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 unsigned long gencount; 160 int timestamp_valid; 161 } nfs_readdir_descriptor_t; 162 163 /* Now we cache directories properly, by stuffing the dirent 164 * data directly in the page cache. 165 * 166 * Inode invalidation due to refresh etc. takes care of 167 * _everything_, no sloppy entry flushing logic, no extraneous 168 * copying, network direct to page cache, the way it was meant 169 * to be. 170 * 171 * NOTE: Dirent information verification is done always by the 172 * page-in of the RPC reply, nowhere else, this simplies 173 * things substantially. 174 */ 175 static 176 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page) 177 { 178 struct file *file = desc->file; 179 struct inode *inode = file->f_path.dentry->d_inode; 180 struct rpc_cred *cred = nfs_file_cred(file); 181 unsigned long timestamp, gencount; 182 int error; 183 184 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n", 185 __func__, (long long)desc->entry->cookie, 186 page->index); 187 188 again: 189 timestamp = jiffies; 190 gencount = nfs_inc_attr_generation_counter(); 191 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page, 192 NFS_SERVER(inode)->dtsize, desc->plus); 193 if (error < 0) { 194 /* We requested READDIRPLUS, but the server doesn't grok it */ 195 if (error == -ENOTSUPP && desc->plus) { 196 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 197 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 198 desc->plus = 0; 199 goto again; 200 } 201 goto error; 202 } 203 desc->timestamp = timestamp; 204 desc->gencount = gencount; 205 desc->timestamp_valid = 1; 206 SetPageUptodate(page); 207 /* Ensure consistent page alignment of the data. 208 * Note: assumes we have exclusive access to this mapping either 209 * through inode->i_mutex or some other mechanism. 210 */ 211 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 212 /* Should never happen */ 213 nfs_zap_mapping(inode, inode->i_mapping); 214 } 215 unlock_page(page); 216 return 0; 217 error: 218 unlock_page(page); 219 return -EIO; 220 } 221 222 static inline 223 int dir_decode(nfs_readdir_descriptor_t *desc) 224 { 225 __be32 *p = desc->ptr; 226 p = desc->decode(p, desc->entry, desc->plus); 227 if (IS_ERR(p)) 228 return PTR_ERR(p); 229 desc->ptr = p; 230 if (desc->timestamp_valid) { 231 desc->entry->fattr->time_start = desc->timestamp; 232 desc->entry->fattr->gencount = desc->gencount; 233 } else 234 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR; 235 return 0; 236 } 237 238 static inline 239 void dir_page_release(nfs_readdir_descriptor_t *desc) 240 { 241 kunmap(desc->page); 242 page_cache_release(desc->page); 243 desc->page = NULL; 244 desc->ptr = NULL; 245 } 246 247 /* 248 * Given a pointer to a buffer that has already been filled by a call 249 * to readdir, find the next entry with cookie '*desc->dir_cookie'. 250 * 251 * If the end of the buffer has been reached, return -EAGAIN, if not, 252 * return the offset within the buffer of the next entry to be 253 * read. 254 */ 255 static inline 256 int find_dirent(nfs_readdir_descriptor_t *desc) 257 { 258 struct nfs_entry *entry = desc->entry; 259 int loop_count = 0, 260 status; 261 262 while((status = dir_decode(desc)) == 0) { 263 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n", 264 __func__, (unsigned long long)entry->cookie); 265 if (entry->prev_cookie == *desc->dir_cookie) 266 break; 267 if (loop_count++ > 200) { 268 loop_count = 0; 269 schedule(); 270 } 271 } 272 return status; 273 } 274 275 /* 276 * Given a pointer to a buffer that has already been filled by a call 277 * to readdir, find the entry at offset 'desc->file->f_pos'. 278 * 279 * If the end of the buffer has been reached, return -EAGAIN, if not, 280 * return the offset within the buffer of the next entry to be 281 * read. 282 */ 283 static inline 284 int find_dirent_index(nfs_readdir_descriptor_t *desc) 285 { 286 struct nfs_entry *entry = desc->entry; 287 int loop_count = 0, 288 status; 289 290 for(;;) { 291 status = dir_decode(desc); 292 if (status) 293 break; 294 295 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n", 296 (unsigned long long)entry->cookie, desc->current_index); 297 298 if (desc->file->f_pos == desc->current_index) { 299 *desc->dir_cookie = entry->cookie; 300 break; 301 } 302 desc->current_index++; 303 if (loop_count++ > 200) { 304 loop_count = 0; 305 schedule(); 306 } 307 } 308 return status; 309 } 310 311 /* 312 * Find the given page, and call find_dirent() or find_dirent_index in 313 * order to try to return the next entry. 314 */ 315 static inline 316 int find_dirent_page(nfs_readdir_descriptor_t *desc) 317 { 318 struct inode *inode = desc->file->f_path.dentry->d_inode; 319 struct page *page; 320 int status; 321 322 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n", 323 __func__, desc->page_index, 324 (long long) *desc->dir_cookie); 325 326 /* If we find the page in the page_cache, we cannot be sure 327 * how fresh the data is, so we will ignore readdir_plus attributes. 328 */ 329 desc->timestamp_valid = 0; 330 page = read_cache_page(inode->i_mapping, desc->page_index, 331 (filler_t *)nfs_readdir_filler, desc); 332 if (IS_ERR(page)) { 333 status = PTR_ERR(page); 334 goto out; 335 } 336 337 /* NOTE: Someone else may have changed the READDIRPLUS flag */ 338 desc->page = page; 339 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 340 if (*desc->dir_cookie != 0) 341 status = find_dirent(desc); 342 else 343 status = find_dirent_index(desc); 344 if (status < 0) 345 dir_page_release(desc); 346 out: 347 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status); 348 return status; 349 } 350 351 /* 352 * Recurse through the page cache pages, and return a 353 * filled nfs_entry structure of the next directory entry if possible. 354 * 355 * The target for the search is '*desc->dir_cookie' if non-0, 356 * 'desc->file->f_pos' otherwise 357 */ 358 static inline 359 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 360 { 361 int loop_count = 0; 362 int res; 363 364 /* Always search-by-index from the beginning of the cache */ 365 if (*desc->dir_cookie == 0) { 366 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n", 367 (long long)desc->file->f_pos); 368 desc->page_index = 0; 369 desc->entry->cookie = desc->entry->prev_cookie = 0; 370 desc->entry->eof = 0; 371 desc->current_index = 0; 372 } else 373 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", 374 (unsigned long long)*desc->dir_cookie); 375 376 for (;;) { 377 res = find_dirent_page(desc); 378 if (res != -EAGAIN) 379 break; 380 /* Align to beginning of next page */ 381 desc->page_index ++; 382 if (loop_count++ > 200) { 383 loop_count = 0; 384 schedule(); 385 } 386 } 387 388 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, res); 389 return res; 390 } 391 392 static inline unsigned int dt_type(struct inode *inode) 393 { 394 return (inode->i_mode >> 12) & 15; 395 } 396 397 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc); 398 399 /* 400 * Once we've found the start of the dirent within a page: fill 'er up... 401 */ 402 static 403 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, 404 filldir_t filldir) 405 { 406 struct file *file = desc->file; 407 struct nfs_entry *entry = desc->entry; 408 struct dentry *dentry = NULL; 409 u64 fileid; 410 int loop_count = 0, 411 res; 412 413 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", 414 (unsigned long long)entry->cookie); 415 416 for(;;) { 417 unsigned d_type = DT_UNKNOWN; 418 /* Note: entry->prev_cookie contains the cookie for 419 * retrieving the current dirent on the server */ 420 fileid = entry->ino; 421 422 /* Get a dentry if we have one */ 423 if (dentry != NULL) 424 dput(dentry); 425 dentry = nfs_readdir_lookup(desc); 426 427 /* Use readdirplus info */ 428 if (dentry != NULL && dentry->d_inode != NULL) { 429 d_type = dt_type(dentry->d_inode); 430 fileid = NFS_FILEID(dentry->d_inode); 431 } 432 433 res = filldir(dirent, entry->name, entry->len, 434 file->f_pos, nfs_compat_user_ino64(fileid), 435 d_type); 436 if (res < 0) 437 break; 438 file->f_pos++; 439 *desc->dir_cookie = entry->cookie; 440 if (dir_decode(desc) != 0) { 441 desc->page_index ++; 442 break; 443 } 444 if (loop_count++ > 200) { 445 loop_count = 0; 446 schedule(); 447 } 448 } 449 dir_page_release(desc); 450 if (dentry != NULL) 451 dput(dentry); 452 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 453 (unsigned long long)*desc->dir_cookie, res); 454 return res; 455 } 456 457 /* 458 * If we cannot find a cookie in our cache, we suspect that this is 459 * because it points to a deleted file, so we ask the server to return 460 * whatever it thinks is the next entry. We then feed this to filldir. 461 * If all goes well, we should then be able to find our way round the 462 * cache on the next call to readdir_search_pagecache(); 463 * 464 * NOTE: we cannot add the anonymous page to the pagecache because 465 * the data it contains might not be page aligned. Besides, 466 * we should already have a complete representation of the 467 * directory in the page cache by the time we get here. 468 */ 469 static inline 470 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, 471 filldir_t filldir) 472 { 473 struct file *file = desc->file; 474 struct inode *inode = file->f_path.dentry->d_inode; 475 struct rpc_cred *cred = nfs_file_cred(file); 476 struct page *page = NULL; 477 int status; 478 unsigned long timestamp, gencount; 479 480 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 481 (unsigned long long)*desc->dir_cookie); 482 483 page = alloc_page(GFP_HIGHUSER); 484 if (!page) { 485 status = -ENOMEM; 486 goto out; 487 } 488 timestamp = jiffies; 489 gencount = nfs_inc_attr_generation_counter(); 490 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, 491 *desc->dir_cookie, page, 492 NFS_SERVER(inode)->dtsize, 493 desc->plus); 494 desc->page = page; 495 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 496 if (status >= 0) { 497 desc->timestamp = timestamp; 498 desc->gencount = gencount; 499 desc->timestamp_valid = 1; 500 if ((status = dir_decode(desc)) == 0) 501 desc->entry->prev_cookie = *desc->dir_cookie; 502 } else 503 status = -EIO; 504 if (status < 0) 505 goto out_release; 506 507 status = nfs_do_filldir(desc, dirent, filldir); 508 509 /* Reset read descriptor so it searches the page cache from 510 * the start upon the next call to readdir_search_pagecache() */ 511 desc->page_index = 0; 512 desc->entry->cookie = desc->entry->prev_cookie = 0; 513 desc->entry->eof = 0; 514 out: 515 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 516 __func__, status); 517 return status; 518 out_release: 519 dir_page_release(desc); 520 goto out; 521 } 522 523 /* The file offset position represents the dirent entry number. A 524 last cookie cache takes care of the common case of reading the 525 whole directory. 526 */ 527 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) 528 { 529 struct dentry *dentry = filp->f_path.dentry; 530 struct inode *inode = dentry->d_inode; 531 nfs_readdir_descriptor_t my_desc, 532 *desc = &my_desc; 533 struct nfs_entry my_entry; 534 struct nfs_fh fh; 535 struct nfs_fattr fattr; 536 long res; 537 538 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n", 539 dentry->d_parent->d_name.name, dentry->d_name.name, 540 (long long)filp->f_pos); 541 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 542 543 /* 544 * filp->f_pos points to the dirent entry number. 545 * *desc->dir_cookie has the cookie for the next entry. We have 546 * to either find the entry with the appropriate number or 547 * revalidate the cookie. 548 */ 549 memset(desc, 0, sizeof(*desc)); 550 551 desc->file = filp; 552 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; 553 desc->decode = NFS_PROTO(inode)->decode_dirent; 554 desc->plus = NFS_USE_READDIRPLUS(inode); 555 556 my_entry.cookie = my_entry.prev_cookie = 0; 557 my_entry.eof = 0; 558 my_entry.fh = &fh; 559 my_entry.fattr = &fattr; 560 nfs_fattr_init(&fattr); 561 desc->entry = &my_entry; 562 563 nfs_block_sillyrename(dentry); 564 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping); 565 if (res < 0) 566 goto out; 567 568 while(!desc->entry->eof) { 569 res = readdir_search_pagecache(desc); 570 571 if (res == -EBADCOOKIE) { 572 /* This means either end of directory */ 573 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { 574 /* Or that the server has 'lost' a cookie */ 575 res = uncached_readdir(desc, dirent, filldir); 576 if (res >= 0) 577 continue; 578 } 579 res = 0; 580 break; 581 } 582 if (res == -ETOOSMALL && desc->plus) { 583 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 584 nfs_zap_caches(inode); 585 desc->plus = 0; 586 desc->entry->eof = 0; 587 continue; 588 } 589 if (res < 0) 590 break; 591 592 res = nfs_do_filldir(desc, dirent, filldir); 593 if (res < 0) { 594 res = 0; 595 break; 596 } 597 } 598 out: 599 nfs_unblock_sillyrename(dentry); 600 if (res > 0) 601 res = 0; 602 dfprintk(FILE, "NFS: readdir(%s/%s) returns %ld\n", 603 dentry->d_parent->d_name.name, dentry->d_name.name, 604 res); 605 return res; 606 } 607 608 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) 609 { 610 struct dentry *dentry = filp->f_path.dentry; 611 struct inode *inode = dentry->d_inode; 612 613 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", 614 dentry->d_parent->d_name.name, 615 dentry->d_name.name, 616 offset, origin); 617 618 mutex_lock(&inode->i_mutex); 619 switch (origin) { 620 case 1: 621 offset += filp->f_pos; 622 case 0: 623 if (offset >= 0) 624 break; 625 default: 626 offset = -EINVAL; 627 goto out; 628 } 629 if (offset != filp->f_pos) { 630 filp->f_pos = offset; 631 nfs_file_open_context(filp)->dir_cookie = 0; 632 } 633 out: 634 mutex_unlock(&inode->i_mutex); 635 return offset; 636 } 637 638 /* 639 * All directory operations under NFS are synchronous, so fsync() 640 * is a dummy operation. 641 */ 642 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) 643 { 644 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", 645 dentry->d_parent->d_name.name, dentry->d_name.name, 646 datasync); 647 648 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); 649 return 0; 650 } 651 652 /** 653 * nfs_force_lookup_revalidate - Mark the directory as having changed 654 * @dir - pointer to directory inode 655 * 656 * This forces the revalidation code in nfs_lookup_revalidate() to do a 657 * full lookup on all child dentries of 'dir' whenever a change occurs 658 * on the server that might have invalidated our dcache. 659 * 660 * The caller should be holding dir->i_lock 661 */ 662 void nfs_force_lookup_revalidate(struct inode *dir) 663 { 664 NFS_I(dir)->cache_change_attribute++; 665 } 666 667 /* 668 * A check for whether or not the parent directory has changed. 669 * In the case it has, we assume that the dentries are untrustworthy 670 * and may need to be looked up again. 671 */ 672 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) 673 { 674 if (IS_ROOT(dentry)) 675 return 1; 676 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 677 return 0; 678 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 679 return 0; 680 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 681 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 682 return 0; 683 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 684 return 0; 685 return 1; 686 } 687 688 /* 689 * Return the intent data that applies to this particular path component 690 * 691 * Note that the current set of intents only apply to the very last 692 * component of the path. 693 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. 694 */ 695 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) 696 { 697 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) 698 return 0; 699 return nd->flags & mask; 700 } 701 702 /* 703 * Use intent information to check whether or not we're going to do 704 * an O_EXCL create using this path component. 705 */ 706 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) 707 { 708 if (NFS_PROTO(dir)->version == 2) 709 return 0; 710 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL); 711 } 712 713 /* 714 * Inode and filehandle revalidation for lookups. 715 * 716 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 717 * or if the intent information indicates that we're about to open this 718 * particular file and the "nocto" mount flag is not set. 719 * 720 */ 721 static inline 722 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) 723 { 724 struct nfs_server *server = NFS_SERVER(inode); 725 726 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags)) 727 return 0; 728 if (nd != NULL) { 729 /* VFS wants an on-the-wire revalidation */ 730 if (nd->flags & LOOKUP_REVAL) 731 goto out_force; 732 /* This is an open(2) */ 733 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && 734 !(server->flags & NFS_MOUNT_NOCTO) && 735 (S_ISREG(inode->i_mode) || 736 S_ISDIR(inode->i_mode))) 737 goto out_force; 738 return 0; 739 } 740 return nfs_revalidate_inode(server, inode); 741 out_force: 742 return __nfs_revalidate_inode(server, inode); 743 } 744 745 /* 746 * We judge how long we want to trust negative 747 * dentries by looking at the parent inode mtime. 748 * 749 * If parent mtime has changed, we revalidate, else we wait for a 750 * period corresponding to the parent's attribute cache timeout value. 751 */ 752 static inline 753 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 754 struct nameidata *nd) 755 { 756 /* Don't revalidate a negative dentry if we're creating a new file */ 757 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) 758 return 0; 759 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 760 return 1; 761 return !nfs_check_verifier(dir, dentry); 762 } 763 764 /* 765 * This is called every time the dcache has a lookup hit, 766 * and we should check whether we can really trust that 767 * lookup. 768 * 769 * NOTE! The hit can be a negative hit too, don't assume 770 * we have an inode! 771 * 772 * If the parent directory is seen to have changed, we throw out the 773 * cached dentry and do a new lookup. 774 */ 775 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) 776 { 777 struct inode *dir; 778 struct inode *inode; 779 struct dentry *parent; 780 int error; 781 struct nfs_fh fhandle; 782 struct nfs_fattr fattr; 783 784 parent = dget_parent(dentry); 785 dir = parent->d_inode; 786 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 787 inode = dentry->d_inode; 788 789 if (!inode) { 790 if (nfs_neg_need_reval(dir, dentry, nd)) 791 goto out_bad; 792 goto out_valid; 793 } 794 795 if (is_bad_inode(inode)) { 796 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", 797 __func__, dentry->d_parent->d_name.name, 798 dentry->d_name.name); 799 goto out_bad; 800 } 801 802 /* Force a full look up iff the parent directory has changed */ 803 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { 804 if (nfs_lookup_verify_inode(inode, nd)) 805 goto out_zap_parent; 806 goto out_valid; 807 } 808 809 if (NFS_STALE(inode)) 810 goto out_bad; 811 812 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 813 if (error) 814 goto out_bad; 815 if (nfs_compare_fh(NFS_FH(inode), &fhandle)) 816 goto out_bad; 817 if ((error = nfs_refresh_inode(inode, &fattr)) != 0) 818 goto out_bad; 819 820 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 821 out_valid: 822 dput(parent); 823 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", 824 __func__, dentry->d_parent->d_name.name, 825 dentry->d_name.name); 826 return 1; 827 out_zap_parent: 828 nfs_zap_caches(dir); 829 out_bad: 830 nfs_mark_for_revalidate(dir); 831 if (inode && S_ISDIR(inode->i_mode)) { 832 /* Purge readdir caches. */ 833 nfs_zap_caches(inode); 834 /* If we have submounts, don't unhash ! */ 835 if (have_submounts(dentry)) 836 goto out_valid; 837 shrink_dcache_parent(dentry); 838 } 839 d_drop(dentry); 840 dput(parent); 841 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", 842 __func__, dentry->d_parent->d_name.name, 843 dentry->d_name.name); 844 return 0; 845 } 846 847 /* 848 * This is called from dput() when d_count is going to 0. 849 */ 850 static int nfs_dentry_delete(struct dentry *dentry) 851 { 852 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", 853 dentry->d_parent->d_name.name, dentry->d_name.name, 854 dentry->d_flags); 855 856 /* Unhash any dentry with a stale inode */ 857 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) 858 return 1; 859 860 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 861 /* Unhash it, so that ->d_iput() would be called */ 862 return 1; 863 } 864 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 865 /* Unhash it, so that ancestors of killed async unlink 866 * files will be cleaned up during umount */ 867 return 1; 868 } 869 return 0; 870 871 } 872 873 static void nfs_drop_nlink(struct inode *inode) 874 { 875 spin_lock(&inode->i_lock); 876 if (inode->i_nlink > 0) 877 drop_nlink(inode); 878 spin_unlock(&inode->i_lock); 879 } 880 881 /* 882 * Called when the dentry loses inode. 883 * We use it to clean up silly-renamed files. 884 */ 885 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 886 { 887 if (S_ISDIR(inode->i_mode)) 888 /* drop any readdir cache as it could easily be old */ 889 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 890 891 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 892 drop_nlink(inode); 893 nfs_complete_unlink(dentry, inode); 894 } 895 iput(inode); 896 } 897 898 struct dentry_operations nfs_dentry_operations = { 899 .d_revalidate = nfs_lookup_revalidate, 900 .d_delete = nfs_dentry_delete, 901 .d_iput = nfs_dentry_iput, 902 }; 903 904 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 905 { 906 struct dentry *res; 907 struct dentry *parent; 908 struct inode *inode = NULL; 909 int error; 910 struct nfs_fh fhandle; 911 struct nfs_fattr fattr; 912 913 dfprintk(VFS, "NFS: lookup(%s/%s)\n", 914 dentry->d_parent->d_name.name, dentry->d_name.name); 915 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 916 917 res = ERR_PTR(-ENAMETOOLONG); 918 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 919 goto out; 920 921 res = ERR_PTR(-ENOMEM); 922 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 923 924 /* 925 * If we're doing an exclusive create, optimize away the lookup 926 * but don't hash the dentry. 927 */ 928 if (nfs_is_exclusive_create(dir, nd)) { 929 d_instantiate(dentry, NULL); 930 res = NULL; 931 goto out; 932 } 933 934 parent = dentry->d_parent; 935 /* Protect against concurrent sillydeletes */ 936 nfs_block_sillyrename(parent); 937 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 938 if (error == -ENOENT) 939 goto no_entry; 940 if (error < 0) { 941 res = ERR_PTR(error); 942 goto out_unblock_sillyrename; 943 } 944 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); 945 res = (struct dentry *)inode; 946 if (IS_ERR(res)) 947 goto out_unblock_sillyrename; 948 949 no_entry: 950 res = d_materialise_unique(dentry, inode); 951 if (res != NULL) { 952 if (IS_ERR(res)) 953 goto out_unblock_sillyrename; 954 dentry = res; 955 } 956 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 957 out_unblock_sillyrename: 958 nfs_unblock_sillyrename(parent); 959 out: 960 return res; 961 } 962 963 #ifdef CONFIG_NFS_V4 964 static int nfs_open_revalidate(struct dentry *, struct nameidata *); 965 966 struct dentry_operations nfs4_dentry_operations = { 967 .d_revalidate = nfs_open_revalidate, 968 .d_delete = nfs_dentry_delete, 969 .d_iput = nfs_dentry_iput, 970 }; 971 972 /* 973 * Use intent information to determine whether we need to substitute 974 * the NFSv4-style stateful OPEN for the LOOKUP call 975 */ 976 static int is_atomic_open(struct inode *dir, struct nameidata *nd) 977 { 978 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) 979 return 0; 980 /* NFS does not (yet) have a stateful open for directories */ 981 if (nd->flags & LOOKUP_DIRECTORY) 982 return 0; 983 /* Are we trying to write to a read only partition? */ 984 if (__mnt_is_readonly(nd->path.mnt) && 985 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) 986 return 0; 987 return 1; 988 } 989 990 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 991 { 992 struct dentry *res = NULL; 993 int error; 994 995 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", 996 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 997 998 /* Check that we are indeed trying to open this file */ 999 if (!is_atomic_open(dir, nd)) 1000 goto no_open; 1001 1002 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { 1003 res = ERR_PTR(-ENAMETOOLONG); 1004 goto out; 1005 } 1006 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1007 1008 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash 1009 * the dentry. */ 1010 if (nd->flags & LOOKUP_EXCL) { 1011 d_instantiate(dentry, NULL); 1012 goto out; 1013 } 1014 1015 /* Open the file on the server */ 1016 res = nfs4_atomic_open(dir, dentry, nd); 1017 if (IS_ERR(res)) { 1018 error = PTR_ERR(res); 1019 switch (error) { 1020 /* Make a negative dentry */ 1021 case -ENOENT: 1022 res = NULL; 1023 goto out; 1024 /* This turned out not to be a regular file */ 1025 case -EISDIR: 1026 case -ENOTDIR: 1027 goto no_open; 1028 case -ELOOP: 1029 if (!(nd->intent.open.flags & O_NOFOLLOW)) 1030 goto no_open; 1031 /* case -EINVAL: */ 1032 default: 1033 goto out; 1034 } 1035 } else if (res != NULL) 1036 dentry = res; 1037 out: 1038 return res; 1039 no_open: 1040 return nfs_lookup(dir, dentry, nd); 1041 } 1042 1043 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) 1044 { 1045 struct dentry *parent = NULL; 1046 struct inode *inode = dentry->d_inode; 1047 struct inode *dir; 1048 int openflags, ret = 0; 1049 1050 parent = dget_parent(dentry); 1051 dir = parent->d_inode; 1052 if (!is_atomic_open(dir, nd)) 1053 goto no_open; 1054 /* We can't create new files in nfs_open_revalidate(), so we 1055 * optimize away revalidation of negative dentries. 1056 */ 1057 if (inode == NULL) { 1058 if (!nfs_neg_need_reval(dir, dentry, nd)) 1059 ret = 1; 1060 goto out; 1061 } 1062 1063 /* NFS only supports OPEN on regular files */ 1064 if (!S_ISREG(inode->i_mode)) 1065 goto no_open; 1066 openflags = nd->intent.open.flags; 1067 /* We cannot do exclusive creation on a positive dentry */ 1068 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) 1069 goto no_open; 1070 /* We can't create new files, or truncate existing ones here */ 1071 openflags &= ~(O_CREAT|O_TRUNC); 1072 1073 /* 1074 * Note: we're not holding inode->i_mutex and so may be racing with 1075 * operations that change the directory. We therefore save the 1076 * change attribute *before* we do the RPC call. 1077 */ 1078 ret = nfs4_open_revalidate(dir, dentry, openflags, nd); 1079 out: 1080 dput(parent); 1081 if (!ret) 1082 d_drop(dentry); 1083 return ret; 1084 no_open: 1085 dput(parent); 1086 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) 1087 return 1; 1088 return nfs_lookup_revalidate(dentry, nd); 1089 } 1090 #endif /* CONFIG_NFSV4 */ 1091 1092 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) 1093 { 1094 struct dentry *parent = desc->file->f_path.dentry; 1095 struct inode *dir = parent->d_inode; 1096 struct nfs_entry *entry = desc->entry; 1097 struct dentry *dentry, *alias; 1098 struct qstr name = { 1099 .name = entry->name, 1100 .len = entry->len, 1101 }; 1102 struct inode *inode; 1103 unsigned long verf = nfs_save_change_attribute(dir); 1104 1105 switch (name.len) { 1106 case 2: 1107 if (name.name[0] == '.' && name.name[1] == '.') 1108 return dget_parent(parent); 1109 break; 1110 case 1: 1111 if (name.name[0] == '.') 1112 return dget(parent); 1113 } 1114 1115 spin_lock(&dir->i_lock); 1116 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) { 1117 spin_unlock(&dir->i_lock); 1118 return NULL; 1119 } 1120 spin_unlock(&dir->i_lock); 1121 1122 name.hash = full_name_hash(name.name, name.len); 1123 dentry = d_lookup(parent, &name); 1124 if (dentry != NULL) { 1125 /* Is this a positive dentry that matches the readdir info? */ 1126 if (dentry->d_inode != NULL && 1127 (NFS_FILEID(dentry->d_inode) == entry->ino || 1128 d_mountpoint(dentry))) { 1129 if (!desc->plus || entry->fh->size == 0) 1130 return dentry; 1131 if (nfs_compare_fh(NFS_FH(dentry->d_inode), 1132 entry->fh) == 0) 1133 goto out_renew; 1134 } 1135 /* No, so d_drop to allow one to be created */ 1136 d_drop(dentry); 1137 dput(dentry); 1138 } 1139 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) 1140 return NULL; 1141 if (name.len > NFS_SERVER(dir)->namelen) 1142 return NULL; 1143 /* Note: caller is already holding the dir->i_mutex! */ 1144 dentry = d_alloc(parent, &name); 1145 if (dentry == NULL) 1146 return NULL; 1147 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1148 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 1149 if (IS_ERR(inode)) { 1150 dput(dentry); 1151 return NULL; 1152 } 1153 1154 alias = d_materialise_unique(dentry, inode); 1155 if (alias != NULL) { 1156 dput(dentry); 1157 if (IS_ERR(alias)) 1158 return NULL; 1159 dentry = alias; 1160 } 1161 1162 out_renew: 1163 nfs_set_verifier(dentry, verf); 1164 return dentry; 1165 } 1166 1167 /* 1168 * Code common to create, mkdir, and mknod. 1169 */ 1170 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1171 struct nfs_fattr *fattr) 1172 { 1173 struct dentry *parent = dget_parent(dentry); 1174 struct inode *dir = parent->d_inode; 1175 struct inode *inode; 1176 int error = -EACCES; 1177 1178 d_drop(dentry); 1179 1180 /* We may have been initialized further down */ 1181 if (dentry->d_inode) 1182 goto out; 1183 if (fhandle->size == 0) { 1184 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1185 if (error) 1186 goto out_error; 1187 } 1188 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1189 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1190 struct nfs_server *server = NFS_SB(dentry->d_sb); 1191 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1192 if (error < 0) 1193 goto out_error; 1194 } 1195 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1196 error = PTR_ERR(inode); 1197 if (IS_ERR(inode)) 1198 goto out_error; 1199 d_add(dentry, inode); 1200 out: 1201 dput(parent); 1202 return 0; 1203 out_error: 1204 nfs_mark_for_revalidate(dir); 1205 dput(parent); 1206 return error; 1207 } 1208 1209 /* 1210 * Following a failed create operation, we drop the dentry rather 1211 * than retain a negative dentry. This avoids a problem in the event 1212 * that the operation succeeded on the server, but an error in the 1213 * reply path made it appear to have failed. 1214 */ 1215 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, 1216 struct nameidata *nd) 1217 { 1218 struct iattr attr; 1219 int error; 1220 int open_flags = 0; 1221 1222 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1223 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1224 1225 attr.ia_mode = mode; 1226 attr.ia_valid = ATTR_MODE; 1227 1228 if ((nd->flags & LOOKUP_CREATE) != 0) 1229 open_flags = nd->intent.open.flags; 1230 1231 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); 1232 if (error != 0) 1233 goto out_err; 1234 return 0; 1235 out_err: 1236 d_drop(dentry); 1237 return error; 1238 } 1239 1240 /* 1241 * See comments for nfs_proc_create regarding failed operations. 1242 */ 1243 static int 1244 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) 1245 { 1246 struct iattr attr; 1247 int status; 1248 1249 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1250 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1251 1252 if (!new_valid_dev(rdev)) 1253 return -EINVAL; 1254 1255 attr.ia_mode = mode; 1256 attr.ia_valid = ATTR_MODE; 1257 1258 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1259 if (status != 0) 1260 goto out_err; 1261 return 0; 1262 out_err: 1263 d_drop(dentry); 1264 return status; 1265 } 1266 1267 /* 1268 * See comments for nfs_proc_create regarding failed operations. 1269 */ 1270 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1271 { 1272 struct iattr attr; 1273 int error; 1274 1275 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1276 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1277 1278 attr.ia_valid = ATTR_MODE; 1279 attr.ia_mode = mode | S_IFDIR; 1280 1281 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1282 if (error != 0) 1283 goto out_err; 1284 return 0; 1285 out_err: 1286 d_drop(dentry); 1287 return error; 1288 } 1289 1290 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1291 { 1292 if (dentry->d_inode != NULL && !d_unhashed(dentry)) 1293 d_delete(dentry); 1294 } 1295 1296 static int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1297 { 1298 int error; 1299 1300 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1301 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1302 1303 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1304 /* Ensure the VFS deletes this inode */ 1305 if (error == 0 && dentry->d_inode != NULL) 1306 clear_nlink(dentry->d_inode); 1307 else if (error == -ENOENT) 1308 nfs_dentry_handle_enoent(dentry); 1309 1310 return error; 1311 } 1312 1313 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) 1314 { 1315 static unsigned int sillycounter; 1316 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2; 1317 const int countersize = sizeof(sillycounter)*2; 1318 const int slen = sizeof(".nfs")+fileidsize+countersize-1; 1319 char silly[slen+1]; 1320 struct qstr qsilly; 1321 struct dentry *sdentry; 1322 int error = -EIO; 1323 1324 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", 1325 dentry->d_parent->d_name.name, dentry->d_name.name, 1326 atomic_read(&dentry->d_count)); 1327 nfs_inc_stats(dir, NFSIOS_SILLYRENAME); 1328 1329 /* 1330 * We don't allow a dentry to be silly-renamed twice. 1331 */ 1332 error = -EBUSY; 1333 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1334 goto out; 1335 1336 sprintf(silly, ".nfs%*.*Lx", 1337 fileidsize, fileidsize, 1338 (unsigned long long)NFS_FILEID(dentry->d_inode)); 1339 1340 /* Return delegation in anticipation of the rename */ 1341 nfs_inode_return_delegation(dentry->d_inode); 1342 1343 sdentry = NULL; 1344 do { 1345 char *suffix = silly + slen - countersize; 1346 1347 dput(sdentry); 1348 sillycounter++; 1349 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); 1350 1351 dfprintk(VFS, "NFS: trying to rename %s to %s\n", 1352 dentry->d_name.name, silly); 1353 1354 sdentry = lookup_one_len(silly, dentry->d_parent, slen); 1355 /* 1356 * N.B. Better to return EBUSY here ... it could be 1357 * dangerous to delete the file while it's in use. 1358 */ 1359 if (IS_ERR(sdentry)) 1360 goto out; 1361 } while(sdentry->d_inode != NULL); /* need negative lookup */ 1362 1363 qsilly.name = silly; 1364 qsilly.len = strlen(silly); 1365 if (dentry->d_inode) { 1366 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1367 dir, &qsilly); 1368 nfs_mark_for_revalidate(dentry->d_inode); 1369 } else 1370 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1371 dir, &qsilly); 1372 if (!error) { 1373 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1374 d_move(dentry, sdentry); 1375 error = nfs_async_unlink(dir, dentry); 1376 /* If we return 0 we don't unlink */ 1377 } 1378 dput(sdentry); 1379 out: 1380 return error; 1381 } 1382 1383 /* 1384 * Remove a file after making sure there are no pending writes, 1385 * and after checking that the file has only one user. 1386 * 1387 * We invalidate the attribute cache and free the inode prior to the operation 1388 * to avoid possible races if the server reuses the inode. 1389 */ 1390 static int nfs_safe_remove(struct dentry *dentry) 1391 { 1392 struct inode *dir = dentry->d_parent->d_inode; 1393 struct inode *inode = dentry->d_inode; 1394 int error = -EBUSY; 1395 1396 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1397 dentry->d_parent->d_name.name, dentry->d_name.name); 1398 1399 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1400 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1401 error = 0; 1402 goto out; 1403 } 1404 1405 if (inode != NULL) { 1406 nfs_inode_return_delegation(inode); 1407 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1408 /* The VFS may want to delete this inode */ 1409 if (error == 0) 1410 nfs_drop_nlink(inode); 1411 nfs_mark_for_revalidate(inode); 1412 } else 1413 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1414 if (error == -ENOENT) 1415 nfs_dentry_handle_enoent(dentry); 1416 out: 1417 return error; 1418 } 1419 1420 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1421 * belongs to an active ".nfs..." file and we return -EBUSY. 1422 * 1423 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1424 */ 1425 static int nfs_unlink(struct inode *dir, struct dentry *dentry) 1426 { 1427 int error; 1428 int need_rehash = 0; 1429 1430 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1431 dir->i_ino, dentry->d_name.name); 1432 1433 spin_lock(&dcache_lock); 1434 spin_lock(&dentry->d_lock); 1435 if (atomic_read(&dentry->d_count) > 1) { 1436 spin_unlock(&dentry->d_lock); 1437 spin_unlock(&dcache_lock); 1438 /* Start asynchronous writeout of the inode */ 1439 write_inode_now(dentry->d_inode, 0); 1440 error = nfs_sillyrename(dir, dentry); 1441 return error; 1442 } 1443 if (!d_unhashed(dentry)) { 1444 __d_drop(dentry); 1445 need_rehash = 1; 1446 } 1447 spin_unlock(&dentry->d_lock); 1448 spin_unlock(&dcache_lock); 1449 error = nfs_safe_remove(dentry); 1450 if (!error || error == -ENOENT) { 1451 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1452 } else if (need_rehash) 1453 d_rehash(dentry); 1454 return error; 1455 } 1456 1457 /* 1458 * To create a symbolic link, most file systems instantiate a new inode, 1459 * add a page to it containing the path, then write it out to the disk 1460 * using prepare_write/commit_write. 1461 * 1462 * Unfortunately the NFS client can't create the in-core inode first 1463 * because it needs a file handle to create an in-core inode (see 1464 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1465 * symlink request has completed on the server. 1466 * 1467 * So instead we allocate a raw page, copy the symname into it, then do 1468 * the SYMLINK request with the page as the buffer. If it succeeds, we 1469 * now have a new file handle and can instantiate an in-core NFS inode 1470 * and move the raw page into its mapping. 1471 */ 1472 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1473 { 1474 struct pagevec lru_pvec; 1475 struct page *page; 1476 char *kaddr; 1477 struct iattr attr; 1478 unsigned int pathlen = strlen(symname); 1479 int error; 1480 1481 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1482 dir->i_ino, dentry->d_name.name, symname); 1483 1484 if (pathlen > PAGE_SIZE) 1485 return -ENAMETOOLONG; 1486 1487 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1488 attr.ia_valid = ATTR_MODE; 1489 1490 page = alloc_page(GFP_HIGHUSER); 1491 if (!page) 1492 return -ENOMEM; 1493 1494 kaddr = kmap_atomic(page, KM_USER0); 1495 memcpy(kaddr, symname, pathlen); 1496 if (pathlen < PAGE_SIZE) 1497 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1498 kunmap_atomic(kaddr, KM_USER0); 1499 1500 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1501 if (error != 0) { 1502 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1503 dir->i_sb->s_id, dir->i_ino, 1504 dentry->d_name.name, symname, error); 1505 d_drop(dentry); 1506 __free_page(page); 1507 return error; 1508 } 1509 1510 /* 1511 * No big deal if we can't add this page to the page cache here. 1512 * READLINK will get the missing page from the server if needed. 1513 */ 1514 pagevec_init(&lru_pvec, 0); 1515 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1516 GFP_KERNEL)) { 1517 pagevec_add(&lru_pvec, page); 1518 pagevec_lru_add_file(&lru_pvec); 1519 SetPageUptodate(page); 1520 unlock_page(page); 1521 } else 1522 __free_page(page); 1523 1524 return 0; 1525 } 1526 1527 static int 1528 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1529 { 1530 struct inode *inode = old_dentry->d_inode; 1531 int error; 1532 1533 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1534 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1535 dentry->d_parent->d_name.name, dentry->d_name.name); 1536 1537 d_drop(dentry); 1538 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1539 if (error == 0) { 1540 atomic_inc(&inode->i_count); 1541 d_add(dentry, inode); 1542 } 1543 return error; 1544 } 1545 1546 /* 1547 * RENAME 1548 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1549 * different file handle for the same inode after a rename (e.g. when 1550 * moving to a different directory). A fail-safe method to do so would 1551 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1552 * rename the old file using the sillyrename stuff. This way, the original 1553 * file in old_dir will go away when the last process iput()s the inode. 1554 * 1555 * FIXED. 1556 * 1557 * It actually works quite well. One needs to have the possibility for 1558 * at least one ".nfs..." file in each directory the file ever gets 1559 * moved or linked to which happens automagically with the new 1560 * implementation that only depends on the dcache stuff instead of 1561 * using the inode layer 1562 * 1563 * Unfortunately, things are a little more complicated than indicated 1564 * above. For a cross-directory move, we want to make sure we can get 1565 * rid of the old inode after the operation. This means there must be 1566 * no pending writes (if it's a file), and the use count must be 1. 1567 * If these conditions are met, we can drop the dentries before doing 1568 * the rename. 1569 */ 1570 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1571 struct inode *new_dir, struct dentry *new_dentry) 1572 { 1573 struct inode *old_inode = old_dentry->d_inode; 1574 struct inode *new_inode = new_dentry->d_inode; 1575 struct dentry *dentry = NULL, *rehash = NULL; 1576 int error = -EBUSY; 1577 1578 /* 1579 * To prevent any new references to the target during the rename, 1580 * we unhash the dentry and free the inode in advance. 1581 */ 1582 if (!d_unhashed(new_dentry)) { 1583 d_drop(new_dentry); 1584 rehash = new_dentry; 1585 } 1586 1587 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1588 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1589 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1590 atomic_read(&new_dentry->d_count)); 1591 1592 /* 1593 * First check whether the target is busy ... we can't 1594 * safely do _any_ rename if the target is in use. 1595 * 1596 * For files, make a copy of the dentry and then do a 1597 * silly-rename. If the silly-rename succeeds, the 1598 * copied dentry is hashed and becomes the new target. 1599 */ 1600 if (!new_inode) 1601 goto go_ahead; 1602 if (S_ISDIR(new_inode->i_mode)) { 1603 error = -EISDIR; 1604 if (!S_ISDIR(old_inode->i_mode)) 1605 goto out; 1606 } else if (atomic_read(&new_dentry->d_count) > 2) { 1607 int err; 1608 /* copy the target dentry's name */ 1609 dentry = d_alloc(new_dentry->d_parent, 1610 &new_dentry->d_name); 1611 if (!dentry) 1612 goto out; 1613 1614 /* silly-rename the existing target ... */ 1615 err = nfs_sillyrename(new_dir, new_dentry); 1616 if (!err) { 1617 new_dentry = rehash = dentry; 1618 new_inode = NULL; 1619 /* instantiate the replacement target */ 1620 d_instantiate(new_dentry, NULL); 1621 } else if (atomic_read(&new_dentry->d_count) > 1) 1622 /* dentry still busy? */ 1623 goto out; 1624 } else 1625 nfs_drop_nlink(new_inode); 1626 1627 go_ahead: 1628 /* 1629 * ... prune child dentries and writebacks if needed. 1630 */ 1631 if (atomic_read(&old_dentry->d_count) > 1) { 1632 if (S_ISREG(old_inode->i_mode)) 1633 nfs_wb_all(old_inode); 1634 shrink_dcache_parent(old_dentry); 1635 } 1636 nfs_inode_return_delegation(old_inode); 1637 1638 if (new_inode != NULL) { 1639 nfs_inode_return_delegation(new_inode); 1640 d_delete(new_dentry); 1641 } 1642 1643 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1644 new_dir, &new_dentry->d_name); 1645 nfs_mark_for_revalidate(old_inode); 1646 out: 1647 if (rehash) 1648 d_rehash(rehash); 1649 if (!error) { 1650 d_move(old_dentry, new_dentry); 1651 nfs_set_verifier(new_dentry, 1652 nfs_save_change_attribute(new_dir)); 1653 } else if (error == -ENOENT) 1654 nfs_dentry_handle_enoent(old_dentry); 1655 1656 /* new dentry created? */ 1657 if (dentry) 1658 dput(dentry); 1659 return error; 1660 } 1661 1662 static DEFINE_SPINLOCK(nfs_access_lru_lock); 1663 static LIST_HEAD(nfs_access_lru_list); 1664 static atomic_long_t nfs_access_nr_entries; 1665 1666 static void nfs_access_free_entry(struct nfs_access_entry *entry) 1667 { 1668 put_rpccred(entry->cred); 1669 kfree(entry); 1670 smp_mb__before_atomic_dec(); 1671 atomic_long_dec(&nfs_access_nr_entries); 1672 smp_mb__after_atomic_dec(); 1673 } 1674 1675 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) 1676 { 1677 LIST_HEAD(head); 1678 struct nfs_inode *nfsi; 1679 struct nfs_access_entry *cache; 1680 1681 restart: 1682 spin_lock(&nfs_access_lru_lock); 1683 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { 1684 struct rw_semaphore *s_umount; 1685 struct inode *inode; 1686 1687 if (nr_to_scan-- == 0) 1688 break; 1689 s_umount = &nfsi->vfs_inode.i_sb->s_umount; 1690 if (!down_read_trylock(s_umount)) 1691 continue; 1692 inode = igrab(&nfsi->vfs_inode); 1693 if (inode == NULL) { 1694 up_read(s_umount); 1695 continue; 1696 } 1697 spin_lock(&inode->i_lock); 1698 if (list_empty(&nfsi->access_cache_entry_lru)) 1699 goto remove_lru_entry; 1700 cache = list_entry(nfsi->access_cache_entry_lru.next, 1701 struct nfs_access_entry, lru); 1702 list_move(&cache->lru, &head); 1703 rb_erase(&cache->rb_node, &nfsi->access_cache); 1704 if (!list_empty(&nfsi->access_cache_entry_lru)) 1705 list_move_tail(&nfsi->access_cache_inode_lru, 1706 &nfs_access_lru_list); 1707 else { 1708 remove_lru_entry: 1709 list_del_init(&nfsi->access_cache_inode_lru); 1710 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1711 } 1712 spin_unlock(&inode->i_lock); 1713 spin_unlock(&nfs_access_lru_lock); 1714 iput(inode); 1715 up_read(s_umount); 1716 goto restart; 1717 } 1718 spin_unlock(&nfs_access_lru_lock); 1719 while (!list_empty(&head)) { 1720 cache = list_entry(head.next, struct nfs_access_entry, lru); 1721 list_del(&cache->lru); 1722 nfs_access_free_entry(cache); 1723 } 1724 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; 1725 } 1726 1727 static void __nfs_access_zap_cache(struct inode *inode) 1728 { 1729 struct nfs_inode *nfsi = NFS_I(inode); 1730 struct rb_root *root_node = &nfsi->access_cache; 1731 struct rb_node *n, *dispose = NULL; 1732 struct nfs_access_entry *entry; 1733 1734 /* Unhook entries from the cache */ 1735 while ((n = rb_first(root_node)) != NULL) { 1736 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1737 rb_erase(n, root_node); 1738 list_del(&entry->lru); 1739 n->rb_left = dispose; 1740 dispose = n; 1741 } 1742 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 1743 spin_unlock(&inode->i_lock); 1744 1745 /* Now kill them all! */ 1746 while (dispose != NULL) { 1747 n = dispose; 1748 dispose = n->rb_left; 1749 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); 1750 } 1751 } 1752 1753 void nfs_access_zap_cache(struct inode *inode) 1754 { 1755 /* Remove from global LRU init */ 1756 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 1757 spin_lock(&nfs_access_lru_lock); 1758 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 1759 spin_unlock(&nfs_access_lru_lock); 1760 } 1761 1762 spin_lock(&inode->i_lock); 1763 /* This will release the spinlock */ 1764 __nfs_access_zap_cache(inode); 1765 } 1766 1767 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 1768 { 1769 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 1770 struct nfs_access_entry *entry; 1771 1772 while (n != NULL) { 1773 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1774 1775 if (cred < entry->cred) 1776 n = n->rb_left; 1777 else if (cred > entry->cred) 1778 n = n->rb_right; 1779 else 1780 return entry; 1781 } 1782 return NULL; 1783 } 1784 1785 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 1786 { 1787 struct nfs_inode *nfsi = NFS_I(inode); 1788 struct nfs_access_entry *cache; 1789 int err = -ENOENT; 1790 1791 spin_lock(&inode->i_lock); 1792 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 1793 goto out_zap; 1794 cache = nfs_access_search_rbtree(inode, cred); 1795 if (cache == NULL) 1796 goto out; 1797 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 1798 goto out_stale; 1799 res->jiffies = cache->jiffies; 1800 res->cred = cache->cred; 1801 res->mask = cache->mask; 1802 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 1803 err = 0; 1804 out: 1805 spin_unlock(&inode->i_lock); 1806 return err; 1807 out_stale: 1808 rb_erase(&cache->rb_node, &nfsi->access_cache); 1809 list_del(&cache->lru); 1810 spin_unlock(&inode->i_lock); 1811 nfs_access_free_entry(cache); 1812 return -ENOENT; 1813 out_zap: 1814 /* This will release the spinlock */ 1815 __nfs_access_zap_cache(inode); 1816 return -ENOENT; 1817 } 1818 1819 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 1820 { 1821 struct nfs_inode *nfsi = NFS_I(inode); 1822 struct rb_root *root_node = &nfsi->access_cache; 1823 struct rb_node **p = &root_node->rb_node; 1824 struct rb_node *parent = NULL; 1825 struct nfs_access_entry *entry; 1826 1827 spin_lock(&inode->i_lock); 1828 while (*p != NULL) { 1829 parent = *p; 1830 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 1831 1832 if (set->cred < entry->cred) 1833 p = &parent->rb_left; 1834 else if (set->cred > entry->cred) 1835 p = &parent->rb_right; 1836 else 1837 goto found; 1838 } 1839 rb_link_node(&set->rb_node, parent, p); 1840 rb_insert_color(&set->rb_node, root_node); 1841 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1842 spin_unlock(&inode->i_lock); 1843 return; 1844 found: 1845 rb_replace_node(parent, &set->rb_node, root_node); 1846 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1847 list_del(&entry->lru); 1848 spin_unlock(&inode->i_lock); 1849 nfs_access_free_entry(entry); 1850 } 1851 1852 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 1853 { 1854 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 1855 if (cache == NULL) 1856 return; 1857 RB_CLEAR_NODE(&cache->rb_node); 1858 cache->jiffies = set->jiffies; 1859 cache->cred = get_rpccred(set->cred); 1860 cache->mask = set->mask; 1861 1862 nfs_access_add_rbtree(inode, cache); 1863 1864 /* Update accounting */ 1865 smp_mb__before_atomic_inc(); 1866 atomic_long_inc(&nfs_access_nr_entries); 1867 smp_mb__after_atomic_inc(); 1868 1869 /* Add inode to global LRU list */ 1870 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 1871 spin_lock(&nfs_access_lru_lock); 1872 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); 1873 spin_unlock(&nfs_access_lru_lock); 1874 } 1875 } 1876 1877 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 1878 { 1879 struct nfs_access_entry cache; 1880 int status; 1881 1882 status = nfs_access_get_cached(inode, cred, &cache); 1883 if (status == 0) 1884 goto out; 1885 1886 /* Be clever: ask server to check for all possible rights */ 1887 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 1888 cache.cred = cred; 1889 cache.jiffies = jiffies; 1890 status = NFS_PROTO(inode)->access(inode, &cache); 1891 if (status != 0) 1892 return status; 1893 nfs_access_add_cache(inode, &cache); 1894 out: 1895 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 1896 return 0; 1897 return -EACCES; 1898 } 1899 1900 static int nfs_open_permission_mask(int openflags) 1901 { 1902 int mask = 0; 1903 1904 if (openflags & FMODE_READ) 1905 mask |= MAY_READ; 1906 if (openflags & FMODE_WRITE) 1907 mask |= MAY_WRITE; 1908 if (openflags & FMODE_EXEC) 1909 mask |= MAY_EXEC; 1910 return mask; 1911 } 1912 1913 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 1914 { 1915 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 1916 } 1917 1918 int nfs_permission(struct inode *inode, int mask) 1919 { 1920 struct rpc_cred *cred; 1921 int res = 0; 1922 1923 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 1924 1925 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 1926 goto out; 1927 /* Is this sys_access() ? */ 1928 if (mask & MAY_ACCESS) 1929 goto force_lookup; 1930 1931 switch (inode->i_mode & S_IFMT) { 1932 case S_IFLNK: 1933 goto out; 1934 case S_IFREG: 1935 /* NFSv4 has atomic_open... */ 1936 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) 1937 && (mask & MAY_OPEN)) 1938 goto out; 1939 break; 1940 case S_IFDIR: 1941 /* 1942 * Optimize away all write operations, since the server 1943 * will check permissions when we perform the op. 1944 */ 1945 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 1946 goto out; 1947 } 1948 1949 force_lookup: 1950 if (!NFS_PROTO(inode)->access) 1951 goto out_notsup; 1952 1953 cred = rpc_lookup_cred(); 1954 if (!IS_ERR(cred)) { 1955 res = nfs_do_access(inode, cred, mask); 1956 put_rpccred(cred); 1957 } else 1958 res = PTR_ERR(cred); 1959 out: 1960 if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) 1961 res = -EACCES; 1962 1963 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 1964 inode->i_sb->s_id, inode->i_ino, mask, res); 1965 return res; 1966 out_notsup: 1967 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1968 if (res == 0) 1969 res = generic_permission(inode, mask, NULL); 1970 goto out; 1971 } 1972 1973 /* 1974 * Local variables: 1975 * version-control: t 1976 * kept-new-versions: 5 1977 * End: 1978 */ 1979