1 /* 2 * linux/fs/nfs/dir.c 3 * 4 * Copyright (C) 1992 Rick Sladkey 5 * 6 * nfs directory handling functions 7 * 8 * 10 Apr 1996 Added silly rename for unlink --okir 9 * 28 Sep 1996 Improved directory cache --okir 10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de 11 * Re-implemented silly rename for unlink, newly implemented 12 * silly rename for nfs_rename() following the suggestions 13 * of Olaf Kirch (okir) found in this file. 14 * Following Linus comments on my original hack, this version 15 * depends only on the dcache stuff and doesn't touch the inode 16 * layer (iput() and friends). 17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM 18 */ 19 20 #include <linux/time.h> 21 #include <linux/errno.h> 22 #include <linux/stat.h> 23 #include <linux/fcntl.h> 24 #include <linux/string.h> 25 #include <linux/kernel.h> 26 #include <linux/slab.h> 27 #include <linux/mm.h> 28 #include <linux/sunrpc/clnt.h> 29 #include <linux/nfs_fs.h> 30 #include <linux/nfs_mount.h> 31 #include <linux/pagemap.h> 32 #include <linux/smp_lock.h> 33 #include <linux/pagevec.h> 34 #include <linux/namei.h> 35 #include <linux/mount.h> 36 #include <linux/sched.h> 37 38 #include "nfs4_fs.h" 39 #include "delegation.h" 40 #include "iostat.h" 41 #include "internal.h" 42 43 /* #define NFS_DEBUG_VERBOSE 1 */ 44 45 static int nfs_opendir(struct inode *, struct file *); 46 static int nfs_readdir(struct file *, void *, filldir_t); 47 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); 48 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); 49 static int nfs_mkdir(struct inode *, struct dentry *, int); 50 static int nfs_rmdir(struct inode *, struct dentry *); 51 static int nfs_unlink(struct inode *, struct dentry *); 52 static int nfs_symlink(struct inode *, struct dentry *, const char *); 53 static int nfs_link(struct dentry *, struct inode *, struct dentry *); 54 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); 55 static int nfs_rename(struct inode *, struct dentry *, 56 struct inode *, struct dentry *); 57 static int nfs_fsync_dir(struct file *, struct dentry *, int); 58 static loff_t nfs_llseek_dir(struct file *, loff_t, int); 59 60 const struct file_operations nfs_dir_operations = { 61 .llseek = nfs_llseek_dir, 62 .read = generic_read_dir, 63 .readdir = nfs_readdir, 64 .open = nfs_opendir, 65 .release = nfs_release, 66 .fsync = nfs_fsync_dir, 67 }; 68 69 const struct inode_operations nfs_dir_inode_operations = { 70 .create = nfs_create, 71 .lookup = nfs_lookup, 72 .link = nfs_link, 73 .unlink = nfs_unlink, 74 .symlink = nfs_symlink, 75 .mkdir = nfs_mkdir, 76 .rmdir = nfs_rmdir, 77 .mknod = nfs_mknod, 78 .rename = nfs_rename, 79 .permission = nfs_permission, 80 .getattr = nfs_getattr, 81 .setattr = nfs_setattr, 82 }; 83 84 #ifdef CONFIG_NFS_V3 85 const struct inode_operations nfs3_dir_inode_operations = { 86 .create = nfs_create, 87 .lookup = nfs_lookup, 88 .link = nfs_link, 89 .unlink = nfs_unlink, 90 .symlink = nfs_symlink, 91 .mkdir = nfs_mkdir, 92 .rmdir = nfs_rmdir, 93 .mknod = nfs_mknod, 94 .rename = nfs_rename, 95 .permission = nfs_permission, 96 .getattr = nfs_getattr, 97 .setattr = nfs_setattr, 98 .listxattr = nfs3_listxattr, 99 .getxattr = nfs3_getxattr, 100 .setxattr = nfs3_setxattr, 101 .removexattr = nfs3_removexattr, 102 }; 103 #endif /* CONFIG_NFS_V3 */ 104 105 #ifdef CONFIG_NFS_V4 106 107 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); 108 const struct inode_operations nfs4_dir_inode_operations = { 109 .create = nfs_create, 110 .lookup = nfs_atomic_lookup, 111 .link = nfs_link, 112 .unlink = nfs_unlink, 113 .symlink = nfs_symlink, 114 .mkdir = nfs_mkdir, 115 .rmdir = nfs_rmdir, 116 .mknod = nfs_mknod, 117 .rename = nfs_rename, 118 .permission = nfs_permission, 119 .getattr = nfs_getattr, 120 .setattr = nfs_setattr, 121 .getxattr = nfs4_getxattr, 122 .setxattr = nfs4_setxattr, 123 .listxattr = nfs4_listxattr, 124 }; 125 126 #endif /* CONFIG_NFS_V4 */ 127 128 /* 129 * Open file 130 */ 131 static int 132 nfs_opendir(struct inode *inode, struct file *filp) 133 { 134 int res; 135 136 dfprintk(VFS, "NFS: opendir(%s/%ld)\n", 137 inode->i_sb->s_id, inode->i_ino); 138 139 lock_kernel(); 140 /* Call generic open code in order to cache credentials */ 141 res = nfs_open(inode, filp); 142 unlock_kernel(); 143 return res; 144 } 145 146 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int); 147 typedef struct { 148 struct file *file; 149 struct page *page; 150 unsigned long page_index; 151 __be32 *ptr; 152 u64 *dir_cookie; 153 loff_t current_index; 154 struct nfs_entry *entry; 155 decode_dirent_t decode; 156 int plus; 157 int error; 158 unsigned long timestamp; 159 int timestamp_valid; 160 } nfs_readdir_descriptor_t; 161 162 /* Now we cache directories properly, by stuffing the dirent 163 * data directly in the page cache. 164 * 165 * Inode invalidation due to refresh etc. takes care of 166 * _everything_, no sloppy entry flushing logic, no extraneous 167 * copying, network direct to page cache, the way it was meant 168 * to be. 169 * 170 * NOTE: Dirent information verification is done always by the 171 * page-in of the RPC reply, nowhere else, this simplies 172 * things substantially. 173 */ 174 static 175 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page) 176 { 177 struct file *file = desc->file; 178 struct inode *inode = file->f_path.dentry->d_inode; 179 struct rpc_cred *cred = nfs_file_cred(file); 180 unsigned long timestamp; 181 int error; 182 183 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n", 184 __FUNCTION__, (long long)desc->entry->cookie, 185 page->index); 186 187 again: 188 timestamp = jiffies; 189 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page, 190 NFS_SERVER(inode)->dtsize, desc->plus); 191 if (error < 0) { 192 /* We requested READDIRPLUS, but the server doesn't grok it */ 193 if (error == -ENOTSUPP && desc->plus) { 194 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 195 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 196 desc->plus = 0; 197 goto again; 198 } 199 goto error; 200 } 201 desc->timestamp = timestamp; 202 desc->timestamp_valid = 1; 203 SetPageUptodate(page); 204 /* Ensure consistent page alignment of the data. 205 * Note: assumes we have exclusive access to this mapping either 206 * through inode->i_mutex or some other mechanism. 207 */ 208 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) { 209 /* Should never happen */ 210 nfs_zap_mapping(inode, inode->i_mapping); 211 } 212 unlock_page(page); 213 return 0; 214 error: 215 unlock_page(page); 216 desc->error = error; 217 return -EIO; 218 } 219 220 static inline 221 int dir_decode(nfs_readdir_descriptor_t *desc) 222 { 223 __be32 *p = desc->ptr; 224 p = desc->decode(p, desc->entry, desc->plus); 225 if (IS_ERR(p)) 226 return PTR_ERR(p); 227 desc->ptr = p; 228 if (desc->timestamp_valid) 229 desc->entry->fattr->time_start = desc->timestamp; 230 else 231 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR; 232 return 0; 233 } 234 235 static inline 236 void dir_page_release(nfs_readdir_descriptor_t *desc) 237 { 238 kunmap(desc->page); 239 page_cache_release(desc->page); 240 desc->page = NULL; 241 desc->ptr = NULL; 242 } 243 244 /* 245 * Given a pointer to a buffer that has already been filled by a call 246 * to readdir, find the next entry with cookie '*desc->dir_cookie'. 247 * 248 * If the end of the buffer has been reached, return -EAGAIN, if not, 249 * return the offset within the buffer of the next entry to be 250 * read. 251 */ 252 static inline 253 int find_dirent(nfs_readdir_descriptor_t *desc) 254 { 255 struct nfs_entry *entry = desc->entry; 256 int loop_count = 0, 257 status; 258 259 while((status = dir_decode(desc)) == 0) { 260 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n", 261 __FUNCTION__, (unsigned long long)entry->cookie); 262 if (entry->prev_cookie == *desc->dir_cookie) 263 break; 264 if (loop_count++ > 200) { 265 loop_count = 0; 266 schedule(); 267 } 268 } 269 return status; 270 } 271 272 /* 273 * Given a pointer to a buffer that has already been filled by a call 274 * to readdir, find the entry at offset 'desc->file->f_pos'. 275 * 276 * If the end of the buffer has been reached, return -EAGAIN, if not, 277 * return the offset within the buffer of the next entry to be 278 * read. 279 */ 280 static inline 281 int find_dirent_index(nfs_readdir_descriptor_t *desc) 282 { 283 struct nfs_entry *entry = desc->entry; 284 int loop_count = 0, 285 status; 286 287 for(;;) { 288 status = dir_decode(desc); 289 if (status) 290 break; 291 292 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n", 293 (unsigned long long)entry->cookie, desc->current_index); 294 295 if (desc->file->f_pos == desc->current_index) { 296 *desc->dir_cookie = entry->cookie; 297 break; 298 } 299 desc->current_index++; 300 if (loop_count++ > 200) { 301 loop_count = 0; 302 schedule(); 303 } 304 } 305 return status; 306 } 307 308 /* 309 * Find the given page, and call find_dirent() or find_dirent_index in 310 * order to try to return the next entry. 311 */ 312 static inline 313 int find_dirent_page(nfs_readdir_descriptor_t *desc) 314 { 315 struct inode *inode = desc->file->f_path.dentry->d_inode; 316 struct page *page; 317 int status; 318 319 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n", 320 __FUNCTION__, desc->page_index, 321 (long long) *desc->dir_cookie); 322 323 /* If we find the page in the page_cache, we cannot be sure 324 * how fresh the data is, so we will ignore readdir_plus attributes. 325 */ 326 desc->timestamp_valid = 0; 327 page = read_cache_page(inode->i_mapping, desc->page_index, 328 (filler_t *)nfs_readdir_filler, desc); 329 if (IS_ERR(page)) { 330 status = PTR_ERR(page); 331 goto out; 332 } 333 334 /* NOTE: Someone else may have changed the READDIRPLUS flag */ 335 desc->page = page; 336 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 337 if (*desc->dir_cookie != 0) 338 status = find_dirent(desc); 339 else 340 status = find_dirent_index(desc); 341 if (status < 0) 342 dir_page_release(desc); 343 out: 344 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status); 345 return status; 346 } 347 348 /* 349 * Recurse through the page cache pages, and return a 350 * filled nfs_entry structure of the next directory entry if possible. 351 * 352 * The target for the search is '*desc->dir_cookie' if non-0, 353 * 'desc->file->f_pos' otherwise 354 */ 355 static inline 356 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 357 { 358 int loop_count = 0; 359 int res; 360 361 /* Always search-by-index from the beginning of the cache */ 362 if (*desc->dir_cookie == 0) { 363 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n", 364 (long long)desc->file->f_pos); 365 desc->page_index = 0; 366 desc->entry->cookie = desc->entry->prev_cookie = 0; 367 desc->entry->eof = 0; 368 desc->current_index = 0; 369 } else 370 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", 371 (unsigned long long)*desc->dir_cookie); 372 373 for (;;) { 374 res = find_dirent_page(desc); 375 if (res != -EAGAIN) 376 break; 377 /* Align to beginning of next page */ 378 desc->page_index ++; 379 if (loop_count++ > 200) { 380 loop_count = 0; 381 schedule(); 382 } 383 } 384 385 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res); 386 return res; 387 } 388 389 static inline unsigned int dt_type(struct inode *inode) 390 { 391 return (inode->i_mode >> 12) & 15; 392 } 393 394 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc); 395 396 /* 397 * Once we've found the start of the dirent within a page: fill 'er up... 398 */ 399 static 400 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, 401 filldir_t filldir) 402 { 403 struct file *file = desc->file; 404 struct nfs_entry *entry = desc->entry; 405 struct dentry *dentry = NULL; 406 u64 fileid; 407 int loop_count = 0, 408 res; 409 410 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", 411 (unsigned long long)entry->cookie); 412 413 for(;;) { 414 unsigned d_type = DT_UNKNOWN; 415 /* Note: entry->prev_cookie contains the cookie for 416 * retrieving the current dirent on the server */ 417 fileid = entry->ino; 418 419 /* Get a dentry if we have one */ 420 if (dentry != NULL) 421 dput(dentry); 422 dentry = nfs_readdir_lookup(desc); 423 424 /* Use readdirplus info */ 425 if (dentry != NULL && dentry->d_inode != NULL) { 426 d_type = dt_type(dentry->d_inode); 427 fileid = NFS_FILEID(dentry->d_inode); 428 } 429 430 res = filldir(dirent, entry->name, entry->len, 431 file->f_pos, nfs_compat_user_ino64(fileid), 432 d_type); 433 if (res < 0) 434 break; 435 file->f_pos++; 436 *desc->dir_cookie = entry->cookie; 437 if (dir_decode(desc) != 0) { 438 desc->page_index ++; 439 break; 440 } 441 if (loop_count++ > 200) { 442 loop_count = 0; 443 schedule(); 444 } 445 } 446 dir_page_release(desc); 447 if (dentry != NULL) 448 dput(dentry); 449 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 450 (unsigned long long)*desc->dir_cookie, res); 451 return res; 452 } 453 454 /* 455 * If we cannot find a cookie in our cache, we suspect that this is 456 * because it points to a deleted file, so we ask the server to return 457 * whatever it thinks is the next entry. We then feed this to filldir. 458 * If all goes well, we should then be able to find our way round the 459 * cache on the next call to readdir_search_pagecache(); 460 * 461 * NOTE: we cannot add the anonymous page to the pagecache because 462 * the data it contains might not be page aligned. Besides, 463 * we should already have a complete representation of the 464 * directory in the page cache by the time we get here. 465 */ 466 static inline 467 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, 468 filldir_t filldir) 469 { 470 struct file *file = desc->file; 471 struct inode *inode = file->f_path.dentry->d_inode; 472 struct rpc_cred *cred = nfs_file_cred(file); 473 struct page *page = NULL; 474 int status; 475 unsigned long timestamp; 476 477 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 478 (unsigned long long)*desc->dir_cookie); 479 480 page = alloc_page(GFP_HIGHUSER); 481 if (!page) { 482 status = -ENOMEM; 483 goto out; 484 } 485 timestamp = jiffies; 486 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie, 487 page, 488 NFS_SERVER(inode)->dtsize, 489 desc->plus); 490 desc->page = page; 491 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 492 if (desc->error >= 0) { 493 desc->timestamp = timestamp; 494 desc->timestamp_valid = 1; 495 if ((status = dir_decode(desc)) == 0) 496 desc->entry->prev_cookie = *desc->dir_cookie; 497 } else 498 status = -EIO; 499 if (status < 0) 500 goto out_release; 501 502 status = nfs_do_filldir(desc, dirent, filldir); 503 504 /* Reset read descriptor so it searches the page cache from 505 * the start upon the next call to readdir_search_pagecache() */ 506 desc->page_index = 0; 507 desc->entry->cookie = desc->entry->prev_cookie = 0; 508 desc->entry->eof = 0; 509 out: 510 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 511 __FUNCTION__, status); 512 return status; 513 out_release: 514 dir_page_release(desc); 515 goto out; 516 } 517 518 /* The file offset position represents the dirent entry number. A 519 last cookie cache takes care of the common case of reading the 520 whole directory. 521 */ 522 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) 523 { 524 struct dentry *dentry = filp->f_path.dentry; 525 struct inode *inode = dentry->d_inode; 526 nfs_readdir_descriptor_t my_desc, 527 *desc = &my_desc; 528 struct nfs_entry my_entry; 529 struct nfs_fh fh; 530 struct nfs_fattr fattr; 531 long res; 532 533 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n", 534 dentry->d_parent->d_name.name, dentry->d_name.name, 535 (long long)filp->f_pos); 536 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 537 538 lock_kernel(); 539 540 /* 541 * filp->f_pos points to the dirent entry number. 542 * *desc->dir_cookie has the cookie for the next entry. We have 543 * to either find the entry with the appropriate number or 544 * revalidate the cookie. 545 */ 546 memset(desc, 0, sizeof(*desc)); 547 548 desc->file = filp; 549 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; 550 desc->decode = NFS_PROTO(inode)->decode_dirent; 551 desc->plus = NFS_USE_READDIRPLUS(inode); 552 553 my_entry.cookie = my_entry.prev_cookie = 0; 554 my_entry.eof = 0; 555 my_entry.fh = &fh; 556 my_entry.fattr = &fattr; 557 nfs_fattr_init(&fattr); 558 desc->entry = &my_entry; 559 560 nfs_block_sillyrename(dentry); 561 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping); 562 if (res < 0) 563 goto out; 564 565 while(!desc->entry->eof) { 566 res = readdir_search_pagecache(desc); 567 568 if (res == -EBADCOOKIE) { 569 /* This means either end of directory */ 570 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { 571 /* Or that the server has 'lost' a cookie */ 572 res = uncached_readdir(desc, dirent, filldir); 573 if (res >= 0) 574 continue; 575 } 576 res = 0; 577 break; 578 } 579 if (res == -ETOOSMALL && desc->plus) { 580 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 581 nfs_zap_caches(inode); 582 desc->plus = 0; 583 desc->entry->eof = 0; 584 continue; 585 } 586 if (res < 0) 587 break; 588 589 res = nfs_do_filldir(desc, dirent, filldir); 590 if (res < 0) { 591 res = 0; 592 break; 593 } 594 } 595 out: 596 nfs_unblock_sillyrename(dentry); 597 unlock_kernel(); 598 if (res > 0) 599 res = 0; 600 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n", 601 dentry->d_parent->d_name.name, dentry->d_name.name, 602 res); 603 return res; 604 } 605 606 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) 607 { 608 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex); 609 switch (origin) { 610 case 1: 611 offset += filp->f_pos; 612 case 0: 613 if (offset >= 0) 614 break; 615 default: 616 offset = -EINVAL; 617 goto out; 618 } 619 if (offset != filp->f_pos) { 620 filp->f_pos = offset; 621 nfs_file_open_context(filp)->dir_cookie = 0; 622 } 623 out: 624 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex); 625 return offset; 626 } 627 628 /* 629 * All directory operations under NFS are synchronous, so fsync() 630 * is a dummy operation. 631 */ 632 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) 633 { 634 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n", 635 dentry->d_parent->d_name.name, dentry->d_name.name, 636 datasync); 637 638 return 0; 639 } 640 641 /* 642 * A check for whether or not the parent directory has changed. 643 * In the case it has, we assume that the dentries are untrustworthy 644 * and may need to be looked up again. 645 */ 646 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) 647 { 648 if (IS_ROOT(dentry)) 649 return 1; 650 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 651 return 0; 652 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 653 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 654 return 0; 655 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 656 return 0; 657 return 1; 658 } 659 660 /* 661 * Return the intent data that applies to this particular path component 662 * 663 * Note that the current set of intents only apply to the very last 664 * component of the path. 665 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. 666 */ 667 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) 668 { 669 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) 670 return 0; 671 return nd->flags & mask; 672 } 673 674 /* 675 * Use intent information to check whether or not we're going to do 676 * an O_EXCL create using this path component. 677 */ 678 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) 679 { 680 if (NFS_PROTO(dir)->version == 2) 681 return 0; 682 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0) 683 return 0; 684 return (nd->intent.open.flags & O_EXCL) != 0; 685 } 686 687 /* 688 * Inode and filehandle revalidation for lookups. 689 * 690 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 691 * or if the intent information indicates that we're about to open this 692 * particular file and the "nocto" mount flag is not set. 693 * 694 */ 695 static inline 696 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) 697 { 698 struct nfs_server *server = NFS_SERVER(inode); 699 700 if (nd != NULL) { 701 /* VFS wants an on-the-wire revalidation */ 702 if (nd->flags & LOOKUP_REVAL) 703 goto out_force; 704 /* This is an open(2) */ 705 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && 706 !(server->flags & NFS_MOUNT_NOCTO) && 707 (S_ISREG(inode->i_mode) || 708 S_ISDIR(inode->i_mode))) 709 goto out_force; 710 return 0; 711 } 712 return nfs_revalidate_inode(server, inode); 713 out_force: 714 return __nfs_revalidate_inode(server, inode); 715 } 716 717 /* 718 * We judge how long we want to trust negative 719 * dentries by looking at the parent inode mtime. 720 * 721 * If parent mtime has changed, we revalidate, else we wait for a 722 * period corresponding to the parent's attribute cache timeout value. 723 */ 724 static inline 725 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 726 struct nameidata *nd) 727 { 728 /* Don't revalidate a negative dentry if we're creating a new file */ 729 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) 730 return 0; 731 return !nfs_check_verifier(dir, dentry); 732 } 733 734 /* 735 * This is called every time the dcache has a lookup hit, 736 * and we should check whether we can really trust that 737 * lookup. 738 * 739 * NOTE! The hit can be a negative hit too, don't assume 740 * we have an inode! 741 * 742 * If the parent directory is seen to have changed, we throw out the 743 * cached dentry and do a new lookup. 744 */ 745 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) 746 { 747 struct inode *dir; 748 struct inode *inode; 749 struct dentry *parent; 750 int error; 751 struct nfs_fh fhandle; 752 struct nfs_fattr fattr; 753 754 parent = dget_parent(dentry); 755 lock_kernel(); 756 dir = parent->d_inode; 757 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 758 inode = dentry->d_inode; 759 760 if (!inode) { 761 if (nfs_neg_need_reval(dir, dentry, nd)) 762 goto out_bad; 763 goto out_valid; 764 } 765 766 if (is_bad_inode(inode)) { 767 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", 768 __FUNCTION__, dentry->d_parent->d_name.name, 769 dentry->d_name.name); 770 goto out_bad; 771 } 772 773 /* Force a full look up iff the parent directory has changed */ 774 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { 775 if (nfs_lookup_verify_inode(inode, nd)) 776 goto out_zap_parent; 777 goto out_valid; 778 } 779 780 if (NFS_STALE(inode)) 781 goto out_bad; 782 783 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 784 if (error) 785 goto out_bad; 786 if (nfs_compare_fh(NFS_FH(inode), &fhandle)) 787 goto out_bad; 788 if ((error = nfs_refresh_inode(inode, &fattr)) != 0) 789 goto out_bad; 790 791 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 792 out_valid: 793 unlock_kernel(); 794 dput(parent); 795 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", 796 __FUNCTION__, dentry->d_parent->d_name.name, 797 dentry->d_name.name); 798 return 1; 799 out_zap_parent: 800 nfs_zap_caches(dir); 801 out_bad: 802 nfs_mark_for_revalidate(dir); 803 if (inode && S_ISDIR(inode->i_mode)) { 804 /* Purge readdir caches. */ 805 nfs_zap_caches(inode); 806 /* If we have submounts, don't unhash ! */ 807 if (have_submounts(dentry)) 808 goto out_valid; 809 shrink_dcache_parent(dentry); 810 } 811 d_drop(dentry); 812 unlock_kernel(); 813 dput(parent); 814 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", 815 __FUNCTION__, dentry->d_parent->d_name.name, 816 dentry->d_name.name); 817 return 0; 818 } 819 820 /* 821 * This is called from dput() when d_count is going to 0. 822 */ 823 static int nfs_dentry_delete(struct dentry *dentry) 824 { 825 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", 826 dentry->d_parent->d_name.name, dentry->d_name.name, 827 dentry->d_flags); 828 829 /* Unhash any dentry with a stale inode */ 830 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) 831 return 1; 832 833 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 834 /* Unhash it, so that ->d_iput() would be called */ 835 return 1; 836 } 837 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 838 /* Unhash it, so that ancestors of killed async unlink 839 * files will be cleaned up during umount */ 840 return 1; 841 } 842 return 0; 843 844 } 845 846 /* 847 * Called when the dentry loses inode. 848 * We use it to clean up silly-renamed files. 849 */ 850 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 851 { 852 nfs_inode_return_delegation(inode); 853 if (S_ISDIR(inode->i_mode)) 854 /* drop any readdir cache as it could easily be old */ 855 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 856 857 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 858 lock_kernel(); 859 drop_nlink(inode); 860 nfs_complete_unlink(dentry, inode); 861 unlock_kernel(); 862 } 863 iput(inode); 864 } 865 866 struct dentry_operations nfs_dentry_operations = { 867 .d_revalidate = nfs_lookup_revalidate, 868 .d_delete = nfs_dentry_delete, 869 .d_iput = nfs_dentry_iput, 870 }; 871 872 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 873 { 874 struct dentry *res; 875 struct dentry *parent; 876 struct inode *inode = NULL; 877 int error; 878 struct nfs_fh fhandle; 879 struct nfs_fattr fattr; 880 881 dfprintk(VFS, "NFS: lookup(%s/%s)\n", 882 dentry->d_parent->d_name.name, dentry->d_name.name); 883 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 884 885 res = ERR_PTR(-ENAMETOOLONG); 886 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 887 goto out; 888 889 res = ERR_PTR(-ENOMEM); 890 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 891 892 lock_kernel(); 893 894 /* 895 * If we're doing an exclusive create, optimize away the lookup 896 * but don't hash the dentry. 897 */ 898 if (nfs_is_exclusive_create(dir, nd)) { 899 d_instantiate(dentry, NULL); 900 res = NULL; 901 goto out_unlock; 902 } 903 904 parent = dentry->d_parent; 905 /* Protect against concurrent sillydeletes */ 906 nfs_block_sillyrename(parent); 907 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 908 if (error == -ENOENT) 909 goto no_entry; 910 if (error < 0) { 911 res = ERR_PTR(error); 912 goto out_unblock_sillyrename; 913 } 914 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); 915 res = (struct dentry *)inode; 916 if (IS_ERR(res)) 917 goto out_unblock_sillyrename; 918 919 no_entry: 920 res = d_materialise_unique(dentry, inode); 921 if (res != NULL) { 922 if (IS_ERR(res)) 923 goto out_unblock_sillyrename; 924 dentry = res; 925 } 926 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 927 out_unblock_sillyrename: 928 nfs_unblock_sillyrename(parent); 929 out_unlock: 930 unlock_kernel(); 931 out: 932 return res; 933 } 934 935 #ifdef CONFIG_NFS_V4 936 static int nfs_open_revalidate(struct dentry *, struct nameidata *); 937 938 struct dentry_operations nfs4_dentry_operations = { 939 .d_revalidate = nfs_open_revalidate, 940 .d_delete = nfs_dentry_delete, 941 .d_iput = nfs_dentry_iput, 942 }; 943 944 /* 945 * Use intent information to determine whether we need to substitute 946 * the NFSv4-style stateful OPEN for the LOOKUP call 947 */ 948 static int is_atomic_open(struct inode *dir, struct nameidata *nd) 949 { 950 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) 951 return 0; 952 /* NFS does not (yet) have a stateful open for directories */ 953 if (nd->flags & LOOKUP_DIRECTORY) 954 return 0; 955 /* Are we trying to write to a read only partition? */ 956 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) 957 return 0; 958 return 1; 959 } 960 961 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 962 { 963 struct dentry *res = NULL; 964 int error; 965 966 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", 967 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 968 969 /* Check that we are indeed trying to open this file */ 970 if (!is_atomic_open(dir, nd)) 971 goto no_open; 972 973 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { 974 res = ERR_PTR(-ENAMETOOLONG); 975 goto out; 976 } 977 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 978 979 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash 980 * the dentry. */ 981 if (nd->intent.open.flags & O_EXCL) { 982 d_instantiate(dentry, NULL); 983 goto out; 984 } 985 986 /* Open the file on the server */ 987 lock_kernel(); 988 res = nfs4_atomic_open(dir, dentry, nd); 989 unlock_kernel(); 990 if (IS_ERR(res)) { 991 error = PTR_ERR(res); 992 switch (error) { 993 /* Make a negative dentry */ 994 case -ENOENT: 995 res = NULL; 996 goto out; 997 /* This turned out not to be a regular file */ 998 case -EISDIR: 999 case -ENOTDIR: 1000 goto no_open; 1001 case -ELOOP: 1002 if (!(nd->intent.open.flags & O_NOFOLLOW)) 1003 goto no_open; 1004 /* case -EINVAL: */ 1005 default: 1006 goto out; 1007 } 1008 } else if (res != NULL) 1009 dentry = res; 1010 out: 1011 return res; 1012 no_open: 1013 return nfs_lookup(dir, dentry, nd); 1014 } 1015 1016 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) 1017 { 1018 struct dentry *parent = NULL; 1019 struct inode *inode = dentry->d_inode; 1020 struct inode *dir; 1021 int openflags, ret = 0; 1022 1023 parent = dget_parent(dentry); 1024 dir = parent->d_inode; 1025 if (!is_atomic_open(dir, nd)) 1026 goto no_open; 1027 /* We can't create new files in nfs_open_revalidate(), so we 1028 * optimize away revalidation of negative dentries. 1029 */ 1030 if (inode == NULL) { 1031 if (!nfs_neg_need_reval(dir, dentry, nd)) 1032 ret = 1; 1033 goto out; 1034 } 1035 1036 /* NFS only supports OPEN on regular files */ 1037 if (!S_ISREG(inode->i_mode)) 1038 goto no_open; 1039 openflags = nd->intent.open.flags; 1040 /* We cannot do exclusive creation on a positive dentry */ 1041 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) 1042 goto no_open; 1043 /* We can't create new files, or truncate existing ones here */ 1044 openflags &= ~(O_CREAT|O_TRUNC); 1045 1046 /* 1047 * Note: we're not holding inode->i_mutex and so may be racing with 1048 * operations that change the directory. We therefore save the 1049 * change attribute *before* we do the RPC call. 1050 */ 1051 lock_kernel(); 1052 ret = nfs4_open_revalidate(dir, dentry, openflags, nd); 1053 unlock_kernel(); 1054 out: 1055 dput(parent); 1056 if (!ret) 1057 d_drop(dentry); 1058 return ret; 1059 no_open: 1060 dput(parent); 1061 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) 1062 return 1; 1063 return nfs_lookup_revalidate(dentry, nd); 1064 } 1065 #endif /* CONFIG_NFSV4 */ 1066 1067 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) 1068 { 1069 struct dentry *parent = desc->file->f_path.dentry; 1070 struct inode *dir = parent->d_inode; 1071 struct nfs_entry *entry = desc->entry; 1072 struct dentry *dentry, *alias; 1073 struct qstr name = { 1074 .name = entry->name, 1075 .len = entry->len, 1076 }; 1077 struct inode *inode; 1078 unsigned long verf = nfs_save_change_attribute(dir); 1079 1080 switch (name.len) { 1081 case 2: 1082 if (name.name[0] == '.' && name.name[1] == '.') 1083 return dget_parent(parent); 1084 break; 1085 case 1: 1086 if (name.name[0] == '.') 1087 return dget(parent); 1088 } 1089 1090 spin_lock(&dir->i_lock); 1091 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) { 1092 spin_unlock(&dir->i_lock); 1093 return NULL; 1094 } 1095 spin_unlock(&dir->i_lock); 1096 1097 name.hash = full_name_hash(name.name, name.len); 1098 dentry = d_lookup(parent, &name); 1099 if (dentry != NULL) { 1100 /* Is this a positive dentry that matches the readdir info? */ 1101 if (dentry->d_inode != NULL && 1102 (NFS_FILEID(dentry->d_inode) == entry->ino || 1103 d_mountpoint(dentry))) { 1104 if (!desc->plus || entry->fh->size == 0) 1105 return dentry; 1106 if (nfs_compare_fh(NFS_FH(dentry->d_inode), 1107 entry->fh) == 0) 1108 goto out_renew; 1109 } 1110 /* No, so d_drop to allow one to be created */ 1111 d_drop(dentry); 1112 dput(dentry); 1113 } 1114 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) 1115 return NULL; 1116 if (name.len > NFS_SERVER(dir)->namelen) 1117 return NULL; 1118 /* Note: caller is already holding the dir->i_mutex! */ 1119 dentry = d_alloc(parent, &name); 1120 if (dentry == NULL) 1121 return NULL; 1122 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1123 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 1124 if (IS_ERR(inode)) { 1125 dput(dentry); 1126 return NULL; 1127 } 1128 1129 alias = d_materialise_unique(dentry, inode); 1130 if (alias != NULL) { 1131 dput(dentry); 1132 if (IS_ERR(alias)) 1133 return NULL; 1134 dentry = alias; 1135 } 1136 1137 out_renew: 1138 nfs_set_verifier(dentry, verf); 1139 return dentry; 1140 } 1141 1142 /* 1143 * Code common to create, mkdir, and mknod. 1144 */ 1145 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1146 struct nfs_fattr *fattr) 1147 { 1148 struct dentry *parent = dget_parent(dentry); 1149 struct inode *dir = parent->d_inode; 1150 struct inode *inode; 1151 int error = -EACCES; 1152 1153 d_drop(dentry); 1154 1155 /* We may have been initialized further down */ 1156 if (dentry->d_inode) 1157 goto out; 1158 if (fhandle->size == 0) { 1159 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1160 if (error) 1161 goto out_error; 1162 } 1163 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1164 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1165 struct nfs_server *server = NFS_SB(dentry->d_sb); 1166 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1167 if (error < 0) 1168 goto out_error; 1169 } 1170 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1171 error = PTR_ERR(inode); 1172 if (IS_ERR(inode)) 1173 goto out_error; 1174 d_add(dentry, inode); 1175 out: 1176 dput(parent); 1177 return 0; 1178 out_error: 1179 nfs_mark_for_revalidate(dir); 1180 dput(parent); 1181 return error; 1182 } 1183 1184 /* 1185 * Following a failed create operation, we drop the dentry rather 1186 * than retain a negative dentry. This avoids a problem in the event 1187 * that the operation succeeded on the server, but an error in the 1188 * reply path made it appear to have failed. 1189 */ 1190 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, 1191 struct nameidata *nd) 1192 { 1193 struct iattr attr; 1194 int error; 1195 int open_flags = 0; 1196 1197 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1198 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1199 1200 attr.ia_mode = mode; 1201 attr.ia_valid = ATTR_MODE; 1202 1203 if ((nd->flags & LOOKUP_CREATE) != 0) 1204 open_flags = nd->intent.open.flags; 1205 1206 lock_kernel(); 1207 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); 1208 if (error != 0) 1209 goto out_err; 1210 unlock_kernel(); 1211 return 0; 1212 out_err: 1213 unlock_kernel(); 1214 d_drop(dentry); 1215 return error; 1216 } 1217 1218 /* 1219 * See comments for nfs_proc_create regarding failed operations. 1220 */ 1221 static int 1222 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) 1223 { 1224 struct iattr attr; 1225 int status; 1226 1227 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1228 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1229 1230 if (!new_valid_dev(rdev)) 1231 return -EINVAL; 1232 1233 attr.ia_mode = mode; 1234 attr.ia_valid = ATTR_MODE; 1235 1236 lock_kernel(); 1237 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1238 if (status != 0) 1239 goto out_err; 1240 unlock_kernel(); 1241 return 0; 1242 out_err: 1243 unlock_kernel(); 1244 d_drop(dentry); 1245 return status; 1246 } 1247 1248 /* 1249 * See comments for nfs_proc_create regarding failed operations. 1250 */ 1251 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1252 { 1253 struct iattr attr; 1254 int error; 1255 1256 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1257 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1258 1259 attr.ia_valid = ATTR_MODE; 1260 attr.ia_mode = mode | S_IFDIR; 1261 1262 lock_kernel(); 1263 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1264 if (error != 0) 1265 goto out_err; 1266 unlock_kernel(); 1267 return 0; 1268 out_err: 1269 d_drop(dentry); 1270 unlock_kernel(); 1271 return error; 1272 } 1273 1274 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1275 { 1276 if (dentry->d_inode != NULL && !d_unhashed(dentry)) 1277 d_delete(dentry); 1278 } 1279 1280 static int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1281 { 1282 int error; 1283 1284 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1285 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1286 1287 lock_kernel(); 1288 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1289 /* Ensure the VFS deletes this inode */ 1290 if (error == 0 && dentry->d_inode != NULL) 1291 clear_nlink(dentry->d_inode); 1292 else if (error == -ENOENT) 1293 nfs_dentry_handle_enoent(dentry); 1294 unlock_kernel(); 1295 1296 return error; 1297 } 1298 1299 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) 1300 { 1301 static unsigned int sillycounter; 1302 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2; 1303 const int countersize = sizeof(sillycounter)*2; 1304 const int slen = sizeof(".nfs")+fileidsize+countersize-1; 1305 char silly[slen+1]; 1306 struct qstr qsilly; 1307 struct dentry *sdentry; 1308 int error = -EIO; 1309 1310 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", 1311 dentry->d_parent->d_name.name, dentry->d_name.name, 1312 atomic_read(&dentry->d_count)); 1313 nfs_inc_stats(dir, NFSIOS_SILLYRENAME); 1314 1315 /* 1316 * We don't allow a dentry to be silly-renamed twice. 1317 */ 1318 error = -EBUSY; 1319 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1320 goto out; 1321 1322 sprintf(silly, ".nfs%*.*Lx", 1323 fileidsize, fileidsize, 1324 (unsigned long long)NFS_FILEID(dentry->d_inode)); 1325 1326 /* Return delegation in anticipation of the rename */ 1327 nfs_inode_return_delegation(dentry->d_inode); 1328 1329 sdentry = NULL; 1330 do { 1331 char *suffix = silly + slen - countersize; 1332 1333 dput(sdentry); 1334 sillycounter++; 1335 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); 1336 1337 dfprintk(VFS, "NFS: trying to rename %s to %s\n", 1338 dentry->d_name.name, silly); 1339 1340 sdentry = lookup_one_len(silly, dentry->d_parent, slen); 1341 /* 1342 * N.B. Better to return EBUSY here ... it could be 1343 * dangerous to delete the file while it's in use. 1344 */ 1345 if (IS_ERR(sdentry)) 1346 goto out; 1347 } while(sdentry->d_inode != NULL); /* need negative lookup */ 1348 1349 qsilly.name = silly; 1350 qsilly.len = strlen(silly); 1351 if (dentry->d_inode) { 1352 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1353 dir, &qsilly); 1354 nfs_mark_for_revalidate(dentry->d_inode); 1355 } else 1356 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1357 dir, &qsilly); 1358 if (!error) { 1359 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1360 d_move(dentry, sdentry); 1361 error = nfs_async_unlink(dir, dentry); 1362 /* If we return 0 we don't unlink */ 1363 } 1364 dput(sdentry); 1365 out: 1366 return error; 1367 } 1368 1369 /* 1370 * Remove a file after making sure there are no pending writes, 1371 * and after checking that the file has only one user. 1372 * 1373 * We invalidate the attribute cache and free the inode prior to the operation 1374 * to avoid possible races if the server reuses the inode. 1375 */ 1376 static int nfs_safe_remove(struct dentry *dentry) 1377 { 1378 struct inode *dir = dentry->d_parent->d_inode; 1379 struct inode *inode = dentry->d_inode; 1380 int error = -EBUSY; 1381 1382 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1383 dentry->d_parent->d_name.name, dentry->d_name.name); 1384 1385 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1386 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1387 error = 0; 1388 goto out; 1389 } 1390 1391 if (inode != NULL) { 1392 nfs_inode_return_delegation(inode); 1393 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1394 /* The VFS may want to delete this inode */ 1395 if (error == 0) 1396 drop_nlink(inode); 1397 nfs_mark_for_revalidate(inode); 1398 } else 1399 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1400 if (error == -ENOENT) 1401 nfs_dentry_handle_enoent(dentry); 1402 out: 1403 return error; 1404 } 1405 1406 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1407 * belongs to an active ".nfs..." file and we return -EBUSY. 1408 * 1409 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1410 */ 1411 static int nfs_unlink(struct inode *dir, struct dentry *dentry) 1412 { 1413 int error; 1414 int need_rehash = 0; 1415 1416 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1417 dir->i_ino, dentry->d_name.name); 1418 1419 lock_kernel(); 1420 spin_lock(&dcache_lock); 1421 spin_lock(&dentry->d_lock); 1422 if (atomic_read(&dentry->d_count) > 1) { 1423 spin_unlock(&dentry->d_lock); 1424 spin_unlock(&dcache_lock); 1425 /* Start asynchronous writeout of the inode */ 1426 write_inode_now(dentry->d_inode, 0); 1427 error = nfs_sillyrename(dir, dentry); 1428 unlock_kernel(); 1429 return error; 1430 } 1431 if (!d_unhashed(dentry)) { 1432 __d_drop(dentry); 1433 need_rehash = 1; 1434 } 1435 spin_unlock(&dentry->d_lock); 1436 spin_unlock(&dcache_lock); 1437 error = nfs_safe_remove(dentry); 1438 if (!error || error == -ENOENT) { 1439 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1440 } else if (need_rehash) 1441 d_rehash(dentry); 1442 unlock_kernel(); 1443 return error; 1444 } 1445 1446 /* 1447 * To create a symbolic link, most file systems instantiate a new inode, 1448 * add a page to it containing the path, then write it out to the disk 1449 * using prepare_write/commit_write. 1450 * 1451 * Unfortunately the NFS client can't create the in-core inode first 1452 * because it needs a file handle to create an in-core inode (see 1453 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1454 * symlink request has completed on the server. 1455 * 1456 * So instead we allocate a raw page, copy the symname into it, then do 1457 * the SYMLINK request with the page as the buffer. If it succeeds, we 1458 * now have a new file handle and can instantiate an in-core NFS inode 1459 * and move the raw page into its mapping. 1460 */ 1461 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1462 { 1463 struct pagevec lru_pvec; 1464 struct page *page; 1465 char *kaddr; 1466 struct iattr attr; 1467 unsigned int pathlen = strlen(symname); 1468 int error; 1469 1470 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1471 dir->i_ino, dentry->d_name.name, symname); 1472 1473 if (pathlen > PAGE_SIZE) 1474 return -ENAMETOOLONG; 1475 1476 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1477 attr.ia_valid = ATTR_MODE; 1478 1479 lock_kernel(); 1480 1481 page = alloc_page(GFP_HIGHUSER); 1482 if (!page) { 1483 unlock_kernel(); 1484 return -ENOMEM; 1485 } 1486 1487 kaddr = kmap_atomic(page, KM_USER0); 1488 memcpy(kaddr, symname, pathlen); 1489 if (pathlen < PAGE_SIZE) 1490 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1491 kunmap_atomic(kaddr, KM_USER0); 1492 1493 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1494 if (error != 0) { 1495 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1496 dir->i_sb->s_id, dir->i_ino, 1497 dentry->d_name.name, symname, error); 1498 d_drop(dentry); 1499 __free_page(page); 1500 unlock_kernel(); 1501 return error; 1502 } 1503 1504 /* 1505 * No big deal if we can't add this page to the page cache here. 1506 * READLINK will get the missing page from the server if needed. 1507 */ 1508 pagevec_init(&lru_pvec, 0); 1509 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1510 GFP_KERNEL)) { 1511 pagevec_add(&lru_pvec, page); 1512 pagevec_lru_add(&lru_pvec); 1513 SetPageUptodate(page); 1514 unlock_page(page); 1515 } else 1516 __free_page(page); 1517 1518 unlock_kernel(); 1519 return 0; 1520 } 1521 1522 static int 1523 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1524 { 1525 struct inode *inode = old_dentry->d_inode; 1526 int error; 1527 1528 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1529 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1530 dentry->d_parent->d_name.name, dentry->d_name.name); 1531 1532 lock_kernel(); 1533 d_drop(dentry); 1534 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1535 if (error == 0) { 1536 atomic_inc(&inode->i_count); 1537 d_add(dentry, inode); 1538 } 1539 unlock_kernel(); 1540 return error; 1541 } 1542 1543 /* 1544 * RENAME 1545 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1546 * different file handle for the same inode after a rename (e.g. when 1547 * moving to a different directory). A fail-safe method to do so would 1548 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1549 * rename the old file using the sillyrename stuff. This way, the original 1550 * file in old_dir will go away when the last process iput()s the inode. 1551 * 1552 * FIXED. 1553 * 1554 * It actually works quite well. One needs to have the possibility for 1555 * at least one ".nfs..." file in each directory the file ever gets 1556 * moved or linked to which happens automagically with the new 1557 * implementation that only depends on the dcache stuff instead of 1558 * using the inode layer 1559 * 1560 * Unfortunately, things are a little more complicated than indicated 1561 * above. For a cross-directory move, we want to make sure we can get 1562 * rid of the old inode after the operation. This means there must be 1563 * no pending writes (if it's a file), and the use count must be 1. 1564 * If these conditions are met, we can drop the dentries before doing 1565 * the rename. 1566 */ 1567 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1568 struct inode *new_dir, struct dentry *new_dentry) 1569 { 1570 struct inode *old_inode = old_dentry->d_inode; 1571 struct inode *new_inode = new_dentry->d_inode; 1572 struct dentry *dentry = NULL, *rehash = NULL; 1573 int error = -EBUSY; 1574 1575 /* 1576 * To prevent any new references to the target during the rename, 1577 * we unhash the dentry and free the inode in advance. 1578 */ 1579 lock_kernel(); 1580 if (!d_unhashed(new_dentry)) { 1581 d_drop(new_dentry); 1582 rehash = new_dentry; 1583 } 1584 1585 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1586 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1587 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1588 atomic_read(&new_dentry->d_count)); 1589 1590 /* 1591 * First check whether the target is busy ... we can't 1592 * safely do _any_ rename if the target is in use. 1593 * 1594 * For files, make a copy of the dentry and then do a 1595 * silly-rename. If the silly-rename succeeds, the 1596 * copied dentry is hashed and becomes the new target. 1597 */ 1598 if (!new_inode) 1599 goto go_ahead; 1600 if (S_ISDIR(new_inode->i_mode)) { 1601 error = -EISDIR; 1602 if (!S_ISDIR(old_inode->i_mode)) 1603 goto out; 1604 } else if (atomic_read(&new_dentry->d_count) > 2) { 1605 int err; 1606 /* copy the target dentry's name */ 1607 dentry = d_alloc(new_dentry->d_parent, 1608 &new_dentry->d_name); 1609 if (!dentry) 1610 goto out; 1611 1612 /* silly-rename the existing target ... */ 1613 err = nfs_sillyrename(new_dir, new_dentry); 1614 if (!err) { 1615 new_dentry = rehash = dentry; 1616 new_inode = NULL; 1617 /* instantiate the replacement target */ 1618 d_instantiate(new_dentry, NULL); 1619 } else if (atomic_read(&new_dentry->d_count) > 1) 1620 /* dentry still busy? */ 1621 goto out; 1622 } else 1623 drop_nlink(new_inode); 1624 1625 go_ahead: 1626 /* 1627 * ... prune child dentries and writebacks if needed. 1628 */ 1629 if (atomic_read(&old_dentry->d_count) > 1) { 1630 if (S_ISREG(old_inode->i_mode)) 1631 nfs_wb_all(old_inode); 1632 shrink_dcache_parent(old_dentry); 1633 } 1634 nfs_inode_return_delegation(old_inode); 1635 1636 if (new_inode != NULL) { 1637 nfs_inode_return_delegation(new_inode); 1638 d_delete(new_dentry); 1639 } 1640 1641 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1642 new_dir, &new_dentry->d_name); 1643 nfs_mark_for_revalidate(old_inode); 1644 out: 1645 if (rehash) 1646 d_rehash(rehash); 1647 if (!error) { 1648 d_move(old_dentry, new_dentry); 1649 nfs_set_verifier(new_dentry, 1650 nfs_save_change_attribute(new_dir)); 1651 } else if (error == -ENOENT) 1652 nfs_dentry_handle_enoent(old_dentry); 1653 1654 /* new dentry created? */ 1655 if (dentry) 1656 dput(dentry); 1657 unlock_kernel(); 1658 return error; 1659 } 1660 1661 static DEFINE_SPINLOCK(nfs_access_lru_lock); 1662 static LIST_HEAD(nfs_access_lru_list); 1663 static atomic_long_t nfs_access_nr_entries; 1664 1665 static void nfs_access_free_entry(struct nfs_access_entry *entry) 1666 { 1667 put_rpccred(entry->cred); 1668 kfree(entry); 1669 smp_mb__before_atomic_dec(); 1670 atomic_long_dec(&nfs_access_nr_entries); 1671 smp_mb__after_atomic_dec(); 1672 } 1673 1674 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) 1675 { 1676 LIST_HEAD(head); 1677 struct nfs_inode *nfsi; 1678 struct nfs_access_entry *cache; 1679 1680 restart: 1681 spin_lock(&nfs_access_lru_lock); 1682 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { 1683 struct inode *inode; 1684 1685 if (nr_to_scan-- == 0) 1686 break; 1687 inode = igrab(&nfsi->vfs_inode); 1688 if (inode == NULL) 1689 continue; 1690 spin_lock(&inode->i_lock); 1691 if (list_empty(&nfsi->access_cache_entry_lru)) 1692 goto remove_lru_entry; 1693 cache = list_entry(nfsi->access_cache_entry_lru.next, 1694 struct nfs_access_entry, lru); 1695 list_move(&cache->lru, &head); 1696 rb_erase(&cache->rb_node, &nfsi->access_cache); 1697 if (!list_empty(&nfsi->access_cache_entry_lru)) 1698 list_move_tail(&nfsi->access_cache_inode_lru, 1699 &nfs_access_lru_list); 1700 else { 1701 remove_lru_entry: 1702 list_del_init(&nfsi->access_cache_inode_lru); 1703 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1704 } 1705 spin_unlock(&inode->i_lock); 1706 spin_unlock(&nfs_access_lru_lock); 1707 iput(inode); 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_FLAGS(inode))) { 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_FLAGS(inode))) { 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 ((cache.mask & mask) == mask) 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, struct nameidata *nd) 1911 { 1912 struct rpc_cred *cred; 1913 int res = 0; 1914 1915 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 1916 1917 if (mask == 0) 1918 goto out; 1919 /* Is this sys_access() ? */ 1920 if (nd != NULL && (nd->flags & LOOKUP_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 && nd != NULL 1930 && (nd->flags & LOOKUP_OPEN)) 1931 goto out; 1932 break; 1933 case S_IFDIR: 1934 /* 1935 * Optimize away all write operations, since the server 1936 * will check permissions when we perform the op. 1937 */ 1938 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 1939 goto out; 1940 } 1941 1942 force_lookup: 1943 lock_kernel(); 1944 1945 if (!NFS_PROTO(inode)->access) 1946 goto out_notsup; 1947 1948 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 1949 if (!IS_ERR(cred)) { 1950 res = nfs_do_access(inode, cred, mask); 1951 put_rpccred(cred); 1952 } else 1953 res = PTR_ERR(cred); 1954 unlock_kernel(); 1955 out: 1956 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 1957 inode->i_sb->s_id, inode->i_ino, mask, res); 1958 return res; 1959 out_notsup: 1960 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1961 if (res == 0) 1962 res = generic_permission(inode, mask, NULL); 1963 unlock_kernel(); 1964 goto out; 1965 } 1966 1967 /* 1968 * Local variables: 1969 * version-control: t 1970 * kept-new-versions: 5 1971 * End: 1972 */ 1973