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 = exp_get(fhp->fh_export); 193 194 /* Lookup the name, but don't follow links */ 195 if (isdotent(name, len)) { 196 if (len==1) 197 dentry = dget(dparent); 198 else if (dparent != exp->ex_path.dentry) 199 dentry = dget_parent(dparent); 200 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp)) 201 dentry = dget(dparent); /* .. == . just like at / */ 202 else { 203 /* checking mountpoint crossing is very different when stepping up */ 204 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry); 205 if (host_err) 206 goto out_nfserr; 207 } 208 } else { 209 /* 210 * In the nfsd4_open() case, this may be held across 211 * subsequent open and delegation acquisition which may 212 * need to take the child's i_mutex: 213 */ 214 fh_lock_nested(fhp, I_MUTEX_PARENT); 215 dentry = lookup_one_len(name, dparent, len); 216 host_err = PTR_ERR(dentry); 217 if (IS_ERR(dentry)) 218 goto out_nfserr; 219 /* 220 * check if we have crossed a mount point ... 221 */ 222 if (nfsd_mountpoint(dentry, exp)) { 223 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) { 224 dput(dentry); 225 goto out_nfserr; 226 } 227 } 228 } 229 *dentry_ret = dentry; 230 *exp_ret = exp; 231 return 0; 232 233 out_nfserr: 234 exp_put(exp); 235 return nfserrno(host_err); 236 } 237 238 /* 239 * Look up one component of a pathname. 240 * N.B. After this call _both_ fhp and resfh need an fh_put 241 * 242 * If the lookup would cross a mountpoint, and the mounted filesystem 243 * is exported to the client with NFSEXP_NOHIDE, then the lookup is 244 * accepted as it stands and the mounted directory is 245 * returned. Otherwise the covered directory is returned. 246 * NOTE: this mountpoint crossing is not supported properly by all 247 * clients and is explicitly disallowed for NFSv3 248 * NeilBrown <neilb@cse.unsw.edu.au> 249 */ 250 __be32 251 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name, 252 unsigned int len, struct svc_fh *resfh) 253 { 254 struct svc_export *exp; 255 struct dentry *dentry; 256 __be32 err; 257 258 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 259 if (err) 260 return err; 261 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry); 262 if (err) 263 return err; 264 err = check_nfsd_access(exp, rqstp); 265 if (err) 266 goto out; 267 /* 268 * Note: we compose the file handle now, but as the 269 * dentry may be negative, it may need to be updated. 270 */ 271 err = fh_compose(resfh, exp, dentry, fhp); 272 if (!err && !dentry->d_inode) 273 err = nfserr_noent; 274 out: 275 dput(dentry); 276 exp_put(exp); 277 return err; 278 } 279 280 /* 281 * Commit metadata changes to stable storage. 282 */ 283 static int 284 commit_metadata(struct svc_fh *fhp) 285 { 286 struct inode *inode = fhp->fh_dentry->d_inode; 287 const struct export_operations *export_ops = inode->i_sb->s_export_op; 288 289 if (!EX_ISSYNC(fhp->fh_export)) 290 return 0; 291 292 if (export_ops->commit_metadata) 293 return export_ops->commit_metadata(inode); 294 return sync_inode_metadata(inode, 1); 295 } 296 297 /* 298 * Go over the attributes and take care of the small differences between 299 * NFS semantics and what Linux expects. 300 */ 301 static void 302 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap) 303 { 304 /* 305 * NFSv2 does not differentiate between "set-[ac]time-to-now" 306 * which only requires access, and "set-[ac]time-to-X" which 307 * requires ownership. 308 * So if it looks like it might be "set both to the same time which 309 * is close to now", and if inode_change_ok fails, then we 310 * convert to "set to now" instead of "set to explicit time" 311 * 312 * We only call inode_change_ok as the last test as technically 313 * it is not an interface that we should be using. 314 */ 315 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET) 316 #define MAX_TOUCH_TIME_ERROR (30*60) 317 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET && 318 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) { 319 /* 320 * Looks probable. 321 * 322 * Now just make sure time is in the right ballpark. 323 * Solaris, at least, doesn't seem to care what the time 324 * request is. We require it be within 30 minutes of now. 325 */ 326 time_t delta = iap->ia_atime.tv_sec - get_seconds(); 327 if (delta < 0) 328 delta = -delta; 329 if (delta < MAX_TOUCH_TIME_ERROR && 330 inode_change_ok(inode, iap) != 0) { 331 /* 332 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME. 333 * This will cause notify_change to set these times 334 * to "now" 335 */ 336 iap->ia_valid &= ~BOTH_TIME_SET; 337 } 338 } 339 340 /* sanitize the mode change */ 341 if (iap->ia_valid & ATTR_MODE) { 342 iap->ia_mode &= S_IALLUGO; 343 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO); 344 } 345 346 /* Revoke setuid/setgid on chown */ 347 if (!S_ISDIR(inode->i_mode) && 348 ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) { 349 iap->ia_valid |= ATTR_KILL_PRIV; 350 if (iap->ia_valid & ATTR_MODE) { 351 /* we're setting mode too, just clear the s*id bits */ 352 iap->ia_mode &= ~S_ISUID; 353 if (iap->ia_mode & S_IXGRP) 354 iap->ia_mode &= ~S_ISGID; 355 } else { 356 /* set ATTR_KILL_* bits and let VFS handle it */ 357 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID); 358 } 359 } 360 } 361 362 static __be32 363 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp, 364 struct iattr *iap) 365 { 366 struct inode *inode = fhp->fh_dentry->d_inode; 367 int host_err; 368 369 if (iap->ia_size < inode->i_size) { 370 __be32 err; 371 372 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 373 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE); 374 if (err) 375 return err; 376 } 377 378 host_err = get_write_access(inode); 379 if (host_err) 380 goto out_nfserrno; 381 382 host_err = locks_verify_truncate(inode, NULL, iap->ia_size); 383 if (host_err) 384 goto out_put_write_access; 385 return 0; 386 387 out_put_write_access: 388 put_write_access(inode); 389 out_nfserrno: 390 return nfserrno(host_err); 391 } 392 393 /* 394 * Set various file attributes. After this call fhp needs an fh_put. 395 */ 396 __be32 397 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap, 398 int check_guard, time_t guardtime) 399 { 400 struct dentry *dentry; 401 struct inode *inode; 402 int accmode = NFSD_MAY_SATTR; 403 umode_t ftype = 0; 404 __be32 err; 405 int host_err; 406 bool get_write_count; 407 int size_change = 0; 408 409 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE)) 410 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE; 411 if (iap->ia_valid & ATTR_SIZE) 412 ftype = S_IFREG; 413 414 /* Callers that do fh_verify should do the fh_want_write: */ 415 get_write_count = !fhp->fh_dentry; 416 417 /* Get inode */ 418 err = fh_verify(rqstp, fhp, ftype, accmode); 419 if (err) 420 goto out; 421 if (get_write_count) { 422 host_err = fh_want_write(fhp); 423 if (host_err) 424 return nfserrno(host_err); 425 } 426 427 dentry = fhp->fh_dentry; 428 inode = dentry->d_inode; 429 430 /* Ignore any mode updates on symlinks */ 431 if (S_ISLNK(inode->i_mode)) 432 iap->ia_valid &= ~ATTR_MODE; 433 434 if (!iap->ia_valid) 435 goto out; 436 437 nfsd_sanitize_attrs(inode, iap); 438 439 /* 440 * The size case is special, it changes the file in addition to the 441 * attributes. 442 */ 443 if (iap->ia_valid & ATTR_SIZE) { 444 err = nfsd_get_write_access(rqstp, fhp, iap); 445 if (err) 446 goto out; 447 size_change = 1; 448 } 449 450 iap->ia_valid |= ATTR_CTIME; 451 452 if (check_guard && guardtime != inode->i_ctime.tv_sec) { 453 err = nfserr_notsync; 454 goto out_put_write_access; 455 } 456 457 fh_lock(fhp); 458 host_err = notify_change(dentry, iap, NULL); 459 fh_unlock(fhp); 460 err = nfserrno(host_err); 461 462 out_put_write_access: 463 if (size_change) 464 put_write_access(inode); 465 if (!err) 466 commit_metadata(fhp); 467 out: 468 return err; 469 } 470 471 #if defined(CONFIG_NFSD_V4) 472 /* 473 * NFS junction information is stored in an extended attribute. 474 */ 475 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs" 476 477 /** 478 * nfsd4_is_junction - Test if an object could be an NFS junction 479 * 480 * @dentry: object to test 481 * 482 * Returns 1 if "dentry" appears to contain NFS junction information. 483 * Otherwise 0 is returned. 484 */ 485 int nfsd4_is_junction(struct dentry *dentry) 486 { 487 struct inode *inode = dentry->d_inode; 488 489 if (inode == NULL) 490 return 0; 491 if (inode->i_mode & S_IXUGO) 492 return 0; 493 if (!(inode->i_mode & S_ISVTX)) 494 return 0; 495 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0) 496 return 0; 497 return 1; 498 } 499 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL 500 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, 501 struct xdr_netobj *label) 502 { 503 __be32 error; 504 int host_error; 505 struct dentry *dentry; 506 507 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR); 508 if (error) 509 return error; 510 511 dentry = fhp->fh_dentry; 512 513 mutex_lock(&dentry->d_inode->i_mutex); 514 host_error = security_inode_setsecctx(dentry, label->data, label->len); 515 mutex_unlock(&dentry->d_inode->i_mutex); 516 return nfserrno(host_error); 517 } 518 #else 519 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, 520 struct xdr_netobj *label) 521 { 522 return nfserr_notsupp; 523 } 524 #endif 525 526 #endif /* defined(CONFIG_NFSD_V4) */ 527 528 #ifdef CONFIG_NFSD_V3 529 /* 530 * Check server access rights to a file system object 531 */ 532 struct accessmap { 533 u32 access; 534 int how; 535 }; 536 static struct accessmap nfs3_regaccess[] = { 537 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 538 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 539 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, 540 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, 541 542 { 0, 0 } 543 }; 544 545 static struct accessmap nfs3_diraccess[] = { 546 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 547 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, 548 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, 549 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, 550 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, 551 552 { 0, 0 } 553 }; 554 555 static struct accessmap nfs3_anyaccess[] = { 556 /* Some clients - Solaris 2.6 at least, make an access call 557 * to the server to check for access for things like /dev/null 558 * (which really, the server doesn't care about). So 559 * We provide simple access checking for them, looking 560 * mainly at mode bits, and we make sure to ignore read-only 561 * filesystem checks 562 */ 563 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 564 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 565 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 566 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 567 568 { 0, 0 } 569 }; 570 571 __be32 572 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) 573 { 574 struct accessmap *map; 575 struct svc_export *export; 576 struct dentry *dentry; 577 u32 query, result = 0, sresult = 0; 578 __be32 error; 579 580 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); 581 if (error) 582 goto out; 583 584 export = fhp->fh_export; 585 dentry = fhp->fh_dentry; 586 587 if (S_ISREG(dentry->d_inode->i_mode)) 588 map = nfs3_regaccess; 589 else if (S_ISDIR(dentry->d_inode->i_mode)) 590 map = nfs3_diraccess; 591 else 592 map = nfs3_anyaccess; 593 594 595 query = *access; 596 for (; map->access; map++) { 597 if (map->access & query) { 598 __be32 err2; 599 600 sresult |= map->access; 601 602 err2 = nfsd_permission(rqstp, export, dentry, map->how); 603 switch (err2) { 604 case nfs_ok: 605 result |= map->access; 606 break; 607 608 /* the following error codes just mean the access was not allowed, 609 * rather than an error occurred */ 610 case nfserr_rofs: 611 case nfserr_acces: 612 case nfserr_perm: 613 /* simply don't "or" in the access bit. */ 614 break; 615 default: 616 error = err2; 617 goto out; 618 } 619 } 620 } 621 *access = result; 622 if (supported) 623 *supported = sresult; 624 625 out: 626 return error; 627 } 628 #endif /* CONFIG_NFSD_V3 */ 629 630 static int nfsd_open_break_lease(struct inode *inode, int access) 631 { 632 unsigned int mode; 633 634 if (access & NFSD_MAY_NOT_BREAK_LEASE) 635 return 0; 636 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY; 637 return break_lease(inode, mode | O_NONBLOCK); 638 } 639 640 /* 641 * Open an existing file or directory. 642 * The may_flags argument indicates the type of open (read/write/lock) 643 * and additional flags. 644 * N.B. After this call fhp needs an fh_put 645 */ 646 __be32 647 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, 648 int may_flags, struct file **filp) 649 { 650 struct path path; 651 struct inode *inode; 652 int flags = O_RDONLY|O_LARGEFILE; 653 __be32 err; 654 int host_err = 0; 655 656 validate_process_creds(); 657 658 /* 659 * If we get here, then the client has already done an "open", 660 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 661 * in case a chmod has now revoked permission. 662 * 663 * Arguably we should also allow the owner override for 664 * directories, but we never have and it doesn't seem to have 665 * caused anyone a problem. If we were to change this, note 666 * also that our filldir callbacks would need a variant of 667 * lookup_one_len that doesn't check permissions. 668 */ 669 if (type == S_IFREG) 670 may_flags |= NFSD_MAY_OWNER_OVERRIDE; 671 err = fh_verify(rqstp, fhp, type, may_flags); 672 if (err) 673 goto out; 674 675 path.mnt = fhp->fh_export->ex_path.mnt; 676 path.dentry = fhp->fh_dentry; 677 inode = path.dentry->d_inode; 678 679 /* Disallow write access to files with the append-only bit set 680 * or any access when mandatory locking enabled 681 */ 682 err = nfserr_perm; 683 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE)) 684 goto out; 685 /* 686 * We must ignore files (but only files) which might have mandatory 687 * locks on them because there is no way to know if the accesser has 688 * the lock. 689 */ 690 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) 691 goto out; 692 693 if (!inode->i_fop) 694 goto out; 695 696 host_err = nfsd_open_break_lease(inode, may_flags); 697 if (host_err) /* NOMEM or WOULDBLOCK */ 698 goto out_nfserr; 699 700 if (may_flags & NFSD_MAY_WRITE) { 701 if (may_flags & NFSD_MAY_READ) 702 flags = O_RDWR|O_LARGEFILE; 703 else 704 flags = O_WRONLY|O_LARGEFILE; 705 } 706 *filp = dentry_open(&path, flags, current_cred()); 707 if (IS_ERR(*filp)) { 708 host_err = PTR_ERR(*filp); 709 *filp = NULL; 710 } else { 711 host_err = ima_file_check(*filp, may_flags); 712 713 if (may_flags & NFSD_MAY_64BIT_COOKIE) 714 (*filp)->f_mode |= FMODE_64BITHASH; 715 else 716 (*filp)->f_mode |= FMODE_32BITHASH; 717 } 718 719 out_nfserr: 720 err = nfserrno(host_err); 721 out: 722 validate_process_creds(); 723 return err; 724 } 725 726 /* 727 * Close a file. 728 */ 729 void 730 nfsd_close(struct file *filp) 731 { 732 fput(filp); 733 } 734 735 /* 736 * Obtain the readahead parameters for the file 737 * specified by (dev, ino). 738 */ 739 740 static inline struct raparms * 741 nfsd_get_raparms(dev_t dev, ino_t ino) 742 { 743 struct raparms *ra, **rap, **frap = NULL; 744 int depth = 0; 745 unsigned int hash; 746 struct raparm_hbucket *rab; 747 748 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 749 rab = &raparm_hash[hash]; 750 751 spin_lock(&rab->pb_lock); 752 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 753 if (ra->p_ino == ino && ra->p_dev == dev) 754 goto found; 755 depth++; 756 if (ra->p_count == 0) 757 frap = rap; 758 } 759 depth = nfsdstats.ra_size; 760 if (!frap) { 761 spin_unlock(&rab->pb_lock); 762 return NULL; 763 } 764 rap = frap; 765 ra = *frap; 766 ra->p_dev = dev; 767 ra->p_ino = ino; 768 ra->p_set = 0; 769 ra->p_hindex = hash; 770 found: 771 if (rap != &rab->pb_head) { 772 *rap = ra->p_next; 773 ra->p_next = rab->pb_head; 774 rab->pb_head = ra; 775 } 776 ra->p_count++; 777 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 778 spin_unlock(&rab->pb_lock); 779 return ra; 780 } 781 782 /* 783 * Grab and keep cached pages associated with a file in the svc_rqst 784 * so that they can be passed to the network sendmsg/sendpage routines 785 * directly. They will be released after the sending has completed. 786 */ 787 static int 788 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 789 struct splice_desc *sd) 790 { 791 struct svc_rqst *rqstp = sd->u.data; 792 struct page **pp = rqstp->rq_next_page; 793 struct page *page = buf->page; 794 size_t size; 795 796 size = sd->len; 797 798 if (rqstp->rq_res.page_len == 0) { 799 get_page(page); 800 put_page(*rqstp->rq_next_page); 801 *(rqstp->rq_next_page++) = page; 802 rqstp->rq_res.page_base = buf->offset; 803 rqstp->rq_res.page_len = size; 804 } else if (page != pp[-1]) { 805 get_page(page); 806 if (*rqstp->rq_next_page) 807 put_page(*rqstp->rq_next_page); 808 *(rqstp->rq_next_page++) = page; 809 rqstp->rq_res.page_len += size; 810 } else 811 rqstp->rq_res.page_len += size; 812 813 return size; 814 } 815 816 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 817 struct splice_desc *sd) 818 { 819 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 820 } 821 822 static __be32 823 nfsd_finish_read(struct file *file, unsigned long *count, int host_err) 824 { 825 if (host_err >= 0) { 826 nfsdstats.io_read += host_err; 827 *count = host_err; 828 fsnotify_access(file); 829 return 0; 830 } else 831 return nfserrno(host_err); 832 } 833 834 __be32 nfsd_splice_read(struct svc_rqst *rqstp, 835 struct file *file, loff_t offset, unsigned long *count) 836 { 837 struct splice_desc sd = { 838 .len = 0, 839 .total_len = *count, 840 .pos = offset, 841 .u.data = rqstp, 842 }; 843 int host_err; 844 845 rqstp->rq_next_page = rqstp->rq_respages + 1; 846 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 847 return nfsd_finish_read(file, count, host_err); 848 } 849 850 __be32 nfsd_readv(struct file *file, loff_t offset, struct kvec *vec, int vlen, 851 unsigned long *count) 852 { 853 mm_segment_t oldfs; 854 int host_err; 855 856 oldfs = get_fs(); 857 set_fs(KERNEL_DS); 858 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 859 set_fs(oldfs); 860 return nfsd_finish_read(file, count, host_err); 861 } 862 863 static __be32 864 nfsd_vfs_read(struct svc_rqst *rqstp, struct file *file, 865 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 866 { 867 if (file->f_op->splice_read && rqstp->rq_splice_ok) 868 return nfsd_splice_read(rqstp, file, offset, count); 869 else 870 return nfsd_readv(file, offset, vec, vlen, count); 871 } 872 873 /* 874 * Gathered writes: If another process is currently writing to the file, 875 * there's a high chance this is another nfsd (triggered by a bulk write 876 * from a client's biod). Rather than syncing the file with each write 877 * request, we sleep for 10 msec. 878 * 879 * I don't know if this roughly approximates C. Juszak's idea of 880 * gathered writes, but it's a nice and simple solution (IMHO), and it 881 * seems to work:-) 882 * 883 * Note: we do this only in the NFSv2 case, since v3 and higher have a 884 * better tool (separate unstable writes and commits) for solving this 885 * problem. 886 */ 887 static int wait_for_concurrent_writes(struct file *file) 888 { 889 struct inode *inode = file_inode(file); 890 static ino_t last_ino; 891 static dev_t last_dev; 892 int err = 0; 893 894 if (atomic_read(&inode->i_writecount) > 1 895 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 896 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 897 msleep(10); 898 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 899 } 900 901 if (inode->i_state & I_DIRTY) { 902 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 903 err = vfs_fsync(file, 0); 904 } 905 last_ino = inode->i_ino; 906 last_dev = inode->i_sb->s_dev; 907 return err; 908 } 909 910 static __be32 911 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 912 loff_t offset, struct kvec *vec, int vlen, 913 unsigned long *cnt, int *stablep) 914 { 915 struct svc_export *exp; 916 struct dentry *dentry; 917 struct inode *inode; 918 mm_segment_t oldfs; 919 __be32 err = 0; 920 int host_err; 921 int stable = *stablep; 922 int use_wgather; 923 loff_t pos = offset; 924 unsigned int pflags = current->flags; 925 926 if (rqstp->rq_local) 927 /* 928 * We want less throttling in balance_dirty_pages() 929 * and shrink_inactive_list() so that nfs to 930 * localhost doesn't cause nfsd to lock up due to all 931 * the client's dirty pages or its congested queue. 932 */ 933 current->flags |= PF_LESS_THROTTLE; 934 935 dentry = file->f_path.dentry; 936 inode = dentry->d_inode; 937 exp = fhp->fh_export; 938 939 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 940 941 if (!EX_ISSYNC(exp)) 942 stable = 0; 943 944 /* Write the data. */ 945 oldfs = get_fs(); set_fs(KERNEL_DS); 946 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos); 947 set_fs(oldfs); 948 if (host_err < 0) 949 goto out_nfserr; 950 *cnt = host_err; 951 nfsdstats.io_write += host_err; 952 fsnotify_modify(file); 953 954 if (stable) { 955 if (use_wgather) 956 host_err = wait_for_concurrent_writes(file); 957 else 958 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0); 959 } 960 961 out_nfserr: 962 dprintk("nfsd: write complete host_err=%d\n", host_err); 963 if (host_err >= 0) 964 err = 0; 965 else 966 err = nfserrno(host_err); 967 if (rqstp->rq_local) 968 tsk_restore_flags(current, pflags, PF_LESS_THROTTLE); 969 return err; 970 } 971 972 __be32 nfsd_get_tmp_read_open(struct svc_rqst *rqstp, struct svc_fh *fhp, 973 struct file **file, struct raparms **ra) 974 { 975 struct inode *inode; 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 return nfs_ok; 990 } 991 992 void nfsd_put_tmp_read_open(struct file *file, struct raparms *ra) 993 { 994 /* Write back readahead params */ 995 if (ra) { 996 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 997 spin_lock(&rab->pb_lock); 998 ra->p_ra = file->f_ra; 999 ra->p_set = 1; 1000 ra->p_count--; 1001 spin_unlock(&rab->pb_lock); 1002 } 1003 nfsd_close(file); 1004 } 1005 1006 /* 1007 * Read data from a file. count must contain the requested read count 1008 * on entry. On return, *count contains the number of bytes actually read. 1009 * N.B. After this call fhp needs an fh_put 1010 */ 1011 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, 1012 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 1013 { 1014 struct file *file; 1015 struct raparms *ra; 1016 __be32 err; 1017 1018 err = nfsd_get_tmp_read_open(rqstp, fhp, &file, &ra); 1019 if (err) 1020 return err; 1021 1022 err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count); 1023 1024 nfsd_put_tmp_read_open(file, ra); 1025 1026 return err; 1027 } 1028 1029 /* 1030 * Write data to a file. 1031 * The stable flag requests synchronous writes. 1032 * N.B. After this call fhp needs an fh_put 1033 */ 1034 __be32 1035 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1036 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1037 int *stablep) 1038 { 1039 __be32 err = 0; 1040 1041 if (file) { 1042 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1043 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1044 if (err) 1045 goto out; 1046 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1047 stablep); 1048 } else { 1049 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1050 if (err) 1051 goto out; 1052 1053 if (cnt) 1054 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1055 cnt, stablep); 1056 nfsd_close(file); 1057 } 1058 out: 1059 return err; 1060 } 1061 1062 #ifdef CONFIG_NFSD_V3 1063 /* 1064 * Commit all pending writes to stable storage. 1065 * 1066 * Note: we only guarantee that data that lies within the range specified 1067 * by the 'offset' and 'count' parameters will be synced. 1068 * 1069 * Unfortunately we cannot lock the file to make sure we return full WCC 1070 * data to the client, as locking happens lower down in the filesystem. 1071 */ 1072 __be32 1073 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1074 loff_t offset, unsigned long count) 1075 { 1076 struct file *file; 1077 loff_t end = LLONG_MAX; 1078 __be32 err = nfserr_inval; 1079 1080 if (offset < 0) 1081 goto out; 1082 if (count != 0) { 1083 end = offset + (loff_t)count - 1; 1084 if (end < offset) 1085 goto out; 1086 } 1087 1088 err = nfsd_open(rqstp, fhp, S_IFREG, 1089 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file); 1090 if (err) 1091 goto out; 1092 if (EX_ISSYNC(fhp->fh_export)) { 1093 int err2 = vfs_fsync_range(file, offset, end, 0); 1094 1095 if (err2 != -EINVAL) 1096 err = nfserrno(err2); 1097 else 1098 err = nfserr_notsupp; 1099 } 1100 1101 nfsd_close(file); 1102 out: 1103 return err; 1104 } 1105 #endif /* CONFIG_NFSD_V3 */ 1106 1107 static __be32 1108 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1109 struct iattr *iap) 1110 { 1111 /* 1112 * Mode has already been set earlier in create: 1113 */ 1114 iap->ia_valid &= ~ATTR_MODE; 1115 /* 1116 * Setting uid/gid works only for root. Irix appears to 1117 * send along the gid on create when it tries to implement 1118 * setgid directories via NFS: 1119 */ 1120 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) 1121 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1122 if (iap->ia_valid) 1123 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1124 return 0; 1125 } 1126 1127 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1128 * setting size to 0 may fail for some specific file systems by the permission 1129 * checking which requires WRITE permission but the mode is 000. 1130 * we ignore the resizing(to 0) on the just new created file, since the size is 1131 * 0 after file created. 1132 * 1133 * call this only after vfs_create() is called. 1134 * */ 1135 static void 1136 nfsd_check_ignore_resizing(struct iattr *iap) 1137 { 1138 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1139 iap->ia_valid &= ~ATTR_SIZE; 1140 } 1141 1142 /* 1143 * Create a file (regular, directory, device, fifo); UNIX sockets 1144 * not yet implemented. 1145 * If the response fh has been verified, the parent directory should 1146 * already be locked. Note that the parent directory is left locked. 1147 * 1148 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1149 */ 1150 __be32 1151 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1152 char *fname, int flen, struct iattr *iap, 1153 int type, dev_t rdev, struct svc_fh *resfhp) 1154 { 1155 struct dentry *dentry, *dchild = NULL; 1156 struct inode *dirp; 1157 __be32 err; 1158 __be32 err2; 1159 int host_err; 1160 1161 err = nfserr_perm; 1162 if (!flen) 1163 goto out; 1164 err = nfserr_exist; 1165 if (isdotent(fname, flen)) 1166 goto out; 1167 1168 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1169 if (err) 1170 goto out; 1171 1172 dentry = fhp->fh_dentry; 1173 dirp = dentry->d_inode; 1174 1175 err = nfserr_notdir; 1176 if (!dirp->i_op->lookup) 1177 goto out; 1178 /* 1179 * Check whether the response file handle has been verified yet. 1180 * If it has, the parent directory should already be locked. 1181 */ 1182 if (!resfhp->fh_dentry) { 1183 host_err = fh_want_write(fhp); 1184 if (host_err) 1185 goto out_nfserr; 1186 1187 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1188 fh_lock_nested(fhp, I_MUTEX_PARENT); 1189 dchild = lookup_one_len(fname, dentry, flen); 1190 host_err = PTR_ERR(dchild); 1191 if (IS_ERR(dchild)) 1192 goto out_nfserr; 1193 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1194 if (err) 1195 goto out; 1196 } else { 1197 /* called from nfsd_proc_create */ 1198 dchild = dget(resfhp->fh_dentry); 1199 if (!fhp->fh_locked) { 1200 /* not actually possible */ 1201 printk(KERN_ERR 1202 "nfsd_create: parent %pd2 not locked!\n", 1203 dentry); 1204 err = nfserr_io; 1205 goto out; 1206 } 1207 } 1208 /* 1209 * Make sure the child dentry is still negative ... 1210 */ 1211 err = nfserr_exist; 1212 if (dchild->d_inode) { 1213 dprintk("nfsd_create: dentry %pd/%pd not negative!\n", 1214 dentry, dchild); 1215 goto out; 1216 } 1217 1218 if (!(iap->ia_valid & ATTR_MODE)) 1219 iap->ia_mode = 0; 1220 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1221 1222 err = nfserr_inval; 1223 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1224 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1225 type); 1226 goto out; 1227 } 1228 1229 /* 1230 * Get the dir op function pointer. 1231 */ 1232 err = 0; 1233 host_err = 0; 1234 switch (type) { 1235 case S_IFREG: 1236 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1237 if (!host_err) 1238 nfsd_check_ignore_resizing(iap); 1239 break; 1240 case S_IFDIR: 1241 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1242 break; 1243 case S_IFCHR: 1244 case S_IFBLK: 1245 case S_IFIFO: 1246 case S_IFSOCK: 1247 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1248 break; 1249 } 1250 if (host_err < 0) 1251 goto out_nfserr; 1252 1253 err = nfsd_create_setattr(rqstp, resfhp, iap); 1254 1255 /* 1256 * nfsd_setattr already committed the child. Transactional filesystems 1257 * had a chance to commit changes for both parent and child 1258 * simultaneously making the following commit_metadata a noop. 1259 */ 1260 err2 = nfserrno(commit_metadata(fhp)); 1261 if (err2) 1262 err = err2; 1263 /* 1264 * Update the file handle to get the new inode info. 1265 */ 1266 if (!err) 1267 err = fh_update(resfhp); 1268 out: 1269 if (dchild && !IS_ERR(dchild)) 1270 dput(dchild); 1271 return err; 1272 1273 out_nfserr: 1274 err = nfserrno(host_err); 1275 goto out; 1276 } 1277 1278 #ifdef CONFIG_NFSD_V3 1279 1280 static inline int nfsd_create_is_exclusive(int createmode) 1281 { 1282 return createmode == NFS3_CREATE_EXCLUSIVE 1283 || createmode == NFS4_CREATE_EXCLUSIVE4_1; 1284 } 1285 1286 /* 1287 * NFSv3 and NFSv4 version of nfsd_create 1288 */ 1289 __be32 1290 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1291 char *fname, int flen, struct iattr *iap, 1292 struct svc_fh *resfhp, int createmode, u32 *verifier, 1293 bool *truncp, bool *created) 1294 { 1295 struct dentry *dentry, *dchild = NULL; 1296 struct inode *dirp; 1297 __be32 err; 1298 int host_err; 1299 __u32 v_mtime=0, v_atime=0; 1300 1301 err = nfserr_perm; 1302 if (!flen) 1303 goto out; 1304 err = nfserr_exist; 1305 if (isdotent(fname, flen)) 1306 goto out; 1307 if (!(iap->ia_valid & ATTR_MODE)) 1308 iap->ia_mode = 0; 1309 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 1310 if (err) 1311 goto out; 1312 1313 dentry = fhp->fh_dentry; 1314 dirp = dentry->d_inode; 1315 1316 /* Get all the sanity checks out of the way before 1317 * we lock the parent. */ 1318 err = nfserr_notdir; 1319 if (!dirp->i_op->lookup) 1320 goto out; 1321 1322 host_err = fh_want_write(fhp); 1323 if (host_err) 1324 goto out_nfserr; 1325 1326 fh_lock_nested(fhp, I_MUTEX_PARENT); 1327 1328 /* 1329 * Compose the response file handle. 1330 */ 1331 dchild = lookup_one_len(fname, dentry, flen); 1332 host_err = PTR_ERR(dchild); 1333 if (IS_ERR(dchild)) 1334 goto out_nfserr; 1335 1336 /* If file doesn't exist, check for permissions to create one */ 1337 if (!dchild->d_inode) { 1338 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1339 if (err) 1340 goto out; 1341 } 1342 1343 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1344 if (err) 1345 goto out; 1346 1347 if (nfsd_create_is_exclusive(createmode)) { 1348 /* solaris7 gets confused (bugid 4218508) if these have 1349 * the high bit set, so just clear the high bits. If this is 1350 * ever changed to use different attrs for storing the 1351 * verifier, then do_open_lookup() will also need to be fixed 1352 * accordingly. 1353 */ 1354 v_mtime = verifier[0]&0x7fffffff; 1355 v_atime = verifier[1]&0x7fffffff; 1356 } 1357 1358 if (dchild->d_inode) { 1359 err = 0; 1360 1361 switch (createmode) { 1362 case NFS3_CREATE_UNCHECKED: 1363 if (! S_ISREG(dchild->d_inode->i_mode)) 1364 goto out; 1365 else if (truncp) { 1366 /* in nfsv4, we need to treat this case a little 1367 * differently. we don't want to truncate the 1368 * file now; this would be wrong if the OPEN 1369 * fails for some other reason. furthermore, 1370 * if the size is nonzero, we should ignore it 1371 * according to spec! 1372 */ 1373 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1374 } 1375 else { 1376 iap->ia_valid &= ATTR_SIZE; 1377 goto set_attr; 1378 } 1379 break; 1380 case NFS3_CREATE_EXCLUSIVE: 1381 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1382 && dchild->d_inode->i_atime.tv_sec == v_atime 1383 && dchild->d_inode->i_size == 0 ) { 1384 if (created) 1385 *created = 1; 1386 break; 1387 } 1388 case NFS4_CREATE_EXCLUSIVE4_1: 1389 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1390 && dchild->d_inode->i_atime.tv_sec == v_atime 1391 && dchild->d_inode->i_size == 0 ) { 1392 if (created) 1393 *created = 1; 1394 goto set_attr; 1395 } 1396 /* fallthru */ 1397 case NFS3_CREATE_GUARDED: 1398 err = nfserr_exist; 1399 } 1400 fh_drop_write(fhp); 1401 goto out; 1402 } 1403 1404 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1405 if (host_err < 0) { 1406 fh_drop_write(fhp); 1407 goto out_nfserr; 1408 } 1409 if (created) 1410 *created = 1; 1411 1412 nfsd_check_ignore_resizing(iap); 1413 1414 if (nfsd_create_is_exclusive(createmode)) { 1415 /* Cram the verifier into atime/mtime */ 1416 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1417 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1418 /* XXX someone who knows this better please fix it for nsec */ 1419 iap->ia_mtime.tv_sec = v_mtime; 1420 iap->ia_atime.tv_sec = v_atime; 1421 iap->ia_mtime.tv_nsec = 0; 1422 iap->ia_atime.tv_nsec = 0; 1423 } 1424 1425 set_attr: 1426 err = nfsd_create_setattr(rqstp, resfhp, iap); 1427 1428 /* 1429 * nfsd_setattr already committed the child (and possibly also the parent). 1430 */ 1431 if (!err) 1432 err = nfserrno(commit_metadata(fhp)); 1433 1434 /* 1435 * Update the filehandle to get the new inode info. 1436 */ 1437 if (!err) 1438 err = fh_update(resfhp); 1439 1440 out: 1441 fh_unlock(fhp); 1442 if (dchild && !IS_ERR(dchild)) 1443 dput(dchild); 1444 fh_drop_write(fhp); 1445 return err; 1446 1447 out_nfserr: 1448 err = nfserrno(host_err); 1449 goto out; 1450 } 1451 #endif /* CONFIG_NFSD_V3 */ 1452 1453 /* 1454 * Read a symlink. On entry, *lenp must contain the maximum path length that 1455 * fits into the buffer. On return, it contains the true length. 1456 * N.B. After this call fhp needs an fh_put 1457 */ 1458 __be32 1459 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1460 { 1461 struct inode *inode; 1462 mm_segment_t oldfs; 1463 __be32 err; 1464 int host_err; 1465 struct path path; 1466 1467 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1468 if (err) 1469 goto out; 1470 1471 path.mnt = fhp->fh_export->ex_path.mnt; 1472 path.dentry = fhp->fh_dentry; 1473 inode = path.dentry->d_inode; 1474 1475 err = nfserr_inval; 1476 if (!inode->i_op->readlink) 1477 goto out; 1478 1479 touch_atime(&path); 1480 /* N.B. Why does this call need a get_fs()?? 1481 * Remove the set_fs and watch the fireworks:-) --okir 1482 */ 1483 1484 oldfs = get_fs(); set_fs(KERNEL_DS); 1485 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp); 1486 set_fs(oldfs); 1487 1488 if (host_err < 0) 1489 goto out_nfserr; 1490 *lenp = host_err; 1491 err = 0; 1492 out: 1493 return err; 1494 1495 out_nfserr: 1496 err = nfserrno(host_err); 1497 goto out; 1498 } 1499 1500 /* 1501 * Create a symlink and look up its inode 1502 * N.B. After this call _both_ fhp and resfhp need an fh_put 1503 */ 1504 __be32 1505 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1506 char *fname, int flen, 1507 char *path, int plen, 1508 struct svc_fh *resfhp, 1509 struct iattr *iap) 1510 { 1511 struct dentry *dentry, *dnew; 1512 __be32 err, cerr; 1513 int host_err; 1514 1515 err = nfserr_noent; 1516 if (!flen || !plen) 1517 goto out; 1518 err = nfserr_exist; 1519 if (isdotent(fname, flen)) 1520 goto out; 1521 1522 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1523 if (err) 1524 goto out; 1525 1526 host_err = fh_want_write(fhp); 1527 if (host_err) 1528 goto out_nfserr; 1529 1530 fh_lock(fhp); 1531 dentry = fhp->fh_dentry; 1532 dnew = lookup_one_len(fname, dentry, flen); 1533 host_err = PTR_ERR(dnew); 1534 if (IS_ERR(dnew)) 1535 goto out_nfserr; 1536 1537 if (unlikely(path[plen] != 0)) { 1538 char *path_alloced = kmalloc(plen+1, GFP_KERNEL); 1539 if (path_alloced == NULL) 1540 host_err = -ENOMEM; 1541 else { 1542 strncpy(path_alloced, path, plen); 1543 path_alloced[plen] = 0; 1544 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced); 1545 kfree(path_alloced); 1546 } 1547 } else 1548 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1549 err = nfserrno(host_err); 1550 if (!err) 1551 err = nfserrno(commit_metadata(fhp)); 1552 fh_unlock(fhp); 1553 1554 fh_drop_write(fhp); 1555 1556 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1557 dput(dnew); 1558 if (err==0) err = cerr; 1559 out: 1560 return err; 1561 1562 out_nfserr: 1563 err = nfserrno(host_err); 1564 goto out; 1565 } 1566 1567 /* 1568 * Create a hardlink 1569 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1570 */ 1571 __be32 1572 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1573 char *name, int len, struct svc_fh *tfhp) 1574 { 1575 struct dentry *ddir, *dnew, *dold; 1576 struct inode *dirp; 1577 __be32 err; 1578 int host_err; 1579 1580 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1581 if (err) 1582 goto out; 1583 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP); 1584 if (err) 1585 goto out; 1586 err = nfserr_isdir; 1587 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode)) 1588 goto out; 1589 err = nfserr_perm; 1590 if (!len) 1591 goto out; 1592 err = nfserr_exist; 1593 if (isdotent(name, len)) 1594 goto out; 1595 1596 host_err = fh_want_write(tfhp); 1597 if (host_err) { 1598 err = nfserrno(host_err); 1599 goto out; 1600 } 1601 1602 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1603 ddir = ffhp->fh_dentry; 1604 dirp = ddir->d_inode; 1605 1606 dnew = lookup_one_len(name, ddir, len); 1607 host_err = PTR_ERR(dnew); 1608 if (IS_ERR(dnew)) 1609 goto out_nfserr; 1610 1611 dold = tfhp->fh_dentry; 1612 1613 err = nfserr_noent; 1614 if (!dold->d_inode) 1615 goto out_dput; 1616 host_err = vfs_link(dold, dirp, dnew, NULL); 1617 if (!host_err) { 1618 err = nfserrno(commit_metadata(ffhp)); 1619 if (!err) 1620 err = nfserrno(commit_metadata(tfhp)); 1621 } else { 1622 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1623 err = nfserr_acces; 1624 else 1625 err = nfserrno(host_err); 1626 } 1627 out_dput: 1628 dput(dnew); 1629 out_unlock: 1630 fh_unlock(ffhp); 1631 fh_drop_write(tfhp); 1632 out: 1633 return err; 1634 1635 out_nfserr: 1636 err = nfserrno(host_err); 1637 goto out_unlock; 1638 } 1639 1640 /* 1641 * Rename a file 1642 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1643 */ 1644 __be32 1645 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1646 struct svc_fh *tfhp, char *tname, int tlen) 1647 { 1648 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1649 struct inode *fdir, *tdir; 1650 __be32 err; 1651 int host_err; 1652 1653 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1654 if (err) 1655 goto out; 1656 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1657 if (err) 1658 goto out; 1659 1660 fdentry = ffhp->fh_dentry; 1661 fdir = fdentry->d_inode; 1662 1663 tdentry = tfhp->fh_dentry; 1664 tdir = tdentry->d_inode; 1665 1666 err = nfserr_perm; 1667 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1668 goto out; 1669 1670 host_err = fh_want_write(ffhp); 1671 if (host_err) { 1672 err = nfserrno(host_err); 1673 goto out; 1674 } 1675 1676 /* cannot use fh_lock as we need deadlock protective ordering 1677 * so do it by hand */ 1678 trap = lock_rename(tdentry, fdentry); 1679 ffhp->fh_locked = tfhp->fh_locked = 1; 1680 fill_pre_wcc(ffhp); 1681 fill_pre_wcc(tfhp); 1682 1683 odentry = lookup_one_len(fname, fdentry, flen); 1684 host_err = PTR_ERR(odentry); 1685 if (IS_ERR(odentry)) 1686 goto out_nfserr; 1687 1688 host_err = -ENOENT; 1689 if (!odentry->d_inode) 1690 goto out_dput_old; 1691 host_err = -EINVAL; 1692 if (odentry == trap) 1693 goto out_dput_old; 1694 1695 ndentry = lookup_one_len(tname, tdentry, tlen); 1696 host_err = PTR_ERR(ndentry); 1697 if (IS_ERR(ndentry)) 1698 goto out_dput_old; 1699 host_err = -ENOTEMPTY; 1700 if (ndentry == trap) 1701 goto out_dput_new; 1702 1703 host_err = -EXDEV; 1704 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1705 goto out_dput_new; 1706 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry) 1707 goto out_dput_new; 1708 1709 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0); 1710 if (!host_err) { 1711 host_err = commit_metadata(tfhp); 1712 if (!host_err) 1713 host_err = commit_metadata(ffhp); 1714 } 1715 out_dput_new: 1716 dput(ndentry); 1717 out_dput_old: 1718 dput(odentry); 1719 out_nfserr: 1720 err = nfserrno(host_err); 1721 /* 1722 * We cannot rely on fh_unlock on the two filehandles, 1723 * as that would do the wrong thing if the two directories 1724 * were the same, so again we do it by hand. 1725 */ 1726 fill_post_wcc(ffhp); 1727 fill_post_wcc(tfhp); 1728 unlock_rename(tdentry, fdentry); 1729 ffhp->fh_locked = tfhp->fh_locked = 0; 1730 fh_drop_write(ffhp); 1731 1732 out: 1733 return err; 1734 } 1735 1736 /* 1737 * Unlink a file or directory 1738 * N.B. After this call fhp needs an fh_put 1739 */ 1740 __be32 1741 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1742 char *fname, int flen) 1743 { 1744 struct dentry *dentry, *rdentry; 1745 struct inode *dirp; 1746 __be32 err; 1747 int host_err; 1748 1749 err = nfserr_acces; 1750 if (!flen || isdotent(fname, flen)) 1751 goto out; 1752 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1753 if (err) 1754 goto out; 1755 1756 host_err = fh_want_write(fhp); 1757 if (host_err) 1758 goto out_nfserr; 1759 1760 fh_lock_nested(fhp, I_MUTEX_PARENT); 1761 dentry = fhp->fh_dentry; 1762 dirp = dentry->d_inode; 1763 1764 rdentry = lookup_one_len(fname, dentry, flen); 1765 host_err = PTR_ERR(rdentry); 1766 if (IS_ERR(rdentry)) 1767 goto out_nfserr; 1768 1769 if (!rdentry->d_inode) { 1770 dput(rdentry); 1771 err = nfserr_noent; 1772 goto out; 1773 } 1774 1775 if (!type) 1776 type = rdentry->d_inode->i_mode & S_IFMT; 1777 1778 if (type != S_IFDIR) 1779 host_err = vfs_unlink(dirp, rdentry, NULL); 1780 else 1781 host_err = vfs_rmdir(dirp, rdentry); 1782 if (!host_err) 1783 host_err = commit_metadata(fhp); 1784 dput(rdentry); 1785 1786 out_nfserr: 1787 err = nfserrno(host_err); 1788 out: 1789 return err; 1790 } 1791 1792 /* 1793 * We do this buffering because we must not call back into the file 1794 * system's ->lookup() method from the filldir callback. That may well 1795 * deadlock a number of file systems. 1796 * 1797 * This is based heavily on the implementation of same in XFS. 1798 */ 1799 struct buffered_dirent { 1800 u64 ino; 1801 loff_t offset; 1802 int namlen; 1803 unsigned int d_type; 1804 char name[]; 1805 }; 1806 1807 struct readdir_data { 1808 struct dir_context ctx; 1809 char *dirent; 1810 size_t used; 1811 int full; 1812 }; 1813 1814 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1815 loff_t offset, u64 ino, unsigned int d_type) 1816 { 1817 struct readdir_data *buf = __buf; 1818 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1819 unsigned int reclen; 1820 1821 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1822 if (buf->used + reclen > PAGE_SIZE) { 1823 buf->full = 1; 1824 return -EINVAL; 1825 } 1826 1827 de->namlen = namlen; 1828 de->offset = offset; 1829 de->ino = ino; 1830 de->d_type = d_type; 1831 memcpy(de->name, name, namlen); 1832 buf->used += reclen; 1833 1834 return 0; 1835 } 1836 1837 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1838 struct readdir_cd *cdp, loff_t *offsetp) 1839 { 1840 struct buffered_dirent *de; 1841 int host_err; 1842 int size; 1843 loff_t offset; 1844 struct readdir_data buf = { 1845 .ctx.actor = nfsd_buffered_filldir, 1846 .dirent = (void *)__get_free_page(GFP_KERNEL) 1847 }; 1848 1849 if (!buf.dirent) 1850 return nfserrno(-ENOMEM); 1851 1852 offset = *offsetp; 1853 1854 while (1) { 1855 struct inode *dir_inode = file_inode(file); 1856 unsigned int reclen; 1857 1858 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1859 buf.used = 0; 1860 buf.full = 0; 1861 1862 host_err = iterate_dir(file, &buf.ctx); 1863 if (buf.full) 1864 host_err = 0; 1865 1866 if (host_err < 0) 1867 break; 1868 1869 size = buf.used; 1870 1871 if (!size) 1872 break; 1873 1874 /* 1875 * Various filldir functions may end up calling back into 1876 * lookup_one_len() and the file system's ->lookup() method. 1877 * These expect i_mutex to be held, as it would within readdir. 1878 */ 1879 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1880 if (host_err) 1881 break; 1882 1883 de = (struct buffered_dirent *)buf.dirent; 1884 while (size > 0) { 1885 offset = de->offset; 1886 1887 if (func(cdp, de->name, de->namlen, de->offset, 1888 de->ino, de->d_type)) 1889 break; 1890 1891 if (cdp->err != nfs_ok) 1892 break; 1893 1894 reclen = ALIGN(sizeof(*de) + de->namlen, 1895 sizeof(u64)); 1896 size -= reclen; 1897 de = (struct buffered_dirent *)((char *)de + reclen); 1898 } 1899 mutex_unlock(&dir_inode->i_mutex); 1900 if (size > 0) /* We bailed out early */ 1901 break; 1902 1903 offset = vfs_llseek(file, 0, SEEK_CUR); 1904 } 1905 1906 free_page((unsigned long)(buf.dirent)); 1907 1908 if (host_err) 1909 return nfserrno(host_err); 1910 1911 *offsetp = offset; 1912 return cdp->err; 1913 } 1914 1915 /* 1916 * Read entries from a directory. 1917 * The NFSv3/4 verifier we ignore for now. 1918 */ 1919 __be32 1920 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 1921 struct readdir_cd *cdp, filldir_t func) 1922 { 1923 __be32 err; 1924 struct file *file; 1925 loff_t offset = *offsetp; 1926 int may_flags = NFSD_MAY_READ; 1927 1928 /* NFSv2 only supports 32 bit cookies */ 1929 if (rqstp->rq_vers > 2) 1930 may_flags |= NFSD_MAY_64BIT_COOKIE; 1931 1932 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file); 1933 if (err) 1934 goto out; 1935 1936 offset = vfs_llseek(file, offset, SEEK_SET); 1937 if (offset < 0) { 1938 err = nfserrno((int)offset); 1939 goto out_close; 1940 } 1941 1942 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 1943 1944 if (err == nfserr_eof || err == nfserr_toosmall) 1945 err = nfs_ok; /* can still be found in ->err */ 1946 out_close: 1947 nfsd_close(file); 1948 out: 1949 return err; 1950 } 1951 1952 /* 1953 * Get file system stats 1954 * N.B. After this call fhp needs an fh_put 1955 */ 1956 __be32 1957 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 1958 { 1959 __be32 err; 1960 1961 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 1962 if (!err) { 1963 struct path path = { 1964 .mnt = fhp->fh_export->ex_path.mnt, 1965 .dentry = fhp->fh_dentry, 1966 }; 1967 if (vfs_statfs(&path, stat)) 1968 err = nfserr_io; 1969 } 1970 return err; 1971 } 1972 1973 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 1974 { 1975 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 1976 } 1977 1978 /* 1979 * Check for a user's access permissions to this inode. 1980 */ 1981 __be32 1982 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 1983 struct dentry *dentry, int acc) 1984 { 1985 struct inode *inode = dentry->d_inode; 1986 int err; 1987 1988 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP) 1989 return 0; 1990 #if 0 1991 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 1992 acc, 1993 (acc & NFSD_MAY_READ)? " read" : "", 1994 (acc & NFSD_MAY_WRITE)? " write" : "", 1995 (acc & NFSD_MAY_EXEC)? " exec" : "", 1996 (acc & NFSD_MAY_SATTR)? " sattr" : "", 1997 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 1998 (acc & NFSD_MAY_LOCK)? " lock" : "", 1999 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 2000 inode->i_mode, 2001 IS_IMMUTABLE(inode)? " immut" : "", 2002 IS_APPEND(inode)? " append" : "", 2003 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2004 dprintk(" owner %d/%d user %d/%d\n", 2005 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2006 #endif 2007 2008 /* Normally we reject any write/sattr etc access on a read-only file 2009 * system. But if it is IRIX doing check on write-access for a 2010 * device special file, we ignore rofs. 2011 */ 2012 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2013 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2014 if (exp_rdonly(rqstp, exp) || 2015 __mnt_is_readonly(exp->ex_path.mnt)) 2016 return nfserr_rofs; 2017 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2018 return nfserr_perm; 2019 } 2020 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2021 return nfserr_perm; 2022 2023 if (acc & NFSD_MAY_LOCK) { 2024 /* If we cannot rely on authentication in NLM requests, 2025 * just allow locks, otherwise require read permission, or 2026 * ownership 2027 */ 2028 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2029 return 0; 2030 else 2031 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2032 } 2033 /* 2034 * The file owner always gets access permission for accesses that 2035 * would normally be checked at open time. This is to make 2036 * file access work even when the client has done a fchmod(fd, 0). 2037 * 2038 * However, `cp foo bar' should fail nevertheless when bar is 2039 * readonly. A sensible way to do this might be to reject all 2040 * attempts to truncate a read-only file, because a creat() call 2041 * always implies file truncation. 2042 * ... but this isn't really fair. A process may reasonably call 2043 * ftruncate on an open file descriptor on a file with perm 000. 2044 * We must trust the client to do permission checking - using "ACCESS" 2045 * with NFSv3. 2046 */ 2047 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2048 uid_eq(inode->i_uid, current_fsuid())) 2049 return 0; 2050 2051 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2052 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2053 2054 /* Allow read access to binaries even when mode 111 */ 2055 if (err == -EACCES && S_ISREG(inode->i_mode) && 2056 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) || 2057 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC))) 2058 err = inode_permission(inode, MAY_EXEC); 2059 2060 return err? nfserrno(err) : 0; 2061 } 2062 2063 void 2064 nfsd_racache_shutdown(void) 2065 { 2066 struct raparms *raparm, *last_raparm; 2067 unsigned int i; 2068 2069 dprintk("nfsd: freeing readahead buffers.\n"); 2070 2071 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2072 raparm = raparm_hash[i].pb_head; 2073 while(raparm) { 2074 last_raparm = raparm; 2075 raparm = raparm->p_next; 2076 kfree(last_raparm); 2077 } 2078 raparm_hash[i].pb_head = NULL; 2079 } 2080 } 2081 /* 2082 * Initialize readahead param cache 2083 */ 2084 int 2085 nfsd_racache_init(int cache_size) 2086 { 2087 int i; 2088 int j = 0; 2089 int nperbucket; 2090 struct raparms **raparm = NULL; 2091 2092 2093 if (raparm_hash[0].pb_head) 2094 return 0; 2095 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2096 nperbucket = max(2, nperbucket); 2097 cache_size = nperbucket * RAPARM_HASH_SIZE; 2098 2099 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2100 2101 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2102 spin_lock_init(&raparm_hash[i].pb_lock); 2103 2104 raparm = &raparm_hash[i].pb_head; 2105 for (j = 0; j < nperbucket; j++) { 2106 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2107 if (!*raparm) 2108 goto out_nomem; 2109 raparm = &(*raparm)->p_next; 2110 } 2111 *raparm = NULL; 2112 } 2113 2114 nfsdstats.ra_size = cache_size; 2115 return 0; 2116 2117 out_nomem: 2118 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2119 nfsd_racache_shutdown(); 2120 return -ENOMEM; 2121 } 2122