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 #include "internal.h" 42 43 /* #define NFS_DEBUG_VERBOSE 1 */ 44 45 static int nfs_opendir(struct inode *, struct file *); 46 static int nfs_readdir(struct file *, void *, filldir_t); 47 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); 48 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); 49 static int nfs_mkdir(struct inode *, struct dentry *, int); 50 static int nfs_rmdir(struct inode *, struct dentry *); 51 static int nfs_unlink(struct inode *, struct dentry *); 52 static int nfs_symlink(struct inode *, struct dentry *, const char *); 53 static int nfs_link(struct dentry *, struct inode *, struct dentry *); 54 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); 55 static int nfs_rename(struct inode *, struct dentry *, 56 struct inode *, struct dentry *); 57 static int nfs_fsync_dir(struct file *, struct dentry *, int); 58 static loff_t nfs_llseek_dir(struct file *, loff_t, int); 59 60 const struct file_operations nfs_dir_operations = { 61 .llseek = nfs_llseek_dir, 62 .read = generic_read_dir, 63 .readdir = nfs_readdir, 64 .open = nfs_opendir, 65 .release = nfs_release, 66 .fsync = nfs_fsync_dir, 67 }; 68 69 const struct inode_operations nfs_dir_inode_operations = { 70 .create = nfs_create, 71 .lookup = nfs_lookup, 72 .link = nfs_link, 73 .unlink = nfs_unlink, 74 .symlink = nfs_symlink, 75 .mkdir = nfs_mkdir, 76 .rmdir = nfs_rmdir, 77 .mknod = nfs_mknod, 78 .rename = nfs_rename, 79 .permission = nfs_permission, 80 .getattr = nfs_getattr, 81 .setattr = nfs_setattr, 82 }; 83 84 #ifdef CONFIG_NFS_V3 85 const struct inode_operations nfs3_dir_inode_operations = { 86 .create = nfs_create, 87 .lookup = nfs_lookup, 88 .link = nfs_link, 89 .unlink = nfs_unlink, 90 .symlink = nfs_symlink, 91 .mkdir = nfs_mkdir, 92 .rmdir = nfs_rmdir, 93 .mknod = nfs_mknod, 94 .rename = nfs_rename, 95 .permission = nfs_permission, 96 .getattr = nfs_getattr, 97 .setattr = nfs_setattr, 98 .listxattr = nfs3_listxattr, 99 .getxattr = nfs3_getxattr, 100 .setxattr = nfs3_setxattr, 101 .removexattr = nfs3_removexattr, 102 }; 103 #endif /* CONFIG_NFS_V3 */ 104 105 #ifdef CONFIG_NFS_V4 106 107 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); 108 const struct inode_operations nfs4_dir_inode_operations = { 109 .create = nfs_create, 110 .lookup = nfs_atomic_lookup, 111 .link = nfs_link, 112 .unlink = nfs_unlink, 113 .symlink = nfs_symlink, 114 .mkdir = nfs_mkdir, 115 .rmdir = nfs_rmdir, 116 .mknod = nfs_mknod, 117 .rename = nfs_rename, 118 .permission = nfs_permission, 119 .getattr = nfs_getattr, 120 .setattr = nfs_setattr, 121 .getxattr = nfs4_getxattr, 122 .setxattr = nfs4_setxattr, 123 .listxattr = nfs4_listxattr, 124 }; 125 126 #endif /* CONFIG_NFS_V4 */ 127 128 /* 129 * Open file 130 */ 131 static int 132 nfs_opendir(struct inode *inode, struct file *filp) 133 { 134 int res; 135 136 dfprintk(VFS, "NFS: opendir(%s/%ld)\n", 137 inode->i_sb->s_id, inode->i_ino); 138 139 lock_kernel(); 140 /* Call generic open code in order to cache credentials */ 141 res = nfs_open(inode, filp); 142 unlock_kernel(); 143 return res; 144 } 145 146 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int); 147 typedef struct { 148 struct file *file; 149 struct page *page; 150 unsigned long page_index; 151 __be32 *ptr; 152 u64 *dir_cookie; 153 loff_t current_index; 154 struct nfs_entry *entry; 155 decode_dirent_t decode; 156 int plus; 157 int error; 158 unsigned long timestamp; 159 int timestamp_valid; 160 } nfs_readdir_descriptor_t; 161 162 /* Now we cache directories properly, by stuffing the dirent 163 * data directly in the page cache. 164 * 165 * Inode invalidation due to refresh etc. takes care of 166 * _everything_, no sloppy entry flushing logic, no extraneous 167 * copying, network direct to page cache, the way it was meant 168 * to be. 169 * 170 * NOTE: Dirent information verification is done always by the 171 * page-in of the RPC reply, nowhere else, this simplies 172 * things substantially. 173 */ 174 static 175 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page) 176 { 177 struct file *file = desc->file; 178 struct inode *inode = file->f_path.dentry->d_inode; 179 struct rpc_cred *cred = nfs_file_cred(file); 180 unsigned long timestamp; 181 int error; 182 183 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n", 184 __FUNCTION__, (long long)desc->entry->cookie, 185 page->index); 186 187 again: 188 timestamp = jiffies; 189 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page, 190 NFS_SERVER(inode)->dtsize, desc->plus); 191 if (error < 0) { 192 /* We requested READDIRPLUS, but the server doesn't grok it */ 193 if (error == -ENOTSUPP && desc->plus) { 194 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 195 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 196 desc->plus = 0; 197 goto again; 198 } 199 goto error; 200 } 201 desc->timestamp = timestamp; 202 desc->timestamp_valid = 1; 203 SetPageUptodate(page); 204 /* Ensure consistent page alignment of the data. 205 * Note: assumes we have exclusive access to this mapping either 206 * through inode->i_mutex or some other mechanism. 207 */ 208 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) { 209 /* Should never happen */ 210 nfs_zap_mapping(inode, inode->i_mapping); 211 } 212 unlock_page(page); 213 return 0; 214 error: 215 unlock_page(page); 216 desc->error = error; 217 return -EIO; 218 } 219 220 static inline 221 int dir_decode(nfs_readdir_descriptor_t *desc) 222 { 223 __be32 *p = desc->ptr; 224 p = desc->decode(p, desc->entry, desc->plus); 225 if (IS_ERR(p)) 226 return PTR_ERR(p); 227 desc->ptr = p; 228 if (desc->timestamp_valid) 229 desc->entry->fattr->time_start = desc->timestamp; 230 else 231 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR; 232 return 0; 233 } 234 235 static inline 236 void dir_page_release(nfs_readdir_descriptor_t *desc) 237 { 238 kunmap(desc->page); 239 page_cache_release(desc->page); 240 desc->page = NULL; 241 desc->ptr = NULL; 242 } 243 244 /* 245 * Given a pointer to a buffer that has already been filled by a call 246 * to readdir, find the next entry with cookie '*desc->dir_cookie'. 247 * 248 * If the end of the buffer has been reached, return -EAGAIN, if not, 249 * return the offset within the buffer of the next entry to be 250 * read. 251 */ 252 static inline 253 int find_dirent(nfs_readdir_descriptor_t *desc) 254 { 255 struct nfs_entry *entry = desc->entry; 256 int loop_count = 0, 257 status; 258 259 while((status = dir_decode(desc)) == 0) { 260 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n", 261 __FUNCTION__, (unsigned long long)entry->cookie); 262 if (entry->prev_cookie == *desc->dir_cookie) 263 break; 264 if (loop_count++ > 200) { 265 loop_count = 0; 266 schedule(); 267 } 268 } 269 return status; 270 } 271 272 /* 273 * Given a pointer to a buffer that has already been filled by a call 274 * to readdir, find the entry at offset 'desc->file->f_pos'. 275 * 276 * If the end of the buffer has been reached, return -EAGAIN, if not, 277 * return the offset within the buffer of the next entry to be 278 * read. 279 */ 280 static inline 281 int find_dirent_index(nfs_readdir_descriptor_t *desc) 282 { 283 struct nfs_entry *entry = desc->entry; 284 int loop_count = 0, 285 status; 286 287 for(;;) { 288 status = dir_decode(desc); 289 if (status) 290 break; 291 292 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n", 293 (unsigned long long)entry->cookie, desc->current_index); 294 295 if (desc->file->f_pos == desc->current_index) { 296 *desc->dir_cookie = entry->cookie; 297 break; 298 } 299 desc->current_index++; 300 if (loop_count++ > 200) { 301 loop_count = 0; 302 schedule(); 303 } 304 } 305 return status; 306 } 307 308 /* 309 * Find the given page, and call find_dirent() or find_dirent_index in 310 * order to try to return the next entry. 311 */ 312 static inline 313 int find_dirent_page(nfs_readdir_descriptor_t *desc) 314 { 315 struct inode *inode = desc->file->f_path.dentry->d_inode; 316 struct page *page; 317 int status; 318 319 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n", 320 __FUNCTION__, desc->page_index, 321 (long long) *desc->dir_cookie); 322 323 /* If we find the page in the page_cache, we cannot be sure 324 * how fresh the data is, so we will ignore readdir_plus attributes. 325 */ 326 desc->timestamp_valid = 0; 327 page = read_cache_page(inode->i_mapping, desc->page_index, 328 (filler_t *)nfs_readdir_filler, desc); 329 if (IS_ERR(page)) { 330 status = PTR_ERR(page); 331 goto out; 332 } 333 334 /* NOTE: Someone else may have changed the READDIRPLUS flag */ 335 desc->page = page; 336 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 337 if (*desc->dir_cookie != 0) 338 status = find_dirent(desc); 339 else 340 status = find_dirent_index(desc); 341 if (status < 0) 342 dir_page_release(desc); 343 out: 344 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status); 345 return status; 346 } 347 348 /* 349 * Recurse through the page cache pages, and return a 350 * filled nfs_entry structure of the next directory entry if possible. 351 * 352 * The target for the search is '*desc->dir_cookie' if non-0, 353 * 'desc->file->f_pos' otherwise 354 */ 355 static inline 356 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 357 { 358 int loop_count = 0; 359 int res; 360 361 /* Always search-by-index from the beginning of the cache */ 362 if (*desc->dir_cookie == 0) { 363 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n", 364 (long long)desc->file->f_pos); 365 desc->page_index = 0; 366 desc->entry->cookie = desc->entry->prev_cookie = 0; 367 desc->entry->eof = 0; 368 desc->current_index = 0; 369 } else 370 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", 371 (unsigned long long)*desc->dir_cookie); 372 373 for (;;) { 374 res = find_dirent_page(desc); 375 if (res != -EAGAIN) 376 break; 377 /* Align to beginning of next page */ 378 desc->page_index ++; 379 if (loop_count++ > 200) { 380 loop_count = 0; 381 schedule(); 382 } 383 } 384 385 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res); 386 return res; 387 } 388 389 static inline unsigned int dt_type(struct inode *inode) 390 { 391 return (inode->i_mode >> 12) & 15; 392 } 393 394 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc); 395 396 /* 397 * Once we've found the start of the dirent within a page: fill 'er up... 398 */ 399 static 400 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, 401 filldir_t filldir) 402 { 403 struct file *file = desc->file; 404 struct nfs_entry *entry = desc->entry; 405 struct dentry *dentry = NULL; 406 u64 fileid; 407 int loop_count = 0, 408 res; 409 410 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", 411 (unsigned long long)entry->cookie); 412 413 for(;;) { 414 unsigned d_type = DT_UNKNOWN; 415 /* Note: entry->prev_cookie contains the cookie for 416 * retrieving the current dirent on the server */ 417 fileid = entry->ino; 418 419 /* Get a dentry if we have one */ 420 if (dentry != NULL) 421 dput(dentry); 422 dentry = nfs_readdir_lookup(desc); 423 424 /* Use readdirplus info */ 425 if (dentry != NULL && dentry->d_inode != NULL) { 426 d_type = dt_type(dentry->d_inode); 427 fileid = NFS_FILEID(dentry->d_inode); 428 } 429 430 res = filldir(dirent, entry->name, entry->len, 431 file->f_pos, nfs_compat_user_ino64(fileid), 432 d_type); 433 if (res < 0) 434 break; 435 file->f_pos++; 436 *desc->dir_cookie = entry->cookie; 437 if (dir_decode(desc) != 0) { 438 desc->page_index ++; 439 break; 440 } 441 if (loop_count++ > 200) { 442 loop_count = 0; 443 schedule(); 444 } 445 } 446 dir_page_release(desc); 447 if (dentry != NULL) 448 dput(dentry); 449 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 450 (unsigned long long)*desc->dir_cookie, res); 451 return res; 452 } 453 454 /* 455 * If we cannot find a cookie in our cache, we suspect that this is 456 * because it points to a deleted file, so we ask the server to return 457 * whatever it thinks is the next entry. We then feed this to filldir. 458 * If all goes well, we should then be able to find our way round the 459 * cache on the next call to readdir_search_pagecache(); 460 * 461 * NOTE: we cannot add the anonymous page to the pagecache because 462 * the data it contains might not be page aligned. Besides, 463 * we should already have a complete representation of the 464 * directory in the page cache by the time we get here. 465 */ 466 static inline 467 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, 468 filldir_t filldir) 469 { 470 struct file *file = desc->file; 471 struct inode *inode = file->f_path.dentry->d_inode; 472 struct rpc_cred *cred = nfs_file_cred(file); 473 struct page *page = NULL; 474 int status; 475 unsigned long timestamp; 476 477 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 478 (unsigned long long)*desc->dir_cookie); 479 480 page = alloc_page(GFP_HIGHUSER); 481 if (!page) { 482 status = -ENOMEM; 483 goto out; 484 } 485 timestamp = jiffies; 486 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie, 487 page, 488 NFS_SERVER(inode)->dtsize, 489 desc->plus); 490 desc->page = page; 491 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 492 if (desc->error >= 0) { 493 desc->timestamp = timestamp; 494 desc->timestamp_valid = 1; 495 if ((status = dir_decode(desc)) == 0) 496 desc->entry->prev_cookie = *desc->dir_cookie; 497 } else 498 status = -EIO; 499 if (status < 0) 500 goto out_release; 501 502 status = nfs_do_filldir(desc, dirent, filldir); 503 504 /* Reset read descriptor so it searches the page cache from 505 * the start upon the next call to readdir_search_pagecache() */ 506 desc->page_index = 0; 507 desc->entry->cookie = desc->entry->prev_cookie = 0; 508 desc->entry->eof = 0; 509 out: 510 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 511 __FUNCTION__, status); 512 return status; 513 out_release: 514 dir_page_release(desc); 515 goto out; 516 } 517 518 /* The file offset position represents the dirent entry number. A 519 last cookie cache takes care of the common case of reading the 520 whole directory. 521 */ 522 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) 523 { 524 struct dentry *dentry = filp->f_path.dentry; 525 struct inode *inode = dentry->d_inode; 526 nfs_readdir_descriptor_t my_desc, 527 *desc = &my_desc; 528 struct nfs_entry my_entry; 529 struct nfs_fh fh; 530 struct nfs_fattr fattr; 531 long res; 532 533 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n", 534 dentry->d_parent->d_name.name, dentry->d_name.name, 535 (long long)filp->f_pos); 536 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 537 538 lock_kernel(); 539 540 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping); 541 if (res < 0) { 542 unlock_kernel(); 543 return res; 544 } 545 546 /* 547 * filp->f_pos points to the dirent entry number. 548 * *desc->dir_cookie has the cookie for the next entry. We have 549 * to either find the entry with the appropriate number or 550 * revalidate the cookie. 551 */ 552 memset(desc, 0, sizeof(*desc)); 553 554 desc->file = filp; 555 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; 556 desc->decode = NFS_PROTO(inode)->decode_dirent; 557 desc->plus = NFS_USE_READDIRPLUS(inode); 558 559 my_entry.cookie = my_entry.prev_cookie = 0; 560 my_entry.eof = 0; 561 my_entry.fh = &fh; 562 my_entry.fattr = &fattr; 563 nfs_fattr_init(&fattr); 564 desc->entry = &my_entry; 565 566 nfs_block_sillyrename(dentry); 567 while(!desc->entry->eof) { 568 res = readdir_search_pagecache(desc); 569 570 if (res == -EBADCOOKIE) { 571 /* This means either end of directory */ 572 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { 573 /* Or that the server has 'lost' a cookie */ 574 res = uncached_readdir(desc, dirent, filldir); 575 if (res >= 0) 576 continue; 577 } 578 res = 0; 579 break; 580 } 581 if (res == -ETOOSMALL && desc->plus) { 582 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 583 nfs_zap_caches(inode); 584 desc->plus = 0; 585 desc->entry->eof = 0; 586 continue; 587 } 588 if (res < 0) 589 break; 590 591 res = nfs_do_filldir(desc, dirent, filldir); 592 if (res < 0) { 593 res = 0; 594 break; 595 } 596 } 597 nfs_unblock_sillyrename(dentry); 598 unlock_kernel(); 599 if (res > 0) 600 res = 0; 601 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n", 602 dentry->d_parent->d_name.name, dentry->d_name.name, 603 res); 604 return res; 605 } 606 607 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) 608 { 609 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex); 610 switch (origin) { 611 case 1: 612 offset += filp->f_pos; 613 case 0: 614 if (offset >= 0) 615 break; 616 default: 617 offset = -EINVAL; 618 goto out; 619 } 620 if (offset != filp->f_pos) { 621 filp->f_pos = offset; 622 nfs_file_open_context(filp)->dir_cookie = 0; 623 } 624 out: 625 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex); 626 return offset; 627 } 628 629 /* 630 * All directory operations under NFS are synchronous, so fsync() 631 * is a dummy operation. 632 */ 633 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) 634 { 635 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n", 636 dentry->d_parent->d_name.name, dentry->d_name.name, 637 datasync); 638 639 return 0; 640 } 641 642 /* 643 * A check for whether or not the parent directory has changed. 644 * In the case it has, we assume that the dentries are untrustworthy 645 * and may need to be looked up again. 646 */ 647 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) 648 { 649 if (IS_ROOT(dentry)) 650 return 1; 651 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 652 return 0; 653 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 654 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 655 return 0; 656 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 657 return 0; 658 return 1; 659 } 660 661 /* 662 * Return the intent data that applies to this particular path component 663 * 664 * Note that the current set of intents only apply to the very last 665 * component of the path. 666 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. 667 */ 668 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) 669 { 670 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) 671 return 0; 672 return nd->flags & mask; 673 } 674 675 /* 676 * Use intent information to check whether or not we're going to do 677 * an O_EXCL create using this path component. 678 */ 679 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) 680 { 681 if (NFS_PROTO(dir)->version == 2) 682 return 0; 683 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0) 684 return 0; 685 return (nd->intent.open.flags & O_EXCL) != 0; 686 } 687 688 /* 689 * Inode and filehandle revalidation for lookups. 690 * 691 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 692 * or if the intent information indicates that we're about to open this 693 * particular file and the "nocto" mount flag is not set. 694 * 695 */ 696 static inline 697 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) 698 { 699 struct nfs_server *server = NFS_SERVER(inode); 700 701 if (nd != NULL) { 702 /* VFS wants an on-the-wire revalidation */ 703 if (nd->flags & LOOKUP_REVAL) 704 goto out_force; 705 /* This is an open(2) */ 706 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && 707 !(server->flags & NFS_MOUNT_NOCTO) && 708 (S_ISREG(inode->i_mode) || 709 S_ISDIR(inode->i_mode))) 710 goto out_force; 711 return 0; 712 } 713 return nfs_revalidate_inode(server, inode); 714 out_force: 715 return __nfs_revalidate_inode(server, inode); 716 } 717 718 /* 719 * We judge how long we want to trust negative 720 * dentries by looking at the parent inode mtime. 721 * 722 * If parent mtime has changed, we revalidate, else we wait for a 723 * period corresponding to the parent's attribute cache timeout value. 724 */ 725 static inline 726 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 727 struct nameidata *nd) 728 { 729 /* Don't revalidate a negative dentry if we're creating a new file */ 730 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) 731 return 0; 732 return !nfs_check_verifier(dir, dentry); 733 } 734 735 /* 736 * This is called every time the dcache has a lookup hit, 737 * and we should check whether we can really trust that 738 * lookup. 739 * 740 * NOTE! The hit can be a negative hit too, don't assume 741 * we have an inode! 742 * 743 * If the parent directory is seen to have changed, we throw out the 744 * cached dentry and do a new lookup. 745 */ 746 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) 747 { 748 struct inode *dir; 749 struct inode *inode; 750 struct dentry *parent; 751 int error; 752 struct nfs_fh fhandle; 753 struct nfs_fattr fattr; 754 755 parent = dget_parent(dentry); 756 lock_kernel(); 757 dir = parent->d_inode; 758 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 759 inode = dentry->d_inode; 760 761 if (!inode) { 762 if (nfs_neg_need_reval(dir, dentry, nd)) 763 goto out_bad; 764 goto out_valid; 765 } 766 767 if (is_bad_inode(inode)) { 768 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", 769 __FUNCTION__, dentry->d_parent->d_name.name, 770 dentry->d_name.name); 771 goto out_bad; 772 } 773 774 /* Force a full look up iff the parent directory has changed */ 775 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { 776 if (nfs_lookup_verify_inode(inode, nd)) 777 goto out_zap_parent; 778 goto out_valid; 779 } 780 781 if (NFS_STALE(inode)) 782 goto out_bad; 783 784 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 785 if (error) 786 goto out_bad; 787 if (nfs_compare_fh(NFS_FH(inode), &fhandle)) 788 goto out_bad; 789 if ((error = nfs_refresh_inode(inode, &fattr)) != 0) 790 goto out_bad; 791 792 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 793 out_valid: 794 unlock_kernel(); 795 dput(parent); 796 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", 797 __FUNCTION__, dentry->d_parent->d_name.name, 798 dentry->d_name.name); 799 return 1; 800 out_zap_parent: 801 nfs_zap_caches(dir); 802 out_bad: 803 nfs_mark_for_revalidate(dir); 804 if (inode && S_ISDIR(inode->i_mode)) { 805 /* Purge readdir caches. */ 806 nfs_zap_caches(inode); 807 /* If we have submounts, don't unhash ! */ 808 if (have_submounts(dentry)) 809 goto out_valid; 810 shrink_dcache_parent(dentry); 811 } 812 d_drop(dentry); 813 unlock_kernel(); 814 dput(parent); 815 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", 816 __FUNCTION__, dentry->d_parent->d_name.name, 817 dentry->d_name.name); 818 return 0; 819 } 820 821 /* 822 * This is called from dput() when d_count is going to 0. 823 */ 824 static int nfs_dentry_delete(struct dentry *dentry) 825 { 826 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", 827 dentry->d_parent->d_name.name, dentry->d_name.name, 828 dentry->d_flags); 829 830 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 831 /* Unhash it, so that ->d_iput() would be called */ 832 return 1; 833 } 834 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 835 /* Unhash it, so that ancestors of killed async unlink 836 * files will be cleaned up during umount */ 837 return 1; 838 } 839 return 0; 840 841 } 842 843 /* 844 * Called when the dentry loses inode. 845 * We use it to clean up silly-renamed files. 846 */ 847 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 848 { 849 nfs_inode_return_delegation(inode); 850 if (S_ISDIR(inode->i_mode)) 851 /* drop any readdir cache as it could easily be old */ 852 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 853 854 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 855 lock_kernel(); 856 drop_nlink(inode); 857 nfs_complete_unlink(dentry, inode); 858 unlock_kernel(); 859 } 860 iput(inode); 861 } 862 863 struct dentry_operations nfs_dentry_operations = { 864 .d_revalidate = nfs_lookup_revalidate, 865 .d_delete = nfs_dentry_delete, 866 .d_iput = nfs_dentry_iput, 867 }; 868 869 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 870 { 871 struct dentry *res; 872 struct dentry *parent; 873 struct inode *inode = NULL; 874 int error; 875 struct nfs_fh fhandle; 876 struct nfs_fattr fattr; 877 878 dfprintk(VFS, "NFS: lookup(%s/%s)\n", 879 dentry->d_parent->d_name.name, dentry->d_name.name); 880 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 881 882 res = ERR_PTR(-ENAMETOOLONG); 883 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 884 goto out; 885 886 res = ERR_PTR(-ENOMEM); 887 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 888 889 lock_kernel(); 890 891 /* 892 * If we're doing an exclusive create, optimize away the lookup 893 * but don't hash the dentry. 894 */ 895 if (nfs_is_exclusive_create(dir, nd)) { 896 d_instantiate(dentry, NULL); 897 res = NULL; 898 goto out_unlock; 899 } 900 901 parent = dentry->d_parent; 902 /* Protect against concurrent sillydeletes */ 903 nfs_block_sillyrename(parent); 904 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 905 if (error == -ENOENT) 906 goto no_entry; 907 if (error < 0) { 908 res = ERR_PTR(error); 909 goto out_unblock_sillyrename; 910 } 911 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); 912 res = (struct dentry *)inode; 913 if (IS_ERR(res)) 914 goto out_unblock_sillyrename; 915 916 no_entry: 917 res = d_materialise_unique(dentry, inode); 918 if (res != NULL) { 919 if (IS_ERR(res)) 920 goto out_unblock_sillyrename; 921 dentry = res; 922 } 923 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 924 out_unblock_sillyrename: 925 nfs_unblock_sillyrename(parent); 926 out_unlock: 927 unlock_kernel(); 928 out: 929 return res; 930 } 931 932 #ifdef CONFIG_NFS_V4 933 static int nfs_open_revalidate(struct dentry *, struct nameidata *); 934 935 struct dentry_operations nfs4_dentry_operations = { 936 .d_revalidate = nfs_open_revalidate, 937 .d_delete = nfs_dentry_delete, 938 .d_iput = nfs_dentry_iput, 939 }; 940 941 /* 942 * Use intent information to determine whether we need to substitute 943 * the NFSv4-style stateful OPEN for the LOOKUP call 944 */ 945 static int is_atomic_open(struct inode *dir, struct nameidata *nd) 946 { 947 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) 948 return 0; 949 /* NFS does not (yet) have a stateful open for directories */ 950 if (nd->flags & LOOKUP_DIRECTORY) 951 return 0; 952 /* Are we trying to write to a read only partition? */ 953 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) 954 return 0; 955 return 1; 956 } 957 958 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 959 { 960 struct dentry *res = NULL; 961 int error; 962 963 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", 964 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 965 966 /* Check that we are indeed trying to open this file */ 967 if (!is_atomic_open(dir, nd)) 968 goto no_open; 969 970 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { 971 res = ERR_PTR(-ENAMETOOLONG); 972 goto out; 973 } 974 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 975 976 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash 977 * the dentry. */ 978 if (nd->intent.open.flags & O_EXCL) { 979 d_instantiate(dentry, NULL); 980 goto out; 981 } 982 983 /* Open the file on the server */ 984 lock_kernel(); 985 res = nfs4_atomic_open(dir, dentry, nd); 986 unlock_kernel(); 987 if (IS_ERR(res)) { 988 error = PTR_ERR(res); 989 switch (error) { 990 /* Make a negative dentry */ 991 case -ENOENT: 992 res = NULL; 993 goto out; 994 /* This turned out not to be a regular file */ 995 case -EISDIR: 996 case -ENOTDIR: 997 goto no_open; 998 case -ELOOP: 999 if (!(nd->intent.open.flags & O_NOFOLLOW)) 1000 goto no_open; 1001 /* case -EINVAL: */ 1002 default: 1003 goto out; 1004 } 1005 } else if (res != NULL) 1006 dentry = res; 1007 out: 1008 return res; 1009 no_open: 1010 return nfs_lookup(dir, dentry, nd); 1011 } 1012 1013 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) 1014 { 1015 struct dentry *parent = NULL; 1016 struct inode *inode = dentry->d_inode; 1017 struct inode *dir; 1018 int openflags, ret = 0; 1019 1020 parent = dget_parent(dentry); 1021 dir = parent->d_inode; 1022 if (!is_atomic_open(dir, nd)) 1023 goto no_open; 1024 /* We can't create new files in nfs_open_revalidate(), so we 1025 * optimize away revalidation of negative dentries. 1026 */ 1027 if (inode == NULL) { 1028 if (!nfs_neg_need_reval(dir, dentry, nd)) 1029 ret = 1; 1030 goto out; 1031 } 1032 1033 /* NFS only supports OPEN on regular files */ 1034 if (!S_ISREG(inode->i_mode)) 1035 goto no_open; 1036 openflags = nd->intent.open.flags; 1037 /* We cannot do exclusive creation on a positive dentry */ 1038 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) 1039 goto no_open; 1040 /* We can't create new files, or truncate existing ones here */ 1041 openflags &= ~(O_CREAT|O_TRUNC); 1042 1043 /* 1044 * Note: we're not holding inode->i_mutex and so may be racing with 1045 * operations that change the directory. We therefore save the 1046 * change attribute *before* we do the RPC call. 1047 */ 1048 lock_kernel(); 1049 ret = nfs4_open_revalidate(dir, dentry, openflags, nd); 1050 unlock_kernel(); 1051 out: 1052 dput(parent); 1053 if (!ret) 1054 d_drop(dentry); 1055 return ret; 1056 no_open: 1057 dput(parent); 1058 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) 1059 return 1; 1060 return nfs_lookup_revalidate(dentry, nd); 1061 } 1062 #endif /* CONFIG_NFSV4 */ 1063 1064 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) 1065 { 1066 struct dentry *parent = desc->file->f_path.dentry; 1067 struct inode *dir = parent->d_inode; 1068 struct nfs_entry *entry = desc->entry; 1069 struct dentry *dentry, *alias; 1070 struct qstr name = { 1071 .name = entry->name, 1072 .len = entry->len, 1073 }; 1074 struct inode *inode; 1075 unsigned long verf = nfs_save_change_attribute(dir); 1076 1077 switch (name.len) { 1078 case 2: 1079 if (name.name[0] == '.' && name.name[1] == '.') 1080 return dget_parent(parent); 1081 break; 1082 case 1: 1083 if (name.name[0] == '.') 1084 return dget(parent); 1085 } 1086 1087 spin_lock(&dir->i_lock); 1088 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) { 1089 spin_unlock(&dir->i_lock); 1090 return NULL; 1091 } 1092 spin_unlock(&dir->i_lock); 1093 1094 name.hash = full_name_hash(name.name, name.len); 1095 dentry = d_lookup(parent, &name); 1096 if (dentry != NULL) { 1097 /* Is this a positive dentry that matches the readdir info? */ 1098 if (dentry->d_inode != NULL && 1099 (NFS_FILEID(dentry->d_inode) == entry->ino || 1100 d_mountpoint(dentry))) { 1101 if (!desc->plus || entry->fh->size == 0) 1102 return dentry; 1103 if (nfs_compare_fh(NFS_FH(dentry->d_inode), 1104 entry->fh) == 0) 1105 goto out_renew; 1106 } 1107 /* No, so d_drop to allow one to be created */ 1108 d_drop(dentry); 1109 dput(dentry); 1110 } 1111 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) 1112 return NULL; 1113 if (name.len > NFS_SERVER(dir)->namelen) 1114 return NULL; 1115 /* Note: caller is already holding the dir->i_mutex! */ 1116 dentry = d_alloc(parent, &name); 1117 if (dentry == NULL) 1118 return NULL; 1119 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1120 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 1121 if (IS_ERR(inode)) { 1122 dput(dentry); 1123 return NULL; 1124 } 1125 1126 alias = d_materialise_unique(dentry, inode); 1127 if (alias != NULL) { 1128 dput(dentry); 1129 if (IS_ERR(alias)) 1130 return NULL; 1131 dentry = alias; 1132 } 1133 1134 out_renew: 1135 nfs_set_verifier(dentry, verf); 1136 return dentry; 1137 } 1138 1139 /* 1140 * Code common to create, mkdir, and mknod. 1141 */ 1142 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1143 struct nfs_fattr *fattr) 1144 { 1145 struct dentry *parent = dget_parent(dentry); 1146 struct inode *dir = parent->d_inode; 1147 struct inode *inode; 1148 int error = -EACCES; 1149 1150 d_drop(dentry); 1151 1152 /* We may have been initialized further down */ 1153 if (dentry->d_inode) 1154 goto out; 1155 if (fhandle->size == 0) { 1156 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1157 if (error) 1158 goto out_error; 1159 } 1160 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1161 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1162 struct nfs_server *server = NFS_SB(dentry->d_sb); 1163 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1164 if (error < 0) 1165 goto out_error; 1166 } 1167 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1168 error = PTR_ERR(inode); 1169 if (IS_ERR(inode)) 1170 goto out_error; 1171 d_add(dentry, inode); 1172 out: 1173 dput(parent); 1174 return 0; 1175 out_error: 1176 nfs_mark_for_revalidate(dir); 1177 dput(parent); 1178 return error; 1179 } 1180 1181 /* 1182 * Following a failed create operation, we drop the dentry rather 1183 * than retain a negative dentry. This avoids a problem in the event 1184 * that the operation succeeded on the server, but an error in the 1185 * reply path made it appear to have failed. 1186 */ 1187 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, 1188 struct nameidata *nd) 1189 { 1190 struct iattr attr; 1191 int error; 1192 int open_flags = 0; 1193 1194 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1195 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1196 1197 attr.ia_mode = mode; 1198 attr.ia_valid = ATTR_MODE; 1199 1200 if ((nd->flags & LOOKUP_CREATE) != 0) 1201 open_flags = nd->intent.open.flags; 1202 1203 lock_kernel(); 1204 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); 1205 if (error != 0) 1206 goto out_err; 1207 unlock_kernel(); 1208 return 0; 1209 out_err: 1210 unlock_kernel(); 1211 d_drop(dentry); 1212 return error; 1213 } 1214 1215 /* 1216 * See comments for nfs_proc_create regarding failed operations. 1217 */ 1218 static int 1219 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) 1220 { 1221 struct iattr attr; 1222 int status; 1223 1224 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1225 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1226 1227 if (!new_valid_dev(rdev)) 1228 return -EINVAL; 1229 1230 attr.ia_mode = mode; 1231 attr.ia_valid = ATTR_MODE; 1232 1233 lock_kernel(); 1234 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1235 if (status != 0) 1236 goto out_err; 1237 unlock_kernel(); 1238 return 0; 1239 out_err: 1240 unlock_kernel(); 1241 d_drop(dentry); 1242 return status; 1243 } 1244 1245 /* 1246 * See comments for nfs_proc_create regarding failed operations. 1247 */ 1248 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1249 { 1250 struct iattr attr; 1251 int error; 1252 1253 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1254 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1255 1256 attr.ia_valid = ATTR_MODE; 1257 attr.ia_mode = mode | S_IFDIR; 1258 1259 lock_kernel(); 1260 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1261 if (error != 0) 1262 goto out_err; 1263 unlock_kernel(); 1264 return 0; 1265 out_err: 1266 d_drop(dentry); 1267 unlock_kernel(); 1268 return error; 1269 } 1270 1271 static int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1272 { 1273 int error; 1274 1275 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1276 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1277 1278 lock_kernel(); 1279 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1280 /* Ensure the VFS deletes this inode */ 1281 if (error == 0 && dentry->d_inode != NULL) 1282 clear_nlink(dentry->d_inode); 1283 unlock_kernel(); 1284 1285 return error; 1286 } 1287 1288 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) 1289 { 1290 static unsigned int sillycounter; 1291 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2; 1292 const int countersize = sizeof(sillycounter)*2; 1293 const int slen = sizeof(".nfs")+fileidsize+countersize-1; 1294 char silly[slen+1]; 1295 struct qstr qsilly; 1296 struct dentry *sdentry; 1297 int error = -EIO; 1298 1299 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", 1300 dentry->d_parent->d_name.name, dentry->d_name.name, 1301 atomic_read(&dentry->d_count)); 1302 nfs_inc_stats(dir, NFSIOS_SILLYRENAME); 1303 1304 /* 1305 * We don't allow a dentry to be silly-renamed twice. 1306 */ 1307 error = -EBUSY; 1308 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1309 goto out; 1310 1311 sprintf(silly, ".nfs%*.*Lx", 1312 fileidsize, fileidsize, 1313 (unsigned long long)NFS_FILEID(dentry->d_inode)); 1314 1315 /* Return delegation in anticipation of the rename */ 1316 nfs_inode_return_delegation(dentry->d_inode); 1317 1318 sdentry = NULL; 1319 do { 1320 char *suffix = silly + slen - countersize; 1321 1322 dput(sdentry); 1323 sillycounter++; 1324 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); 1325 1326 dfprintk(VFS, "NFS: trying to rename %s to %s\n", 1327 dentry->d_name.name, silly); 1328 1329 sdentry = lookup_one_len(silly, dentry->d_parent, slen); 1330 /* 1331 * N.B. Better to return EBUSY here ... it could be 1332 * dangerous to delete the file while it's in use. 1333 */ 1334 if (IS_ERR(sdentry)) 1335 goto out; 1336 } while(sdentry->d_inode != NULL); /* need negative lookup */ 1337 1338 qsilly.name = silly; 1339 qsilly.len = strlen(silly); 1340 if (dentry->d_inode) { 1341 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1342 dir, &qsilly); 1343 nfs_mark_for_revalidate(dentry->d_inode); 1344 } else 1345 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1346 dir, &qsilly); 1347 if (!error) { 1348 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1349 d_move(dentry, sdentry); 1350 error = nfs_async_unlink(dir, dentry); 1351 /* If we return 0 we don't unlink */ 1352 } 1353 dput(sdentry); 1354 out: 1355 return error; 1356 } 1357 1358 /* 1359 * Remove a file after making sure there are no pending writes, 1360 * and after checking that the file has only one user. 1361 * 1362 * We invalidate the attribute cache and free the inode prior to the operation 1363 * to avoid possible races if the server reuses the inode. 1364 */ 1365 static int nfs_safe_remove(struct dentry *dentry) 1366 { 1367 struct inode *dir = dentry->d_parent->d_inode; 1368 struct inode *inode = dentry->d_inode; 1369 int error = -EBUSY; 1370 1371 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1372 dentry->d_parent->d_name.name, dentry->d_name.name); 1373 1374 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1375 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1376 error = 0; 1377 goto out; 1378 } 1379 1380 if (inode != NULL) { 1381 nfs_inode_return_delegation(inode); 1382 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1383 /* The VFS may want to delete this inode */ 1384 if (error == 0) 1385 drop_nlink(inode); 1386 nfs_mark_for_revalidate(inode); 1387 } else 1388 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1389 out: 1390 return error; 1391 } 1392 1393 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1394 * belongs to an active ".nfs..." file and we return -EBUSY. 1395 * 1396 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1397 */ 1398 static int nfs_unlink(struct inode *dir, struct dentry *dentry) 1399 { 1400 int error; 1401 int need_rehash = 0; 1402 1403 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1404 dir->i_ino, dentry->d_name.name); 1405 1406 lock_kernel(); 1407 spin_lock(&dcache_lock); 1408 spin_lock(&dentry->d_lock); 1409 if (atomic_read(&dentry->d_count) > 1) { 1410 spin_unlock(&dentry->d_lock); 1411 spin_unlock(&dcache_lock); 1412 /* Start asynchronous writeout of the inode */ 1413 write_inode_now(dentry->d_inode, 0); 1414 error = nfs_sillyrename(dir, dentry); 1415 unlock_kernel(); 1416 return error; 1417 } 1418 if (!d_unhashed(dentry)) { 1419 __d_drop(dentry); 1420 need_rehash = 1; 1421 } 1422 spin_unlock(&dentry->d_lock); 1423 spin_unlock(&dcache_lock); 1424 error = nfs_safe_remove(dentry); 1425 if (!error) { 1426 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1427 } else if (need_rehash) 1428 d_rehash(dentry); 1429 unlock_kernel(); 1430 return error; 1431 } 1432 1433 /* 1434 * To create a symbolic link, most file systems instantiate a new inode, 1435 * add a page to it containing the path, then write it out to the disk 1436 * using prepare_write/commit_write. 1437 * 1438 * Unfortunately the NFS client can't create the in-core inode first 1439 * because it needs a file handle to create an in-core inode (see 1440 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1441 * symlink request has completed on the server. 1442 * 1443 * So instead we allocate a raw page, copy the symname into it, then do 1444 * the SYMLINK request with the page as the buffer. If it succeeds, we 1445 * now have a new file handle and can instantiate an in-core NFS inode 1446 * and move the raw page into its mapping. 1447 */ 1448 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1449 { 1450 struct pagevec lru_pvec; 1451 struct page *page; 1452 char *kaddr; 1453 struct iattr attr; 1454 unsigned int pathlen = strlen(symname); 1455 int error; 1456 1457 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1458 dir->i_ino, dentry->d_name.name, symname); 1459 1460 if (pathlen > PAGE_SIZE) 1461 return -ENAMETOOLONG; 1462 1463 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1464 attr.ia_valid = ATTR_MODE; 1465 1466 lock_kernel(); 1467 1468 page = alloc_page(GFP_HIGHUSER); 1469 if (!page) { 1470 unlock_kernel(); 1471 return -ENOMEM; 1472 } 1473 1474 kaddr = kmap_atomic(page, KM_USER0); 1475 memcpy(kaddr, symname, pathlen); 1476 if (pathlen < PAGE_SIZE) 1477 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1478 kunmap_atomic(kaddr, KM_USER0); 1479 1480 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1481 if (error != 0) { 1482 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1483 dir->i_sb->s_id, dir->i_ino, 1484 dentry->d_name.name, symname, error); 1485 d_drop(dentry); 1486 __free_page(page); 1487 unlock_kernel(); 1488 return error; 1489 } 1490 1491 /* 1492 * No big deal if we can't add this page to the page cache here. 1493 * READLINK will get the missing page from the server if needed. 1494 */ 1495 pagevec_init(&lru_pvec, 0); 1496 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1497 GFP_KERNEL)) { 1498 pagevec_add(&lru_pvec, page); 1499 pagevec_lru_add(&lru_pvec); 1500 SetPageUptodate(page); 1501 unlock_page(page); 1502 } else 1503 __free_page(page); 1504 1505 unlock_kernel(); 1506 return 0; 1507 } 1508 1509 static int 1510 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1511 { 1512 struct inode *inode = old_dentry->d_inode; 1513 int error; 1514 1515 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1516 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1517 dentry->d_parent->d_name.name, dentry->d_name.name); 1518 1519 lock_kernel(); 1520 d_drop(dentry); 1521 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1522 if (error == 0) { 1523 atomic_inc(&inode->i_count); 1524 d_add(dentry, inode); 1525 } 1526 unlock_kernel(); 1527 return error; 1528 } 1529 1530 /* 1531 * RENAME 1532 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1533 * different file handle for the same inode after a rename (e.g. when 1534 * moving to a different directory). A fail-safe method to do so would 1535 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1536 * rename the old file using the sillyrename stuff. This way, the original 1537 * file in old_dir will go away when the last process iput()s the inode. 1538 * 1539 * FIXED. 1540 * 1541 * It actually works quite well. One needs to have the possibility for 1542 * at least one ".nfs..." file in each directory the file ever gets 1543 * moved or linked to which happens automagically with the new 1544 * implementation that only depends on the dcache stuff instead of 1545 * using the inode layer 1546 * 1547 * Unfortunately, things are a little more complicated than indicated 1548 * above. For a cross-directory move, we want to make sure we can get 1549 * rid of the old inode after the operation. This means there must be 1550 * no pending writes (if it's a file), and the use count must be 1. 1551 * If these conditions are met, we can drop the dentries before doing 1552 * the rename. 1553 */ 1554 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1555 struct inode *new_dir, struct dentry *new_dentry) 1556 { 1557 struct inode *old_inode = old_dentry->d_inode; 1558 struct inode *new_inode = new_dentry->d_inode; 1559 struct dentry *dentry = NULL, *rehash = NULL; 1560 int error = -EBUSY; 1561 1562 /* 1563 * To prevent any new references to the target during the rename, 1564 * we unhash the dentry and free the inode in advance. 1565 */ 1566 lock_kernel(); 1567 if (!d_unhashed(new_dentry)) { 1568 d_drop(new_dentry); 1569 rehash = new_dentry; 1570 } 1571 1572 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1573 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1574 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1575 atomic_read(&new_dentry->d_count)); 1576 1577 /* 1578 * First check whether the target is busy ... we can't 1579 * safely do _any_ rename if the target is in use. 1580 * 1581 * For files, make a copy of the dentry and then do a 1582 * silly-rename. If the silly-rename succeeds, the 1583 * copied dentry is hashed and becomes the new target. 1584 */ 1585 if (!new_inode) 1586 goto go_ahead; 1587 if (S_ISDIR(new_inode->i_mode)) { 1588 error = -EISDIR; 1589 if (!S_ISDIR(old_inode->i_mode)) 1590 goto out; 1591 } else if (atomic_read(&new_dentry->d_count) > 2) { 1592 int err; 1593 /* copy the target dentry's name */ 1594 dentry = d_alloc(new_dentry->d_parent, 1595 &new_dentry->d_name); 1596 if (!dentry) 1597 goto out; 1598 1599 /* silly-rename the existing target ... */ 1600 err = nfs_sillyrename(new_dir, new_dentry); 1601 if (!err) { 1602 new_dentry = rehash = dentry; 1603 new_inode = NULL; 1604 /* instantiate the replacement target */ 1605 d_instantiate(new_dentry, NULL); 1606 } else if (atomic_read(&new_dentry->d_count) > 1) 1607 /* dentry still busy? */ 1608 goto out; 1609 } else 1610 drop_nlink(new_inode); 1611 1612 go_ahead: 1613 /* 1614 * ... prune child dentries and writebacks if needed. 1615 */ 1616 if (atomic_read(&old_dentry->d_count) > 1) { 1617 if (S_ISREG(old_inode->i_mode)) 1618 nfs_wb_all(old_inode); 1619 shrink_dcache_parent(old_dentry); 1620 } 1621 nfs_inode_return_delegation(old_inode); 1622 1623 if (new_inode != NULL) { 1624 nfs_inode_return_delegation(new_inode); 1625 d_delete(new_dentry); 1626 } 1627 1628 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1629 new_dir, &new_dentry->d_name); 1630 nfs_mark_for_revalidate(old_inode); 1631 out: 1632 if (rehash) 1633 d_rehash(rehash); 1634 if (!error) { 1635 d_move(old_dentry, new_dentry); 1636 nfs_set_verifier(new_dentry, 1637 nfs_save_change_attribute(new_dir)); 1638 } 1639 1640 /* new dentry created? */ 1641 if (dentry) 1642 dput(dentry); 1643 unlock_kernel(); 1644 return error; 1645 } 1646 1647 static DEFINE_SPINLOCK(nfs_access_lru_lock); 1648 static LIST_HEAD(nfs_access_lru_list); 1649 static atomic_long_t nfs_access_nr_entries; 1650 1651 static void nfs_access_free_entry(struct nfs_access_entry *entry) 1652 { 1653 put_rpccred(entry->cred); 1654 kfree(entry); 1655 smp_mb__before_atomic_dec(); 1656 atomic_long_dec(&nfs_access_nr_entries); 1657 smp_mb__after_atomic_dec(); 1658 } 1659 1660 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) 1661 { 1662 LIST_HEAD(head); 1663 struct nfs_inode *nfsi; 1664 struct nfs_access_entry *cache; 1665 1666 restart: 1667 spin_lock(&nfs_access_lru_lock); 1668 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { 1669 struct inode *inode; 1670 1671 if (nr_to_scan-- == 0) 1672 break; 1673 inode = igrab(&nfsi->vfs_inode); 1674 if (inode == NULL) 1675 continue; 1676 spin_lock(&inode->i_lock); 1677 if (list_empty(&nfsi->access_cache_entry_lru)) 1678 goto remove_lru_entry; 1679 cache = list_entry(nfsi->access_cache_entry_lru.next, 1680 struct nfs_access_entry, lru); 1681 list_move(&cache->lru, &head); 1682 rb_erase(&cache->rb_node, &nfsi->access_cache); 1683 if (!list_empty(&nfsi->access_cache_entry_lru)) 1684 list_move_tail(&nfsi->access_cache_inode_lru, 1685 &nfs_access_lru_list); 1686 else { 1687 remove_lru_entry: 1688 list_del_init(&nfsi->access_cache_inode_lru); 1689 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1690 } 1691 spin_unlock(&inode->i_lock); 1692 spin_unlock(&nfs_access_lru_lock); 1693 iput(inode); 1694 goto restart; 1695 } 1696 spin_unlock(&nfs_access_lru_lock); 1697 while (!list_empty(&head)) { 1698 cache = list_entry(head.next, struct nfs_access_entry, lru); 1699 list_del(&cache->lru); 1700 nfs_access_free_entry(cache); 1701 } 1702 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; 1703 } 1704 1705 static void __nfs_access_zap_cache(struct inode *inode) 1706 { 1707 struct nfs_inode *nfsi = NFS_I(inode); 1708 struct rb_root *root_node = &nfsi->access_cache; 1709 struct rb_node *n, *dispose = NULL; 1710 struct nfs_access_entry *entry; 1711 1712 /* Unhook entries from the cache */ 1713 while ((n = rb_first(root_node)) != NULL) { 1714 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1715 rb_erase(n, root_node); 1716 list_del(&entry->lru); 1717 n->rb_left = dispose; 1718 dispose = n; 1719 } 1720 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 1721 spin_unlock(&inode->i_lock); 1722 1723 /* Now kill them all! */ 1724 while (dispose != NULL) { 1725 n = dispose; 1726 dispose = n->rb_left; 1727 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); 1728 } 1729 } 1730 1731 void nfs_access_zap_cache(struct inode *inode) 1732 { 1733 /* Remove from global LRU init */ 1734 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1735 spin_lock(&nfs_access_lru_lock); 1736 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 1737 spin_unlock(&nfs_access_lru_lock); 1738 } 1739 1740 spin_lock(&inode->i_lock); 1741 /* This will release the spinlock */ 1742 __nfs_access_zap_cache(inode); 1743 } 1744 1745 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 1746 { 1747 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 1748 struct nfs_access_entry *entry; 1749 1750 while (n != NULL) { 1751 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1752 1753 if (cred < entry->cred) 1754 n = n->rb_left; 1755 else if (cred > entry->cred) 1756 n = n->rb_right; 1757 else 1758 return entry; 1759 } 1760 return NULL; 1761 } 1762 1763 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 1764 { 1765 struct nfs_inode *nfsi = NFS_I(inode); 1766 struct nfs_access_entry *cache; 1767 int err = -ENOENT; 1768 1769 spin_lock(&inode->i_lock); 1770 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 1771 goto out_zap; 1772 cache = nfs_access_search_rbtree(inode, cred); 1773 if (cache == NULL) 1774 goto out; 1775 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 1776 goto out_stale; 1777 res->jiffies = cache->jiffies; 1778 res->cred = cache->cred; 1779 res->mask = cache->mask; 1780 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 1781 err = 0; 1782 out: 1783 spin_unlock(&inode->i_lock); 1784 return err; 1785 out_stale: 1786 rb_erase(&cache->rb_node, &nfsi->access_cache); 1787 list_del(&cache->lru); 1788 spin_unlock(&inode->i_lock); 1789 nfs_access_free_entry(cache); 1790 return -ENOENT; 1791 out_zap: 1792 /* This will release the spinlock */ 1793 __nfs_access_zap_cache(inode); 1794 return -ENOENT; 1795 } 1796 1797 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 1798 { 1799 struct nfs_inode *nfsi = NFS_I(inode); 1800 struct rb_root *root_node = &nfsi->access_cache; 1801 struct rb_node **p = &root_node->rb_node; 1802 struct rb_node *parent = NULL; 1803 struct nfs_access_entry *entry; 1804 1805 spin_lock(&inode->i_lock); 1806 while (*p != NULL) { 1807 parent = *p; 1808 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 1809 1810 if (set->cred < entry->cred) 1811 p = &parent->rb_left; 1812 else if (set->cred > entry->cred) 1813 p = &parent->rb_right; 1814 else 1815 goto found; 1816 } 1817 rb_link_node(&set->rb_node, parent, p); 1818 rb_insert_color(&set->rb_node, root_node); 1819 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1820 spin_unlock(&inode->i_lock); 1821 return; 1822 found: 1823 rb_replace_node(parent, &set->rb_node, root_node); 1824 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1825 list_del(&entry->lru); 1826 spin_unlock(&inode->i_lock); 1827 nfs_access_free_entry(entry); 1828 } 1829 1830 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 1831 { 1832 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 1833 if (cache == NULL) 1834 return; 1835 RB_CLEAR_NODE(&cache->rb_node); 1836 cache->jiffies = set->jiffies; 1837 cache->cred = get_rpccred(set->cred); 1838 cache->mask = set->mask; 1839 1840 nfs_access_add_rbtree(inode, cache); 1841 1842 /* Update accounting */ 1843 smp_mb__before_atomic_inc(); 1844 atomic_long_inc(&nfs_access_nr_entries); 1845 smp_mb__after_atomic_inc(); 1846 1847 /* Add inode to global LRU list */ 1848 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1849 spin_lock(&nfs_access_lru_lock); 1850 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); 1851 spin_unlock(&nfs_access_lru_lock); 1852 } 1853 } 1854 1855 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 1856 { 1857 struct nfs_access_entry cache; 1858 int status; 1859 1860 status = nfs_access_get_cached(inode, cred, &cache); 1861 if (status == 0) 1862 goto out; 1863 1864 /* Be clever: ask server to check for all possible rights */ 1865 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 1866 cache.cred = cred; 1867 cache.jiffies = jiffies; 1868 status = NFS_PROTO(inode)->access(inode, &cache); 1869 if (status != 0) 1870 return status; 1871 nfs_access_add_cache(inode, &cache); 1872 out: 1873 if ((cache.mask & mask) == mask) 1874 return 0; 1875 return -EACCES; 1876 } 1877 1878 static int nfs_open_permission_mask(int openflags) 1879 { 1880 int mask = 0; 1881 1882 if (openflags & FMODE_READ) 1883 mask |= MAY_READ; 1884 if (openflags & FMODE_WRITE) 1885 mask |= MAY_WRITE; 1886 if (openflags & FMODE_EXEC) 1887 mask |= MAY_EXEC; 1888 return mask; 1889 } 1890 1891 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 1892 { 1893 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 1894 } 1895 1896 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd) 1897 { 1898 struct rpc_cred *cred; 1899 int res = 0; 1900 1901 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 1902 1903 if (mask == 0) 1904 goto out; 1905 /* Is this sys_access() ? */ 1906 if (nd != NULL && (nd->flags & LOOKUP_ACCESS)) 1907 goto force_lookup; 1908 1909 switch (inode->i_mode & S_IFMT) { 1910 case S_IFLNK: 1911 goto out; 1912 case S_IFREG: 1913 /* NFSv4 has atomic_open... */ 1914 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) 1915 && nd != NULL 1916 && (nd->flags & LOOKUP_OPEN)) 1917 goto out; 1918 break; 1919 case S_IFDIR: 1920 /* 1921 * Optimize away all write operations, since the server 1922 * will check permissions when we perform the op. 1923 */ 1924 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 1925 goto out; 1926 } 1927 1928 force_lookup: 1929 lock_kernel(); 1930 1931 if (!NFS_PROTO(inode)->access) 1932 goto out_notsup; 1933 1934 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 1935 if (!IS_ERR(cred)) { 1936 res = nfs_do_access(inode, cred, mask); 1937 put_rpccred(cred); 1938 } else 1939 res = PTR_ERR(cred); 1940 unlock_kernel(); 1941 out: 1942 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 1943 inode->i_sb->s_id, inode->i_ino, mask, res); 1944 return res; 1945 out_notsup: 1946 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1947 if (res == 0) 1948 res = generic_permission(inode, mask, NULL); 1949 unlock_kernel(); 1950 goto out; 1951 } 1952 1953 /* 1954 * Local variables: 1955 * version-control: t 1956 * kept-new-versions: 5 1957 * End: 1958 */ 1959