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 37 #include "nfs4_fs.h" 38 #include "delegation.h" 39 #include "iostat.h" 40 41 #define NFS_PARANOIA 1 42 /* #define NFS_DEBUG_VERBOSE 1 */ 43 44 static int nfs_opendir(struct inode *, struct file *); 45 static int nfs_readdir(struct file *, void *, filldir_t); 46 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); 47 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); 48 static int nfs_mkdir(struct inode *, struct dentry *, int); 49 static int nfs_rmdir(struct inode *, struct dentry *); 50 static int nfs_unlink(struct inode *, struct dentry *); 51 static int nfs_symlink(struct inode *, struct dentry *, const char *); 52 static int nfs_link(struct dentry *, struct inode *, struct dentry *); 53 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); 54 static int nfs_rename(struct inode *, struct dentry *, 55 struct inode *, struct dentry *); 56 static int nfs_fsync_dir(struct file *, struct dentry *, int); 57 static loff_t nfs_llseek_dir(struct file *, loff_t, int); 58 59 const struct file_operations nfs_dir_operations = { 60 .llseek = nfs_llseek_dir, 61 .read = generic_read_dir, 62 .readdir = nfs_readdir, 63 .open = nfs_opendir, 64 .release = nfs_release, 65 .fsync = nfs_fsync_dir, 66 }; 67 68 struct inode_operations nfs_dir_inode_operations = { 69 .create = nfs_create, 70 .lookup = nfs_lookup, 71 .link = nfs_link, 72 .unlink = nfs_unlink, 73 .symlink = nfs_symlink, 74 .mkdir = nfs_mkdir, 75 .rmdir = nfs_rmdir, 76 .mknod = nfs_mknod, 77 .rename = nfs_rename, 78 .permission = nfs_permission, 79 .getattr = nfs_getattr, 80 .setattr = nfs_setattr, 81 }; 82 83 #ifdef CONFIG_NFS_V3 84 struct inode_operations nfs3_dir_inode_operations = { 85 .create = nfs_create, 86 .lookup = nfs_lookup, 87 .link = nfs_link, 88 .unlink = nfs_unlink, 89 .symlink = nfs_symlink, 90 .mkdir = nfs_mkdir, 91 .rmdir = nfs_rmdir, 92 .mknod = nfs_mknod, 93 .rename = nfs_rename, 94 .permission = nfs_permission, 95 .getattr = nfs_getattr, 96 .setattr = nfs_setattr, 97 .listxattr = nfs3_listxattr, 98 .getxattr = nfs3_getxattr, 99 .setxattr = nfs3_setxattr, 100 .removexattr = nfs3_removexattr, 101 }; 102 #endif /* CONFIG_NFS_V3 */ 103 104 #ifdef CONFIG_NFS_V4 105 106 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); 107 struct inode_operations nfs4_dir_inode_operations = { 108 .create = nfs_create, 109 .lookup = nfs_atomic_lookup, 110 .link = nfs_link, 111 .unlink = nfs_unlink, 112 .symlink = nfs_symlink, 113 .mkdir = nfs_mkdir, 114 .rmdir = nfs_rmdir, 115 .mknod = nfs_mknod, 116 .rename = nfs_rename, 117 .permission = nfs_permission, 118 .getattr = nfs_getattr, 119 .setattr = nfs_setattr, 120 .getxattr = nfs4_getxattr, 121 .setxattr = nfs4_setxattr, 122 .listxattr = nfs4_listxattr, 123 }; 124 125 #endif /* CONFIG_NFS_V4 */ 126 127 /* 128 * Open file 129 */ 130 static int 131 nfs_opendir(struct inode *inode, struct file *filp) 132 { 133 int res; 134 135 dfprintk(VFS, "NFS: opendir(%s/%ld)\n", 136 inode->i_sb->s_id, inode->i_ino); 137 138 lock_kernel(); 139 /* Call generic open code in order to cache credentials */ 140 res = nfs_open(inode, filp); 141 unlock_kernel(); 142 return res; 143 } 144 145 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int); 146 typedef struct { 147 struct file *file; 148 struct page *page; 149 unsigned long page_index; 150 __be32 *ptr; 151 u64 *dir_cookie; 152 loff_t current_index; 153 struct nfs_entry *entry; 154 decode_dirent_t decode; 155 int plus; 156 int error; 157 } nfs_readdir_descriptor_t; 158 159 /* Now we cache directories properly, by stuffing the dirent 160 * data directly in the page cache. 161 * 162 * Inode invalidation due to refresh etc. takes care of 163 * _everything_, no sloppy entry flushing logic, no extraneous 164 * copying, network direct to page cache, the way it was meant 165 * to be. 166 * 167 * NOTE: Dirent information verification is done always by the 168 * page-in of the RPC reply, nowhere else, this simplies 169 * things substantially. 170 */ 171 static 172 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page) 173 { 174 struct file *file = desc->file; 175 struct inode *inode = file->f_path.dentry->d_inode; 176 struct rpc_cred *cred = nfs_file_cred(file); 177 unsigned long timestamp; 178 int error; 179 180 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n", 181 __FUNCTION__, (long long)desc->entry->cookie, 182 page->index); 183 184 again: 185 timestamp = jiffies; 186 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page, 187 NFS_SERVER(inode)->dtsize, desc->plus); 188 if (error < 0) { 189 /* We requested READDIRPLUS, but the server doesn't grok it */ 190 if (error == -ENOTSUPP && desc->plus) { 191 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 192 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 193 desc->plus = 0; 194 goto again; 195 } 196 goto error; 197 } 198 SetPageUptodate(page); 199 spin_lock(&inode->i_lock); 200 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; 201 spin_unlock(&inode->i_lock); 202 /* Ensure consistent page alignment of the data. 203 * Note: assumes we have exclusive access to this mapping either 204 * through inode->i_mutex or some other mechanism. 205 */ 206 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) { 207 /* Should never happen */ 208 nfs_zap_mapping(inode, inode->i_mapping); 209 } 210 unlock_page(page); 211 return 0; 212 error: 213 SetPageError(page); 214 unlock_page(page); 215 nfs_zap_caches(inode); 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 return 0; 229 } 230 231 static inline 232 void dir_page_release(nfs_readdir_descriptor_t *desc) 233 { 234 kunmap(desc->page); 235 page_cache_release(desc->page); 236 desc->page = NULL; 237 desc->ptr = NULL; 238 } 239 240 /* 241 * Given a pointer to a buffer that has already been filled by a call 242 * to readdir, find the next entry with cookie '*desc->dir_cookie'. 243 * 244 * If the end of the buffer has been reached, return -EAGAIN, if not, 245 * return the offset within the buffer of the next entry to be 246 * read. 247 */ 248 static inline 249 int find_dirent(nfs_readdir_descriptor_t *desc) 250 { 251 struct nfs_entry *entry = desc->entry; 252 int loop_count = 0, 253 status; 254 255 while((status = dir_decode(desc)) == 0) { 256 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n", 257 __FUNCTION__, (unsigned long long)entry->cookie); 258 if (entry->prev_cookie == *desc->dir_cookie) 259 break; 260 if (loop_count++ > 200) { 261 loop_count = 0; 262 schedule(); 263 } 264 } 265 return status; 266 } 267 268 /* 269 * Given a pointer to a buffer that has already been filled by a call 270 * to readdir, find the entry at offset 'desc->file->f_pos'. 271 * 272 * If the end of the buffer has been reached, return -EAGAIN, if not, 273 * return the offset within the buffer of the next entry to be 274 * read. 275 */ 276 static inline 277 int find_dirent_index(nfs_readdir_descriptor_t *desc) 278 { 279 struct nfs_entry *entry = desc->entry; 280 int loop_count = 0, 281 status; 282 283 for(;;) { 284 status = dir_decode(desc); 285 if (status) 286 break; 287 288 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n", 289 (unsigned long long)entry->cookie, desc->current_index); 290 291 if (desc->file->f_pos == desc->current_index) { 292 *desc->dir_cookie = entry->cookie; 293 break; 294 } 295 desc->current_index++; 296 if (loop_count++ > 200) { 297 loop_count = 0; 298 schedule(); 299 } 300 } 301 return status; 302 } 303 304 /* 305 * Find the given page, and call find_dirent() or find_dirent_index in 306 * order to try to return the next entry. 307 */ 308 static inline 309 int find_dirent_page(nfs_readdir_descriptor_t *desc) 310 { 311 struct inode *inode = desc->file->f_path.dentry->d_inode; 312 struct page *page; 313 int status; 314 315 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n", 316 __FUNCTION__, desc->page_index, 317 (long long) *desc->dir_cookie); 318 319 page = read_cache_page(inode->i_mapping, desc->page_index, 320 (filler_t *)nfs_readdir_filler, desc); 321 if (IS_ERR(page)) { 322 status = PTR_ERR(page); 323 goto out; 324 } 325 if (!PageUptodate(page)) 326 goto read_error; 327 328 /* NOTE: Someone else may have changed the READDIRPLUS flag */ 329 desc->page = page; 330 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 331 if (*desc->dir_cookie != 0) 332 status = find_dirent(desc); 333 else 334 status = find_dirent_index(desc); 335 if (status < 0) 336 dir_page_release(desc); 337 out: 338 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status); 339 return status; 340 read_error: 341 page_cache_release(page); 342 return -EIO; 343 } 344 345 /* 346 * Recurse through the page cache pages, and return a 347 * filled nfs_entry structure of the next directory entry if possible. 348 * 349 * The target for the search is '*desc->dir_cookie' if non-0, 350 * 'desc->file->f_pos' otherwise 351 */ 352 static inline 353 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 354 { 355 int loop_count = 0; 356 int res; 357 358 /* Always search-by-index from the beginning of the cache */ 359 if (*desc->dir_cookie == 0) { 360 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n", 361 (long long)desc->file->f_pos); 362 desc->page_index = 0; 363 desc->entry->cookie = desc->entry->prev_cookie = 0; 364 desc->entry->eof = 0; 365 desc->current_index = 0; 366 } else 367 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", 368 (unsigned long long)*desc->dir_cookie); 369 370 for (;;) { 371 res = find_dirent_page(desc); 372 if (res != -EAGAIN) 373 break; 374 /* Align to beginning of next page */ 375 desc->page_index ++; 376 if (loop_count++ > 200) { 377 loop_count = 0; 378 schedule(); 379 } 380 } 381 382 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res); 383 return res; 384 } 385 386 static inline unsigned int dt_type(struct inode *inode) 387 { 388 return (inode->i_mode >> 12) & 15; 389 } 390 391 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc); 392 393 /* 394 * Once we've found the start of the dirent within a page: fill 'er up... 395 */ 396 static 397 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, 398 filldir_t filldir) 399 { 400 struct file *file = desc->file; 401 struct nfs_entry *entry = desc->entry; 402 struct dentry *dentry = NULL; 403 unsigned long fileid; 404 int loop_count = 0, 405 res; 406 407 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", 408 (unsigned long long)entry->cookie); 409 410 for(;;) { 411 unsigned d_type = DT_UNKNOWN; 412 /* Note: entry->prev_cookie contains the cookie for 413 * retrieving the current dirent on the server */ 414 fileid = nfs_fileid_to_ino_t(entry->ino); 415 416 /* Get a dentry if we have one */ 417 if (dentry != NULL) 418 dput(dentry); 419 dentry = nfs_readdir_lookup(desc); 420 421 /* Use readdirplus info */ 422 if (dentry != NULL && dentry->d_inode != NULL) { 423 d_type = dt_type(dentry->d_inode); 424 fileid = dentry->d_inode->i_ino; 425 } 426 427 res = filldir(dirent, entry->name, entry->len, 428 file->f_pos, fileid, d_type); 429 if (res < 0) 430 break; 431 file->f_pos++; 432 *desc->dir_cookie = entry->cookie; 433 if (dir_decode(desc) != 0) { 434 desc->page_index ++; 435 break; 436 } 437 if (loop_count++ > 200) { 438 loop_count = 0; 439 schedule(); 440 } 441 } 442 dir_page_release(desc); 443 if (dentry != NULL) 444 dput(dentry); 445 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 446 (unsigned long long)*desc->dir_cookie, res); 447 return res; 448 } 449 450 /* 451 * If we cannot find a cookie in our cache, we suspect that this is 452 * because it points to a deleted file, so we ask the server to return 453 * whatever it thinks is the next entry. We then feed this to filldir. 454 * If all goes well, we should then be able to find our way round the 455 * cache on the next call to readdir_search_pagecache(); 456 * 457 * NOTE: we cannot add the anonymous page to the pagecache because 458 * the data it contains might not be page aligned. Besides, 459 * we should already have a complete representation of the 460 * directory in the page cache by the time we get here. 461 */ 462 static inline 463 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, 464 filldir_t filldir) 465 { 466 struct file *file = desc->file; 467 struct inode *inode = file->f_path.dentry->d_inode; 468 struct rpc_cred *cred = nfs_file_cred(file); 469 struct page *page = NULL; 470 int status; 471 472 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 473 (unsigned long long)*desc->dir_cookie); 474 475 page = alloc_page(GFP_HIGHUSER); 476 if (!page) { 477 status = -ENOMEM; 478 goto out; 479 } 480 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie, 481 page, 482 NFS_SERVER(inode)->dtsize, 483 desc->plus); 484 spin_lock(&inode->i_lock); 485 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; 486 spin_unlock(&inode->i_lock); 487 desc->page = page; 488 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ 489 if (desc->error >= 0) { 490 if ((status = dir_decode(desc)) == 0) 491 desc->entry->prev_cookie = *desc->dir_cookie; 492 } else 493 status = -EIO; 494 if (status < 0) 495 goto out_release; 496 497 status = nfs_do_filldir(desc, dirent, filldir); 498 499 /* Reset read descriptor so it searches the page cache from 500 * the start upon the next call to readdir_search_pagecache() */ 501 desc->page_index = 0; 502 desc->entry->cookie = desc->entry->prev_cookie = 0; 503 desc->entry->eof = 0; 504 out: 505 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 506 __FUNCTION__, status); 507 return status; 508 out_release: 509 dir_page_release(desc); 510 goto out; 511 } 512 513 /* The file offset position represents the dirent entry number. A 514 last cookie cache takes care of the common case of reading the 515 whole directory. 516 */ 517 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) 518 { 519 struct dentry *dentry = filp->f_path.dentry; 520 struct inode *inode = dentry->d_inode; 521 nfs_readdir_descriptor_t my_desc, 522 *desc = &my_desc; 523 struct nfs_entry my_entry; 524 struct nfs_fh fh; 525 struct nfs_fattr fattr; 526 long res; 527 528 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n", 529 dentry->d_parent->d_name.name, dentry->d_name.name, 530 (long long)filp->f_pos); 531 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 532 533 lock_kernel(); 534 535 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping); 536 if (res < 0) { 537 unlock_kernel(); 538 return res; 539 } 540 541 /* 542 * filp->f_pos points to the dirent entry number. 543 * *desc->dir_cookie has the cookie for the next entry. We have 544 * to either find the entry with the appropriate number or 545 * revalidate the cookie. 546 */ 547 memset(desc, 0, sizeof(*desc)); 548 549 desc->file = filp; 550 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie; 551 desc->decode = NFS_PROTO(inode)->decode_dirent; 552 desc->plus = NFS_USE_READDIRPLUS(inode); 553 554 my_entry.cookie = my_entry.prev_cookie = 0; 555 my_entry.eof = 0; 556 my_entry.fh = &fh; 557 my_entry.fattr = &fattr; 558 nfs_fattr_init(&fattr); 559 desc->entry = &my_entry; 560 561 while(!desc->entry->eof) { 562 res = readdir_search_pagecache(desc); 563 564 if (res == -EBADCOOKIE) { 565 /* This means either end of directory */ 566 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { 567 /* Or that the server has 'lost' a cookie */ 568 res = uncached_readdir(desc, dirent, filldir); 569 if (res >= 0) 570 continue; 571 } 572 res = 0; 573 break; 574 } 575 if (res == -ETOOSMALL && desc->plus) { 576 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); 577 nfs_zap_caches(inode); 578 desc->plus = 0; 579 desc->entry->eof = 0; 580 continue; 581 } 582 if (res < 0) 583 break; 584 585 res = nfs_do_filldir(desc, dirent, filldir); 586 if (res < 0) { 587 res = 0; 588 break; 589 } 590 } 591 unlock_kernel(); 592 if (res > 0) 593 res = 0; 594 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n", 595 dentry->d_parent->d_name.name, dentry->d_name.name, 596 res); 597 return res; 598 } 599 600 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) 601 { 602 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex); 603 switch (origin) { 604 case 1: 605 offset += filp->f_pos; 606 case 0: 607 if (offset >= 0) 608 break; 609 default: 610 offset = -EINVAL; 611 goto out; 612 } 613 if (offset != filp->f_pos) { 614 filp->f_pos = offset; 615 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0; 616 } 617 out: 618 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex); 619 return offset; 620 } 621 622 /* 623 * All directory operations under NFS are synchronous, so fsync() 624 * is a dummy operation. 625 */ 626 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) 627 { 628 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n", 629 dentry->d_parent->d_name.name, dentry->d_name.name, 630 datasync); 631 632 return 0; 633 } 634 635 /* 636 * A check for whether or not the parent directory has changed. 637 * In the case it has, we assume that the dentries are untrustworthy 638 * and may need to be looked up again. 639 */ 640 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry) 641 { 642 if (IS_ROOT(dentry)) 643 return 1; 644 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0 645 || nfs_attribute_timeout(dir)) 646 return 0; 647 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata); 648 } 649 650 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf) 651 { 652 dentry->d_fsdata = (void *)verf; 653 } 654 655 /* 656 * Whenever an NFS operation succeeds, we know that the dentry 657 * is valid, so we update the revalidation timestamp. 658 */ 659 static inline void nfs_renew_times(struct dentry * dentry) 660 { 661 dentry->d_time = jiffies; 662 } 663 664 /* 665 * Return the intent data that applies to this particular path component 666 * 667 * Note that the current set of intents only apply to the very last 668 * component of the path. 669 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. 670 */ 671 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) 672 { 673 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) 674 return 0; 675 return nd->flags & mask; 676 } 677 678 /* 679 * Inode and filehandle revalidation for lookups. 680 * 681 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 682 * or if the intent information indicates that we're about to open this 683 * particular file and the "nocto" mount flag is not set. 684 * 685 */ 686 static inline 687 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) 688 { 689 struct nfs_server *server = NFS_SERVER(inode); 690 691 if (nd != NULL) { 692 /* VFS wants an on-the-wire revalidation */ 693 if (nd->flags & LOOKUP_REVAL) 694 goto out_force; 695 /* This is an open(2) */ 696 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && 697 !(server->flags & NFS_MOUNT_NOCTO) && 698 (S_ISREG(inode->i_mode) || 699 S_ISDIR(inode->i_mode))) 700 goto out_force; 701 } 702 return nfs_revalidate_inode(server, inode); 703 out_force: 704 return __nfs_revalidate_inode(server, inode); 705 } 706 707 /* 708 * We judge how long we want to trust negative 709 * dentries by looking at the parent inode mtime. 710 * 711 * If parent mtime has changed, we revalidate, else we wait for a 712 * period corresponding to the parent's attribute cache timeout value. 713 */ 714 static inline 715 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 716 struct nameidata *nd) 717 { 718 /* Don't revalidate a negative dentry if we're creating a new file */ 719 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) 720 return 0; 721 return !nfs_check_verifier(dir, dentry); 722 } 723 724 /* 725 * This is called every time the dcache has a lookup hit, 726 * and we should check whether we can really trust that 727 * lookup. 728 * 729 * NOTE! The hit can be a negative hit too, don't assume 730 * we have an inode! 731 * 732 * If the parent directory is seen to have changed, we throw out the 733 * cached dentry and do a new lookup. 734 */ 735 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) 736 { 737 struct inode *dir; 738 struct inode *inode; 739 struct dentry *parent; 740 int error; 741 struct nfs_fh fhandle; 742 struct nfs_fattr fattr; 743 unsigned long verifier; 744 745 parent = dget_parent(dentry); 746 lock_kernel(); 747 dir = parent->d_inode; 748 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 749 inode = dentry->d_inode; 750 751 if (!inode) { 752 if (nfs_neg_need_reval(dir, dentry, nd)) 753 goto out_bad; 754 goto out_valid; 755 } 756 757 if (is_bad_inode(inode)) { 758 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", 759 __FUNCTION__, dentry->d_parent->d_name.name, 760 dentry->d_name.name); 761 goto out_bad; 762 } 763 764 /* Revalidate parent directory attribute cache */ 765 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 766 goto out_zap_parent; 767 768 /* Force a full look up iff the parent directory has changed */ 769 if (nfs_check_verifier(dir, dentry)) { 770 if (nfs_lookup_verify_inode(inode, nd)) 771 goto out_zap_parent; 772 goto out_valid; 773 } 774 775 if (NFS_STALE(inode)) 776 goto out_bad; 777 778 verifier = nfs_save_change_attribute(dir); 779 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 780 if (error) 781 goto out_bad; 782 if (nfs_compare_fh(NFS_FH(inode), &fhandle)) 783 goto out_bad; 784 if ((error = nfs_refresh_inode(inode, &fattr)) != 0) 785 goto out_bad; 786 787 nfs_renew_times(dentry); 788 nfs_set_verifier(dentry, verifier); 789 out_valid: 790 unlock_kernel(); 791 dput(parent); 792 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", 793 __FUNCTION__, dentry->d_parent->d_name.name, 794 dentry->d_name.name); 795 return 1; 796 out_zap_parent: 797 nfs_zap_caches(dir); 798 out_bad: 799 NFS_CACHEINV(dir); 800 if (inode && S_ISDIR(inode->i_mode)) { 801 /* Purge readdir caches. */ 802 nfs_zap_caches(inode); 803 /* If we have submounts, don't unhash ! */ 804 if (have_submounts(dentry)) 805 goto out_valid; 806 shrink_dcache_parent(dentry); 807 } 808 d_drop(dentry); 809 unlock_kernel(); 810 dput(parent); 811 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", 812 __FUNCTION__, dentry->d_parent->d_name.name, 813 dentry->d_name.name); 814 return 0; 815 } 816 817 /* 818 * This is called from dput() when d_count is going to 0. 819 */ 820 static int nfs_dentry_delete(struct dentry *dentry) 821 { 822 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", 823 dentry->d_parent->d_name.name, dentry->d_name.name, 824 dentry->d_flags); 825 826 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 827 /* Unhash it, so that ->d_iput() would be called */ 828 return 1; 829 } 830 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 831 /* Unhash it, so that ancestors of killed async unlink 832 * files will be cleaned up during umount */ 833 return 1; 834 } 835 return 0; 836 837 } 838 839 /* 840 * Called when the dentry loses inode. 841 * We use it to clean up silly-renamed files. 842 */ 843 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 844 { 845 nfs_inode_return_delegation(inode); 846 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 847 lock_kernel(); 848 drop_nlink(inode); 849 nfs_complete_unlink(dentry); 850 unlock_kernel(); 851 } 852 /* When creating a negative dentry, we want to renew d_time */ 853 nfs_renew_times(dentry); 854 iput(inode); 855 } 856 857 struct dentry_operations nfs_dentry_operations = { 858 .d_revalidate = nfs_lookup_revalidate, 859 .d_delete = nfs_dentry_delete, 860 .d_iput = nfs_dentry_iput, 861 }; 862 863 /* 864 * Use intent information to check whether or not we're going to do 865 * an O_EXCL create using this path component. 866 */ 867 static inline 868 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) 869 { 870 if (NFS_PROTO(dir)->version == 2) 871 return 0; 872 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0) 873 return 0; 874 return (nd->intent.open.flags & O_EXCL) != 0; 875 } 876 877 static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir, 878 struct nfs_fh *fh, struct nfs_fattr *fattr) 879 { 880 struct nfs_server *server = NFS_SERVER(dir); 881 882 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid)) 883 /* Revalidate fsid on root dir */ 884 return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode); 885 return 0; 886 } 887 888 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 889 { 890 struct dentry *res; 891 struct inode *inode = NULL; 892 int error; 893 struct nfs_fh fhandle; 894 struct nfs_fattr fattr; 895 896 dfprintk(VFS, "NFS: lookup(%s/%s)\n", 897 dentry->d_parent->d_name.name, dentry->d_name.name); 898 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 899 900 res = ERR_PTR(-ENAMETOOLONG); 901 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 902 goto out; 903 904 res = ERR_PTR(-ENOMEM); 905 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 906 907 lock_kernel(); 908 909 /* 910 * If we're doing an exclusive create, optimize away the lookup 911 * but don't hash the dentry. 912 */ 913 if (nfs_is_exclusive_create(dir, nd)) { 914 d_instantiate(dentry, NULL); 915 res = NULL; 916 goto out_unlock; 917 } 918 919 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); 920 if (error == -ENOENT) 921 goto no_entry; 922 if (error < 0) { 923 res = ERR_PTR(error); 924 goto out_unlock; 925 } 926 error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr); 927 if (error < 0) { 928 res = ERR_PTR(error); 929 goto out_unlock; 930 } 931 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); 932 res = (struct dentry *)inode; 933 if (IS_ERR(res)) 934 goto out_unlock; 935 936 no_entry: 937 res = d_materialise_unique(dentry, inode); 938 if (res != NULL) { 939 struct dentry *parent; 940 if (IS_ERR(res)) 941 goto out_unlock; 942 /* Was a directory renamed! */ 943 parent = dget_parent(res); 944 if (!IS_ROOT(parent)) 945 nfs_mark_for_revalidate(parent->d_inode); 946 dput(parent); 947 dentry = res; 948 } 949 nfs_renew_times(dentry); 950 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 951 out_unlock: 952 unlock_kernel(); 953 out: 954 return res; 955 } 956 957 #ifdef CONFIG_NFS_V4 958 static int nfs_open_revalidate(struct dentry *, struct nameidata *); 959 960 struct dentry_operations nfs4_dentry_operations = { 961 .d_revalidate = nfs_open_revalidate, 962 .d_delete = nfs_dentry_delete, 963 .d_iput = nfs_dentry_iput, 964 }; 965 966 /* 967 * Use intent information to determine whether we need to substitute 968 * the NFSv4-style stateful OPEN for the LOOKUP call 969 */ 970 static int is_atomic_open(struct inode *dir, struct nameidata *nd) 971 { 972 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) 973 return 0; 974 /* NFS does not (yet) have a stateful open for directories */ 975 if (nd->flags & LOOKUP_DIRECTORY) 976 return 0; 977 /* Are we trying to write to a read only partition? */ 978 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) 979 return 0; 980 return 1; 981 } 982 983 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 984 { 985 struct dentry *res = NULL; 986 int error; 987 988 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", 989 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 990 991 /* Check that we are indeed trying to open this file */ 992 if (!is_atomic_open(dir, nd)) 993 goto no_open; 994 995 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { 996 res = ERR_PTR(-ENAMETOOLONG); 997 goto out; 998 } 999 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1000 1001 /* Let vfs_create() deal with O_EXCL */ 1002 if (nd->intent.open.flags & O_EXCL) { 1003 d_add(dentry, NULL); 1004 goto out; 1005 } 1006 1007 /* Open the file on the server */ 1008 lock_kernel(); 1009 /* Revalidate parent directory attribute cache */ 1010 error = nfs_revalidate_inode(NFS_SERVER(dir), dir); 1011 if (error < 0) { 1012 res = ERR_PTR(error); 1013 unlock_kernel(); 1014 goto out; 1015 } 1016 1017 if (nd->intent.open.flags & O_CREAT) { 1018 nfs_begin_data_update(dir); 1019 res = nfs4_atomic_open(dir, dentry, nd); 1020 nfs_end_data_update(dir); 1021 } else 1022 res = nfs4_atomic_open(dir, dentry, nd); 1023 unlock_kernel(); 1024 if (IS_ERR(res)) { 1025 error = PTR_ERR(res); 1026 switch (error) { 1027 /* Make a negative dentry */ 1028 case -ENOENT: 1029 res = NULL; 1030 goto out; 1031 /* This turned out not to be a regular file */ 1032 case -EISDIR: 1033 case -ENOTDIR: 1034 goto no_open; 1035 case -ELOOP: 1036 if (!(nd->intent.open.flags & O_NOFOLLOW)) 1037 goto no_open; 1038 /* case -EINVAL: */ 1039 default: 1040 goto out; 1041 } 1042 } else if (res != NULL) 1043 dentry = res; 1044 nfs_renew_times(dentry); 1045 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1046 out: 1047 return res; 1048 no_open: 1049 return nfs_lookup(dir, dentry, nd); 1050 } 1051 1052 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) 1053 { 1054 struct dentry *parent = NULL; 1055 struct inode *inode = dentry->d_inode; 1056 struct inode *dir; 1057 unsigned long verifier; 1058 int openflags, ret = 0; 1059 1060 parent = dget_parent(dentry); 1061 dir = parent->d_inode; 1062 if (!is_atomic_open(dir, nd)) 1063 goto no_open; 1064 /* We can't create new files in nfs_open_revalidate(), so we 1065 * optimize away revalidation of negative dentries. 1066 */ 1067 if (inode == NULL) 1068 goto out; 1069 /* NFS only supports OPEN on regular files */ 1070 if (!S_ISREG(inode->i_mode)) 1071 goto no_open; 1072 openflags = nd->intent.open.flags; 1073 /* We cannot do exclusive creation on a positive dentry */ 1074 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) 1075 goto no_open; 1076 /* We can't create new files, or truncate existing ones here */ 1077 openflags &= ~(O_CREAT|O_TRUNC); 1078 1079 /* 1080 * Note: we're not holding inode->i_mutex and so may be racing with 1081 * operations that change the directory. We therefore save the 1082 * change attribute *before* we do the RPC call. 1083 */ 1084 lock_kernel(); 1085 verifier = nfs_save_change_attribute(dir); 1086 ret = nfs4_open_revalidate(dir, dentry, openflags, nd); 1087 if (!ret) 1088 nfs_set_verifier(dentry, verifier); 1089 unlock_kernel(); 1090 out: 1091 dput(parent); 1092 if (!ret) 1093 d_drop(dentry); 1094 return ret; 1095 no_open: 1096 dput(parent); 1097 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) 1098 return 1; 1099 return nfs_lookup_revalidate(dentry, nd); 1100 } 1101 #endif /* CONFIG_NFSV4 */ 1102 1103 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) 1104 { 1105 struct dentry *parent = desc->file->f_path.dentry; 1106 struct inode *dir = parent->d_inode; 1107 struct nfs_entry *entry = desc->entry; 1108 struct dentry *dentry, *alias; 1109 struct qstr name = { 1110 .name = entry->name, 1111 .len = entry->len, 1112 }; 1113 struct inode *inode; 1114 1115 switch (name.len) { 1116 case 2: 1117 if (name.name[0] == '.' && name.name[1] == '.') 1118 return dget_parent(parent); 1119 break; 1120 case 1: 1121 if (name.name[0] == '.') 1122 return dget(parent); 1123 } 1124 name.hash = full_name_hash(name.name, name.len); 1125 dentry = d_lookup(parent, &name); 1126 if (dentry != NULL) 1127 return dentry; 1128 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) 1129 return NULL; 1130 /* Note: caller is already holding the dir->i_mutex! */ 1131 dentry = d_alloc(parent, &name); 1132 if (dentry == NULL) 1133 return NULL; 1134 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1135 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 1136 if (IS_ERR(inode)) { 1137 dput(dentry); 1138 return NULL; 1139 } 1140 1141 alias = d_materialise_unique(dentry, inode); 1142 if (alias != NULL) { 1143 dput(dentry); 1144 if (IS_ERR(alias)) 1145 return NULL; 1146 dentry = alias; 1147 } 1148 1149 nfs_renew_times(dentry); 1150 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1151 return dentry; 1152 } 1153 1154 /* 1155 * Code common to create, mkdir, and mknod. 1156 */ 1157 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1158 struct nfs_fattr *fattr) 1159 { 1160 struct inode *inode; 1161 int error = -EACCES; 1162 1163 /* We may have been initialized further down */ 1164 if (dentry->d_inode) 1165 return 0; 1166 if (fhandle->size == 0) { 1167 struct inode *dir = dentry->d_parent->d_inode; 1168 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1169 if (error) 1170 return error; 1171 } 1172 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1173 struct nfs_server *server = NFS_SB(dentry->d_sb); 1174 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1175 if (error < 0) 1176 return error; 1177 } 1178 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1179 error = PTR_ERR(inode); 1180 if (IS_ERR(inode)) 1181 return error; 1182 d_instantiate(dentry, inode); 1183 if (d_unhashed(dentry)) 1184 d_rehash(dentry); 1185 return 0; 1186 } 1187 1188 /* 1189 * Following a failed create operation, we drop the dentry rather 1190 * than retain a negative dentry. This avoids a problem in the event 1191 * that the operation succeeded on the server, but an error in the 1192 * reply path made it appear to have failed. 1193 */ 1194 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, 1195 struct nameidata *nd) 1196 { 1197 struct iattr attr; 1198 int error; 1199 int open_flags = 0; 1200 1201 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1202 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1203 1204 attr.ia_mode = mode; 1205 attr.ia_valid = ATTR_MODE; 1206 1207 if (nd && (nd->flags & LOOKUP_CREATE)) 1208 open_flags = nd->intent.open.flags; 1209 1210 lock_kernel(); 1211 nfs_begin_data_update(dir); 1212 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); 1213 nfs_end_data_update(dir); 1214 if (error != 0) 1215 goto out_err; 1216 nfs_renew_times(dentry); 1217 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1218 unlock_kernel(); 1219 return 0; 1220 out_err: 1221 unlock_kernel(); 1222 d_drop(dentry); 1223 return error; 1224 } 1225 1226 /* 1227 * See comments for nfs_proc_create regarding failed operations. 1228 */ 1229 static int 1230 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) 1231 { 1232 struct iattr attr; 1233 int status; 1234 1235 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1236 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1237 1238 if (!new_valid_dev(rdev)) 1239 return -EINVAL; 1240 1241 attr.ia_mode = mode; 1242 attr.ia_valid = ATTR_MODE; 1243 1244 lock_kernel(); 1245 nfs_begin_data_update(dir); 1246 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1247 nfs_end_data_update(dir); 1248 if (status != 0) 1249 goto out_err; 1250 nfs_renew_times(dentry); 1251 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1252 unlock_kernel(); 1253 return 0; 1254 out_err: 1255 unlock_kernel(); 1256 d_drop(dentry); 1257 return status; 1258 } 1259 1260 /* 1261 * See comments for nfs_proc_create regarding failed operations. 1262 */ 1263 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1264 { 1265 struct iattr attr; 1266 int error; 1267 1268 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1269 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1270 1271 attr.ia_valid = ATTR_MODE; 1272 attr.ia_mode = mode | S_IFDIR; 1273 1274 lock_kernel(); 1275 nfs_begin_data_update(dir); 1276 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1277 nfs_end_data_update(dir); 1278 if (error != 0) 1279 goto out_err; 1280 nfs_renew_times(dentry); 1281 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1282 unlock_kernel(); 1283 return 0; 1284 out_err: 1285 d_drop(dentry); 1286 unlock_kernel(); 1287 return error; 1288 } 1289 1290 static int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1291 { 1292 int error; 1293 1294 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1295 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1296 1297 lock_kernel(); 1298 nfs_begin_data_update(dir); 1299 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1300 /* Ensure the VFS deletes this inode */ 1301 if (error == 0 && dentry->d_inode != NULL) 1302 clear_nlink(dentry->d_inode); 1303 nfs_end_data_update(dir); 1304 unlock_kernel(); 1305 1306 return error; 1307 } 1308 1309 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) 1310 { 1311 static unsigned int sillycounter; 1312 const int i_inosize = sizeof(dir->i_ino)*2; 1313 const int countersize = sizeof(sillycounter)*2; 1314 const int slen = sizeof(".nfs") + i_inosize + countersize - 1; 1315 char silly[slen+1]; 1316 struct qstr qsilly; 1317 struct dentry *sdentry; 1318 int error = -EIO; 1319 1320 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", 1321 dentry->d_parent->d_name.name, dentry->d_name.name, 1322 atomic_read(&dentry->d_count)); 1323 nfs_inc_stats(dir, NFSIOS_SILLYRENAME); 1324 1325 #ifdef NFS_PARANOIA 1326 if (!dentry->d_inode) 1327 printk("NFS: silly-renaming %s/%s, negative dentry??\n", 1328 dentry->d_parent->d_name.name, dentry->d_name.name); 1329 #endif 1330 /* 1331 * We don't allow a dentry to be silly-renamed twice. 1332 */ 1333 error = -EBUSY; 1334 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1335 goto out; 1336 1337 sprintf(silly, ".nfs%*.*lx", 1338 i_inosize, i_inosize, dentry->d_inode->i_ino); 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 nfs_begin_data_update(dir); 1366 if (dentry->d_inode) { 1367 nfs_begin_data_update(dentry->d_inode); 1368 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1369 dir, &qsilly); 1370 nfs_mark_for_revalidate(dentry->d_inode); 1371 nfs_end_data_update(dentry->d_inode); 1372 } else 1373 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1374 dir, &qsilly); 1375 nfs_end_data_update(dir); 1376 if (!error) { 1377 nfs_renew_times(dentry); 1378 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1379 d_move(dentry, sdentry); 1380 error = nfs_async_unlink(dentry); 1381 /* If we return 0 we don't unlink */ 1382 } 1383 dput(sdentry); 1384 out: 1385 return error; 1386 } 1387 1388 /* 1389 * Remove a file after making sure there are no pending writes, 1390 * and after checking that the file has only one user. 1391 * 1392 * We invalidate the attribute cache and free the inode prior to the operation 1393 * to avoid possible races if the server reuses the inode. 1394 */ 1395 static int nfs_safe_remove(struct dentry *dentry) 1396 { 1397 struct inode *dir = dentry->d_parent->d_inode; 1398 struct inode *inode = dentry->d_inode; 1399 int error = -EBUSY; 1400 1401 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1402 dentry->d_parent->d_name.name, dentry->d_name.name); 1403 1404 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1405 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1406 error = 0; 1407 goto out; 1408 } 1409 1410 nfs_begin_data_update(dir); 1411 if (inode != NULL) { 1412 nfs_inode_return_delegation(inode); 1413 nfs_begin_data_update(inode); 1414 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1415 /* The VFS may want to delete this inode */ 1416 if (error == 0) 1417 drop_nlink(inode); 1418 nfs_mark_for_revalidate(inode); 1419 nfs_end_data_update(inode); 1420 } else 1421 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1422 nfs_end_data_update(dir); 1423 out: 1424 return error; 1425 } 1426 1427 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1428 * belongs to an active ".nfs..." file and we return -EBUSY. 1429 * 1430 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1431 */ 1432 static int nfs_unlink(struct inode *dir, struct dentry *dentry) 1433 { 1434 int error; 1435 int need_rehash = 0; 1436 1437 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1438 dir->i_ino, dentry->d_name.name); 1439 1440 lock_kernel(); 1441 spin_lock(&dcache_lock); 1442 spin_lock(&dentry->d_lock); 1443 if (atomic_read(&dentry->d_count) > 1) { 1444 spin_unlock(&dentry->d_lock); 1445 spin_unlock(&dcache_lock); 1446 /* Start asynchronous writeout of the inode */ 1447 write_inode_now(dentry->d_inode, 0); 1448 error = nfs_sillyrename(dir, dentry); 1449 unlock_kernel(); 1450 return error; 1451 } 1452 if (!d_unhashed(dentry)) { 1453 __d_drop(dentry); 1454 need_rehash = 1; 1455 } 1456 spin_unlock(&dentry->d_lock); 1457 spin_unlock(&dcache_lock); 1458 error = nfs_safe_remove(dentry); 1459 if (!error) { 1460 nfs_renew_times(dentry); 1461 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1462 } else if (need_rehash) 1463 d_rehash(dentry); 1464 unlock_kernel(); 1465 return error; 1466 } 1467 1468 /* 1469 * To create a symbolic link, most file systems instantiate a new inode, 1470 * add a page to it containing the path, then write it out to the disk 1471 * using prepare_write/commit_write. 1472 * 1473 * Unfortunately the NFS client can't create the in-core inode first 1474 * because it needs a file handle to create an in-core inode (see 1475 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1476 * symlink request has completed on the server. 1477 * 1478 * So instead we allocate a raw page, copy the symname into it, then do 1479 * the SYMLINK request with the page as the buffer. If it succeeds, we 1480 * now have a new file handle and can instantiate an in-core NFS inode 1481 * and move the raw page into its mapping. 1482 */ 1483 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1484 { 1485 struct pagevec lru_pvec; 1486 struct page *page; 1487 char *kaddr; 1488 struct iattr attr; 1489 unsigned int pathlen = strlen(symname); 1490 int error; 1491 1492 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1493 dir->i_ino, dentry->d_name.name, symname); 1494 1495 if (pathlen > PAGE_SIZE) 1496 return -ENAMETOOLONG; 1497 1498 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1499 attr.ia_valid = ATTR_MODE; 1500 1501 lock_kernel(); 1502 1503 page = alloc_page(GFP_KERNEL); 1504 if (!page) { 1505 unlock_kernel(); 1506 return -ENOMEM; 1507 } 1508 1509 kaddr = kmap_atomic(page, KM_USER0); 1510 memcpy(kaddr, symname, pathlen); 1511 if (pathlen < PAGE_SIZE) 1512 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1513 kunmap_atomic(kaddr, KM_USER0); 1514 1515 nfs_begin_data_update(dir); 1516 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1517 nfs_end_data_update(dir); 1518 if (error != 0) { 1519 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1520 dir->i_sb->s_id, dir->i_ino, 1521 dentry->d_name.name, symname, error); 1522 d_drop(dentry); 1523 __free_page(page); 1524 unlock_kernel(); 1525 return error; 1526 } 1527 1528 /* 1529 * No big deal if we can't add this page to the page cache here. 1530 * READLINK will get the missing page from the server if needed. 1531 */ 1532 pagevec_init(&lru_pvec, 0); 1533 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1534 GFP_KERNEL)) { 1535 pagevec_add(&lru_pvec, page); 1536 pagevec_lru_add(&lru_pvec); 1537 SetPageUptodate(page); 1538 unlock_page(page); 1539 } else 1540 __free_page(page); 1541 1542 unlock_kernel(); 1543 return 0; 1544 } 1545 1546 static int 1547 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1548 { 1549 struct inode *inode = old_dentry->d_inode; 1550 int error; 1551 1552 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1553 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1554 dentry->d_parent->d_name.name, dentry->d_name.name); 1555 1556 lock_kernel(); 1557 nfs_begin_data_update(dir); 1558 nfs_begin_data_update(inode); 1559 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1560 if (error == 0) { 1561 atomic_inc(&inode->i_count); 1562 d_instantiate(dentry, inode); 1563 } 1564 nfs_end_data_update(inode); 1565 nfs_end_data_update(dir); 1566 unlock_kernel(); 1567 return error; 1568 } 1569 1570 /* 1571 * RENAME 1572 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1573 * different file handle for the same inode after a rename (e.g. when 1574 * moving to a different directory). A fail-safe method to do so would 1575 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1576 * rename the old file using the sillyrename stuff. This way, the original 1577 * file in old_dir will go away when the last process iput()s the inode. 1578 * 1579 * FIXED. 1580 * 1581 * It actually works quite well. One needs to have the possibility for 1582 * at least one ".nfs..." file in each directory the file ever gets 1583 * moved or linked to which happens automagically with the new 1584 * implementation that only depends on the dcache stuff instead of 1585 * using the inode layer 1586 * 1587 * Unfortunately, things are a little more complicated than indicated 1588 * above. For a cross-directory move, we want to make sure we can get 1589 * rid of the old inode after the operation. This means there must be 1590 * no pending writes (if it's a file), and the use count must be 1. 1591 * If these conditions are met, we can drop the dentries before doing 1592 * the rename. 1593 */ 1594 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1595 struct inode *new_dir, struct dentry *new_dentry) 1596 { 1597 struct inode *old_inode = old_dentry->d_inode; 1598 struct inode *new_inode = new_dentry->d_inode; 1599 struct dentry *dentry = NULL, *rehash = NULL; 1600 int error = -EBUSY; 1601 1602 /* 1603 * To prevent any new references to the target during the rename, 1604 * we unhash the dentry and free the inode in advance. 1605 */ 1606 lock_kernel(); 1607 if (!d_unhashed(new_dentry)) { 1608 d_drop(new_dentry); 1609 rehash = new_dentry; 1610 } 1611 1612 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1613 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1614 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1615 atomic_read(&new_dentry->d_count)); 1616 1617 /* 1618 * First check whether the target is busy ... we can't 1619 * safely do _any_ rename if the target is in use. 1620 * 1621 * For files, make a copy of the dentry and then do a 1622 * silly-rename. If the silly-rename succeeds, the 1623 * copied dentry is hashed and becomes the new target. 1624 */ 1625 if (!new_inode) 1626 goto go_ahead; 1627 if (S_ISDIR(new_inode->i_mode)) { 1628 error = -EISDIR; 1629 if (!S_ISDIR(old_inode->i_mode)) 1630 goto out; 1631 } else if (atomic_read(&new_dentry->d_count) > 2) { 1632 int err; 1633 /* copy the target dentry's name */ 1634 dentry = d_alloc(new_dentry->d_parent, 1635 &new_dentry->d_name); 1636 if (!dentry) 1637 goto out; 1638 1639 /* silly-rename the existing target ... */ 1640 err = nfs_sillyrename(new_dir, new_dentry); 1641 if (!err) { 1642 new_dentry = rehash = dentry; 1643 new_inode = NULL; 1644 /* instantiate the replacement target */ 1645 d_instantiate(new_dentry, NULL); 1646 } else if (atomic_read(&new_dentry->d_count) > 1) { 1647 /* dentry still busy? */ 1648 #ifdef NFS_PARANOIA 1649 printk("nfs_rename: target %s/%s busy, d_count=%d\n", 1650 new_dentry->d_parent->d_name.name, 1651 new_dentry->d_name.name, 1652 atomic_read(&new_dentry->d_count)); 1653 #endif 1654 goto out; 1655 } 1656 } else 1657 drop_nlink(new_inode); 1658 1659 go_ahead: 1660 /* 1661 * ... prune child dentries and writebacks if needed. 1662 */ 1663 if (atomic_read(&old_dentry->d_count) > 1) { 1664 nfs_wb_all(old_inode); 1665 shrink_dcache_parent(old_dentry); 1666 } 1667 nfs_inode_return_delegation(old_inode); 1668 1669 if (new_inode != NULL) { 1670 nfs_inode_return_delegation(new_inode); 1671 d_delete(new_dentry); 1672 } 1673 1674 nfs_begin_data_update(old_dir); 1675 nfs_begin_data_update(new_dir); 1676 nfs_begin_data_update(old_inode); 1677 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1678 new_dir, &new_dentry->d_name); 1679 nfs_mark_for_revalidate(old_inode); 1680 nfs_end_data_update(old_inode); 1681 nfs_end_data_update(new_dir); 1682 nfs_end_data_update(old_dir); 1683 out: 1684 if (rehash) 1685 d_rehash(rehash); 1686 if (!error) { 1687 d_move(old_dentry, new_dentry); 1688 nfs_renew_times(new_dentry); 1689 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir)); 1690 } 1691 1692 /* new dentry created? */ 1693 if (dentry) 1694 dput(dentry); 1695 unlock_kernel(); 1696 return error; 1697 } 1698 1699 static DEFINE_SPINLOCK(nfs_access_lru_lock); 1700 static LIST_HEAD(nfs_access_lru_list); 1701 static atomic_long_t nfs_access_nr_entries; 1702 1703 static void nfs_access_free_entry(struct nfs_access_entry *entry) 1704 { 1705 put_rpccred(entry->cred); 1706 kfree(entry); 1707 smp_mb__before_atomic_dec(); 1708 atomic_long_dec(&nfs_access_nr_entries); 1709 smp_mb__after_atomic_dec(); 1710 } 1711 1712 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) 1713 { 1714 LIST_HEAD(head); 1715 struct nfs_inode *nfsi; 1716 struct nfs_access_entry *cache; 1717 1718 spin_lock(&nfs_access_lru_lock); 1719 restart: 1720 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { 1721 struct inode *inode; 1722 1723 if (nr_to_scan-- == 0) 1724 break; 1725 inode = igrab(&nfsi->vfs_inode); 1726 if (inode == NULL) 1727 continue; 1728 spin_lock(&inode->i_lock); 1729 if (list_empty(&nfsi->access_cache_entry_lru)) 1730 goto remove_lru_entry; 1731 cache = list_entry(nfsi->access_cache_entry_lru.next, 1732 struct nfs_access_entry, lru); 1733 list_move(&cache->lru, &head); 1734 rb_erase(&cache->rb_node, &nfsi->access_cache); 1735 if (!list_empty(&nfsi->access_cache_entry_lru)) 1736 list_move_tail(&nfsi->access_cache_inode_lru, 1737 &nfs_access_lru_list); 1738 else { 1739 remove_lru_entry: 1740 list_del_init(&nfsi->access_cache_inode_lru); 1741 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1742 } 1743 spin_unlock(&inode->i_lock); 1744 iput(inode); 1745 goto restart; 1746 } 1747 spin_unlock(&nfs_access_lru_lock); 1748 while (!list_empty(&head)) { 1749 cache = list_entry(head.next, struct nfs_access_entry, lru); 1750 list_del(&cache->lru); 1751 nfs_access_free_entry(cache); 1752 } 1753 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; 1754 } 1755 1756 static void __nfs_access_zap_cache(struct inode *inode) 1757 { 1758 struct nfs_inode *nfsi = NFS_I(inode); 1759 struct rb_root *root_node = &nfsi->access_cache; 1760 struct rb_node *n, *dispose = NULL; 1761 struct nfs_access_entry *entry; 1762 1763 /* Unhook entries from the cache */ 1764 while ((n = rb_first(root_node)) != NULL) { 1765 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1766 rb_erase(n, root_node); 1767 list_del(&entry->lru); 1768 n->rb_left = dispose; 1769 dispose = n; 1770 } 1771 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 1772 spin_unlock(&inode->i_lock); 1773 1774 /* Now kill them all! */ 1775 while (dispose != NULL) { 1776 n = dispose; 1777 dispose = n->rb_left; 1778 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); 1779 } 1780 } 1781 1782 void nfs_access_zap_cache(struct inode *inode) 1783 { 1784 /* Remove from global LRU init */ 1785 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1786 spin_lock(&nfs_access_lru_lock); 1787 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 1788 spin_unlock(&nfs_access_lru_lock); 1789 } 1790 1791 spin_lock(&inode->i_lock); 1792 /* This will release the spinlock */ 1793 __nfs_access_zap_cache(inode); 1794 } 1795 1796 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 1797 { 1798 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 1799 struct nfs_access_entry *entry; 1800 1801 while (n != NULL) { 1802 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1803 1804 if (cred < entry->cred) 1805 n = n->rb_left; 1806 else if (cred > entry->cred) 1807 n = n->rb_right; 1808 else 1809 return entry; 1810 } 1811 return NULL; 1812 } 1813 1814 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 1815 { 1816 struct nfs_inode *nfsi = NFS_I(inode); 1817 struct nfs_access_entry *cache; 1818 int err = -ENOENT; 1819 1820 spin_lock(&inode->i_lock); 1821 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 1822 goto out_zap; 1823 cache = nfs_access_search_rbtree(inode, cred); 1824 if (cache == NULL) 1825 goto out; 1826 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))) 1827 goto out_stale; 1828 res->jiffies = cache->jiffies; 1829 res->cred = cache->cred; 1830 res->mask = cache->mask; 1831 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 1832 err = 0; 1833 out: 1834 spin_unlock(&inode->i_lock); 1835 return err; 1836 out_stale: 1837 rb_erase(&cache->rb_node, &nfsi->access_cache); 1838 list_del(&cache->lru); 1839 spin_unlock(&inode->i_lock); 1840 nfs_access_free_entry(cache); 1841 return -ENOENT; 1842 out_zap: 1843 /* This will release the spinlock */ 1844 __nfs_access_zap_cache(inode); 1845 return -ENOENT; 1846 } 1847 1848 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 1849 { 1850 struct nfs_inode *nfsi = NFS_I(inode); 1851 struct rb_root *root_node = &nfsi->access_cache; 1852 struct rb_node **p = &root_node->rb_node; 1853 struct rb_node *parent = NULL; 1854 struct nfs_access_entry *entry; 1855 1856 spin_lock(&inode->i_lock); 1857 while (*p != NULL) { 1858 parent = *p; 1859 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 1860 1861 if (set->cred < entry->cred) 1862 p = &parent->rb_left; 1863 else if (set->cred > entry->cred) 1864 p = &parent->rb_right; 1865 else 1866 goto found; 1867 } 1868 rb_link_node(&set->rb_node, parent, p); 1869 rb_insert_color(&set->rb_node, root_node); 1870 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1871 spin_unlock(&inode->i_lock); 1872 return; 1873 found: 1874 rb_replace_node(parent, &set->rb_node, root_node); 1875 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1876 list_del(&entry->lru); 1877 spin_unlock(&inode->i_lock); 1878 nfs_access_free_entry(entry); 1879 } 1880 1881 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 1882 { 1883 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 1884 if (cache == NULL) 1885 return; 1886 RB_CLEAR_NODE(&cache->rb_node); 1887 cache->jiffies = set->jiffies; 1888 cache->cred = get_rpccred(set->cred); 1889 cache->mask = set->mask; 1890 1891 nfs_access_add_rbtree(inode, cache); 1892 1893 /* Update accounting */ 1894 smp_mb__before_atomic_inc(); 1895 atomic_long_inc(&nfs_access_nr_entries); 1896 smp_mb__after_atomic_inc(); 1897 1898 /* Add inode to global LRU list */ 1899 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1900 spin_lock(&nfs_access_lru_lock); 1901 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); 1902 spin_unlock(&nfs_access_lru_lock); 1903 } 1904 } 1905 1906 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 1907 { 1908 struct nfs_access_entry cache; 1909 int status; 1910 1911 status = nfs_access_get_cached(inode, cred, &cache); 1912 if (status == 0) 1913 goto out; 1914 1915 /* Be clever: ask server to check for all possible rights */ 1916 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 1917 cache.cred = cred; 1918 cache.jiffies = jiffies; 1919 status = NFS_PROTO(inode)->access(inode, &cache); 1920 if (status != 0) 1921 return status; 1922 nfs_access_add_cache(inode, &cache); 1923 out: 1924 if ((cache.mask & mask) == mask) 1925 return 0; 1926 return -EACCES; 1927 } 1928 1929 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd) 1930 { 1931 struct rpc_cred *cred; 1932 int res = 0; 1933 1934 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 1935 1936 if (mask == 0) 1937 goto out; 1938 /* Is this sys_access() ? */ 1939 if (nd != NULL && (nd->flags & LOOKUP_ACCESS)) 1940 goto force_lookup; 1941 1942 switch (inode->i_mode & S_IFMT) { 1943 case S_IFLNK: 1944 goto out; 1945 case S_IFREG: 1946 /* NFSv4 has atomic_open... */ 1947 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) 1948 && nd != NULL 1949 && (nd->flags & LOOKUP_OPEN)) 1950 goto out; 1951 break; 1952 case S_IFDIR: 1953 /* 1954 * Optimize away all write operations, since the server 1955 * will check permissions when we perform the op. 1956 */ 1957 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 1958 goto out; 1959 } 1960 1961 force_lookup: 1962 lock_kernel(); 1963 1964 if (!NFS_PROTO(inode)->access) 1965 goto out_notsup; 1966 1967 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 1968 if (!IS_ERR(cred)) { 1969 res = nfs_do_access(inode, cred, mask); 1970 put_rpccred(cred); 1971 } else 1972 res = PTR_ERR(cred); 1973 unlock_kernel(); 1974 out: 1975 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 1976 inode->i_sb->s_id, inode->i_ino, mask, res); 1977 return res; 1978 out_notsup: 1979 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1980 if (res == 0) 1981 res = generic_permission(inode, mask, NULL); 1982 unlock_kernel(); 1983 goto out; 1984 } 1985 1986 /* 1987 * Local variables: 1988 * version-control: t 1989 * kept-new-versions: 5 1990 * End: 1991 */ 1992