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, nfs_compat_user_ino64(fileid), 431 d_type); 432 if (res < 0) 433 break; 434 file->f_pos++; 435 *desc->dir_cookie = entry->cookie; 436 if (dir_decode(desc) != 0) { 437 desc->page_index ++; 438 break; 439 } 440 if (loop_count++ > 200) { 441 loop_count = 0; 442 schedule(); 443 } 444 } 445 dir_page_release(desc); 446 if (dentry != NULL) 447 dput(dentry); 448 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 449 (unsigned long long)*desc->dir_cookie, res); 450 return res; 451 } 452 453 /* 454 * If we cannot find a cookie in our cache, we suspect that this is 455 * because it points to a deleted file, so we ask the server to return 456 * whatever it thinks is the next entry. We then feed this to filldir. 457 * If all goes well, we should then be able to find our way round the 458 * cache on the next call to readdir_search_pagecache(); 459 * 460 * NOTE: we cannot add the anonymous page to the pagecache because 461 * the data it contains might not be page aligned. Besides, 462 * we should already have a complete representation of the 463 * directory in the page cache by the time we get here. 464 */ 465 static inline 466 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, 467 filldir_t filldir) 468 { 469 struct file *file = desc->file; 470 struct inode *inode = file->f_path.dentry->d_inode; 471 struct rpc_cred *cred = nfs_file_cred(file); 472 struct page *page = NULL; 473 int status; 474 unsigned long timestamp; 475 476 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 477 (unsigned long long)*desc->dir_cookie); 478 479 page = alloc_page(GFP_HIGHUSER); 480 if (!page) { 481 status = -ENOMEM; 482 goto out; 483 } 484 timestamp = jiffies; 485 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie, 486 page, 487 NFS_SERVER(inode)->dtsize, 488 desc->plus); 489 desc->page = page; 490 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 491 if (desc->error >= 0) { 492 desc->timestamp = timestamp; 493 desc->timestamp_valid = 1; 494 if ((status = dir_decode(desc)) == 0) 495 desc->entry->prev_cookie = *desc->dir_cookie; 496 } else 497 status = -EIO; 498 if (status < 0) 499 goto out_release; 500 501 status = nfs_do_filldir(desc, dirent, filldir); 502 503 /* Reset read descriptor so it searches the page cache from 504 * the start upon the next call to readdir_search_pagecache() */ 505 desc->page_index = 0; 506 desc->entry->cookie = desc->entry->prev_cookie = 0; 507 desc->entry->eof = 0; 508 out: 509 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 510 __FUNCTION__, status); 511 return status; 512 out_release: 513 dir_page_release(desc); 514 goto out; 515 } 516 517 /* The file offset position represents the dirent entry number. A 518 last cookie cache takes care of the common case of reading the 519 whole directory. 520 */ 521 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) 522 { 523 struct dentry *dentry = filp->f_path.dentry; 524 struct inode *inode = dentry->d_inode; 525 nfs_readdir_descriptor_t my_desc, 526 *desc = &my_desc; 527 struct nfs_entry my_entry; 528 struct nfs_fh fh; 529 struct nfs_fattr fattr; 530 long res; 531 532 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n", 533 dentry->d_parent->d_name.name, dentry->d_name.name, 534 (long long)filp->f_pos); 535 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 536 537 lock_kernel(); 538 539 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping); 540 if (res < 0) { 541 unlock_kernel(); 542 return res; 543 } 544 545 /* 546 * filp->f_pos points to the dirent entry number. 547 * *desc->dir_cookie has the cookie for the next entry. We have 548 * to either find the entry with the appropriate number or 549 * revalidate the cookie. 550 */ 551 memset(desc, 0, sizeof(*desc)); 552 553 desc->file = filp; 554 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; 555 desc->decode = NFS_PROTO(inode)->decode_dirent; 556 desc->plus = NFS_USE_READDIRPLUS(inode); 557 558 my_entry.cookie = my_entry.prev_cookie = 0; 559 my_entry.eof = 0; 560 my_entry.fh = &fh; 561 my_entry.fattr = &fattr; 562 nfs_fattr_init(&fattr); 563 desc->entry = &my_entry; 564 565 nfs_block_sillyrename(dentry); 566 while(!desc->entry->eof) { 567 res = readdir_search_pagecache(desc); 568 569 if (res == -EBADCOOKIE) { 570 /* This means either end of directory */ 571 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { 572 /* Or that the server has 'lost' a cookie */ 573 res = uncached_readdir(desc, dirent, filldir); 574 if (res >= 0) 575 continue; 576 } 577 res = 0; 578 break; 579 } 580 if (res == -ETOOSMALL && desc->plus) { 581 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 582 nfs_zap_caches(inode); 583 desc->plus = 0; 584 desc->entry->eof = 0; 585 continue; 586 } 587 if (res < 0) 588 break; 589 590 res = nfs_do_filldir(desc, dirent, filldir); 591 if (res < 0) { 592 res = 0; 593 break; 594 } 595 } 596 nfs_unblock_sillyrename(dentry); 597 unlock_kernel(); 598 if (res > 0) 599 res = 0; 600 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n", 601 dentry->d_parent->d_name.name, dentry->d_name.name, 602 res); 603 return res; 604 } 605 606 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) 607 { 608 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex); 609 switch (origin) { 610 case 1: 611 offset += filp->f_pos; 612 case 0: 613 if (offset >= 0) 614 break; 615 default: 616 offset = -EINVAL; 617 goto out; 618 } 619 if (offset != filp->f_pos) { 620 filp->f_pos = offset; 621 nfs_file_open_context(filp)->dir_cookie = 0; 622 } 623 out: 624 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex); 625 return offset; 626 } 627 628 /* 629 * All directory operations under NFS are synchronous, so fsync() 630 * is a dummy operation. 631 */ 632 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) 633 { 634 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n", 635 dentry->d_parent->d_name.name, dentry->d_name.name, 636 datasync); 637 638 return 0; 639 } 640 641 /* 642 * A check for whether or not the parent directory has changed. 643 * In the case it has, we assume that the dentries are untrustworthy 644 * and may need to be looked up again. 645 */ 646 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) 647 { 648 if (IS_ROOT(dentry)) 649 return 1; 650 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 651 return 0; 652 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 653 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 654 return 0; 655 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 656 return 0; 657 return 1; 658 } 659 660 /* 661 * Return the intent data that applies to this particular path component 662 * 663 * Note that the current set of intents only apply to the very last 664 * component of the path. 665 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. 666 */ 667 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) 668 { 669 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) 670 return 0; 671 return nd->flags & mask; 672 } 673 674 /* 675 * Use intent information to check whether or not we're going to do 676 * an O_EXCL create using this path component. 677 */ 678 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) 679 { 680 if (NFS_PROTO(dir)->version == 2) 681 return 0; 682 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0) 683 return 0; 684 return (nd->intent.open.flags & O_EXCL) != 0; 685 } 686 687 /* 688 * Inode and filehandle revalidation for lookups. 689 * 690 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 691 * or if the intent information indicates that we're about to open this 692 * particular file and the "nocto" mount flag is not set. 693 * 694 */ 695 static inline 696 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) 697 { 698 struct nfs_server *server = NFS_SERVER(inode); 699 700 if (nd != NULL) { 701 /* VFS wants an on-the-wire revalidation */ 702 if (nd->flags & LOOKUP_REVAL) 703 goto out_force; 704 /* This is an open(2) */ 705 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && 706 !(server->flags & NFS_MOUNT_NOCTO) && 707 (S_ISREG(inode->i_mode) || 708 S_ISDIR(inode->i_mode))) 709 goto out_force; 710 return 0; 711 } 712 return nfs_revalidate_inode(server, inode); 713 out_force: 714 return __nfs_revalidate_inode(server, inode); 715 } 716 717 /* 718 * We judge how long we want to trust negative 719 * dentries by looking at the parent inode mtime. 720 * 721 * If parent mtime has changed, we revalidate, else we wait for a 722 * period corresponding to the parent's attribute cache timeout value. 723 */ 724 static inline 725 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 726 struct nameidata *nd) 727 { 728 /* Don't revalidate a negative dentry if we're creating a new file */ 729 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) 730 return 0; 731 return !nfs_check_verifier(dir, dentry); 732 } 733 734 /* 735 * This is called every time the dcache has a lookup hit, 736 * and we should check whether we can really trust that 737 * lookup. 738 * 739 * NOTE! The hit can be a negative hit too, don't assume 740 * we have an inode! 741 * 742 * If the parent directory is seen to have changed, we throw out the 743 * cached dentry and do a new lookup. 744 */ 745 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) 746 { 747 struct inode *dir; 748 struct inode *inode; 749 struct dentry *parent; 750 int error; 751 struct nfs_fh fhandle; 752 struct nfs_fattr fattr; 753 754 parent = dget_parent(dentry); 755 lock_kernel(); 756 dir = parent->d_inode; 757 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 758 inode = dentry->d_inode; 759 760 if (!inode) { 761 if (nfs_neg_need_reval(dir, dentry, nd)) 762 goto out_bad; 763 goto out_valid; 764 } 765 766 if (is_bad_inode(inode)) { 767 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", 768 __FUNCTION__, dentry->d_parent->d_name.name, 769 dentry->d_name.name); 770 goto out_bad; 771 } 772 773 /* Force a full look up iff the parent directory has changed */ 774 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { 775 if (nfs_lookup_verify_inode(inode, nd)) 776 goto out_zap_parent; 777 goto out_valid; 778 } 779 780 if (NFS_STALE(inode)) 781 goto out_bad; 782 783 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 784 if (error) 785 goto out_bad; 786 if (nfs_compare_fh(NFS_FH(inode), &fhandle)) 787 goto out_bad; 788 if ((error = nfs_refresh_inode(inode, &fattr)) != 0) 789 goto out_bad; 790 791 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 792 out_valid: 793 unlock_kernel(); 794 dput(parent); 795 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", 796 __FUNCTION__, dentry->d_parent->d_name.name, 797 dentry->d_name.name); 798 return 1; 799 out_zap_parent: 800 nfs_zap_caches(dir); 801 out_bad: 802 nfs_mark_for_revalidate(dir); 803 if (inode && S_ISDIR(inode->i_mode)) { 804 /* Purge readdir caches. */ 805 nfs_zap_caches(inode); 806 /* If we have submounts, don't unhash ! */ 807 if (have_submounts(dentry)) 808 goto out_valid; 809 shrink_dcache_parent(dentry); 810 } 811 d_drop(dentry); 812 unlock_kernel(); 813 dput(parent); 814 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", 815 __FUNCTION__, dentry->d_parent->d_name.name, 816 dentry->d_name.name); 817 return 0; 818 } 819 820 /* 821 * This is called from dput() when d_count is going to 0. 822 */ 823 static int nfs_dentry_delete(struct dentry *dentry) 824 { 825 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", 826 dentry->d_parent->d_name.name, dentry->d_name.name, 827 dentry->d_flags); 828 829 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 830 /* Unhash it, so that ->d_iput() would be called */ 831 return 1; 832 } 833 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 834 /* Unhash it, so that ancestors of killed async unlink 835 * files will be cleaned up during umount */ 836 return 1; 837 } 838 return 0; 839 840 } 841 842 /* 843 * Called when the dentry loses inode. 844 * We use it to clean up silly-renamed files. 845 */ 846 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 847 { 848 nfs_inode_return_delegation(inode); 849 if (S_ISDIR(inode->i_mode)) 850 /* drop any readdir cache as it could easily be old */ 851 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 852 853 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 854 lock_kernel(); 855 drop_nlink(inode); 856 nfs_complete_unlink(dentry, inode); 857 unlock_kernel(); 858 } 859 iput(inode); 860 } 861 862 struct dentry_operations nfs_dentry_operations = { 863 .d_revalidate = nfs_lookup_revalidate, 864 .d_delete = nfs_dentry_delete, 865 .d_iput = nfs_dentry_iput, 866 }; 867 868 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 869 { 870 struct dentry *res; 871 struct dentry *parent; 872 struct inode *inode = NULL; 873 int error; 874 struct nfs_fh fhandle; 875 struct nfs_fattr fattr; 876 877 dfprintk(VFS, "NFS: lookup(%s/%s)\n", 878 dentry->d_parent->d_name.name, dentry->d_name.name); 879 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 880 881 res = ERR_PTR(-ENAMETOOLONG); 882 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 883 goto out; 884 885 res = ERR_PTR(-ENOMEM); 886 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 887 888 lock_kernel(); 889 890 /* 891 * If we're doing an exclusive create, optimize away the lookup 892 * but don't hash the dentry. 893 */ 894 if (nfs_is_exclusive_create(dir, nd)) { 895 d_instantiate(dentry, NULL); 896 res = NULL; 897 goto out_unlock; 898 } 899 900 parent = dentry->d_parent; 901 /* Protect against concurrent sillydeletes */ 902 nfs_block_sillyrename(parent); 903 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 904 if (error == -ENOENT) 905 goto no_entry; 906 if (error < 0) { 907 res = ERR_PTR(error); 908 goto out_unblock_sillyrename; 909 } 910 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); 911 res = (struct dentry *)inode; 912 if (IS_ERR(res)) 913 goto out_unblock_sillyrename; 914 915 no_entry: 916 res = d_materialise_unique(dentry, inode); 917 if (res != NULL) { 918 if (IS_ERR(res)) 919 goto out_unblock_sillyrename; 920 dentry = res; 921 } 922 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 923 out_unblock_sillyrename: 924 nfs_unblock_sillyrename(parent); 925 out_unlock: 926 unlock_kernel(); 927 out: 928 return res; 929 } 930 931 #ifdef CONFIG_NFS_V4 932 static int nfs_open_revalidate(struct dentry *, struct nameidata *); 933 934 struct dentry_operations nfs4_dentry_operations = { 935 .d_revalidate = nfs_open_revalidate, 936 .d_delete = nfs_dentry_delete, 937 .d_iput = nfs_dentry_iput, 938 }; 939 940 /* 941 * Use intent information to determine whether we need to substitute 942 * the NFSv4-style stateful OPEN for the LOOKUP call 943 */ 944 static int is_atomic_open(struct inode *dir, struct nameidata *nd) 945 { 946 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) 947 return 0; 948 /* NFS does not (yet) have a stateful open for directories */ 949 if (nd->flags & LOOKUP_DIRECTORY) 950 return 0; 951 /* Are we trying to write to a read only partition? */ 952 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) 953 return 0; 954 return 1; 955 } 956 957 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 958 { 959 struct dentry *res = NULL; 960 int error; 961 962 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", 963 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 964 965 /* Check that we are indeed trying to open this file */ 966 if (!is_atomic_open(dir, nd)) 967 goto no_open; 968 969 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { 970 res = ERR_PTR(-ENAMETOOLONG); 971 goto out; 972 } 973 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 974 975 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash 976 * the dentry. */ 977 if (nd->intent.open.flags & O_EXCL) { 978 d_instantiate(dentry, NULL); 979 goto out; 980 } 981 982 /* Open the file on the server */ 983 lock_kernel(); 984 res = nfs4_atomic_open(dir, dentry, nd); 985 unlock_kernel(); 986 if (IS_ERR(res)) { 987 error = PTR_ERR(res); 988 switch (error) { 989 /* Make a negative dentry */ 990 case -ENOENT: 991 res = NULL; 992 goto out; 993 /* This turned out not to be a regular file */ 994 case -EISDIR: 995 case -ENOTDIR: 996 goto no_open; 997 case -ELOOP: 998 if (!(nd->intent.open.flags & O_NOFOLLOW)) 999 goto no_open; 1000 /* case -EINVAL: */ 1001 default: 1002 goto out; 1003 } 1004 } else if (res != NULL) 1005 dentry = res; 1006 out: 1007 return res; 1008 no_open: 1009 return nfs_lookup(dir, dentry, nd); 1010 } 1011 1012 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) 1013 { 1014 struct dentry *parent = NULL; 1015 struct inode *inode = dentry->d_inode; 1016 struct inode *dir; 1017 int openflags, ret = 0; 1018 1019 parent = dget_parent(dentry); 1020 dir = parent->d_inode; 1021 if (!is_atomic_open(dir, nd)) 1022 goto no_open; 1023 /* We can't create new files in nfs_open_revalidate(), so we 1024 * optimize away revalidation of negative dentries. 1025 */ 1026 if (inode == NULL) { 1027 if (!nfs_neg_need_reval(dir, dentry, nd)) 1028 ret = 1; 1029 goto out; 1030 } 1031 1032 /* NFS only supports OPEN on regular files */ 1033 if (!S_ISREG(inode->i_mode)) 1034 goto no_open; 1035 openflags = nd->intent.open.flags; 1036 /* We cannot do exclusive creation on a positive dentry */ 1037 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) 1038 goto no_open; 1039 /* We can't create new files, or truncate existing ones here */ 1040 openflags &= ~(O_CREAT|O_TRUNC); 1041 1042 /* 1043 * Note: we're not holding inode->i_mutex and so may be racing with 1044 * operations that change the directory. We therefore save the 1045 * change attribute *before* we do the RPC call. 1046 */ 1047 lock_kernel(); 1048 ret = nfs4_open_revalidate(dir, dentry, openflags, nd); 1049 unlock_kernel(); 1050 out: 1051 dput(parent); 1052 if (!ret) 1053 d_drop(dentry); 1054 return ret; 1055 no_open: 1056 dput(parent); 1057 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) 1058 return 1; 1059 return nfs_lookup_revalidate(dentry, nd); 1060 } 1061 #endif /* CONFIG_NFSV4 */ 1062 1063 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) 1064 { 1065 struct dentry *parent = desc->file->f_path.dentry; 1066 struct inode *dir = parent->d_inode; 1067 struct nfs_entry *entry = desc->entry; 1068 struct dentry *dentry, *alias; 1069 struct qstr name = { 1070 .name = entry->name, 1071 .len = entry->len, 1072 }; 1073 struct inode *inode; 1074 unsigned long verf = nfs_save_change_attribute(dir); 1075 1076 switch (name.len) { 1077 case 2: 1078 if (name.name[0] == '.' && name.name[1] == '.') 1079 return dget_parent(parent); 1080 break; 1081 case 1: 1082 if (name.name[0] == '.') 1083 return dget(parent); 1084 } 1085 1086 spin_lock(&dir->i_lock); 1087 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) { 1088 spin_unlock(&dir->i_lock); 1089 return NULL; 1090 } 1091 spin_unlock(&dir->i_lock); 1092 1093 name.hash = full_name_hash(name.name, name.len); 1094 dentry = d_lookup(parent, &name); 1095 if (dentry != NULL) { 1096 /* Is this a positive dentry that matches the readdir info? */ 1097 if (dentry->d_inode != NULL && 1098 (NFS_FILEID(dentry->d_inode) == entry->ino || 1099 d_mountpoint(dentry))) { 1100 if (!desc->plus || entry->fh->size == 0) 1101 return dentry; 1102 if (nfs_compare_fh(NFS_FH(dentry->d_inode), 1103 entry->fh) == 0) 1104 goto out_renew; 1105 } 1106 /* No, so d_drop to allow one to be created */ 1107 d_drop(dentry); 1108 dput(dentry); 1109 } 1110 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) 1111 return NULL; 1112 if (name.len > NFS_SERVER(dir)->namelen) 1113 return NULL; 1114 /* Note: caller is already holding the dir->i_mutex! */ 1115 dentry = d_alloc(parent, &name); 1116 if (dentry == NULL) 1117 return NULL; 1118 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1119 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 1120 if (IS_ERR(inode)) { 1121 dput(dentry); 1122 return NULL; 1123 } 1124 1125 alias = d_materialise_unique(dentry, inode); 1126 if (alias != NULL) { 1127 dput(dentry); 1128 if (IS_ERR(alias)) 1129 return NULL; 1130 dentry = alias; 1131 } 1132 1133 out_renew: 1134 nfs_set_verifier(dentry, verf); 1135 return dentry; 1136 } 1137 1138 /* 1139 * Code common to create, mkdir, and mknod. 1140 */ 1141 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1142 struct nfs_fattr *fattr) 1143 { 1144 struct dentry *parent = dget_parent(dentry); 1145 struct inode *dir = parent->d_inode; 1146 struct inode *inode; 1147 int error = -EACCES; 1148 1149 d_drop(dentry); 1150 1151 /* We may have been initialized further down */ 1152 if (dentry->d_inode) 1153 goto out; 1154 if (fhandle->size == 0) { 1155 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1156 if (error) 1157 goto out_error; 1158 } 1159 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1160 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1161 struct nfs_server *server = NFS_SB(dentry->d_sb); 1162 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1163 if (error < 0) 1164 goto out_error; 1165 } 1166 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1167 error = PTR_ERR(inode); 1168 if (IS_ERR(inode)) 1169 goto out_error; 1170 d_add(dentry, inode); 1171 out: 1172 dput(parent); 1173 return 0; 1174 out_error: 1175 nfs_mark_for_revalidate(dir); 1176 dput(parent); 1177 return error; 1178 } 1179 1180 /* 1181 * Following a failed create operation, we drop the dentry rather 1182 * than retain a negative dentry. This avoids a problem in the event 1183 * that the operation succeeded on the server, but an error in the 1184 * reply path made it appear to have failed. 1185 */ 1186 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, 1187 struct nameidata *nd) 1188 { 1189 struct iattr attr; 1190 int error; 1191 int open_flags = 0; 1192 1193 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1194 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1195 1196 attr.ia_mode = mode; 1197 attr.ia_valid = ATTR_MODE; 1198 1199 if ((nd->flags & LOOKUP_CREATE) != 0) 1200 open_flags = nd->intent.open.flags; 1201 1202 lock_kernel(); 1203 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); 1204 if (error != 0) 1205 goto out_err; 1206 unlock_kernel(); 1207 return 0; 1208 out_err: 1209 unlock_kernel(); 1210 d_drop(dentry); 1211 return error; 1212 } 1213 1214 /* 1215 * See comments for nfs_proc_create regarding failed operations. 1216 */ 1217 static int 1218 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) 1219 { 1220 struct iattr attr; 1221 int status; 1222 1223 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1224 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1225 1226 if (!new_valid_dev(rdev)) 1227 return -EINVAL; 1228 1229 attr.ia_mode = mode; 1230 attr.ia_valid = ATTR_MODE; 1231 1232 lock_kernel(); 1233 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1234 if (status != 0) 1235 goto out_err; 1236 unlock_kernel(); 1237 return 0; 1238 out_err: 1239 unlock_kernel(); 1240 d_drop(dentry); 1241 return status; 1242 } 1243 1244 /* 1245 * See comments for nfs_proc_create regarding failed operations. 1246 */ 1247 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1248 { 1249 struct iattr attr; 1250 int error; 1251 1252 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1253 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1254 1255 attr.ia_valid = ATTR_MODE; 1256 attr.ia_mode = mode | S_IFDIR; 1257 1258 lock_kernel(); 1259 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1260 if (error != 0) 1261 goto out_err; 1262 unlock_kernel(); 1263 return 0; 1264 out_err: 1265 d_drop(dentry); 1266 unlock_kernel(); 1267 return error; 1268 } 1269 1270 static int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1271 { 1272 int error; 1273 1274 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1275 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1276 1277 lock_kernel(); 1278 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1279 /* Ensure the VFS deletes this inode */ 1280 if (error == 0 && dentry->d_inode != NULL) 1281 clear_nlink(dentry->d_inode); 1282 unlock_kernel(); 1283 1284 return error; 1285 } 1286 1287 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) 1288 { 1289 static unsigned int sillycounter; 1290 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2; 1291 const int countersize = sizeof(sillycounter)*2; 1292 const int slen = sizeof(".nfs")+fileidsize+countersize-1; 1293 char silly[slen+1]; 1294 struct qstr qsilly; 1295 struct dentry *sdentry; 1296 int error = -EIO; 1297 1298 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", 1299 dentry->d_parent->d_name.name, dentry->d_name.name, 1300 atomic_read(&dentry->d_count)); 1301 nfs_inc_stats(dir, NFSIOS_SILLYRENAME); 1302 1303 /* 1304 * We don't allow a dentry to be silly-renamed twice. 1305 */ 1306 error = -EBUSY; 1307 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1308 goto out; 1309 1310 sprintf(silly, ".nfs%*.*Lx", 1311 fileidsize, fileidsize, 1312 (unsigned long long)NFS_FILEID(dentry->d_inode)); 1313 1314 /* Return delegation in anticipation of the rename */ 1315 nfs_inode_return_delegation(dentry->d_inode); 1316 1317 sdentry = NULL; 1318 do { 1319 char *suffix = silly + slen - countersize; 1320 1321 dput(sdentry); 1322 sillycounter++; 1323 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); 1324 1325 dfprintk(VFS, "NFS: trying to rename %s to %s\n", 1326 dentry->d_name.name, silly); 1327 1328 sdentry = lookup_one_len(silly, dentry->d_parent, slen); 1329 /* 1330 * N.B. Better to return EBUSY here ... it could be 1331 * dangerous to delete the file while it's in use. 1332 */ 1333 if (IS_ERR(sdentry)) 1334 goto out; 1335 } while(sdentry->d_inode != NULL); /* need negative lookup */ 1336 1337 qsilly.name = silly; 1338 qsilly.len = strlen(silly); 1339 if (dentry->d_inode) { 1340 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1341 dir, &qsilly); 1342 nfs_mark_for_revalidate(dentry->d_inode); 1343 } else 1344 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1345 dir, &qsilly); 1346 if (!error) { 1347 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1348 d_move(dentry, sdentry); 1349 error = nfs_async_unlink(dir, dentry); 1350 /* If we return 0 we don't unlink */ 1351 } 1352 dput(sdentry); 1353 out: 1354 return error; 1355 } 1356 1357 /* 1358 * Remove a file after making sure there are no pending writes, 1359 * and after checking that the file has only one user. 1360 * 1361 * We invalidate the attribute cache and free the inode prior to the operation 1362 * to avoid possible races if the server reuses the inode. 1363 */ 1364 static int nfs_safe_remove(struct dentry *dentry) 1365 { 1366 struct inode *dir = dentry->d_parent->d_inode; 1367 struct inode *inode = dentry->d_inode; 1368 int error = -EBUSY; 1369 1370 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1371 dentry->d_parent->d_name.name, dentry->d_name.name); 1372 1373 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1374 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1375 error = 0; 1376 goto out; 1377 } 1378 1379 if (inode != NULL) { 1380 nfs_inode_return_delegation(inode); 1381 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1382 /* The VFS may want to delete this inode */ 1383 if (error == 0) 1384 drop_nlink(inode); 1385 nfs_mark_for_revalidate(inode); 1386 } else 1387 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1388 out: 1389 return error; 1390 } 1391 1392 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1393 * belongs to an active ".nfs..." file and we return -EBUSY. 1394 * 1395 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1396 */ 1397 static int nfs_unlink(struct inode *dir, struct dentry *dentry) 1398 { 1399 int error; 1400 int need_rehash = 0; 1401 1402 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1403 dir->i_ino, dentry->d_name.name); 1404 1405 lock_kernel(); 1406 spin_lock(&dcache_lock); 1407 spin_lock(&dentry->d_lock); 1408 if (atomic_read(&dentry->d_count) > 1) { 1409 spin_unlock(&dentry->d_lock); 1410 spin_unlock(&dcache_lock); 1411 /* Start asynchronous writeout of the inode */ 1412 write_inode_now(dentry->d_inode, 0); 1413 error = nfs_sillyrename(dir, dentry); 1414 unlock_kernel(); 1415 return error; 1416 } 1417 if (!d_unhashed(dentry)) { 1418 __d_drop(dentry); 1419 need_rehash = 1; 1420 } 1421 spin_unlock(&dentry->d_lock); 1422 spin_unlock(&dcache_lock); 1423 error = nfs_safe_remove(dentry); 1424 if (!error) { 1425 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1426 } else if (need_rehash) 1427 d_rehash(dentry); 1428 unlock_kernel(); 1429 return error; 1430 } 1431 1432 /* 1433 * To create a symbolic link, most file systems instantiate a new inode, 1434 * add a page to it containing the path, then write it out to the disk 1435 * using prepare_write/commit_write. 1436 * 1437 * Unfortunately the NFS client can't create the in-core inode first 1438 * because it needs a file handle to create an in-core inode (see 1439 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1440 * symlink request has completed on the server. 1441 * 1442 * So instead we allocate a raw page, copy the symname into it, then do 1443 * the SYMLINK request with the page as the buffer. If it succeeds, we 1444 * now have a new file handle and can instantiate an in-core NFS inode 1445 * and move the raw page into its mapping. 1446 */ 1447 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1448 { 1449 struct pagevec lru_pvec; 1450 struct page *page; 1451 char *kaddr; 1452 struct iattr attr; 1453 unsigned int pathlen = strlen(symname); 1454 int error; 1455 1456 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1457 dir->i_ino, dentry->d_name.name, symname); 1458 1459 if (pathlen > PAGE_SIZE) 1460 return -ENAMETOOLONG; 1461 1462 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1463 attr.ia_valid = ATTR_MODE; 1464 1465 lock_kernel(); 1466 1467 page = alloc_page(GFP_HIGHUSER); 1468 if (!page) { 1469 unlock_kernel(); 1470 return -ENOMEM; 1471 } 1472 1473 kaddr = kmap_atomic(page, KM_USER0); 1474 memcpy(kaddr, symname, pathlen); 1475 if (pathlen < PAGE_SIZE) 1476 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1477 kunmap_atomic(kaddr, KM_USER0); 1478 1479 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1480 if (error != 0) { 1481 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1482 dir->i_sb->s_id, dir->i_ino, 1483 dentry->d_name.name, symname, error); 1484 d_drop(dentry); 1485 __free_page(page); 1486 unlock_kernel(); 1487 return error; 1488 } 1489 1490 /* 1491 * No big deal if we can't add this page to the page cache here. 1492 * READLINK will get the missing page from the server if needed. 1493 */ 1494 pagevec_init(&lru_pvec, 0); 1495 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1496 GFP_KERNEL)) { 1497 pagevec_add(&lru_pvec, page); 1498 pagevec_lru_add(&lru_pvec); 1499 SetPageUptodate(page); 1500 unlock_page(page); 1501 } else 1502 __free_page(page); 1503 1504 unlock_kernel(); 1505 return 0; 1506 } 1507 1508 static int 1509 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1510 { 1511 struct inode *inode = old_dentry->d_inode; 1512 int error; 1513 1514 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1515 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1516 dentry->d_parent->d_name.name, dentry->d_name.name); 1517 1518 lock_kernel(); 1519 d_drop(dentry); 1520 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1521 if (error == 0) { 1522 atomic_inc(&inode->i_count); 1523 d_add(dentry, inode); 1524 } 1525 unlock_kernel(); 1526 return error; 1527 } 1528 1529 /* 1530 * RENAME 1531 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1532 * different file handle for the same inode after a rename (e.g. when 1533 * moving to a different directory). A fail-safe method to do so would 1534 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1535 * rename the old file using the sillyrename stuff. This way, the original 1536 * file in old_dir will go away when the last process iput()s the inode. 1537 * 1538 * FIXED. 1539 * 1540 * It actually works quite well. One needs to have the possibility for 1541 * at least one ".nfs..." file in each directory the file ever gets 1542 * moved or linked to which happens automagically with the new 1543 * implementation that only depends on the dcache stuff instead of 1544 * using the inode layer 1545 * 1546 * Unfortunately, things are a little more complicated than indicated 1547 * above. For a cross-directory move, we want to make sure we can get 1548 * rid of the old inode after the operation. This means there must be 1549 * no pending writes (if it's a file), and the use count must be 1. 1550 * If these conditions are met, we can drop the dentries before doing 1551 * the rename. 1552 */ 1553 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1554 struct inode *new_dir, struct dentry *new_dentry) 1555 { 1556 struct inode *old_inode = old_dentry->d_inode; 1557 struct inode *new_inode = new_dentry->d_inode; 1558 struct dentry *dentry = NULL, *rehash = NULL; 1559 int error = -EBUSY; 1560 1561 /* 1562 * To prevent any new references to the target during the rename, 1563 * we unhash the dentry and free the inode in advance. 1564 */ 1565 lock_kernel(); 1566 if (!d_unhashed(new_dentry)) { 1567 d_drop(new_dentry); 1568 rehash = new_dentry; 1569 } 1570 1571 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1572 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1573 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1574 atomic_read(&new_dentry->d_count)); 1575 1576 /* 1577 * First check whether the target is busy ... we can't 1578 * safely do _any_ rename if the target is in use. 1579 * 1580 * For files, make a copy of the dentry and then do a 1581 * silly-rename. If the silly-rename succeeds, the 1582 * copied dentry is hashed and becomes the new target. 1583 */ 1584 if (!new_inode) 1585 goto go_ahead; 1586 if (S_ISDIR(new_inode->i_mode)) { 1587 error = -EISDIR; 1588 if (!S_ISDIR(old_inode->i_mode)) 1589 goto out; 1590 } else if (atomic_read(&new_dentry->d_count) > 2) { 1591 int err; 1592 /* copy the target dentry's name */ 1593 dentry = d_alloc(new_dentry->d_parent, 1594 &new_dentry->d_name); 1595 if (!dentry) 1596 goto out; 1597 1598 /* silly-rename the existing target ... */ 1599 err = nfs_sillyrename(new_dir, new_dentry); 1600 if (!err) { 1601 new_dentry = rehash = dentry; 1602 new_inode = NULL; 1603 /* instantiate the replacement target */ 1604 d_instantiate(new_dentry, NULL); 1605 } else if (atomic_read(&new_dentry->d_count) > 1) 1606 /* dentry still busy? */ 1607 goto out; 1608 } else 1609 drop_nlink(new_inode); 1610 1611 go_ahead: 1612 /* 1613 * ... prune child dentries and writebacks if needed. 1614 */ 1615 if (atomic_read(&old_dentry->d_count) > 1) { 1616 if (S_ISREG(old_inode->i_mode)) 1617 nfs_wb_all(old_inode); 1618 shrink_dcache_parent(old_dentry); 1619 } 1620 nfs_inode_return_delegation(old_inode); 1621 1622 if (new_inode != NULL) { 1623 nfs_inode_return_delegation(new_inode); 1624 d_delete(new_dentry); 1625 } 1626 1627 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1628 new_dir, &new_dentry->d_name); 1629 nfs_mark_for_revalidate(old_inode); 1630 out: 1631 if (rehash) 1632 d_rehash(rehash); 1633 if (!error) { 1634 d_move(old_dentry, new_dentry); 1635 nfs_set_verifier(new_dentry, 1636 nfs_save_change_attribute(new_dir)); 1637 } 1638 1639 /* new dentry created? */ 1640 if (dentry) 1641 dput(dentry); 1642 unlock_kernel(); 1643 return error; 1644 } 1645 1646 static DEFINE_SPINLOCK(nfs_access_lru_lock); 1647 static LIST_HEAD(nfs_access_lru_list); 1648 static atomic_long_t nfs_access_nr_entries; 1649 1650 static void nfs_access_free_entry(struct nfs_access_entry *entry) 1651 { 1652 put_rpccred(entry->cred); 1653 kfree(entry); 1654 smp_mb__before_atomic_dec(); 1655 atomic_long_dec(&nfs_access_nr_entries); 1656 smp_mb__after_atomic_dec(); 1657 } 1658 1659 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) 1660 { 1661 LIST_HEAD(head); 1662 struct nfs_inode *nfsi; 1663 struct nfs_access_entry *cache; 1664 1665 restart: 1666 spin_lock(&nfs_access_lru_lock); 1667 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { 1668 struct inode *inode; 1669 1670 if (nr_to_scan-- == 0) 1671 break; 1672 inode = igrab(&nfsi->vfs_inode); 1673 if (inode == NULL) 1674 continue; 1675 spin_lock(&inode->i_lock); 1676 if (list_empty(&nfsi->access_cache_entry_lru)) 1677 goto remove_lru_entry; 1678 cache = list_entry(nfsi->access_cache_entry_lru.next, 1679 struct nfs_access_entry, lru); 1680 list_move(&cache->lru, &head); 1681 rb_erase(&cache->rb_node, &nfsi->access_cache); 1682 if (!list_empty(&nfsi->access_cache_entry_lru)) 1683 list_move_tail(&nfsi->access_cache_inode_lru, 1684 &nfs_access_lru_list); 1685 else { 1686 remove_lru_entry: 1687 list_del_init(&nfsi->access_cache_inode_lru); 1688 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1689 } 1690 spin_unlock(&inode->i_lock); 1691 spin_unlock(&nfs_access_lru_lock); 1692 iput(inode); 1693 goto restart; 1694 } 1695 spin_unlock(&nfs_access_lru_lock); 1696 while (!list_empty(&head)) { 1697 cache = list_entry(head.next, struct nfs_access_entry, lru); 1698 list_del(&cache->lru); 1699 nfs_access_free_entry(cache); 1700 } 1701 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; 1702 } 1703 1704 static void __nfs_access_zap_cache(struct inode *inode) 1705 { 1706 struct nfs_inode *nfsi = NFS_I(inode); 1707 struct rb_root *root_node = &nfsi->access_cache; 1708 struct rb_node *n, *dispose = NULL; 1709 struct nfs_access_entry *entry; 1710 1711 /* Unhook entries from the cache */ 1712 while ((n = rb_first(root_node)) != NULL) { 1713 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1714 rb_erase(n, root_node); 1715 list_del(&entry->lru); 1716 n->rb_left = dispose; 1717 dispose = n; 1718 } 1719 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 1720 spin_unlock(&inode->i_lock); 1721 1722 /* Now kill them all! */ 1723 while (dispose != NULL) { 1724 n = dispose; 1725 dispose = n->rb_left; 1726 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); 1727 } 1728 } 1729 1730 void nfs_access_zap_cache(struct inode *inode) 1731 { 1732 /* Remove from global LRU init */ 1733 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1734 spin_lock(&nfs_access_lru_lock); 1735 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 1736 spin_unlock(&nfs_access_lru_lock); 1737 } 1738 1739 spin_lock(&inode->i_lock); 1740 /* This will release the spinlock */ 1741 __nfs_access_zap_cache(inode); 1742 } 1743 1744 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 1745 { 1746 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 1747 struct nfs_access_entry *entry; 1748 1749 while (n != NULL) { 1750 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1751 1752 if (cred < entry->cred) 1753 n = n->rb_left; 1754 else if (cred > entry->cred) 1755 n = n->rb_right; 1756 else 1757 return entry; 1758 } 1759 return NULL; 1760 } 1761 1762 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 1763 { 1764 struct nfs_inode *nfsi = NFS_I(inode); 1765 struct nfs_access_entry *cache; 1766 int err = -ENOENT; 1767 1768 spin_lock(&inode->i_lock); 1769 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 1770 goto out_zap; 1771 cache = nfs_access_search_rbtree(inode, cred); 1772 if (cache == NULL) 1773 goto out; 1774 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 1775 goto out_stale; 1776 res->jiffies = cache->jiffies; 1777 res->cred = cache->cred; 1778 res->mask = cache->mask; 1779 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 1780 err = 0; 1781 out: 1782 spin_unlock(&inode->i_lock); 1783 return err; 1784 out_stale: 1785 rb_erase(&cache->rb_node, &nfsi->access_cache); 1786 list_del(&cache->lru); 1787 spin_unlock(&inode->i_lock); 1788 nfs_access_free_entry(cache); 1789 return -ENOENT; 1790 out_zap: 1791 /* This will release the spinlock */ 1792 __nfs_access_zap_cache(inode); 1793 return -ENOENT; 1794 } 1795 1796 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 1797 { 1798 struct nfs_inode *nfsi = NFS_I(inode); 1799 struct rb_root *root_node = &nfsi->access_cache; 1800 struct rb_node **p = &root_node->rb_node; 1801 struct rb_node *parent = NULL; 1802 struct nfs_access_entry *entry; 1803 1804 spin_lock(&inode->i_lock); 1805 while (*p != NULL) { 1806 parent = *p; 1807 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 1808 1809 if (set->cred < entry->cred) 1810 p = &parent->rb_left; 1811 else if (set->cred > entry->cred) 1812 p = &parent->rb_right; 1813 else 1814 goto found; 1815 } 1816 rb_link_node(&set->rb_node, parent, p); 1817 rb_insert_color(&set->rb_node, root_node); 1818 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1819 spin_unlock(&inode->i_lock); 1820 return; 1821 found: 1822 rb_replace_node(parent, &set->rb_node, root_node); 1823 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1824 list_del(&entry->lru); 1825 spin_unlock(&inode->i_lock); 1826 nfs_access_free_entry(entry); 1827 } 1828 1829 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 1830 { 1831 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 1832 if (cache == NULL) 1833 return; 1834 RB_CLEAR_NODE(&cache->rb_node); 1835 cache->jiffies = set->jiffies; 1836 cache->cred = get_rpccred(set->cred); 1837 cache->mask = set->mask; 1838 1839 nfs_access_add_rbtree(inode, cache); 1840 1841 /* Update accounting */ 1842 smp_mb__before_atomic_inc(); 1843 atomic_long_inc(&nfs_access_nr_entries); 1844 smp_mb__after_atomic_inc(); 1845 1846 /* Add inode to global LRU list */ 1847 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1848 spin_lock(&nfs_access_lru_lock); 1849 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); 1850 spin_unlock(&nfs_access_lru_lock); 1851 } 1852 } 1853 1854 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 1855 { 1856 struct nfs_access_entry cache; 1857 int status; 1858 1859 status = nfs_access_get_cached(inode, cred, &cache); 1860 if (status == 0) 1861 goto out; 1862 1863 /* Be clever: ask server to check for all possible rights */ 1864 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 1865 cache.cred = cred; 1866 cache.jiffies = jiffies; 1867 status = NFS_PROTO(inode)->access(inode, &cache); 1868 if (status != 0) 1869 return status; 1870 nfs_access_add_cache(inode, &cache); 1871 out: 1872 if ((cache.mask & mask) == mask) 1873 return 0; 1874 return -EACCES; 1875 } 1876 1877 static int nfs_open_permission_mask(int openflags) 1878 { 1879 int mask = 0; 1880 1881 if (openflags & FMODE_READ) 1882 mask |= MAY_READ; 1883 if (openflags & FMODE_WRITE) 1884 mask |= MAY_WRITE; 1885 if (openflags & FMODE_EXEC) 1886 mask |= MAY_EXEC; 1887 return mask; 1888 } 1889 1890 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 1891 { 1892 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 1893 } 1894 1895 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd) 1896 { 1897 struct rpc_cred *cred; 1898 int res = 0; 1899 1900 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 1901 1902 if (mask == 0) 1903 goto out; 1904 /* Is this sys_access() ? */ 1905 if (nd != NULL && (nd->flags & LOOKUP_ACCESS)) 1906 goto force_lookup; 1907 1908 switch (inode->i_mode & S_IFMT) { 1909 case S_IFLNK: 1910 goto out; 1911 case S_IFREG: 1912 /* NFSv4 has atomic_open... */ 1913 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) 1914 && nd != NULL 1915 && (nd->flags & LOOKUP_OPEN)) 1916 goto out; 1917 break; 1918 case S_IFDIR: 1919 /* 1920 * Optimize away all write operations, since the server 1921 * will check permissions when we perform the op. 1922 */ 1923 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 1924 goto out; 1925 } 1926 1927 force_lookup: 1928 lock_kernel(); 1929 1930 if (!NFS_PROTO(inode)->access) 1931 goto out_notsup; 1932 1933 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 1934 if (!IS_ERR(cred)) { 1935 res = nfs_do_access(inode, cred, mask); 1936 put_rpccred(cred); 1937 } else 1938 res = PTR_ERR(cred); 1939 unlock_kernel(); 1940 out: 1941 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 1942 inode->i_sb->s_id, inode->i_ino, mask, res); 1943 return res; 1944 out_notsup: 1945 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1946 if (res == 0) 1947 res = generic_permission(inode, mask, NULL); 1948 unlock_kernel(); 1949 goto out; 1950 } 1951 1952 /* 1953 * Local variables: 1954 * version-control: t 1955 * kept-new-versions: 5 1956 * End: 1957 */ 1958