1 /* 2 * File operations used by nfsd. Some of these have been ripped from 3 * other parts of the kernel because they weren't exported, others 4 * are partial duplicates with added or changed functionality. 5 * 6 * Note that several functions dget() the dentry upon which they want 7 * to act, most notably those that create directory entries. Response 8 * dentry's are dput()'d if necessary in the release callback. 9 * So if you notice code paths that apparently fail to dput() the 10 * dentry, don't worry--they have been taken care of. 11 * 12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de> 13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp> 14 */ 15 16 #include <linux/fs.h> 17 #include <linux/file.h> 18 #include <linux/splice.h> 19 #include <linux/fcntl.h> 20 #include <linux/namei.h> 21 #include <linux/delay.h> 22 #include <linux/fsnotify.h> 23 #include <linux/posix_acl_xattr.h> 24 #include <linux/xattr.h> 25 #include <linux/jhash.h> 26 #include <linux/ima.h> 27 #include <linux/slab.h> 28 #include <asm/uaccess.h> 29 #include <linux/exportfs.h> 30 #include <linux/writeback.h> 31 #include <linux/security.h> 32 33 #ifdef CONFIG_NFSD_V3 34 #include "xdr3.h" 35 #endif /* CONFIG_NFSD_V3 */ 36 37 #ifdef CONFIG_NFSD_V4 38 #include "acl.h" 39 #include "idmap.h" 40 #endif /* CONFIG_NFSD_V4 */ 41 42 #include "nfsd.h" 43 #include "vfs.h" 44 45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP 46 47 48 /* 49 * This is a cache of readahead params that help us choose the proper 50 * readahead strategy. Initially, we set all readahead parameters to 0 51 * and let the VFS handle things. 52 * If you increase the number of cached files very much, you'll need to 53 * add a hash table here. 54 */ 55 struct raparms { 56 struct raparms *p_next; 57 unsigned int p_count; 58 ino_t p_ino; 59 dev_t p_dev; 60 int p_set; 61 struct file_ra_state p_ra; 62 unsigned int p_hindex; 63 }; 64 65 struct raparm_hbucket { 66 struct raparms *pb_head; 67 spinlock_t pb_lock; 68 } ____cacheline_aligned_in_smp; 69 70 #define RAPARM_HASH_BITS 4 71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) 72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) 73 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; 74 75 /* 76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed 77 * a mount point. 78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged, 79 * or nfs_ok having possibly changed *dpp and *expp 80 */ 81 int 82 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 83 struct svc_export **expp) 84 { 85 struct svc_export *exp = *expp, *exp2 = NULL; 86 struct dentry *dentry = *dpp; 87 struct path path = {.mnt = mntget(exp->ex_path.mnt), 88 .dentry = dget(dentry)}; 89 int err = 0; 90 91 err = follow_down(&path); 92 if (err < 0) 93 goto out; 94 95 exp2 = rqst_exp_get_by_name(rqstp, &path); 96 if (IS_ERR(exp2)) { 97 err = PTR_ERR(exp2); 98 /* 99 * We normally allow NFS clients to continue 100 * "underneath" a mountpoint that is not exported. 101 * The exception is V4ROOT, where no traversal is ever 102 * allowed without an explicit export of the new 103 * directory. 104 */ 105 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT)) 106 err = 0; 107 path_put(&path); 108 goto out; 109 } 110 if (nfsd_v4client(rqstp) || 111 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) { 112 /* successfully crossed mount point */ 113 /* 114 * This is subtle: path.dentry is *not* on path.mnt 115 * at this point. The only reason we are safe is that 116 * original mnt is pinned down by exp, so we should 117 * put path *before* putting exp 118 */ 119 *dpp = path.dentry; 120 path.dentry = dentry; 121 *expp = exp2; 122 exp2 = exp; 123 } 124 path_put(&path); 125 exp_put(exp2); 126 out: 127 return err; 128 } 129 130 static void follow_to_parent(struct path *path) 131 { 132 struct dentry *dp; 133 134 while (path->dentry == path->mnt->mnt_root && follow_up(path)) 135 ; 136 dp = dget_parent(path->dentry); 137 dput(path->dentry); 138 path->dentry = dp; 139 } 140 141 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp) 142 { 143 struct svc_export *exp2; 144 struct path path = {.mnt = mntget((*exp)->ex_path.mnt), 145 .dentry = dget(dparent)}; 146 147 follow_to_parent(&path); 148 149 exp2 = rqst_exp_parent(rqstp, &path); 150 if (PTR_ERR(exp2) == -ENOENT) { 151 *dentryp = dget(dparent); 152 } else if (IS_ERR(exp2)) { 153 path_put(&path); 154 return PTR_ERR(exp2); 155 } else { 156 *dentryp = dget(path.dentry); 157 exp_put(*exp); 158 *exp = exp2; 159 } 160 path_put(&path); 161 return 0; 162 } 163 164 /* 165 * For nfsd purposes, we treat V4ROOT exports as though there was an 166 * export at *every* directory. 167 */ 168 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp) 169 { 170 if (d_mountpoint(dentry)) 171 return 1; 172 if (nfsd4_is_junction(dentry)) 173 return 1; 174 if (!(exp->ex_flags & NFSEXP_V4ROOT)) 175 return 0; 176 return dentry->d_inode != NULL; 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 int host_err; 188 189 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); 190 191 dparent = fhp->fh_dentry; 192 exp = fhp->fh_export; 193 exp_get(exp); 194 195 /* Lookup the name, but don't follow links */ 196 if (isdotent(name, len)) { 197 if (len==1) 198 dentry = dget(dparent); 199 else if (dparent != exp->ex_path.dentry) 200 dentry = dget_parent(dparent); 201 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp)) 202 dentry = dget(dparent); /* .. == . just like at / */ 203 else { 204 /* checking mountpoint crossing is very different when stepping up */ 205 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry); 206 if (host_err) 207 goto out_nfserr; 208 } 209 } else { 210 /* 211 * In the nfsd4_open() case, this may be held across 212 * subsequent open and delegation acquisition which may 213 * need to take the child's i_mutex: 214 */ 215 fh_lock_nested(fhp, I_MUTEX_PARENT); 216 dentry = lookup_one_len(name, dparent, len); 217 host_err = PTR_ERR(dentry); 218 if (IS_ERR(dentry)) 219 goto out_nfserr; 220 /* 221 * check if we have crossed a mount point ... 222 */ 223 if (nfsd_mountpoint(dentry, exp)) { 224 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { 225 dput(dentry); 226 goto out_nfserr; 227 } 228 } 229 } 230 *dentry_ret = dentry; 231 *exp_ret = exp; 232 return 0; 233 234 out_nfserr: 235 exp_put(exp); 236 return nfserrno(host_err); 237 } 238 239 /* 240 * Look up one component of a pathname. 241 * N.B. After this call _both_ fhp and resfh need an fh_put 242 * 243 * If the lookup would cross a mountpoint, and the mounted filesystem 244 * is exported to the client with NFSEXP_NOHIDE, then the lookup is 245 * accepted as it stands and the mounted directory is 246 * returned. Otherwise the covered directory is returned. 247 * NOTE: this mountpoint crossing is not supported properly by all 248 * clients and is explicitly disallowed for NFSv3 249 * NeilBrown <neilb@cse.unsw.edu.au> 250 */ 251 __be32 252 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, 253 unsigned int len, struct svc_fh *resfh) 254 { 255 struct svc_export *exp; 256 struct dentry *dentry; 257 __be32 err; 258 259 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 260 if (err) 261 return err; 262 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry); 263 if (err) 264 return err; 265 err = check_nfsd_access(exp, rqstp); 266 if (err) 267 goto out; 268 /* 269 * Note: we compose the file handle now, but as the 270 * dentry may be negative, it may need to be updated. 271 */ 272 err = fh_compose(resfh, exp, dentry, fhp); 273 if (!err && !dentry->d_inode) 274 err = nfserr_noent; 275 out: 276 dput(dentry); 277 exp_put(exp); 278 return err; 279 } 280 281 /* 282 * Commit metadata changes to stable storage. 283 */ 284 static int 285 commit_metadata(struct svc_fh *fhp) 286 { 287 struct inode *inode = fhp->fh_dentry->d_inode; 288 const struct export_operations *export_ops = inode->i_sb->s_export_op; 289 290 if (!EX_ISSYNC(fhp->fh_export)) 291 return 0; 292 293 if (export_ops->commit_metadata) 294 return export_ops->commit_metadata(inode); 295 return sync_inode_metadata(inode, 1); 296 } 297 298 /* 299 * Go over the attributes and take care of the small differences between 300 * NFS semantics and what Linux expects. 301 */ 302 static void 303 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap) 304 { 305 /* 306 * NFSv2 does not differentiate between "set-[ac]time-to-now" 307 * which only requires access, and "set-[ac]time-to-X" which 308 * requires ownership. 309 * So if it looks like it might be "set both to the same time which 310 * is close to now", and if inode_change_ok fails, then we 311 * convert to "set to now" instead of "set to explicit time" 312 * 313 * We only call inode_change_ok as the last test as technically 314 * it is not an interface that we should be using. 315 */ 316 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) 317 #define MAX_TOUCH_TIME_ERROR (30*60) 318 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET && 319 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) { 320 /* 321 * Looks probable. 322 * 323 * Now just make sure time is in the right ballpark. 324 * Solaris, at least, doesn't seem to care what the time 325 * request is. We require it be within 30 minutes of now. 326 */ 327 time_t delta = iap->ia_atime.tv_sec - get_seconds(); 328 if (delta < 0) 329 delta = -delta; 330 if (delta < MAX_TOUCH_TIME_ERROR && 331 inode_change_ok(inode, iap) != 0) { 332 /* 333 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME. 334 * This will cause notify_change to set these times 335 * to "now" 336 */ 337 iap->ia_valid &= ~BOTH_TIME_SET; 338 } 339 } 340 341 /* sanitize the mode change */ 342 if (iap->ia_valid & ATTR_MODE) { 343 iap->ia_mode &= S_IALLUGO; 344 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO); 345 } 346 347 /* Revoke setuid/setgid on chown */ 348 if (!S_ISDIR(inode->i_mode) && 349 ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) { 350 iap->ia_valid |= ATTR_KILL_PRIV; 351 if (iap->ia_valid & ATTR_MODE) { 352 /* we're setting mode too, just clear the s*id bits */ 353 iap->ia_mode &= ~S_ISUID; 354 if (iap->ia_mode & S_IXGRP) 355 iap->ia_mode &= ~S_ISGID; 356 } else { 357 /* set ATTR_KILL_* bits and let VFS handle it */ 358 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID); 359 } 360 } 361 } 362 363 static __be32 364 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp, 365 struct iattr *iap) 366 { 367 struct inode *inode = fhp->fh_dentry->d_inode; 368 int host_err; 369 370 if (iap->ia_size < inode->i_size) { 371 __be32 err; 372 373 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 374 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE); 375 if (err) 376 return err; 377 } 378 379 host_err = get_write_access(inode); 380 if (host_err) 381 goto out_nfserrno; 382 383 host_err = locks_verify_truncate(inode, NULL, iap->ia_size); 384 if (host_err) 385 goto out_put_write_access; 386 return 0; 387 388 out_put_write_access: 389 put_write_access(inode); 390 out_nfserrno: 391 return nfserrno(host_err); 392 } 393 394 /* 395 * Set various file attributes. After this call fhp needs an fh_put. 396 */ 397 __be32 398 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, 399 int check_guard, time_t guardtime) 400 { 401 struct dentry *dentry; 402 struct inode *inode; 403 int accmode = NFSD_MAY_SATTR; 404 umode_t ftype = 0; 405 __be32 err; 406 int host_err; 407 bool get_write_count; 408 int size_change = 0; 409 410 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) 411 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE; 412 if (iap->ia_valid & ATTR_SIZE) 413 ftype = S_IFREG; 414 415 /* Callers that do fh_verify should do the fh_want_write: */ 416 get_write_count = !fhp->fh_dentry; 417 418 /* Get inode */ 419 err = fh_verify(rqstp, fhp, ftype, accmode); 420 if (err) 421 goto out; 422 if (get_write_count) { 423 host_err = fh_want_write(fhp); 424 if (host_err) 425 return nfserrno(host_err); 426 } 427 428 dentry = fhp->fh_dentry; 429 inode = dentry->d_inode; 430 431 /* Ignore any mode updates on symlinks */ 432 if (S_ISLNK(inode->i_mode)) 433 iap->ia_valid &= ~ATTR_MODE; 434 435 if (!iap->ia_valid) 436 goto out; 437 438 nfsd_sanitize_attrs(inode, iap); 439 440 /* 441 * The size case is special, it changes the file in addition to the 442 * attributes. 443 */ 444 if (iap->ia_valid & ATTR_SIZE) { 445 err = nfsd_get_write_access(rqstp, fhp, iap); 446 if (err) 447 goto out; 448 size_change = 1; 449 } 450 451 iap->ia_valid |= ATTR_CTIME; 452 453 if (check_guard && guardtime != inode->i_ctime.tv_sec) { 454 err = nfserr_notsync; 455 goto out_put_write_access; 456 } 457 458 fh_lock(fhp); 459 host_err = notify_change(dentry, iap, NULL); 460 fh_unlock(fhp); 461 err = nfserrno(host_err); 462 463 out_put_write_access: 464 if (size_change) 465 put_write_access(inode); 466 if (!err) 467 commit_metadata(fhp); 468 out: 469 return err; 470 } 471 472 #if defined(CONFIG_NFSD_V4) 473 /* 474 * NFS junction information is stored in an extended attribute. 475 */ 476 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs" 477 478 /** 479 * nfsd4_is_junction - Test if an object could be an NFS junction 480 * 481 * @dentry: object to test 482 * 483 * Returns 1 if "dentry" appears to contain NFS junction information. 484 * Otherwise 0 is returned. 485 */ 486 int nfsd4_is_junction(struct dentry *dentry) 487 { 488 struct inode *inode = dentry->d_inode; 489 490 if (inode == NULL) 491 return 0; 492 if (inode->i_mode & S_IXUGO) 493 return 0; 494 if (!(inode->i_mode & S_ISVTX)) 495 return 0; 496 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0) 497 return 0; 498 return 1; 499 } 500 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL 501 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, 502 struct xdr_netobj *label) 503 { 504 __be32 error; 505 int host_error; 506 struct dentry *dentry; 507 508 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR); 509 if (error) 510 return error; 511 512 dentry = fhp->fh_dentry; 513 514 mutex_lock(&dentry->d_inode->i_mutex); 515 host_error = security_inode_setsecctx(dentry, label->data, label->len); 516 mutex_unlock(&dentry->d_inode->i_mutex); 517 return nfserrno(host_error); 518 } 519 #else 520 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, 521 struct xdr_netobj *label) 522 { 523 return nfserr_notsupp; 524 } 525 #endif 526 527 #endif /* defined(CONFIG_NFSD_V4) */ 528 529 #ifdef CONFIG_NFSD_V3 530 /* 531 * Check server access rights to a file system object 532 */ 533 struct accessmap { 534 u32 access; 535 int how; 536 }; 537 static struct accessmap nfs3_regaccess[] = { 538 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 539 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 540 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, 541 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, 542 543 { 0, 0 } 544 }; 545 546 static struct accessmap nfs3_diraccess[] = { 547 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 548 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, 549 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, 550 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, 551 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, 552 553 { 0, 0 } 554 }; 555 556 static struct accessmap nfs3_anyaccess[] = { 557 /* Some clients - Solaris 2.6 at least, make an access call 558 * to the server to check for access for things like /dev/null 559 * (which really, the server doesn't care about). So 560 * We provide simple access checking for them, looking 561 * mainly at mode bits, and we make sure to ignore read-only 562 * filesystem checks 563 */ 564 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 565 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 566 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 567 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 568 569 { 0, 0 } 570 }; 571 572 __be32 573 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) 574 { 575 struct accessmap *map; 576 struct svc_export *export; 577 struct dentry *dentry; 578 u32 query, result = 0, sresult = 0; 579 __be32 error; 580 581 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); 582 if (error) 583 goto out; 584 585 export = fhp->fh_export; 586 dentry = fhp->fh_dentry; 587 588 if (S_ISREG(dentry->d_inode->i_mode)) 589 map = nfs3_regaccess; 590 else if (S_ISDIR(dentry->d_inode->i_mode)) 591 map = nfs3_diraccess; 592 else 593 map = nfs3_anyaccess; 594 595 596 query = *access; 597 for (; map->access; map++) { 598 if (map->access & query) { 599 __be32 err2; 600 601 sresult |= map->access; 602 603 err2 = nfsd_permission(rqstp, export, dentry, map->how); 604 switch (err2) { 605 case nfs_ok: 606 result |= map->access; 607 break; 608 609 /* the following error codes just mean the access was not allowed, 610 * rather than an error occurred */ 611 case nfserr_rofs: 612 case nfserr_acces: 613 case nfserr_perm: 614 /* simply don't "or" in the access bit. */ 615 break; 616 default: 617 error = err2; 618 goto out; 619 } 620 } 621 } 622 *access = result; 623 if (supported) 624 *supported = sresult; 625 626 out: 627 return error; 628 } 629 #endif /* CONFIG_NFSD_V3 */ 630 631 static int nfsd_open_break_lease(struct inode *inode, int access) 632 { 633 unsigned int mode; 634 635 if (access & NFSD_MAY_NOT_BREAK_LEASE) 636 return 0; 637 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY; 638 return break_lease(inode, mode | O_NONBLOCK); 639 } 640 641 /* 642 * Open an existing file or directory. 643 * The may_flags argument indicates the type of open (read/write/lock) 644 * and additional flags. 645 * N.B. After this call fhp needs an fh_put 646 */ 647 __be32 648 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, 649 int may_flags, struct file **filp) 650 { 651 struct path path; 652 struct inode *inode; 653 int flags = O_RDONLY|O_LARGEFILE; 654 __be32 err; 655 int host_err = 0; 656 657 validate_process_creds(); 658 659 /* 660 * If we get here, then the client has already done an "open", 661 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 662 * in case a chmod has now revoked permission. 663 * 664 * Arguably we should also allow the owner override for 665 * directories, but we never have and it doesn't seem to have 666 * caused anyone a problem. If we were to change this, note 667 * also that our filldir callbacks would need a variant of 668 * lookup_one_len that doesn't check permissions. 669 */ 670 if (type == S_IFREG) 671 may_flags |= NFSD_MAY_OWNER_OVERRIDE; 672 err = fh_verify(rqstp, fhp, type, may_flags); 673 if (err) 674 goto out; 675 676 path.mnt = fhp->fh_export->ex_path.mnt; 677 path.dentry = fhp->fh_dentry; 678 inode = path.dentry->d_inode; 679 680 /* Disallow write access to files with the append-only bit set 681 * or any access when mandatory locking enabled 682 */ 683 err = nfserr_perm; 684 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE)) 685 goto out; 686 /* 687 * We must ignore files (but only files) which might have mandatory 688 * locks on them because there is no way to know if the accesser has 689 * the lock. 690 */ 691 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) 692 goto out; 693 694 if (!inode->i_fop) 695 goto out; 696 697 host_err = nfsd_open_break_lease(inode, may_flags); 698 if (host_err) /* NOMEM or WOULDBLOCK */ 699 goto out_nfserr; 700 701 if (may_flags & NFSD_MAY_WRITE) { 702 if (may_flags & NFSD_MAY_READ) 703 flags = O_RDWR|O_LARGEFILE; 704 else 705 flags = O_WRONLY|O_LARGEFILE; 706 } 707 *filp = dentry_open(&path, flags, current_cred()); 708 if (IS_ERR(*filp)) { 709 host_err = PTR_ERR(*filp); 710 *filp = NULL; 711 } else { 712 host_err = ima_file_check(*filp, may_flags); 713 714 if (may_flags & NFSD_MAY_64BIT_COOKIE) 715 (*filp)->f_mode |= FMODE_64BITHASH; 716 else 717 (*filp)->f_mode |= FMODE_32BITHASH; 718 } 719 720 out_nfserr: 721 err = nfserrno(host_err); 722 out: 723 validate_process_creds(); 724 return err; 725 } 726 727 /* 728 * Close a file. 729 */ 730 void 731 nfsd_close(struct file *filp) 732 { 733 fput(filp); 734 } 735 736 /* 737 * Obtain the readahead parameters for the file 738 * specified by (dev, ino). 739 */ 740 741 static inline struct raparms * 742 nfsd_get_raparms(dev_t dev, ino_t ino) 743 { 744 struct raparms *ra, **rap, **frap = NULL; 745 int depth = 0; 746 unsigned int hash; 747 struct raparm_hbucket *rab; 748 749 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 750 rab = &raparm_hash[hash]; 751 752 spin_lock(&rab->pb_lock); 753 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 754 if (ra->p_ino == ino && ra->p_dev == dev) 755 goto found; 756 depth++; 757 if (ra->p_count == 0) 758 frap = rap; 759 } 760 depth = nfsdstats.ra_size; 761 if (!frap) { 762 spin_unlock(&rab->pb_lock); 763 return NULL; 764 } 765 rap = frap; 766 ra = *frap; 767 ra->p_dev = dev; 768 ra->p_ino = ino; 769 ra->p_set = 0; 770 ra->p_hindex = hash; 771 found: 772 if (rap != &rab->pb_head) { 773 *rap = ra->p_next; 774 ra->p_next = rab->pb_head; 775 rab->pb_head = ra; 776 } 777 ra->p_count++; 778 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 779 spin_unlock(&rab->pb_lock); 780 return ra; 781 } 782 783 /* 784 * Grab and keep cached pages associated with a file in the svc_rqst 785 * so that they can be passed to the network sendmsg/sendpage routines 786 * directly. They will be released after the sending has completed. 787 */ 788 static int 789 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 790 struct splice_desc *sd) 791 { 792 struct svc_rqst *rqstp = sd->u.data; 793 struct page **pp = rqstp->rq_next_page; 794 struct page *page = buf->page; 795 size_t size; 796 797 size = sd->len; 798 799 if (rqstp->rq_res.page_len == 0) { 800 get_page(page); 801 put_page(*rqstp->rq_next_page); 802 *(rqstp->rq_next_page++) = page; 803 rqstp->rq_res.page_base = buf->offset; 804 rqstp->rq_res.page_len = size; 805 } else if (page != pp[-1]) { 806 get_page(page); 807 if (*rqstp->rq_next_page) 808 put_page(*rqstp->rq_next_page); 809 *(rqstp->rq_next_page++) = page; 810 rqstp->rq_res.page_len += size; 811 } else 812 rqstp->rq_res.page_len += size; 813 814 return size; 815 } 816 817 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 818 struct splice_desc *sd) 819 { 820 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 821 } 822 823 static __be32 824 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 825 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 826 { 827 mm_segment_t oldfs; 828 __be32 err; 829 int host_err; 830 831 err = nfserr_perm; 832 833 if (file->f_op->splice_read && rqstp->rq_splice_ok) { 834 struct splice_desc sd = { 835 .len = 0, 836 .total_len = *count, 837 .pos = offset, 838 .u.data = rqstp, 839 }; 840 841 rqstp->rq_next_page = rqstp->rq_respages + 1; 842 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 843 } else { 844 oldfs = get_fs(); 845 set_fs(KERNEL_DS); 846 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 847 set_fs(oldfs); 848 } 849 850 if (host_err >= 0) { 851 nfsdstats.io_read += host_err; 852 *count = host_err; 853 err = 0; 854 fsnotify_access(file); 855 } else 856 err = nfserrno(host_err); 857 return err; 858 } 859 860 static void kill_suid(struct dentry *dentry) 861 { 862 struct iattr ia; 863 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 864 865 mutex_lock(&dentry->d_inode->i_mutex); 866 /* 867 * Note we call this on write, so notify_change will not 868 * encounter any conflicting delegations: 869 */ 870 notify_change(dentry, &ia, NULL); 871 mutex_unlock(&dentry->d_inode->i_mutex); 872 } 873 874 /* 875 * Gathered writes: If another process is currently writing to the file, 876 * there's a high chance this is another nfsd (triggered by a bulk write 877 * from a client's biod). Rather than syncing the file with each write 878 * request, we sleep for 10 msec. 879 * 880 * I don't know if this roughly approximates C. Juszak's idea of 881 * gathered writes, but it's a nice and simple solution (IMHO), and it 882 * seems to work:-) 883 * 884 * Note: we do this only in the NFSv2 case, since v3 and higher have a 885 * better tool (separate unstable writes and commits) for solving this 886 * problem. 887 */ 888 static int wait_for_concurrent_writes(struct file *file) 889 { 890 struct inode *inode = file_inode(file); 891 static ino_t last_ino; 892 static dev_t last_dev; 893 int err = 0; 894 895 if (atomic_read(&inode->i_writecount) > 1 896 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 897 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 898 msleep(10); 899 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 900 } 901 902 if (inode->i_state & I_DIRTY) { 903 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 904 err = vfs_fsync(file, 0); 905 } 906 last_ino = inode->i_ino; 907 last_dev = inode->i_sb->s_dev; 908 return err; 909 } 910 911 static __be32 912 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 913 loff_t offset, struct kvec *vec, int vlen, 914 unsigned long *cnt, int *stablep) 915 { 916 struct svc_export *exp; 917 struct dentry *dentry; 918 struct inode *inode; 919 mm_segment_t oldfs; 920 __be32 err = 0; 921 int host_err; 922 int stable = *stablep; 923 int use_wgather; 924 loff_t pos = offset; 925 926 dentry = file->f_path.dentry; 927 inode = dentry->d_inode; 928 exp = fhp->fh_export; 929 930 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 931 932 if (!EX_ISSYNC(exp)) 933 stable = 0; 934 935 /* Write the data. */ 936 oldfs = get_fs(); set_fs(KERNEL_DS); 937 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos); 938 set_fs(oldfs); 939 if (host_err < 0) 940 goto out_nfserr; 941 *cnt = host_err; 942 nfsdstats.io_write += host_err; 943 fsnotify_modify(file); 944 945 /* clear setuid/setgid flag after write */ 946 if (inode->i_mode & (S_ISUID | S_ISGID)) 947 kill_suid(dentry); 948 949 if (stable) { 950 if (use_wgather) 951 host_err = wait_for_concurrent_writes(file); 952 else 953 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0); 954 } 955 956 out_nfserr: 957 dprintk("nfsd: write complete host_err=%d\n", host_err); 958 if (host_err >= 0) 959 err = 0; 960 else 961 err = nfserrno(host_err); 962 return err; 963 } 964 965 /* 966 * Read data from a file. count must contain the requested read count 967 * on entry. On return, *count contains the number of bytes actually read. 968 * N.B. After this call fhp needs an fh_put 969 */ 970 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, 971 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 972 { 973 struct file *file; 974 struct inode *inode; 975 struct raparms *ra; 976 __be32 err; 977 978 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); 979 if (err) 980 return err; 981 982 inode = file_inode(file); 983 984 /* Get readahead parameters */ 985 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 986 987 if (ra && ra->p_set) 988 file->f_ra = ra->p_ra; 989 990 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 991 992 /* Write back readahead params */ 993 if (ra) { 994 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 995 spin_lock(&rab->pb_lock); 996 ra->p_ra = file->f_ra; 997 ra->p_set = 1; 998 ra->p_count--; 999 spin_unlock(&rab->pb_lock); 1000 } 1001 1002 nfsd_close(file); 1003 return err; 1004 } 1005 1006 /* As above, but use the provided file descriptor. */ 1007 __be32 1008 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1009 loff_t offset, struct kvec *vec, int vlen, 1010 unsigned long *count) 1011 { 1012 __be32 err; 1013 1014 if (file) { 1015 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1016 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE); 1017 if (err) 1018 goto out; 1019 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count); 1020 } else /* Note file may still be NULL in NFSv4 special stateid case: */ 1021 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count); 1022 out: 1023 return err; 1024 } 1025 1026 /* 1027 * Write data to a file. 1028 * The stable flag requests synchronous writes. 1029 * N.B. After this call fhp needs an fh_put 1030 */ 1031 __be32 1032 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1033 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1034 int *stablep) 1035 { 1036 __be32 err = 0; 1037 1038 if (file) { 1039 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1040 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1041 if (err) 1042 goto out; 1043 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1044 stablep); 1045 } else { 1046 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1047 if (err) 1048 goto out; 1049 1050 if (cnt) 1051 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1052 cnt, stablep); 1053 nfsd_close(file); 1054 } 1055 out: 1056 return err; 1057 } 1058 1059 #ifdef CONFIG_NFSD_V3 1060 /* 1061 * Commit all pending writes to stable storage. 1062 * 1063 * Note: we only guarantee that data that lies within the range specified 1064 * by the 'offset' and 'count' parameters will be synced. 1065 * 1066 * Unfortunately we cannot lock the file to make sure we return full WCC 1067 * data to the client, as locking happens lower down in the filesystem. 1068 */ 1069 __be32 1070 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1071 loff_t offset, unsigned long count) 1072 { 1073 struct file *file; 1074 loff_t end = LLONG_MAX; 1075 __be32 err = nfserr_inval; 1076 1077 if (offset < 0) 1078 goto out; 1079 if (count != 0) { 1080 end = offset + (loff_t)count - 1; 1081 if (end < offset) 1082 goto out; 1083 } 1084 1085 err = nfsd_open(rqstp, fhp, S_IFREG, 1086 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file); 1087 if (err) 1088 goto out; 1089 if (EX_ISSYNC(fhp->fh_export)) { 1090 int err2 = vfs_fsync_range(file, offset, end, 0); 1091 1092 if (err2 != -EINVAL) 1093 err = nfserrno(err2); 1094 else 1095 err = nfserr_notsupp; 1096 } 1097 1098 nfsd_close(file); 1099 out: 1100 return err; 1101 } 1102 #endif /* CONFIG_NFSD_V3 */ 1103 1104 static __be32 1105 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1106 struct iattr *iap) 1107 { 1108 /* 1109 * Mode has already been set earlier in create: 1110 */ 1111 iap->ia_valid &= ~ATTR_MODE; 1112 /* 1113 * Setting uid/gid works only for root. Irix appears to 1114 * send along the gid on create when it tries to implement 1115 * setgid directories via NFS: 1116 */ 1117 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) 1118 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1119 if (iap->ia_valid) 1120 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1121 return 0; 1122 } 1123 1124 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1125 * setting size to 0 may fail for some specific file systems by the permission 1126 * checking which requires WRITE permission but the mode is 000. 1127 * we ignore the resizing(to 0) on the just new created file, since the size is 1128 * 0 after file created. 1129 * 1130 * call this only after vfs_create() is called. 1131 * */ 1132 static void 1133 nfsd_check_ignore_resizing(struct iattr *iap) 1134 { 1135 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1136 iap->ia_valid &= ~ATTR_SIZE; 1137 } 1138 1139 /* 1140 * Create a file (regular, directory, device, fifo); UNIX sockets 1141 * not yet implemented. 1142 * If the response fh has been verified, the parent directory should 1143 * already be locked. Note that the parent directory is left locked. 1144 * 1145 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1146 */ 1147 __be32 1148 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1149 char *fname, int flen, struct iattr *iap, 1150 int type, dev_t rdev, struct svc_fh *resfhp) 1151 { 1152 struct dentry *dentry, *dchild = NULL; 1153 struct inode *dirp; 1154 __be32 err; 1155 __be32 err2; 1156 int host_err; 1157 1158 err = nfserr_perm; 1159 if (!flen) 1160 goto out; 1161 err = nfserr_exist; 1162 if (isdotent(fname, flen)) 1163 goto out; 1164 1165 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1166 if (err) 1167 goto out; 1168 1169 dentry = fhp->fh_dentry; 1170 dirp = dentry->d_inode; 1171 1172 err = nfserr_notdir; 1173 if (!dirp->i_op->lookup) 1174 goto out; 1175 /* 1176 * Check whether the response file handle has been verified yet. 1177 * If it has, the parent directory should already be locked. 1178 */ 1179 if (!resfhp->fh_dentry) { 1180 host_err = fh_want_write(fhp); 1181 if (host_err) 1182 goto out_nfserr; 1183 1184 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1185 fh_lock_nested(fhp, I_MUTEX_PARENT); 1186 dchild = lookup_one_len(fname, dentry, flen); 1187 host_err = PTR_ERR(dchild); 1188 if (IS_ERR(dchild)) 1189 goto out_nfserr; 1190 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1191 if (err) 1192 goto out; 1193 } else { 1194 /* called from nfsd_proc_create */ 1195 dchild = dget(resfhp->fh_dentry); 1196 if (!fhp->fh_locked) { 1197 /* not actually possible */ 1198 printk(KERN_ERR 1199 "nfsd_create: parent %pd2 not locked!\n", 1200 dentry); 1201 err = nfserr_io; 1202 goto out; 1203 } 1204 } 1205 /* 1206 * Make sure the child dentry is still negative ... 1207 */ 1208 err = nfserr_exist; 1209 if (dchild->d_inode) { 1210 dprintk("nfsd_create: dentry %pd/%pd not negative!\n", 1211 dentry, dchild); 1212 goto out; 1213 } 1214 1215 if (!(iap->ia_valid & ATTR_MODE)) 1216 iap->ia_mode = 0; 1217 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1218 1219 err = nfserr_inval; 1220 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1221 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1222 type); 1223 goto out; 1224 } 1225 1226 /* 1227 * Get the dir op function pointer. 1228 */ 1229 err = 0; 1230 host_err = 0; 1231 switch (type) { 1232 case S_IFREG: 1233 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1234 if (!host_err) 1235 nfsd_check_ignore_resizing(iap); 1236 break; 1237 case S_IFDIR: 1238 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1239 break; 1240 case S_IFCHR: 1241 case S_IFBLK: 1242 case S_IFIFO: 1243 case S_IFSOCK: 1244 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1245 break; 1246 } 1247 if (host_err < 0) 1248 goto out_nfserr; 1249 1250 err = nfsd_create_setattr(rqstp, resfhp, iap); 1251 1252 /* 1253 * nfsd_setattr already committed the child. Transactional filesystems 1254 * had a chance to commit changes for both parent and child 1255 * simultaneously making the following commit_metadata a noop. 1256 */ 1257 err2 = nfserrno(commit_metadata(fhp)); 1258 if (err2) 1259 err = err2; 1260 /* 1261 * Update the file handle to get the new inode info. 1262 */ 1263 if (!err) 1264 err = fh_update(resfhp); 1265 out: 1266 if (dchild && !IS_ERR(dchild)) 1267 dput(dchild); 1268 return err; 1269 1270 out_nfserr: 1271 err = nfserrno(host_err); 1272 goto out; 1273 } 1274 1275 #ifdef CONFIG_NFSD_V3 1276 1277 static inline int nfsd_create_is_exclusive(int createmode) 1278 { 1279 return createmode == NFS3_CREATE_EXCLUSIVE 1280 || createmode == NFS4_CREATE_EXCLUSIVE4_1; 1281 } 1282 1283 /* 1284 * NFSv3 and NFSv4 version of nfsd_create 1285 */ 1286 __be32 1287 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1288 char *fname, int flen, struct iattr *iap, 1289 struct svc_fh *resfhp, int createmode, u32 *verifier, 1290 bool *truncp, bool *created) 1291 { 1292 struct dentry *dentry, *dchild = NULL; 1293 struct inode *dirp; 1294 __be32 err; 1295 int host_err; 1296 __u32 v_mtime=0, v_atime=0; 1297 1298 err = nfserr_perm; 1299 if (!flen) 1300 goto out; 1301 err = nfserr_exist; 1302 if (isdotent(fname, flen)) 1303 goto out; 1304 if (!(iap->ia_valid & ATTR_MODE)) 1305 iap->ia_mode = 0; 1306 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 1307 if (err) 1308 goto out; 1309 1310 dentry = fhp->fh_dentry; 1311 dirp = dentry->d_inode; 1312 1313 /* Get all the sanity checks out of the way before 1314 * we lock the parent. */ 1315 err = nfserr_notdir; 1316 if (!dirp->i_op->lookup) 1317 goto out; 1318 1319 host_err = fh_want_write(fhp); 1320 if (host_err) 1321 goto out_nfserr; 1322 1323 fh_lock_nested(fhp, I_MUTEX_PARENT); 1324 1325 /* 1326 * Compose the response file handle. 1327 */ 1328 dchild = lookup_one_len(fname, dentry, flen); 1329 host_err = PTR_ERR(dchild); 1330 if (IS_ERR(dchild)) 1331 goto out_nfserr; 1332 1333 /* If file doesn't exist, check for permissions to create one */ 1334 if (!dchild->d_inode) { 1335 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1336 if (err) 1337 goto out; 1338 } 1339 1340 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1341 if (err) 1342 goto out; 1343 1344 if (nfsd_create_is_exclusive(createmode)) { 1345 /* solaris7 gets confused (bugid 4218508) if these have 1346 * the high bit set, so just clear the high bits. If this is 1347 * ever changed to use different attrs for storing the 1348 * verifier, then do_open_lookup() will also need to be fixed 1349 * accordingly. 1350 */ 1351 v_mtime = verifier[0]&0x7fffffff; 1352 v_atime = verifier[1]&0x7fffffff; 1353 } 1354 1355 if (dchild->d_inode) { 1356 err = 0; 1357 1358 switch (createmode) { 1359 case NFS3_CREATE_UNCHECKED: 1360 if (! S_ISREG(dchild->d_inode->i_mode)) 1361 goto out; 1362 else if (truncp) { 1363 /* in nfsv4, we need to treat this case a little 1364 * differently. we don't want to truncate the 1365 * file now; this would be wrong if the OPEN 1366 * fails for some other reason. furthermore, 1367 * if the size is nonzero, we should ignore it 1368 * according to spec! 1369 */ 1370 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1371 } 1372 else { 1373 iap->ia_valid &= ATTR_SIZE; 1374 goto set_attr; 1375 } 1376 break; 1377 case NFS3_CREATE_EXCLUSIVE: 1378 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1379 && dchild->d_inode->i_atime.tv_sec == v_atime 1380 && dchild->d_inode->i_size == 0 ) { 1381 if (created) 1382 *created = 1; 1383 break; 1384 } 1385 case NFS4_CREATE_EXCLUSIVE4_1: 1386 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1387 && dchild->d_inode->i_atime.tv_sec == v_atime 1388 && dchild->d_inode->i_size == 0 ) { 1389 if (created) 1390 *created = 1; 1391 goto set_attr; 1392 } 1393 /* fallthru */ 1394 case NFS3_CREATE_GUARDED: 1395 err = nfserr_exist; 1396 } 1397 fh_drop_write(fhp); 1398 goto out; 1399 } 1400 1401 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1402 if (host_err < 0) { 1403 fh_drop_write(fhp); 1404 goto out_nfserr; 1405 } 1406 if (created) 1407 *created = 1; 1408 1409 nfsd_check_ignore_resizing(iap); 1410 1411 if (nfsd_create_is_exclusive(createmode)) { 1412 /* Cram the verifier into atime/mtime */ 1413 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1414 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1415 /* XXX someone who knows this better please fix it for nsec */ 1416 iap->ia_mtime.tv_sec = v_mtime; 1417 iap->ia_atime.tv_sec = v_atime; 1418 iap->ia_mtime.tv_nsec = 0; 1419 iap->ia_atime.tv_nsec = 0; 1420 } 1421 1422 set_attr: 1423 err = nfsd_create_setattr(rqstp, resfhp, iap); 1424 1425 /* 1426 * nfsd_setattr already committed the child (and possibly also the parent). 1427 */ 1428 if (!err) 1429 err = nfserrno(commit_metadata(fhp)); 1430 1431 /* 1432 * Update the filehandle to get the new inode info. 1433 */ 1434 if (!err) 1435 err = fh_update(resfhp); 1436 1437 out: 1438 fh_unlock(fhp); 1439 if (dchild && !IS_ERR(dchild)) 1440 dput(dchild); 1441 fh_drop_write(fhp); 1442 return err; 1443 1444 out_nfserr: 1445 err = nfserrno(host_err); 1446 goto out; 1447 } 1448 #endif /* CONFIG_NFSD_V3 */ 1449 1450 /* 1451 * Read a symlink. On entry, *lenp must contain the maximum path length that 1452 * fits into the buffer. On return, it contains the true length. 1453 * N.B. After this call fhp needs an fh_put 1454 */ 1455 __be32 1456 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1457 { 1458 struct inode *inode; 1459 mm_segment_t oldfs; 1460 __be32 err; 1461 int host_err; 1462 struct path path; 1463 1464 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1465 if (err) 1466 goto out; 1467 1468 path.mnt = fhp->fh_export->ex_path.mnt; 1469 path.dentry = fhp->fh_dentry; 1470 inode = path.dentry->d_inode; 1471 1472 err = nfserr_inval; 1473 if (!inode->i_op->readlink) 1474 goto out; 1475 1476 touch_atime(&path); 1477 /* N.B. Why does this call need a get_fs()?? 1478 * Remove the set_fs and watch the fireworks:-) --okir 1479 */ 1480 1481 oldfs = get_fs(); set_fs(KERNEL_DS); 1482 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp); 1483 set_fs(oldfs); 1484 1485 if (host_err < 0) 1486 goto out_nfserr; 1487 *lenp = host_err; 1488 err = 0; 1489 out: 1490 return err; 1491 1492 out_nfserr: 1493 err = nfserrno(host_err); 1494 goto out; 1495 } 1496 1497 /* 1498 * Create a symlink and look up its inode 1499 * N.B. After this call _both_ fhp and resfhp need an fh_put 1500 */ 1501 __be32 1502 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1503 char *fname, int flen, 1504 char *path, int plen, 1505 struct svc_fh *resfhp, 1506 struct iattr *iap) 1507 { 1508 struct dentry *dentry, *dnew; 1509 __be32 err, cerr; 1510 int host_err; 1511 1512 err = nfserr_noent; 1513 if (!flen || !plen) 1514 goto out; 1515 err = nfserr_exist; 1516 if (isdotent(fname, flen)) 1517 goto out; 1518 1519 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1520 if (err) 1521 goto out; 1522 1523 host_err = fh_want_write(fhp); 1524 if (host_err) 1525 goto out_nfserr; 1526 1527 fh_lock(fhp); 1528 dentry = fhp->fh_dentry; 1529 dnew = lookup_one_len(fname, dentry, flen); 1530 host_err = PTR_ERR(dnew); 1531 if (IS_ERR(dnew)) 1532 goto out_nfserr; 1533 1534 if (unlikely(path[plen] != 0)) { 1535 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1536 if (path_alloced == NULL) 1537 host_err = -ENOMEM; 1538 else { 1539 strncpy(path_alloced, path, plen); 1540 path_alloced[plen] = 0; 1541 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); 1542 kfree(path_alloced); 1543 } 1544 } else 1545 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1546 err = nfserrno(host_err); 1547 if (!err) 1548 err = nfserrno(commit_metadata(fhp)); 1549 fh_unlock(fhp); 1550 1551 fh_drop_write(fhp); 1552 1553 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1554 dput(dnew); 1555 if (err==0) err = cerr; 1556 out: 1557 return err; 1558 1559 out_nfserr: 1560 err = nfserrno(host_err); 1561 goto out; 1562 } 1563 1564 /* 1565 * Create a hardlink 1566 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1567 */ 1568 __be32 1569 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1570 char *name, int len, struct svc_fh *tfhp) 1571 { 1572 struct dentry *ddir, *dnew, *dold; 1573 struct inode *dirp; 1574 __be32 err; 1575 int host_err; 1576 1577 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1578 if (err) 1579 goto out; 1580 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP); 1581 if (err) 1582 goto out; 1583 err = nfserr_isdir; 1584 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode)) 1585 goto out; 1586 err = nfserr_perm; 1587 if (!len) 1588 goto out; 1589 err = nfserr_exist; 1590 if (isdotent(name, len)) 1591 goto out; 1592 1593 host_err = fh_want_write(tfhp); 1594 if (host_err) { 1595 err = nfserrno(host_err); 1596 goto out; 1597 } 1598 1599 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1600 ddir = ffhp->fh_dentry; 1601 dirp = ddir->d_inode; 1602 1603 dnew = lookup_one_len(name, ddir, len); 1604 host_err = PTR_ERR(dnew); 1605 if (IS_ERR(dnew)) 1606 goto out_nfserr; 1607 1608 dold = tfhp->fh_dentry; 1609 1610 err = nfserr_noent; 1611 if (!dold->d_inode) 1612 goto out_dput; 1613 host_err = vfs_link(dold, dirp, dnew, NULL); 1614 if (!host_err) { 1615 err = nfserrno(commit_metadata(ffhp)); 1616 if (!err) 1617 err = nfserrno(commit_metadata(tfhp)); 1618 } else { 1619 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1620 err = nfserr_acces; 1621 else 1622 err = nfserrno(host_err); 1623 } 1624 out_dput: 1625 dput(dnew); 1626 out_unlock: 1627 fh_unlock(ffhp); 1628 fh_drop_write(tfhp); 1629 out: 1630 return err; 1631 1632 out_nfserr: 1633 err = nfserrno(host_err); 1634 goto out_unlock; 1635 } 1636 1637 /* 1638 * Rename a file 1639 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1640 */ 1641 __be32 1642 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1643 struct svc_fh *tfhp, char *tname, int tlen) 1644 { 1645 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1646 struct inode *fdir, *tdir; 1647 __be32 err; 1648 int host_err; 1649 1650 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1651 if (err) 1652 goto out; 1653 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1654 if (err) 1655 goto out; 1656 1657 fdentry = ffhp->fh_dentry; 1658 fdir = fdentry->d_inode; 1659 1660 tdentry = tfhp->fh_dentry; 1661 tdir = tdentry->d_inode; 1662 1663 err = nfserr_perm; 1664 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1665 goto out; 1666 1667 host_err = fh_want_write(ffhp); 1668 if (host_err) { 1669 err = nfserrno(host_err); 1670 goto out; 1671 } 1672 1673 /* cannot use fh_lock as we need deadlock protective ordering 1674 * so do it by hand */ 1675 trap = lock_rename(tdentry, fdentry); 1676 ffhp->fh_locked = tfhp->fh_locked = 1; 1677 fill_pre_wcc(ffhp); 1678 fill_pre_wcc(tfhp); 1679 1680 odentry = lookup_one_len(fname, fdentry, flen); 1681 host_err = PTR_ERR(odentry); 1682 if (IS_ERR(odentry)) 1683 goto out_nfserr; 1684 1685 host_err = -ENOENT; 1686 if (!odentry->d_inode) 1687 goto out_dput_old; 1688 host_err = -EINVAL; 1689 if (odentry == trap) 1690 goto out_dput_old; 1691 1692 ndentry = lookup_one_len(tname, tdentry, tlen); 1693 host_err = PTR_ERR(ndentry); 1694 if (IS_ERR(ndentry)) 1695 goto out_dput_old; 1696 host_err = -ENOTEMPTY; 1697 if (ndentry == trap) 1698 goto out_dput_new; 1699 1700 host_err = -EXDEV; 1701 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1702 goto out_dput_new; 1703 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry) 1704 goto out_dput_new; 1705 1706 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0); 1707 if (!host_err) { 1708 host_err = commit_metadata(tfhp); 1709 if (!host_err) 1710 host_err = commit_metadata(ffhp); 1711 } 1712 out_dput_new: 1713 dput(ndentry); 1714 out_dput_old: 1715 dput(odentry); 1716 out_nfserr: 1717 err = nfserrno(host_err); 1718 /* 1719 * We cannot rely on fh_unlock on the two filehandles, 1720 * as that would do the wrong thing if the two directories 1721 * were the same, so again we do it by hand. 1722 */ 1723 fill_post_wcc(ffhp); 1724 fill_post_wcc(tfhp); 1725 unlock_rename(tdentry, fdentry); 1726 ffhp->fh_locked = tfhp->fh_locked = 0; 1727 fh_drop_write(ffhp); 1728 1729 out: 1730 return err; 1731 } 1732 1733 /* 1734 * Unlink a file or directory 1735 * N.B. After this call fhp needs an fh_put 1736 */ 1737 __be32 1738 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1739 char *fname, int flen) 1740 { 1741 struct dentry *dentry, *rdentry; 1742 struct inode *dirp; 1743 __be32 err; 1744 int host_err; 1745 1746 err = nfserr_acces; 1747 if (!flen || isdotent(fname, flen)) 1748 goto out; 1749 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1750 if (err) 1751 goto out; 1752 1753 host_err = fh_want_write(fhp); 1754 if (host_err) 1755 goto out_nfserr; 1756 1757 fh_lock_nested(fhp, I_MUTEX_PARENT); 1758 dentry = fhp->fh_dentry; 1759 dirp = dentry->d_inode; 1760 1761 rdentry = lookup_one_len(fname, dentry, flen); 1762 host_err = PTR_ERR(rdentry); 1763 if (IS_ERR(rdentry)) 1764 goto out_nfserr; 1765 1766 if (!rdentry->d_inode) { 1767 dput(rdentry); 1768 err = nfserr_noent; 1769 goto out; 1770 } 1771 1772 if (!type) 1773 type = rdentry->d_inode->i_mode & S_IFMT; 1774 1775 if (type != S_IFDIR) 1776 host_err = vfs_unlink(dirp, rdentry, NULL); 1777 else 1778 host_err = vfs_rmdir(dirp, rdentry); 1779 if (!host_err) 1780 host_err = commit_metadata(fhp); 1781 dput(rdentry); 1782 1783 out_nfserr: 1784 err = nfserrno(host_err); 1785 out: 1786 return err; 1787 } 1788 1789 /* 1790 * We do this buffering because we must not call back into the file 1791 * system's ->lookup() method from the filldir callback. That may well 1792 * deadlock a number of file systems. 1793 * 1794 * This is based heavily on the implementation of same in XFS. 1795 */ 1796 struct buffered_dirent { 1797 u64 ino; 1798 loff_t offset; 1799 int namlen; 1800 unsigned int d_type; 1801 char name[]; 1802 }; 1803 1804 struct readdir_data { 1805 struct dir_context ctx; 1806 char *dirent; 1807 size_t used; 1808 int full; 1809 }; 1810 1811 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1812 loff_t offset, u64 ino, unsigned int d_type) 1813 { 1814 struct readdir_data *buf = __buf; 1815 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1816 unsigned int reclen; 1817 1818 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1819 if (buf->used + reclen > PAGE_SIZE) { 1820 buf->full = 1; 1821 return -EINVAL; 1822 } 1823 1824 de->namlen = namlen; 1825 de->offset = offset; 1826 de->ino = ino; 1827 de->d_type = d_type; 1828 memcpy(de->name, name, namlen); 1829 buf->used += reclen; 1830 1831 return 0; 1832 } 1833 1834 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1835 struct readdir_cd *cdp, loff_t *offsetp) 1836 { 1837 struct buffered_dirent *de; 1838 int host_err; 1839 int size; 1840 loff_t offset; 1841 struct readdir_data buf = { 1842 .ctx.actor = nfsd_buffered_filldir, 1843 .dirent = (void *)__get_free_page(GFP_KERNEL) 1844 }; 1845 1846 if (!buf.dirent) 1847 return nfserrno(-ENOMEM); 1848 1849 offset = *offsetp; 1850 1851 while (1) { 1852 struct inode *dir_inode = file_inode(file); 1853 unsigned int reclen; 1854 1855 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1856 buf.used = 0; 1857 buf.full = 0; 1858 1859 host_err = iterate_dir(file, &buf.ctx); 1860 if (buf.full) 1861 host_err = 0; 1862 1863 if (host_err < 0) 1864 break; 1865 1866 size = buf.used; 1867 1868 if (!size) 1869 break; 1870 1871 /* 1872 * Various filldir functions may end up calling back into 1873 * lookup_one_len() and the file system's ->lookup() method. 1874 * These expect i_mutex to be held, as it would within readdir. 1875 */ 1876 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1877 if (host_err) 1878 break; 1879 1880 de = (struct buffered_dirent *)buf.dirent; 1881 while (size > 0) { 1882 offset = de->offset; 1883 1884 if (func(cdp, de->name, de->namlen, de->offset, 1885 de->ino, de->d_type)) 1886 break; 1887 1888 if (cdp->err != nfs_ok) 1889 break; 1890 1891 reclen = ALIGN(sizeof(*de) + de->namlen, 1892 sizeof(u64)); 1893 size -= reclen; 1894 de = (struct buffered_dirent *)((char *)de + reclen); 1895 } 1896 mutex_unlock(&dir_inode->i_mutex); 1897 if (size > 0) /* We bailed out early */ 1898 break; 1899 1900 offset = vfs_llseek(file, 0, SEEK_CUR); 1901 } 1902 1903 free_page((unsigned long)(buf.dirent)); 1904 1905 if (host_err) 1906 return nfserrno(host_err); 1907 1908 *offsetp = offset; 1909 return cdp->err; 1910 } 1911 1912 /* 1913 * Read entries from a directory. 1914 * The NFSv3/4 verifier we ignore for now. 1915 */ 1916 __be32 1917 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 1918 struct readdir_cd *cdp, filldir_t func) 1919 { 1920 __be32 err; 1921 struct file *file; 1922 loff_t offset = *offsetp; 1923 int may_flags = NFSD_MAY_READ; 1924 1925 /* NFSv2 only supports 32 bit cookies */ 1926 if (rqstp->rq_vers > 2) 1927 may_flags |= NFSD_MAY_64BIT_COOKIE; 1928 1929 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file); 1930 if (err) 1931 goto out; 1932 1933 offset = vfs_llseek(file, offset, SEEK_SET); 1934 if (offset < 0) { 1935 err = nfserrno((int)offset); 1936 goto out_close; 1937 } 1938 1939 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 1940 1941 if (err == nfserr_eof || err == nfserr_toosmall) 1942 err = nfs_ok; /* can still be found in ->err */ 1943 out_close: 1944 nfsd_close(file); 1945 out: 1946 return err; 1947 } 1948 1949 /* 1950 * Get file system stats 1951 * N.B. After this call fhp needs an fh_put 1952 */ 1953 __be32 1954 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 1955 { 1956 __be32 err; 1957 1958 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 1959 if (!err) { 1960 struct path path = { 1961 .mnt = fhp->fh_export->ex_path.mnt, 1962 .dentry = fhp->fh_dentry, 1963 }; 1964 if (vfs_statfs(&path, stat)) 1965 err = nfserr_io; 1966 } 1967 return err; 1968 } 1969 1970 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 1971 { 1972 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 1973 } 1974 1975 /* 1976 * Check for a user's access permissions to this inode. 1977 */ 1978 __be32 1979 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 1980 struct dentry *dentry, int acc) 1981 { 1982 struct inode *inode = dentry->d_inode; 1983 int err; 1984 1985 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP) 1986 return 0; 1987 #if 0 1988 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 1989 acc, 1990 (acc & NFSD_MAY_READ)? " read" : "", 1991 (acc & NFSD_MAY_WRITE)? " write" : "", 1992 (acc & NFSD_MAY_EXEC)? " exec" : "", 1993 (acc & NFSD_MAY_SATTR)? " sattr" : "", 1994 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 1995 (acc & NFSD_MAY_LOCK)? " lock" : "", 1996 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 1997 inode->i_mode, 1998 IS_IMMUTABLE(inode)? " immut" : "", 1999 IS_APPEND(inode)? " append" : "", 2000 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2001 dprintk(" owner %d/%d user %d/%d\n", 2002 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2003 #endif 2004 2005 /* Normally we reject any write/sattr etc access on a read-only file 2006 * system. But if it is IRIX doing check on write-access for a 2007 * device special file, we ignore rofs. 2008 */ 2009 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2010 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2011 if (exp_rdonly(rqstp, exp) || 2012 __mnt_is_readonly(exp->ex_path.mnt)) 2013 return nfserr_rofs; 2014 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2015 return nfserr_perm; 2016 } 2017 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2018 return nfserr_perm; 2019 2020 if (acc & NFSD_MAY_LOCK) { 2021 /* If we cannot rely on authentication in NLM requests, 2022 * just allow locks, otherwise require read permission, or 2023 * ownership 2024 */ 2025 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2026 return 0; 2027 else 2028 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2029 } 2030 /* 2031 * The file owner always gets access permission for accesses that 2032 * would normally be checked at open time. This is to make 2033 * file access work even when the client has done a fchmod(fd, 0). 2034 * 2035 * However, `cp foo bar' should fail nevertheless when bar is 2036 * readonly. A sensible way to do this might be to reject all 2037 * attempts to truncate a read-only file, because a creat() call 2038 * always implies file truncation. 2039 * ... but this isn't really fair. A process may reasonably call 2040 * ftruncate on an open file descriptor on a file with perm 000. 2041 * We must trust the client to do permission checking - using "ACCESS" 2042 * with NFSv3. 2043 */ 2044 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2045 uid_eq(inode->i_uid, current_fsuid())) 2046 return 0; 2047 2048 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2049 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2050 2051 /* Allow read access to binaries even when mode 111 */ 2052 if (err == -EACCES && S_ISREG(inode->i_mode) && 2053 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) || 2054 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC))) 2055 err = inode_permission(inode, MAY_EXEC); 2056 2057 return err? nfserrno(err) : 0; 2058 } 2059 2060 void 2061 nfsd_racache_shutdown(void) 2062 { 2063 struct raparms *raparm, *last_raparm; 2064 unsigned int i; 2065 2066 dprintk("nfsd: freeing readahead buffers.\n"); 2067 2068 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2069 raparm = raparm_hash[i].pb_head; 2070 while(raparm) { 2071 last_raparm = raparm; 2072 raparm = raparm->p_next; 2073 kfree(last_raparm); 2074 } 2075 raparm_hash[i].pb_head = NULL; 2076 } 2077 } 2078 /* 2079 * Initialize readahead param cache 2080 */ 2081 int 2082 nfsd_racache_init(int cache_size) 2083 { 2084 int i; 2085 int j = 0; 2086 int nperbucket; 2087 struct raparms **raparm = NULL; 2088 2089 2090 if (raparm_hash[0].pb_head) 2091 return 0; 2092 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2093 if (nperbucket < 2) 2094 nperbucket = 2; 2095 cache_size = nperbucket * RAPARM_HASH_SIZE; 2096 2097 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2098 2099 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2100 spin_lock_init(&raparm_hash[i].pb_lock); 2101 2102 raparm = &raparm_hash[i].pb_head; 2103 for (j = 0; j < nperbucket; j++) { 2104 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2105 if (!*raparm) 2106 goto out_nomem; 2107 raparm = &(*raparm)->p_next; 2108 } 2109 *raparm = NULL; 2110 } 2111 2112 nfsdstats.ra_size = cache_size; 2113 return 0; 2114 2115 out_nomem: 2116 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2117 nfsd_racache_shutdown(); 2118 return -ENOMEM; 2119 } 2120