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/module.h> 21 #include <linux/time.h> 22 #include <linux/errno.h> 23 #include <linux/stat.h> 24 #include <linux/fcntl.h> 25 #include <linux/string.h> 26 #include <linux/kernel.h> 27 #include <linux/slab.h> 28 #include <linux/mm.h> 29 #include <linux/sunrpc/clnt.h> 30 #include <linux/nfs_fs.h> 31 #include <linux/nfs_mount.h> 32 #include <linux/pagemap.h> 33 #include <linux/pagevec.h> 34 #include <linux/namei.h> 35 #include <linux/mount.h> 36 #include <linux/sched.h> 37 #include <linux/kmemleak.h> 38 #include <linux/xattr.h> 39 40 #include "delegation.h" 41 #include "iostat.h" 42 #include "internal.h" 43 #include "fscache.h" 44 45 /* #define NFS_DEBUG_VERBOSE 1 */ 46 47 static int nfs_opendir(struct inode *, struct file *); 48 static int nfs_closedir(struct inode *, struct file *); 49 static int nfs_readdir(struct file *, void *, filldir_t); 50 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); 51 static loff_t nfs_llseek_dir(struct file *, loff_t, int); 52 static void nfs_readdir_clear_array(struct page*); 53 54 const struct file_operations nfs_dir_operations = { 55 .llseek = nfs_llseek_dir, 56 .read = generic_read_dir, 57 .readdir = nfs_readdir, 58 .open = nfs_opendir, 59 .release = nfs_closedir, 60 .fsync = nfs_fsync_dir, 61 }; 62 63 const struct address_space_operations nfs_dir_aops = { 64 .freepage = nfs_readdir_clear_array, 65 }; 66 67 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred) 68 { 69 struct nfs_open_dir_context *ctx; 70 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 71 if (ctx != NULL) { 72 ctx->duped = 0; 73 ctx->attr_gencount = NFS_I(dir)->attr_gencount; 74 ctx->dir_cookie = 0; 75 ctx->dup_cookie = 0; 76 ctx->cred = get_rpccred(cred); 77 return ctx; 78 } 79 return ERR_PTR(-ENOMEM); 80 } 81 82 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx) 83 { 84 put_rpccred(ctx->cred); 85 kfree(ctx); 86 } 87 88 /* 89 * Open file 90 */ 91 static int 92 nfs_opendir(struct inode *inode, struct file *filp) 93 { 94 int res = 0; 95 struct nfs_open_dir_context *ctx; 96 struct rpc_cred *cred; 97 98 dfprintk(FILE, "NFS: open dir(%s/%s)\n", 99 filp->f_path.dentry->d_parent->d_name.name, 100 filp->f_path.dentry->d_name.name); 101 102 nfs_inc_stats(inode, NFSIOS_VFSOPEN); 103 104 cred = rpc_lookup_cred(); 105 if (IS_ERR(cred)) 106 return PTR_ERR(cred); 107 ctx = alloc_nfs_open_dir_context(inode, cred); 108 if (IS_ERR(ctx)) { 109 res = PTR_ERR(ctx); 110 goto out; 111 } 112 filp->private_data = ctx; 113 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { 114 /* This is a mountpoint, so d_revalidate will never 115 * have been called, so we need to refresh the 116 * inode (for close-open consistency) ourselves. 117 */ 118 __nfs_revalidate_inode(NFS_SERVER(inode), inode); 119 } 120 out: 121 put_rpccred(cred); 122 return res; 123 } 124 125 static int 126 nfs_closedir(struct inode *inode, struct file *filp) 127 { 128 put_nfs_open_dir_context(filp->private_data); 129 return 0; 130 } 131 132 struct nfs_cache_array_entry { 133 u64 cookie; 134 u64 ino; 135 struct qstr string; 136 unsigned char d_type; 137 }; 138 139 struct nfs_cache_array { 140 int size; 141 int eof_index; 142 u64 last_cookie; 143 struct nfs_cache_array_entry array[0]; 144 }; 145 146 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int); 147 typedef struct { 148 struct file *file; 149 struct page *page; 150 unsigned long page_index; 151 u64 *dir_cookie; 152 u64 last_cookie; 153 loff_t current_index; 154 decode_dirent_t decode; 155 156 unsigned long timestamp; 157 unsigned long gencount; 158 unsigned int cache_entry_index; 159 unsigned int plus:1; 160 unsigned int eof:1; 161 } nfs_readdir_descriptor_t; 162 163 /* 164 * The caller is responsible for calling nfs_readdir_release_array(page) 165 */ 166 static 167 struct nfs_cache_array *nfs_readdir_get_array(struct page *page) 168 { 169 void *ptr; 170 if (page == NULL) 171 return ERR_PTR(-EIO); 172 ptr = kmap(page); 173 if (ptr == NULL) 174 return ERR_PTR(-ENOMEM); 175 return ptr; 176 } 177 178 static 179 void nfs_readdir_release_array(struct page *page) 180 { 181 kunmap(page); 182 } 183 184 /* 185 * we are freeing strings created by nfs_add_to_readdir_array() 186 */ 187 static 188 void nfs_readdir_clear_array(struct page *page) 189 { 190 struct nfs_cache_array *array; 191 int i; 192 193 array = kmap_atomic(page); 194 for (i = 0; i < array->size; i++) 195 kfree(array->array[i].string.name); 196 kunmap_atomic(array); 197 } 198 199 /* 200 * the caller is responsible for freeing qstr.name 201 * when called by nfs_readdir_add_to_array, the strings will be freed in 202 * nfs_clear_readdir_array() 203 */ 204 static 205 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) 206 { 207 string->len = len; 208 string->name = kmemdup(name, len, GFP_KERNEL); 209 if (string->name == NULL) 210 return -ENOMEM; 211 /* 212 * Avoid a kmemleak false positive. The pointer to the name is stored 213 * in a page cache page which kmemleak does not scan. 214 */ 215 kmemleak_not_leak(string->name); 216 string->hash = full_name_hash(name, len); 217 return 0; 218 } 219 220 static 221 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 222 { 223 struct nfs_cache_array *array = nfs_readdir_get_array(page); 224 struct nfs_cache_array_entry *cache_entry; 225 int ret; 226 227 if (IS_ERR(array)) 228 return PTR_ERR(array); 229 230 cache_entry = &array->array[array->size]; 231 232 /* Check that this entry lies within the page bounds */ 233 ret = -ENOSPC; 234 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) 235 goto out; 236 237 cache_entry->cookie = entry->prev_cookie; 238 cache_entry->ino = entry->ino; 239 cache_entry->d_type = entry->d_type; 240 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); 241 if (ret) 242 goto out; 243 array->last_cookie = entry->cookie; 244 array->size++; 245 if (entry->eof != 0) 246 array->eof_index = array->size; 247 out: 248 nfs_readdir_release_array(page); 249 return ret; 250 } 251 252 static 253 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 254 { 255 loff_t diff = desc->file->f_pos - desc->current_index; 256 unsigned int index; 257 258 if (diff < 0) 259 goto out_eof; 260 if (diff >= array->size) { 261 if (array->eof_index >= 0) 262 goto out_eof; 263 return -EAGAIN; 264 } 265 266 index = (unsigned int)diff; 267 *desc->dir_cookie = array->array[index].cookie; 268 desc->cache_entry_index = index; 269 return 0; 270 out_eof: 271 desc->eof = 1; 272 return -EBADCOOKIE; 273 } 274 275 static 276 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 277 { 278 int i; 279 loff_t new_pos; 280 int status = -EAGAIN; 281 282 for (i = 0; i < array->size; i++) { 283 if (array->array[i].cookie == *desc->dir_cookie) { 284 struct nfs_inode *nfsi = NFS_I(desc->file->f_path.dentry->d_inode); 285 struct nfs_open_dir_context *ctx = desc->file->private_data; 286 287 new_pos = desc->current_index + i; 288 if (ctx->attr_gencount != nfsi->attr_gencount 289 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) { 290 ctx->duped = 0; 291 ctx->attr_gencount = nfsi->attr_gencount; 292 } else if (new_pos < desc->file->f_pos) { 293 if (ctx->duped > 0 294 && ctx->dup_cookie == *desc->dir_cookie) { 295 if (printk_ratelimit()) { 296 pr_notice("NFS: directory %s/%s contains a readdir loop." 297 "Please contact your server vendor. " 298 "The file: %s has duplicate cookie %llu\n", 299 desc->file->f_dentry->d_parent->d_name.name, 300 desc->file->f_dentry->d_name.name, 301 array->array[i].string.name, 302 *desc->dir_cookie); 303 } 304 status = -ELOOP; 305 goto out; 306 } 307 ctx->dup_cookie = *desc->dir_cookie; 308 ctx->duped = -1; 309 } 310 desc->file->f_pos = new_pos; 311 desc->cache_entry_index = i; 312 return 0; 313 } 314 } 315 if (array->eof_index >= 0) { 316 status = -EBADCOOKIE; 317 if (*desc->dir_cookie == array->last_cookie) 318 desc->eof = 1; 319 } 320 out: 321 return status; 322 } 323 324 static 325 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) 326 { 327 struct nfs_cache_array *array; 328 int status; 329 330 array = nfs_readdir_get_array(desc->page); 331 if (IS_ERR(array)) { 332 status = PTR_ERR(array); 333 goto out; 334 } 335 336 if (*desc->dir_cookie == 0) 337 status = nfs_readdir_search_for_pos(array, desc); 338 else 339 status = nfs_readdir_search_for_cookie(array, desc); 340 341 if (status == -EAGAIN) { 342 desc->last_cookie = array->last_cookie; 343 desc->current_index += array->size; 344 desc->page_index++; 345 } 346 nfs_readdir_release_array(desc->page); 347 out: 348 return status; 349 } 350 351 /* Fill a page with xdr information before transferring to the cache page */ 352 static 353 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, 354 struct nfs_entry *entry, struct file *file, struct inode *inode) 355 { 356 struct nfs_open_dir_context *ctx = file->private_data; 357 struct rpc_cred *cred = ctx->cred; 358 unsigned long timestamp, gencount; 359 int error; 360 361 again: 362 timestamp = jiffies; 363 gencount = nfs_inc_attr_generation_counter(); 364 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages, 365 NFS_SERVER(inode)->dtsize, desc->plus); 366 if (error < 0) { 367 /* We requested READDIRPLUS, but the server doesn't grok it */ 368 if (error == -ENOTSUPP && desc->plus) { 369 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 370 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 371 desc->plus = 0; 372 goto again; 373 } 374 goto error; 375 } 376 desc->timestamp = timestamp; 377 desc->gencount = gencount; 378 error: 379 return error; 380 } 381 382 static int xdr_decode(nfs_readdir_descriptor_t *desc, 383 struct nfs_entry *entry, struct xdr_stream *xdr) 384 { 385 int error; 386 387 error = desc->decode(xdr, entry, desc->plus); 388 if (error) 389 return error; 390 entry->fattr->time_start = desc->timestamp; 391 entry->fattr->gencount = desc->gencount; 392 return 0; 393 } 394 395 static 396 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 397 { 398 if (dentry->d_inode == NULL) 399 goto different; 400 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0) 401 goto different; 402 return 1; 403 different: 404 return 0; 405 } 406 407 static 408 bool nfs_use_readdirplus(struct inode *dir, struct file *filp) 409 { 410 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 411 return false; 412 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 413 return true; 414 if (filp->f_pos == 0) 415 return true; 416 return false; 417 } 418 419 /* 420 * This function is called by the lookup code to request the use of 421 * readdirplus to accelerate any future lookups in the same 422 * directory. 423 */ 424 static 425 void nfs_advise_use_readdirplus(struct inode *dir) 426 { 427 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags); 428 } 429 430 static 431 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) 432 { 433 struct qstr filename = QSTR_INIT(entry->name, entry->len); 434 struct dentry *dentry; 435 struct dentry *alias; 436 struct inode *dir = parent->d_inode; 437 struct inode *inode; 438 439 if (filename.name[0] == '.') { 440 if (filename.len == 1) 441 return; 442 if (filename.len == 2 && filename.name[1] == '.') 443 return; 444 } 445 filename.hash = full_name_hash(filename.name, filename.len); 446 447 dentry = d_lookup(parent, &filename); 448 if (dentry != NULL) { 449 if (nfs_same_file(dentry, entry)) { 450 nfs_refresh_inode(dentry->d_inode, entry->fattr); 451 goto out; 452 } else { 453 if (d_invalidate(dentry) != 0) 454 goto out; 455 dput(dentry); 456 } 457 } 458 459 dentry = d_alloc(parent, &filename); 460 if (dentry == NULL) 461 return; 462 463 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 464 if (IS_ERR(inode)) 465 goto out; 466 467 alias = d_materialise_unique(dentry, inode); 468 if (IS_ERR(alias)) 469 goto out; 470 else if (alias) { 471 nfs_set_verifier(alias, nfs_save_change_attribute(dir)); 472 dput(alias); 473 } else 474 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 475 476 out: 477 dput(dentry); 478 } 479 480 /* Perform conversion from xdr to cache array */ 481 static 482 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 483 struct page **xdr_pages, struct page *page, unsigned int buflen) 484 { 485 struct xdr_stream stream; 486 struct xdr_buf buf; 487 struct page *scratch; 488 struct nfs_cache_array *array; 489 unsigned int count = 0; 490 int status; 491 492 scratch = alloc_page(GFP_KERNEL); 493 if (scratch == NULL) 494 return -ENOMEM; 495 496 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 497 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 498 499 do { 500 status = xdr_decode(desc, entry, &stream); 501 if (status != 0) { 502 if (status == -EAGAIN) 503 status = 0; 504 break; 505 } 506 507 count++; 508 509 if (desc->plus != 0) 510 nfs_prime_dcache(desc->file->f_path.dentry, entry); 511 512 status = nfs_readdir_add_to_array(entry, page); 513 if (status != 0) 514 break; 515 } while (!entry->eof); 516 517 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { 518 array = nfs_readdir_get_array(page); 519 if (!IS_ERR(array)) { 520 array->eof_index = array->size; 521 status = 0; 522 nfs_readdir_release_array(page); 523 } else 524 status = PTR_ERR(array); 525 } 526 527 put_page(scratch); 528 return status; 529 } 530 531 static 532 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) 533 { 534 unsigned int i; 535 for (i = 0; i < npages; i++) 536 put_page(pages[i]); 537 } 538 539 static 540 void nfs_readdir_free_large_page(void *ptr, struct page **pages, 541 unsigned int npages) 542 { 543 nfs_readdir_free_pagearray(pages, npages); 544 } 545 546 /* 547 * nfs_readdir_large_page will allocate pages that must be freed with a call 548 * to nfs_readdir_free_large_page 549 */ 550 static 551 int nfs_readdir_large_page(struct page **pages, unsigned int npages) 552 { 553 unsigned int i; 554 555 for (i = 0; i < npages; i++) { 556 struct page *page = alloc_page(GFP_KERNEL); 557 if (page == NULL) 558 goto out_freepages; 559 pages[i] = page; 560 } 561 return 0; 562 563 out_freepages: 564 nfs_readdir_free_pagearray(pages, i); 565 return -ENOMEM; 566 } 567 568 static 569 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) 570 { 571 struct page *pages[NFS_MAX_READDIR_PAGES]; 572 void *pages_ptr = NULL; 573 struct nfs_entry entry; 574 struct file *file = desc->file; 575 struct nfs_cache_array *array; 576 int status = -ENOMEM; 577 unsigned int array_size = ARRAY_SIZE(pages); 578 579 entry.prev_cookie = 0; 580 entry.cookie = desc->last_cookie; 581 entry.eof = 0; 582 entry.fh = nfs_alloc_fhandle(); 583 entry.fattr = nfs_alloc_fattr(); 584 entry.server = NFS_SERVER(inode); 585 if (entry.fh == NULL || entry.fattr == NULL) 586 goto out; 587 588 array = nfs_readdir_get_array(page); 589 if (IS_ERR(array)) { 590 status = PTR_ERR(array); 591 goto out; 592 } 593 memset(array, 0, sizeof(struct nfs_cache_array)); 594 array->eof_index = -1; 595 596 status = nfs_readdir_large_page(pages, array_size); 597 if (status < 0) 598 goto out_release_array; 599 do { 600 unsigned int pglen; 601 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 602 603 if (status < 0) 604 break; 605 pglen = status; 606 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 607 if (status < 0) { 608 if (status == -ENOSPC) 609 status = 0; 610 break; 611 } 612 } while (array->eof_index < 0); 613 614 nfs_readdir_free_large_page(pages_ptr, pages, array_size); 615 out_release_array: 616 nfs_readdir_release_array(page); 617 out: 618 nfs_free_fattr(entry.fattr); 619 nfs_free_fhandle(entry.fh); 620 return status; 621 } 622 623 /* 624 * Now we cache directories properly, by converting xdr information 625 * to an array that can be used for lookups later. This results in 626 * fewer cache pages, since we can store more information on each page. 627 * We only need to convert from xdr once so future lookups are much simpler 628 */ 629 static 630 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) 631 { 632 struct inode *inode = desc->file->f_path.dentry->d_inode; 633 int ret; 634 635 ret = nfs_readdir_xdr_to_array(desc, page, inode); 636 if (ret < 0) 637 goto error; 638 SetPageUptodate(page); 639 640 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 641 /* Should never happen */ 642 nfs_zap_mapping(inode, inode->i_mapping); 643 } 644 unlock_page(page); 645 return 0; 646 error: 647 unlock_page(page); 648 return ret; 649 } 650 651 static 652 void cache_page_release(nfs_readdir_descriptor_t *desc) 653 { 654 if (!desc->page->mapping) 655 nfs_readdir_clear_array(desc->page); 656 page_cache_release(desc->page); 657 desc->page = NULL; 658 } 659 660 static 661 struct page *get_cache_page(nfs_readdir_descriptor_t *desc) 662 { 663 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping, 664 desc->page_index, (filler_t *)nfs_readdir_filler, desc); 665 } 666 667 /* 668 * Returns 0 if desc->dir_cookie was found on page desc->page_index 669 */ 670 static 671 int find_cache_page(nfs_readdir_descriptor_t *desc) 672 { 673 int res; 674 675 desc->page = get_cache_page(desc); 676 if (IS_ERR(desc->page)) 677 return PTR_ERR(desc->page); 678 679 res = nfs_readdir_search_array(desc); 680 if (res != 0) 681 cache_page_release(desc); 682 return res; 683 } 684 685 /* Search for desc->dir_cookie from the beginning of the page cache */ 686 static inline 687 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 688 { 689 int res; 690 691 if (desc->page_index == 0) { 692 desc->current_index = 0; 693 desc->last_cookie = 0; 694 } 695 do { 696 res = find_cache_page(desc); 697 } while (res == -EAGAIN); 698 return res; 699 } 700 701 /* 702 * Once we've found the start of the dirent within a page: fill 'er up... 703 */ 704 static 705 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, 706 filldir_t filldir) 707 { 708 struct file *file = desc->file; 709 int i = 0; 710 int res = 0; 711 struct nfs_cache_array *array = NULL; 712 struct nfs_open_dir_context *ctx = file->private_data; 713 714 array = nfs_readdir_get_array(desc->page); 715 if (IS_ERR(array)) { 716 res = PTR_ERR(array); 717 goto out; 718 } 719 720 for (i = desc->cache_entry_index; i < array->size; i++) { 721 struct nfs_cache_array_entry *ent; 722 723 ent = &array->array[i]; 724 if (filldir(dirent, ent->string.name, ent->string.len, 725 file->f_pos, nfs_compat_user_ino64(ent->ino), 726 ent->d_type) < 0) { 727 desc->eof = 1; 728 break; 729 } 730 file->f_pos++; 731 if (i < (array->size-1)) 732 *desc->dir_cookie = array->array[i+1].cookie; 733 else 734 *desc->dir_cookie = array->last_cookie; 735 if (ctx->duped != 0) 736 ctx->duped = 1; 737 } 738 if (array->eof_index >= 0) 739 desc->eof = 1; 740 741 nfs_readdir_release_array(desc->page); 742 out: 743 cache_page_release(desc); 744 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 745 (unsigned long long)*desc->dir_cookie, res); 746 return res; 747 } 748 749 /* 750 * If we cannot find a cookie in our cache, we suspect that this is 751 * because it points to a deleted file, so we ask the server to return 752 * whatever it thinks is the next entry. We then feed this to filldir. 753 * If all goes well, we should then be able to find our way round the 754 * cache on the next call to readdir_search_pagecache(); 755 * 756 * NOTE: we cannot add the anonymous page to the pagecache because 757 * the data it contains might not be page aligned. Besides, 758 * we should already have a complete representation of the 759 * directory in the page cache by the time we get here. 760 */ 761 static inline 762 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, 763 filldir_t filldir) 764 { 765 struct page *page = NULL; 766 int status; 767 struct inode *inode = desc->file->f_path.dentry->d_inode; 768 struct nfs_open_dir_context *ctx = desc->file->private_data; 769 770 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 771 (unsigned long long)*desc->dir_cookie); 772 773 page = alloc_page(GFP_HIGHUSER); 774 if (!page) { 775 status = -ENOMEM; 776 goto out; 777 } 778 779 desc->page_index = 0; 780 desc->last_cookie = *desc->dir_cookie; 781 desc->page = page; 782 ctx->duped = 0; 783 784 status = nfs_readdir_xdr_to_array(desc, page, inode); 785 if (status < 0) 786 goto out_release; 787 788 status = nfs_do_filldir(desc, dirent, filldir); 789 790 out: 791 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 792 __func__, status); 793 return status; 794 out_release: 795 cache_page_release(desc); 796 goto out; 797 } 798 799 /* The file offset position represents the dirent entry number. A 800 last cookie cache takes care of the common case of reading the 801 whole directory. 802 */ 803 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) 804 { 805 struct dentry *dentry = filp->f_path.dentry; 806 struct inode *inode = dentry->d_inode; 807 nfs_readdir_descriptor_t my_desc, 808 *desc = &my_desc; 809 struct nfs_open_dir_context *dir_ctx = filp->private_data; 810 int res; 811 812 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n", 813 dentry->d_parent->d_name.name, dentry->d_name.name, 814 (long long)filp->f_pos); 815 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 816 817 /* 818 * filp->f_pos points to the dirent entry number. 819 * *desc->dir_cookie has the cookie for the next entry. We have 820 * to either find the entry with the appropriate number or 821 * revalidate the cookie. 822 */ 823 memset(desc, 0, sizeof(*desc)); 824 825 desc->file = filp; 826 desc->dir_cookie = &dir_ctx->dir_cookie; 827 desc->decode = NFS_PROTO(inode)->decode_dirent; 828 desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0; 829 830 nfs_block_sillyrename(dentry); 831 res = nfs_revalidate_mapping(inode, filp->f_mapping); 832 if (res < 0) 833 goto out; 834 835 do { 836 res = readdir_search_pagecache(desc); 837 838 if (res == -EBADCOOKIE) { 839 res = 0; 840 /* This means either end of directory */ 841 if (*desc->dir_cookie && desc->eof == 0) { 842 /* Or that the server has 'lost' a cookie */ 843 res = uncached_readdir(desc, dirent, filldir); 844 if (res == 0) 845 continue; 846 } 847 break; 848 } 849 if (res == -ETOOSMALL && desc->plus) { 850 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 851 nfs_zap_caches(inode); 852 desc->page_index = 0; 853 desc->plus = 0; 854 desc->eof = 0; 855 continue; 856 } 857 if (res < 0) 858 break; 859 860 res = nfs_do_filldir(desc, dirent, filldir); 861 if (res < 0) 862 break; 863 } while (!desc->eof); 864 out: 865 nfs_unblock_sillyrename(dentry); 866 if (res > 0) 867 res = 0; 868 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n", 869 dentry->d_parent->d_name.name, dentry->d_name.name, 870 res); 871 return res; 872 } 873 874 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 875 { 876 struct dentry *dentry = filp->f_path.dentry; 877 struct inode *inode = dentry->d_inode; 878 struct nfs_open_dir_context *dir_ctx = filp->private_data; 879 880 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", 881 dentry->d_parent->d_name.name, 882 dentry->d_name.name, 883 offset, whence); 884 885 mutex_lock(&inode->i_mutex); 886 switch (whence) { 887 case 1: 888 offset += filp->f_pos; 889 case 0: 890 if (offset >= 0) 891 break; 892 default: 893 offset = -EINVAL; 894 goto out; 895 } 896 if (offset != filp->f_pos) { 897 filp->f_pos = offset; 898 dir_ctx->dir_cookie = 0; 899 dir_ctx->duped = 0; 900 } 901 out: 902 mutex_unlock(&inode->i_mutex); 903 return offset; 904 } 905 906 /* 907 * All directory operations under NFS are synchronous, so fsync() 908 * is a dummy operation. 909 */ 910 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 911 int datasync) 912 { 913 struct dentry *dentry = filp->f_path.dentry; 914 struct inode *inode = dentry->d_inode; 915 916 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", 917 dentry->d_parent->d_name.name, dentry->d_name.name, 918 datasync); 919 920 mutex_lock(&inode->i_mutex); 921 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); 922 mutex_unlock(&inode->i_mutex); 923 return 0; 924 } 925 926 /** 927 * nfs_force_lookup_revalidate - Mark the directory as having changed 928 * @dir - pointer to directory inode 929 * 930 * This forces the revalidation code in nfs_lookup_revalidate() to do a 931 * full lookup on all child dentries of 'dir' whenever a change occurs 932 * on the server that might have invalidated our dcache. 933 * 934 * The caller should be holding dir->i_lock 935 */ 936 void nfs_force_lookup_revalidate(struct inode *dir) 937 { 938 NFS_I(dir)->cache_change_attribute++; 939 } 940 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 941 942 /* 943 * A check for whether or not the parent directory has changed. 944 * In the case it has, we assume that the dentries are untrustworthy 945 * and may need to be looked up again. 946 */ 947 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) 948 { 949 if (IS_ROOT(dentry)) 950 return 1; 951 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 952 return 0; 953 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 954 return 0; 955 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 956 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 957 return 0; 958 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 959 return 0; 960 return 1; 961 } 962 963 /* 964 * Use intent information to check whether or not we're going to do 965 * an O_EXCL create using this path component. 966 */ 967 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 968 { 969 if (NFS_PROTO(dir)->version == 2) 970 return 0; 971 return flags & LOOKUP_EXCL; 972 } 973 974 /* 975 * Inode and filehandle revalidation for lookups. 976 * 977 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 978 * or if the intent information indicates that we're about to open this 979 * particular file and the "nocto" mount flag is not set. 980 * 981 */ 982 static 983 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 984 { 985 struct nfs_server *server = NFS_SERVER(inode); 986 int ret; 987 988 if (IS_AUTOMOUNT(inode)) 989 return 0; 990 /* VFS wants an on-the-wire revalidation */ 991 if (flags & LOOKUP_REVAL) 992 goto out_force; 993 /* This is an open(2) */ 994 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && 995 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) 996 goto out_force; 997 out: 998 return (inode->i_nlink == 0) ? -ENOENT : 0; 999 out_force: 1000 ret = __nfs_revalidate_inode(server, inode); 1001 if (ret != 0) 1002 return ret; 1003 goto out; 1004 } 1005 1006 /* 1007 * We judge how long we want to trust negative 1008 * dentries by looking at the parent inode mtime. 1009 * 1010 * If parent mtime has changed, we revalidate, else we wait for a 1011 * period corresponding to the parent's attribute cache timeout value. 1012 */ 1013 static inline 1014 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1015 unsigned int flags) 1016 { 1017 /* Don't revalidate a negative dentry if we're creating a new file */ 1018 if (flags & LOOKUP_CREATE) 1019 return 0; 1020 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1021 return 1; 1022 return !nfs_check_verifier(dir, dentry); 1023 } 1024 1025 /* 1026 * This is called every time the dcache has a lookup hit, 1027 * and we should check whether we can really trust that 1028 * lookup. 1029 * 1030 * NOTE! The hit can be a negative hit too, don't assume 1031 * we have an inode! 1032 * 1033 * If the parent directory is seen to have changed, we throw out the 1034 * cached dentry and do a new lookup. 1035 */ 1036 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1037 { 1038 struct inode *dir; 1039 struct inode *inode; 1040 struct dentry *parent; 1041 struct nfs_fh *fhandle = NULL; 1042 struct nfs_fattr *fattr = NULL; 1043 int error; 1044 1045 if (flags & LOOKUP_RCU) 1046 return -ECHILD; 1047 1048 parent = dget_parent(dentry); 1049 dir = parent->d_inode; 1050 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1051 inode = dentry->d_inode; 1052 1053 if (!inode) { 1054 if (nfs_neg_need_reval(dir, dentry, flags)) 1055 goto out_bad; 1056 goto out_valid_noent; 1057 } 1058 1059 if (is_bad_inode(inode)) { 1060 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", 1061 __func__, dentry->d_parent->d_name.name, 1062 dentry->d_name.name); 1063 goto out_bad; 1064 } 1065 1066 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) 1067 goto out_set_verifier; 1068 1069 /* Force a full look up iff the parent directory has changed */ 1070 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) { 1071 if (nfs_lookup_verify_inode(inode, flags)) 1072 goto out_zap_parent; 1073 goto out_valid; 1074 } 1075 1076 if (NFS_STALE(inode)) 1077 goto out_bad; 1078 1079 error = -ENOMEM; 1080 fhandle = nfs_alloc_fhandle(); 1081 fattr = nfs_alloc_fattr(); 1082 if (fhandle == NULL || fattr == NULL) 1083 goto out_error; 1084 1085 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1086 if (error) 1087 goto out_bad; 1088 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1089 goto out_bad; 1090 if ((error = nfs_refresh_inode(inode, fattr)) != 0) 1091 goto out_bad; 1092 1093 nfs_free_fattr(fattr); 1094 nfs_free_fhandle(fhandle); 1095 out_set_verifier: 1096 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1097 out_valid: 1098 /* Success: notify readdir to use READDIRPLUS */ 1099 nfs_advise_use_readdirplus(dir); 1100 out_valid_noent: 1101 dput(parent); 1102 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", 1103 __func__, dentry->d_parent->d_name.name, 1104 dentry->d_name.name); 1105 return 1; 1106 out_zap_parent: 1107 nfs_zap_caches(dir); 1108 out_bad: 1109 nfs_free_fattr(fattr); 1110 nfs_free_fhandle(fhandle); 1111 nfs_mark_for_revalidate(dir); 1112 if (inode && S_ISDIR(inode->i_mode)) { 1113 /* Purge readdir caches. */ 1114 nfs_zap_caches(inode); 1115 /* If we have submounts, don't unhash ! */ 1116 if (have_submounts(dentry)) 1117 goto out_valid; 1118 if (dentry->d_flags & DCACHE_DISCONNECTED) 1119 goto out_valid; 1120 shrink_dcache_parent(dentry); 1121 } 1122 d_drop(dentry); 1123 dput(parent); 1124 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", 1125 __func__, dentry->d_parent->d_name.name, 1126 dentry->d_name.name); 1127 return 0; 1128 out_error: 1129 nfs_free_fattr(fattr); 1130 nfs_free_fhandle(fhandle); 1131 dput(parent); 1132 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n", 1133 __func__, dentry->d_parent->d_name.name, 1134 dentry->d_name.name, error); 1135 return error; 1136 } 1137 1138 /* 1139 * This is called from dput() when d_count is going to 0. 1140 */ 1141 static int nfs_dentry_delete(const struct dentry *dentry) 1142 { 1143 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", 1144 dentry->d_parent->d_name.name, dentry->d_name.name, 1145 dentry->d_flags); 1146 1147 /* Unhash any dentry with a stale inode */ 1148 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) 1149 return 1; 1150 1151 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1152 /* Unhash it, so that ->d_iput() would be called */ 1153 return 1; 1154 } 1155 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 1156 /* Unhash it, so that ancestors of killed async unlink 1157 * files will be cleaned up during umount */ 1158 return 1; 1159 } 1160 return 0; 1161 1162 } 1163 1164 /* Ensure that we revalidate inode->i_nlink */ 1165 static void nfs_drop_nlink(struct inode *inode) 1166 { 1167 spin_lock(&inode->i_lock); 1168 /* drop the inode if we're reasonably sure this is the last link */ 1169 if (inode->i_nlink == 1) 1170 clear_nlink(inode); 1171 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; 1172 spin_unlock(&inode->i_lock); 1173 } 1174 1175 /* 1176 * Called when the dentry loses inode. 1177 * We use it to clean up silly-renamed files. 1178 */ 1179 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1180 { 1181 if (S_ISDIR(inode->i_mode)) 1182 /* drop any readdir cache as it could easily be old */ 1183 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1184 1185 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1186 nfs_complete_unlink(dentry, inode); 1187 nfs_drop_nlink(inode); 1188 } 1189 iput(inode); 1190 } 1191 1192 static void nfs_d_release(struct dentry *dentry) 1193 { 1194 /* free cached devname value, if it survived that far */ 1195 if (unlikely(dentry->d_fsdata)) { 1196 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1197 WARN_ON(1); 1198 else 1199 kfree(dentry->d_fsdata); 1200 } 1201 } 1202 1203 const struct dentry_operations nfs_dentry_operations = { 1204 .d_revalidate = nfs_lookup_revalidate, 1205 .d_delete = nfs_dentry_delete, 1206 .d_iput = nfs_dentry_iput, 1207 .d_automount = nfs_d_automount, 1208 .d_release = nfs_d_release, 1209 }; 1210 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1211 1212 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1213 { 1214 struct dentry *res; 1215 struct dentry *parent; 1216 struct inode *inode = NULL; 1217 struct nfs_fh *fhandle = NULL; 1218 struct nfs_fattr *fattr = NULL; 1219 int error; 1220 1221 dfprintk(VFS, "NFS: lookup(%s/%s)\n", 1222 dentry->d_parent->d_name.name, dentry->d_name.name); 1223 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1224 1225 res = ERR_PTR(-ENAMETOOLONG); 1226 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1227 goto out; 1228 1229 /* 1230 * If we're doing an exclusive create, optimize away the lookup 1231 * but don't hash the dentry. 1232 */ 1233 if (nfs_is_exclusive_create(dir, flags)) { 1234 d_instantiate(dentry, NULL); 1235 res = NULL; 1236 goto out; 1237 } 1238 1239 res = ERR_PTR(-ENOMEM); 1240 fhandle = nfs_alloc_fhandle(); 1241 fattr = nfs_alloc_fattr(); 1242 if (fhandle == NULL || fattr == NULL) 1243 goto out; 1244 1245 parent = dentry->d_parent; 1246 /* Protect against concurrent sillydeletes */ 1247 nfs_block_sillyrename(parent); 1248 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1249 if (error == -ENOENT) 1250 goto no_entry; 1251 if (error < 0) { 1252 res = ERR_PTR(error); 1253 goto out_unblock_sillyrename; 1254 } 1255 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1256 res = ERR_CAST(inode); 1257 if (IS_ERR(res)) 1258 goto out_unblock_sillyrename; 1259 1260 /* Success: notify readdir to use READDIRPLUS */ 1261 nfs_advise_use_readdirplus(dir); 1262 1263 no_entry: 1264 res = d_materialise_unique(dentry, inode); 1265 if (res != NULL) { 1266 if (IS_ERR(res)) 1267 goto out_unblock_sillyrename; 1268 dentry = res; 1269 } 1270 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1271 out_unblock_sillyrename: 1272 nfs_unblock_sillyrename(parent); 1273 out: 1274 nfs_free_fattr(fattr); 1275 nfs_free_fhandle(fhandle); 1276 return res; 1277 } 1278 EXPORT_SYMBOL_GPL(nfs_lookup); 1279 1280 #if IS_ENABLED(CONFIG_NFS_V4) 1281 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1282 1283 const struct dentry_operations nfs4_dentry_operations = { 1284 .d_revalidate = nfs4_lookup_revalidate, 1285 .d_delete = nfs_dentry_delete, 1286 .d_iput = nfs_dentry_iput, 1287 .d_automount = nfs_d_automount, 1288 .d_release = nfs_d_release, 1289 }; 1290 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1291 1292 static fmode_t flags_to_mode(int flags) 1293 { 1294 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1295 if ((flags & O_ACCMODE) != O_WRONLY) 1296 res |= FMODE_READ; 1297 if ((flags & O_ACCMODE) != O_RDONLY) 1298 res |= FMODE_WRITE; 1299 return res; 1300 } 1301 1302 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags) 1303 { 1304 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags)); 1305 } 1306 1307 static int do_open(struct inode *inode, struct file *filp) 1308 { 1309 nfs_fscache_set_inode_cookie(inode, filp); 1310 return 0; 1311 } 1312 1313 static int nfs_finish_open(struct nfs_open_context *ctx, 1314 struct dentry *dentry, 1315 struct file *file, unsigned open_flags, 1316 int *opened) 1317 { 1318 int err; 1319 1320 if (ctx->dentry != dentry) { 1321 dput(ctx->dentry); 1322 ctx->dentry = dget(dentry); 1323 } 1324 1325 /* If the open_intent is for execute, we have an extra check to make */ 1326 if (ctx->mode & FMODE_EXEC) { 1327 err = nfs_may_open(dentry->d_inode, ctx->cred, open_flags); 1328 if (err < 0) 1329 goto out; 1330 } 1331 1332 err = finish_open(file, dentry, do_open, opened); 1333 if (err) 1334 goto out; 1335 nfs_file_set_open_context(file, ctx); 1336 1337 out: 1338 put_nfs_open_context(ctx); 1339 return err; 1340 } 1341 1342 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1343 struct file *file, unsigned open_flags, 1344 umode_t mode, int *opened) 1345 { 1346 struct nfs_open_context *ctx; 1347 struct dentry *res; 1348 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1349 struct inode *inode; 1350 int err; 1351 1352 /* Expect a negative dentry */ 1353 BUG_ON(dentry->d_inode); 1354 1355 dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n", 1356 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1357 1358 /* NFS only supports OPEN on regular files */ 1359 if ((open_flags & O_DIRECTORY)) { 1360 if (!d_unhashed(dentry)) { 1361 /* 1362 * Hashed negative dentry with O_DIRECTORY: dentry was 1363 * revalidated and is fine, no need to perform lookup 1364 * again 1365 */ 1366 return -ENOENT; 1367 } 1368 goto no_open; 1369 } 1370 1371 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1372 return -ENAMETOOLONG; 1373 1374 if (open_flags & O_CREAT) { 1375 attr.ia_valid |= ATTR_MODE; 1376 attr.ia_mode = mode & ~current_umask(); 1377 } 1378 if (open_flags & O_TRUNC) { 1379 attr.ia_valid |= ATTR_SIZE; 1380 attr.ia_size = 0; 1381 } 1382 1383 ctx = create_nfs_open_context(dentry, open_flags); 1384 err = PTR_ERR(ctx); 1385 if (IS_ERR(ctx)) 1386 goto out; 1387 1388 nfs_block_sillyrename(dentry->d_parent); 1389 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr); 1390 d_drop(dentry); 1391 if (IS_ERR(inode)) { 1392 nfs_unblock_sillyrename(dentry->d_parent); 1393 put_nfs_open_context(ctx); 1394 err = PTR_ERR(inode); 1395 switch (err) { 1396 case -ENOENT: 1397 d_add(dentry, NULL); 1398 break; 1399 case -EISDIR: 1400 case -ENOTDIR: 1401 goto no_open; 1402 case -ELOOP: 1403 if (!(open_flags & O_NOFOLLOW)) 1404 goto no_open; 1405 break; 1406 /* case -EINVAL: */ 1407 default: 1408 break; 1409 } 1410 goto out; 1411 } 1412 res = d_add_unique(dentry, inode); 1413 if (res != NULL) 1414 dentry = res; 1415 1416 nfs_unblock_sillyrename(dentry->d_parent); 1417 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1418 1419 err = nfs_finish_open(ctx, dentry, file, open_flags, opened); 1420 1421 dput(res); 1422 out: 1423 return err; 1424 1425 no_open: 1426 res = nfs_lookup(dir, dentry, 0); 1427 err = PTR_ERR(res); 1428 if (IS_ERR(res)) 1429 goto out; 1430 1431 return finish_no_open(file, res); 1432 } 1433 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1434 1435 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1436 { 1437 struct dentry *parent = NULL; 1438 struct inode *inode; 1439 struct inode *dir; 1440 int ret = 0; 1441 1442 if (flags & LOOKUP_RCU) 1443 return -ECHILD; 1444 1445 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1446 goto no_open; 1447 if (d_mountpoint(dentry)) 1448 goto no_open; 1449 1450 inode = dentry->d_inode; 1451 parent = dget_parent(dentry); 1452 dir = parent->d_inode; 1453 1454 /* We can't create new files in nfs_open_revalidate(), so we 1455 * optimize away revalidation of negative dentries. 1456 */ 1457 if (inode == NULL) { 1458 if (!nfs_neg_need_reval(dir, dentry, flags)) 1459 ret = 1; 1460 goto out; 1461 } 1462 1463 /* NFS only supports OPEN on regular files */ 1464 if (!S_ISREG(inode->i_mode)) 1465 goto no_open_dput; 1466 /* We cannot do exclusive creation on a positive dentry */ 1467 if (flags & LOOKUP_EXCL) 1468 goto no_open_dput; 1469 1470 /* Let f_op->open() actually open (and revalidate) the file */ 1471 ret = 1; 1472 1473 out: 1474 dput(parent); 1475 return ret; 1476 1477 no_open_dput: 1478 dput(parent); 1479 no_open: 1480 return nfs_lookup_revalidate(dentry, flags); 1481 } 1482 1483 #endif /* CONFIG_NFSV4 */ 1484 1485 /* 1486 * Code common to create, mkdir, and mknod. 1487 */ 1488 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1489 struct nfs_fattr *fattr) 1490 { 1491 struct dentry *parent = dget_parent(dentry); 1492 struct inode *dir = parent->d_inode; 1493 struct inode *inode; 1494 int error = -EACCES; 1495 1496 d_drop(dentry); 1497 1498 /* We may have been initialized further down */ 1499 if (dentry->d_inode) 1500 goto out; 1501 if (fhandle->size == 0) { 1502 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); 1503 if (error) 1504 goto out_error; 1505 } 1506 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1507 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1508 struct nfs_server *server = NFS_SB(dentry->d_sb); 1509 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); 1510 if (error < 0) 1511 goto out_error; 1512 } 1513 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1514 error = PTR_ERR(inode); 1515 if (IS_ERR(inode)) 1516 goto out_error; 1517 d_add(dentry, inode); 1518 out: 1519 dput(parent); 1520 return 0; 1521 out_error: 1522 nfs_mark_for_revalidate(dir); 1523 dput(parent); 1524 return error; 1525 } 1526 EXPORT_SYMBOL_GPL(nfs_instantiate); 1527 1528 /* 1529 * Following a failed create operation, we drop the dentry rather 1530 * than retain a negative dentry. This avoids a problem in the event 1531 * that the operation succeeded on the server, but an error in the 1532 * reply path made it appear to have failed. 1533 */ 1534 int nfs_create(struct inode *dir, struct dentry *dentry, 1535 umode_t mode, bool excl) 1536 { 1537 struct iattr attr; 1538 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 1539 int error; 1540 1541 dfprintk(VFS, "NFS: create(%s/%ld), %s\n", 1542 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1543 1544 attr.ia_mode = mode; 1545 attr.ia_valid = ATTR_MODE; 1546 1547 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1548 if (error != 0) 1549 goto out_err; 1550 return 0; 1551 out_err: 1552 d_drop(dentry); 1553 return error; 1554 } 1555 EXPORT_SYMBOL_GPL(nfs_create); 1556 1557 /* 1558 * See comments for nfs_proc_create regarding failed operations. 1559 */ 1560 int 1561 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) 1562 { 1563 struct iattr attr; 1564 int status; 1565 1566 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", 1567 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1568 1569 if (!new_valid_dev(rdev)) 1570 return -EINVAL; 1571 1572 attr.ia_mode = mode; 1573 attr.ia_valid = ATTR_MODE; 1574 1575 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1576 if (status != 0) 1577 goto out_err; 1578 return 0; 1579 out_err: 1580 d_drop(dentry); 1581 return status; 1582 } 1583 EXPORT_SYMBOL_GPL(nfs_mknod); 1584 1585 /* 1586 * See comments for nfs_proc_create regarding failed operations. 1587 */ 1588 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 1589 { 1590 struct iattr attr; 1591 int error; 1592 1593 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", 1594 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1595 1596 attr.ia_valid = ATTR_MODE; 1597 attr.ia_mode = mode | S_IFDIR; 1598 1599 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1600 if (error != 0) 1601 goto out_err; 1602 return 0; 1603 out_err: 1604 d_drop(dentry); 1605 return error; 1606 } 1607 EXPORT_SYMBOL_GPL(nfs_mkdir); 1608 1609 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1610 { 1611 if (dentry->d_inode != NULL && !d_unhashed(dentry)) 1612 d_delete(dentry); 1613 } 1614 1615 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1616 { 1617 int error; 1618 1619 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", 1620 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); 1621 1622 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1623 /* Ensure the VFS deletes this inode */ 1624 if (error == 0 && dentry->d_inode != NULL) 1625 clear_nlink(dentry->d_inode); 1626 else if (error == -ENOENT) 1627 nfs_dentry_handle_enoent(dentry); 1628 1629 return error; 1630 } 1631 EXPORT_SYMBOL_GPL(nfs_rmdir); 1632 1633 /* 1634 * Remove a file after making sure there are no pending writes, 1635 * and after checking that the file has only one user. 1636 * 1637 * We invalidate the attribute cache and free the inode prior to the operation 1638 * to avoid possible races if the server reuses the inode. 1639 */ 1640 static int nfs_safe_remove(struct dentry *dentry) 1641 { 1642 struct inode *dir = dentry->d_parent->d_inode; 1643 struct inode *inode = dentry->d_inode; 1644 int error = -EBUSY; 1645 1646 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", 1647 dentry->d_parent->d_name.name, dentry->d_name.name); 1648 1649 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1650 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1651 error = 0; 1652 goto out; 1653 } 1654 1655 if (inode != NULL) { 1656 NFS_PROTO(inode)->return_delegation(inode); 1657 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1658 if (error == 0) 1659 nfs_drop_nlink(inode); 1660 } else 1661 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1662 if (error == -ENOENT) 1663 nfs_dentry_handle_enoent(dentry); 1664 out: 1665 return error; 1666 } 1667 1668 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1669 * belongs to an active ".nfs..." file and we return -EBUSY. 1670 * 1671 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1672 */ 1673 int nfs_unlink(struct inode *dir, struct dentry *dentry) 1674 { 1675 int error; 1676 int need_rehash = 0; 1677 1678 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, 1679 dir->i_ino, dentry->d_name.name); 1680 1681 spin_lock(&dentry->d_lock); 1682 if (dentry->d_count > 1) { 1683 spin_unlock(&dentry->d_lock); 1684 /* Start asynchronous writeout of the inode */ 1685 write_inode_now(dentry->d_inode, 0); 1686 error = nfs_sillyrename(dir, dentry); 1687 return error; 1688 } 1689 if (!d_unhashed(dentry)) { 1690 __d_drop(dentry); 1691 need_rehash = 1; 1692 } 1693 spin_unlock(&dentry->d_lock); 1694 error = nfs_safe_remove(dentry); 1695 if (!error || error == -ENOENT) { 1696 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1697 } else if (need_rehash) 1698 d_rehash(dentry); 1699 return error; 1700 } 1701 EXPORT_SYMBOL_GPL(nfs_unlink); 1702 1703 /* 1704 * To create a symbolic link, most file systems instantiate a new inode, 1705 * add a page to it containing the path, then write it out to the disk 1706 * using prepare_write/commit_write. 1707 * 1708 * Unfortunately the NFS client can't create the in-core inode first 1709 * because it needs a file handle to create an in-core inode (see 1710 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1711 * symlink request has completed on the server. 1712 * 1713 * So instead we allocate a raw page, copy the symname into it, then do 1714 * the SYMLINK request with the page as the buffer. If it succeeds, we 1715 * now have a new file handle and can instantiate an in-core NFS inode 1716 * and move the raw page into its mapping. 1717 */ 1718 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1719 { 1720 struct pagevec lru_pvec; 1721 struct page *page; 1722 char *kaddr; 1723 struct iattr attr; 1724 unsigned int pathlen = strlen(symname); 1725 int error; 1726 1727 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, 1728 dir->i_ino, dentry->d_name.name, symname); 1729 1730 if (pathlen > PAGE_SIZE) 1731 return -ENAMETOOLONG; 1732 1733 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1734 attr.ia_valid = ATTR_MODE; 1735 1736 page = alloc_page(GFP_HIGHUSER); 1737 if (!page) 1738 return -ENOMEM; 1739 1740 kaddr = kmap_atomic(page); 1741 memcpy(kaddr, symname, pathlen); 1742 if (pathlen < PAGE_SIZE) 1743 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1744 kunmap_atomic(kaddr); 1745 1746 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1747 if (error != 0) { 1748 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", 1749 dir->i_sb->s_id, dir->i_ino, 1750 dentry->d_name.name, symname, error); 1751 d_drop(dentry); 1752 __free_page(page); 1753 return error; 1754 } 1755 1756 /* 1757 * No big deal if we can't add this page to the page cache here. 1758 * READLINK will get the missing page from the server if needed. 1759 */ 1760 pagevec_init(&lru_pvec, 0); 1761 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, 1762 GFP_KERNEL)) { 1763 pagevec_add(&lru_pvec, page); 1764 pagevec_lru_add_file(&lru_pvec); 1765 SetPageUptodate(page); 1766 unlock_page(page); 1767 } else 1768 __free_page(page); 1769 1770 return 0; 1771 } 1772 EXPORT_SYMBOL_GPL(nfs_symlink); 1773 1774 int 1775 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1776 { 1777 struct inode *inode = old_dentry->d_inode; 1778 int error; 1779 1780 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", 1781 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1782 dentry->d_parent->d_name.name, dentry->d_name.name); 1783 1784 NFS_PROTO(inode)->return_delegation(inode); 1785 1786 d_drop(dentry); 1787 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1788 if (error == 0) { 1789 ihold(inode); 1790 d_add(dentry, inode); 1791 } 1792 return error; 1793 } 1794 EXPORT_SYMBOL_GPL(nfs_link); 1795 1796 /* 1797 * RENAME 1798 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1799 * different file handle for the same inode after a rename (e.g. when 1800 * moving to a different directory). A fail-safe method to do so would 1801 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1802 * rename the old file using the sillyrename stuff. This way, the original 1803 * file in old_dir will go away when the last process iput()s the inode. 1804 * 1805 * FIXED. 1806 * 1807 * It actually works quite well. One needs to have the possibility for 1808 * at least one ".nfs..." file in each directory the file ever gets 1809 * moved or linked to which happens automagically with the new 1810 * implementation that only depends on the dcache stuff instead of 1811 * using the inode layer 1812 * 1813 * Unfortunately, things are a little more complicated than indicated 1814 * above. For a cross-directory move, we want to make sure we can get 1815 * rid of the old inode after the operation. This means there must be 1816 * no pending writes (if it's a file), and the use count must be 1. 1817 * If these conditions are met, we can drop the dentries before doing 1818 * the rename. 1819 */ 1820 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1821 struct inode *new_dir, struct dentry *new_dentry) 1822 { 1823 struct inode *old_inode = old_dentry->d_inode; 1824 struct inode *new_inode = new_dentry->d_inode; 1825 struct dentry *dentry = NULL, *rehash = NULL; 1826 int error = -EBUSY; 1827 1828 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", 1829 old_dentry->d_parent->d_name.name, old_dentry->d_name.name, 1830 new_dentry->d_parent->d_name.name, new_dentry->d_name.name, 1831 new_dentry->d_count); 1832 1833 /* 1834 * For non-directories, check whether the target is busy and if so, 1835 * make a copy of the dentry and then do a silly-rename. If the 1836 * silly-rename succeeds, the copied dentry is hashed and becomes 1837 * the new target. 1838 */ 1839 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 1840 /* 1841 * To prevent any new references to the target during the 1842 * rename, we unhash the dentry in advance. 1843 */ 1844 if (!d_unhashed(new_dentry)) { 1845 d_drop(new_dentry); 1846 rehash = new_dentry; 1847 } 1848 1849 if (new_dentry->d_count > 2) { 1850 int err; 1851 1852 /* copy the target dentry's name */ 1853 dentry = d_alloc(new_dentry->d_parent, 1854 &new_dentry->d_name); 1855 if (!dentry) 1856 goto out; 1857 1858 /* silly-rename the existing target ... */ 1859 err = nfs_sillyrename(new_dir, new_dentry); 1860 if (err) 1861 goto out; 1862 1863 new_dentry = dentry; 1864 rehash = NULL; 1865 new_inode = NULL; 1866 } 1867 } 1868 1869 NFS_PROTO(old_inode)->return_delegation(old_inode); 1870 if (new_inode != NULL) 1871 NFS_PROTO(new_inode)->return_delegation(new_inode); 1872 1873 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, 1874 new_dir, &new_dentry->d_name); 1875 nfs_mark_for_revalidate(old_inode); 1876 out: 1877 if (rehash) 1878 d_rehash(rehash); 1879 if (!error) { 1880 if (new_inode != NULL) 1881 nfs_drop_nlink(new_inode); 1882 d_move(old_dentry, new_dentry); 1883 nfs_set_verifier(new_dentry, 1884 nfs_save_change_attribute(new_dir)); 1885 } else if (error == -ENOENT) 1886 nfs_dentry_handle_enoent(old_dentry); 1887 1888 /* new dentry created? */ 1889 if (dentry) 1890 dput(dentry); 1891 return error; 1892 } 1893 EXPORT_SYMBOL_GPL(nfs_rename); 1894 1895 static DEFINE_SPINLOCK(nfs_access_lru_lock); 1896 static LIST_HEAD(nfs_access_lru_list); 1897 static atomic_long_t nfs_access_nr_entries; 1898 1899 static void nfs_access_free_entry(struct nfs_access_entry *entry) 1900 { 1901 put_rpccred(entry->cred); 1902 kfree(entry); 1903 smp_mb__before_atomic_dec(); 1904 atomic_long_dec(&nfs_access_nr_entries); 1905 smp_mb__after_atomic_dec(); 1906 } 1907 1908 static void nfs_access_free_list(struct list_head *head) 1909 { 1910 struct nfs_access_entry *cache; 1911 1912 while (!list_empty(head)) { 1913 cache = list_entry(head->next, struct nfs_access_entry, lru); 1914 list_del(&cache->lru); 1915 nfs_access_free_entry(cache); 1916 } 1917 } 1918 1919 int nfs_access_cache_shrinker(struct shrinker *shrink, 1920 struct shrink_control *sc) 1921 { 1922 LIST_HEAD(head); 1923 struct nfs_inode *nfsi, *next; 1924 struct nfs_access_entry *cache; 1925 int nr_to_scan = sc->nr_to_scan; 1926 gfp_t gfp_mask = sc->gfp_mask; 1927 1928 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 1929 return (nr_to_scan == 0) ? 0 : -1; 1930 1931 spin_lock(&nfs_access_lru_lock); 1932 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 1933 struct inode *inode; 1934 1935 if (nr_to_scan-- == 0) 1936 break; 1937 inode = &nfsi->vfs_inode; 1938 spin_lock(&inode->i_lock); 1939 if (list_empty(&nfsi->access_cache_entry_lru)) 1940 goto remove_lru_entry; 1941 cache = list_entry(nfsi->access_cache_entry_lru.next, 1942 struct nfs_access_entry, lru); 1943 list_move(&cache->lru, &head); 1944 rb_erase(&cache->rb_node, &nfsi->access_cache); 1945 if (!list_empty(&nfsi->access_cache_entry_lru)) 1946 list_move_tail(&nfsi->access_cache_inode_lru, 1947 &nfs_access_lru_list); 1948 else { 1949 remove_lru_entry: 1950 list_del_init(&nfsi->access_cache_inode_lru); 1951 smp_mb__before_clear_bit(); 1952 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 1953 smp_mb__after_clear_bit(); 1954 } 1955 spin_unlock(&inode->i_lock); 1956 } 1957 spin_unlock(&nfs_access_lru_lock); 1958 nfs_access_free_list(&head); 1959 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; 1960 } 1961 1962 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 1963 { 1964 struct rb_root *root_node = &nfsi->access_cache; 1965 struct rb_node *n; 1966 struct nfs_access_entry *entry; 1967 1968 /* Unhook entries from the cache */ 1969 while ((n = rb_first(root_node)) != NULL) { 1970 entry = rb_entry(n, struct nfs_access_entry, rb_node); 1971 rb_erase(n, root_node); 1972 list_move(&entry->lru, head); 1973 } 1974 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 1975 } 1976 1977 void nfs_access_zap_cache(struct inode *inode) 1978 { 1979 LIST_HEAD(head); 1980 1981 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 1982 return; 1983 /* Remove from global LRU init */ 1984 spin_lock(&nfs_access_lru_lock); 1985 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 1986 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 1987 1988 spin_lock(&inode->i_lock); 1989 __nfs_access_zap_cache(NFS_I(inode), &head); 1990 spin_unlock(&inode->i_lock); 1991 spin_unlock(&nfs_access_lru_lock); 1992 nfs_access_free_list(&head); 1993 } 1994 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 1995 1996 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 1997 { 1998 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 1999 struct nfs_access_entry *entry; 2000 2001 while (n != NULL) { 2002 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2003 2004 if (cred < entry->cred) 2005 n = n->rb_left; 2006 else if (cred > entry->cred) 2007 n = n->rb_right; 2008 else 2009 return entry; 2010 } 2011 return NULL; 2012 } 2013 2014 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2015 { 2016 struct nfs_inode *nfsi = NFS_I(inode); 2017 struct nfs_access_entry *cache; 2018 int err = -ENOENT; 2019 2020 spin_lock(&inode->i_lock); 2021 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2022 goto out_zap; 2023 cache = nfs_access_search_rbtree(inode, cred); 2024 if (cache == NULL) 2025 goto out; 2026 if (!nfs_have_delegated_attributes(inode) && 2027 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 2028 goto out_stale; 2029 res->jiffies = cache->jiffies; 2030 res->cred = cache->cred; 2031 res->mask = cache->mask; 2032 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2033 err = 0; 2034 out: 2035 spin_unlock(&inode->i_lock); 2036 return err; 2037 out_stale: 2038 rb_erase(&cache->rb_node, &nfsi->access_cache); 2039 list_del(&cache->lru); 2040 spin_unlock(&inode->i_lock); 2041 nfs_access_free_entry(cache); 2042 return -ENOENT; 2043 out_zap: 2044 spin_unlock(&inode->i_lock); 2045 nfs_access_zap_cache(inode); 2046 return -ENOENT; 2047 } 2048 2049 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2050 { 2051 struct nfs_inode *nfsi = NFS_I(inode); 2052 struct rb_root *root_node = &nfsi->access_cache; 2053 struct rb_node **p = &root_node->rb_node; 2054 struct rb_node *parent = NULL; 2055 struct nfs_access_entry *entry; 2056 2057 spin_lock(&inode->i_lock); 2058 while (*p != NULL) { 2059 parent = *p; 2060 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2061 2062 if (set->cred < entry->cred) 2063 p = &parent->rb_left; 2064 else if (set->cred > entry->cred) 2065 p = &parent->rb_right; 2066 else 2067 goto found; 2068 } 2069 rb_link_node(&set->rb_node, parent, p); 2070 rb_insert_color(&set->rb_node, root_node); 2071 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2072 spin_unlock(&inode->i_lock); 2073 return; 2074 found: 2075 rb_replace_node(parent, &set->rb_node, root_node); 2076 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2077 list_del(&entry->lru); 2078 spin_unlock(&inode->i_lock); 2079 nfs_access_free_entry(entry); 2080 } 2081 2082 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2083 { 2084 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2085 if (cache == NULL) 2086 return; 2087 RB_CLEAR_NODE(&cache->rb_node); 2088 cache->jiffies = set->jiffies; 2089 cache->cred = get_rpccred(set->cred); 2090 cache->mask = set->mask; 2091 2092 nfs_access_add_rbtree(inode, cache); 2093 2094 /* Update accounting */ 2095 smp_mb__before_atomic_inc(); 2096 atomic_long_inc(&nfs_access_nr_entries); 2097 smp_mb__after_atomic_inc(); 2098 2099 /* Add inode to global LRU list */ 2100 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2101 spin_lock(&nfs_access_lru_lock); 2102 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2103 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2104 &nfs_access_lru_list); 2105 spin_unlock(&nfs_access_lru_lock); 2106 } 2107 } 2108 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2109 2110 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2111 { 2112 entry->mask = 0; 2113 if (access_result & NFS4_ACCESS_READ) 2114 entry->mask |= MAY_READ; 2115 if (access_result & 2116 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) 2117 entry->mask |= MAY_WRITE; 2118 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) 2119 entry->mask |= MAY_EXEC; 2120 } 2121 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2122 2123 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 2124 { 2125 struct nfs_access_entry cache; 2126 int status; 2127 2128 status = nfs_access_get_cached(inode, cred, &cache); 2129 if (status == 0) 2130 goto out; 2131 2132 /* Be clever: ask server to check for all possible rights */ 2133 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 2134 cache.cred = cred; 2135 cache.jiffies = jiffies; 2136 status = NFS_PROTO(inode)->access(inode, &cache); 2137 if (status != 0) { 2138 if (status == -ESTALE) { 2139 nfs_zap_caches(inode); 2140 if (!S_ISDIR(inode->i_mode)) 2141 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 2142 } 2143 return status; 2144 } 2145 nfs_access_add_cache(inode, &cache); 2146 out: 2147 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2148 return 0; 2149 return -EACCES; 2150 } 2151 2152 static int nfs_open_permission_mask(int openflags) 2153 { 2154 int mask = 0; 2155 2156 if (openflags & __FMODE_EXEC) { 2157 /* ONLY check exec rights */ 2158 mask = MAY_EXEC; 2159 } else { 2160 if ((openflags & O_ACCMODE) != O_WRONLY) 2161 mask |= MAY_READ; 2162 if ((openflags & O_ACCMODE) != O_RDONLY) 2163 mask |= MAY_WRITE; 2164 } 2165 2166 return mask; 2167 } 2168 2169 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 2170 { 2171 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2172 } 2173 EXPORT_SYMBOL_GPL(nfs_may_open); 2174 2175 int nfs_permission(struct inode *inode, int mask) 2176 { 2177 struct rpc_cred *cred; 2178 int res = 0; 2179 2180 if (mask & MAY_NOT_BLOCK) 2181 return -ECHILD; 2182 2183 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2184 2185 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2186 goto out; 2187 /* Is this sys_access() ? */ 2188 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2189 goto force_lookup; 2190 2191 switch (inode->i_mode & S_IFMT) { 2192 case S_IFLNK: 2193 goto out; 2194 case S_IFREG: 2195 /* NFSv4 has atomic_open... */ 2196 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) 2197 && (mask & MAY_OPEN) 2198 && !(mask & MAY_EXEC)) 2199 goto out; 2200 break; 2201 case S_IFDIR: 2202 /* 2203 * Optimize away all write operations, since the server 2204 * will check permissions when we perform the op. 2205 */ 2206 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2207 goto out; 2208 } 2209 2210 force_lookup: 2211 if (!NFS_PROTO(inode)->access) 2212 goto out_notsup; 2213 2214 cred = rpc_lookup_cred(); 2215 if (!IS_ERR(cred)) { 2216 res = nfs_do_access(inode, cred, mask); 2217 put_rpccred(cred); 2218 } else 2219 res = PTR_ERR(cred); 2220 out: 2221 if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) 2222 res = -EACCES; 2223 2224 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", 2225 inode->i_sb->s_id, inode->i_ino, mask, res); 2226 return res; 2227 out_notsup: 2228 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2229 if (res == 0) 2230 res = generic_permission(inode, mask); 2231 goto out; 2232 } 2233 EXPORT_SYMBOL_GPL(nfs_permission); 2234 2235 /* 2236 * Local variables: 2237 * version-control: t 2238 * kept-new-versions: 5 2239 * End: 2240 */ 2241