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