1 /* 2 * linux/fs/nfs/dir.c 3 * 4 * Copyright (C) 1992 Rick Sladkey 5 * 6 * nfs directory handling functions 7 * 8 * 10 Apr 1996 Added silly rename for unlink --okir 9 * 28 Sep 1996 Improved directory cache --okir 10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de 11 * Re-implemented silly rename for unlink, newly implemented 12 * silly rename for nfs_rename() following the suggestions 13 * of Olaf Kirch (okir) found in this file. 14 * Following Linus comments on my original hack, this version 15 * depends only on the dcache stuff and doesn't touch the inode 16 * layer (iput() and friends). 17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM 18 */ 19 20 #include <linux/time.h> 21 #include <linux/errno.h> 22 #include <linux/stat.h> 23 #include <linux/fcntl.h> 24 #include <linux/string.h> 25 #include <linux/kernel.h> 26 #include <linux/slab.h> 27 #include <linux/mm.h> 28 #include <linux/sunrpc/clnt.h> 29 #include <linux/nfs_fs.h> 30 #include <linux/nfs_mount.h> 31 #include <linux/pagemap.h> 32 #include <linux/smp_lock.h> 33 #include <linux/pagevec.h> 34 #include <linux/namei.h> 35 #include <linux/mount.h> 36 #include <linux/sched.h> 37 38 #include "nfs4_fs.h" 39 #include "delegation.h" 40 #include "iostat.h" 41 42 /* #define NFS_DEBUG_VERBOSE 1 */ 43 44 static int nfs_opendir(struct inode *, struct file *); 45 static int nfs_readdir(struct file *, void *, filldir_t); 46 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); 47 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); 48 static int nfs_mkdir(struct inode *, struct dentry *, int); 49 static int nfs_rmdir(struct inode *, struct dentry *); 50 static int nfs_unlink(struct inode *, struct dentry *); 51 static int nfs_symlink(struct inode *, struct dentry *, const char *); 52 static int nfs_link(struct dentry *, struct inode *, struct dentry *); 53 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); 54 static int nfs_rename(struct inode *, struct dentry *, 55 struct inode *, struct dentry *); 56 static int nfs_fsync_dir(struct file *, struct dentry *, int); 57 static loff_t nfs_llseek_dir(struct file *, loff_t, int); 58 59 const struct file_operations nfs_dir_operations = { 60 .llseek = nfs_llseek_dir, 61 .read = generic_read_dir, 62 .readdir = nfs_readdir, 63 .open = nfs_opendir, 64 .release = nfs_release, 65 .fsync = nfs_fsync_dir, 66 }; 67 68 const struct inode_operations nfs_dir_inode_operations = { 69 .create = nfs_create, 70 .lookup = nfs_lookup, 71 .link = nfs_link, 72 .unlink = nfs_unlink, 73 .symlink = nfs_symlink, 74 .mkdir = nfs_mkdir, 75 .rmdir = nfs_rmdir, 76 .mknod = nfs_mknod, 77 .rename = nfs_rename, 78 .permission = nfs_permission, 79 .getattr = nfs_getattr, 80 .setattr = nfs_setattr, 81 }; 82 83 #ifdef CONFIG_NFS_V3 84 const struct inode_operations nfs3_dir_inode_operations = { 85 .create = nfs_create, 86 .lookup = nfs_lookup, 87 .link = nfs_link, 88 .unlink = nfs_unlink, 89 .symlink = nfs_symlink, 90 .mkdir = nfs_mkdir, 91 .rmdir = nfs_rmdir, 92 .mknod = nfs_mknod, 93 .rename = nfs_rename, 94 .permission = nfs_permission, 95 .getattr = nfs_getattr, 96 .setattr = nfs_setattr, 97 .listxattr = nfs3_listxattr, 98 .getxattr = nfs3_getxattr, 99 .setxattr = nfs3_setxattr, 100 .removexattr = nfs3_removexattr, 101 }; 102 #endif /* CONFIG_NFS_V3 */ 103 104 #ifdef CONFIG_NFS_V4 105 106 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); 107 const struct inode_operations nfs4_dir_inode_operations = { 108 .create = nfs_create, 109 .lookup = nfs_atomic_lookup, 110 .link = nfs_link, 111 .unlink = nfs_unlink, 112 .symlink = nfs_symlink, 113 .mkdir = nfs_mkdir, 114 .rmdir = nfs_rmdir, 115 .mknod = nfs_mknod, 116 .rename = nfs_rename, 117 .permission = nfs_permission, 118 .getattr = nfs_getattr, 119 .setattr = nfs_setattr, 120 .getxattr = nfs4_getxattr, 121 .setxattr = nfs4_setxattr, 122 .listxattr = nfs4_listxattr, 123 }; 124 125 #endif /* CONFIG_NFS_V4 */ 126 127 /* 128 * Open file 129 */ 130 static int 131 nfs_opendir(struct inode *inode, struct file *filp) 132 { 133 int res; 134 135 dfprintk(VFS, "NFS: opendir(%s/%ld)\n", 136 inode->i_sb->s_id, inode->i_ino); 137 138 lock_kernel(); 139 /* Call generic open code in order to cache credentials */ 140 res = nfs_open(inode, filp); 141 unlock_kernel(); 142 return res; 143 } 144 145 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int); 146 typedef struct { 147 struct file *file; 148 struct page *page; 149 unsigned long page_index; 150 __be32 *ptr; 151 u64 *dir_cookie; 152 loff_t current_index; 153 struct nfs_entry *entry; 154 decode_dirent_t decode; 155 int plus; 156 int error; 157 unsigned long timestamp; 158 int timestamp_valid; 159 } nfs_readdir_descriptor_t; 160 161 /* Now we cache directories properly, by stuffing the dirent 162 * data directly in the page cache. 163 * 164 * Inode invalidation due to refresh etc. takes care of 165 * _everything_, no sloppy entry flushing logic, no extraneous 166 * copying, network direct to page cache, the way it was meant 167 * to be. 168 * 169 * NOTE: Dirent information verification is done always by the 170 * page-in of the RPC reply, nowhere else, this simplies 171 * things substantially. 172 */ 173 static 174 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page) 175 { 176 struct file *file = desc->file; 177 struct inode *inode = file->f_path.dentry->d_inode; 178 struct rpc_cred *cred = nfs_file_cred(file); 179 unsigned long timestamp; 180 int error; 181 182 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n", 183 __FUNCTION__, (long long)desc->entry->cookie, 184 page->index); 185 186 again: 187 timestamp = jiffies; 188 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page, 189 NFS_SERVER(inode)->dtsize, desc->plus); 190 if (error < 0) { 191 /* We requested READDIRPLUS, but the server doesn't grok it */ 192 if (error == -ENOTSUPP && desc->plus) { 193 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 194 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 195 desc->plus = 0; 196 goto again; 197 } 198 goto error; 199 } 200 desc->timestamp = timestamp; 201 desc->timestamp_valid = 1; 202 SetPageUptodate(page); 203 /* Ensure consistent page alignment of the data. 204 * Note: assumes we have exclusive access to this mapping either 205 * through inode->i_mutex or some other mechanism. 206 */ 207 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) { 208 /* Should never happen */ 209 nfs_zap_mapping(inode, inode->i_mapping); 210 } 211 unlock_page(page); 212 return 0; 213 error: 214 unlock_page(page); 215 desc->error = error; 216 return -EIO; 217 } 218 219 static inline 220 int dir_decode(nfs_readdir_descriptor_t *desc) 221 { 222 __be32 *p = desc->ptr; 223 p = desc->decode(p, desc->entry, desc->plus); 224 if (IS_ERR(p)) 225 return PTR_ERR(p); 226 desc->ptr = p; 227 if (desc->timestamp_valid) 228 desc->entry->fattr->time_start = desc->timestamp; 229 else 230 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR; 231 return 0; 232 } 233 234 static inline 235 void dir_page_release(nfs_readdir_descriptor_t *desc) 236 { 237 kunmap(desc->page); 238 page_cache_release(desc->page); 239 desc->page = NULL; 240 desc->ptr = NULL; 241 } 242 243 /* 244 * Given a pointer to a buffer that has already been filled by a call 245 * to readdir, find the next entry with cookie '*desc->dir_cookie'. 246 * 247 * If the end of the buffer has been reached, return -EAGAIN, if not, 248 * return the offset within the buffer of the next entry to be 249 * read. 250 */ 251 static inline 252 int find_dirent(nfs_readdir_descriptor_t *desc) 253 { 254 struct nfs_entry *entry = desc->entry; 255 int loop_count = 0, 256 status; 257 258 while((status = dir_decode(desc)) == 0) { 259 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n", 260 __FUNCTION__, (unsigned long long)entry->cookie); 261 if (entry->prev_cookie == *desc->dir_cookie) 262 break; 263 if (loop_count++ > 200) { 264 loop_count = 0; 265 schedule(); 266 } 267 } 268 return status; 269 } 270 271 /* 272 * Given a pointer to a buffer that has already been filled by a call 273 * to readdir, find the entry at offset 'desc->file->f_pos'. 274 * 275 * If the end of the buffer has been reached, return -EAGAIN, if not, 276 * return the offset within the buffer of the next entry to be 277 * read. 278 */ 279 static inline 280 int find_dirent_index(nfs_readdir_descriptor_t *desc) 281 { 282 struct nfs_entry *entry = desc->entry; 283 int loop_count = 0, 284 status; 285 286 for(;;) { 287 status = dir_decode(desc); 288 if (status) 289 break; 290 291 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n", 292 (unsigned long long)entry->cookie, desc->current_index); 293 294 if (desc->file->f_pos == desc->current_index) { 295 *desc->dir_cookie = entry->cookie; 296 break; 297 } 298 desc->current_index++; 299 if (loop_count++ > 200) { 300 loop_count = 0; 301 schedule(); 302 } 303 } 304 return status; 305 } 306 307 /* 308 * Find the given page, and call find_dirent() or find_dirent_index in 309 * order to try to return the next entry. 310 */ 311 static inline 312 int find_dirent_page(nfs_readdir_descriptor_t *desc) 313 { 314 struct inode *inode = desc->file->f_path.dentry->d_inode; 315 struct page *page; 316 int status; 317 318 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n", 319 __FUNCTION__, desc->page_index, 320 (long long) *desc->dir_cookie); 321 322 /* If we find the page in the page_cache, we cannot be sure 323 * how fresh the data is, so we will ignore readdir_plus attributes. 324 */ 325 desc->timestamp_valid = 0; 326 page = read_cache_page(inode->i_mapping, desc->page_index, 327 (filler_t *)nfs_readdir_filler, desc); 328 if (IS_ERR(page)) { 329 status = PTR_ERR(page); 330 goto out; 331 } 332 333 /* NOTE: Someone else may have changed the READDIRPLUS flag */ 334 desc->page = page; 335 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 336 if (*desc->dir_cookie != 0) 337 status = find_dirent(desc); 338 else 339 status = find_dirent_index(desc); 340 if (status < 0) 341 dir_page_release(desc); 342 out: 343 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status); 344 return status; 345 } 346 347 /* 348 * Recurse through the page cache pages, and return a 349 * filled nfs_entry structure of the next directory entry if possible. 350 * 351 * The target for the search is '*desc->dir_cookie' if non-0, 352 * 'desc->file->f_pos' otherwise 353 */ 354 static inline 355 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 356 { 357 int loop_count = 0; 358 int res; 359 360 /* Always search-by-index from the beginning of the cache */ 361 if (*desc->dir_cookie == 0) { 362 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n", 363 (long long)desc->file->f_pos); 364 desc->page_index = 0; 365 desc->entry->cookie = desc->entry->prev_cookie = 0; 366 desc->entry->eof = 0; 367 desc->current_index = 0; 368 } else 369 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", 370 (unsigned long long)*desc->dir_cookie); 371 372 for (;;) { 373 res = find_dirent_page(desc); 374 if (res != -EAGAIN) 375 break; 376 /* Align to beginning of next page */ 377 desc->page_index ++; 378 if (loop_count++ > 200) { 379 loop_count = 0; 380 schedule(); 381 } 382 } 383 384 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res); 385 return res; 386 } 387 388 static inline unsigned int dt_type(struct inode *inode) 389 { 390 return (inode->i_mode >> 12) & 15; 391 } 392 393 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc); 394 395 /* 396 * Once we've found the start of the dirent within a page: fill 'er up... 397 */ 398 static 399 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, 400 filldir_t filldir) 401 { 402 struct file *file = desc->file; 403 struct nfs_entry *entry = desc->entry; 404 struct dentry *dentry = NULL; 405 u64 fileid; 406 int loop_count = 0, 407 res; 408 409 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", 410 (unsigned long long)entry->cookie); 411 412 for(;;) { 413 unsigned d_type = DT_UNKNOWN; 414 /* Note: entry->prev_cookie contains the cookie for 415 * retrieving the current dirent on the server */ 416 fileid = entry->ino; 417 418 /* Get a dentry if we have one */ 419 if (dentry != NULL) 420 dput(dentry); 421 dentry = nfs_readdir_lookup(desc); 422 423 /* Use readdirplus info */ 424 if (dentry != NULL && dentry->d_inode != NULL) { 425 d_type = dt_type(dentry->d_inode); 426 fileid = NFS_FILEID(dentry->d_inode); 427 } 428 429 res = filldir(dirent, entry->name, entry->len, 430 file->f_pos, fileid, d_type); 431 if (res < 0) 432 break; 433 file->f_pos++; 434 *desc->dir_cookie = entry->cookie; 435 if (dir_decode(desc) != 0) { 436 desc->page_index ++; 437 break; 438 } 439 if (loop_count++ > 200) { 440 loop_count = 0; 441 schedule(); 442 } 443 } 444 dir_page_release(desc); 445 if (dentry != NULL) 446 dput(dentry); 447 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 448 (unsigned long long)*desc->dir_cookie, res); 449 return res; 450 } 451 452 /* 453 * If we cannot find a cookie in our cache, we suspect that this is 454 * because it points to a deleted file, so we ask the server to return 455 * whatever it thinks is the next entry. We then feed this to filldir. 456 * If all goes well, we should then be able to find our way round the 457 * cache on the next call to readdir_search_pagecache(); 458 * 459 * NOTE: we cannot add the anonymous page to the pagecache because 460 * the data it contains might not be page aligned. Besides, 461 * we should already have a complete representation of the 462 * directory in the page cache by the time we get here. 463 */ 464 static inline 465 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, 466 filldir_t filldir) 467 { 468 struct file *file = desc->file; 469 struct inode *inode = file->f_path.dentry->d_inode; 470 struct rpc_cred *cred = nfs_file_cred(file); 471 struct page *page = NULL; 472 int status; 473 unsigned long timestamp; 474 475 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 476 (unsigned long long)*desc->dir_cookie); 477 478 page = alloc_page(GFP_HIGHUSER); 479 if (!page) { 480 status = -ENOMEM; 481 goto out; 482 } 483 timestamp = jiffies; 484 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie, 485 page, 486 NFS_SERVER(inode)->dtsize, 487 desc->plus); 488 desc->page = page; 489 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 490 if (desc->error >= 0) { 491 desc->timestamp = timestamp; 492 desc->timestamp_valid = 1; 493 if ((status = dir_decode(desc)) == 0) 494 desc->entry->prev_cookie = *desc->dir_cookie; 495 } else 496 status = -EIO; 497 if (status < 0) 498 goto out_release; 499 500 status = nfs_do_filldir(desc, dirent, filldir); 501 502 /* Reset read descriptor so it searches the page cache from 503 * the start upon the next call to readdir_search_pagecache() */ 504 desc->page_index = 0; 505 desc->entry->cookie = desc->entry->prev_cookie = 0; 506 desc->entry->eof = 0; 507 out: 508 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 509 __FUNCTION__, status); 510 return status; 511 out_release: 512 dir_page_release(desc); 513 goto out; 514 } 515 516 /* The file offset position represents the dirent entry number. A 517 last cookie cache takes care of the common case of reading the 518 whole directory. 519 */ 520 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) 521 { 522 struct dentry *dentry = filp->f_path.dentry; 523 struct inode *inode = dentry->d_inode; 524 nfs_readdir_descriptor_t my_desc, 525 *desc = &my_desc; 526 struct nfs_entry my_entry; 527 struct nfs_fh fh; 528 struct nfs_fattr fattr; 529 long res; 530 531 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n", 532 dentry->d_parent->d_name.name, dentry->d_name.name, 533 (long long)filp->f_pos); 534 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 535 536 lock_kernel(); 537 538 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping); 539 if (res < 0) { 540 unlock_kernel(); 541 return res; 542 } 543 544 /* 545 * filp->f_pos points to the dirent entry number. 546 * *desc->dir_cookie has the cookie for the next entry. We have 547 * to either find the entry with the appropriate number or 548 * revalidate the cookie. 549 */ 550 memset(desc, 0, sizeof(*desc)); 551 552 desc->file = filp; 553 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; 554 desc->decode = NFS_PROTO(inode)->decode_dirent; 555 desc->plus = NFS_USE_READDIRPLUS(inode); 556 557 my_entry.cookie = my_entry.prev_cookie = 0; 558 my_entry.eof = 0; 559 my_entry.fh = &fh; 560 my_entry.fattr = &fattr; 561 nfs_fattr_init(&fattr); 562 desc->entry = &my_entry; 563 564 while(!desc->entry->eof) { 565 res = readdir_search_pagecache(desc); 566 567 if (res == -EBADCOOKIE) { 568 /* This means either end of directory */ 569 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { 570 /* Or that the server has 'lost' a cookie */ 571 res = uncached_readdir(desc, dirent, filldir); 572 if (res >= 0) 573 continue; 574 } 575 res = 0; 576 break; 577 } 578 if (res == -ETOOSMALL && desc->plus) { 579 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 580 nfs_zap_caches(inode); 581 desc->plus = 0; 582 desc->entry->eof = 0; 583 continue; 584 } 585 if (res < 0) 586 break; 587 588 res = nfs_do_filldir(desc, dirent, filldir); 589 if (res < 0) { 590 res = 0; 591 break; 592 } 593 } 594 unlock_kernel(); 595 if (res > 0) 596 res = 0; 597 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n", 598 dentry->d_parent->d_name.name, dentry->d_name.name, 599 res); 600 return res; 601 } 602 603 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) 604 { 605 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex); 606 switch (origin) { 607 case 1: 608 offset += filp->f_pos; 609 case 0: 610 if (offset >= 0) 611 break; 612 default: 613 offset = -EINVAL; 614 goto out; 615 } 616 if (offset != filp->f_pos) { 617 filp->f_pos = offset; 618 nfs_file_open_context(filp)->dir_cookie = 0; 619 } 620 out: 621 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex); 622 return offset; 623 } 624 625 /* 626 * All directory operations under NFS are synchronous, so fsync() 627 * is a dummy operation. 628 */ 629 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) 630 { 631 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n", 632 dentry->d_parent->d_name.name, dentry->d_name.name, 633 datasync); 634 635 return 0; 636 } 637 638 /* 639 * A check for whether or not the parent directory has changed. 640 * In the case it has, we assume that the dentries are untrustworthy 641 * and may need to be looked up again. 642 */ 643 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) 644 { 645 if (IS_ROOT(dentry)) 646 return 1; 647 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 648 return 0; 649 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 650 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 651 return 0; 652 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 653 return 0; 654 return 1; 655 } 656 657 /* 658 * Return the intent data that applies to this particular path component 659 * 660 * Note that the current set of intents only apply to the very last 661 * component of the path. 662 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. 663 */ 664 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) 665 { 666 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) 667 return 0; 668 return nd->flags & mask; 669 } 670 671 /* 672 * Use intent information to check whether or not we're going to do 673 * an O_EXCL create using this path component. 674 */ 675 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) 676 { 677 if (NFS_PROTO(dir)->version == 2) 678 return 0; 679 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0) 680 return 0; 681 return (nd->intent.open.flags & O_EXCL) != 0; 682 } 683 684 /* 685 * Inode and filehandle revalidation for lookups. 686 * 687 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 688 * or if the intent information indicates that we're about to open this 689 * particular file and the "nocto" mount flag is not set. 690 * 691 */ 692 static inline 693 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) 694 { 695 struct nfs_server *server = NFS_SERVER(inode); 696 697 if (nd != NULL) { 698 /* VFS wants an on-the-wire revalidation */ 699 if (nd->flags & LOOKUP_REVAL) 700 goto out_force; 701 /* This is an open(2) */ 702 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && 703 !(server->flags & NFS_MOUNT_NOCTO) && 704 (S_ISREG(inode->i_mode) || 705 S_ISDIR(inode->i_mode))) 706 goto out_force; 707 return 0; 708 } 709 return nfs_revalidate_inode(server, inode); 710 out_force: 711 return __nfs_revalidate_inode(server, inode); 712 } 713 714 /* 715 * We judge how long we want to trust negative 716 * dentries by looking at the parent inode mtime. 717 * 718 * If parent mtime has changed, we revalidate, else we wait for a 719 * period corresponding to the parent's attribute cache timeout value. 720 */ 721 static inline 722 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 723 struct nameidata *nd) 724 { 725 /* Don't revalidate a negative dentry if we're creating a new file */ 726 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) 727 return 0; 728 return !nfs_check_verifier(dir, dentry); 729 } 730 731 /* 732 * This is called every time the dcache has a lookup hit, 733 * and we should check whether we can really trust that 734 * lookup. 735 * 736 * NOTE! The hit can be a negative hit too, don't assume 737 * we have an inode! 738 * 739 * If the parent directory is seen to have changed, we throw out the 740 * cached dentry and do a new lookup. 741 */ 742 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) 743 { 744 struct inode *dir; 745 struct inode *inode; 746 struct dentry *parent; 747 int error; 748 struct nfs_fh fhandle; 749 struct nfs_fattr fattr; 750 751 parent = dget_parent(dentry); 752 lock_kernel(); 753 dir = parent->d_inode; 754 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 755 inode = dentry->d_inode; 756 757 if (!inode) { 758 if (nfs_neg_need_reval(dir, dentry, nd)) 759 goto out_bad; 760 goto out_valid; 761 } 762 763 if (is_bad_inode(inode)) { 764 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", 765 __FUNCTION__, dentry->d_parent->d_name.name, 766 dentry->d_name.name); 767 goto out_bad; 768 } 769 770 /* Force a full look up iff the parent directory has changed */ 771 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { 772 if (nfs_lookup_verify_inode(inode, nd)) 773 goto out_zap_parent; 774 goto out_valid; 775 } 776 777 if (NFS_STALE(inode)) 778 goto out_bad; 779 780 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 781 if (error) 782 goto out_bad; 783 if (nfs_compare_fh(NFS_FH(inode), &fhandle)) 784 goto out_bad; 785 if ((error = nfs_refresh_inode(inode, &fattr)) != 0) 786 goto out_bad; 787 788 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 789 out_valid: 790 unlock_kernel(); 791 dput(parent); 792 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", 793 __FUNCTION__, dentry->d_parent->d_name.name, 794 dentry->d_name.name); 795 return 1; 796 out_zap_parent: 797 nfs_zap_caches(dir); 798 out_bad: 799 nfs_mark_for_revalidate(dir); 800 if (inode && S_ISDIR(inode->i_mode)) { 801 /* Purge readdir caches. */ 802 nfs_zap_caches(inode); 803 /* If we have submounts, don't unhash ! */ 804 if (have_submounts(dentry)) 805 goto out_valid; 806 shrink_dcache_parent(dentry); 807 } 808 d_drop(dentry); 809 unlock_kernel(); 810 dput(parent); 811 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", 812 __FUNCTION__, dentry->d_parent->d_name.name, 813 dentry->d_name.name); 814 return 0; 815 } 816 817 /* 818 * This is called from dput() when d_count is going to 0. 819 */ 820 static int nfs_dentry_delete(struct dentry *dentry) 821 { 822 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", 823 dentry->d_parent->d_name.name, dentry->d_name.name, 824 dentry->d_flags); 825 826 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 827 /* Unhash it, so that ->d_iput() would be called */ 828 return 1; 829 } 830 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 831 /* Unhash it, so that ancestors of killed async unlink 832 * files will be cleaned up during umount */ 833 return 1; 834 } 835 return 0; 836 837 } 838 839 /* 840 * Called when the dentry loses inode. 841 * We use it to clean up silly-renamed files. 842 */ 843 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 844 { 845 nfs_inode_return_delegation(inode); 846 if (S_ISDIR(inode->i_mode)) 847 /* drop any readdir cache as it could easily be old */ 848 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 849 850 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 851 lock_kernel(); 852 drop_nlink(inode); 853 nfs_complete_unlink(dentry, inode); 854 unlock_kernel(); 855 } 856 iput(inode); 857 } 858 859 struct dentry_operations nfs_dentry_operations = { 860 .d_revalidate = nfs_lookup_revalidate, 861 .d_delete = nfs_dentry_delete, 862 .d_iput = nfs_dentry_iput, 863 }; 864 865 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 866 { 867 struct dentry *res; 868 struct inode *inode = NULL; 869 int error; 870 struct nfs_fh fhandle; 871 struct nfs_fattr fattr; 872 873 dfprintk(VFS, "NFS: lookup(%s/%s)\n", 874 dentry->d_parent->d_name.name, dentry->d_name.name); 875 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 876 877 res = ERR_PTR(-ENAMETOOLONG); 878 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 879 goto out; 880 881 res = ERR_PTR(-ENOMEM); 882 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 883 884 lock_kernel(); 885 886 /* 887 * If we're doing an exclusive create, optimize away the lookup 888 * but don't hash the dentry. 889 */ 890 if (nfs_is_exclusive_create(dir, nd)) { 891 d_instantiate(dentry, NULL); 892 res = NULL; 893 goto out_unlock; 894 } 895 896 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 897 if (error == -ENOENT) 898 goto no_entry; 899 if (error < 0) { 900 res = ERR_PTR(error); 901 goto out_unlock; 902 } 903 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); 904 res = (struct dentry *)inode; 905 if (IS_ERR(res)) 906 goto out_unlock; 907 908 no_entry: 909 res = d_materialise_unique(dentry, inode); 910 if (res != NULL) { 911 if (IS_ERR(res)) 912 goto out_unlock; 913 dentry = res; 914 } 915 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 916 out_unlock: 917 unlock_kernel(); 918 out: 919 return res; 920 } 921 922 #ifdef CONFIG_NFS_V4 923 static int nfs_open_revalidate(struct dentry *, struct nameidata *); 924 925 struct dentry_operations nfs4_dentry_operations = { 926 .d_revalidate = nfs_open_revalidate, 927 .d_delete = nfs_dentry_delete, 928 .d_iput = nfs_dentry_iput, 929 }; 930 931 /* 932 * Use intent information to determine whether we need to substitute 933 * the NFSv4-style stateful OPEN for the LOOKUP call 934 */ 935 static int is_atomic_open(struct inode *dir, struct nameidata *nd) 936 { 937 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) 938 return 0; 939 /* NFS does not (yet) have a stateful open for directories */ 940 if (nd->flags & LOOKUP_DIRECTORY) 941 return 0; 942 /* Are we trying to write to a read only partition? */ 943 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) 944 return 0; 945 return 1; 946 } 947 948 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 949 { 950 struct dentry *res = NULL; 951 int error; 952 953 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", 954 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 955 956 /* Check that we are indeed trying to open this file */ 957 if (!is_atomic_open(dir, nd)) 958 goto no_open; 959 960 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { 961 res = ERR_PTR(-ENAMETOOLONG); 962 goto out; 963 } 964 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 965 966 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash 967 * the dentry. */ 968 if (nd->intent.open.flags & O_EXCL) { 969 d_instantiate(dentry, NULL); 970 goto out; 971 } 972 973 /* Open the file on the server */ 974 lock_kernel(); 975 res = nfs4_atomic_open(dir, dentry, nd); 976 unlock_kernel(); 977 if (IS_ERR(res)) { 978 error = PTR_ERR(res); 979 switch (error) { 980 /* Make a negative dentry */ 981 case -ENOENT: 982 res = NULL; 983 goto out; 984 /* This turned out not to be a regular file */ 985 case -EISDIR: 986 case -ENOTDIR: 987 goto no_open; 988 case -ELOOP: 989 if (!(nd->intent.open.flags & O_NOFOLLOW)) 990 goto no_open; 991 /* case -EINVAL: */ 992 default: 993 goto out; 994 } 995 } else if (res != NULL) 996 dentry = res; 997 out: 998 return res; 999 no_open: 1000 return nfs_lookup(dir, dentry, nd); 1001 } 1002 1003 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) 1004 { 1005 struct dentry *parent = NULL; 1006 struct inode *inode = dentry->d_inode; 1007 struct inode *dir; 1008 int openflags, ret = 0; 1009 1010 parent = dget_parent(dentry); 1011 dir = parent->d_inode; 1012 if (!is_atomic_open(dir, nd)) 1013 goto no_open; 1014 /* We can't create new files in nfs_open_revalidate(), so we 1015 * optimize away revalidation of negative dentries. 1016 */ 1017 if (inode == NULL) { 1018 if (!nfs_neg_need_reval(dir, dentry, nd)) 1019 ret = 1; 1020 goto out; 1021 } 1022 1023 /* NFS only supports OPEN on regular files */ 1024 if (!S_ISREG(inode->i_mode)) 1025 goto no_open; 1026 openflags = nd->intent.open.flags; 1027 /* We cannot do exclusive creation on a positive dentry */ 1028 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) 1029 goto no_open; 1030 /* We can't create new files, or truncate existing ones here */ 1031 openflags &= ~(O_CREAT|O_TRUNC); 1032 1033 /* 1034 * Note: we're not holding inode->i_mutex and so may be racing with 1035 * operations that change the directory. We therefore save the 1036 * change attribute *before* we do the RPC call. 1037 */ 1038 lock_kernel(); 1039 ret = nfs4_open_revalidate(dir, dentry, openflags, nd); 1040 unlock_kernel(); 1041 out: 1042 dput(parent); 1043 if (!ret) 1044 d_drop(dentry); 1045 return ret; 1046 no_open: 1047 dput(parent); 1048 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) 1049 return 1; 1050 return nfs_lookup_revalidate(dentry, nd); 1051 } 1052 #endif /* CONFIG_NFSV4 */ 1053 1054 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) 1055 { 1056 struct dentry *parent = desc->file->f_path.dentry; 1057 struct inode *dir = parent->d_inode; 1058 struct nfs_entry *entry = desc->entry; 1059 struct dentry *dentry, *alias; 1060 struct qstr name = { 1061 .name = entry->name, 1062 .len = entry->len, 1063 }; 1064 struct inode *inode; 1065 unsigned long verf = nfs_save_change_attribute(dir); 1066 1067 switch (name.len) { 1068 case 2: 1069 if (name.name[0] == '.' && name.name[1] == '.') 1070 return dget_parent(parent); 1071 break; 1072 case 1: 1073 if (name.name[0] == '.') 1074 return dget(parent); 1075 } 1076 1077 spin_lock(&dir->i_lock); 1078 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) { 1079 spin_unlock(&dir->i_lock); 1080 return NULL; 1081 } 1082 spin_unlock(&dir->i_lock); 1083 1084 name.hash = full_name_hash(name.name, name.len); 1085 dentry = d_lookup(parent, &name); 1086 if (dentry != NULL) { 1087 /* Is this a positive dentry that matches the readdir info? */ 1088 if (dentry->d_inode != NULL && 1089 (NFS_FILEID(dentry->d_inode) == entry->ino || 1090 d_mountpoint(dentry))) { 1091 if (!desc->plus || entry->fh->size == 0) 1092 return dentry; 1093 if (nfs_compare_fh(NFS_FH(dentry->d_inode), 1094 entry->fh) == 0) 1095 goto out_renew; 1096 } 1097 /* No, so d_drop to allow one to be created */ 1098 d_drop(dentry); 1099 dput(dentry); 1100 } 1101 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) 1102 return NULL; 1103 if (name.len > NFS_SERVER(dir)->namelen) 1104 return NULL; 1105 /* Note: caller is already holding the dir->i_mutex! */ 1106 dentry = d_alloc(parent, &name); 1107 if (dentry == NULL) 1108 return NULL; 1109 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1110 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 1111 if (IS_ERR(inode)) { 1112 dput(dentry); 1113 return NULL; 1114 } 1115 1116 alias = d_materialise_unique(dentry, inode); 1117 if (alias != NULL) { 1118 dput(dentry); 1119 if (IS_ERR(alias)) 1120 return NULL; 1121 dentry = alias; 1122 } 1123 1124 out_renew: 1125 nfs_set_verifier(dentry, verf); 1126 return dentry; 1127 } 1128 1129 /* 1130 * Code common to create, mkdir, and mknod. 1131 */ 1132 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1133 struct nfs_fattr *fattr) 1134 { 1135 struct dentry *parent = dget_parent(dentry); 1136 struct inode *dir = parent->d_inode; 1137 struct inode *inode; 1138 int error = -EACCES; 1139 1140 d_drop(dentry); 1141 1142 /* We may have been initialized further down */ 1143 if (dentry->d_inode) 1144 goto out; 1145 if (fhandle->size == 0) { 1146 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1147 if (error) 1148 goto out_error; 1149 } 1150 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1151 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1152 struct nfs_server *server = NFS_SB(dentry->d_sb); 1153 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1154 if (error < 0) 1155 goto out_error; 1156 } 1157 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1158 error = PTR_ERR(inode); 1159 if (IS_ERR(inode)) 1160 goto out_error; 1161 d_add(dentry, inode); 1162 out: 1163 dput(parent); 1164 return 0; 1165 out_error: 1166 nfs_mark_for_revalidate(dir); 1167 dput(parent); 1168 return error; 1169 } 1170 1171 /* 1172 * Following a failed create operation, we drop the dentry rather 1173 * than retain a negative dentry. This avoids a problem in the event 1174 * that the operation succeeded on the server, but an error in the 1175 * reply path made it appear to have failed. 1176 */ 1177 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, 1178 struct nameidata *nd) 1179 { 1180 struct iattr attr; 1181 int error; 1182 int open_flags = 0; 1183 1184 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1185 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1186 1187 attr.ia_mode = mode; 1188 attr.ia_valid = ATTR_MODE; 1189 1190 if ((nd->flags & LOOKUP_CREATE) != 0) 1191 open_flags = nd->intent.open.flags; 1192 1193 lock_kernel(); 1194 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); 1195 if (error != 0) 1196 goto out_err; 1197 unlock_kernel(); 1198 return 0; 1199 out_err: 1200 unlock_kernel(); 1201 d_drop(dentry); 1202 return error; 1203 } 1204 1205 /* 1206 * See comments for nfs_proc_create regarding failed operations. 1207 */ 1208 static int 1209 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) 1210 { 1211 struct iattr attr; 1212 int status; 1213 1214 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1215 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1216 1217 if (!new_valid_dev(rdev)) 1218 return -EINVAL; 1219 1220 attr.ia_mode = mode; 1221 attr.ia_valid = ATTR_MODE; 1222 1223 lock_kernel(); 1224 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1225 if (status != 0) 1226 goto out_err; 1227 unlock_kernel(); 1228 return 0; 1229 out_err: 1230 unlock_kernel(); 1231 d_drop(dentry); 1232 return status; 1233 } 1234 1235 /* 1236 * See comments for nfs_proc_create regarding failed operations. 1237 */ 1238 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1239 { 1240 struct iattr attr; 1241 int error; 1242 1243 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1244 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1245 1246 attr.ia_valid = ATTR_MODE; 1247 attr.ia_mode = mode | S_IFDIR; 1248 1249 lock_kernel(); 1250 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1251 if (error != 0) 1252 goto out_err; 1253 unlock_kernel(); 1254 return 0; 1255 out_err: 1256 d_drop(dentry); 1257 unlock_kernel(); 1258 return error; 1259 } 1260 1261 static int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1262 { 1263 int error; 1264 1265 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1266 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1267 1268 lock_kernel(); 1269 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1270 /* Ensure the VFS deletes this inode */ 1271 if (error == 0 && dentry->d_inode != NULL) 1272 clear_nlink(dentry->d_inode); 1273 unlock_kernel(); 1274 1275 return error; 1276 } 1277 1278 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) 1279 { 1280 static unsigned int sillycounter; 1281 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2; 1282 const int countersize = sizeof(sillycounter)*2; 1283 const int slen = sizeof(".nfs")+fileidsize+countersize-1; 1284 char silly[slen+1]; 1285 struct qstr qsilly; 1286 struct dentry *sdentry; 1287 int error = -EIO; 1288 1289 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", 1290 dentry->d_parent->d_name.name, dentry->d_name.name, 1291 atomic_read(&dentry->d_count)); 1292 nfs_inc_stats(dir, NFSIOS_SILLYRENAME); 1293 1294 /* 1295 * We don't allow a dentry to be silly-renamed twice. 1296 */ 1297 error = -EBUSY; 1298 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1299 goto out; 1300 1301 sprintf(silly, ".nfs%*.*Lx", 1302 fileidsize, fileidsize, 1303 (unsigned long long)NFS_FILEID(dentry->d_inode)); 1304 1305 /* Return delegation in anticipation of the rename */ 1306 nfs_inode_return_delegation(dentry->d_inode); 1307 1308 sdentry = NULL; 1309 do { 1310 char *suffix = silly + slen - countersize; 1311 1312 dput(sdentry); 1313 sillycounter++; 1314 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); 1315 1316 dfprintk(VFS, "NFS: trying to rename %s to %s\n", 1317 dentry->d_name.name, silly); 1318 1319 sdentry = lookup_one_len(silly, dentry->d_parent, slen); 1320 /* 1321 * N.B. Better to return EBUSY here ... it could be 1322 * dangerous to delete the file while it's in use. 1323 */ 1324 if (IS_ERR(sdentry)) 1325 goto out; 1326 } while(sdentry->d_inode != NULL); /* need negative lookup */ 1327 1328 qsilly.name = silly; 1329 qsilly.len = strlen(silly); 1330 if (dentry->d_inode) { 1331 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1332 dir, &qsilly); 1333 nfs_mark_for_revalidate(dentry->d_inode); 1334 } else 1335 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1336 dir, &qsilly); 1337 if (!error) { 1338 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1339 d_move(dentry, sdentry); 1340 error = nfs_async_unlink(dir, dentry); 1341 /* If we return 0 we don't unlink */ 1342 } 1343 dput(sdentry); 1344 out: 1345 return error; 1346 } 1347 1348 /* 1349 * Remove a file after making sure there are no pending writes, 1350 * and after checking that the file has only one user. 1351 * 1352 * We invalidate the attribute cache and free the inode prior to the operation 1353 * to avoid possible races if the server reuses the inode. 1354 */ 1355 static int nfs_safe_remove(struct dentry *dentry) 1356 { 1357 struct inode *dir = dentry->d_parent->d_inode; 1358 struct inode *inode = dentry->d_inode; 1359 int error = -EBUSY; 1360 1361 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1362 dentry->d_parent->d_name.name, dentry->d_name.name); 1363 1364 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1365 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1366 error = 0; 1367 goto out; 1368 } 1369 1370 if (inode != NULL) { 1371 nfs_inode_return_delegation(inode); 1372 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1373 /* The VFS may want to delete this inode */ 1374 if (error == 0) 1375 drop_nlink(inode); 1376 nfs_mark_for_revalidate(inode); 1377 } else 1378 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1379 out: 1380 return error; 1381 } 1382 1383 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1384 * belongs to an active ".nfs..." file and we return -EBUSY. 1385 * 1386 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1387 */ 1388 static int nfs_unlink(struct inode *dir, struct dentry *dentry) 1389 { 1390 int error; 1391 int need_rehash = 0; 1392 1393 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1394 dir->i_ino, dentry->d_name.name); 1395 1396 lock_kernel(); 1397 spin_lock(&dcache_lock); 1398 spin_lock(&dentry->d_lock); 1399 if (atomic_read(&dentry->d_count) > 1) { 1400 spin_unlock(&dentry->d_lock); 1401 spin_unlock(&dcache_lock); 1402 /* Start asynchronous writeout of the inode */ 1403 write_inode_now(dentry->d_inode, 0); 1404 error = nfs_sillyrename(dir, dentry); 1405 unlock_kernel(); 1406 return error; 1407 } 1408 if (!d_unhashed(dentry)) { 1409 __d_drop(dentry); 1410 need_rehash = 1; 1411 } 1412 spin_unlock(&dentry->d_lock); 1413 spin_unlock(&dcache_lock); 1414 error = nfs_safe_remove(dentry); 1415 if (!error) { 1416 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1417 } else if (need_rehash) 1418 d_rehash(dentry); 1419 unlock_kernel(); 1420 return error; 1421 } 1422 1423 /* 1424 * To create a symbolic link, most file systems instantiate a new inode, 1425 * add a page to it containing the path, then write it out to the disk 1426 * using prepare_write/commit_write. 1427 * 1428 * Unfortunately the NFS client can't create the in-core inode first 1429 * because it needs a file handle to create an in-core inode (see 1430 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1431 * symlink request has completed on the server. 1432 * 1433 * So instead we allocate a raw page, copy the symname into it, then do 1434 * the SYMLINK request with the page as the buffer. If it succeeds, we 1435 * now have a new file handle and can instantiate an in-core NFS inode 1436 * and move the raw page into its mapping. 1437 */ 1438 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1439 { 1440 struct pagevec lru_pvec; 1441 struct page *page; 1442 char *kaddr; 1443 struct iattr attr; 1444 unsigned int pathlen = strlen(symname); 1445 int error; 1446 1447 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1448 dir->i_ino, dentry->d_name.name, symname); 1449 1450 if (pathlen > PAGE_SIZE) 1451 return -ENAMETOOLONG; 1452 1453 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1454 attr.ia_valid = ATTR_MODE; 1455 1456 lock_kernel(); 1457 1458 page = alloc_page(GFP_HIGHUSER); 1459 if (!page) { 1460 unlock_kernel(); 1461 return -ENOMEM; 1462 } 1463 1464 kaddr = kmap_atomic(page, KM_USER0); 1465 memcpy(kaddr, symname, pathlen); 1466 if (pathlen < PAGE_SIZE) 1467 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1468 kunmap_atomic(kaddr, KM_USER0); 1469 1470 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1471 if (error != 0) { 1472 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1473 dir->i_sb->s_id, dir->i_ino, 1474 dentry->d_name.name, symname, error); 1475 d_drop(dentry); 1476 __free_page(page); 1477 unlock_kernel(); 1478 return error; 1479 } 1480 1481 /* 1482 * No big deal if we can't add this page to the page cache here. 1483 * READLINK will get the missing page from the server if needed. 1484 */ 1485 pagevec_init(&lru_pvec, 0); 1486 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1487 GFP_KERNEL)) { 1488 pagevec_add(&lru_pvec, page); 1489 pagevec_lru_add(&lru_pvec); 1490 SetPageUptodate(page); 1491 unlock_page(page); 1492 } else 1493 __free_page(page); 1494 1495 unlock_kernel(); 1496 return 0; 1497 } 1498 1499 static int 1500 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1501 { 1502 struct inode *inode = old_dentry->d_inode; 1503 int error; 1504 1505 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1506 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1507 dentry->d_parent->d_name.name, dentry->d_name.name); 1508 1509 lock_kernel(); 1510 d_drop(dentry); 1511 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1512 if (error == 0) { 1513 atomic_inc(&inode->i_count); 1514 d_add(dentry, inode); 1515 } 1516 unlock_kernel(); 1517 return error; 1518 } 1519 1520 /* 1521 * RENAME 1522 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1523 * different file handle for the same inode after a rename (e.g. when 1524 * moving to a different directory). A fail-safe method to do so would 1525 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1526 * rename the old file using the sillyrename stuff. This way, the original 1527 * file in old_dir will go away when the last process iput()s the inode. 1528 * 1529 * FIXED. 1530 * 1531 * It actually works quite well. One needs to have the possibility for 1532 * at least one ".nfs..." file in each directory the file ever gets 1533 * moved or linked to which happens automagically with the new 1534 * implementation that only depends on the dcache stuff instead of 1535 * using the inode layer 1536 * 1537 * Unfortunately, things are a little more complicated than indicated 1538 * above. For a cross-directory move, we want to make sure we can get 1539 * rid of the old inode after the operation. This means there must be 1540 * no pending writes (if it's a file), and the use count must be 1. 1541 * If these conditions are met, we can drop the dentries before doing 1542 * the rename. 1543 */ 1544 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1545 struct inode *new_dir, struct dentry *new_dentry) 1546 { 1547 struct inode *old_inode = old_dentry->d_inode; 1548 struct inode *new_inode = new_dentry->d_inode; 1549 struct dentry *dentry = NULL, *rehash = NULL; 1550 int error = -EBUSY; 1551 1552 /* 1553 * To prevent any new references to the target during the rename, 1554 * we unhash the dentry and free the inode in advance. 1555 */ 1556 lock_kernel(); 1557 if (!d_unhashed(new_dentry)) { 1558 d_drop(new_dentry); 1559 rehash = new_dentry; 1560 } 1561 1562 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1563 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1564 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1565 atomic_read(&new_dentry->d_count)); 1566 1567 /* 1568 * First check whether the target is busy ... we can't 1569 * safely do _any_ rename if the target is in use. 1570 * 1571 * For files, make a copy of the dentry and then do a 1572 * silly-rename. If the silly-rename succeeds, the 1573 * copied dentry is hashed and becomes the new target. 1574 */ 1575 if (!new_inode) 1576 goto go_ahead; 1577 if (S_ISDIR(new_inode->i_mode)) { 1578 error = -EISDIR; 1579 if (!S_ISDIR(old_inode->i_mode)) 1580 goto out; 1581 } else if (atomic_read(&new_dentry->d_count) > 2) { 1582 int err; 1583 /* copy the target dentry's name */ 1584 dentry = d_alloc(new_dentry->d_parent, 1585 &new_dentry->d_name); 1586 if (!dentry) 1587 goto out; 1588 1589 /* silly-rename the existing target ... */ 1590 err = nfs_sillyrename(new_dir, new_dentry); 1591 if (!err) { 1592 new_dentry = rehash = dentry; 1593 new_inode = NULL; 1594 /* instantiate the replacement target */ 1595 d_instantiate(new_dentry, NULL); 1596 } else if (atomic_read(&new_dentry->d_count) > 1) 1597 /* dentry still busy? */ 1598 goto out; 1599 } else 1600 drop_nlink(new_inode); 1601 1602 go_ahead: 1603 /* 1604 * ... prune child dentries and writebacks if needed. 1605 */ 1606 if (atomic_read(&old_dentry->d_count) > 1) { 1607 if (S_ISREG(old_inode->i_mode)) 1608 nfs_wb_all(old_inode); 1609 shrink_dcache_parent(old_dentry); 1610 } 1611 nfs_inode_return_delegation(old_inode); 1612 1613 if (new_inode != NULL) { 1614 nfs_inode_return_delegation(new_inode); 1615 d_delete(new_dentry); 1616 } 1617 1618 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1619 new_dir, &new_dentry->d_name); 1620 nfs_mark_for_revalidate(old_inode); 1621 out: 1622 if (rehash) 1623 d_rehash(rehash); 1624 if (!error) { 1625 d_move(old_dentry, new_dentry); 1626 nfs_set_verifier(new_dentry, 1627 nfs_save_change_attribute(new_dir)); 1628 } 1629 1630 /* new dentry created? */ 1631 if (dentry) 1632 dput(dentry); 1633 unlock_kernel(); 1634 return error; 1635 } 1636 1637 static DEFINE_SPINLOCK(nfs_access_lru_lock); 1638 static LIST_HEAD(nfs_access_lru_list); 1639 static atomic_long_t nfs_access_nr_entries; 1640 1641 static void nfs_access_free_entry(struct nfs_access_entry *entry) 1642 { 1643 put_rpccred(entry->cred); 1644 kfree(entry); 1645 smp_mb__before_atomic_dec(); 1646 atomic_long_dec(&nfs_access_nr_entries); 1647 smp_mb__after_atomic_dec(); 1648 } 1649 1650 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) 1651 { 1652 LIST_HEAD(head); 1653 struct nfs_inode *nfsi; 1654 struct nfs_access_entry *cache; 1655 1656 restart: 1657 spin_lock(&nfs_access_lru_lock); 1658 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { 1659 struct inode *inode; 1660 1661 if (nr_to_scan-- == 0) 1662 break; 1663 inode = igrab(&nfsi->vfs_inode); 1664 if (inode == NULL) 1665 continue; 1666 spin_lock(&inode->i_lock); 1667 if (list_empty(&nfsi->access_cache_entry_lru)) 1668 goto remove_lru_entry; 1669 cache = list_entry(nfsi->access_cache_entry_lru.next, 1670 struct nfs_access_entry, lru); 1671 list_move(&cache->lru, &head); 1672 rb_erase(&cache->rb_node, &nfsi->access_cache); 1673 if (!list_empty(&nfsi->access_cache_entry_lru)) 1674 list_move_tail(&nfsi->access_cache_inode_lru, 1675 &nfs_access_lru_list); 1676 else { 1677 remove_lru_entry: 1678 list_del_init(&nfsi->access_cache_inode_lru); 1679 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1680 } 1681 spin_unlock(&inode->i_lock); 1682 spin_unlock(&nfs_access_lru_lock); 1683 iput(inode); 1684 goto restart; 1685 } 1686 spin_unlock(&nfs_access_lru_lock); 1687 while (!list_empty(&head)) { 1688 cache = list_entry(head.next, struct nfs_access_entry, lru); 1689 list_del(&cache->lru); 1690 nfs_access_free_entry(cache); 1691 } 1692 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; 1693 } 1694 1695 static void __nfs_access_zap_cache(struct inode *inode) 1696 { 1697 struct nfs_inode *nfsi = NFS_I(inode); 1698 struct rb_root *root_node = &nfsi->access_cache; 1699 struct rb_node *n, *dispose = NULL; 1700 struct nfs_access_entry *entry; 1701 1702 /* Unhook entries from the cache */ 1703 while ((n = rb_first(root_node)) != NULL) { 1704 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1705 rb_erase(n, root_node); 1706 list_del(&entry->lru); 1707 n->rb_left = dispose; 1708 dispose = n; 1709 } 1710 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 1711 spin_unlock(&inode->i_lock); 1712 1713 /* Now kill them all! */ 1714 while (dispose != NULL) { 1715 n = dispose; 1716 dispose = n->rb_left; 1717 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); 1718 } 1719 } 1720 1721 void nfs_access_zap_cache(struct inode *inode) 1722 { 1723 /* Remove from global LRU init */ 1724 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1725 spin_lock(&nfs_access_lru_lock); 1726 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 1727 spin_unlock(&nfs_access_lru_lock); 1728 } 1729 1730 spin_lock(&inode->i_lock); 1731 /* This will release the spinlock */ 1732 __nfs_access_zap_cache(inode); 1733 } 1734 1735 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 1736 { 1737 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 1738 struct nfs_access_entry *entry; 1739 1740 while (n != NULL) { 1741 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1742 1743 if (cred < entry->cred) 1744 n = n->rb_left; 1745 else if (cred > entry->cred) 1746 n = n->rb_right; 1747 else 1748 return entry; 1749 } 1750 return NULL; 1751 } 1752 1753 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 1754 { 1755 struct nfs_inode *nfsi = NFS_I(inode); 1756 struct nfs_access_entry *cache; 1757 int err = -ENOENT; 1758 1759 spin_lock(&inode->i_lock); 1760 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 1761 goto out_zap; 1762 cache = nfs_access_search_rbtree(inode, cred); 1763 if (cache == NULL) 1764 goto out; 1765 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))) 1766 goto out_stale; 1767 res->jiffies = cache->jiffies; 1768 res->cred = cache->cred; 1769 res->mask = cache->mask; 1770 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 1771 err = 0; 1772 out: 1773 spin_unlock(&inode->i_lock); 1774 return err; 1775 out_stale: 1776 rb_erase(&cache->rb_node, &nfsi->access_cache); 1777 list_del(&cache->lru); 1778 spin_unlock(&inode->i_lock); 1779 nfs_access_free_entry(cache); 1780 return -ENOENT; 1781 out_zap: 1782 /* This will release the spinlock */ 1783 __nfs_access_zap_cache(inode); 1784 return -ENOENT; 1785 } 1786 1787 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 1788 { 1789 struct nfs_inode *nfsi = NFS_I(inode); 1790 struct rb_root *root_node = &nfsi->access_cache; 1791 struct rb_node **p = &root_node->rb_node; 1792 struct rb_node *parent = NULL; 1793 struct nfs_access_entry *entry; 1794 1795 spin_lock(&inode->i_lock); 1796 while (*p != NULL) { 1797 parent = *p; 1798 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 1799 1800 if (set->cred < entry->cred) 1801 p = &parent->rb_left; 1802 else if (set->cred > entry->cred) 1803 p = &parent->rb_right; 1804 else 1805 goto found; 1806 } 1807 rb_link_node(&set->rb_node, parent, p); 1808 rb_insert_color(&set->rb_node, root_node); 1809 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1810 spin_unlock(&inode->i_lock); 1811 return; 1812 found: 1813 rb_replace_node(parent, &set->rb_node, root_node); 1814 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1815 list_del(&entry->lru); 1816 spin_unlock(&inode->i_lock); 1817 nfs_access_free_entry(entry); 1818 } 1819 1820 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 1821 { 1822 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 1823 if (cache == NULL) 1824 return; 1825 RB_CLEAR_NODE(&cache->rb_node); 1826 cache->jiffies = set->jiffies; 1827 cache->cred = get_rpccred(set->cred); 1828 cache->mask = set->mask; 1829 1830 nfs_access_add_rbtree(inode, cache); 1831 1832 /* Update accounting */ 1833 smp_mb__before_atomic_inc(); 1834 atomic_long_inc(&nfs_access_nr_entries); 1835 smp_mb__after_atomic_inc(); 1836 1837 /* Add inode to global LRU list */ 1838 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1839 spin_lock(&nfs_access_lru_lock); 1840 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); 1841 spin_unlock(&nfs_access_lru_lock); 1842 } 1843 } 1844 1845 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 1846 { 1847 struct nfs_access_entry cache; 1848 int status; 1849 1850 status = nfs_access_get_cached(inode, cred, &cache); 1851 if (status == 0) 1852 goto out; 1853 1854 /* Be clever: ask server to check for all possible rights */ 1855 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 1856 cache.cred = cred; 1857 cache.jiffies = jiffies; 1858 status = NFS_PROTO(inode)->access(inode, &cache); 1859 if (status != 0) 1860 return status; 1861 nfs_access_add_cache(inode, &cache); 1862 out: 1863 if ((cache.mask & mask) == mask) 1864 return 0; 1865 return -EACCES; 1866 } 1867 1868 static int nfs_open_permission_mask(int openflags) 1869 { 1870 int mask = 0; 1871 1872 if (openflags & FMODE_READ) 1873 mask |= MAY_READ; 1874 if (openflags & FMODE_WRITE) 1875 mask |= MAY_WRITE; 1876 if (openflags & FMODE_EXEC) 1877 mask |= MAY_EXEC; 1878 return mask; 1879 } 1880 1881 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 1882 { 1883 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 1884 } 1885 1886 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd) 1887 { 1888 struct rpc_cred *cred; 1889 int res = 0; 1890 1891 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 1892 1893 if (mask == 0) 1894 goto out; 1895 /* Is this sys_access() ? */ 1896 if (nd != NULL && (nd->flags & LOOKUP_ACCESS)) 1897 goto force_lookup; 1898 1899 switch (inode->i_mode & S_IFMT) { 1900 case S_IFLNK: 1901 goto out; 1902 case S_IFREG: 1903 /* NFSv4 has atomic_open... */ 1904 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) 1905 && nd != NULL 1906 && (nd->flags & LOOKUP_OPEN)) 1907 goto out; 1908 break; 1909 case S_IFDIR: 1910 /* 1911 * Optimize away all write operations, since the server 1912 * will check permissions when we perform the op. 1913 */ 1914 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 1915 goto out; 1916 } 1917 1918 force_lookup: 1919 lock_kernel(); 1920 1921 if (!NFS_PROTO(inode)->access) 1922 goto out_notsup; 1923 1924 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 1925 if (!IS_ERR(cred)) { 1926 res = nfs_do_access(inode, cred, mask); 1927 put_rpccred(cred); 1928 } else 1929 res = PTR_ERR(cred); 1930 unlock_kernel(); 1931 out: 1932 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 1933 inode->i_sb->s_id, inode->i_ino, mask, res); 1934 return res; 1935 out_notsup: 1936 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1937 if (res == 0) 1938 res = generic_permission(inode, mask, NULL); 1939 unlock_kernel(); 1940 goto out; 1941 } 1942 1943 /* 1944 * Local variables: 1945 * version-control: t 1946 * kept-new-versions: 5 1947 * End: 1948 */ 1949