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_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_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_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_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_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_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(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_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_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 if (IS_ERR(res)) 940 goto out_unlock; 941 dentry = res; 942 } 943 nfs_renew_times(dentry); 944 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 945 out_unlock: 946 unlock_kernel(); 947 out: 948 return res; 949 } 950 951 #ifdef CONFIG_NFS_V4 952 static int nfs_open_revalidate(struct dentry *, struct nameidata *); 953 954 struct dentry_operations nfs4_dentry_operations = { 955 .d_revalidate = nfs_open_revalidate, 956 .d_delete = nfs_dentry_delete, 957 .d_iput = nfs_dentry_iput, 958 }; 959 960 /* 961 * Use intent information to determine whether we need to substitute 962 * the NFSv4-style stateful OPEN for the LOOKUP call 963 */ 964 static int is_atomic_open(struct inode *dir, struct nameidata *nd) 965 { 966 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) 967 return 0; 968 /* NFS does not (yet) have a stateful open for directories */ 969 if (nd->flags & LOOKUP_DIRECTORY) 970 return 0; 971 /* Are we trying to write to a read only partition? */ 972 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) 973 return 0; 974 return 1; 975 } 976 977 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) 978 { 979 struct dentry *res = NULL; 980 int error; 981 982 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", 983 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 984 985 /* Check that we are indeed trying to open this file */ 986 if (!is_atomic_open(dir, nd)) 987 goto no_open; 988 989 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { 990 res = ERR_PTR(-ENAMETOOLONG); 991 goto out; 992 } 993 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 994 995 /* Let vfs_create() deal with O_EXCL */ 996 if (nd->intent.open.flags & O_EXCL) { 997 d_add(dentry, NULL); 998 goto out; 999 } 1000 1001 /* Open the file on the server */ 1002 lock_kernel(); 1003 /* Revalidate parent directory attribute cache */ 1004 error = nfs_revalidate_inode(NFS_SERVER(dir), dir); 1005 if (error < 0) { 1006 res = ERR_PTR(error); 1007 unlock_kernel(); 1008 goto out; 1009 } 1010 1011 if (nd->intent.open.flags & O_CREAT) { 1012 nfs_begin_data_update(dir); 1013 res = nfs4_atomic_open(dir, dentry, nd); 1014 nfs_end_data_update(dir); 1015 } else 1016 res = nfs4_atomic_open(dir, dentry, nd); 1017 unlock_kernel(); 1018 if (IS_ERR(res)) { 1019 error = PTR_ERR(res); 1020 switch (error) { 1021 /* Make a negative dentry */ 1022 case -ENOENT: 1023 res = NULL; 1024 goto out; 1025 /* This turned out not to be a regular file */ 1026 case -EISDIR: 1027 case -ENOTDIR: 1028 goto no_open; 1029 case -ELOOP: 1030 if (!(nd->intent.open.flags & O_NOFOLLOW)) 1031 goto no_open; 1032 /* case -EINVAL: */ 1033 default: 1034 goto out; 1035 } 1036 } else if (res != NULL) 1037 dentry = res; 1038 nfs_renew_times(dentry); 1039 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1040 out: 1041 return res; 1042 no_open: 1043 return nfs_lookup(dir, dentry, nd); 1044 } 1045 1046 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) 1047 { 1048 struct dentry *parent = NULL; 1049 struct inode *inode = dentry->d_inode; 1050 struct inode *dir; 1051 unsigned long verifier; 1052 int openflags, ret = 0; 1053 1054 parent = dget_parent(dentry); 1055 dir = parent->d_inode; 1056 if (!is_atomic_open(dir, nd)) 1057 goto no_open; 1058 /* We can't create new files in nfs_open_revalidate(), so we 1059 * optimize away revalidation of negative dentries. 1060 */ 1061 if (inode == NULL) 1062 goto out; 1063 /* NFS only supports OPEN on regular files */ 1064 if (!S_ISREG(inode->i_mode)) 1065 goto no_open; 1066 openflags = nd->intent.open.flags; 1067 /* We cannot do exclusive creation on a positive dentry */ 1068 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) 1069 goto no_open; 1070 /* We can't create new files, or truncate existing ones here */ 1071 openflags &= ~(O_CREAT|O_TRUNC); 1072 1073 /* 1074 * Note: we're not holding inode->i_mutex and so may be racing with 1075 * operations that change the directory. We therefore save the 1076 * change attribute *before* we do the RPC call. 1077 */ 1078 lock_kernel(); 1079 verifier = nfs_save_change_attribute(dir); 1080 ret = nfs4_open_revalidate(dir, dentry, openflags, nd); 1081 if (!ret) 1082 nfs_set_verifier(dentry, verifier); 1083 unlock_kernel(); 1084 out: 1085 dput(parent); 1086 if (!ret) 1087 d_drop(dentry); 1088 return ret; 1089 no_open: 1090 dput(parent); 1091 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) 1092 return 1; 1093 return nfs_lookup_revalidate(dentry, nd); 1094 } 1095 #endif /* CONFIG_NFSV4 */ 1096 1097 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) 1098 { 1099 struct dentry *parent = desc->file->f_dentry; 1100 struct inode *dir = parent->d_inode; 1101 struct nfs_entry *entry = desc->entry; 1102 struct dentry *dentry, *alias; 1103 struct qstr name = { 1104 .name = entry->name, 1105 .len = entry->len, 1106 }; 1107 struct inode *inode; 1108 1109 switch (name.len) { 1110 case 2: 1111 if (name.name[0] == '.' && name.name[1] == '.') 1112 return dget_parent(parent); 1113 break; 1114 case 1: 1115 if (name.name[0] == '.') 1116 return dget(parent); 1117 } 1118 name.hash = full_name_hash(name.name, name.len); 1119 dentry = d_lookup(parent, &name); 1120 if (dentry != NULL) 1121 return dentry; 1122 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) 1123 return NULL; 1124 /* Note: caller is already holding the dir->i_mutex! */ 1125 dentry = d_alloc(parent, &name); 1126 if (dentry == NULL) 1127 return NULL; 1128 dentry->d_op = NFS_PROTO(dir)->dentry_ops; 1129 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 1130 if (IS_ERR(inode)) { 1131 dput(dentry); 1132 return NULL; 1133 } 1134 1135 alias = d_materialise_unique(dentry, inode); 1136 if (alias != NULL) { 1137 dput(dentry); 1138 if (IS_ERR(alias)) 1139 return NULL; 1140 dentry = alias; 1141 } 1142 1143 nfs_renew_times(dentry); 1144 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1145 return dentry; 1146 } 1147 1148 /* 1149 * Code common to create, mkdir, and mknod. 1150 */ 1151 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1152 struct nfs_fattr *fattr) 1153 { 1154 struct inode *inode; 1155 int error = -EACCES; 1156 1157 /* We may have been initialized further down */ 1158 if (dentry->d_inode) 1159 return 0; 1160 if (fhandle->size == 0) { 1161 struct inode *dir = dentry->d_parent->d_inode; 1162 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1163 if (error) 1164 return error; 1165 } 1166 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1167 struct nfs_server *server = NFS_SB(dentry->d_sb); 1168 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1169 if (error < 0) 1170 return error; 1171 } 1172 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1173 error = PTR_ERR(inode); 1174 if (IS_ERR(inode)) 1175 return error; 1176 d_instantiate(dentry, inode); 1177 if (d_unhashed(dentry)) 1178 d_rehash(dentry); 1179 return 0; 1180 } 1181 1182 /* 1183 * Following a failed create operation, we drop the dentry rather 1184 * than retain a negative dentry. This avoids a problem in the event 1185 * that the operation succeeded on the server, but an error in the 1186 * reply path made it appear to have failed. 1187 */ 1188 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, 1189 struct nameidata *nd) 1190 { 1191 struct iattr attr; 1192 int error; 1193 int open_flags = 0; 1194 1195 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1196 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1197 1198 attr.ia_mode = mode; 1199 attr.ia_valid = ATTR_MODE; 1200 1201 if (nd && (nd->flags & LOOKUP_CREATE)) 1202 open_flags = nd->intent.open.flags; 1203 1204 lock_kernel(); 1205 nfs_begin_data_update(dir); 1206 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); 1207 nfs_end_data_update(dir); 1208 if (error != 0) 1209 goto out_err; 1210 nfs_renew_times(dentry); 1211 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1212 unlock_kernel(); 1213 return 0; 1214 out_err: 1215 unlock_kernel(); 1216 d_drop(dentry); 1217 return error; 1218 } 1219 1220 /* 1221 * See comments for nfs_proc_create regarding failed operations. 1222 */ 1223 static int 1224 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) 1225 { 1226 struct iattr attr; 1227 int status; 1228 1229 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1230 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1231 1232 if (!new_valid_dev(rdev)) 1233 return -EINVAL; 1234 1235 attr.ia_mode = mode; 1236 attr.ia_valid = ATTR_MODE; 1237 1238 lock_kernel(); 1239 nfs_begin_data_update(dir); 1240 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1241 nfs_end_data_update(dir); 1242 if (status != 0) 1243 goto out_err; 1244 nfs_renew_times(dentry); 1245 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1246 unlock_kernel(); 1247 return 0; 1248 out_err: 1249 unlock_kernel(); 1250 d_drop(dentry); 1251 return status; 1252 } 1253 1254 /* 1255 * See comments for nfs_proc_create regarding failed operations. 1256 */ 1257 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1258 { 1259 struct iattr attr; 1260 int error; 1261 1262 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1263 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1264 1265 attr.ia_valid = ATTR_MODE; 1266 attr.ia_mode = mode | S_IFDIR; 1267 1268 lock_kernel(); 1269 nfs_begin_data_update(dir); 1270 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1271 nfs_end_data_update(dir); 1272 if (error != 0) 1273 goto out_err; 1274 nfs_renew_times(dentry); 1275 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1276 unlock_kernel(); 1277 return 0; 1278 out_err: 1279 d_drop(dentry); 1280 unlock_kernel(); 1281 return error; 1282 } 1283 1284 static int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1285 { 1286 int error; 1287 1288 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1289 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1290 1291 lock_kernel(); 1292 nfs_begin_data_update(dir); 1293 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1294 /* Ensure the VFS deletes this inode */ 1295 if (error == 0 && dentry->d_inode != NULL) 1296 clear_nlink(dentry->d_inode); 1297 nfs_end_data_update(dir); 1298 unlock_kernel(); 1299 1300 return error; 1301 } 1302 1303 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) 1304 { 1305 static unsigned int sillycounter; 1306 const int i_inosize = sizeof(dir->i_ino)*2; 1307 const int countersize = sizeof(sillycounter)*2; 1308 const int slen = sizeof(".nfs") + i_inosize + countersize - 1; 1309 char silly[slen+1]; 1310 struct qstr qsilly; 1311 struct dentry *sdentry; 1312 int error = -EIO; 1313 1314 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", 1315 dentry->d_parent->d_name.name, dentry->d_name.name, 1316 atomic_read(&dentry->d_count)); 1317 nfs_inc_stats(dir, NFSIOS_SILLYRENAME); 1318 1319 #ifdef NFS_PARANOIA 1320 if (!dentry->d_inode) 1321 printk("NFS: silly-renaming %s/%s, negative dentry??\n", 1322 dentry->d_parent->d_name.name, dentry->d_name.name); 1323 #endif 1324 /* 1325 * We don't allow a dentry to be silly-renamed twice. 1326 */ 1327 error = -EBUSY; 1328 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1329 goto out; 1330 1331 sprintf(silly, ".nfs%*.*lx", 1332 i_inosize, i_inosize, dentry->d_inode->i_ino); 1333 1334 /* Return delegation in anticipation of the rename */ 1335 nfs_inode_return_delegation(dentry->d_inode); 1336 1337 sdentry = NULL; 1338 do { 1339 char *suffix = silly + slen - countersize; 1340 1341 dput(sdentry); 1342 sillycounter++; 1343 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); 1344 1345 dfprintk(VFS, "NFS: trying to rename %s to %s\n", 1346 dentry->d_name.name, silly); 1347 1348 sdentry = lookup_one_len(silly, dentry->d_parent, slen); 1349 /* 1350 * N.B. Better to return EBUSY here ... it could be 1351 * dangerous to delete the file while it's in use. 1352 */ 1353 if (IS_ERR(sdentry)) 1354 goto out; 1355 } while(sdentry->d_inode != NULL); /* need negative lookup */ 1356 1357 qsilly.name = silly; 1358 qsilly.len = strlen(silly); 1359 nfs_begin_data_update(dir); 1360 if (dentry->d_inode) { 1361 nfs_begin_data_update(dentry->d_inode); 1362 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1363 dir, &qsilly); 1364 nfs_mark_for_revalidate(dentry->d_inode); 1365 nfs_end_data_update(dentry->d_inode); 1366 } else 1367 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, 1368 dir, &qsilly); 1369 nfs_end_data_update(dir); 1370 if (!error) { 1371 nfs_renew_times(dentry); 1372 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1373 d_move(dentry, sdentry); 1374 error = nfs_async_unlink(dentry); 1375 /* If we return 0 we don't unlink */ 1376 } 1377 dput(sdentry); 1378 out: 1379 return error; 1380 } 1381 1382 /* 1383 * Remove a file after making sure there are no pending writes, 1384 * and after checking that the file has only one user. 1385 * 1386 * We invalidate the attribute cache and free the inode prior to the operation 1387 * to avoid possible races if the server reuses the inode. 1388 */ 1389 static int nfs_safe_remove(struct dentry *dentry) 1390 { 1391 struct inode *dir = dentry->d_parent->d_inode; 1392 struct inode *inode = dentry->d_inode; 1393 int error = -EBUSY; 1394 1395 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1396 dentry->d_parent->d_name.name, dentry->d_name.name); 1397 1398 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1399 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1400 error = 0; 1401 goto out; 1402 } 1403 1404 nfs_begin_data_update(dir); 1405 if (inode != NULL) { 1406 nfs_inode_return_delegation(inode); 1407 nfs_begin_data_update(inode); 1408 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1409 /* The VFS may want to delete this inode */ 1410 if (error == 0) 1411 drop_nlink(inode); 1412 nfs_mark_for_revalidate(inode); 1413 nfs_end_data_update(inode); 1414 } else 1415 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1416 nfs_end_data_update(dir); 1417 out: 1418 return error; 1419 } 1420 1421 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1422 * belongs to an active ".nfs..." file and we return -EBUSY. 1423 * 1424 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1425 */ 1426 static int nfs_unlink(struct inode *dir, struct dentry *dentry) 1427 { 1428 int error; 1429 int need_rehash = 0; 1430 1431 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1432 dir->i_ino, dentry->d_name.name); 1433 1434 lock_kernel(); 1435 spin_lock(&dcache_lock); 1436 spin_lock(&dentry->d_lock); 1437 if (atomic_read(&dentry->d_count) > 1) { 1438 spin_unlock(&dentry->d_lock); 1439 spin_unlock(&dcache_lock); 1440 error = nfs_sillyrename(dir, dentry); 1441 unlock_kernel(); 1442 return error; 1443 } 1444 if (!d_unhashed(dentry)) { 1445 __d_drop(dentry); 1446 need_rehash = 1; 1447 } 1448 spin_unlock(&dentry->d_lock); 1449 spin_unlock(&dcache_lock); 1450 error = nfs_safe_remove(dentry); 1451 if (!error) { 1452 nfs_renew_times(dentry); 1453 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1454 } else if (need_rehash) 1455 d_rehash(dentry); 1456 unlock_kernel(); 1457 return error; 1458 } 1459 1460 /* 1461 * To create a symbolic link, most file systems instantiate a new inode, 1462 * add a page to it containing the path, then write it out to the disk 1463 * using prepare_write/commit_write. 1464 * 1465 * Unfortunately the NFS client can't create the in-core inode first 1466 * because it needs a file handle to create an in-core inode (see 1467 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1468 * symlink request has completed on the server. 1469 * 1470 * So instead we allocate a raw page, copy the symname into it, then do 1471 * the SYMLINK request with the page as the buffer. If it succeeds, we 1472 * now have a new file handle and can instantiate an in-core NFS inode 1473 * and move the raw page into its mapping. 1474 */ 1475 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1476 { 1477 struct pagevec lru_pvec; 1478 struct page *page; 1479 char *kaddr; 1480 struct iattr attr; 1481 unsigned int pathlen = strlen(symname); 1482 int error; 1483 1484 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1485 dir->i_ino, dentry->d_name.name, symname); 1486 1487 if (pathlen > PAGE_SIZE) 1488 return -ENAMETOOLONG; 1489 1490 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1491 attr.ia_valid = ATTR_MODE; 1492 1493 lock_kernel(); 1494 1495 page = alloc_page(GFP_KERNEL); 1496 if (!page) { 1497 unlock_kernel(); 1498 return -ENOMEM; 1499 } 1500 1501 kaddr = kmap_atomic(page, KM_USER0); 1502 memcpy(kaddr, symname, pathlen); 1503 if (pathlen < PAGE_SIZE) 1504 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1505 kunmap_atomic(kaddr, KM_USER0); 1506 1507 nfs_begin_data_update(dir); 1508 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1509 nfs_end_data_update(dir); 1510 if (error != 0) { 1511 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1512 dir->i_sb->s_id, dir->i_ino, 1513 dentry->d_name.name, symname, error); 1514 d_drop(dentry); 1515 __free_page(page); 1516 unlock_kernel(); 1517 return error; 1518 } 1519 1520 /* 1521 * No big deal if we can't add this page to the page cache here. 1522 * READLINK will get the missing page from the server if needed. 1523 */ 1524 pagevec_init(&lru_pvec, 0); 1525 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1526 GFP_KERNEL)) { 1527 pagevec_add(&lru_pvec, page); 1528 pagevec_lru_add(&lru_pvec); 1529 SetPageUptodate(page); 1530 unlock_page(page); 1531 } else 1532 __free_page(page); 1533 1534 unlock_kernel(); 1535 return 0; 1536 } 1537 1538 static int 1539 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1540 { 1541 struct inode *inode = old_dentry->d_inode; 1542 int error; 1543 1544 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1545 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1546 dentry->d_parent->d_name.name, dentry->d_name.name); 1547 1548 lock_kernel(); 1549 nfs_begin_data_update(dir); 1550 nfs_begin_data_update(inode); 1551 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1552 if (error == 0) { 1553 atomic_inc(&inode->i_count); 1554 d_instantiate(dentry, inode); 1555 } 1556 nfs_end_data_update(inode); 1557 nfs_end_data_update(dir); 1558 unlock_kernel(); 1559 return error; 1560 } 1561 1562 /* 1563 * RENAME 1564 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1565 * different file handle for the same inode after a rename (e.g. when 1566 * moving to a different directory). A fail-safe method to do so would 1567 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1568 * rename the old file using the sillyrename stuff. This way, the original 1569 * file in old_dir will go away when the last process iput()s the inode. 1570 * 1571 * FIXED. 1572 * 1573 * It actually works quite well. One needs to have the possibility for 1574 * at least one ".nfs..." file in each directory the file ever gets 1575 * moved or linked to which happens automagically with the new 1576 * implementation that only depends on the dcache stuff instead of 1577 * using the inode layer 1578 * 1579 * Unfortunately, things are a little more complicated than indicated 1580 * above. For a cross-directory move, we want to make sure we can get 1581 * rid of the old inode after the operation. This means there must be 1582 * no pending writes (if it's a file), and the use count must be 1. 1583 * If these conditions are met, we can drop the dentries before doing 1584 * the rename. 1585 */ 1586 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1587 struct inode *new_dir, struct dentry *new_dentry) 1588 { 1589 struct inode *old_inode = old_dentry->d_inode; 1590 struct inode *new_inode = new_dentry->d_inode; 1591 struct dentry *dentry = NULL, *rehash = NULL; 1592 int error = -EBUSY; 1593 1594 /* 1595 * To prevent any new references to the target during the rename, 1596 * we unhash the dentry and free the inode in advance. 1597 */ 1598 lock_kernel(); 1599 if (!d_unhashed(new_dentry)) { 1600 d_drop(new_dentry); 1601 rehash = new_dentry; 1602 } 1603 1604 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1605 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1606 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1607 atomic_read(&new_dentry->d_count)); 1608 1609 /* 1610 * First check whether the target is busy ... we can't 1611 * safely do _any_ rename if the target is in use. 1612 * 1613 * For files, make a copy of the dentry and then do a 1614 * silly-rename. If the silly-rename succeeds, the 1615 * copied dentry is hashed and becomes the new target. 1616 */ 1617 if (!new_inode) 1618 goto go_ahead; 1619 if (S_ISDIR(new_inode->i_mode)) { 1620 error = -EISDIR; 1621 if (!S_ISDIR(old_inode->i_mode)) 1622 goto out; 1623 } else if (atomic_read(&new_dentry->d_count) > 2) { 1624 int err; 1625 /* copy the target dentry's name */ 1626 dentry = d_alloc(new_dentry->d_parent, 1627 &new_dentry->d_name); 1628 if (!dentry) 1629 goto out; 1630 1631 /* silly-rename the existing target ... */ 1632 err = nfs_sillyrename(new_dir, new_dentry); 1633 if (!err) { 1634 new_dentry = rehash = dentry; 1635 new_inode = NULL; 1636 /* instantiate the replacement target */ 1637 d_instantiate(new_dentry, NULL); 1638 } else if (atomic_read(&new_dentry->d_count) > 1) { 1639 /* dentry still busy? */ 1640 #ifdef NFS_PARANOIA 1641 printk("nfs_rename: target %s/%s busy, d_count=%d\n", 1642 new_dentry->d_parent->d_name.name, 1643 new_dentry->d_name.name, 1644 atomic_read(&new_dentry->d_count)); 1645 #endif 1646 goto out; 1647 } 1648 } else 1649 drop_nlink(new_inode); 1650 1651 go_ahead: 1652 /* 1653 * ... prune child dentries and writebacks if needed. 1654 */ 1655 if (atomic_read(&old_dentry->d_count) > 1) { 1656 nfs_wb_all(old_inode); 1657 shrink_dcache_parent(old_dentry); 1658 } 1659 nfs_inode_return_delegation(old_inode); 1660 1661 if (new_inode != NULL) { 1662 nfs_inode_return_delegation(new_inode); 1663 d_delete(new_dentry); 1664 } 1665 1666 nfs_begin_data_update(old_dir); 1667 nfs_begin_data_update(new_dir); 1668 nfs_begin_data_update(old_inode); 1669 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1670 new_dir, &new_dentry->d_name); 1671 nfs_mark_for_revalidate(old_inode); 1672 nfs_end_data_update(old_inode); 1673 nfs_end_data_update(new_dir); 1674 nfs_end_data_update(old_dir); 1675 out: 1676 if (rehash) 1677 d_rehash(rehash); 1678 if (!error) { 1679 d_move(old_dentry, new_dentry); 1680 nfs_renew_times(new_dentry); 1681 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir)); 1682 } 1683 1684 /* new dentry created? */ 1685 if (dentry) 1686 dput(dentry); 1687 unlock_kernel(); 1688 return error; 1689 } 1690 1691 static DEFINE_SPINLOCK(nfs_access_lru_lock); 1692 static LIST_HEAD(nfs_access_lru_list); 1693 static atomic_long_t nfs_access_nr_entries; 1694 1695 static void nfs_access_free_entry(struct nfs_access_entry *entry) 1696 { 1697 put_rpccred(entry->cred); 1698 kfree(entry); 1699 smp_mb__before_atomic_dec(); 1700 atomic_long_dec(&nfs_access_nr_entries); 1701 smp_mb__after_atomic_dec(); 1702 } 1703 1704 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) 1705 { 1706 LIST_HEAD(head); 1707 struct nfs_inode *nfsi; 1708 struct nfs_access_entry *cache; 1709 1710 spin_lock(&nfs_access_lru_lock); 1711 restart: 1712 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { 1713 struct inode *inode; 1714 1715 if (nr_to_scan-- == 0) 1716 break; 1717 inode = igrab(&nfsi->vfs_inode); 1718 if (inode == NULL) 1719 continue; 1720 spin_lock(&inode->i_lock); 1721 if (list_empty(&nfsi->access_cache_entry_lru)) 1722 goto remove_lru_entry; 1723 cache = list_entry(nfsi->access_cache_entry_lru.next, 1724 struct nfs_access_entry, lru); 1725 list_move(&cache->lru, &head); 1726 rb_erase(&cache->rb_node, &nfsi->access_cache); 1727 if (!list_empty(&nfsi->access_cache_entry_lru)) 1728 list_move_tail(&nfsi->access_cache_inode_lru, 1729 &nfs_access_lru_list); 1730 else { 1731 remove_lru_entry: 1732 list_del_init(&nfsi->access_cache_inode_lru); 1733 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1734 } 1735 spin_unlock(&inode->i_lock); 1736 iput(inode); 1737 goto restart; 1738 } 1739 spin_unlock(&nfs_access_lru_lock); 1740 while (!list_empty(&head)) { 1741 cache = list_entry(head.next, struct nfs_access_entry, lru); 1742 list_del(&cache->lru); 1743 nfs_access_free_entry(cache); 1744 } 1745 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; 1746 } 1747 1748 static void __nfs_access_zap_cache(struct inode *inode) 1749 { 1750 struct nfs_inode *nfsi = NFS_I(inode); 1751 struct rb_root *root_node = &nfsi->access_cache; 1752 struct rb_node *n, *dispose = NULL; 1753 struct nfs_access_entry *entry; 1754 1755 /* Unhook entries from the cache */ 1756 while ((n = rb_first(root_node)) != NULL) { 1757 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1758 rb_erase(n, root_node); 1759 list_del(&entry->lru); 1760 n->rb_left = dispose; 1761 dispose = n; 1762 } 1763 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 1764 spin_unlock(&inode->i_lock); 1765 1766 /* Now kill them all! */ 1767 while (dispose != NULL) { 1768 n = dispose; 1769 dispose = n->rb_left; 1770 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); 1771 } 1772 } 1773 1774 void nfs_access_zap_cache(struct inode *inode) 1775 { 1776 /* Remove from global LRU init */ 1777 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1778 spin_lock(&nfs_access_lru_lock); 1779 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 1780 spin_unlock(&nfs_access_lru_lock); 1781 } 1782 1783 spin_lock(&inode->i_lock); 1784 /* This will release the spinlock */ 1785 __nfs_access_zap_cache(inode); 1786 } 1787 1788 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 1789 { 1790 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 1791 struct nfs_access_entry *entry; 1792 1793 while (n != NULL) { 1794 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1795 1796 if (cred < entry->cred) 1797 n = n->rb_left; 1798 else if (cred > entry->cred) 1799 n = n->rb_right; 1800 else 1801 return entry; 1802 } 1803 return NULL; 1804 } 1805 1806 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 1807 { 1808 struct nfs_inode *nfsi = NFS_I(inode); 1809 struct nfs_access_entry *cache; 1810 int err = -ENOENT; 1811 1812 spin_lock(&inode->i_lock); 1813 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 1814 goto out_zap; 1815 cache = nfs_access_search_rbtree(inode, cred); 1816 if (cache == NULL) 1817 goto out; 1818 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))) 1819 goto out_stale; 1820 res->jiffies = cache->jiffies; 1821 res->cred = cache->cred; 1822 res->mask = cache->mask; 1823 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 1824 err = 0; 1825 out: 1826 spin_unlock(&inode->i_lock); 1827 return err; 1828 out_stale: 1829 rb_erase(&cache->rb_node, &nfsi->access_cache); 1830 list_del(&cache->lru); 1831 spin_unlock(&inode->i_lock); 1832 nfs_access_free_entry(cache); 1833 return -ENOENT; 1834 out_zap: 1835 /* This will release the spinlock */ 1836 __nfs_access_zap_cache(inode); 1837 return -ENOENT; 1838 } 1839 1840 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 1841 { 1842 struct nfs_inode *nfsi = NFS_I(inode); 1843 struct rb_root *root_node = &nfsi->access_cache; 1844 struct rb_node **p = &root_node->rb_node; 1845 struct rb_node *parent = NULL; 1846 struct nfs_access_entry *entry; 1847 1848 spin_lock(&inode->i_lock); 1849 while (*p != NULL) { 1850 parent = *p; 1851 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 1852 1853 if (set->cred < entry->cred) 1854 p = &parent->rb_left; 1855 else if (set->cred > entry->cred) 1856 p = &parent->rb_right; 1857 else 1858 goto found; 1859 } 1860 rb_link_node(&set->rb_node, parent, p); 1861 rb_insert_color(&set->rb_node, root_node); 1862 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1863 spin_unlock(&inode->i_lock); 1864 return; 1865 found: 1866 rb_replace_node(parent, &set->rb_node, root_node); 1867 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 1868 list_del(&entry->lru); 1869 spin_unlock(&inode->i_lock); 1870 nfs_access_free_entry(entry); 1871 } 1872 1873 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 1874 { 1875 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 1876 if (cache == NULL) 1877 return; 1878 RB_CLEAR_NODE(&cache->rb_node); 1879 cache->jiffies = set->jiffies; 1880 cache->cred = get_rpccred(set->cred); 1881 cache->mask = set->mask; 1882 1883 nfs_access_add_rbtree(inode, cache); 1884 1885 /* Update accounting */ 1886 smp_mb__before_atomic_inc(); 1887 atomic_long_inc(&nfs_access_nr_entries); 1888 smp_mb__after_atomic_inc(); 1889 1890 /* Add inode to global LRU list */ 1891 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { 1892 spin_lock(&nfs_access_lru_lock); 1893 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); 1894 spin_unlock(&nfs_access_lru_lock); 1895 } 1896 } 1897 1898 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 1899 { 1900 struct nfs_access_entry cache; 1901 int status; 1902 1903 status = nfs_access_get_cached(inode, cred, &cache); 1904 if (status == 0) 1905 goto out; 1906 1907 /* Be clever: ask server to check for all possible rights */ 1908 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 1909 cache.cred = cred; 1910 cache.jiffies = jiffies; 1911 status = NFS_PROTO(inode)->access(inode, &cache); 1912 if (status != 0) 1913 return status; 1914 nfs_access_add_cache(inode, &cache); 1915 out: 1916 if ((cache.mask & mask) == mask) 1917 return 0; 1918 return -EACCES; 1919 } 1920 1921 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd) 1922 { 1923 struct rpc_cred *cred; 1924 int res = 0; 1925 1926 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 1927 1928 if (mask == 0) 1929 goto out; 1930 /* Is this sys_access() ? */ 1931 if (nd != NULL && (nd->flags & LOOKUP_ACCESS)) 1932 goto force_lookup; 1933 1934 switch (inode->i_mode & S_IFMT) { 1935 case S_IFLNK: 1936 goto out; 1937 case S_IFREG: 1938 /* NFSv4 has atomic_open... */ 1939 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) 1940 && nd != NULL 1941 && (nd->flags & LOOKUP_OPEN)) 1942 goto out; 1943 break; 1944 case S_IFDIR: 1945 /* 1946 * Optimize away all write operations, since the server 1947 * will check permissions when we perform the op. 1948 */ 1949 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 1950 goto out; 1951 } 1952 1953 force_lookup: 1954 lock_kernel(); 1955 1956 if (!NFS_PROTO(inode)->access) 1957 goto out_notsup; 1958 1959 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); 1960 if (!IS_ERR(cred)) { 1961 res = nfs_do_access(inode, cred, mask); 1962 put_rpccred(cred); 1963 } else 1964 res = PTR_ERR(cred); 1965 unlock_kernel(); 1966 out: 1967 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 1968 inode->i_sb->s_id, inode->i_ino, mask, res); 1969 return res; 1970 out_notsup: 1971 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1972 if (res == 0) 1973 res = generic_permission(inode, mask, NULL); 1974 unlock_kernel(); 1975 goto out; 1976 } 1977 1978 /* 1979 * Local variables: 1980 * version-control: t 1981 * kept-new-versions: 5 1982 * End: 1983 */ 1984