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/swap.h> 37 #include <linux/sched.h> 38 #include <linux/kmemleak.h> 39 #include <linux/xattr.h> 40 41 #include "delegation.h" 42 #include "iostat.h" 43 #include "internal.h" 44 #include "fscache.h" 45 46 #include "nfstrace.h" 47 48 /* #define NFS_DEBUG_VERBOSE 1 */ 49 50 static int nfs_opendir(struct inode *, struct file *); 51 static int nfs_closedir(struct inode *, struct file *); 52 static int nfs_readdir(struct file *, struct dir_context *); 53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); 54 static loff_t nfs_llseek_dir(struct file *, loff_t, int); 55 static void nfs_readdir_clear_array(struct page*); 56 57 const struct file_operations nfs_dir_operations = { 58 .llseek = nfs_llseek_dir, 59 .read = generic_read_dir, 60 .iterate = nfs_readdir, 61 .open = nfs_opendir, 62 .release = nfs_closedir, 63 .fsync = nfs_fsync_dir, 64 }; 65 66 const struct address_space_operations nfs_dir_aops = { 67 .freepage = nfs_readdir_clear_array, 68 }; 69 70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred) 71 { 72 struct nfs_inode *nfsi = NFS_I(dir); 73 struct nfs_open_dir_context *ctx; 74 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 75 if (ctx != NULL) { 76 ctx->duped = 0; 77 ctx->attr_gencount = nfsi->attr_gencount; 78 ctx->dir_cookie = 0; 79 ctx->dup_cookie = 0; 80 ctx->cred = get_rpccred(cred); 81 spin_lock(&dir->i_lock); 82 list_add(&ctx->list, &nfsi->open_files); 83 spin_unlock(&dir->i_lock); 84 return ctx; 85 } 86 return ERR_PTR(-ENOMEM); 87 } 88 89 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx) 90 { 91 spin_lock(&dir->i_lock); 92 list_del(&ctx->list); 93 spin_unlock(&dir->i_lock); 94 put_rpccred(ctx->cred); 95 kfree(ctx); 96 } 97 98 /* 99 * Open file 100 */ 101 static int 102 nfs_opendir(struct inode *inode, struct file *filp) 103 { 104 int res = 0; 105 struct nfs_open_dir_context *ctx; 106 struct rpc_cred *cred; 107 108 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); 109 110 nfs_inc_stats(inode, NFSIOS_VFSOPEN); 111 112 cred = rpc_lookup_cred(); 113 if (IS_ERR(cred)) 114 return PTR_ERR(cred); 115 ctx = alloc_nfs_open_dir_context(inode, cred); 116 if (IS_ERR(ctx)) { 117 res = PTR_ERR(ctx); 118 goto out; 119 } 120 filp->private_data = ctx; 121 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { 122 /* This is a mountpoint, so d_revalidate will never 123 * have been called, so we need to refresh the 124 * inode (for close-open consistency) ourselves. 125 */ 126 __nfs_revalidate_inode(NFS_SERVER(inode), inode); 127 } 128 out: 129 put_rpccred(cred); 130 return res; 131 } 132 133 static int 134 nfs_closedir(struct inode *inode, struct file *filp) 135 { 136 put_nfs_open_dir_context(filp->f_path.dentry->d_inode, filp->private_data); 137 return 0; 138 } 139 140 struct nfs_cache_array_entry { 141 u64 cookie; 142 u64 ino; 143 struct qstr string; 144 unsigned char d_type; 145 }; 146 147 struct nfs_cache_array { 148 int size; 149 int eof_index; 150 u64 last_cookie; 151 struct nfs_cache_array_entry array[0]; 152 }; 153 154 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int); 155 typedef struct { 156 struct file *file; 157 struct page *page; 158 struct dir_context *ctx; 159 unsigned long page_index; 160 u64 *dir_cookie; 161 u64 last_cookie; 162 loff_t current_index; 163 decode_dirent_t decode; 164 165 unsigned long timestamp; 166 unsigned long gencount; 167 unsigned int cache_entry_index; 168 unsigned int plus:1; 169 unsigned int eof:1; 170 } nfs_readdir_descriptor_t; 171 172 /* 173 * The caller is responsible for calling nfs_readdir_release_array(page) 174 */ 175 static 176 struct nfs_cache_array *nfs_readdir_get_array(struct page *page) 177 { 178 void *ptr; 179 if (page == NULL) 180 return ERR_PTR(-EIO); 181 ptr = kmap(page); 182 if (ptr == NULL) 183 return ERR_PTR(-ENOMEM); 184 return ptr; 185 } 186 187 static 188 void nfs_readdir_release_array(struct page *page) 189 { 190 kunmap(page); 191 } 192 193 /* 194 * we are freeing strings created by nfs_add_to_readdir_array() 195 */ 196 static 197 void nfs_readdir_clear_array(struct page *page) 198 { 199 struct nfs_cache_array *array; 200 int i; 201 202 array = kmap_atomic(page); 203 for (i = 0; i < array->size; i++) 204 kfree(array->array[i].string.name); 205 kunmap_atomic(array); 206 } 207 208 /* 209 * the caller is responsible for freeing qstr.name 210 * when called by nfs_readdir_add_to_array, the strings will be freed in 211 * nfs_clear_readdir_array() 212 */ 213 static 214 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) 215 { 216 string->len = len; 217 string->name = kmemdup(name, len, GFP_KERNEL); 218 if (string->name == NULL) 219 return -ENOMEM; 220 /* 221 * Avoid a kmemleak false positive. The pointer to the name is stored 222 * in a page cache page which kmemleak does not scan. 223 */ 224 kmemleak_not_leak(string->name); 225 string->hash = full_name_hash(name, len); 226 return 0; 227 } 228 229 static 230 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 231 { 232 struct nfs_cache_array *array = nfs_readdir_get_array(page); 233 struct nfs_cache_array_entry *cache_entry; 234 int ret; 235 236 if (IS_ERR(array)) 237 return PTR_ERR(array); 238 239 cache_entry = &array->array[array->size]; 240 241 /* Check that this entry lies within the page bounds */ 242 ret = -ENOSPC; 243 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) 244 goto out; 245 246 cache_entry->cookie = entry->prev_cookie; 247 cache_entry->ino = entry->ino; 248 cache_entry->d_type = entry->d_type; 249 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); 250 if (ret) 251 goto out; 252 array->last_cookie = entry->cookie; 253 array->size++; 254 if (entry->eof != 0) 255 array->eof_index = array->size; 256 out: 257 nfs_readdir_release_array(page); 258 return ret; 259 } 260 261 static 262 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 263 { 264 loff_t diff = desc->ctx->pos - desc->current_index; 265 unsigned int index; 266 267 if (diff < 0) 268 goto out_eof; 269 if (diff >= array->size) { 270 if (array->eof_index >= 0) 271 goto out_eof; 272 return -EAGAIN; 273 } 274 275 index = (unsigned int)diff; 276 *desc->dir_cookie = array->array[index].cookie; 277 desc->cache_entry_index = index; 278 return 0; 279 out_eof: 280 desc->eof = 1; 281 return -EBADCOOKIE; 282 } 283 284 static bool 285 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) 286 { 287 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 288 return false; 289 smp_rmb(); 290 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); 291 } 292 293 static 294 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 295 { 296 int i; 297 loff_t new_pos; 298 int status = -EAGAIN; 299 300 for (i = 0; i < array->size; i++) { 301 if (array->array[i].cookie == *desc->dir_cookie) { 302 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); 303 struct nfs_open_dir_context *ctx = desc->file->private_data; 304 305 new_pos = desc->current_index + i; 306 if (ctx->attr_gencount != nfsi->attr_gencount || 307 !nfs_readdir_inode_mapping_valid(nfsi)) { 308 ctx->duped = 0; 309 ctx->attr_gencount = nfsi->attr_gencount; 310 } else if (new_pos < desc->ctx->pos) { 311 if (ctx->duped > 0 312 && ctx->dup_cookie == *desc->dir_cookie) { 313 if (printk_ratelimit()) { 314 pr_notice("NFS: directory %pD2 contains a readdir loop." 315 "Please contact your server vendor. " 316 "The file: %.*s has duplicate cookie %llu\n", 317 desc->file, array->array[i].string.len, 318 array->array[i].string.name, *desc->dir_cookie); 319 } 320 status = -ELOOP; 321 goto out; 322 } 323 ctx->dup_cookie = *desc->dir_cookie; 324 ctx->duped = -1; 325 } 326 desc->ctx->pos = new_pos; 327 desc->cache_entry_index = i; 328 return 0; 329 } 330 } 331 if (array->eof_index >= 0) { 332 status = -EBADCOOKIE; 333 if (*desc->dir_cookie == array->last_cookie) 334 desc->eof = 1; 335 } 336 out: 337 return status; 338 } 339 340 static 341 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) 342 { 343 struct nfs_cache_array *array; 344 int status; 345 346 array = nfs_readdir_get_array(desc->page); 347 if (IS_ERR(array)) { 348 status = PTR_ERR(array); 349 goto out; 350 } 351 352 if (*desc->dir_cookie == 0) 353 status = nfs_readdir_search_for_pos(array, desc); 354 else 355 status = nfs_readdir_search_for_cookie(array, desc); 356 357 if (status == -EAGAIN) { 358 desc->last_cookie = array->last_cookie; 359 desc->current_index += array->size; 360 desc->page_index++; 361 } 362 nfs_readdir_release_array(desc->page); 363 out: 364 return status; 365 } 366 367 /* Fill a page with xdr information before transferring to the cache page */ 368 static 369 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, 370 struct nfs_entry *entry, struct file *file, struct inode *inode) 371 { 372 struct nfs_open_dir_context *ctx = file->private_data; 373 struct rpc_cred *cred = ctx->cred; 374 unsigned long timestamp, gencount; 375 int error; 376 377 again: 378 timestamp = jiffies; 379 gencount = nfs_inc_attr_generation_counter(); 380 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages, 381 NFS_SERVER(inode)->dtsize, desc->plus); 382 if (error < 0) { 383 /* We requested READDIRPLUS, but the server doesn't grok it */ 384 if (error == -ENOTSUPP && desc->plus) { 385 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 386 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 387 desc->plus = 0; 388 goto again; 389 } 390 goto error; 391 } 392 desc->timestamp = timestamp; 393 desc->gencount = gencount; 394 error: 395 return error; 396 } 397 398 static int xdr_decode(nfs_readdir_descriptor_t *desc, 399 struct nfs_entry *entry, struct xdr_stream *xdr) 400 { 401 int error; 402 403 error = desc->decode(xdr, entry, desc->plus); 404 if (error) 405 return error; 406 entry->fattr->time_start = desc->timestamp; 407 entry->fattr->gencount = desc->gencount; 408 return 0; 409 } 410 411 static 412 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 413 { 414 if (dentry->d_inode == NULL) 415 goto different; 416 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0) 417 goto different; 418 return 1; 419 different: 420 return 0; 421 } 422 423 static 424 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) 425 { 426 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 427 return false; 428 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 429 return true; 430 if (ctx->pos == 0) 431 return true; 432 return false; 433 } 434 435 /* 436 * This function is called by the lookup code to request the use of 437 * readdirplus to accelerate any future lookups in the same 438 * directory. 439 */ 440 static 441 void nfs_advise_use_readdirplus(struct inode *dir) 442 { 443 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags); 444 } 445 446 /* 447 * This function is mainly for use by nfs_getattr(). 448 * 449 * If this is an 'ls -l', we want to force use of readdirplus. 450 * Do this by checking if there is an active file descriptor 451 * and calling nfs_advise_use_readdirplus, then forcing a 452 * cache flush. 453 */ 454 void nfs_force_use_readdirplus(struct inode *dir) 455 { 456 if (!list_empty(&NFS_I(dir)->open_files)) { 457 nfs_advise_use_readdirplus(dir); 458 nfs_zap_mapping(dir, dir->i_mapping); 459 } 460 } 461 462 static 463 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) 464 { 465 struct qstr filename = QSTR_INIT(entry->name, entry->len); 466 struct dentry *dentry; 467 struct dentry *alias; 468 struct inode *dir = parent->d_inode; 469 struct inode *inode; 470 int status; 471 472 if (filename.name[0] == '.') { 473 if (filename.len == 1) 474 return; 475 if (filename.len == 2 && filename.name[1] == '.') 476 return; 477 } 478 filename.hash = full_name_hash(filename.name, filename.len); 479 480 dentry = d_lookup(parent, &filename); 481 if (dentry != NULL) { 482 if (nfs_same_file(dentry, entry)) { 483 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 484 status = nfs_refresh_inode(dentry->d_inode, entry->fattr); 485 if (!status) 486 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label); 487 goto out; 488 } else { 489 d_invalidate(dentry); 490 dput(dentry); 491 } 492 } 493 494 dentry = d_alloc(parent, &filename); 495 if (dentry == NULL) 496 return; 497 498 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 499 if (IS_ERR(inode)) 500 goto out; 501 502 alias = d_materialise_unique(dentry, inode); 503 if (IS_ERR(alias)) 504 goto out; 505 else if (alias) { 506 nfs_set_verifier(alias, nfs_save_change_attribute(dir)); 507 dput(alias); 508 } else 509 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 510 511 out: 512 dput(dentry); 513 } 514 515 /* Perform conversion from xdr to cache array */ 516 static 517 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 518 struct page **xdr_pages, struct page *page, unsigned int buflen) 519 { 520 struct xdr_stream stream; 521 struct xdr_buf buf; 522 struct page *scratch; 523 struct nfs_cache_array *array; 524 unsigned int count = 0; 525 int status; 526 527 scratch = alloc_page(GFP_KERNEL); 528 if (scratch == NULL) 529 return -ENOMEM; 530 531 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 532 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 533 534 do { 535 status = xdr_decode(desc, entry, &stream); 536 if (status != 0) { 537 if (status == -EAGAIN) 538 status = 0; 539 break; 540 } 541 542 count++; 543 544 if (desc->plus != 0) 545 nfs_prime_dcache(desc->file->f_path.dentry, entry); 546 547 status = nfs_readdir_add_to_array(entry, page); 548 if (status != 0) 549 break; 550 } while (!entry->eof); 551 552 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { 553 array = nfs_readdir_get_array(page); 554 if (!IS_ERR(array)) { 555 array->eof_index = array->size; 556 status = 0; 557 nfs_readdir_release_array(page); 558 } else 559 status = PTR_ERR(array); 560 } 561 562 put_page(scratch); 563 return status; 564 } 565 566 static 567 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) 568 { 569 unsigned int i; 570 for (i = 0; i < npages; i++) 571 put_page(pages[i]); 572 } 573 574 static 575 void nfs_readdir_free_large_page(void *ptr, struct page **pages, 576 unsigned int npages) 577 { 578 nfs_readdir_free_pagearray(pages, npages); 579 } 580 581 /* 582 * nfs_readdir_large_page will allocate pages that must be freed with a call 583 * to nfs_readdir_free_large_page 584 */ 585 static 586 int nfs_readdir_large_page(struct page **pages, unsigned int npages) 587 { 588 unsigned int i; 589 590 for (i = 0; i < npages; i++) { 591 struct page *page = alloc_page(GFP_KERNEL); 592 if (page == NULL) 593 goto out_freepages; 594 pages[i] = page; 595 } 596 return 0; 597 598 out_freepages: 599 nfs_readdir_free_pagearray(pages, i); 600 return -ENOMEM; 601 } 602 603 static 604 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) 605 { 606 struct page *pages[NFS_MAX_READDIR_PAGES]; 607 void *pages_ptr = NULL; 608 struct nfs_entry entry; 609 struct file *file = desc->file; 610 struct nfs_cache_array *array; 611 int status = -ENOMEM; 612 unsigned int array_size = ARRAY_SIZE(pages); 613 614 entry.prev_cookie = 0; 615 entry.cookie = desc->last_cookie; 616 entry.eof = 0; 617 entry.fh = nfs_alloc_fhandle(); 618 entry.fattr = nfs_alloc_fattr(); 619 entry.server = NFS_SERVER(inode); 620 if (entry.fh == NULL || entry.fattr == NULL) 621 goto out; 622 623 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 624 if (IS_ERR(entry.label)) { 625 status = PTR_ERR(entry.label); 626 goto out; 627 } 628 629 array = nfs_readdir_get_array(page); 630 if (IS_ERR(array)) { 631 status = PTR_ERR(array); 632 goto out_label_free; 633 } 634 memset(array, 0, sizeof(struct nfs_cache_array)); 635 array->eof_index = -1; 636 637 status = nfs_readdir_large_page(pages, array_size); 638 if (status < 0) 639 goto out_release_array; 640 do { 641 unsigned int pglen; 642 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 643 644 if (status < 0) 645 break; 646 pglen = status; 647 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 648 if (status < 0) { 649 if (status == -ENOSPC) 650 status = 0; 651 break; 652 } 653 } while (array->eof_index < 0); 654 655 nfs_readdir_free_large_page(pages_ptr, pages, array_size); 656 out_release_array: 657 nfs_readdir_release_array(page); 658 out_label_free: 659 nfs4_label_free(entry.label); 660 out: 661 nfs_free_fattr(entry.fattr); 662 nfs_free_fhandle(entry.fh); 663 return status; 664 } 665 666 /* 667 * Now we cache directories properly, by converting xdr information 668 * to an array that can be used for lookups later. This results in 669 * fewer cache pages, since we can store more information on each page. 670 * We only need to convert from xdr once so future lookups are much simpler 671 */ 672 static 673 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) 674 { 675 struct inode *inode = file_inode(desc->file); 676 int ret; 677 678 ret = nfs_readdir_xdr_to_array(desc, page, inode); 679 if (ret < 0) 680 goto error; 681 SetPageUptodate(page); 682 683 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 684 /* Should never happen */ 685 nfs_zap_mapping(inode, inode->i_mapping); 686 } 687 unlock_page(page); 688 return 0; 689 error: 690 unlock_page(page); 691 return ret; 692 } 693 694 static 695 void cache_page_release(nfs_readdir_descriptor_t *desc) 696 { 697 if (!desc->page->mapping) 698 nfs_readdir_clear_array(desc->page); 699 page_cache_release(desc->page); 700 desc->page = NULL; 701 } 702 703 static 704 struct page *get_cache_page(nfs_readdir_descriptor_t *desc) 705 { 706 return read_cache_page(file_inode(desc->file)->i_mapping, 707 desc->page_index, (filler_t *)nfs_readdir_filler, desc); 708 } 709 710 /* 711 * Returns 0 if desc->dir_cookie was found on page desc->page_index 712 */ 713 static 714 int find_cache_page(nfs_readdir_descriptor_t *desc) 715 { 716 int res; 717 718 desc->page = get_cache_page(desc); 719 if (IS_ERR(desc->page)) 720 return PTR_ERR(desc->page); 721 722 res = nfs_readdir_search_array(desc); 723 if (res != 0) 724 cache_page_release(desc); 725 return res; 726 } 727 728 /* Search for desc->dir_cookie from the beginning of the page cache */ 729 static inline 730 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 731 { 732 int res; 733 734 if (desc->page_index == 0) { 735 desc->current_index = 0; 736 desc->last_cookie = 0; 737 } 738 do { 739 res = find_cache_page(desc); 740 } while (res == -EAGAIN); 741 return res; 742 } 743 744 /* 745 * Once we've found the start of the dirent within a page: fill 'er up... 746 */ 747 static 748 int nfs_do_filldir(nfs_readdir_descriptor_t *desc) 749 { 750 struct file *file = desc->file; 751 int i = 0; 752 int res = 0; 753 struct nfs_cache_array *array = NULL; 754 struct nfs_open_dir_context *ctx = file->private_data; 755 756 array = nfs_readdir_get_array(desc->page); 757 if (IS_ERR(array)) { 758 res = PTR_ERR(array); 759 goto out; 760 } 761 762 for (i = desc->cache_entry_index; i < array->size; i++) { 763 struct nfs_cache_array_entry *ent; 764 765 ent = &array->array[i]; 766 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, 767 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 768 desc->eof = 1; 769 break; 770 } 771 desc->ctx->pos++; 772 if (i < (array->size-1)) 773 *desc->dir_cookie = array->array[i+1].cookie; 774 else 775 *desc->dir_cookie = array->last_cookie; 776 if (ctx->duped != 0) 777 ctx->duped = 1; 778 } 779 if (array->eof_index >= 0) 780 desc->eof = 1; 781 782 nfs_readdir_release_array(desc->page); 783 out: 784 cache_page_release(desc); 785 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 786 (unsigned long long)*desc->dir_cookie, res); 787 return res; 788 } 789 790 /* 791 * If we cannot find a cookie in our cache, we suspect that this is 792 * because it points to a deleted file, so we ask the server to return 793 * whatever it thinks is the next entry. We then feed this to filldir. 794 * If all goes well, we should then be able to find our way round the 795 * cache on the next call to readdir_search_pagecache(); 796 * 797 * NOTE: we cannot add the anonymous page to the pagecache because 798 * the data it contains might not be page aligned. Besides, 799 * we should already have a complete representation of the 800 * directory in the page cache by the time we get here. 801 */ 802 static inline 803 int uncached_readdir(nfs_readdir_descriptor_t *desc) 804 { 805 struct page *page = NULL; 806 int status; 807 struct inode *inode = file_inode(desc->file); 808 struct nfs_open_dir_context *ctx = desc->file->private_data; 809 810 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 811 (unsigned long long)*desc->dir_cookie); 812 813 page = alloc_page(GFP_HIGHUSER); 814 if (!page) { 815 status = -ENOMEM; 816 goto out; 817 } 818 819 desc->page_index = 0; 820 desc->last_cookie = *desc->dir_cookie; 821 desc->page = page; 822 ctx->duped = 0; 823 824 status = nfs_readdir_xdr_to_array(desc, page, inode); 825 if (status < 0) 826 goto out_release; 827 828 status = nfs_do_filldir(desc); 829 830 out: 831 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 832 __func__, status); 833 return status; 834 out_release: 835 cache_page_release(desc); 836 goto out; 837 } 838 839 static bool nfs_dir_mapping_need_revalidate(struct inode *dir) 840 { 841 struct nfs_inode *nfsi = NFS_I(dir); 842 843 if (nfs_attribute_cache_expired(dir)) 844 return true; 845 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 846 return true; 847 return false; 848 } 849 850 /* The file offset position represents the dirent entry number. A 851 last cookie cache takes care of the common case of reading the 852 whole directory. 853 */ 854 static int nfs_readdir(struct file *file, struct dir_context *ctx) 855 { 856 struct dentry *dentry = file->f_path.dentry; 857 struct inode *inode = dentry->d_inode; 858 nfs_readdir_descriptor_t my_desc, 859 *desc = &my_desc; 860 struct nfs_open_dir_context *dir_ctx = file->private_data; 861 int res = 0; 862 863 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 864 file, (long long)ctx->pos); 865 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 866 867 /* 868 * ctx->pos points to the dirent entry number. 869 * *desc->dir_cookie has the cookie for the next entry. We have 870 * to either find the entry with the appropriate number or 871 * revalidate the cookie. 872 */ 873 memset(desc, 0, sizeof(*desc)); 874 875 desc->file = file; 876 desc->ctx = ctx; 877 desc->dir_cookie = &dir_ctx->dir_cookie; 878 desc->decode = NFS_PROTO(inode)->decode_dirent; 879 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0; 880 881 nfs_block_sillyrename(dentry); 882 if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode)) 883 res = nfs_revalidate_mapping(inode, file->f_mapping); 884 if (res < 0) 885 goto out; 886 887 do { 888 res = readdir_search_pagecache(desc); 889 890 if (res == -EBADCOOKIE) { 891 res = 0; 892 /* This means either end of directory */ 893 if (*desc->dir_cookie && desc->eof == 0) { 894 /* Or that the server has 'lost' a cookie */ 895 res = uncached_readdir(desc); 896 if (res == 0) 897 continue; 898 } 899 break; 900 } 901 if (res == -ETOOSMALL && desc->plus) { 902 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 903 nfs_zap_caches(inode); 904 desc->page_index = 0; 905 desc->plus = 0; 906 desc->eof = 0; 907 continue; 908 } 909 if (res < 0) 910 break; 911 912 res = nfs_do_filldir(desc); 913 if (res < 0) 914 break; 915 } while (!desc->eof); 916 out: 917 nfs_unblock_sillyrename(dentry); 918 if (res > 0) 919 res = 0; 920 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 921 return res; 922 } 923 924 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 925 { 926 struct inode *inode = file_inode(filp); 927 struct nfs_open_dir_context *dir_ctx = filp->private_data; 928 929 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 930 filp, offset, whence); 931 932 mutex_lock(&inode->i_mutex); 933 switch (whence) { 934 case 1: 935 offset += filp->f_pos; 936 case 0: 937 if (offset >= 0) 938 break; 939 default: 940 offset = -EINVAL; 941 goto out; 942 } 943 if (offset != filp->f_pos) { 944 filp->f_pos = offset; 945 dir_ctx->dir_cookie = 0; 946 dir_ctx->duped = 0; 947 } 948 out: 949 mutex_unlock(&inode->i_mutex); 950 return offset; 951 } 952 953 /* 954 * All directory operations under NFS are synchronous, so fsync() 955 * is a dummy operation. 956 */ 957 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 958 int datasync) 959 { 960 struct inode *inode = file_inode(filp); 961 962 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 963 964 mutex_lock(&inode->i_mutex); 965 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 966 mutex_unlock(&inode->i_mutex); 967 return 0; 968 } 969 970 /** 971 * nfs_force_lookup_revalidate - Mark the directory as having changed 972 * @dir - pointer to directory inode 973 * 974 * This forces the revalidation code in nfs_lookup_revalidate() to do a 975 * full lookup on all child dentries of 'dir' whenever a change occurs 976 * on the server that might have invalidated our dcache. 977 * 978 * The caller should be holding dir->i_lock 979 */ 980 void nfs_force_lookup_revalidate(struct inode *dir) 981 { 982 NFS_I(dir)->cache_change_attribute++; 983 } 984 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 985 986 /* 987 * A check for whether or not the parent directory has changed. 988 * In the case it has, we assume that the dentries are untrustworthy 989 * and may need to be looked up again. 990 * If rcu_walk prevents us from performing a full check, return 0. 991 */ 992 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 993 int rcu_walk) 994 { 995 int ret; 996 997 if (IS_ROOT(dentry)) 998 return 1; 999 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1000 return 0; 1001 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1002 return 0; 1003 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1004 if (rcu_walk) 1005 ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir); 1006 else 1007 ret = nfs_revalidate_inode(NFS_SERVER(dir), dir); 1008 if (ret < 0) 1009 return 0; 1010 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1011 return 0; 1012 return 1; 1013 } 1014 1015 /* 1016 * Use intent information to check whether or not we're going to do 1017 * an O_EXCL create using this path component. 1018 */ 1019 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1020 { 1021 if (NFS_PROTO(dir)->version == 2) 1022 return 0; 1023 return flags & LOOKUP_EXCL; 1024 } 1025 1026 /* 1027 * Inode and filehandle revalidation for lookups. 1028 * 1029 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1030 * or if the intent information indicates that we're about to open this 1031 * particular file and the "nocto" mount flag is not set. 1032 * 1033 */ 1034 static 1035 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1036 { 1037 struct nfs_server *server = NFS_SERVER(inode); 1038 int ret; 1039 1040 if (IS_AUTOMOUNT(inode)) 1041 return 0; 1042 /* VFS wants an on-the-wire revalidation */ 1043 if (flags & LOOKUP_REVAL) 1044 goto out_force; 1045 /* This is an open(2) */ 1046 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && 1047 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) 1048 goto out_force; 1049 out: 1050 return (inode->i_nlink == 0) ? -ENOENT : 0; 1051 out_force: 1052 if (flags & LOOKUP_RCU) 1053 return -ECHILD; 1054 ret = __nfs_revalidate_inode(server, inode); 1055 if (ret != 0) 1056 return ret; 1057 goto out; 1058 } 1059 1060 /* 1061 * We judge how long we want to trust negative 1062 * dentries by looking at the parent inode mtime. 1063 * 1064 * If parent mtime has changed, we revalidate, else we wait for a 1065 * period corresponding to the parent's attribute cache timeout value. 1066 * 1067 * If LOOKUP_RCU prevents us from performing a full check, return 1 1068 * suggesting a reval is needed. 1069 */ 1070 static inline 1071 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1072 unsigned int flags) 1073 { 1074 /* Don't revalidate a negative dentry if we're creating a new file */ 1075 if (flags & LOOKUP_CREATE) 1076 return 0; 1077 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1078 return 1; 1079 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1080 } 1081 1082 /* 1083 * This is called every time the dcache has a lookup hit, 1084 * and we should check whether we can really trust that 1085 * lookup. 1086 * 1087 * NOTE! The hit can be a negative hit too, don't assume 1088 * we have an inode! 1089 * 1090 * If the parent directory is seen to have changed, we throw out the 1091 * cached dentry and do a new lookup. 1092 */ 1093 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1094 { 1095 struct inode *dir; 1096 struct inode *inode; 1097 struct dentry *parent; 1098 struct nfs_fh *fhandle = NULL; 1099 struct nfs_fattr *fattr = NULL; 1100 struct nfs4_label *label = NULL; 1101 int error; 1102 1103 if (flags & LOOKUP_RCU) { 1104 parent = ACCESS_ONCE(dentry->d_parent); 1105 dir = ACCESS_ONCE(parent->d_inode); 1106 if (!dir) 1107 return -ECHILD; 1108 } else { 1109 parent = dget_parent(dentry); 1110 dir = parent->d_inode; 1111 } 1112 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1113 inode = dentry->d_inode; 1114 1115 if (!inode) { 1116 if (nfs_neg_need_reval(dir, dentry, flags)) { 1117 if (flags & LOOKUP_RCU) 1118 return -ECHILD; 1119 goto out_bad; 1120 } 1121 goto out_valid_noent; 1122 } 1123 1124 if (is_bad_inode(inode)) { 1125 if (flags & LOOKUP_RCU) 1126 return -ECHILD; 1127 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1128 __func__, dentry); 1129 goto out_bad; 1130 } 1131 1132 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) 1133 goto out_set_verifier; 1134 1135 /* Force a full look up iff the parent directory has changed */ 1136 if (!nfs_is_exclusive_create(dir, flags) && 1137 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1138 1139 if (nfs_lookup_verify_inode(inode, flags)) { 1140 if (flags & LOOKUP_RCU) 1141 return -ECHILD; 1142 goto out_zap_parent; 1143 } 1144 goto out_valid; 1145 } 1146 1147 if (flags & LOOKUP_RCU) 1148 return -ECHILD; 1149 1150 if (NFS_STALE(inode)) 1151 goto out_bad; 1152 1153 error = -ENOMEM; 1154 fhandle = nfs_alloc_fhandle(); 1155 fattr = nfs_alloc_fattr(); 1156 if (fhandle == NULL || fattr == NULL) 1157 goto out_error; 1158 1159 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 1160 if (IS_ERR(label)) 1161 goto out_error; 1162 1163 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1164 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1165 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1166 if (error) 1167 goto out_bad; 1168 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1169 goto out_bad; 1170 if ((error = nfs_refresh_inode(inode, fattr)) != 0) 1171 goto out_bad; 1172 1173 nfs_setsecurity(inode, fattr, label); 1174 1175 nfs_free_fattr(fattr); 1176 nfs_free_fhandle(fhandle); 1177 nfs4_label_free(label); 1178 1179 out_set_verifier: 1180 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1181 out_valid: 1182 /* Success: notify readdir to use READDIRPLUS */ 1183 nfs_advise_use_readdirplus(dir); 1184 out_valid_noent: 1185 if (flags & LOOKUP_RCU) { 1186 if (parent != ACCESS_ONCE(dentry->d_parent)) 1187 return -ECHILD; 1188 } else 1189 dput(parent); 1190 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1191 __func__, dentry); 1192 return 1; 1193 out_zap_parent: 1194 nfs_zap_caches(dir); 1195 out_bad: 1196 WARN_ON(flags & LOOKUP_RCU); 1197 nfs_free_fattr(fattr); 1198 nfs_free_fhandle(fhandle); 1199 nfs4_label_free(label); 1200 nfs_mark_for_revalidate(dir); 1201 if (inode && S_ISDIR(inode->i_mode)) { 1202 /* Purge readdir caches. */ 1203 nfs_zap_caches(inode); 1204 /* 1205 * We can't d_drop the root of a disconnected tree: 1206 * its d_hash is on the s_anon list and d_drop() would hide 1207 * it from shrink_dcache_for_unmount(), leading to busy 1208 * inodes on unmount and further oopses. 1209 */ 1210 if (IS_ROOT(dentry)) 1211 goto out_valid; 1212 } 1213 dput(parent); 1214 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1215 __func__, dentry); 1216 return 0; 1217 out_error: 1218 WARN_ON(flags & LOOKUP_RCU); 1219 nfs_free_fattr(fattr); 1220 nfs_free_fhandle(fhandle); 1221 nfs4_label_free(label); 1222 dput(parent); 1223 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1224 __func__, dentry, error); 1225 return error; 1226 } 1227 1228 /* 1229 * A weaker form of d_revalidate for revalidating just the dentry->d_inode 1230 * when we don't really care about the dentry name. This is called when a 1231 * pathwalk ends on a dentry that was not found via a normal lookup in the 1232 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1233 * 1234 * In this situation, we just want to verify that the inode itself is OK 1235 * since the dentry might have changed on the server. 1236 */ 1237 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1238 { 1239 int error; 1240 struct inode *inode = dentry->d_inode; 1241 1242 /* 1243 * I believe we can only get a negative dentry here in the case of a 1244 * procfs-style symlink. Just assume it's correct for now, but we may 1245 * eventually need to do something more here. 1246 */ 1247 if (!inode) { 1248 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1249 __func__, dentry); 1250 return 1; 1251 } 1252 1253 if (is_bad_inode(inode)) { 1254 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1255 __func__, dentry); 1256 return 0; 1257 } 1258 1259 error = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1260 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1261 __func__, inode->i_ino, error ? "invalid" : "valid"); 1262 return !error; 1263 } 1264 1265 /* 1266 * This is called from dput() when d_count is going to 0. 1267 */ 1268 static int nfs_dentry_delete(const struct dentry *dentry) 1269 { 1270 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1271 dentry, dentry->d_flags); 1272 1273 /* Unhash any dentry with a stale inode */ 1274 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) 1275 return 1; 1276 1277 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1278 /* Unhash it, so that ->d_iput() would be called */ 1279 return 1; 1280 } 1281 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 1282 /* Unhash it, so that ancestors of killed async unlink 1283 * files will be cleaned up during umount */ 1284 return 1; 1285 } 1286 return 0; 1287 1288 } 1289 1290 /* Ensure that we revalidate inode->i_nlink */ 1291 static void nfs_drop_nlink(struct inode *inode) 1292 { 1293 spin_lock(&inode->i_lock); 1294 /* drop the inode if we're reasonably sure this is the last link */ 1295 if (inode->i_nlink == 1) 1296 clear_nlink(inode); 1297 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; 1298 spin_unlock(&inode->i_lock); 1299 } 1300 1301 /* 1302 * Called when the dentry loses inode. 1303 * We use it to clean up silly-renamed files. 1304 */ 1305 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1306 { 1307 if (S_ISDIR(inode->i_mode)) 1308 /* drop any readdir cache as it could easily be old */ 1309 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1310 1311 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1312 nfs_complete_unlink(dentry, inode); 1313 nfs_drop_nlink(inode); 1314 } 1315 iput(inode); 1316 } 1317 1318 static void nfs_d_release(struct dentry *dentry) 1319 { 1320 /* free cached devname value, if it survived that far */ 1321 if (unlikely(dentry->d_fsdata)) { 1322 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1323 WARN_ON(1); 1324 else 1325 kfree(dentry->d_fsdata); 1326 } 1327 } 1328 1329 const struct dentry_operations nfs_dentry_operations = { 1330 .d_revalidate = nfs_lookup_revalidate, 1331 .d_weak_revalidate = nfs_weak_revalidate, 1332 .d_delete = nfs_dentry_delete, 1333 .d_iput = nfs_dentry_iput, 1334 .d_automount = nfs_d_automount, 1335 .d_release = nfs_d_release, 1336 }; 1337 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1338 1339 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1340 { 1341 struct dentry *res; 1342 struct dentry *parent; 1343 struct inode *inode = NULL; 1344 struct nfs_fh *fhandle = NULL; 1345 struct nfs_fattr *fattr = NULL; 1346 struct nfs4_label *label = NULL; 1347 int error; 1348 1349 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1350 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1351 1352 res = ERR_PTR(-ENAMETOOLONG); 1353 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1354 goto out; 1355 1356 /* 1357 * If we're doing an exclusive create, optimize away the lookup 1358 * but don't hash the dentry. 1359 */ 1360 if (nfs_is_exclusive_create(dir, flags)) { 1361 d_instantiate(dentry, NULL); 1362 res = NULL; 1363 goto out; 1364 } 1365 1366 res = ERR_PTR(-ENOMEM); 1367 fhandle = nfs_alloc_fhandle(); 1368 fattr = nfs_alloc_fattr(); 1369 if (fhandle == NULL || fattr == NULL) 1370 goto out; 1371 1372 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1373 if (IS_ERR(label)) 1374 goto out; 1375 1376 parent = dentry->d_parent; 1377 /* Protect against concurrent sillydeletes */ 1378 trace_nfs_lookup_enter(dir, dentry, flags); 1379 nfs_block_sillyrename(parent); 1380 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1381 if (error == -ENOENT) 1382 goto no_entry; 1383 if (error < 0) { 1384 res = ERR_PTR(error); 1385 goto out_unblock_sillyrename; 1386 } 1387 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1388 res = ERR_CAST(inode); 1389 if (IS_ERR(res)) 1390 goto out_unblock_sillyrename; 1391 1392 /* Success: notify readdir to use READDIRPLUS */ 1393 nfs_advise_use_readdirplus(dir); 1394 1395 no_entry: 1396 res = d_materialise_unique(dentry, inode); 1397 if (res != NULL) { 1398 if (IS_ERR(res)) 1399 goto out_unblock_sillyrename; 1400 dentry = res; 1401 } 1402 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1403 out_unblock_sillyrename: 1404 nfs_unblock_sillyrename(parent); 1405 trace_nfs_lookup_exit(dir, dentry, flags, error); 1406 nfs4_label_free(label); 1407 out: 1408 nfs_free_fattr(fattr); 1409 nfs_free_fhandle(fhandle); 1410 return res; 1411 } 1412 EXPORT_SYMBOL_GPL(nfs_lookup); 1413 1414 #if IS_ENABLED(CONFIG_NFS_V4) 1415 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1416 1417 const struct dentry_operations nfs4_dentry_operations = { 1418 .d_revalidate = nfs4_lookup_revalidate, 1419 .d_delete = nfs_dentry_delete, 1420 .d_iput = nfs_dentry_iput, 1421 .d_automount = nfs_d_automount, 1422 .d_release = nfs_d_release, 1423 }; 1424 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1425 1426 static fmode_t flags_to_mode(int flags) 1427 { 1428 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1429 if ((flags & O_ACCMODE) != O_WRONLY) 1430 res |= FMODE_READ; 1431 if ((flags & O_ACCMODE) != O_RDONLY) 1432 res |= FMODE_WRITE; 1433 return res; 1434 } 1435 1436 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags) 1437 { 1438 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags)); 1439 } 1440 1441 static int do_open(struct inode *inode, struct file *filp) 1442 { 1443 nfs_fscache_open_file(inode, filp); 1444 return 0; 1445 } 1446 1447 static int nfs_finish_open(struct nfs_open_context *ctx, 1448 struct dentry *dentry, 1449 struct file *file, unsigned open_flags, 1450 int *opened) 1451 { 1452 int err; 1453 1454 if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) 1455 *opened |= FILE_CREATED; 1456 1457 err = finish_open(file, dentry, do_open, opened); 1458 if (err) 1459 goto out; 1460 nfs_file_set_open_context(file, ctx); 1461 1462 out: 1463 return err; 1464 } 1465 1466 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1467 struct file *file, unsigned open_flags, 1468 umode_t mode, int *opened) 1469 { 1470 struct nfs_open_context *ctx; 1471 struct dentry *res; 1472 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1473 struct inode *inode; 1474 unsigned int lookup_flags = 0; 1475 int err; 1476 1477 /* Expect a negative dentry */ 1478 BUG_ON(dentry->d_inode); 1479 1480 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1481 dir->i_sb->s_id, dir->i_ino, dentry); 1482 1483 err = nfs_check_flags(open_flags); 1484 if (err) 1485 return err; 1486 1487 /* NFS only supports OPEN on regular files */ 1488 if ((open_flags & O_DIRECTORY)) { 1489 if (!d_unhashed(dentry)) { 1490 /* 1491 * Hashed negative dentry with O_DIRECTORY: dentry was 1492 * revalidated and is fine, no need to perform lookup 1493 * again 1494 */ 1495 return -ENOENT; 1496 } 1497 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1498 goto no_open; 1499 } 1500 1501 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1502 return -ENAMETOOLONG; 1503 1504 if (open_flags & O_CREAT) { 1505 attr.ia_valid |= ATTR_MODE; 1506 attr.ia_mode = mode & ~current_umask(); 1507 } 1508 if (open_flags & O_TRUNC) { 1509 attr.ia_valid |= ATTR_SIZE; 1510 attr.ia_size = 0; 1511 } 1512 1513 ctx = create_nfs_open_context(dentry, open_flags); 1514 err = PTR_ERR(ctx); 1515 if (IS_ERR(ctx)) 1516 goto out; 1517 1518 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1519 nfs_block_sillyrename(dentry->d_parent); 1520 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened); 1521 nfs_unblock_sillyrename(dentry->d_parent); 1522 if (IS_ERR(inode)) { 1523 err = PTR_ERR(inode); 1524 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1525 put_nfs_open_context(ctx); 1526 switch (err) { 1527 case -ENOENT: 1528 d_drop(dentry); 1529 d_add(dentry, NULL); 1530 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1531 break; 1532 case -EISDIR: 1533 case -ENOTDIR: 1534 goto no_open; 1535 case -ELOOP: 1536 if (!(open_flags & O_NOFOLLOW)) 1537 goto no_open; 1538 break; 1539 /* case -EINVAL: */ 1540 default: 1541 break; 1542 } 1543 goto out; 1544 } 1545 1546 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened); 1547 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1548 put_nfs_open_context(ctx); 1549 out: 1550 return err; 1551 1552 no_open: 1553 res = nfs_lookup(dir, dentry, lookup_flags); 1554 err = PTR_ERR(res); 1555 if (IS_ERR(res)) 1556 goto out; 1557 1558 return finish_no_open(file, res); 1559 } 1560 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1561 1562 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1563 { 1564 struct inode *inode; 1565 int ret = 0; 1566 1567 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1568 goto no_open; 1569 if (d_mountpoint(dentry)) 1570 goto no_open; 1571 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) 1572 goto no_open; 1573 1574 inode = dentry->d_inode; 1575 1576 /* We can't create new files in nfs_open_revalidate(), so we 1577 * optimize away revalidation of negative dentries. 1578 */ 1579 if (inode == NULL) { 1580 struct dentry *parent; 1581 struct inode *dir; 1582 1583 if (flags & LOOKUP_RCU) { 1584 parent = ACCESS_ONCE(dentry->d_parent); 1585 dir = ACCESS_ONCE(parent->d_inode); 1586 if (!dir) 1587 return -ECHILD; 1588 } else { 1589 parent = dget_parent(dentry); 1590 dir = parent->d_inode; 1591 } 1592 if (!nfs_neg_need_reval(dir, dentry, flags)) 1593 ret = 1; 1594 else if (flags & LOOKUP_RCU) 1595 ret = -ECHILD; 1596 if (!(flags & LOOKUP_RCU)) 1597 dput(parent); 1598 else if (parent != ACCESS_ONCE(dentry->d_parent)) 1599 return -ECHILD; 1600 goto out; 1601 } 1602 1603 /* NFS only supports OPEN on regular files */ 1604 if (!S_ISREG(inode->i_mode)) 1605 goto no_open; 1606 /* We cannot do exclusive creation on a positive dentry */ 1607 if (flags & LOOKUP_EXCL) 1608 goto no_open; 1609 1610 /* Let f_op->open() actually open (and revalidate) the file */ 1611 ret = 1; 1612 1613 out: 1614 return ret; 1615 1616 no_open: 1617 return nfs_lookup_revalidate(dentry, flags); 1618 } 1619 1620 #endif /* CONFIG_NFSV4 */ 1621 1622 /* 1623 * Code common to create, mkdir, and mknod. 1624 */ 1625 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1626 struct nfs_fattr *fattr, 1627 struct nfs4_label *label) 1628 { 1629 struct dentry *parent = dget_parent(dentry); 1630 struct inode *dir = parent->d_inode; 1631 struct inode *inode; 1632 int error = -EACCES; 1633 1634 d_drop(dentry); 1635 1636 /* We may have been initialized further down */ 1637 if (dentry->d_inode) 1638 goto out; 1639 if (fhandle->size == 0) { 1640 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); 1641 if (error) 1642 goto out_error; 1643 } 1644 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1645 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1646 struct nfs_server *server = NFS_SB(dentry->d_sb); 1647 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); 1648 if (error < 0) 1649 goto out_error; 1650 } 1651 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1652 error = PTR_ERR(inode); 1653 if (IS_ERR(inode)) 1654 goto out_error; 1655 d_add(dentry, inode); 1656 out: 1657 dput(parent); 1658 return 0; 1659 out_error: 1660 nfs_mark_for_revalidate(dir); 1661 dput(parent); 1662 return error; 1663 } 1664 EXPORT_SYMBOL_GPL(nfs_instantiate); 1665 1666 /* 1667 * Following a failed create operation, we drop the dentry rather 1668 * than retain a negative dentry. This avoids a problem in the event 1669 * that the operation succeeded on the server, but an error in the 1670 * reply path made it appear to have failed. 1671 */ 1672 int nfs_create(struct inode *dir, struct dentry *dentry, 1673 umode_t mode, bool excl) 1674 { 1675 struct iattr attr; 1676 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 1677 int error; 1678 1679 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 1680 dir->i_sb->s_id, dir->i_ino, dentry); 1681 1682 attr.ia_mode = mode; 1683 attr.ia_valid = ATTR_MODE; 1684 1685 trace_nfs_create_enter(dir, dentry, open_flags); 1686 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1687 trace_nfs_create_exit(dir, dentry, open_flags, error); 1688 if (error != 0) 1689 goto out_err; 1690 return 0; 1691 out_err: 1692 d_drop(dentry); 1693 return error; 1694 } 1695 EXPORT_SYMBOL_GPL(nfs_create); 1696 1697 /* 1698 * See comments for nfs_proc_create regarding failed operations. 1699 */ 1700 int 1701 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) 1702 { 1703 struct iattr attr; 1704 int status; 1705 1706 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 1707 dir->i_sb->s_id, dir->i_ino, dentry); 1708 1709 if (!new_valid_dev(rdev)) 1710 return -EINVAL; 1711 1712 attr.ia_mode = mode; 1713 attr.ia_valid = ATTR_MODE; 1714 1715 trace_nfs_mknod_enter(dir, dentry); 1716 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1717 trace_nfs_mknod_exit(dir, dentry, status); 1718 if (status != 0) 1719 goto out_err; 1720 return 0; 1721 out_err: 1722 d_drop(dentry); 1723 return status; 1724 } 1725 EXPORT_SYMBOL_GPL(nfs_mknod); 1726 1727 /* 1728 * See comments for nfs_proc_create regarding failed operations. 1729 */ 1730 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 1731 { 1732 struct iattr attr; 1733 int error; 1734 1735 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 1736 dir->i_sb->s_id, dir->i_ino, dentry); 1737 1738 attr.ia_valid = ATTR_MODE; 1739 attr.ia_mode = mode | S_IFDIR; 1740 1741 trace_nfs_mkdir_enter(dir, dentry); 1742 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1743 trace_nfs_mkdir_exit(dir, dentry, error); 1744 if (error != 0) 1745 goto out_err; 1746 return 0; 1747 out_err: 1748 d_drop(dentry); 1749 return error; 1750 } 1751 EXPORT_SYMBOL_GPL(nfs_mkdir); 1752 1753 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1754 { 1755 if (dentry->d_inode != NULL && !d_unhashed(dentry)) 1756 d_delete(dentry); 1757 } 1758 1759 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1760 { 1761 int error; 1762 1763 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 1764 dir->i_sb->s_id, dir->i_ino, dentry); 1765 1766 trace_nfs_rmdir_enter(dir, dentry); 1767 if (dentry->d_inode) { 1768 nfs_wait_on_sillyrename(dentry); 1769 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1770 /* Ensure the VFS deletes this inode */ 1771 switch (error) { 1772 case 0: 1773 clear_nlink(dentry->d_inode); 1774 break; 1775 case -ENOENT: 1776 nfs_dentry_handle_enoent(dentry); 1777 } 1778 } else 1779 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1780 trace_nfs_rmdir_exit(dir, dentry, error); 1781 1782 return error; 1783 } 1784 EXPORT_SYMBOL_GPL(nfs_rmdir); 1785 1786 /* 1787 * Remove a file after making sure there are no pending writes, 1788 * and after checking that the file has only one user. 1789 * 1790 * We invalidate the attribute cache and free the inode prior to the operation 1791 * to avoid possible races if the server reuses the inode. 1792 */ 1793 static int nfs_safe_remove(struct dentry *dentry) 1794 { 1795 struct inode *dir = dentry->d_parent->d_inode; 1796 struct inode *inode = dentry->d_inode; 1797 int error = -EBUSY; 1798 1799 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 1800 1801 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1802 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1803 error = 0; 1804 goto out; 1805 } 1806 1807 trace_nfs_remove_enter(dir, dentry); 1808 if (inode != NULL) { 1809 NFS_PROTO(inode)->return_delegation(inode); 1810 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1811 if (error == 0) 1812 nfs_drop_nlink(inode); 1813 } else 1814 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1815 if (error == -ENOENT) 1816 nfs_dentry_handle_enoent(dentry); 1817 trace_nfs_remove_exit(dir, dentry, error); 1818 out: 1819 return error; 1820 } 1821 1822 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1823 * belongs to an active ".nfs..." file and we return -EBUSY. 1824 * 1825 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1826 */ 1827 int nfs_unlink(struct inode *dir, struct dentry *dentry) 1828 { 1829 int error; 1830 int need_rehash = 0; 1831 1832 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 1833 dir->i_ino, dentry); 1834 1835 trace_nfs_unlink_enter(dir, dentry); 1836 spin_lock(&dentry->d_lock); 1837 if (d_count(dentry) > 1) { 1838 spin_unlock(&dentry->d_lock); 1839 /* Start asynchronous writeout of the inode */ 1840 write_inode_now(dentry->d_inode, 0); 1841 error = nfs_sillyrename(dir, dentry); 1842 goto out; 1843 } 1844 if (!d_unhashed(dentry)) { 1845 __d_drop(dentry); 1846 need_rehash = 1; 1847 } 1848 spin_unlock(&dentry->d_lock); 1849 error = nfs_safe_remove(dentry); 1850 if (!error || error == -ENOENT) { 1851 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1852 } else if (need_rehash) 1853 d_rehash(dentry); 1854 out: 1855 trace_nfs_unlink_exit(dir, dentry, error); 1856 return error; 1857 } 1858 EXPORT_SYMBOL_GPL(nfs_unlink); 1859 1860 /* 1861 * To create a symbolic link, most file systems instantiate a new inode, 1862 * add a page to it containing the path, then write it out to the disk 1863 * using prepare_write/commit_write. 1864 * 1865 * Unfortunately the NFS client can't create the in-core inode first 1866 * because it needs a file handle to create an in-core inode (see 1867 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1868 * symlink request has completed on the server. 1869 * 1870 * So instead we allocate a raw page, copy the symname into it, then do 1871 * the SYMLINK request with the page as the buffer. If it succeeds, we 1872 * now have a new file handle and can instantiate an in-core NFS inode 1873 * and move the raw page into its mapping. 1874 */ 1875 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1876 { 1877 struct page *page; 1878 char *kaddr; 1879 struct iattr attr; 1880 unsigned int pathlen = strlen(symname); 1881 int error; 1882 1883 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 1884 dir->i_ino, dentry, symname); 1885 1886 if (pathlen > PAGE_SIZE) 1887 return -ENAMETOOLONG; 1888 1889 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1890 attr.ia_valid = ATTR_MODE; 1891 1892 page = alloc_page(GFP_HIGHUSER); 1893 if (!page) 1894 return -ENOMEM; 1895 1896 kaddr = kmap_atomic(page); 1897 memcpy(kaddr, symname, pathlen); 1898 if (pathlen < PAGE_SIZE) 1899 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1900 kunmap_atomic(kaddr); 1901 1902 trace_nfs_symlink_enter(dir, dentry); 1903 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1904 trace_nfs_symlink_exit(dir, dentry, error); 1905 if (error != 0) { 1906 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 1907 dir->i_sb->s_id, dir->i_ino, 1908 dentry, symname, error); 1909 d_drop(dentry); 1910 __free_page(page); 1911 return error; 1912 } 1913 1914 /* 1915 * No big deal if we can't add this page to the page cache here. 1916 * READLINK will get the missing page from the server if needed. 1917 */ 1918 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0, 1919 GFP_KERNEL)) { 1920 SetPageUptodate(page); 1921 unlock_page(page); 1922 /* 1923 * add_to_page_cache_lru() grabs an extra page refcount. 1924 * Drop it here to avoid leaking this page later. 1925 */ 1926 page_cache_release(page); 1927 } else 1928 __free_page(page); 1929 1930 return 0; 1931 } 1932 EXPORT_SYMBOL_GPL(nfs_symlink); 1933 1934 int 1935 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1936 { 1937 struct inode *inode = old_dentry->d_inode; 1938 int error; 1939 1940 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 1941 old_dentry, dentry); 1942 1943 trace_nfs_link_enter(inode, dir, dentry); 1944 NFS_PROTO(inode)->return_delegation(inode); 1945 1946 d_drop(dentry); 1947 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1948 if (error == 0) { 1949 ihold(inode); 1950 d_add(dentry, inode); 1951 } 1952 trace_nfs_link_exit(inode, dir, dentry, error); 1953 return error; 1954 } 1955 EXPORT_SYMBOL_GPL(nfs_link); 1956 1957 /* 1958 * RENAME 1959 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1960 * different file handle for the same inode after a rename (e.g. when 1961 * moving to a different directory). A fail-safe method to do so would 1962 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1963 * rename the old file using the sillyrename stuff. This way, the original 1964 * file in old_dir will go away when the last process iput()s the inode. 1965 * 1966 * FIXED. 1967 * 1968 * It actually works quite well. One needs to have the possibility for 1969 * at least one ".nfs..." file in each directory the file ever gets 1970 * moved or linked to which happens automagically with the new 1971 * implementation that only depends on the dcache stuff instead of 1972 * using the inode layer 1973 * 1974 * Unfortunately, things are a little more complicated than indicated 1975 * above. For a cross-directory move, we want to make sure we can get 1976 * rid of the old inode after the operation. This means there must be 1977 * no pending writes (if it's a file), and the use count must be 1. 1978 * If these conditions are met, we can drop the dentries before doing 1979 * the rename. 1980 */ 1981 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1982 struct inode *new_dir, struct dentry *new_dentry) 1983 { 1984 struct inode *old_inode = old_dentry->d_inode; 1985 struct inode *new_inode = new_dentry->d_inode; 1986 struct dentry *dentry = NULL, *rehash = NULL; 1987 struct rpc_task *task; 1988 int error = -EBUSY; 1989 1990 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 1991 old_dentry, new_dentry, 1992 d_count(new_dentry)); 1993 1994 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 1995 /* 1996 * For non-directories, check whether the target is busy and if so, 1997 * make a copy of the dentry and then do a silly-rename. If the 1998 * silly-rename succeeds, the copied dentry is hashed and becomes 1999 * the new target. 2000 */ 2001 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2002 /* 2003 * To prevent any new references to the target during the 2004 * rename, we unhash the dentry in advance. 2005 */ 2006 if (!d_unhashed(new_dentry)) { 2007 d_drop(new_dentry); 2008 rehash = new_dentry; 2009 } 2010 2011 if (d_count(new_dentry) > 2) { 2012 int err; 2013 2014 /* copy the target dentry's name */ 2015 dentry = d_alloc(new_dentry->d_parent, 2016 &new_dentry->d_name); 2017 if (!dentry) 2018 goto out; 2019 2020 /* silly-rename the existing target ... */ 2021 err = nfs_sillyrename(new_dir, new_dentry); 2022 if (err) 2023 goto out; 2024 2025 new_dentry = dentry; 2026 rehash = NULL; 2027 new_inode = NULL; 2028 } 2029 } 2030 2031 NFS_PROTO(old_inode)->return_delegation(old_inode); 2032 if (new_inode != NULL) 2033 NFS_PROTO(new_inode)->return_delegation(new_inode); 2034 2035 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2036 if (IS_ERR(task)) { 2037 error = PTR_ERR(task); 2038 goto out; 2039 } 2040 2041 error = rpc_wait_for_completion_task(task); 2042 if (error == 0) 2043 error = task->tk_status; 2044 rpc_put_task(task); 2045 nfs_mark_for_revalidate(old_inode); 2046 out: 2047 if (rehash) 2048 d_rehash(rehash); 2049 trace_nfs_rename_exit(old_dir, old_dentry, 2050 new_dir, new_dentry, error); 2051 if (!error) { 2052 if (new_inode != NULL) 2053 nfs_drop_nlink(new_inode); 2054 d_move(old_dentry, new_dentry); 2055 nfs_set_verifier(new_dentry, 2056 nfs_save_change_attribute(new_dir)); 2057 } else if (error == -ENOENT) 2058 nfs_dentry_handle_enoent(old_dentry); 2059 2060 /* new dentry created? */ 2061 if (dentry) 2062 dput(dentry); 2063 return error; 2064 } 2065 EXPORT_SYMBOL_GPL(nfs_rename); 2066 2067 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2068 static LIST_HEAD(nfs_access_lru_list); 2069 static atomic_long_t nfs_access_nr_entries; 2070 2071 static unsigned long nfs_access_max_cachesize = ULONG_MAX; 2072 module_param(nfs_access_max_cachesize, ulong, 0644); 2073 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2074 2075 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2076 { 2077 put_rpccred(entry->cred); 2078 kfree_rcu(entry, rcu_head); 2079 smp_mb__before_atomic(); 2080 atomic_long_dec(&nfs_access_nr_entries); 2081 smp_mb__after_atomic(); 2082 } 2083 2084 static void nfs_access_free_list(struct list_head *head) 2085 { 2086 struct nfs_access_entry *cache; 2087 2088 while (!list_empty(head)) { 2089 cache = list_entry(head->next, struct nfs_access_entry, lru); 2090 list_del(&cache->lru); 2091 nfs_access_free_entry(cache); 2092 } 2093 } 2094 2095 static unsigned long 2096 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2097 { 2098 LIST_HEAD(head); 2099 struct nfs_inode *nfsi, *next; 2100 struct nfs_access_entry *cache; 2101 long freed = 0; 2102 2103 spin_lock(&nfs_access_lru_lock); 2104 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2105 struct inode *inode; 2106 2107 if (nr_to_scan-- == 0) 2108 break; 2109 inode = &nfsi->vfs_inode; 2110 spin_lock(&inode->i_lock); 2111 if (list_empty(&nfsi->access_cache_entry_lru)) 2112 goto remove_lru_entry; 2113 cache = list_entry(nfsi->access_cache_entry_lru.next, 2114 struct nfs_access_entry, lru); 2115 list_move(&cache->lru, &head); 2116 rb_erase(&cache->rb_node, &nfsi->access_cache); 2117 freed++; 2118 if (!list_empty(&nfsi->access_cache_entry_lru)) 2119 list_move_tail(&nfsi->access_cache_inode_lru, 2120 &nfs_access_lru_list); 2121 else { 2122 remove_lru_entry: 2123 list_del_init(&nfsi->access_cache_inode_lru); 2124 smp_mb__before_atomic(); 2125 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2126 smp_mb__after_atomic(); 2127 } 2128 spin_unlock(&inode->i_lock); 2129 } 2130 spin_unlock(&nfs_access_lru_lock); 2131 nfs_access_free_list(&head); 2132 return freed; 2133 } 2134 2135 unsigned long 2136 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2137 { 2138 int nr_to_scan = sc->nr_to_scan; 2139 gfp_t gfp_mask = sc->gfp_mask; 2140 2141 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2142 return SHRINK_STOP; 2143 return nfs_do_access_cache_scan(nr_to_scan); 2144 } 2145 2146 2147 unsigned long 2148 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2149 { 2150 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2151 } 2152 2153 static void 2154 nfs_access_cache_enforce_limit(void) 2155 { 2156 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2157 unsigned long diff; 2158 unsigned int nr_to_scan; 2159 2160 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2161 return; 2162 nr_to_scan = 100; 2163 diff = nr_entries - nfs_access_max_cachesize; 2164 if (diff < nr_to_scan) 2165 nr_to_scan = diff; 2166 nfs_do_access_cache_scan(nr_to_scan); 2167 } 2168 2169 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2170 { 2171 struct rb_root *root_node = &nfsi->access_cache; 2172 struct rb_node *n; 2173 struct nfs_access_entry *entry; 2174 2175 /* Unhook entries from the cache */ 2176 while ((n = rb_first(root_node)) != NULL) { 2177 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2178 rb_erase(n, root_node); 2179 list_move(&entry->lru, head); 2180 } 2181 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2182 } 2183 2184 void nfs_access_zap_cache(struct inode *inode) 2185 { 2186 LIST_HEAD(head); 2187 2188 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2189 return; 2190 /* Remove from global LRU init */ 2191 spin_lock(&nfs_access_lru_lock); 2192 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2193 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2194 2195 spin_lock(&inode->i_lock); 2196 __nfs_access_zap_cache(NFS_I(inode), &head); 2197 spin_unlock(&inode->i_lock); 2198 spin_unlock(&nfs_access_lru_lock); 2199 nfs_access_free_list(&head); 2200 } 2201 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2202 2203 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 2204 { 2205 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2206 struct nfs_access_entry *entry; 2207 2208 while (n != NULL) { 2209 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2210 2211 if (cred < entry->cred) 2212 n = n->rb_left; 2213 else if (cred > entry->cred) 2214 n = n->rb_right; 2215 else 2216 return entry; 2217 } 2218 return NULL; 2219 } 2220 2221 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2222 { 2223 struct nfs_inode *nfsi = NFS_I(inode); 2224 struct nfs_access_entry *cache; 2225 int err = -ENOENT; 2226 2227 spin_lock(&inode->i_lock); 2228 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2229 goto out_zap; 2230 cache = nfs_access_search_rbtree(inode, cred); 2231 if (cache == NULL) 2232 goto out; 2233 if (!nfs_have_delegated_attributes(inode) && 2234 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 2235 goto out_stale; 2236 res->jiffies = cache->jiffies; 2237 res->cred = cache->cred; 2238 res->mask = cache->mask; 2239 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2240 err = 0; 2241 out: 2242 spin_unlock(&inode->i_lock); 2243 return err; 2244 out_stale: 2245 rb_erase(&cache->rb_node, &nfsi->access_cache); 2246 list_del(&cache->lru); 2247 spin_unlock(&inode->i_lock); 2248 nfs_access_free_entry(cache); 2249 return -ENOENT; 2250 out_zap: 2251 spin_unlock(&inode->i_lock); 2252 nfs_access_zap_cache(inode); 2253 return -ENOENT; 2254 } 2255 2256 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2257 { 2258 /* Only check the most recently returned cache entry, 2259 * but do it without locking. 2260 */ 2261 struct nfs_inode *nfsi = NFS_I(inode); 2262 struct nfs_access_entry *cache; 2263 int err = -ECHILD; 2264 struct list_head *lh; 2265 2266 rcu_read_lock(); 2267 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2268 goto out; 2269 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); 2270 cache = list_entry(lh, struct nfs_access_entry, lru); 2271 if (lh == &nfsi->access_cache_entry_lru || 2272 cred != cache->cred) 2273 cache = NULL; 2274 if (cache == NULL) 2275 goto out; 2276 if (!nfs_have_delegated_attributes(inode) && 2277 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 2278 goto out; 2279 res->jiffies = cache->jiffies; 2280 res->cred = cache->cred; 2281 res->mask = cache->mask; 2282 err = 0; 2283 out: 2284 rcu_read_unlock(); 2285 return err; 2286 } 2287 2288 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2289 { 2290 struct nfs_inode *nfsi = NFS_I(inode); 2291 struct rb_root *root_node = &nfsi->access_cache; 2292 struct rb_node **p = &root_node->rb_node; 2293 struct rb_node *parent = NULL; 2294 struct nfs_access_entry *entry; 2295 2296 spin_lock(&inode->i_lock); 2297 while (*p != NULL) { 2298 parent = *p; 2299 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2300 2301 if (set->cred < entry->cred) 2302 p = &parent->rb_left; 2303 else if (set->cred > entry->cred) 2304 p = &parent->rb_right; 2305 else 2306 goto found; 2307 } 2308 rb_link_node(&set->rb_node, parent, p); 2309 rb_insert_color(&set->rb_node, root_node); 2310 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2311 spin_unlock(&inode->i_lock); 2312 return; 2313 found: 2314 rb_replace_node(parent, &set->rb_node, root_node); 2315 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2316 list_del(&entry->lru); 2317 spin_unlock(&inode->i_lock); 2318 nfs_access_free_entry(entry); 2319 } 2320 2321 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2322 { 2323 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2324 if (cache == NULL) 2325 return; 2326 RB_CLEAR_NODE(&cache->rb_node); 2327 cache->jiffies = set->jiffies; 2328 cache->cred = get_rpccred(set->cred); 2329 cache->mask = set->mask; 2330 2331 /* The above field assignments must be visible 2332 * before this item appears on the lru. We cannot easily 2333 * use rcu_assign_pointer, so just force the memory barrier. 2334 */ 2335 smp_wmb(); 2336 nfs_access_add_rbtree(inode, cache); 2337 2338 /* Update accounting */ 2339 smp_mb__before_atomic(); 2340 atomic_long_inc(&nfs_access_nr_entries); 2341 smp_mb__after_atomic(); 2342 2343 /* Add inode to global LRU list */ 2344 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2345 spin_lock(&nfs_access_lru_lock); 2346 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2347 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2348 &nfs_access_lru_list); 2349 spin_unlock(&nfs_access_lru_lock); 2350 } 2351 nfs_access_cache_enforce_limit(); 2352 } 2353 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2354 2355 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2356 { 2357 entry->mask = 0; 2358 if (access_result & NFS4_ACCESS_READ) 2359 entry->mask |= MAY_READ; 2360 if (access_result & 2361 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) 2362 entry->mask |= MAY_WRITE; 2363 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) 2364 entry->mask |= MAY_EXEC; 2365 } 2366 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2367 2368 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 2369 { 2370 struct nfs_access_entry cache; 2371 int status; 2372 2373 trace_nfs_access_enter(inode); 2374 2375 status = nfs_access_get_cached_rcu(inode, cred, &cache); 2376 if (status != 0) 2377 status = nfs_access_get_cached(inode, cred, &cache); 2378 if (status == 0) 2379 goto out_cached; 2380 2381 status = -ECHILD; 2382 if (mask & MAY_NOT_BLOCK) 2383 goto out; 2384 2385 /* Be clever: ask server to check for all possible rights */ 2386 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 2387 cache.cred = cred; 2388 cache.jiffies = jiffies; 2389 status = NFS_PROTO(inode)->access(inode, &cache); 2390 if (status != 0) { 2391 if (status == -ESTALE) { 2392 nfs_zap_caches(inode); 2393 if (!S_ISDIR(inode->i_mode)) 2394 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 2395 } 2396 goto out; 2397 } 2398 nfs_access_add_cache(inode, &cache); 2399 out_cached: 2400 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2401 status = -EACCES; 2402 out: 2403 trace_nfs_access_exit(inode, status); 2404 return status; 2405 } 2406 2407 static int nfs_open_permission_mask(int openflags) 2408 { 2409 int mask = 0; 2410 2411 if (openflags & __FMODE_EXEC) { 2412 /* ONLY check exec rights */ 2413 mask = MAY_EXEC; 2414 } else { 2415 if ((openflags & O_ACCMODE) != O_WRONLY) 2416 mask |= MAY_READ; 2417 if ((openflags & O_ACCMODE) != O_RDONLY) 2418 mask |= MAY_WRITE; 2419 } 2420 2421 return mask; 2422 } 2423 2424 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 2425 { 2426 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2427 } 2428 EXPORT_SYMBOL_GPL(nfs_may_open); 2429 2430 int nfs_permission(struct inode *inode, int mask) 2431 { 2432 struct rpc_cred *cred; 2433 int res = 0; 2434 2435 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2436 2437 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2438 goto out; 2439 /* Is this sys_access() ? */ 2440 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2441 goto force_lookup; 2442 2443 switch (inode->i_mode & S_IFMT) { 2444 case S_IFLNK: 2445 goto out; 2446 case S_IFREG: 2447 break; 2448 case S_IFDIR: 2449 /* 2450 * Optimize away all write operations, since the server 2451 * will check permissions when we perform the op. 2452 */ 2453 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2454 goto out; 2455 } 2456 2457 force_lookup: 2458 if (!NFS_PROTO(inode)->access) 2459 goto out_notsup; 2460 2461 /* Always try fast lookups first */ 2462 rcu_read_lock(); 2463 cred = rpc_lookup_cred_nonblock(); 2464 if (!IS_ERR(cred)) 2465 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); 2466 else 2467 res = PTR_ERR(cred); 2468 rcu_read_unlock(); 2469 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { 2470 /* Fast lookup failed, try the slow way */ 2471 cred = rpc_lookup_cred(); 2472 if (!IS_ERR(cred)) { 2473 res = nfs_do_access(inode, cred, mask); 2474 put_rpccred(cred); 2475 } else 2476 res = PTR_ERR(cred); 2477 } 2478 out: 2479 if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) 2480 res = -EACCES; 2481 2482 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2483 inode->i_sb->s_id, inode->i_ino, mask, res); 2484 return res; 2485 out_notsup: 2486 if (mask & MAY_NOT_BLOCK) 2487 return -ECHILD; 2488 2489 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2490 if (res == 0) 2491 res = generic_permission(inode, mask); 2492 goto out; 2493 } 2494 EXPORT_SYMBOL_GPL(nfs_permission); 2495 2496 /* 2497 * Local variables: 2498 * version-control: t 2499 * kept-new-versions: 5 2500 * End: 2501 */ 2502