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/falloc.h> 20 #include <linux/fcntl.h> 21 #include <linux/namei.h> 22 #include <linux/delay.h> 23 #include <linux/fsnotify.h> 24 #include <linux/posix_acl_xattr.h> 25 #include <linux/xattr.h> 26 #include <linux/jhash.h> 27 #include <linux/ima.h> 28 #include <linux/slab.h> 29 #include <asm/uaccess.h> 30 #include <linux/exportfs.h> 31 #include <linux/writeback.h> 32 #include <linux/security.h> 33 34 #ifdef CONFIG_NFSD_V3 35 #include "xdr3.h" 36 #endif /* CONFIG_NFSD_V3 */ 37 38 #ifdef CONFIG_NFSD_V4 39 #include "acl.h" 40 #include "idmap.h" 41 #endif /* CONFIG_NFSD_V4 */ 42 43 #include "nfsd.h" 44 #include "vfs.h" 45 46 #define NFSDDBG_FACILITY NFSDDBG_FILEOP 47 48 49 /* 50 * This is a cache of readahead params that help us choose the proper 51 * readahead strategy. Initially, we set all readahead parameters to 0 52 * and let the VFS handle things. 53 * If you increase the number of cached files very much, you'll need to 54 * add a hash table here. 55 */ 56 struct raparms { 57 struct raparms *p_next; 58 unsigned int p_count; 59 ino_t p_ino; 60 dev_t p_dev; 61 int p_set; 62 struct file_ra_state p_ra; 63 unsigned int p_hindex; 64 }; 65 66 struct raparm_hbucket { 67 struct raparms *pb_head; 68 spinlock_t pb_lock; 69 } ____cacheline_aligned_in_smp; 70 71 #define RAPARM_HASH_BITS 4 72 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS) 73 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1) 74 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE]; 75 76 /* 77 * Called from nfsd_lookup and encode_dirent. Check if we have crossed 78 * a mount point. 79 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged, 80 * or nfs_ok having possibly changed *dpp and *expp 81 */ 82 int 83 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 84 struct svc_export **expp) 85 { 86 struct svc_export *exp = *expp, *exp2 = NULL; 87 struct dentry *dentry = *dpp; 88 struct path path = {.mnt = mntget(exp->ex_path.mnt), 89 .dentry = dget(dentry)}; 90 int err = 0; 91 92 err = follow_down(&path); 93 if (err < 0) 94 goto out; 95 96 exp2 = rqst_exp_get_by_name(rqstp, &path); 97 if (IS_ERR(exp2)) { 98 err = PTR_ERR(exp2); 99 /* 100 * We normally allow NFS clients to continue 101 * "underneath" a mountpoint that is not exported. 102 * The exception is V4ROOT, where no traversal is ever 103 * allowed without an explicit export of the new 104 * directory. 105 */ 106 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT)) 107 err = 0; 108 path_put(&path); 109 goto out; 110 } 111 if (nfsd_v4client(rqstp) || 112 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) { 113 /* successfully crossed mount point */ 114 /* 115 * This is subtle: path.dentry is *not* on path.mnt 116 * at this point. The only reason we are safe is that 117 * original mnt is pinned down by exp, so we should 118 * put path *before* putting exp 119 */ 120 *dpp = path.dentry; 121 path.dentry = dentry; 122 *expp = exp2; 123 exp2 = exp; 124 } 125 path_put(&path); 126 exp_put(exp2); 127 out: 128 return err; 129 } 130 131 static void follow_to_parent(struct path *path) 132 { 133 struct dentry *dp; 134 135 while (path->dentry == path->mnt->mnt_root && follow_up(path)) 136 ; 137 dp = dget_parent(path->dentry); 138 dput(path->dentry); 139 path->dentry = dp; 140 } 141 142 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp) 143 { 144 struct svc_export *exp2; 145 struct path path = {.mnt = mntget((*exp)->ex_path.mnt), 146 .dentry = dget(dparent)}; 147 148 follow_to_parent(&path); 149 150 exp2 = rqst_exp_parent(rqstp, &path); 151 if (PTR_ERR(exp2) == -ENOENT) { 152 *dentryp = dget(dparent); 153 } else if (IS_ERR(exp2)) { 154 path_put(&path); 155 return PTR_ERR(exp2); 156 } else { 157 *dentryp = dget(path.dentry); 158 exp_put(*exp); 159 *exp = exp2; 160 } 161 path_put(&path); 162 return 0; 163 } 164 165 /* 166 * For nfsd purposes, we treat V4ROOT exports as though there was an 167 * export at *every* directory. 168 */ 169 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp) 170 { 171 if (d_mountpoint(dentry)) 172 return 1; 173 if (nfsd4_is_junction(dentry)) 174 return 1; 175 if (!(exp->ex_flags & NFSEXP_V4ROOT)) 176 return 0; 177 return dentry->d_inode != NULL; 178 } 179 180 __be32 181 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp, 182 const char *name, unsigned int len, 183 struct svc_export **exp_ret, struct dentry **dentry_ret) 184 { 185 struct svc_export *exp; 186 struct dentry *dparent; 187 struct dentry *dentry; 188 int host_err; 189 190 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name); 191 192 dparent = fhp->fh_dentry; 193 exp = exp_get(fhp->fh_export); 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 * RFC5661, Section 18.30.4: 452 * Changing the size of a file with SETATTR indirectly 453 * changes the time_modify and change attributes. 454 * 455 * (and similar for the older RFCs) 456 */ 457 if (iap->ia_size != i_size_read(inode)) 458 iap->ia_valid |= ATTR_MTIME; 459 } 460 461 iap->ia_valid |= ATTR_CTIME; 462 463 if (check_guard && guardtime != inode->i_ctime.tv_sec) { 464 err = nfserr_notsync; 465 goto out_put_write_access; 466 } 467 468 fh_lock(fhp); 469 host_err = notify_change(dentry, iap, NULL); 470 fh_unlock(fhp); 471 err = nfserrno(host_err); 472 473 out_put_write_access: 474 if (size_change) 475 put_write_access(inode); 476 if (!err) 477 err = nfserrno(commit_metadata(fhp)); 478 out: 479 return err; 480 } 481 482 #if defined(CONFIG_NFSD_V4) 483 /* 484 * NFS junction information is stored in an extended attribute. 485 */ 486 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs" 487 488 /** 489 * nfsd4_is_junction - Test if an object could be an NFS junction 490 * 491 * @dentry: object to test 492 * 493 * Returns 1 if "dentry" appears to contain NFS junction information. 494 * Otherwise 0 is returned. 495 */ 496 int nfsd4_is_junction(struct dentry *dentry) 497 { 498 struct inode *inode = dentry->d_inode; 499 500 if (inode == NULL) 501 return 0; 502 if (inode->i_mode & S_IXUGO) 503 return 0; 504 if (!(inode->i_mode & S_ISVTX)) 505 return 0; 506 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0) 507 return 0; 508 return 1; 509 } 510 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL 511 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, 512 struct xdr_netobj *label) 513 { 514 __be32 error; 515 int host_error; 516 struct dentry *dentry; 517 518 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR); 519 if (error) 520 return error; 521 522 dentry = fhp->fh_dentry; 523 524 mutex_lock(&dentry->d_inode->i_mutex); 525 host_error = security_inode_setsecctx(dentry, label->data, label->len); 526 mutex_unlock(&dentry->d_inode->i_mutex); 527 return nfserrno(host_error); 528 } 529 #else 530 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp, 531 struct xdr_netobj *label) 532 { 533 return nfserr_notsupp; 534 } 535 #endif 536 537 __be32 nfsd4_vfs_fallocate(struct svc_rqst *rqstp, struct svc_fh *fhp, 538 struct file *file, loff_t offset, loff_t len, 539 int flags) 540 { 541 __be32 err; 542 int error; 543 544 if (!S_ISREG(file_inode(file)->i_mode)) 545 return nfserr_inval; 546 547 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, NFSD_MAY_WRITE); 548 if (err) 549 return err; 550 551 error = vfs_fallocate(file, flags, offset, len); 552 if (!error) 553 error = commit_metadata(fhp); 554 555 return nfserrno(error); 556 } 557 #endif /* defined(CONFIG_NFSD_V4) */ 558 559 #ifdef CONFIG_NFSD_V3 560 /* 561 * Check server access rights to a file system object 562 */ 563 struct accessmap { 564 u32 access; 565 int how; 566 }; 567 static struct accessmap nfs3_regaccess[] = { 568 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 569 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 570 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC }, 571 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE }, 572 573 { 0, 0 } 574 }; 575 576 static struct accessmap nfs3_diraccess[] = { 577 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 578 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC }, 579 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC}, 580 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE }, 581 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE }, 582 583 { 0, 0 } 584 }; 585 586 static struct accessmap nfs3_anyaccess[] = { 587 /* Some clients - Solaris 2.6 at least, make an access call 588 * to the server to check for access for things like /dev/null 589 * (which really, the server doesn't care about). So 590 * We provide simple access checking for them, looking 591 * mainly at mode bits, and we make sure to ignore read-only 592 * filesystem checks 593 */ 594 { NFS3_ACCESS_READ, NFSD_MAY_READ }, 595 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC }, 596 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 597 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS }, 598 599 { 0, 0 } 600 }; 601 602 __be32 603 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported) 604 { 605 struct accessmap *map; 606 struct svc_export *export; 607 struct dentry *dentry; 608 u32 query, result = 0, sresult = 0; 609 __be32 error; 610 611 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP); 612 if (error) 613 goto out; 614 615 export = fhp->fh_export; 616 dentry = fhp->fh_dentry; 617 618 if (S_ISREG(dentry->d_inode->i_mode)) 619 map = nfs3_regaccess; 620 else if (S_ISDIR(dentry->d_inode->i_mode)) 621 map = nfs3_diraccess; 622 else 623 map = nfs3_anyaccess; 624 625 626 query = *access; 627 for (; map->access; map++) { 628 if (map->access & query) { 629 __be32 err2; 630 631 sresult |= map->access; 632 633 err2 = nfsd_permission(rqstp, export, dentry, map->how); 634 switch (err2) { 635 case nfs_ok: 636 result |= map->access; 637 break; 638 639 /* the following error codes just mean the access was not allowed, 640 * rather than an error occurred */ 641 case nfserr_rofs: 642 case nfserr_acces: 643 case nfserr_perm: 644 /* simply don't "or" in the access bit. */ 645 break; 646 default: 647 error = err2; 648 goto out; 649 } 650 } 651 } 652 *access = result; 653 if (supported) 654 *supported = sresult; 655 656 out: 657 return error; 658 } 659 #endif /* CONFIG_NFSD_V3 */ 660 661 static int nfsd_open_break_lease(struct inode *inode, int access) 662 { 663 unsigned int mode; 664 665 if (access & NFSD_MAY_NOT_BREAK_LEASE) 666 return 0; 667 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY; 668 return break_lease(inode, mode | O_NONBLOCK); 669 } 670 671 /* 672 * Open an existing file or directory. 673 * The may_flags argument indicates the type of open (read/write/lock) 674 * and additional flags. 675 * N.B. After this call fhp needs an fh_put 676 */ 677 __be32 678 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, 679 int may_flags, struct file **filp) 680 { 681 struct path path; 682 struct inode *inode; 683 struct file *file; 684 int flags = O_RDONLY|O_LARGEFILE; 685 __be32 err; 686 int host_err = 0; 687 688 validate_process_creds(); 689 690 /* 691 * If we get here, then the client has already done an "open", 692 * and (hopefully) checked permission - so allow OWNER_OVERRIDE 693 * in case a chmod has now revoked permission. 694 * 695 * Arguably we should also allow the owner override for 696 * directories, but we never have and it doesn't seem to have 697 * caused anyone a problem. If we were to change this, note 698 * also that our filldir callbacks would need a variant of 699 * lookup_one_len that doesn't check permissions. 700 */ 701 if (type == S_IFREG) 702 may_flags |= NFSD_MAY_OWNER_OVERRIDE; 703 err = fh_verify(rqstp, fhp, type, may_flags); 704 if (err) 705 goto out; 706 707 path.mnt = fhp->fh_export->ex_path.mnt; 708 path.dentry = fhp->fh_dentry; 709 inode = path.dentry->d_inode; 710 711 /* Disallow write access to files with the append-only bit set 712 * or any access when mandatory locking enabled 713 */ 714 err = nfserr_perm; 715 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE)) 716 goto out; 717 /* 718 * We must ignore files (but only files) which might have mandatory 719 * locks on them because there is no way to know if the accesser has 720 * the lock. 721 */ 722 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode)) 723 goto out; 724 725 if (!inode->i_fop) 726 goto out; 727 728 host_err = nfsd_open_break_lease(inode, may_flags); 729 if (host_err) /* NOMEM or WOULDBLOCK */ 730 goto out_nfserr; 731 732 if (may_flags & NFSD_MAY_WRITE) { 733 if (may_flags & NFSD_MAY_READ) 734 flags = O_RDWR|O_LARGEFILE; 735 else 736 flags = O_WRONLY|O_LARGEFILE; 737 } 738 739 file = dentry_open(&path, flags, current_cred()); 740 if (IS_ERR(file)) { 741 host_err = PTR_ERR(file); 742 goto out_nfserr; 743 } 744 745 host_err = ima_file_check(file, may_flags, 0); 746 if (host_err) { 747 nfsd_close(file); 748 goto out_nfserr; 749 } 750 751 if (may_flags & NFSD_MAY_64BIT_COOKIE) 752 file->f_mode |= FMODE_64BITHASH; 753 else 754 file->f_mode |= FMODE_32BITHASH; 755 756 *filp = file; 757 out_nfserr: 758 err = nfserrno(host_err); 759 out: 760 validate_process_creds(); 761 return err; 762 } 763 764 /* 765 * Close a file. 766 */ 767 void 768 nfsd_close(struct file *filp) 769 { 770 fput(filp); 771 } 772 773 /* 774 * Obtain the readahead parameters for the file 775 * specified by (dev, ino). 776 */ 777 778 static inline struct raparms * 779 nfsd_get_raparms(dev_t dev, ino_t ino) 780 { 781 struct raparms *ra, **rap, **frap = NULL; 782 int depth = 0; 783 unsigned int hash; 784 struct raparm_hbucket *rab; 785 786 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK; 787 rab = &raparm_hash[hash]; 788 789 spin_lock(&rab->pb_lock); 790 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) { 791 if (ra->p_ino == ino && ra->p_dev == dev) 792 goto found; 793 depth++; 794 if (ra->p_count == 0) 795 frap = rap; 796 } 797 depth = nfsdstats.ra_size; 798 if (!frap) { 799 spin_unlock(&rab->pb_lock); 800 return NULL; 801 } 802 rap = frap; 803 ra = *frap; 804 ra->p_dev = dev; 805 ra->p_ino = ino; 806 ra->p_set = 0; 807 ra->p_hindex = hash; 808 found: 809 if (rap != &rab->pb_head) { 810 *rap = ra->p_next; 811 ra->p_next = rab->pb_head; 812 rab->pb_head = ra; 813 } 814 ra->p_count++; 815 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++; 816 spin_unlock(&rab->pb_lock); 817 return ra; 818 } 819 820 /* 821 * Grab and keep cached pages associated with a file in the svc_rqst 822 * so that they can be passed to the network sendmsg/sendpage routines 823 * directly. They will be released after the sending has completed. 824 */ 825 static int 826 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf, 827 struct splice_desc *sd) 828 { 829 struct svc_rqst *rqstp = sd->u.data; 830 struct page **pp = rqstp->rq_next_page; 831 struct page *page = buf->page; 832 size_t size; 833 834 size = sd->len; 835 836 if (rqstp->rq_res.page_len == 0) { 837 get_page(page); 838 put_page(*rqstp->rq_next_page); 839 *(rqstp->rq_next_page++) = page; 840 rqstp->rq_res.page_base = buf->offset; 841 rqstp->rq_res.page_len = size; 842 } else if (page != pp[-1]) { 843 get_page(page); 844 if (*rqstp->rq_next_page) 845 put_page(*rqstp->rq_next_page); 846 *(rqstp->rq_next_page++) = page; 847 rqstp->rq_res.page_len += size; 848 } else 849 rqstp->rq_res.page_len += size; 850 851 return size; 852 } 853 854 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe, 855 struct splice_desc *sd) 856 { 857 return __splice_from_pipe(pipe, sd, nfsd_splice_actor); 858 } 859 860 static __be32 861 nfsd_finish_read(struct file *file, unsigned long *count, int host_err) 862 { 863 if (host_err >= 0) { 864 nfsdstats.io_read += host_err; 865 *count = host_err; 866 fsnotify_access(file); 867 return 0; 868 } else 869 return nfserrno(host_err); 870 } 871 872 __be32 nfsd_splice_read(struct svc_rqst *rqstp, 873 struct file *file, loff_t offset, unsigned long *count) 874 { 875 struct splice_desc sd = { 876 .len = 0, 877 .total_len = *count, 878 .pos = offset, 879 .u.data = rqstp, 880 }; 881 int host_err; 882 883 rqstp->rq_next_page = rqstp->rq_respages + 1; 884 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor); 885 return nfsd_finish_read(file, count, host_err); 886 } 887 888 __be32 nfsd_readv(struct file *file, loff_t offset, struct kvec *vec, int vlen, 889 unsigned long *count) 890 { 891 mm_segment_t oldfs; 892 int host_err; 893 894 oldfs = get_fs(); 895 set_fs(KERNEL_DS); 896 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset); 897 set_fs(oldfs); 898 return nfsd_finish_read(file, count, host_err); 899 } 900 901 static __be32 902 nfsd_vfs_read(struct svc_rqst *rqstp, struct file *file, 903 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 904 { 905 if (file->f_op->splice_read && rqstp->rq_splice_ok) 906 return nfsd_splice_read(rqstp, file, offset, count); 907 else 908 return nfsd_readv(file, offset, vec, vlen, count); 909 } 910 911 /* 912 * Gathered writes: If another process is currently writing to the file, 913 * there's a high chance this is another nfsd (triggered by a bulk write 914 * from a client's biod). Rather than syncing the file with each write 915 * request, we sleep for 10 msec. 916 * 917 * I don't know if this roughly approximates C. Juszak's idea of 918 * gathered writes, but it's a nice and simple solution (IMHO), and it 919 * seems to work:-) 920 * 921 * Note: we do this only in the NFSv2 case, since v3 and higher have a 922 * better tool (separate unstable writes and commits) for solving this 923 * problem. 924 */ 925 static int wait_for_concurrent_writes(struct file *file) 926 { 927 struct inode *inode = file_inode(file); 928 static ino_t last_ino; 929 static dev_t last_dev; 930 int err = 0; 931 932 if (atomic_read(&inode->i_writecount) > 1 933 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) { 934 dprintk("nfsd: write defer %d\n", task_pid_nr(current)); 935 msleep(10); 936 dprintk("nfsd: write resume %d\n", task_pid_nr(current)); 937 } 938 939 if (inode->i_state & I_DIRTY) { 940 dprintk("nfsd: write sync %d\n", task_pid_nr(current)); 941 err = vfs_fsync(file, 0); 942 } 943 last_ino = inode->i_ino; 944 last_dev = inode->i_sb->s_dev; 945 return err; 946 } 947 948 static __be32 949 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 950 loff_t offset, struct kvec *vec, int vlen, 951 unsigned long *cnt, int *stablep) 952 { 953 struct svc_export *exp; 954 struct dentry *dentry; 955 struct inode *inode; 956 mm_segment_t oldfs; 957 __be32 err = 0; 958 int host_err; 959 int stable = *stablep; 960 int use_wgather; 961 loff_t pos = offset; 962 loff_t end = LLONG_MAX; 963 unsigned int pflags = current->flags; 964 965 if (test_bit(RQ_LOCAL, &rqstp->rq_flags)) 966 /* 967 * We want less throttling in balance_dirty_pages() 968 * and shrink_inactive_list() so that nfs to 969 * localhost doesn't cause nfsd to lock up due to all 970 * the client's dirty pages or its congested queue. 971 */ 972 current->flags |= PF_LESS_THROTTLE; 973 974 dentry = file->f_path.dentry; 975 inode = dentry->d_inode; 976 exp = fhp->fh_export; 977 978 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 979 980 if (!EX_ISSYNC(exp)) 981 stable = 0; 982 983 /* Write the data. */ 984 oldfs = get_fs(); set_fs(KERNEL_DS); 985 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos); 986 set_fs(oldfs); 987 if (host_err < 0) 988 goto out_nfserr; 989 *cnt = host_err; 990 nfsdstats.io_write += host_err; 991 fsnotify_modify(file); 992 993 if (stable) { 994 if (use_wgather) { 995 host_err = wait_for_concurrent_writes(file); 996 } else { 997 if (*cnt) 998 end = offset + *cnt - 1; 999 host_err = vfs_fsync_range(file, offset, end, 0); 1000 } 1001 } 1002 1003 out_nfserr: 1004 dprintk("nfsd: write complete host_err=%d\n", host_err); 1005 if (host_err >= 0) 1006 err = 0; 1007 else 1008 err = nfserrno(host_err); 1009 if (test_bit(RQ_LOCAL, &rqstp->rq_flags)) 1010 tsk_restore_flags(current, pflags, PF_LESS_THROTTLE); 1011 return err; 1012 } 1013 1014 __be32 nfsd_get_tmp_read_open(struct svc_rqst *rqstp, struct svc_fh *fhp, 1015 struct file **file, struct raparms **ra) 1016 { 1017 struct inode *inode; 1018 __be32 err; 1019 1020 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, file); 1021 if (err) 1022 return err; 1023 1024 inode = file_inode(*file); 1025 1026 /* Get readahead parameters */ 1027 *ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 1028 1029 if (*ra && (*ra)->p_set) 1030 (*file)->f_ra = (*ra)->p_ra; 1031 return nfs_ok; 1032 } 1033 1034 void nfsd_put_tmp_read_open(struct file *file, struct raparms *ra) 1035 { 1036 /* Write back readahead params */ 1037 if (ra) { 1038 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 1039 spin_lock(&rab->pb_lock); 1040 ra->p_ra = file->f_ra; 1041 ra->p_set = 1; 1042 ra->p_count--; 1043 spin_unlock(&rab->pb_lock); 1044 } 1045 nfsd_close(file); 1046 } 1047 1048 /* 1049 * Read data from a file. count must contain the requested read count 1050 * on entry. On return, *count contains the number of bytes actually read. 1051 * N.B. After this call fhp needs an fh_put 1052 */ 1053 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, 1054 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 1055 { 1056 struct file *file; 1057 struct raparms *ra; 1058 __be32 err; 1059 1060 err = nfsd_get_tmp_read_open(rqstp, fhp, &file, &ra); 1061 if (err) 1062 return err; 1063 1064 err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count); 1065 1066 nfsd_put_tmp_read_open(file, ra); 1067 1068 return err; 1069 } 1070 1071 /* 1072 * Write data to a file. 1073 * The stable flag requests synchronous writes. 1074 * N.B. After this call fhp needs an fh_put 1075 */ 1076 __be32 1077 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1078 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1079 int *stablep) 1080 { 1081 __be32 err = 0; 1082 1083 if (file) { 1084 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1085 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1086 if (err) 1087 goto out; 1088 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1089 stablep); 1090 } else { 1091 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1092 if (err) 1093 goto out; 1094 1095 if (cnt) 1096 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1097 cnt, stablep); 1098 nfsd_close(file); 1099 } 1100 out: 1101 return err; 1102 } 1103 1104 #ifdef CONFIG_NFSD_V3 1105 /* 1106 * Commit all pending writes to stable storage. 1107 * 1108 * Note: we only guarantee that data that lies within the range specified 1109 * by the 'offset' and 'count' parameters will be synced. 1110 * 1111 * Unfortunately we cannot lock the file to make sure we return full WCC 1112 * data to the client, as locking happens lower down in the filesystem. 1113 */ 1114 __be32 1115 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1116 loff_t offset, unsigned long count) 1117 { 1118 struct file *file; 1119 loff_t end = LLONG_MAX; 1120 __be32 err = nfserr_inval; 1121 1122 if (offset < 0) 1123 goto out; 1124 if (count != 0) { 1125 end = offset + (loff_t)count - 1; 1126 if (end < offset) 1127 goto out; 1128 } 1129 1130 err = nfsd_open(rqstp, fhp, S_IFREG, 1131 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file); 1132 if (err) 1133 goto out; 1134 if (EX_ISSYNC(fhp->fh_export)) { 1135 int err2 = vfs_fsync_range(file, offset, end, 0); 1136 1137 if (err2 != -EINVAL) 1138 err = nfserrno(err2); 1139 else 1140 err = nfserr_notsupp; 1141 } 1142 1143 nfsd_close(file); 1144 out: 1145 return err; 1146 } 1147 #endif /* CONFIG_NFSD_V3 */ 1148 1149 static __be32 1150 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1151 struct iattr *iap) 1152 { 1153 /* 1154 * Mode has already been set earlier in create: 1155 */ 1156 iap->ia_valid &= ~ATTR_MODE; 1157 /* 1158 * Setting uid/gid works only for root. Irix appears to 1159 * send along the gid on create when it tries to implement 1160 * setgid directories via NFS: 1161 */ 1162 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) 1163 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1164 if (iap->ia_valid) 1165 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1166 /* Callers expect file metadata to be committed here */ 1167 return nfserrno(commit_metadata(resfhp)); 1168 } 1169 1170 /* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1171 * setting size to 0 may fail for some specific file systems by the permission 1172 * checking which requires WRITE permission but the mode is 000. 1173 * we ignore the resizing(to 0) on the just new created file, since the size is 1174 * 0 after file created. 1175 * 1176 * call this only after vfs_create() is called. 1177 * */ 1178 static void 1179 nfsd_check_ignore_resizing(struct iattr *iap) 1180 { 1181 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1182 iap->ia_valid &= ~ATTR_SIZE; 1183 } 1184 1185 /* 1186 * Create a file (regular, directory, device, fifo); UNIX sockets 1187 * not yet implemented. 1188 * If the response fh has been verified, the parent directory should 1189 * already be locked. Note that the parent directory is left locked. 1190 * 1191 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1192 */ 1193 __be32 1194 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1195 char *fname, int flen, struct iattr *iap, 1196 int type, dev_t rdev, struct svc_fh *resfhp) 1197 { 1198 struct dentry *dentry, *dchild = NULL; 1199 struct inode *dirp; 1200 __be32 err; 1201 __be32 err2; 1202 int host_err; 1203 1204 err = nfserr_perm; 1205 if (!flen) 1206 goto out; 1207 err = nfserr_exist; 1208 if (isdotent(fname, flen)) 1209 goto out; 1210 1211 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1212 if (err) 1213 goto out; 1214 1215 dentry = fhp->fh_dentry; 1216 dirp = dentry->d_inode; 1217 1218 err = nfserr_notdir; 1219 if (!dirp->i_op->lookup) 1220 goto out; 1221 /* 1222 * Check whether the response file handle has been verified yet. 1223 * If it has, the parent directory should already be locked. 1224 */ 1225 if (!resfhp->fh_dentry) { 1226 host_err = fh_want_write(fhp); 1227 if (host_err) 1228 goto out_nfserr; 1229 1230 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1231 fh_lock_nested(fhp, I_MUTEX_PARENT); 1232 dchild = lookup_one_len(fname, dentry, flen); 1233 host_err = PTR_ERR(dchild); 1234 if (IS_ERR(dchild)) 1235 goto out_nfserr; 1236 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1237 if (err) 1238 goto out; 1239 } else { 1240 /* called from nfsd_proc_create */ 1241 dchild = dget(resfhp->fh_dentry); 1242 if (!fhp->fh_locked) { 1243 /* not actually possible */ 1244 printk(KERN_ERR 1245 "nfsd_create: parent %pd2 not locked!\n", 1246 dentry); 1247 err = nfserr_io; 1248 goto out; 1249 } 1250 } 1251 /* 1252 * Make sure the child dentry is still negative ... 1253 */ 1254 err = nfserr_exist; 1255 if (dchild->d_inode) { 1256 dprintk("nfsd_create: dentry %pd/%pd not negative!\n", 1257 dentry, dchild); 1258 goto out; 1259 } 1260 1261 if (!(iap->ia_valid & ATTR_MODE)) 1262 iap->ia_mode = 0; 1263 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1264 1265 err = nfserr_inval; 1266 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1267 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1268 type); 1269 goto out; 1270 } 1271 1272 /* 1273 * Get the dir op function pointer. 1274 */ 1275 err = 0; 1276 host_err = 0; 1277 switch (type) { 1278 case S_IFREG: 1279 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1280 if (!host_err) 1281 nfsd_check_ignore_resizing(iap); 1282 break; 1283 case S_IFDIR: 1284 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1285 break; 1286 case S_IFCHR: 1287 case S_IFBLK: 1288 case S_IFIFO: 1289 case S_IFSOCK: 1290 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1291 break; 1292 } 1293 if (host_err < 0) 1294 goto out_nfserr; 1295 1296 err = nfsd_create_setattr(rqstp, resfhp, iap); 1297 1298 /* 1299 * nfsd_create_setattr already committed the child. Transactional 1300 * filesystems had a chance to commit changes for both parent and 1301 * child * simultaneously making the following commit_metadata a 1302 * noop. 1303 */ 1304 err2 = nfserrno(commit_metadata(fhp)); 1305 if (err2) 1306 err = err2; 1307 /* 1308 * Update the file handle to get the new inode info. 1309 */ 1310 if (!err) 1311 err = fh_update(resfhp); 1312 out: 1313 if (dchild && !IS_ERR(dchild)) 1314 dput(dchild); 1315 return err; 1316 1317 out_nfserr: 1318 err = nfserrno(host_err); 1319 goto out; 1320 } 1321 1322 #ifdef CONFIG_NFSD_V3 1323 1324 static inline int nfsd_create_is_exclusive(int createmode) 1325 { 1326 return createmode == NFS3_CREATE_EXCLUSIVE 1327 || createmode == NFS4_CREATE_EXCLUSIVE4_1; 1328 } 1329 1330 /* 1331 * NFSv3 and NFSv4 version of nfsd_create 1332 */ 1333 __be32 1334 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1335 char *fname, int flen, struct iattr *iap, 1336 struct svc_fh *resfhp, int createmode, u32 *verifier, 1337 bool *truncp, bool *created) 1338 { 1339 struct dentry *dentry, *dchild = NULL; 1340 struct inode *dirp; 1341 __be32 err; 1342 int host_err; 1343 __u32 v_mtime=0, v_atime=0; 1344 1345 err = nfserr_perm; 1346 if (!flen) 1347 goto out; 1348 err = nfserr_exist; 1349 if (isdotent(fname, flen)) 1350 goto out; 1351 if (!(iap->ia_valid & ATTR_MODE)) 1352 iap->ia_mode = 0; 1353 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 1354 if (err) 1355 goto out; 1356 1357 dentry = fhp->fh_dentry; 1358 dirp = dentry->d_inode; 1359 1360 /* Get all the sanity checks out of the way before 1361 * we lock the parent. */ 1362 err = nfserr_notdir; 1363 if (!dirp->i_op->lookup) 1364 goto out; 1365 1366 host_err = fh_want_write(fhp); 1367 if (host_err) 1368 goto out_nfserr; 1369 1370 fh_lock_nested(fhp, I_MUTEX_PARENT); 1371 1372 /* 1373 * Compose the response file handle. 1374 */ 1375 dchild = lookup_one_len(fname, dentry, flen); 1376 host_err = PTR_ERR(dchild); 1377 if (IS_ERR(dchild)) 1378 goto out_nfserr; 1379 1380 /* If file doesn't exist, check for permissions to create one */ 1381 if (!dchild->d_inode) { 1382 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1383 if (err) 1384 goto out; 1385 } 1386 1387 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1388 if (err) 1389 goto out; 1390 1391 if (nfsd_create_is_exclusive(createmode)) { 1392 /* solaris7 gets confused (bugid 4218508) if these have 1393 * the high bit set, so just clear the high bits. If this is 1394 * ever changed to use different attrs for storing the 1395 * verifier, then do_open_lookup() will also need to be fixed 1396 * accordingly. 1397 */ 1398 v_mtime = verifier[0]&0x7fffffff; 1399 v_atime = verifier[1]&0x7fffffff; 1400 } 1401 1402 if (dchild->d_inode) { 1403 err = 0; 1404 1405 switch (createmode) { 1406 case NFS3_CREATE_UNCHECKED: 1407 if (! S_ISREG(dchild->d_inode->i_mode)) 1408 goto out; 1409 else if (truncp) { 1410 /* in nfsv4, we need to treat this case a little 1411 * differently. we don't want to truncate the 1412 * file now; this would be wrong if the OPEN 1413 * fails for some other reason. furthermore, 1414 * if the size is nonzero, we should ignore it 1415 * according to spec! 1416 */ 1417 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1418 } 1419 else { 1420 iap->ia_valid &= ATTR_SIZE; 1421 goto set_attr; 1422 } 1423 break; 1424 case NFS3_CREATE_EXCLUSIVE: 1425 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1426 && dchild->d_inode->i_atime.tv_sec == v_atime 1427 && dchild->d_inode->i_size == 0 ) { 1428 if (created) 1429 *created = 1; 1430 break; 1431 } 1432 case NFS4_CREATE_EXCLUSIVE4_1: 1433 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime 1434 && dchild->d_inode->i_atime.tv_sec == v_atime 1435 && dchild->d_inode->i_size == 0 ) { 1436 if (created) 1437 *created = 1; 1438 goto set_attr; 1439 } 1440 /* fallthru */ 1441 case NFS3_CREATE_GUARDED: 1442 err = nfserr_exist; 1443 } 1444 fh_drop_write(fhp); 1445 goto out; 1446 } 1447 1448 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1449 if (host_err < 0) { 1450 fh_drop_write(fhp); 1451 goto out_nfserr; 1452 } 1453 if (created) 1454 *created = 1; 1455 1456 nfsd_check_ignore_resizing(iap); 1457 1458 if (nfsd_create_is_exclusive(createmode)) { 1459 /* Cram the verifier into atime/mtime */ 1460 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1461 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1462 /* XXX someone who knows this better please fix it for nsec */ 1463 iap->ia_mtime.tv_sec = v_mtime; 1464 iap->ia_atime.tv_sec = v_atime; 1465 iap->ia_mtime.tv_nsec = 0; 1466 iap->ia_atime.tv_nsec = 0; 1467 } 1468 1469 set_attr: 1470 err = nfsd_create_setattr(rqstp, resfhp, iap); 1471 1472 /* 1473 * nfsd_create_setattr already committed the child 1474 * (and possibly also the parent). 1475 */ 1476 if (!err) 1477 err = nfserrno(commit_metadata(fhp)); 1478 1479 /* 1480 * Update the filehandle to get the new inode info. 1481 */ 1482 if (!err) 1483 err = fh_update(resfhp); 1484 1485 out: 1486 fh_unlock(fhp); 1487 if (dchild && !IS_ERR(dchild)) 1488 dput(dchild); 1489 fh_drop_write(fhp); 1490 return err; 1491 1492 out_nfserr: 1493 err = nfserrno(host_err); 1494 goto out; 1495 } 1496 #endif /* CONFIG_NFSD_V3 */ 1497 1498 /* 1499 * Read a symlink. On entry, *lenp must contain the maximum path length that 1500 * fits into the buffer. On return, it contains the true length. 1501 * N.B. After this call fhp needs an fh_put 1502 */ 1503 __be32 1504 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1505 { 1506 struct inode *inode; 1507 mm_segment_t oldfs; 1508 __be32 err; 1509 int host_err; 1510 struct path path; 1511 1512 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1513 if (err) 1514 goto out; 1515 1516 path.mnt = fhp->fh_export->ex_path.mnt; 1517 path.dentry = fhp->fh_dentry; 1518 inode = path.dentry->d_inode; 1519 1520 err = nfserr_inval; 1521 if (!inode->i_op->readlink) 1522 goto out; 1523 1524 touch_atime(&path); 1525 /* N.B. Why does this call need a get_fs()?? 1526 * Remove the set_fs and watch the fireworks:-) --okir 1527 */ 1528 1529 oldfs = get_fs(); set_fs(KERNEL_DS); 1530 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp); 1531 set_fs(oldfs); 1532 1533 if (host_err < 0) 1534 goto out_nfserr; 1535 *lenp = host_err; 1536 err = 0; 1537 out: 1538 return err; 1539 1540 out_nfserr: 1541 err = nfserrno(host_err); 1542 goto out; 1543 } 1544 1545 /* 1546 * Create a symlink and look up its inode 1547 * N.B. After this call _both_ fhp and resfhp need an fh_put 1548 */ 1549 __be32 1550 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1551 char *fname, int flen, 1552 char *path, 1553 struct svc_fh *resfhp) 1554 { 1555 struct dentry *dentry, *dnew; 1556 __be32 err, cerr; 1557 int host_err; 1558 1559 err = nfserr_noent; 1560 if (!flen || path[0] == '\0') 1561 goto out; 1562 err = nfserr_exist; 1563 if (isdotent(fname, flen)) 1564 goto out; 1565 1566 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1567 if (err) 1568 goto out; 1569 1570 host_err = fh_want_write(fhp); 1571 if (host_err) 1572 goto out_nfserr; 1573 1574 fh_lock(fhp); 1575 dentry = fhp->fh_dentry; 1576 dnew = lookup_one_len(fname, dentry, flen); 1577 host_err = PTR_ERR(dnew); 1578 if (IS_ERR(dnew)) 1579 goto out_nfserr; 1580 1581 host_err = vfs_symlink(dentry->d_inode, dnew, path); 1582 err = nfserrno(host_err); 1583 if (!err) 1584 err = nfserrno(commit_metadata(fhp)); 1585 fh_unlock(fhp); 1586 1587 fh_drop_write(fhp); 1588 1589 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1590 dput(dnew); 1591 if (err==0) err = cerr; 1592 out: 1593 return err; 1594 1595 out_nfserr: 1596 err = nfserrno(host_err); 1597 goto out; 1598 } 1599 1600 /* 1601 * Create a hardlink 1602 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1603 */ 1604 __be32 1605 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1606 char *name, int len, struct svc_fh *tfhp) 1607 { 1608 struct dentry *ddir, *dnew, *dold; 1609 struct inode *dirp; 1610 __be32 err; 1611 int host_err; 1612 1613 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1614 if (err) 1615 goto out; 1616 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP); 1617 if (err) 1618 goto out; 1619 err = nfserr_isdir; 1620 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode)) 1621 goto out; 1622 err = nfserr_perm; 1623 if (!len) 1624 goto out; 1625 err = nfserr_exist; 1626 if (isdotent(name, len)) 1627 goto out; 1628 1629 host_err = fh_want_write(tfhp); 1630 if (host_err) { 1631 err = nfserrno(host_err); 1632 goto out; 1633 } 1634 1635 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1636 ddir = ffhp->fh_dentry; 1637 dirp = ddir->d_inode; 1638 1639 dnew = lookup_one_len(name, ddir, len); 1640 host_err = PTR_ERR(dnew); 1641 if (IS_ERR(dnew)) 1642 goto out_nfserr; 1643 1644 dold = tfhp->fh_dentry; 1645 1646 err = nfserr_noent; 1647 if (!dold->d_inode) 1648 goto out_dput; 1649 host_err = vfs_link(dold, dirp, dnew, NULL); 1650 if (!host_err) { 1651 err = nfserrno(commit_metadata(ffhp)); 1652 if (!err) 1653 err = nfserrno(commit_metadata(tfhp)); 1654 } else { 1655 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1656 err = nfserr_acces; 1657 else 1658 err = nfserrno(host_err); 1659 } 1660 out_dput: 1661 dput(dnew); 1662 out_unlock: 1663 fh_unlock(ffhp); 1664 fh_drop_write(tfhp); 1665 out: 1666 return err; 1667 1668 out_nfserr: 1669 err = nfserrno(host_err); 1670 goto out_unlock; 1671 } 1672 1673 /* 1674 * Rename a file 1675 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1676 */ 1677 __be32 1678 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1679 struct svc_fh *tfhp, char *tname, int tlen) 1680 { 1681 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1682 struct inode *fdir, *tdir; 1683 __be32 err; 1684 int host_err; 1685 1686 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1687 if (err) 1688 goto out; 1689 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1690 if (err) 1691 goto out; 1692 1693 fdentry = ffhp->fh_dentry; 1694 fdir = fdentry->d_inode; 1695 1696 tdentry = tfhp->fh_dentry; 1697 tdir = tdentry->d_inode; 1698 1699 err = nfserr_perm; 1700 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1701 goto out; 1702 1703 host_err = fh_want_write(ffhp); 1704 if (host_err) { 1705 err = nfserrno(host_err); 1706 goto out; 1707 } 1708 1709 /* cannot use fh_lock as we need deadlock protective ordering 1710 * so do it by hand */ 1711 trap = lock_rename(tdentry, fdentry); 1712 ffhp->fh_locked = tfhp->fh_locked = 1; 1713 fill_pre_wcc(ffhp); 1714 fill_pre_wcc(tfhp); 1715 1716 odentry = lookup_one_len(fname, fdentry, flen); 1717 host_err = PTR_ERR(odentry); 1718 if (IS_ERR(odentry)) 1719 goto out_nfserr; 1720 1721 host_err = -ENOENT; 1722 if (!odentry->d_inode) 1723 goto out_dput_old; 1724 host_err = -EINVAL; 1725 if (odentry == trap) 1726 goto out_dput_old; 1727 1728 ndentry = lookup_one_len(tname, tdentry, tlen); 1729 host_err = PTR_ERR(ndentry); 1730 if (IS_ERR(ndentry)) 1731 goto out_dput_old; 1732 host_err = -ENOTEMPTY; 1733 if (ndentry == trap) 1734 goto out_dput_new; 1735 1736 host_err = -EXDEV; 1737 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1738 goto out_dput_new; 1739 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry) 1740 goto out_dput_new; 1741 1742 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0); 1743 if (!host_err) { 1744 host_err = commit_metadata(tfhp); 1745 if (!host_err) 1746 host_err = commit_metadata(ffhp); 1747 } 1748 out_dput_new: 1749 dput(ndentry); 1750 out_dput_old: 1751 dput(odentry); 1752 out_nfserr: 1753 err = nfserrno(host_err); 1754 /* 1755 * We cannot rely on fh_unlock on the two filehandles, 1756 * as that would do the wrong thing if the two directories 1757 * were the same, so again we do it by hand. 1758 */ 1759 fill_post_wcc(ffhp); 1760 fill_post_wcc(tfhp); 1761 unlock_rename(tdentry, fdentry); 1762 ffhp->fh_locked = tfhp->fh_locked = 0; 1763 fh_drop_write(ffhp); 1764 1765 out: 1766 return err; 1767 } 1768 1769 /* 1770 * Unlink a file or directory 1771 * N.B. After this call fhp needs an fh_put 1772 */ 1773 __be32 1774 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1775 char *fname, int flen) 1776 { 1777 struct dentry *dentry, *rdentry; 1778 struct inode *dirp; 1779 __be32 err; 1780 int host_err; 1781 1782 err = nfserr_acces; 1783 if (!flen || isdotent(fname, flen)) 1784 goto out; 1785 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1786 if (err) 1787 goto out; 1788 1789 host_err = fh_want_write(fhp); 1790 if (host_err) 1791 goto out_nfserr; 1792 1793 fh_lock_nested(fhp, I_MUTEX_PARENT); 1794 dentry = fhp->fh_dentry; 1795 dirp = dentry->d_inode; 1796 1797 rdentry = lookup_one_len(fname, dentry, flen); 1798 host_err = PTR_ERR(rdentry); 1799 if (IS_ERR(rdentry)) 1800 goto out_nfserr; 1801 1802 if (!rdentry->d_inode) { 1803 dput(rdentry); 1804 err = nfserr_noent; 1805 goto out; 1806 } 1807 1808 if (!type) 1809 type = rdentry->d_inode->i_mode & S_IFMT; 1810 1811 if (type != S_IFDIR) 1812 host_err = vfs_unlink(dirp, rdentry, NULL); 1813 else 1814 host_err = vfs_rmdir(dirp, rdentry); 1815 if (!host_err) 1816 host_err = commit_metadata(fhp); 1817 dput(rdentry); 1818 1819 out_nfserr: 1820 err = nfserrno(host_err); 1821 out: 1822 return err; 1823 } 1824 1825 /* 1826 * We do this buffering because we must not call back into the file 1827 * system's ->lookup() method from the filldir callback. That may well 1828 * deadlock a number of file systems. 1829 * 1830 * This is based heavily on the implementation of same in XFS. 1831 */ 1832 struct buffered_dirent { 1833 u64 ino; 1834 loff_t offset; 1835 int namlen; 1836 unsigned int d_type; 1837 char name[]; 1838 }; 1839 1840 struct readdir_data { 1841 struct dir_context ctx; 1842 char *dirent; 1843 size_t used; 1844 int full; 1845 }; 1846 1847 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen, 1848 loff_t offset, u64 ino, unsigned int d_type) 1849 { 1850 struct readdir_data *buf = __buf; 1851 struct buffered_dirent *de = (void *)(buf->dirent + buf->used); 1852 unsigned int reclen; 1853 1854 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64)); 1855 if (buf->used + reclen > PAGE_SIZE) { 1856 buf->full = 1; 1857 return -EINVAL; 1858 } 1859 1860 de->namlen = namlen; 1861 de->offset = offset; 1862 de->ino = ino; 1863 de->d_type = d_type; 1864 memcpy(de->name, name, namlen); 1865 buf->used += reclen; 1866 1867 return 0; 1868 } 1869 1870 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func, 1871 struct readdir_cd *cdp, loff_t *offsetp) 1872 { 1873 struct buffered_dirent *de; 1874 int host_err; 1875 int size; 1876 loff_t offset; 1877 struct readdir_data buf = { 1878 .ctx.actor = nfsd_buffered_filldir, 1879 .dirent = (void *)__get_free_page(GFP_KERNEL) 1880 }; 1881 1882 if (!buf.dirent) 1883 return nfserrno(-ENOMEM); 1884 1885 offset = *offsetp; 1886 1887 while (1) { 1888 struct inode *dir_inode = file_inode(file); 1889 unsigned int reclen; 1890 1891 cdp->err = nfserr_eof; /* will be cleared on successful read */ 1892 buf.used = 0; 1893 buf.full = 0; 1894 1895 host_err = iterate_dir(file, &buf.ctx); 1896 if (buf.full) 1897 host_err = 0; 1898 1899 if (host_err < 0) 1900 break; 1901 1902 size = buf.used; 1903 1904 if (!size) 1905 break; 1906 1907 /* 1908 * Various filldir functions may end up calling back into 1909 * lookup_one_len() and the file system's ->lookup() method. 1910 * These expect i_mutex to be held, as it would within readdir. 1911 */ 1912 host_err = mutex_lock_killable(&dir_inode->i_mutex); 1913 if (host_err) 1914 break; 1915 1916 de = (struct buffered_dirent *)buf.dirent; 1917 while (size > 0) { 1918 offset = de->offset; 1919 1920 if (func(cdp, de->name, de->namlen, de->offset, 1921 de->ino, de->d_type)) 1922 break; 1923 1924 if (cdp->err != nfs_ok) 1925 break; 1926 1927 reclen = ALIGN(sizeof(*de) + de->namlen, 1928 sizeof(u64)); 1929 size -= reclen; 1930 de = (struct buffered_dirent *)((char *)de + reclen); 1931 } 1932 mutex_unlock(&dir_inode->i_mutex); 1933 if (size > 0) /* We bailed out early */ 1934 break; 1935 1936 offset = vfs_llseek(file, 0, SEEK_CUR); 1937 } 1938 1939 free_page((unsigned long)(buf.dirent)); 1940 1941 if (host_err) 1942 return nfserrno(host_err); 1943 1944 *offsetp = offset; 1945 return cdp->err; 1946 } 1947 1948 /* 1949 * Read entries from a directory. 1950 * The NFSv3/4 verifier we ignore for now. 1951 */ 1952 __be32 1953 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 1954 struct readdir_cd *cdp, filldir_t func) 1955 { 1956 __be32 err; 1957 struct file *file; 1958 loff_t offset = *offsetp; 1959 int may_flags = NFSD_MAY_READ; 1960 1961 /* NFSv2 only supports 32 bit cookies */ 1962 if (rqstp->rq_vers > 2) 1963 may_flags |= NFSD_MAY_64BIT_COOKIE; 1964 1965 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file); 1966 if (err) 1967 goto out; 1968 1969 offset = vfs_llseek(file, offset, SEEK_SET); 1970 if (offset < 0) { 1971 err = nfserrno((int)offset); 1972 goto out_close; 1973 } 1974 1975 err = nfsd_buffered_readdir(file, func, cdp, offsetp); 1976 1977 if (err == nfserr_eof || err == nfserr_toosmall) 1978 err = nfs_ok; /* can still be found in ->err */ 1979 out_close: 1980 nfsd_close(file); 1981 out: 1982 return err; 1983 } 1984 1985 /* 1986 * Get file system stats 1987 * N.B. After this call fhp needs an fh_put 1988 */ 1989 __be32 1990 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access) 1991 { 1992 __be32 err; 1993 1994 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access); 1995 if (!err) { 1996 struct path path = { 1997 .mnt = fhp->fh_export->ex_path.mnt, 1998 .dentry = fhp->fh_dentry, 1999 }; 2000 if (vfs_statfs(&path, stat)) 2001 err = nfserr_io; 2002 } 2003 return err; 2004 } 2005 2006 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp) 2007 { 2008 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY; 2009 } 2010 2011 /* 2012 * Check for a user's access permissions to this inode. 2013 */ 2014 __be32 2015 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 2016 struct dentry *dentry, int acc) 2017 { 2018 struct inode *inode = dentry->d_inode; 2019 int err; 2020 2021 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP) 2022 return 0; 2023 #if 0 2024 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n", 2025 acc, 2026 (acc & NFSD_MAY_READ)? " read" : "", 2027 (acc & NFSD_MAY_WRITE)? " write" : "", 2028 (acc & NFSD_MAY_EXEC)? " exec" : "", 2029 (acc & NFSD_MAY_SATTR)? " sattr" : "", 2030 (acc & NFSD_MAY_TRUNC)? " trunc" : "", 2031 (acc & NFSD_MAY_LOCK)? " lock" : "", 2032 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "", 2033 inode->i_mode, 2034 IS_IMMUTABLE(inode)? " immut" : "", 2035 IS_APPEND(inode)? " append" : "", 2036 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : ""); 2037 dprintk(" owner %d/%d user %d/%d\n", 2038 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid()); 2039 #endif 2040 2041 /* Normally we reject any write/sattr etc access on a read-only file 2042 * system. But if it is IRIX doing check on write-access for a 2043 * device special file, we ignore rofs. 2044 */ 2045 if (!(acc & NFSD_MAY_LOCAL_ACCESS)) 2046 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) { 2047 if (exp_rdonly(rqstp, exp) || 2048 __mnt_is_readonly(exp->ex_path.mnt)) 2049 return nfserr_rofs; 2050 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode)) 2051 return nfserr_perm; 2052 } 2053 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode)) 2054 return nfserr_perm; 2055 2056 if (acc & NFSD_MAY_LOCK) { 2057 /* If we cannot rely on authentication in NLM requests, 2058 * just allow locks, otherwise require read permission, or 2059 * ownership 2060 */ 2061 if (exp->ex_flags & NFSEXP_NOAUTHNLM) 2062 return 0; 2063 else 2064 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE; 2065 } 2066 /* 2067 * The file owner always gets access permission for accesses that 2068 * would normally be checked at open time. This is to make 2069 * file access work even when the client has done a fchmod(fd, 0). 2070 * 2071 * However, `cp foo bar' should fail nevertheless when bar is 2072 * readonly. A sensible way to do this might be to reject all 2073 * attempts to truncate a read-only file, because a creat() call 2074 * always implies file truncation. 2075 * ... but this isn't really fair. A process may reasonably call 2076 * ftruncate on an open file descriptor on a file with perm 000. 2077 * We must trust the client to do permission checking - using "ACCESS" 2078 * with NFSv3. 2079 */ 2080 if ((acc & NFSD_MAY_OWNER_OVERRIDE) && 2081 uid_eq(inode->i_uid, current_fsuid())) 2082 return 0; 2083 2084 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */ 2085 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC)); 2086 2087 /* Allow read access to binaries even when mode 111 */ 2088 if (err == -EACCES && S_ISREG(inode->i_mode) && 2089 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) || 2090 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC))) 2091 err = inode_permission(inode, MAY_EXEC); 2092 2093 return err? nfserrno(err) : 0; 2094 } 2095 2096 void 2097 nfsd_racache_shutdown(void) 2098 { 2099 struct raparms *raparm, *last_raparm; 2100 unsigned int i; 2101 2102 dprintk("nfsd: freeing readahead buffers.\n"); 2103 2104 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2105 raparm = raparm_hash[i].pb_head; 2106 while(raparm) { 2107 last_raparm = raparm; 2108 raparm = raparm->p_next; 2109 kfree(last_raparm); 2110 } 2111 raparm_hash[i].pb_head = NULL; 2112 } 2113 } 2114 /* 2115 * Initialize readahead param cache 2116 */ 2117 int 2118 nfsd_racache_init(int cache_size) 2119 { 2120 int i; 2121 int j = 0; 2122 int nperbucket; 2123 struct raparms **raparm = NULL; 2124 2125 2126 if (raparm_hash[0].pb_head) 2127 return 0; 2128 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE); 2129 nperbucket = max(2, nperbucket); 2130 cache_size = nperbucket * RAPARM_HASH_SIZE; 2131 2132 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size); 2133 2134 for (i = 0; i < RAPARM_HASH_SIZE; i++) { 2135 spin_lock_init(&raparm_hash[i].pb_lock); 2136 2137 raparm = &raparm_hash[i].pb_head; 2138 for (j = 0; j < nperbucket; j++) { 2139 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL); 2140 if (!*raparm) 2141 goto out_nomem; 2142 raparm = &(*raparm)->p_next; 2143 } 2144 *raparm = NULL; 2145 } 2146 2147 nfsdstats.ra_size = cache_size; 2148 return 0; 2149 2150 out_nomem: 2151 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2152 nfsd_racache_shutdown(); 2153 return -ENOMEM; 2154 } 2155