changed to use different attrs for storing the 1428 * verifier, then do_open_lookup() will also need to be fixed 1429 * accordingly. 1430 */ 1431 v_mtime = verifier[0]&0x7fffffff; 1432 v_atime = verifier[1]&0x7fffffff; 1433 } 1434 1435 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1436 if (host_err) 1437 goto out_nfserr; 1438 if (dchild->d_inode) { 1439 err = 0; 1440 1441 switch (createmode) { 1442 case NFS3_CREATE_UNCHECKED: 1443 if (! S_ISREG(dchild->d_inode->i_mode)) 1444 err = nfserr_exist; 1445 else if (truncp) { 1446 /* in nfsv4, we need to treat this case a little 1447 * differently. we don't want to truncate the 1448 * file now; this would be wrong if the OPEN 1449 * fails for some other reason. furthermore, 1450 * if the size is nonzero, we should ignore it 1451 * according to spec! 1452 */ 1453 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1454 } 1455 else { 1456 iap->ia_valid &= ATTR_SIZE; 1457 goto set_attr; 1458 } 1459 break; 1460 case NFS3_CREATE_EXCLUSIVE: 1461 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1462 && dchild->d_inode->i_atime.tv_sec == v_atime 1463 && dchild->d_inode->i_size == 0 ) 1464 break; 1465 /* fallthru */ 1466 case NFS3_CREATE_GUARDED: 1467 err = nfserr_exist; 1468 } 1469 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1470 goto out; 1471 } 1472 1473 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL); 1474 if (host_err < 0) { 1475 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1476 goto out_nfserr; 1477 } 1478 if (created) 1479 *created = 1; 1480 1481 if (EX_ISSYNC(fhp->fh_export)) { 1482 err = nfserrno(nfsd_sync_dir(dentry)); 1483 /* setattr will sync the child (or not) */ 1484 } 1485 1486 nfsd_check_ignore_resizing(iap); 1487 1488 if (createmode == NFS3_CREATE_EXCLUSIVE) { 1489 /* Cram the verifier into atime/mtime */ 1490 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1491 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1492 /* XXX someone who knows this better please fix it for nsec */ 1493 iap->ia_mtime.tv_sec = v_mtime; 1494 iap->ia_atime.tv_sec = v_atime; 1495 iap->ia_mtime.tv_nsec = 0; 1496 iap->ia_atime.tv_nsec = 0; 1497 } 1498 1499 set_attr: 1500 err2 = nfsd_create_setattr(rqstp, resfhp, iap); 1501 if (err2) 1502 err = err2; 1503 1504 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1505 /* 1506 * Update the filehandle to get the new inode info. 1507 */ 1508 if (!err) 1509 err = fh_update(resfhp); 1510 1511 out: 1512 fh_unlock(fhp); 1513 if (dchild && !IS_ERR(dchild)) 1514 dput(dchild); 1515 return err; 1516 1517 out_nfserr: 1518 err = nfserrno(host_err); 1519 goto out; 1520 } 1521 #endif /* CONFIG_NFSD_V3 */ 1522 1523 /* 1524 * Read a symlink. On entry, *lenp must contain the maximum path length that 1525 * fits into the buffer. On return, it contains the true length. 1526 * N.B. After this call fhp needs an fh_put 1527 */ 1528 __be32 1529 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1530 { 1531 struct dentry *dentry; 1532 struct inode *inode; 1533 mm_segment_t oldfs; 1534 __be32 err; 1535 int host_err; 1536 1537 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1538 if (err) 1539 goto out; 1540 1541 dentry = fhp->fh_dentry; 1542 inode = dentry->d_inode; 1543 1544 err = nfserr_inval; 1545 if (!inode->i_op->readlink) 1546 goto out; 1547 1548 touch_atime(fhp->fh_export->ex_path.mnt, dentry); 1549 /* N.B. Why does this call need a get_fs()?? 1550 * Remove the set_fs and watch the fireworks:-) --okir 1551 */ 1552 1553 oldfs = get_fs(); set_fs(KERNEL_DS); 1554 host_err = inode->i_op->readlink(dentry, buf, *lenp); 1555 set_fs(oldfs); 1556 1557 if (host_err < 0) 1558 goto out_nfserr; 1559 *lenp = host_err; 1560 err = 0; 1561 out: 1562 return err; 1563 1564 out_nfserr: 1565 err = nfserrno(host_err); 1566 goto out; 1567 } 1568 1569 /* 1570 * Create a symlink and look up its inode 1571 * N.B. After this call _both_ fhp and resfhp need an fh_put 1572 */ 1573 __be32 1574 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1575 char *fname, int flen, 1576 char *path, int plen, 1577 struct svc_fh *resfhp, 1578 struct iattr *iap) 1579 { 1580 struct dentry *dentry, *dnew; 1581 __be32 err, cerr; 1582 int host_err; 1583 1584 err = nfserr_noent; 1585 if (!flen || !plen) 1586 goto out; 1587 err = nfserr_exist; 1588 if (isdotent(fname, flen)) 1589 goto out; 1590 1591 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1592 if (err) 1593 goto out; 1594 fh_lock(fhp); 1595 dentry = fhp->fh_dentry; 1596 dnew = lookup_one_len(fname, dentry, flen); 1597 host_err = PTR_ERR(dnew); 1598 if (IS_ERR(dnew)) 1599 goto out_nfserr; 1600 1601 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1602 if (host_err) 1603 goto out_nfserr; 1604 1605 if (unlikely(path[plen] != 0)) { 1606 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1607 if (path_alloced == NULL) 1608 host_err = -ENOMEM; 1609 else { 1610 strncpy(path_alloced, path, plen); 1611 path_alloced[plen] = 0; 1612 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); 1613 kfree(path_alloced); 1614 } 1615 } else 1616 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1617 1618 if (!host_err) { 1619 if (EX_ISSYNC(fhp->fh_export)) 1620 host_err = nfsd_sync_dir(dentry); 1621 } 1622 err = nfserrno(host_err); 1623 fh_unlock(fhp); 1624 1625 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1626 1627 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1628 dput(dnew); 1629 if (err==0) err = cerr; 1630 out: 1631 return err; 1632 1633 out_nfserr: 1634 err = nfserrno(host_err); 1635 goto out; 1636 } 1637 1638 /* 1639 * Create a hardlink 1640 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1641 */ 1642 __be32 1643 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1644 char *name, int len, struct svc_fh *tfhp) 1645 { 1646 struct dentry *ddir, *dnew, *dold; 1647 struct inode *dirp, *dest; 1648 __be32 err; 1649 int host_err; 1650 1651 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1652 if (err) 1653 goto out; 1654 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP); 1655 if (err) 1656 goto out; 1657 1658 err = nfserr_perm; 1659 if (!len) 1660 goto out; 1661 err = nfserr_exist; 1662 if (isdotent(name, len)) 1663 goto out; 1664 1665 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1666 ddir = ffhp->fh_dentry; 1667 dirp = ddir->d_inode; 1668 1669 dnew = lookup_one_len(name, ddir, len); 1670 host_err = PTR_ERR(dnew); 1671 if (IS_ERR(dnew)) 1672 goto out_nfserr; 1673 1674 dold = tfhp->fh_dentry; 1675 dest = dold->d_inode; 1676 1677 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt); 1678 if (host_err) { 1679 err = nfserrno(host_err); 1680 goto out_dput; 1681 } 1682 host_err = vfs_link(dold, dirp, dnew); 1683 if (!host_err) { 1684 if (EX_ISSYNC(ffhp->fh_export)) { 1685 err = nfserrno(nfsd_sync_dir(ddir)); 1686 write_inode_now(dest, 1); 1687 } 1688 err = 0; 1689 } else { 1690 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1691 err = nfserr_acces; 1692 else 1693 err = nfserrno(host_err); 1694 } 1695 mnt_drop_write(tfhp->fh_export->ex_path.mnt); 1696 out_dput: 1697 dput(dnew); 1698 out_unlock: 1699 fh_unlock(ffhp); 1700 out: 1701 return err; 1702 1703 out_nfserr: 1704 err = nfserrno(host_err); 1705 goto out_unlock; 1706 } 1707 1708 /* 1709 * Rename a file 1710 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1711 */ 1712 __be32 1713 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1714 struct svc_fh *tfhp, char *tname, int tlen) 1715 { 1716 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1717 struct inode *fdir, *tdir; 1718 __be32 err; 1719 int host_err; 1720 1721 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1722 if (err) 1723 goto out; 1724 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1725 if (err) 1726 goto out; 1727 1728 fdentry = ffhp->fh_dentry; 1729 fdir = fdentry->d_inode; 1730 1731 tdentry = tfhp->fh_dentry; 1732 tdir = tdentry->d_inode; 1733 1734 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev; 1735 if (ffhp->fh_export != tfhp->fh_export) 1736 goto out; 1737 1738 err = nfserr_perm; 1739 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1740 goto out; 1741 1742 /* cannot use fh_lock as we need deadlock protective ordering 1743 * so do it by hand */ 1744 trap = lock_rename(tdentry, fdentry); 1745 ffhp->fh_locked = tfhp->fh_locked = 1; 1746 fill_pre_wcc(ffhp); 1747 fill_pre_wcc(tfhp); 1748 1749 odentry = lookup_one_len(fname, fdentry, flen); 1750 host_err = PTR_ERR(odentry); 1751 if (IS_ERR(odentry)) 1752 goto out_nfserr; 1753 1754 host_err = -ENOENT; 1755 if (!odentry->d_inode) 1756 goto out_dput_old; 1757 host_err = -EINVAL; 1758 if (odentry == trap) 1759 goto out_dput_old; 1760 1761 ndentry = lookup_one_len(tname, tdentry, tlen); 1762 host_err = PTR_ERR(ndentry); 1763 if (IS_ERR(ndentry)) 1764 goto out_dput_old; 1765 host_err = -ENOTEMPTY; 1766 if (ndentry == trap) 1767 goto out_dput_new; 1768 1769 if (svc_msnfs(ffhp) && 1770 ((atomic_read(&odentry->d_count) > 1) 1771 || (atomic_read(&ndentry->d_count) > 1))) { 1772 host_err = -EPERM; 1773 goto out_dput_new; 1774 } 1775 1776 host_err = -EXDEV; 1777 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1778 goto out_dput_new; 1779 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt); 1780 if (host_err) 1781 goto out_dput_new; 1782 1783 host_err = vfs_rename(fdir, odentry, tdir, ndentry); 1784 if (!host_err && EX_ISSYNC(tfhp->fh_export)) { 1785 host_err = nfsd_sync_dir(tdentry); 1786 if (!host_err) 1787 host_err = nfsd_sync_dir(fdentry); 1788 } 1789 1790 mnt_drop_write(ffhp->fh_export->ex_path.mnt); 1791 1792 out_dput_new: 1793 dput(ndentry); 1794 out_dput_old: 1795 dput(odentry); 1796 out_nfserr: 1797 err = nfserrno(host_err); 1798 1799 /* we cannot reply on fh_unlock on the two filehandles, 1800 * as that would do the wrong thing if the two directories 1801 * were the same, so again we do it by hand 1802 */ 1803 fill_post_wcc(ffhp); 1804 fill_post_wcc(tfhp); 1805 unlock_rename(tdentry, fdentry); 1806 ffhp->fh_locked = tfhp->fh_locked = 0; 1807 1808 out: 1809 return err; 1810 } 1811 1812 /* 1813 * Unlink a file or directory 1814 * N.B. After this call fhp needs an fh_put 1815 */ 1816 __be32 1817 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1818 char *fname, int flen) 1819 { 1820 struct dentry *dentry, *rdentry; 1821 struct inode *dirp; 1822 __be32 err; 1823 int host_err; 1824 1825 err = nfserr_acces; 1826 if (!flen || isdotent(fname, flen)) 1827 goto out; 1828 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1829 if (err) 1830 goto out; 1831 1832 fh_lock_nested(fhp, I_MUTEX_PARENT); 1833 dentry = fhp->fh_dentry; 1834 dirp = dentry->d_inode; 1835 1836 rdentry = lookup_one_len(fname, dentry, flen); 1837 host_err = PTR_ERR(rdentry); 1838 if (IS_ERR(rdentry)) 1839 goto out_nfserr; 1840 1841 if (!rdentry->d_inode) { 1842 dput(rdentry); 1843 err = nfserr_noent; 1844 goto out; 1845 } 1846 1847 if (!type) 1848 type = rdentry->d_inode->i_mode & S_IFMT; 1849 1850 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt); 1851 if (host_err) 1852 goto out_nfserr; 1853 1854 if (type != S_IFDIR) { /* It's UNLINK */ 1855 #ifdef MSNFS 1856 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) && 1857 (atomic_read(&rdentry->d_count) > 1)) { 1858 host_err = -EPERM; 1859 } else 1860 #endif 1861 host_err = vfs_unlink(dirp, rdentry); 1862 } else { /* It's RMDIR */ 1863 host_err = vfs_rmdir(dirp, rdentry); 1864 } 1865 1866 dput(rdentry); 1867 1868 if (host_err) 1869 goto out_drop; 1870 if (EX_ISSYNC(fhp->fh_export)) 1871 host_err = nfsd_sync_dir(dentry); 1872 1873 out_drop: 1874 mnt_drop_write(fhp->fh_export->ex_path.mnt); 1875 out_nfserr: 1876 err = nfserrno(host_err); 1877 out: 1878 return err; 1879 } 1880 1881 /* 1882 * We do this buffering because we must not call back into the file 1883 * system's ->lookup() method from the filldir callback. That may well 1884 * deadlock a number of file systems. 1885 * 1886 * This is based heavily on the implementation of same in XFS. 1887 */ 1888 struct buffered_dirent { 1889 u64 ino; 1890 loff_t offset; 1891 int namlen; 1892 unsigned int d_type; 1893 char name[]; 1894 }; 1895 1896 struct readdir_data { 1897 char *dirent; 1898 size_t used; 1899 int full; 1900 }; 1901 1902 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1903 loff_t offset, u64 ino, unsigned int d_type) 1904 { 1905 struct readdir_data *buf = __buf; 1906 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1907 unsigned int reclen; 1908 1909 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1910 if (buf->used + reclen > PAGE_SIZE) { 1911 buf->full = 1; 1912 return -EINVAL; 1913 } 1914 1915 de->namlen = namlen; 1916 de->offset = offset; 1917 de->ino = ino; 1918 de->d_type = d_type; 1919 memcpy(de->name, name, namlen); 1920 buf->used += reclen; 1921 1922 return 0; 1923 } 1924 1925 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1926 struct readdir_cd *cdp, loff_t *offsetp) 1927 { 1928 struct readdir_data buf; 1929 struct buffered_dirent *de; 1930 int host_err; 1931 int size; 1932 loff_t offset; 1933 1934 buf.dirent = (void *)__get_free_page(GFP_KERNEL); 1935 if (!buf.dirent) 1936 return nfserrno(-ENOMEM); 1937 1938 offset = *offsetp; 1939 1940 while (1) { 1941 struct inode *dir_inode = file->f_path.dentry->d_inode; 1942 unsigned int reclen; 1943 1944 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1945 buf.used = 0; 1946 buf.full = 0; 1947 1948 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf); 1949 if (buf.full) 1950 host_err = 0; 1951 1952 if (host_err < 0) 1953 break; 1954 1955 size = buf.used; 1956 1957 if (!size) 1958 break; 1959 1960 /* 1961 * Various filldir functions may end up calling back into 1962 * lookup_one_len() and the file system's ->lookup() method. 1963 * These expect i_mutex to be held, as it would within readdir. 1964 */ 1965 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1966 if (host_err) 1967 break; 1968 1969 de = (struct buffered_dirent *)buf.dirent; 1970 while (size > 0) { 1971 offset = de->offset; 1972 1973 if (func(cdp, de->name, de->namlen, de->offset, 1974 de->ino, de->d_type)) 1975 break; 1976 1977 if (cdp->err != nfs_ok) 1978 break; 1979 1980 reclen = ALIGN(sizeof(*de) + de->namlen, 1981 sizeof(u64)); 1982 size -= reclen; 1983 de = (struct buffered_dirent *)((char *)de + reclen); 1984 } 1985 mutex_unlock(&dir_inode->i_mutex); 1986 if (size > 0) /* We bailed out early */ 1987 break; 1988 1989 offset = vfs_llseek(file, 0, SEEK_CUR); 1990 } 1991 1992 free_page((unsigned long)(buf.dirent)); 1993 1994 if (host_err) 1995 return nfserrno(host_err); 1996 1997 *offsetp = offset; 1998 return cdp->err; 1999 } 2000 2001 /* 2002 * Read entries from a directory. 2003 * The NFSv3/4 verifier we ignore for now. 2004 */ 2005 __be32 2006 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 2007 struct readdir_cd *cdp, filldir_t func) 2008 { 2009 __be32 err; 2010 struct file *file; 2011 loff_t offset = *offsetp; 2012 2013 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file); 2014 if (err) 2015 goto out; 2016 2017 offset = vfs_llseek(file, offset, 0); 2018 if (offset < 0) { 2019 err = nfserrno((int)offset); 2020 goto out_close; 2021 } 2022 2023 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 2024 2025 if (err == nfserr_eof || err == nfserr_toosmall) 2026 err = nfs_ok; /* can still be found in ->err */ 2027 out_close: 2028 nfsd_close(file); 2029 out: 2030 return err; 2031 } 2032 2033 /* 2034 * Get file system stats 2035 * N.B. After this call fhp needs an fh_put 2036 */ 2037 __be32 2038 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 2039 { 2040 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 2041 if (!err && vfs_statfs(fhp->fh_dentry,stat)) 2042 err = nfserr_io; 2043 return err; 2044 } 2045 2046 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 2047 { 2048 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 2049 } 2050 2051 /* 2052 * Check for a user's access permissions to this inode. 2053 */ 2054 __be32 2055 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 2056 struct dentry *dentry, int acc) 2057 { 2058 struct inode *inode = dentry->d_inode; 2059 struct path path; 2060 int err; 2061 2062 if (acc == NFSD_MAY_NOP) 2063 return 0; 2064 #if 0 2065 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 2066 acc, 2067 (acc & NFSD_MAY_READ)? " read" : "", 2068 (acc & NFSD_MAY_WRITE)? " write" : "", 2069 (acc & NFSD_MAY_EXEC)? " exec" : "", 2070 (acc & NFSD_MAY_SATTR)? " sattr" : "", 2071 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 2072 (acc & NFSD_MAY_LOCK)? " lock" : "", 2073 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 2074 inode->i_mode, 2075 IS_IMMUTABLE(inode)? " immut" : "", 2076 IS_APPEND(inode)? " append" : "", 2077 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2078 dprintk(" owner %d/%d user %d/%d\n", 2079 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2080 #endif 2081 2082 /* Normally we reject any write/sattr etc access on a read-only file 2083 * system. But if it is IRIX doing check on write-access for a 2084 * device special file, we ignore rofs. 2085 */ 2086 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2087 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2088 if (exp_rdonly(rqstp, exp) || 2089 __mnt_is_readonly(exp->ex_path.mnt)) 2090 return nfserr_rofs; 2091 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2092 return nfserr_perm; 2093 } 2094 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2095 return nfserr_perm; 2096 2097 if (acc & NFSD_MAY_LOCK) { 2098 /* If we cannot rely on authentication in NLM requests, 2099 * just allow locks, otherwise require read permission, or 2100 * ownership 2101 */ 2102 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2103 return 0; 2104 else 2105 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2106 } 2107 /* 2108 * The file owner always gets access permission for accesses that 2109 * would normally be checked at open time. This is to make 2110 * file access work even when the client has done a fchmod(fd, 0). 2111 * 2112 * However, `cp foo bar' should fail nevertheless when bar is 2113 * readonly. A sensible way to do this might be to reject all 2114 * attempts to truncate a read-only file, because a creat() call 2115 * always implies file truncation. 2116 * ... but this isn't really fair. A process may reasonably call 2117 * ftruncate on an open file descriptor on a file with perm 000. 2118 * We must trust the client to do permission checking - using "ACCESS" 2119 * with NFSv3. 2120 */ 2121 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2122 inode->i_uid == current_fsuid()) 2123 return 0; 2124 2125 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2126 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2127 2128 /* Allow read access to binaries even when mode 111 */ 2129 if (err == -EACCES && S_ISREG(inode->i_mode) && 2130 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE)) 2131 err = inode_permission(inode, MAY_EXEC); 2132 if (err) 2133 goto nfsd_out; 2134 2135 /* Do integrity (permission) checking now, but defer incrementing 2136 * IMA counts to the actual file open. 2137 */ 2138 path.mnt = exp->ex_path.mnt; 2139 path.dentry = dentry; 2140 err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC), 2141 IMA_COUNT_LEAVE); 2142 nfsd_out: 2143 return err? nfserrno(err) : 0; 2144 } 2145 2146 void 2147 nfsd_racache_shutdown(void) 2148 { 2149 struct raparms *raparm, *last_raparm; 2150 unsigned int i; 2151 2152 dprintk("nfsd: freeing readahead buffers.\n"); 2153 2154 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2155 raparm = raparm_hash[i].pb_head; 2156 while(raparm) { 2157 last_raparm = raparm; 2158 raparm = raparm->p_next; 2159 kfree(last_raparm); 2160 } 2161 raparm_hash[i].pb_head = NULL; 2162 } 2163 } 2164 /* 2165 * Initialize readahead param cache 2166 */ 2167 int 2168 nfsd_racache_init(int cache_size) 2169 { 2170 int i; 2171 int j = 0; 2172 int nperbucket; 2173 struct raparms **raparm = NULL; 2174 2175 2176 if (raparm_hash[0].pb_head) 2177 return 0; 2178 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2179 if (nperbucket < 2) 2180 nperbucket = 2; 2181 cache_size = nperbucket * RAPARM_HASH_SIZE; 2182 2183 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2184 2185 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2186 spin_lock_init(&raparm_hash[i].pb_lock); 2187 2188 raparm = &raparm_hash[i].pb_head; 2189 for (j = 0; j < nperbucket; j++) { 2190 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2191 if (!*raparm) 2192 goto out_nomem; 2193 raparm = &(*raparm)->p_next; 2194 } 2195 *raparm = NULL; 2196 } 2197 2198 nfsdstats.ra_size = cache_size; 2199 return 0; 2200 2201 out_nomem: 2202 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2203 nfsd_racache_shutdown(); 2204 return -ENOMEM; 2205 } 2206 2207 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 2208 struct posix_acl * 2209 nfsd_get_posix_acl(struct svc_fh *fhp, int type) 2210 { 2211 struct inode *inode = fhp->fh_dentry->d_inode; 2212 char *name; 2213 void *value = NULL; 2214 ssize_t size; 2215 struct posix_acl *acl; 2216 2217 if (!IS_POSIXACL(inode)) 2218 return ERR_PTR(-EOPNOTSUPP); 2219 2220 switch (type) { 2221 case ACL_TYPE_ACCESS: 2222 name = POSIX_ACL_XATTR_ACCESS; 2223 break; 2224 case ACL_TYPE_DEFAULT: 2225 name = POSIX_ACL_XATTR_DEFAULT; 2226 break; 2227 default: 2228 return ERR_PTR(-EOPNOTSUPP); 2229 } 2230 2231 size = nfsd_getxattr(fhp->fh_dentry, name, &value); 2232 if (size < 0) 2233 return ERR_PTR(size); 2234 2235 acl = posix_acl_from_xattr(value, size); 2236 kfree(value); 2237 return acl; 2238 } 2239 2240 int 2241 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl) 2242 { 2243 struct inode *inode = fhp->fh_dentry->d_inode; 2244 char *name; 2245 void *value = NULL; 2246 size_t size; 2247 int error; 2248 2249 if (!IS_POSIXACL(inode) || 2250 !inode->i_op->setxattr || !inode->i_op->removexattr) 2251 return -EOPNOTSUPP; 2252 switch(type) { 2253 case ACL_TYPE_ACCESS: 2254 name = POSIX_ACL_XATTR_ACCESS; 2255 break; 2256 case ACL_TYPE_DEFAULT: 2257 name = POSIX_ACL_XATTR_DEFAULT; 2258 break; 2259 default: 2260 return -EOPNOTSUPP; 2261 } 2262 2263 if (acl && acl->a_count) { 2264 size = posix_acl_xattr_size(acl->a_count); 2265 value = kmalloc(size, GFP_KERNEL); 2266 if (!value) 2267 return -ENOMEM; 2268 error = posix_acl_to_xattr(acl, value, size); 2269 if (error < 0) 2270 goto getout; 2271 size = error; 2272 } else 2273 size = 0; 2274 2275 error = mnt_want_write(fhp->fh_export->ex_path.mnt); 2276 if (error) 2277 goto getout; 2278 if (size) 2279 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0); 2280 else { 2281 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT) 2282 error = 0; 2283 else { 2284 error = vfs_removexattr(fhp->fh_dentry, name); 2285 if (error == -ENODATA) 2286 error = 0; 2287 } 2288 } 2289 mnt_drop_write(fhp->fh_export->ex_path.mnt); 2290 2291 getout: 2292 kfree(value); 2293 return error; 2294 } 2295 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 2296