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 */ 56struct 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 66struct 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) 74static 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 */ 82int 83nfsd_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); 127out: 128 return err; 129} 130 131static 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 142static 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 */ 169int 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 d_inode(dentry) != NULL; 178} 179 180__be32 181nfsd_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 234out_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 252nfsd_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 && d_really_is_negative(dentry)) 274 err = nfserr_noent; 275out: 276 dput(dentry); 277 exp_put(exp); 278 return err; 279} 280 281/* 282 * Commit metadata changes to stable storage. 283 */ 284static int 285commit_metadata(struct svc_fh *fhp) 286{ 287 struct inode *inode = d_inode(fhp->fh_dentry); 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 */ 302static void 303nfsd_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 363static __be32 364nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp, 365 struct iattr *iap) 366{ 367 struct inode *inode = d_inode(fhp->fh_dentry); 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 388out_put_write_access: 389 put_write_access(inode); 390out_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 398nfsd_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 = d_inode(dentry); 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 473out_put_write_access: 474 if (size_change) 475 put_write_access(inode); 476 if (!err) 477 err = nfserrno(commit_metadata(fhp)); 478out: 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 */ 496int nfsd4_is_junction(struct dentry *dentry) 497{ 498 struct inode *inode = d_inode(dentry); 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(&d_inode(dentry)->i_mutex); 525 host_error = security_inode_setsecctx(dentry, label->data, label->len); 526 mutex_unlock(&d_inode(dentry)->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 */ 563struct accessmap { 564 u32 access; 565 int how; 566}; 567static 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 576static 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 586static 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 603nfsd_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 (d_is_reg(dentry)) 619 map = nfs3_regaccess; 620 else if (d_is_dir(dentry)) 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 661static 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 678nfsd_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 = d_inode(path.dentry); 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; 757out_nfserr: 758 err = nfserrno(host_err); 759out: 760 validate_process_creds(); 761 return err; 762} 763 764/* 765 * Close a file. 766 */ 767void 768nfsd_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 778static inline struct raparms * 779nfsd_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; 808found: 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 */ 825static int 826nfsd_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 854static 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 860static __be32 861nfsd_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 901static __be32 902nfsd_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 && test_bit(RQ_SPLICE_OK, &rqstp->rq_flags)) 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 */ 925static 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 948static __be32 949nfsd_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 inode *inode; 955 mm_segment_t oldfs; 956 __be32 err = 0; 957 int host_err; 958 int stable = *stablep; 959 int use_wgather; 960 loff_t pos = offset; 961 loff_t end = LLONG_MAX; 962 unsigned int pflags = current->flags; 963 964 if (test_bit(RQ_LOCAL, &rqstp->rq_flags)) 965 /* 966 * We want less throttling in balance_dirty_pages() 967 * and shrink_inactive_list() so that nfs to 968 * localhost doesn't cause nfsd to lock up due to all 969 * the client's dirty pages or its congested queue. 970 */ 971 current->flags |= PF_LESS_THROTTLE; 972 973 inode = file_inode(file); 974 exp = fhp->fh_export; 975 976 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp); 977 978 if (!EX_ISSYNC(exp)) 979 stable = 0; 980 981 /* Write the data. */ 982 oldfs = get_fs(); set_fs(KERNEL_DS); 983 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos); 984 set_fs(oldfs); 985 if (host_err < 0) 986 goto out_nfserr; 987 *cnt = host_err; 988 nfsdstats.io_write += host_err; 989 fsnotify_modify(file); 990 991 if (stable) { 992 if (use_wgather) { 993 host_err = wait_for_concurrent_writes(file); 994 } else { 995 if (*cnt) 996 end = offset + *cnt - 1; 997 host_err = vfs_fsync_range(file, offset, end, 0); 998 } 999 } 1000 1001out_nfserr: 1002 dprintk("nfsd: write complete host_err=%d\n", host_err); 1003 if (host_err >= 0) 1004 err = 0; 1005 else 1006 err = nfserrno(host_err); 1007 if (test_bit(RQ_LOCAL, &rqstp->rq_flags)) 1008 tsk_restore_flags(current, pflags, PF_LESS_THROTTLE); 1009 return err; 1010} 1011 1012__be32 nfsd_get_tmp_read_open(struct svc_rqst *rqstp, struct svc_fh *fhp, 1013 struct file **file, struct raparms **ra) 1014{ 1015 struct inode *inode; 1016 __be32 err; 1017 1018 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, file); 1019 if (err) 1020 return err; 1021 1022 inode = file_inode(*file); 1023 1024 /* Get readahead parameters */ 1025 *ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino); 1026 1027 if (*ra && (*ra)->p_set) 1028 (*file)->f_ra = (*ra)->p_ra; 1029 return nfs_ok; 1030} 1031 1032void nfsd_put_tmp_read_open(struct file *file, struct raparms *ra) 1033{ 1034 /* Write back readahead params */ 1035 if (ra) { 1036 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex]; 1037 spin_lock(&rab->pb_lock); 1038 ra->p_ra = file->f_ra; 1039 ra->p_set = 1; 1040 ra->p_count--; 1041 spin_unlock(&rab->pb_lock); 1042 } 1043 nfsd_close(file); 1044} 1045 1046/* 1047 * Read data from a file. count must contain the requested read count 1048 * on entry. On return, *count contains the number of bytes actually read. 1049 * N.B. After this call fhp needs an fh_put 1050 */ 1051__be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, 1052 loff_t offset, struct kvec *vec, int vlen, unsigned long *count) 1053{ 1054 struct file *file; 1055 struct raparms *ra; 1056 __be32 err; 1057 1058 err = nfsd_get_tmp_read_open(rqstp, fhp, &file, &ra); 1059 if (err) 1060 return err; 1061 1062 err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count); 1063 1064 nfsd_put_tmp_read_open(file, ra); 1065 1066 return err; 1067} 1068 1069/* 1070 * Write data to a file. 1071 * The stable flag requests synchronous writes. 1072 * N.B. After this call fhp needs an fh_put 1073 */ 1074__be32 1075nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file, 1076 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt, 1077 int *stablep) 1078{ 1079 __be32 err = 0; 1080 1081 if (file) { 1082 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry, 1083 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE); 1084 if (err) 1085 goto out; 1086 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt, 1087 stablep); 1088 } else { 1089 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file); 1090 if (err) 1091 goto out; 1092 1093 if (cnt) 1094 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, 1095 cnt, stablep); 1096 nfsd_close(file); 1097 } 1098out: 1099 return err; 1100} 1101 1102#ifdef CONFIG_NFSD_V3 1103/* 1104 * Commit all pending writes to stable storage. 1105 * 1106 * Note: we only guarantee that data that lies within the range specified 1107 * by the 'offset' and 'count' parameters will be synced. 1108 * 1109 * Unfortunately we cannot lock the file to make sure we return full WCC 1110 * data to the client, as locking happens lower down in the filesystem. 1111 */ 1112__be32 1113nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp, 1114 loff_t offset, unsigned long count) 1115{ 1116 struct file *file; 1117 loff_t end = LLONG_MAX; 1118 __be32 err = nfserr_inval; 1119 1120 if (offset < 0) 1121 goto out; 1122 if (count != 0) { 1123 end = offset + (loff_t)count - 1; 1124 if (end < offset) 1125 goto out; 1126 } 1127 1128 err = nfsd_open(rqstp, fhp, S_IFREG, 1129 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file); 1130 if (err) 1131 goto out; 1132 if (EX_ISSYNC(fhp->fh_export)) { 1133 int err2 = vfs_fsync_range(file, offset, end, 0); 1134 1135 if (err2 != -EINVAL) 1136 err = nfserrno(err2); 1137 else 1138 err = nfserr_notsupp; 1139 } 1140 1141 nfsd_close(file); 1142out: 1143 return err; 1144} 1145#endif /* CONFIG_NFSD_V3 */ 1146 1147static __be32 1148nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp, 1149 struct iattr *iap) 1150{ 1151 /* 1152 * Mode has already been set earlier in create: 1153 */ 1154 iap->ia_valid &= ~ATTR_MODE; 1155 /* 1156 * Setting uid/gid works only for root. Irix appears to 1157 * send along the gid on create when it tries to implement 1158 * setgid directories via NFS: 1159 */ 1160 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) 1161 iap->ia_valid &= ~(ATTR_UID|ATTR_GID); 1162 if (iap->ia_valid) 1163 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0); 1164 /* Callers expect file metadata to be committed here */ 1165 return nfserrno(commit_metadata(resfhp)); 1166} 1167 1168/* HPUX client sometimes creates a file in mode 000, and sets size to 0. 1169 * setting size to 0 may fail for some specific file systems by the permission 1170 * checking which requires WRITE permission but the mode is 000. 1171 * we ignore the resizing(to 0) on the just new created file, since the size is 1172 * 0 after file created. 1173 * 1174 * call this only after vfs_create() is called. 1175 * */ 1176static void 1177nfsd_check_ignore_resizing(struct iattr *iap) 1178{ 1179 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0)) 1180 iap->ia_valid &= ~ATTR_SIZE; 1181} 1182 1183/* 1184 * Create a file (regular, directory, device, fifo); UNIX sockets 1185 * not yet implemented. 1186 * If the response fh has been verified, the parent directory should 1187 * already be locked. Note that the parent directory is left locked. 1188 * 1189 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp 1190 */ 1191__be32 1192nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1193 char *fname, int flen, struct iattr *iap, 1194 int type, dev_t rdev, struct svc_fh *resfhp) 1195{ 1196 struct dentry *dentry, *dchild = NULL; 1197 struct inode *dirp; 1198 __be32 err; 1199 __be32 err2; 1200 int host_err; 1201 1202 err = nfserr_perm; 1203 if (!flen) 1204 goto out; 1205 err = nfserr_exist; 1206 if (isdotent(fname, flen)) 1207 goto out; 1208 1209 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1210 if (err) 1211 goto out; 1212 1213 dentry = fhp->fh_dentry; 1214 dirp = d_inode(dentry); 1215 1216 err = nfserr_notdir; 1217 if (!dirp->i_op->lookup) 1218 goto out; 1219 /* 1220 * Check whether the response file handle has been verified yet. 1221 * If it has, the parent directory should already be locked. 1222 */ 1223 if (!resfhp->fh_dentry) { 1224 host_err = fh_want_write(fhp); 1225 if (host_err) 1226 goto out_nfserr; 1227 1228 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */ 1229 fh_lock_nested(fhp, I_MUTEX_PARENT); 1230 dchild = lookup_one_len(fname, dentry, flen); 1231 host_err = PTR_ERR(dchild); 1232 if (IS_ERR(dchild)) 1233 goto out_nfserr; 1234 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1235 if (err) 1236 goto out; 1237 } else { 1238 /* called from nfsd_proc_create */ 1239 dchild = dget(resfhp->fh_dentry); 1240 if (!fhp->fh_locked) { 1241 /* not actually possible */ 1242 printk(KERN_ERR 1243 "nfsd_create: parent %pd2 not locked!\n", 1244 dentry); 1245 err = nfserr_io; 1246 goto out; 1247 } 1248 } 1249 /* 1250 * Make sure the child dentry is still negative ... 1251 */ 1252 err = nfserr_exist; 1253 if (d_really_is_positive(dchild)) { 1254 dprintk("nfsd_create: dentry %pd/%pd not negative!\n", 1255 dentry, dchild); 1256 goto out; 1257 } 1258 1259 if (!(iap->ia_valid & ATTR_MODE)) 1260 iap->ia_mode = 0; 1261 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type; 1262 1263 err = nfserr_inval; 1264 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) { 1265 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n", 1266 type); 1267 goto out; 1268 } 1269 1270 /* 1271 * Get the dir op function pointer. 1272 */ 1273 err = 0; 1274 host_err = 0; 1275 switch (type) { 1276 case S_IFREG: 1277 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1278 if (!host_err) 1279 nfsd_check_ignore_resizing(iap); 1280 break; 1281 case S_IFDIR: 1282 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode); 1283 break; 1284 case S_IFCHR: 1285 case S_IFBLK: 1286 case S_IFIFO: 1287 case S_IFSOCK: 1288 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev); 1289 break; 1290 } 1291 if (host_err < 0) 1292 goto out_nfserr; 1293 1294 err = nfsd_create_setattr(rqstp, resfhp, iap); 1295 1296 /* 1297 * nfsd_create_setattr already committed the child. Transactional 1298 * filesystems had a chance to commit changes for both parent and 1299 * child * simultaneously making the following commit_metadata a 1300 * noop. 1301 */ 1302 err2 = nfserrno(commit_metadata(fhp)); 1303 if (err2) 1304 err = err2; 1305 /* 1306 * Update the file handle to get the new inode info. 1307 */ 1308 if (!err) 1309 err = fh_update(resfhp); 1310out: 1311 if (dchild && !IS_ERR(dchild)) 1312 dput(dchild); 1313 return err; 1314 1315out_nfserr: 1316 err = nfserrno(host_err); 1317 goto out; 1318} 1319 1320#ifdef CONFIG_NFSD_V3 1321 1322static inline int nfsd_create_is_exclusive(int createmode) 1323{ 1324 return createmode == NFS3_CREATE_EXCLUSIVE 1325 || createmode == NFS4_CREATE_EXCLUSIVE4_1; 1326} 1327 1328/* 1329 * NFSv3 and NFSv4 version of nfsd_create 1330 */ 1331__be32 1332do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp, 1333 char *fname, int flen, struct iattr *iap, 1334 struct svc_fh *resfhp, int createmode, u32 *verifier, 1335 bool *truncp, bool *created) 1336{ 1337 struct dentry *dentry, *dchild = NULL; 1338 struct inode *dirp; 1339 __be32 err; 1340 int host_err; 1341 __u32 v_mtime=0, v_atime=0; 1342 1343 err = nfserr_perm; 1344 if (!flen) 1345 goto out; 1346 err = nfserr_exist; 1347 if (isdotent(fname, flen)) 1348 goto out; 1349 if (!(iap->ia_valid & ATTR_MODE)) 1350 iap->ia_mode = 0; 1351 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC); 1352 if (err) 1353 goto out; 1354 1355 dentry = fhp->fh_dentry; 1356 dirp = d_inode(dentry); 1357 1358 /* Get all the sanity checks out of the way before 1359 * we lock the parent. */ 1360 err = nfserr_notdir; 1361 if (!dirp->i_op->lookup) 1362 goto out; 1363 1364 host_err = fh_want_write(fhp); 1365 if (host_err) 1366 goto out_nfserr; 1367 1368 fh_lock_nested(fhp, I_MUTEX_PARENT); 1369 1370 /* 1371 * Compose the response file handle. 1372 */ 1373 dchild = lookup_one_len(fname, dentry, flen); 1374 host_err = PTR_ERR(dchild); 1375 if (IS_ERR(dchild)) 1376 goto out_nfserr; 1377 1378 /* If file doesn't exist, check for permissions to create one */ 1379 if (d_really_is_negative(dchild)) { 1380 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1381 if (err) 1382 goto out; 1383 } 1384 1385 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp); 1386 if (err) 1387 goto out; 1388 1389 if (nfsd_create_is_exclusive(createmode)) { 1390 /* solaris7 gets confused (bugid 4218508) if these have 1391 * the high bit set, so just clear the high bits. If this is 1392 * ever changed to use different attrs for storing the 1393 * verifier, then do_open_lookup() will also need to be fixed 1394 * accordingly. 1395 */ 1396 v_mtime = verifier[0]&0x7fffffff; 1397 v_atime = verifier[1]&0x7fffffff; 1398 } 1399 1400 if (d_really_is_positive(dchild)) { 1401 err = 0; 1402 1403 switch (createmode) { 1404 case NFS3_CREATE_UNCHECKED: 1405 if (! d_is_reg(dchild)) 1406 goto out; 1407 else if (truncp) { 1408 /* in nfsv4, we need to treat this case a little 1409 * differently. we don't want to truncate the 1410 * file now; this would be wrong if the OPEN 1411 * fails for some other reason. furthermore, 1412 * if the size is nonzero, we should ignore it 1413 * according to spec! 1414 */ 1415 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size; 1416 } 1417 else { 1418 iap->ia_valid &= ATTR_SIZE; 1419 goto set_attr; 1420 } 1421 break; 1422 case NFS3_CREATE_EXCLUSIVE: 1423 if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime 1424 && d_inode(dchild)->i_atime.tv_sec == v_atime 1425 && d_inode(dchild)->i_size == 0 ) { 1426 if (created) 1427 *created = 1; 1428 break; 1429 } 1430 case NFS4_CREATE_EXCLUSIVE4_1: 1431 if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime 1432 && d_inode(dchild)->i_atime.tv_sec == v_atime 1433 && d_inode(dchild)->i_size == 0 ) { 1434 if (created) 1435 *created = 1; 1436 goto set_attr; 1437 } 1438 /* fallthru */ 1439 case NFS3_CREATE_GUARDED: 1440 err = nfserr_exist; 1441 } 1442 fh_drop_write(fhp); 1443 goto out; 1444 } 1445 1446 host_err = vfs_create(dirp, dchild, iap->ia_mode, true); 1447 if (host_err < 0) { 1448 fh_drop_write(fhp); 1449 goto out_nfserr; 1450 } 1451 if (created) 1452 *created = 1; 1453 1454 nfsd_check_ignore_resizing(iap); 1455 1456 if (nfsd_create_is_exclusive(createmode)) { 1457 /* Cram the verifier into atime/mtime */ 1458 iap->ia_valid = ATTR_MTIME|ATTR_ATIME 1459 | ATTR_MTIME_SET|ATTR_ATIME_SET; 1460 /* XXX someone who knows this better please fix it for nsec */ 1461 iap->ia_mtime.tv_sec = v_mtime; 1462 iap->ia_atime.tv_sec = v_atime; 1463 iap->ia_mtime.tv_nsec = 0; 1464 iap->ia_atime.tv_nsec = 0; 1465 } 1466 1467 set_attr: 1468 err = nfsd_create_setattr(rqstp, resfhp, iap); 1469 1470 /* 1471 * nfsd_create_setattr already committed the child 1472 * (and possibly also the parent). 1473 */ 1474 if (!err) 1475 err = nfserrno(commit_metadata(fhp)); 1476 1477 /* 1478 * Update the filehandle to get the new inode info. 1479 */ 1480 if (!err) 1481 err = fh_update(resfhp); 1482 1483 out: 1484 fh_unlock(fhp); 1485 if (dchild && !IS_ERR(dchild)) 1486 dput(dchild); 1487 fh_drop_write(fhp); 1488 return err; 1489 1490 out_nfserr: 1491 err = nfserrno(host_err); 1492 goto out; 1493} 1494#endif /* CONFIG_NFSD_V3 */ 1495 1496/* 1497 * Read a symlink. On entry, *lenp must contain the maximum path length that 1498 * fits into the buffer. On return, it contains the true length. 1499 * N.B. After this call fhp needs an fh_put 1500 */ 1501__be32 1502nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp) 1503{ 1504 struct inode *inode; 1505 mm_segment_t oldfs; 1506 __be32 err; 1507 int host_err; 1508 struct path path; 1509 1510 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP); 1511 if (err) 1512 goto out; 1513 1514 path.mnt = fhp->fh_export->ex_path.mnt; 1515 path.dentry = fhp->fh_dentry; 1516 inode = d_inode(path.dentry); 1517 1518 err = nfserr_inval; 1519 if (!inode->i_op->readlink) 1520 goto out; 1521 1522 touch_atime(&path); 1523 /* N.B. Why does this call need a get_fs()?? 1524 * Remove the set_fs and watch the fireworks:-) --okir 1525 */ 1526 1527 oldfs = get_fs(); set_fs(KERNEL_DS); 1528 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp); 1529 set_fs(oldfs); 1530 1531 if (host_err < 0) 1532 goto out_nfserr; 1533 *lenp = host_err; 1534 err = 0; 1535out: 1536 return err; 1537 1538out_nfserr: 1539 err = nfserrno(host_err); 1540 goto out; 1541} 1542 1543/* 1544 * Create a symlink and look up its inode 1545 * N.B. After this call _both_ fhp and resfhp need an fh_put 1546 */ 1547__be32 1548nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp, 1549 char *fname, int flen, 1550 char *path, 1551 struct svc_fh *resfhp) 1552{ 1553 struct dentry *dentry, *dnew; 1554 __be32 err, cerr; 1555 int host_err; 1556 1557 err = nfserr_noent; 1558 if (!flen || path[0] == '\0') 1559 goto out; 1560 err = nfserr_exist; 1561 if (isdotent(fname, flen)) 1562 goto out; 1563 1564 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE); 1565 if (err) 1566 goto out; 1567 1568 host_err = fh_want_write(fhp); 1569 if (host_err) 1570 goto out_nfserr; 1571 1572 fh_lock(fhp); 1573 dentry = fhp->fh_dentry; 1574 dnew = lookup_one_len(fname, dentry, flen); 1575 host_err = PTR_ERR(dnew); 1576 if (IS_ERR(dnew)) 1577 goto out_nfserr; 1578 1579 host_err = vfs_symlink(d_inode(dentry), dnew, path); 1580 err = nfserrno(host_err); 1581 if (!err) 1582 err = nfserrno(commit_metadata(fhp)); 1583 fh_unlock(fhp); 1584 1585 fh_drop_write(fhp); 1586 1587 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp); 1588 dput(dnew); 1589 if (err==0) err = cerr; 1590out: 1591 return err; 1592 1593out_nfserr: 1594 err = nfserrno(host_err); 1595 goto out; 1596} 1597 1598/* 1599 * Create a hardlink 1600 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1601 */ 1602__be32 1603nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp, 1604 char *name, int len, struct svc_fh *tfhp) 1605{ 1606 struct dentry *ddir, *dnew, *dold; 1607 struct inode *dirp; 1608 __be32 err; 1609 int host_err; 1610 1611 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE); 1612 if (err) 1613 goto out; 1614 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP); 1615 if (err) 1616 goto out; 1617 err = nfserr_isdir; 1618 if (d_is_dir(tfhp->fh_dentry)) 1619 goto out; 1620 err = nfserr_perm; 1621 if (!len) 1622 goto out; 1623 err = nfserr_exist; 1624 if (isdotent(name, len)) 1625 goto out; 1626 1627 host_err = fh_want_write(tfhp); 1628 if (host_err) { 1629 err = nfserrno(host_err); 1630 goto out; 1631 } 1632 1633 fh_lock_nested(ffhp, I_MUTEX_PARENT); 1634 ddir = ffhp->fh_dentry; 1635 dirp = d_inode(ddir); 1636 1637 dnew = lookup_one_len(name, ddir, len); 1638 host_err = PTR_ERR(dnew); 1639 if (IS_ERR(dnew)) 1640 goto out_nfserr; 1641 1642 dold = tfhp->fh_dentry; 1643 1644 err = nfserr_noent; 1645 if (d_really_is_negative(dold)) 1646 goto out_dput; 1647 host_err = vfs_link(dold, dirp, dnew, NULL); 1648 if (!host_err) { 1649 err = nfserrno(commit_metadata(ffhp)); 1650 if (!err) 1651 err = nfserrno(commit_metadata(tfhp)); 1652 } else { 1653 if (host_err == -EXDEV && rqstp->rq_vers == 2) 1654 err = nfserr_acces; 1655 else 1656 err = nfserrno(host_err); 1657 } 1658out_dput: 1659 dput(dnew); 1660out_unlock: 1661 fh_unlock(ffhp); 1662 fh_drop_write(tfhp); 1663out: 1664 return err; 1665 1666out_nfserr: 1667 err = nfserrno(host_err); 1668 goto out_unlock; 1669} 1670 1671/* 1672 * Rename a file 1673 * N.B. After this call _both_ ffhp and tfhp need an fh_put 1674 */ 1675__be32 1676nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen, 1677 struct svc_fh *tfhp, char *tname, int tlen) 1678{ 1679 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap; 1680 struct inode *fdir, *tdir; 1681 __be32 err; 1682 int host_err; 1683 1684 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE); 1685 if (err) 1686 goto out; 1687 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE); 1688 if (err) 1689 goto out; 1690 1691 fdentry = ffhp->fh_dentry; 1692 fdir = d_inode(fdentry); 1693 1694 tdentry = tfhp->fh_dentry; 1695 tdir = d_inode(tdentry); 1696 1697 err = nfserr_perm; 1698 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen)) 1699 goto out; 1700 1701 host_err = fh_want_write(ffhp); 1702 if (host_err) { 1703 err = nfserrno(host_err); 1704 goto out; 1705 } 1706 1707 /* cannot use fh_lock as we need deadlock protective ordering 1708 * so do it by hand */ 1709 trap = lock_rename(tdentry, fdentry); 1710 ffhp->fh_locked = tfhp->fh_locked = 1; 1711 fill_pre_wcc(ffhp); 1712 fill_pre_wcc(tfhp); 1713 1714 odentry = lookup_one_len(fname, fdentry, flen); 1715 host_err = PTR_ERR(odentry); 1716 if (IS_ERR(odentry)) 1717 goto out_nfserr; 1718 1719 host_err = -ENOENT; 1720 if (d_really_is_negative(odentry)) 1721 goto out_dput_old; 1722 host_err = -EINVAL; 1723 if (odentry == trap) 1724 goto out_dput_old; 1725 1726 ndentry = lookup_one_len(tname, tdentry, tlen); 1727 host_err = PTR_ERR(ndentry); 1728 if (IS_ERR(ndentry)) 1729 goto out_dput_old; 1730 host_err = -ENOTEMPTY; 1731 if (ndentry == trap) 1732 goto out_dput_new; 1733 1734 host_err = -EXDEV; 1735 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt) 1736 goto out_dput_new; 1737 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry) 1738 goto out_dput_new; 1739 1740 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0); 1741 if (!host_err) { 1742 host_err = commit_metadata(tfhp); 1743 if (!host_err) 1744 host_err = commit_metadata(ffhp); 1745 } 1746 out_dput_new: 1747 dput(ndentry); 1748 out_dput_old: 1749 dput(odentry); 1750 out_nfserr: 1751 err = nfserrno(host_err); 1752 /* 1753 * We cannot rely on fh_unlock on the two filehandles, 1754 * as that would do the wrong thing if the two directories 1755 * were the same, so again we do it by hand. 1756 */ 1757 fill_post_wcc(ffhp); 1758 fill_post_wcc(tfhp); 1759 unlock_rename(tdentry, fdentry); 1760 ffhp->fh_locked = tfhp->fh_locked = 0; 1761 fh_drop_write(ffhp); 1762 1763out: 1764 return err; 1765} 1766 1767/* 1768 * Unlink a file or directory 1769 * N.B. After this call fhp needs an fh_put 1770 */ 1771__be32 1772nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, 1773 char *fname, int flen) 1774{ 1775 struct dentry *dentry, *rdentry; 1776 struct inode *dirp; 1777 __be32 err; 1778 int host_err; 1779 1780 err = nfserr_acces; 1781 if (!flen || isdotent(fname, flen)) 1782 goto out; 1783 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE); 1784 if (err) 1785 goto out; 1786 1787 host_err = fh_want_write(fhp); 1788 if (host_err) 1789 goto out_nfserr; 1790 1791 fh_lock_nested(fhp, I_MUTEX_PARENT); 1792 dentry = fhp->fh_dentry; 1793 dirp = d_inode(dentry); 1794 1795 rdentry = lookup_one_len(fname, dentry, flen); 1796 host_err = PTR_ERR(rdentry); 1797 if (IS_ERR(rdentry)) 1798 goto out_nfserr; 1799 1800 if (d_really_is_negative(rdentry)) { 1801 dput(rdentry); 1802 err = nfserr_noent; 1803 goto out; 1804 } 1805 1806 if (!type) 1807 type = d_inode(rdentry)->i_mode & S_IFMT; 1808 1809 if (type != S_IFDIR) 1810 host_err = vfs_unlink(dirp, rdentry, NULL); 1811 else 1812 host_err = vfs_rmdir(dirp, rdentry); 1813 if (!host_err) 1814 host_err = commit_metadata(fhp); 1815 dput(rdentry); 1816 1817out_nfserr: 1818 err = nfserrno(host_err); 1819out: 1820 return err; 1821} 1822 1823/* 1824 * We do this buffering because we must not call back into the file 1825 * system's ->lookup() method from the filldir callback. That may well 1826 * deadlock a number of file systems. 1827 * 1828 * This is based heavily on the implementation of same in XFS. 1829 */ 1830struct buffered_dirent { 1831 u64 ino; 1832 loff_t offset; 1833 int namlen; 1834 unsigned int d_type; 1835 char name[]; 1836}; 1837 1838struct readdir_data { 1839 struct dir_context ctx; 1840 char *dirent; 1841 size_t used; 1842 int full; 1843}; 1844 1845static int nfsd_buffered_filldir(struct dir_context *ctx, const char *name, 1846 int namlen, loff_t offset, u64 ino, 1847 unsigned int d_type) 1848{ 1849 struct readdir_data *buf = 1850 container_of(ctx, struct readdir_data, ctx); 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 1870static __be32 nfsd_buffered_readdir(struct file *file, nfsd_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 1953nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 1954 struct readdir_cd *cdp, nfsd_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 */ 1979out_close: 1980 nfsd_close(file); 1981out: 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 1990nfsd_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 2006static 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 2015nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp, 2016 struct dentry *dentry, int acc) 2017{ 2018 struct inode *inode = d_inode(dentry); 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 2096void 2097nfsd_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 */ 2117int 2118nfsd_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 2150out_nomem: 2151 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n"); 2152 nfsd_racache_shutdown(); 2153 return -ENOMEM; 2154} 2155