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(FILE, "NFS: open dir(%s/%s)\n", 137 filp->f_path.dentry->d_parent->d_name.name, 138 filp->f_path.dentry->d_name.name); 139 140 nfs_inc_stats(inode, NFSIOS_VFSOPEN); 141 142 /* Call generic open code in order to cache credentials */ 143 res = nfs_open(inode, filp); 144 return res; 145 } 146 147 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int); 148 typedef struct { 149 struct file *file; 150 struct page *page; 151 unsigned long page_index; 152 __be32 *ptr; 153 u64 *dir_cookie; 154 loff_t current_index; 155 struct nfs_entry *entry; 156 decode_dirent_t decode; 157 int plus; 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 __func__, (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_I(inode)->flags); 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 (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -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 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 __func__, (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 __func__, 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", __func__, 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", __func__, 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 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, 486 *desc->dir_cookie, 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 (status >= 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 __func__, 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(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\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 /* 538 * filp->f_pos points to the dirent entry number. 539 * *desc->dir_cookie has the cookie for the next entry. We have 540 * to either find the entry with the appropriate number or 541 * revalidate the cookie. 542 */ 543 memset(desc, 0, sizeof(*desc)); 544 545 desc->file = filp; 546 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; 547 desc->decode = NFS_PROTO(inode)->decode_dirent; 548 desc->plus = NFS_USE_READDIRPLUS(inode); 549 550 my_entry.cookie = my_entry.prev_cookie = 0; 551 my_entry.eof = 0; 552 my_entry.fh = &fh; 553 my_entry.fattr = &fattr; 554 nfs_fattr_init(&fattr); 555 desc->entry = &my_entry; 556 557 nfs_block_sillyrename(dentry); 558 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping); 559 if (res < 0) 560 goto out; 561 562 while(!desc->entry->eof) { 563 res = readdir_search_pagecache(desc); 564 565 if (res == -EBADCOOKIE) { 566 /* This means either end of directory */ 567 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { 568 /* Or that the server has 'lost' a cookie */ 569 res = uncached_readdir(desc, dirent, filldir); 570 if (res >= 0) 571 continue; 572 } 573 res = 0; 574 break; 575 } 576 if (res == -ETOOSMALL && desc->plus) { 577 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 578 nfs_zap_caches(inode); 579 desc->plus = 0; 580 desc->entry->eof = 0; 581 continue; 582 } 583 if (res < 0) 584 break; 585 586 res = nfs_do_filldir(desc, dirent, filldir); 587 if (res < 0) { 588 res = 0; 589 break; 590 } 591 } 592 out: 593 nfs_unblock_sillyrename(dentry); 594 if (res > 0) 595 res = 0; 596 dfprintk(FILE, "NFS: readdir(%s/%s) returns %ld\n", 597 dentry->d_parent->d_name.name, dentry->d_name.name, 598 res); 599 return res; 600 } 601 602 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) 603 { 604 struct dentry *dentry = filp->f_path.dentry; 605 struct inode *inode = dentry->d_inode; 606 607 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", 608 dentry->d_parent->d_name.name, 609 dentry->d_name.name, 610 offset, origin); 611 612 mutex_lock(&inode->i_mutex); 613 switch (origin) { 614 case 1: 615 offset += filp->f_pos; 616 case 0: 617 if (offset >= 0) 618 break; 619 default: 620 offset = -EINVAL; 621 goto out; 622 } 623 if (offset != filp->f_pos) { 624 filp->f_pos = offset; 625 nfs_file_open_context(filp)->dir_cookie = 0; 626 } 627 out: 628 mutex_unlock(&inode->i_mutex); 629 return offset; 630 } 631 632 /* 633 * All directory operations under NFS are synchronous, so fsync() 634 * is a dummy operation. 635 */ 636 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) 637 { 638 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", 639 dentry->d_parent->d_name.name, dentry->d_name.name, 640 datasync); 641 642 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); 643 return 0; 644 } 645 646 /** 647 * nfs_force_lookup_revalidate - Mark the directory as having changed 648 * @dir - pointer to directory inode 649 * 650 * This forces the revalidation code in nfs_lookup_revalidate() to do a 651 * full lookup on all child dentries of 'dir' whenever a change occurs 652 * on the server that might have invalidated our dcache. 653 * 654 * The caller should be holding dir->i_lock 655 */ 656 void nfs_force_lookup_revalidate(struct inode *dir) 657 { 658 NFS_I(dir)->cache_change_attribute = jiffies; 659 } 660 661 /* 662 * A check for whether or not the parent directory has changed. 663 * In the case it has, we assume that the dentries are untrustworthy 664 * and may need to be looked up again. 665 */ 666 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) 667 { 668 if (IS_ROOT(dentry)) 669 return 1; 670 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 671 return 0; 672 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 673 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 674 return 0; 675 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 676 return 0; 677 return 1; 678 } 679 680 /* 681 * Return the intent data that applies to this particular path component 682 * 683 * Note that the current set of intents only apply to the very last 684 * component of the path. 685 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. 686 */ 687 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) 688 { 689 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) 690 return 0; 691 return nd->flags & mask; 692 } 693 694 /* 695 * Use intent information to check whether or not we're going to do 696 * an O_EXCL create using this path component. 697 */ 698 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) 699 { 700 if (NFS_PROTO(dir)->version == 2) 701 return 0; 702 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0) 703 return 0; 704 return (nd->intent.open.flags & O_EXCL) != 0; 705 } 706 707 /* 708 * Inode and filehandle revalidation for lookups. 709 * 710 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 711 * or if the intent information indicates that we're about to open this 712 * particular file and the "nocto" mount flag is not set. 713 * 714 */ 715 static inline 716 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) 717 { 718 struct nfs_server *server = NFS_SERVER(inode); 719 720 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags)) 721 return 0; 722 if (nd != NULL) { 723 /* VFS wants an on-the-wire revalidation */ 724 if (nd->flags & LOOKUP_REVAL) 725 goto out_force; 726 /* This is an open(2) */ 727 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && 728 !(server->flags & NFS_MOUNT_NOCTO) && 729 (S_ISREG(inode->i_mode) || 730 S_ISDIR(inode->i_mode))) 731 goto out_force; 732 return 0; 733 } 734 return nfs_revalidate_inode(server, inode); 735 out_force: 736 return __nfs_revalidate_inode(server, inode); 737 } 738 739 /* 740 * We judge how long we want to trust negative 741 * dentries by looking at the parent inode mtime. 742 * 743 * If parent mtime has changed, we revalidate, else we wait for a 744 * period corresponding to the parent's attribute cache timeout value. 745 */ 746 static inline 747 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 748 struct nameidata *nd) 749 { 750 /* Don't revalidate a negative dentry if we're creating a new file */ 751 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) 752 return 0; 753 return !nfs_check_verifier(dir, dentry); 754 } 755 756 /* 757 * This is called every time the dcache has a lookup hit, 758 * and we should check whether we can really trust that 759 * lookup. 760 * 761 * NOTE! The hit can be a negative hit too, don't assume 762 * we have an inode! 763 * 764 * If the parent directory is seen to have changed, we throw out the 765 * cached dentry and do a new lookup. 766 */ 767 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) 768 { 769 struct inode *dir; 770 struct inode *inode; 771 struct dentry *parent; 772 int error; 773 struct nfs_fh fhandle; 774 struct nfs_fattr fattr; 775 776 parent = dget_parent(dentry); 777 dir = parent->d_inode; 778 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 779 inode = dentry->d_inode; 780 781 if (!inode) { 782 if (nfs_neg_need_reval(dir, dentry, nd)) 783 goto out_bad; 784 goto out_valid; 785 } 786 787 if (is_bad_inode(inode)) { 788 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", 789 __func__, dentry->d_parent->d_name.name, 790 dentry->d_name.name); 791 goto out_bad; 792 } 793 794 /* Force a full look up iff the parent directory has changed */ 795 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { 796 if (nfs_lookup_verify_inode(inode, nd)) 797 goto out_zap_parent; 798 goto out_valid; 799 } 800 801 if (NFS_STALE(inode)) 802 goto out_bad; 803 804 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 805 if (error) 806 goto out_bad; 807 if (nfs_compare_fh(NFS_FH(inode), &fhandle)) 808 goto out_bad; 809 if ((error = nfs_refresh_inode(inode, &fattr)) != 0) 810 goto out_bad; 811 812 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 813 out_valid: 814 dput(parent); 815 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", 816 __func__, dentry->d_parent->d_name.name, 817 dentry->d_name.name); 818 return 1; 819 out_zap_parent: 820 nfs_zap_caches(dir); 821 out_bad: 822 nfs_mark_for_revalidate(dir); 823 if (inode && S_ISDIR(inode->i_mode)) { 824 /* Purge readdir caches. */ 825 nfs_zap_caches(inode); 826 /* If we have submounts, don't unhash ! */ 827 if (have_submounts(dentry)) 828 goto out_valid; 829 shrink_dcache_parent(dentry); 830 } 831 d_drop(dentry); 832 dput(parent); 833 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", 834 __func__, dentry->d_parent->d_name.name, 835 dentry->d_name.name); 836 return 0; 837 } 838 839 /* 840 * This is called from dput() when d_count is going to 0. 841 */ 842 static int nfs_dentry_delete(struct dentry *dentry) 843 { 844 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", 845 dentry->d_parent->d_name.name, dentry->d_name.name, 846 dentry->d_flags); 847 848 /* Unhash any dentry with a stale inode */ 849 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) 850 return 1; 851 852 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 853 /* Unhash it, so that ->d_iput() would be called */ 854 return 1; 855 } 856 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 857 /* Unhash it, so that ancestors of killed async unlink 858 * files will be cleaned up during umount */ 859 return 1; 860 } 861 return 0; 862 863 } 864 865 static void nfs_drop_nlink(struct inode *inode) 866 { 867 spin_lock(&inode->i_lock); 868 if (inode->i_nlink > 0) 869 drop_nlink(inode); 870 spin_unlock(&inode->i_lock); 871 } 872 873 /* 874 * Called when the dentry loses inode. 875 * We use it to clean up silly-renamed files. 876 */ 877 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 878 { 879 if (S_ISDIR(inode->i_mode)) 880 /* drop any readdir cache as it could easily be old */ 881 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 882 883 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 884 drop_nlink(inode); 885 nfs_complete_unlink(dentry, inode); 886 } 887 iput(inode); 888 } 889 890 struct dentry_operations nfs_dentry_operations = { 891 .d_revalidate = nfs_lookup_revalidate, 892 .d_delete = nfs_dentry_delete, 893 .d_iput = nfs_dentry_iput, 894 }; 895 896 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 897 { 898 struct dentry *res; 899 struct dentry *parent; 900 struct inode *inode = NULL; 901 int error; 902 struct nfs_fh fhandle; 903 struct nfs_fattr fattr; 904 905 dfprintk(VFS, "NFS: lookup(%s/%s)\n", 906 dentry->d_parent->d_name.name, dentry->d_name.name); 907 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 908 909 res = ERR_PTR(-ENAMETOOLONG); 910 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 911 goto out; 912 913 res = ERR_PTR(-ENOMEM); 914 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 915 916 /* 917 * If we're doing an exclusive create, optimize away the lookup 918 * but don't hash the dentry. 919 */ 920 if (nfs_is_exclusive_create(dir, nd)) { 921 d_instantiate(dentry, NULL); 922 res = NULL; 923 goto out; 924 } 925 926 parent = dentry->d_parent; 927 /* Protect against concurrent sillydeletes */ 928 nfs_block_sillyrename(parent); 929 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 930 if (error == -ENOENT) 931 goto no_entry; 932 if (error < 0) { 933 res = ERR_PTR(error); 934 goto out_unblock_sillyrename; 935 } 936 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); 937 res = (struct dentry *)inode; 938 if (IS_ERR(res)) 939 goto out_unblock_sillyrename; 940 941 no_entry: 942 res = d_materialise_unique(dentry, inode); 943 if (res != NULL) { 944 if (IS_ERR(res)) 945 goto out_unblock_sillyrename; 946 dentry = res; 947 } 948 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 949 out_unblock_sillyrename: 950 nfs_unblock_sillyrename(parent); 951 out: 952 return res; 953 } 954 955 #ifdef CONFIG_NFS_V4 956 static int nfs_open_revalidate(struct dentry *, struct nameidata *); 957 958 struct dentry_operations nfs4_dentry_operations = { 959 .d_revalidate = nfs_open_revalidate, 960 .d_delete = nfs_dentry_delete, 961 .d_iput = nfs_dentry_iput, 962 }; 963 964 /* 965 * Use intent information to determine whether we need to substitute 966 * the NFSv4-style stateful OPEN for the LOOKUP call 967 */ 968 static int is_atomic_open(struct inode *dir, struct nameidata *nd) 969 { 970 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) 971 return 0; 972 /* NFS does not (yet) have a stateful open for directories */ 973 if (nd->flags & LOOKUP_DIRECTORY) 974 return 0; 975 /* Are we trying to write to a read only partition? */ 976 if (__mnt_is_readonly(nd->path.mnt) && 977 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) 978 return 0; 979 return 1; 980 } 981 982 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 983 { 984 struct dentry *res = NULL; 985 int error; 986 987 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", 988 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 989 990 /* Check that we are indeed trying to open this file */ 991 if (!is_atomic_open(dir, nd)) 992 goto no_open; 993 994 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { 995 res = ERR_PTR(-ENAMETOOLONG); 996 goto out; 997 } 998 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 999 1000 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash 1001 * the dentry. */ 1002 if (nd->intent.open.flags & O_EXCL) { 1003 d_instantiate(dentry, NULL); 1004 goto out; 1005 } 1006 1007 /* Open the file on the server */ 1008 res = nfs4_atomic_open(dir, dentry, nd); 1009 if (IS_ERR(res)) { 1010 error = PTR_ERR(res); 1011 switch (error) { 1012 /* Make a negative dentry */ 1013 case -ENOENT: 1014 res = NULL; 1015 goto out; 1016 /* This turned out not to be a regular file */ 1017 case -EISDIR: 1018 case -ENOTDIR: 1019 goto no_open; 1020 case -ELOOP: 1021 if (!(nd->intent.open.flags & O_NOFOLLOW)) 1022 goto no_open; 1023 /* case -EINVAL: */ 1024 default: 1025 goto out; 1026 } 1027 } else if (res != NULL) 1028 dentry = res; 1029 out: 1030 return res; 1031 no_open: 1032 return nfs_lookup(dir, dentry, nd); 1033 } 1034 1035 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) 1036 { 1037 struct dentry *parent = NULL; 1038 struct inode *inode = dentry->d_inode; 1039 struct inode *dir; 1040 int openflags, ret = 0; 1041 1042 parent = dget_parent(dentry); 1043 dir = parent->d_inode; 1044 if (!is_atomic_open(dir, nd)) 1045 goto no_open; 1046 /* We can't create new files in nfs_open_revalidate(), so we 1047 * optimize away revalidation of negative dentries. 1048 */ 1049 if (inode == NULL) { 1050 if (!nfs_neg_need_reval(dir, dentry, nd)) 1051 ret = 1; 1052 goto out; 1053 } 1054 1055 /* NFS only supports OPEN on regular files */ 1056 if (!S_ISREG(inode->i_mode)) 1057 goto no_open; 1058 openflags = nd->intent.open.flags; 1059 /* We cannot do exclusive creation on a positive dentry */ 1060 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) 1061 goto no_open; 1062 /* We can't create new files, or truncate existing ones here */ 1063 openflags &= ~(O_CREAT|O_TRUNC); 1064 1065 /* 1066 * Note: we're not holding inode->i_mutex and so may be racing with 1067 * operations that change the directory. We therefore save the 1068 * change attribute *before* we do the RPC call. 1069 */ 1070 ret = nfs4_open_revalidate(dir, dentry, openflags, nd); 1071 out: 1072 dput(parent); 1073 if (!ret) 1074 d_drop(dentry); 1075 return ret; 1076 no_open: 1077 dput(parent); 1078 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) 1079 return 1; 1080 return nfs_lookup_revalidate(dentry, nd); 1081 } 1082 #endif /* CONFIG_NFSV4 */ 1083 1084 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) 1085 { 1086 struct dentry *parent = desc->file->f_path.dentry; 1087 struct inode *dir = parent->d_inode; 1088 struct nfs_entry *entry = desc->entry; 1089 struct dentry *dentry, *alias; 1090 struct qstr name = { 1091 .name = entry->name, 1092 .len = entry->len, 1093 }; 1094 struct inode *inode; 1095 unsigned long verf = nfs_save_change_attribute(dir); 1096 1097 switch (name.len) { 1098 case 2: 1099 if (name.name[0] == '.' && name.name[1] == '.') 1100 return dget_parent(parent); 1101 break; 1102 case 1: 1103 if (name.name[0] == '.') 1104 return dget(parent); 1105 } 1106 1107 spin_lock(&dir->i_lock); 1108 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) { 1109 spin_unlock(&dir->i_lock); 1110 return NULL; 1111 } 1112 spin_unlock(&dir->i_lock); 1113 1114 name.hash = full_name_hash(name.name, name.len); 1115 dentry = d_lookup(parent, &name); 1116 if (dentry != NULL) { 1117 /* Is this a positive dentry that matches the readdir info? */ 1118 if (dentry->d_inode != NULL && 1119 (NFS_FILEID(dentry->d_inode) == entry->ino || 1120 d_mountpoint(dentry))) { 1121 if (!desc->plus || entry->fh->size == 0) 1122 return dentry; 1123 if (nfs_compare_fh(NFS_FH(dentry->d_inode), 1124 entry->fh) == 0) 1125 goto out_renew; 1126 } 1127 /* No, so d_drop to allow one to be created */ 1128 d_drop(dentry); 1129 dput(dentry); 1130 } 1131 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) 1132 return NULL; 1133 if (name.len > NFS_SERVER(dir)->namelen) 1134 return NULL; 1135 /* Note: caller is already holding the dir->i_mutex! */ 1136 dentry = d_alloc(parent, &name); 1137 if (dentry == NULL) 1138 return NULL; 1139 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1140 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 1141 if (IS_ERR(inode)) { 1142 dput(dentry); 1143 return NULL; 1144 } 1145 1146 alias = d_materialise_unique(dentry, inode); 1147 if (alias != NULL) { 1148 dput(dentry); 1149 if (IS_ERR(alias)) 1150 return NULL; 1151 dentry = alias; 1152 } 1153 1154 out_renew: 1155 nfs_set_verifier(dentry, verf); 1156 return dentry; 1157 } 1158 1159 /* 1160 * Code common to create, mkdir, and mknod. 1161 */ 1162 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1163 struct nfs_fattr *fattr) 1164 { 1165 struct dentry *parent = dget_parent(dentry); 1166 struct inode *dir = parent->d_inode; 1167 struct inode *inode; 1168 int error = -EACCES; 1169 1170 d_drop(dentry); 1171 1172 /* We may have been initialized further down */ 1173 if (dentry->d_inode) 1174 goto out; 1175 if (fhandle->size == 0) { 1176 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1177 if (error) 1178 goto out_error; 1179 } 1180 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1181 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1182 struct nfs_server *server = NFS_SB(dentry->d_sb); 1183 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1184 if (error < 0) 1185 goto out_error; 1186 } 1187 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1188 error = PTR_ERR(inode); 1189 if (IS_ERR(inode)) 1190 goto out_error; 1191 d_add(dentry, inode); 1192 out: 1193 dput(parent); 1194 return 0; 1195 out_error: 1196 nfs_mark_for_revalidate(dir); 1197 dput(parent); 1198 return error; 1199 } 1200 1201 /* 1202 * Following a failed create operation, we drop the dentry rather 1203 * than retain a negative dentry. This avoids a problem in the event 1204 * that the operation succeeded on the server, but an error in the 1205 * reply path made it appear to have failed. 1206 */ 1207 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, 1208 struct nameidata *nd) 1209 { 1210 struct iattr attr; 1211 int error; 1212 int open_flags = 0; 1213 1214 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1215 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1216 1217 attr.ia_mode = mode; 1218 attr.ia_valid = ATTR_MODE; 1219 1220 if ((nd->flags & LOOKUP_CREATE) != 0) 1221 open_flags = nd->intent.open.flags; 1222 1223 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); 1224 if (error != 0) 1225 goto out_err; 1226 return 0; 1227 out_err: 1228 d_drop(dentry); 1229 return error; 1230 } 1231 1232 /* 1233 * See comments for nfs_proc_create regarding failed operations. 1234 */ 1235 static int 1236 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) 1237 { 1238 struct iattr attr; 1239 int status; 1240 1241 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1242 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1243 1244 if (!new_valid_dev(rdev)) 1245 return -EINVAL; 1246 1247 attr.ia_mode = mode; 1248 attr.ia_valid = ATTR_MODE; 1249 1250 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1251 if (status != 0) 1252 goto out_err; 1253 return 0; 1254 out_err: 1255 d_drop(dentry); 1256 return status; 1257 } 1258 1259 /* 1260 * See comments for nfs_proc_create regarding failed operations. 1261 */ 1262 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1263 { 1264 struct iattr attr; 1265 int error; 1266 1267 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1268 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1269 1270 attr.ia_valid = ATTR_MODE; 1271 attr.ia_mode = mode | S_IFDIR; 1272 1273 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1274 if (error != 0) 1275 goto out_err; 1276 return 0; 1277 out_err: 1278 d_drop(dentry); 1279 return error; 1280 } 1281 1282 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1283 { 1284 if (dentry->d_inode != NULL && !d_unhashed(dentry)) 1285 d_delete(dentry); 1286 } 1287 1288 static int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1289 { 1290 int error; 1291 1292 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1293 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1294 1295 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1296 /* Ensure the VFS deletes this inode */ 1297 if (error == 0 && dentry->d_inode != NULL) 1298 clear_nlink(dentry->d_inode); 1299 else if (error == -ENOENT) 1300 nfs_dentry_handle_enoent(dentry); 1301 1302 return error; 1303 } 1304 1305 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) 1306 { 1307 static unsigned int sillycounter; 1308 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2; 1309 const int countersize = sizeof(sillycounter)*2; 1310 const int slen = sizeof(".nfs")+fileidsize+countersize-1; 1311 char silly[slen+1]; 1312 struct qstr qsilly; 1313 struct dentry *sdentry; 1314 int error = -EIO; 1315 1316 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", 1317 dentry->d_parent->d_name.name, dentry->d_name.name, 1318 atomic_read(&dentry->d_count)); 1319 nfs_inc_stats(dir, NFSIOS_SILLYRENAME); 1320 1321 /* 1322 * We don't allow a dentry to be silly-renamed twice. 1323 */ 1324 error = -EBUSY; 1325 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1326 goto out; 1327 1328 sprintf(silly, ".nfs%*.*Lx", 1329 fileidsize, fileidsize, 1330 (unsigned long long)NFS_FILEID(dentry->d_inode)); 1331 1332 /* Return delegation in anticipation of the rename */ 1333 nfs_inode_return_delegation(dentry->d_inode); 1334 1335 sdentry = NULL; 1336 do { 1337 char *suffix = silly + slen - countersize; 1338 1339 dput(sdentry); 1340 sillycounter++; 1341 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); 1342 1343 dfprintk(VFS, "NFS: trying to rename %s to %s\n", 1344 dentry->d_name.name, silly); 1345 1346 sdentry = lookup_one_len(silly, dentry->d_parent, slen); 1347 /* 1348 * N.B. Better to return EBUSY here ... it could be 1349 * dangerous to delete the file while it's in use. 1350 */ 1351 if (IS_ERR(sdentry)) 1352 goto out; 1353 } while(sdentry->d_inode != NULL); /* need negative lookup */ 1354 1355 qsilly.name = silly; 1356 qsilly.len = strlen(silly); 1357 if (dentry->d_inode) { 1358 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1359 dir, &qsilly); 1360 nfs_mark_for_revalidate(dentry->d_inode); 1361 } else 1362 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1363 dir, &qsilly); 1364 if (!error) { 1365 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1366 d_move(dentry, sdentry); 1367 error = nfs_async_unlink(dir, dentry); 1368 /* If we return 0 we don't unlink */ 1369 } 1370 dput(sdentry); 1371 out: 1372 return error; 1373 } 1374 1375 /* 1376 * Remove a file after making sure there are no pending writes, 1377 * and after checking that the file has only one user. 1378 * 1379 * We invalidate the attribute cache and free the inode prior to the operation 1380 * to avoid possible races if the server reuses the inode. 1381 */ 1382 static int nfs_safe_remove(struct dentry *dentry) 1383 { 1384 struct inode *dir = dentry->d_parent->d_inode; 1385 struct inode *inode = dentry->d_inode; 1386 int error = -EBUSY; 1387 1388 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1389 dentry->d_parent->d_name.name, dentry->d_name.name); 1390 1391 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1392 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1393 error = 0; 1394 goto out; 1395 } 1396 1397 if (inode != NULL) { 1398 nfs_inode_return_delegation(inode); 1399 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1400 /* The VFS may want to delete this inode */ 1401 if (error == 0) 1402 nfs_drop_nlink(inode); 1403 nfs_mark_for_revalidate(inode); 1404 } else 1405 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1406 if (error == -ENOENT) 1407 nfs_dentry_handle_enoent(dentry); 1408 out: 1409 return error; 1410 } 1411 1412 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1413 * belongs to an active ".nfs..." file and we return -EBUSY. 1414 * 1415 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1416 */ 1417 static int nfs_unlink(struct inode *dir, struct dentry *dentry) 1418 { 1419 int error; 1420 int need_rehash = 0; 1421 1422 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1423 dir->i_ino, dentry->d_name.name); 1424 1425 spin_lock(&dcache_lock); 1426 spin_lock(&dentry->d_lock); 1427 if (atomic_read(&dentry->d_count) > 1) { 1428 spin_unlock(&dentry->d_lock); 1429 spin_unlock(&dcache_lock); 1430 /* Start asynchronous writeout of the inode */ 1431 write_inode_now(dentry->d_inode, 0); 1432 error = nfs_sillyrename(dir, dentry); 1433 return error; 1434 } 1435 if (!d_unhashed(dentry)) { 1436 __d_drop(dentry); 1437 need_rehash = 1; 1438 } 1439 spin_unlock(&dentry->d_lock); 1440 spin_unlock(&dcache_lock); 1441 error = nfs_safe_remove(dentry); 1442 if (!error || error == -ENOENT) { 1443 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1444 } else if (need_rehash) 1445 d_rehash(dentry); 1446 return error; 1447 } 1448 1449 /* 1450 * To create a symbolic link, most file systems instantiate a new inode, 1451 * add a page to it containing the path, then write it out to the disk 1452 * using prepare_write/commit_write. 1453 * 1454 * Unfortunately the NFS client can't create the in-core inode first 1455 * because it needs a file handle to create an in-core inode (see 1456 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1457 * symlink request has completed on the server. 1458 * 1459 * So instead we allocate a raw page, copy the symname into it, then do 1460 * the SYMLINK request with the page as the buffer. If it succeeds, we 1461 * now have a new file handle and can instantiate an in-core NFS inode 1462 * and move the raw page into its mapping. 1463 */ 1464 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1465 { 1466 struct pagevec lru_pvec; 1467 struct page *page; 1468 char *kaddr; 1469 struct iattr attr; 1470 unsigned int pathlen = strlen(symname); 1471 int error; 1472 1473 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1474 dir->i_ino, dentry->d_name.name, symname); 1475 1476 if (pathlen > PAGE_SIZE) 1477 return -ENAMETOOLONG; 1478 1479 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1480 attr.ia_valid = ATTR_MODE; 1481 1482 page = alloc_page(GFP_HIGHUSER); 1483 if (!page) 1484 return -ENOMEM; 1485 1486 kaddr = kmap_atomic(page, KM_USER0); 1487 memcpy(kaddr, symname, pathlen); 1488 if (pathlen < PAGE_SIZE) 1489 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1490 kunmap_atomic(kaddr, KM_USER0); 1491 1492 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1493 if (error != 0) { 1494 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1495 dir->i_sb->s_id, dir->i_ino, 1496 dentry->d_name.name, symname, error); 1497 d_drop(dentry); 1498 __free_page(page); 1499 return error; 1500 } 1501 1502 /* 1503 * No big deal if we can't add this page to the page cache here. 1504 * READLINK will get the missing page from the server if needed. 1505 */ 1506 pagevec_init(&lru_pvec, 0); 1507 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1508 GFP_KERNEL)) { 1509 pagevec_add(&lru_pvec, page); 1510 pagevec_lru_add(&lru_pvec); 1511 SetPageUptodate(page); 1512 unlock_page(page); 1513 } else 1514 __free_page(page); 1515 1516 return 0; 1517 } 1518 1519 static int 1520 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1521 { 1522 struct inode *inode = old_dentry->d_inode; 1523 int error; 1524 1525 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1526 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1527 dentry->d_parent->d_name.name, dentry->d_name.name); 1528 1529 d_drop(dentry); 1530 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1531 if (error == 0) { 1532 atomic_inc(&inode->i_count); 1533 d_add(dentry, inode); 1534 } 1535 return error; 1536 } 1537 1538 /* 1539 * RENAME 1540 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1541 * different file handle for the same inode after a rename (e.g. when 1542 * moving to a different directory). A fail-safe method to do so would 1543 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1544 * rename the old file using the sillyrename stuff. This way, the original 1545 * file in old_dir will go away when the last process iput()s the inode. 1546 * 1547 * FIXED. 1548 * 1549 * It actually works quite well. One needs to have the possibility for 1550 * at least one ".nfs..." file in each directory the file ever gets 1551 * moved or linked to which happens automagically with the new 1552 * implementation that only depends on the dcache stuff instead of 1553 * using the inode layer 1554 * 1555 * Unfortunately, things are a little more complicated than indicated 1556 * above. For a cross-directory move, we want to make sure we can get 1557 * rid of the old inode after the operation. This means there must be 1558 * no pending writes (if it's a file), and the use count must be 1. 1559 * If these conditions are met, we can drop the dentries before doing 1560 * the rename. 1561 */ 1562 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1563 struct inode *new_dir, struct dentry *new_dentry) 1564 { 1565 struct inode *old_inode = old_dentry->d_inode; 1566 struct inode *new_inode = new_dentry->d_inode; 1567 struct dentry *dentry = NULL, *rehash = NULL; 1568 int error = -EBUSY; 1569 1570 /* 1571 * To prevent any new references to the target during the rename, 1572 * we unhash the dentry and free the inode in advance. 1573 */ 1574 if (!d_unhashed(new_dentry)) { 1575 d_drop(new_dentry); 1576 rehash = new_dentry; 1577 } 1578 1579 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1580 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1581 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1582 atomic_read(&new_dentry->d_count)); 1583 1584 /* 1585 * First check whether the target is busy ... we can't 1586 * safely do _any_ rename if the target is in use. 1587 * 1588 * For files, make a copy of the dentry and then do a 1589 * silly-rename. If the silly-rename succeeds, the 1590 * copied dentry is hashed and becomes the new target. 1591 */ 1592 if (!new_inode) 1593 goto go_ahead; 1594 if (S_ISDIR(new_inode->i_mode)) { 1595 error = -EISDIR; 1596 if (!S_ISDIR(old_inode->i_mode)) 1597 goto out; 1598 } else if (atomic_read(&new_dentry->d_count) > 2) { 1599 int err; 1600 /* copy the target dentry's name */ 1601 dentry = d_alloc(new_dentry->d_parent, 1602 &new_dentry->d_name); 1603 if (!dentry) 1604 goto out; 1605 1606 /* silly-rename the existing target ... */ 1607 err = nfs_sillyrename(new_dir, new_dentry); 1608 if (!err) { 1609 new_dentry = rehash = dentry; 1610 new_inode = NULL; 1611 /* instantiate the replacement target */ 1612 d_instantiate(new_dentry, NULL); 1613 } else if (atomic_read(&new_dentry->d_count) > 1) 1614 /* dentry still busy? */ 1615 goto out; 1616 } else 1617 nfs_drop_nlink(new_inode); 1618 1619 go_ahead: 1620 /* 1621 * ... prune child dentries and writebacks if needed. 1622 */ 1623 if (atomic_read(&old_dentry->d_count) > 1) { 1624 if (S_ISREG(old_inode->i_mode)) 1625 nfs_wb_all(old_inode); 1626 shrink_dcache_parent(old_dentry); 1627 } 1628 nfs_inode_return_delegation(old_inode); 1629 1630 if (new_inode != NULL) { 1631 nfs_inode_return_delegation(new_inode); 1632 d_delete(new_dentry); 1633 } 1634 1635 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1636 new_dir, &new_dentry->d_name); 1637 nfs_mark_for_revalidate(old_inode); 1638 out: 1639 if (rehash) 1640 d_rehash(rehash); 1641 if (!error) { 1642 d_move(old_dentry, new_dentry); 1643 nfs_set_verifier(new_dentry, 1644 nfs_save_change_attribute(new_dir)); 1645 } else if (error == -ENOENT) 1646 nfs_dentry_handle_enoent(old_dentry); 1647 1648 /* new dentry created? */ 1649 if (dentry) 1650 dput(dentry); 1651 return error; 1652 } 1653 1654 static DEFINE_SPINLOCK(nfs_access_lru_lock); 1655 static LIST_HEAD(nfs_access_lru_list); 1656 static atomic_long_t nfs_access_nr_entries; 1657 1658 static void nfs_access_free_entry(struct nfs_access_entry *entry) 1659 { 1660 put_rpccred(entry->cred); 1661 kfree(entry); 1662 smp_mb__before_atomic_dec(); 1663 atomic_long_dec(&nfs_access_nr_entries); 1664 smp_mb__after_atomic_dec(); 1665 } 1666 1667 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) 1668 { 1669 LIST_HEAD(head); 1670 struct nfs_inode *nfsi; 1671 struct nfs_access_entry *cache; 1672 1673 restart: 1674 spin_lock(&nfs_access_lru_lock); 1675 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { 1676 struct rw_semaphore *s_umount; 1677 struct inode *inode; 1678 1679 if (nr_to_scan-- == 0) 1680 break; 1681 s_umount = &nfsi->vfs_inode.i_sb->s_umount; 1682 if (!down_read_trylock(s_umount)) 1683 continue; 1684 inode = igrab(&nfsi->vfs_inode); 1685 if (inode == NULL) { 1686 up_read(s_umount); 1687 continue; 1688 } 1689 spin_lock(&inode->i_lock); 1690 if (list_empty(&nfsi->access_cache_entry_lru)) 1691 goto remove_lru_entry; 1692 cache = list_entry(nfsi->access_cache_entry_lru.next, 1693 struct nfs_access_entry, lru); 1694 list_move(&cache->lru, &head); 1695 rb_erase(&cache->rb_node, &nfsi->access_cache); 1696 if (!list_empty(&nfsi->access_cache_entry_lru)) 1697 list_move_tail(&nfsi->access_cache_inode_lru, 1698 &nfs_access_lru_list); 1699 else { 1700 remove_lru_entry: 1701 list_del_init(&nfsi->access_cache_inode_lru); 1702 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1703 } 1704 spin_unlock(&inode->i_lock); 1705 spin_unlock(&nfs_access_lru_lock); 1706 iput(inode); 1707 up_read(s_umount); 1708 goto restart; 1709 } 1710 spin_unlock(&nfs_access_lru_lock); 1711 while (!list_empty(&head)) { 1712 cache = list_entry(head.next, struct nfs_access_entry, lru); 1713 list_del(&cache->lru); 1714 nfs_access_free_entry(cache); 1715 } 1716 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; 1717 } 1718 1719 static void __nfs_access_zap_cache(struct inode *inode) 1720 { 1721 struct nfs_inode *nfsi = NFS_I(inode); 1722 struct rb_root *root_node = &nfsi->access_cache; 1723 struct rb_node *n, *dispose = NULL; 1724 struct nfs_access_entry *entry; 1725 1726 /* Unhook entries from the cache */ 1727 while ((n = rb_first(root_node)) != NULL) { 1728 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1729 rb_erase(n, root_node); 1730 list_del(&entry->lru); 1731 n->rb_left = dispose; 1732 dispose = n; 1733 } 1734 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 1735 spin_unlock(&inode->i_lock); 1736 1737 /* Now kill them all! */ 1738 while (dispose != NULL) { 1739 n = dispose; 1740 dispose = n->rb_left; 1741 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); 1742 } 1743 } 1744 1745 void nfs_access_zap_cache(struct inode *inode) 1746 { 1747 /* Remove from global LRU init */ 1748 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 1749 spin_lock(&nfs_access_lru_lock); 1750 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 1751 spin_unlock(&nfs_access_lru_lock); 1752 } 1753 1754 spin_lock(&inode->i_lock); 1755 /* This will release the spinlock */ 1756 __nfs_access_zap_cache(inode); 1757 } 1758 1759 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 1760 { 1761 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 1762 struct nfs_access_entry *entry; 1763 1764 while (n != NULL) { 1765 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1766 1767 if (cred < entry->cred) 1768 n = n->rb_left; 1769 else if (cred > entry->cred) 1770 n = n->rb_right; 1771 else 1772 return entry; 1773 } 1774 return NULL; 1775 } 1776 1777 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 1778 { 1779 struct nfs_inode *nfsi = NFS_I(inode); 1780 struct nfs_access_entry *cache; 1781 int err = -ENOENT; 1782 1783 spin_lock(&inode->i_lock); 1784 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 1785 goto out_zap; 1786 cache = nfs_access_search_rbtree(inode, cred); 1787 if (cache == NULL) 1788 goto out; 1789 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 1790 goto out_stale; 1791 res->jiffies = cache->jiffies; 1792 res->cred = cache->cred; 1793 res->mask = cache->mask; 1794 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 1795 err = 0; 1796 out: 1797 spin_unlock(&inode->i_lock); 1798 return err; 1799 out_stale: 1800 rb_erase(&cache->rb_node, &nfsi->access_cache); 1801 list_del(&cache->lru); 1802 spin_unlock(&inode->i_lock); 1803 nfs_access_free_entry(cache); 1804 return -ENOENT; 1805 out_zap: 1806 /* This will release the spinlock */ 1807 __nfs_access_zap_cache(inode); 1808 return -ENOENT; 1809 } 1810 1811 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 1812 { 1813 struct nfs_inode *nfsi = NFS_I(inode); 1814 struct rb_root *root_node = &nfsi->access_cache; 1815 struct rb_node **p = &root_node->rb_node; 1816 struct rb_node *parent = NULL; 1817 struct nfs_access_entry *entry; 1818 1819 spin_lock(&inode->i_lock); 1820 while (*p != NULL) { 1821 parent = *p; 1822 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 1823 1824 if (set->cred < entry->cred) 1825 p = &parent->rb_left; 1826 else if (set->cred > entry->cred) 1827 p = &parent->rb_right; 1828 else 1829 goto found; 1830 } 1831 rb_link_node(&set->rb_node, parent, p); 1832 rb_insert_color(&set->rb_node, root_node); 1833 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1834 spin_unlock(&inode->i_lock); 1835 return; 1836 found: 1837 rb_replace_node(parent, &set->rb_node, root_node); 1838 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1839 list_del(&entry->lru); 1840 spin_unlock(&inode->i_lock); 1841 nfs_access_free_entry(entry); 1842 } 1843 1844 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 1845 { 1846 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 1847 if (cache == NULL) 1848 return; 1849 RB_CLEAR_NODE(&cache->rb_node); 1850 cache->jiffies = set->jiffies; 1851 cache->cred = get_rpccred(set->cred); 1852 cache->mask = set->mask; 1853 1854 nfs_access_add_rbtree(inode, cache); 1855 1856 /* Update accounting */ 1857 smp_mb__before_atomic_inc(); 1858 atomic_long_inc(&nfs_access_nr_entries); 1859 smp_mb__after_atomic_inc(); 1860 1861 /* Add inode to global LRU list */ 1862 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 1863 spin_lock(&nfs_access_lru_lock); 1864 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); 1865 spin_unlock(&nfs_access_lru_lock); 1866 } 1867 } 1868 1869 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 1870 { 1871 struct nfs_access_entry cache; 1872 int status; 1873 1874 status = nfs_access_get_cached(inode, cred, &cache); 1875 if (status == 0) 1876 goto out; 1877 1878 /* Be clever: ask server to check for all possible rights */ 1879 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 1880 cache.cred = cred; 1881 cache.jiffies = jiffies; 1882 status = NFS_PROTO(inode)->access(inode, &cache); 1883 if (status != 0) 1884 return status; 1885 nfs_access_add_cache(inode, &cache); 1886 out: 1887 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 1888 return 0; 1889 return -EACCES; 1890 } 1891 1892 static int nfs_open_permission_mask(int openflags) 1893 { 1894 int mask = 0; 1895 1896 if (openflags & FMODE_READ) 1897 mask |= MAY_READ; 1898 if (openflags & FMODE_WRITE) 1899 mask |= MAY_WRITE; 1900 if (openflags & FMODE_EXEC) 1901 mask |= MAY_EXEC; 1902 return mask; 1903 } 1904 1905 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 1906 { 1907 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 1908 } 1909 1910 int nfs_permission(struct inode *inode, int mask) 1911 { 1912 struct rpc_cred *cred; 1913 int res = 0; 1914 1915 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 1916 1917 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 1918 goto out; 1919 /* Is this sys_access() ? */ 1920 if (mask & MAY_ACCESS) 1921 goto force_lookup; 1922 1923 switch (inode->i_mode & S_IFMT) { 1924 case S_IFLNK: 1925 goto out; 1926 case S_IFREG: 1927 /* NFSv4 has atomic_open... */ 1928 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) 1929 && (mask & MAY_OPEN)) 1930 goto out; 1931 break; 1932 case S_IFDIR: 1933 /* 1934 * Optimize away all write operations, since the server 1935 * will check permissions when we perform the op. 1936 */ 1937 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 1938 goto out; 1939 } 1940 1941 force_lookup: 1942 if (!NFS_PROTO(inode)->access) 1943 goto out_notsup; 1944 1945 cred = rpc_lookup_cred(); 1946 if (!IS_ERR(cred)) { 1947 res = nfs_do_access(inode, cred, mask); 1948 put_rpccred(cred); 1949 } else 1950 res = PTR_ERR(cred); 1951 out: 1952 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 1953 inode->i_sb->s_id, inode->i_ino, mask, res); 1954 return res; 1955 out_notsup: 1956 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1957 if (res == 0) 1958 res = generic_permission(inode, mask, NULL); 1959 goto out; 1960 } 1961 1962 /* 1963 * Local variables: 1964 * version-control: t 1965 * kept-new-versions: 5 1966 * End: 1967 */ 1968