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 out: 122 put_rpccred(cred); 123 return res; 124 } 125 126 static int 127 nfs_closedir(struct inode *inode, struct file *filp) 128 { 129 put_nfs_open_dir_context(file_inode(filp), filp->private_data); 130 return 0; 131 } 132 133 struct nfs_cache_array_entry { 134 u64 cookie; 135 u64 ino; 136 struct qstr string; 137 unsigned char d_type; 138 }; 139 140 struct nfs_cache_array { 141 int size; 142 int eof_index; 143 u64 last_cookie; 144 struct nfs_cache_array_entry array[0]; 145 }; 146 147 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, bool); 148 typedef struct { 149 struct file *file; 150 struct page *page; 151 struct dir_context *ctx; 152 unsigned long page_index; 153 u64 *dir_cookie; 154 u64 last_cookie; 155 loff_t current_index; 156 decode_dirent_t decode; 157 158 unsigned long timestamp; 159 unsigned long gencount; 160 unsigned int cache_entry_index; 161 bool plus; 162 bool eof; 163 } nfs_readdir_descriptor_t; 164 165 /* 166 * we are freeing strings created by nfs_add_to_readdir_array() 167 */ 168 static 169 void nfs_readdir_clear_array(struct page *page) 170 { 171 struct nfs_cache_array *array; 172 int i; 173 174 array = kmap_atomic(page); 175 for (i = 0; i < array->size; i++) 176 kfree(array->array[i].string.name); 177 kunmap_atomic(array); 178 } 179 180 /* 181 * the caller is responsible for freeing qstr.name 182 * when called by nfs_readdir_add_to_array, the strings will be freed in 183 * nfs_clear_readdir_array() 184 */ 185 static 186 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) 187 { 188 string->len = len; 189 string->name = kmemdup(name, len, GFP_KERNEL); 190 if (string->name == NULL) 191 return -ENOMEM; 192 /* 193 * Avoid a kmemleak false positive. The pointer to the name is stored 194 * in a page cache page which kmemleak does not scan. 195 */ 196 kmemleak_not_leak(string->name); 197 string->hash = full_name_hash(NULL, name, len); 198 return 0; 199 } 200 201 static 202 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 203 { 204 struct nfs_cache_array *array = kmap(page); 205 struct nfs_cache_array_entry *cache_entry; 206 int ret; 207 208 cache_entry = &array->array[array->size]; 209 210 /* Check that this entry lies within the page bounds */ 211 ret = -ENOSPC; 212 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) 213 goto out; 214 215 cache_entry->cookie = entry->prev_cookie; 216 cache_entry->ino = entry->ino; 217 cache_entry->d_type = entry->d_type; 218 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); 219 if (ret) 220 goto out; 221 array->last_cookie = entry->cookie; 222 array->size++; 223 if (entry->eof != 0) 224 array->eof_index = array->size; 225 out: 226 kunmap(page); 227 return ret; 228 } 229 230 static 231 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 232 { 233 loff_t diff = desc->ctx->pos - desc->current_index; 234 unsigned int index; 235 236 if (diff < 0) 237 goto out_eof; 238 if (diff >= array->size) { 239 if (array->eof_index >= 0) 240 goto out_eof; 241 return -EAGAIN; 242 } 243 244 index = (unsigned int)diff; 245 *desc->dir_cookie = array->array[index].cookie; 246 desc->cache_entry_index = index; 247 return 0; 248 out_eof: 249 desc->eof = true; 250 return -EBADCOOKIE; 251 } 252 253 static bool 254 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) 255 { 256 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 257 return false; 258 smp_rmb(); 259 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); 260 } 261 262 static 263 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 264 { 265 int i; 266 loff_t new_pos; 267 int status = -EAGAIN; 268 269 for (i = 0; i < array->size; i++) { 270 if (array->array[i].cookie == *desc->dir_cookie) { 271 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); 272 struct nfs_open_dir_context *ctx = desc->file->private_data; 273 274 new_pos = desc->current_index + i; 275 if (ctx->attr_gencount != nfsi->attr_gencount || 276 !nfs_readdir_inode_mapping_valid(nfsi)) { 277 ctx->duped = 0; 278 ctx->attr_gencount = nfsi->attr_gencount; 279 } else if (new_pos < desc->ctx->pos) { 280 if (ctx->duped > 0 281 && ctx->dup_cookie == *desc->dir_cookie) { 282 if (printk_ratelimit()) { 283 pr_notice("NFS: directory %pD2 contains a readdir loop." 284 "Please contact your server vendor. " 285 "The file: %.*s has duplicate cookie %llu\n", 286 desc->file, array->array[i].string.len, 287 array->array[i].string.name, *desc->dir_cookie); 288 } 289 status = -ELOOP; 290 goto out; 291 } 292 ctx->dup_cookie = *desc->dir_cookie; 293 ctx->duped = -1; 294 } 295 desc->ctx->pos = new_pos; 296 desc->cache_entry_index = i; 297 return 0; 298 } 299 } 300 if (array->eof_index >= 0) { 301 status = -EBADCOOKIE; 302 if (*desc->dir_cookie == array->last_cookie) 303 desc->eof = true; 304 } 305 out: 306 return status; 307 } 308 309 static 310 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) 311 { 312 struct nfs_cache_array *array; 313 int status; 314 315 array = kmap(desc->page); 316 317 if (*desc->dir_cookie == 0) 318 status = nfs_readdir_search_for_pos(array, desc); 319 else 320 status = nfs_readdir_search_for_cookie(array, desc); 321 322 if (status == -EAGAIN) { 323 desc->last_cookie = array->last_cookie; 324 desc->current_index += array->size; 325 desc->page_index++; 326 } 327 kunmap(desc->page); 328 return status; 329 } 330 331 /* Fill a page with xdr information before transferring to the cache page */ 332 static 333 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, 334 struct nfs_entry *entry, struct file *file, struct inode *inode) 335 { 336 struct nfs_open_dir_context *ctx = file->private_data; 337 struct rpc_cred *cred = ctx->cred; 338 unsigned long timestamp, gencount; 339 int error; 340 341 again: 342 timestamp = jiffies; 343 gencount = nfs_inc_attr_generation_counter(); 344 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages, 345 NFS_SERVER(inode)->dtsize, desc->plus); 346 if (error < 0) { 347 /* We requested READDIRPLUS, but the server doesn't grok it */ 348 if (error == -ENOTSUPP && desc->plus) { 349 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 350 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 351 desc->plus = false; 352 goto again; 353 } 354 goto error; 355 } 356 desc->timestamp = timestamp; 357 desc->gencount = gencount; 358 error: 359 return error; 360 } 361 362 static int xdr_decode(nfs_readdir_descriptor_t *desc, 363 struct nfs_entry *entry, struct xdr_stream *xdr) 364 { 365 int error; 366 367 error = desc->decode(xdr, entry, desc->plus); 368 if (error) 369 return error; 370 entry->fattr->time_start = desc->timestamp; 371 entry->fattr->gencount = desc->gencount; 372 return 0; 373 } 374 375 /* Match file and dirent using either filehandle or fileid 376 * Note: caller is responsible for checking the fsid 377 */ 378 static 379 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 380 { 381 struct inode *inode; 382 struct nfs_inode *nfsi; 383 384 if (d_really_is_negative(dentry)) 385 return 0; 386 387 inode = d_inode(dentry); 388 if (is_bad_inode(inode) || NFS_STALE(inode)) 389 return 0; 390 391 nfsi = NFS_I(inode); 392 if (entry->fattr->fileid != nfsi->fileid) 393 return 0; 394 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) 395 return 0; 396 return 1; 397 } 398 399 static 400 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) 401 { 402 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 403 return false; 404 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 405 return true; 406 if (ctx->pos == 0) 407 return true; 408 return false; 409 } 410 411 /* 412 * This function is called by the lookup and getattr code to request the 413 * use of readdirplus to accelerate any future lookups in the same 414 * directory. 415 */ 416 void nfs_advise_use_readdirplus(struct inode *dir) 417 { 418 struct nfs_inode *nfsi = NFS_I(dir); 419 420 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 421 !list_empty(&nfsi->open_files)) 422 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 423 } 424 425 /* 426 * This function is mainly for use by nfs_getattr(). 427 * 428 * If this is an 'ls -l', we want to force use of readdirplus. 429 * Do this by checking if there is an active file descriptor 430 * and calling nfs_advise_use_readdirplus, then forcing a 431 * cache flush. 432 */ 433 void nfs_force_use_readdirplus(struct inode *dir) 434 { 435 struct nfs_inode *nfsi = NFS_I(dir); 436 437 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 438 !list_empty(&nfsi->open_files)) { 439 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 440 invalidate_mapping_pages(dir->i_mapping, 0, -1); 441 } 442 } 443 444 static 445 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) 446 { 447 struct qstr filename = QSTR_INIT(entry->name, entry->len); 448 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 449 struct dentry *dentry; 450 struct dentry *alias; 451 struct inode *dir = d_inode(parent); 452 struct inode *inode; 453 int status; 454 455 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 456 return; 457 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 458 return; 459 if (filename.len == 0) 460 return; 461 /* Validate that the name doesn't contain any illegal '\0' */ 462 if (strnlen(filename.name, filename.len) != filename.len) 463 return; 464 /* ...or '/' */ 465 if (strnchr(filename.name, filename.len, '/')) 466 return; 467 if (filename.name[0] == '.') { 468 if (filename.len == 1) 469 return; 470 if (filename.len == 2 && filename.name[1] == '.') 471 return; 472 } 473 filename.hash = full_name_hash(parent, filename.name, filename.len); 474 475 dentry = d_lookup(parent, &filename); 476 again: 477 if (!dentry) { 478 dentry = d_alloc_parallel(parent, &filename, &wq); 479 if (IS_ERR(dentry)) 480 return; 481 } 482 if (!d_in_lookup(dentry)) { 483 /* Is there a mountpoint here? If so, just exit */ 484 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 485 &entry->fattr->fsid)) 486 goto out; 487 if (nfs_same_file(dentry, entry)) { 488 if (!entry->fh->size) 489 goto out; 490 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 491 status = nfs_refresh_inode(d_inode(dentry), entry->fattr); 492 if (!status) 493 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); 494 goto out; 495 } else { 496 d_invalidate(dentry); 497 dput(dentry); 498 dentry = NULL; 499 goto again; 500 } 501 } 502 if (!entry->fh->size) { 503 d_lookup_done(dentry); 504 goto out; 505 } 506 507 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 508 alias = d_splice_alias(inode, dentry); 509 d_lookup_done(dentry); 510 if (alias) { 511 if (IS_ERR(alias)) 512 goto out; 513 dput(dentry); 514 dentry = alias; 515 } 516 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 517 out: 518 dput(dentry); 519 } 520 521 /* Perform conversion from xdr to cache array */ 522 static 523 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 524 struct page **xdr_pages, struct page *page, unsigned int buflen) 525 { 526 struct xdr_stream stream; 527 struct xdr_buf buf; 528 struct page *scratch; 529 struct nfs_cache_array *array; 530 unsigned int count = 0; 531 int status; 532 533 scratch = alloc_page(GFP_KERNEL); 534 if (scratch == NULL) 535 return -ENOMEM; 536 537 if (buflen == 0) 538 goto out_nopages; 539 540 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 541 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 542 543 do { 544 status = xdr_decode(desc, entry, &stream); 545 if (status != 0) { 546 if (status == -EAGAIN) 547 status = 0; 548 break; 549 } 550 551 count++; 552 553 if (desc->plus) 554 nfs_prime_dcache(file_dentry(desc->file), entry); 555 556 status = nfs_readdir_add_to_array(entry, page); 557 if (status != 0) 558 break; 559 } while (!entry->eof); 560 561 out_nopages: 562 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { 563 array = kmap(page); 564 array->eof_index = array->size; 565 status = 0; 566 kunmap(page); 567 } 568 569 put_page(scratch); 570 return status; 571 } 572 573 static 574 void nfs_readdir_free_pages(struct page **pages, unsigned int npages) 575 { 576 unsigned int i; 577 for (i = 0; i < npages; i++) 578 put_page(pages[i]); 579 } 580 581 /* 582 * nfs_readdir_large_page will allocate pages that must be freed with a call 583 * to nfs_readdir_free_pagearray 584 */ 585 static 586 int nfs_readdir_alloc_pages(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_pages(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 struct nfs_entry entry; 608 struct file *file = desc->file; 609 struct nfs_cache_array *array; 610 int status = -ENOMEM; 611 unsigned int array_size = ARRAY_SIZE(pages); 612 613 entry.prev_cookie = 0; 614 entry.cookie = desc->last_cookie; 615 entry.eof = 0; 616 entry.fh = nfs_alloc_fhandle(); 617 entry.fattr = nfs_alloc_fattr(); 618 entry.server = NFS_SERVER(inode); 619 if (entry.fh == NULL || entry.fattr == NULL) 620 goto out; 621 622 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 623 if (IS_ERR(entry.label)) { 624 status = PTR_ERR(entry.label); 625 goto out; 626 } 627 628 array = kmap(page); 629 memset(array, 0, sizeof(struct nfs_cache_array)); 630 array->eof_index = -1; 631 632 status = nfs_readdir_alloc_pages(pages, array_size); 633 if (status < 0) 634 goto out_release_array; 635 do { 636 unsigned int pglen; 637 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 638 639 if (status < 0) 640 break; 641 pglen = status; 642 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 643 if (status < 0) { 644 if (status == -ENOSPC) 645 status = 0; 646 break; 647 } 648 } while (array->eof_index < 0); 649 650 nfs_readdir_free_pages(pages, array_size); 651 out_release_array: 652 kunmap(page); 653 nfs4_label_free(entry.label); 654 out: 655 nfs_free_fattr(entry.fattr); 656 nfs_free_fhandle(entry.fh); 657 return status; 658 } 659 660 /* 661 * Now we cache directories properly, by converting xdr information 662 * to an array that can be used for lookups later. This results in 663 * fewer cache pages, since we can store more information on each page. 664 * We only need to convert from xdr once so future lookups are much simpler 665 */ 666 static 667 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) 668 { 669 struct inode *inode = file_inode(desc->file); 670 int ret; 671 672 ret = nfs_readdir_xdr_to_array(desc, page, inode); 673 if (ret < 0) 674 goto error; 675 SetPageUptodate(page); 676 677 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 678 /* Should never happen */ 679 nfs_zap_mapping(inode, inode->i_mapping); 680 } 681 unlock_page(page); 682 return 0; 683 error: 684 unlock_page(page); 685 return ret; 686 } 687 688 static 689 void cache_page_release(nfs_readdir_descriptor_t *desc) 690 { 691 if (!desc->page->mapping) 692 nfs_readdir_clear_array(desc->page); 693 put_page(desc->page); 694 desc->page = NULL; 695 } 696 697 static 698 struct page *get_cache_page(nfs_readdir_descriptor_t *desc) 699 { 700 return read_cache_page(desc->file->f_mapping, 701 desc->page_index, (filler_t *)nfs_readdir_filler, desc); 702 } 703 704 /* 705 * Returns 0 if desc->dir_cookie was found on page desc->page_index 706 */ 707 static 708 int find_cache_page(nfs_readdir_descriptor_t *desc) 709 { 710 int res; 711 712 desc->page = get_cache_page(desc); 713 if (IS_ERR(desc->page)) 714 return PTR_ERR(desc->page); 715 716 res = nfs_readdir_search_array(desc); 717 if (res != 0) 718 cache_page_release(desc); 719 return res; 720 } 721 722 /* Search for desc->dir_cookie from the beginning of the page cache */ 723 static inline 724 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 725 { 726 int res; 727 728 if (desc->page_index == 0) { 729 desc->current_index = 0; 730 desc->last_cookie = 0; 731 } 732 do { 733 res = find_cache_page(desc); 734 } while (res == -EAGAIN); 735 return res; 736 } 737 738 /* 739 * Once we've found the start of the dirent within a page: fill 'er up... 740 */ 741 static 742 int nfs_do_filldir(nfs_readdir_descriptor_t *desc) 743 { 744 struct file *file = desc->file; 745 int i = 0; 746 int res = 0; 747 struct nfs_cache_array *array = NULL; 748 struct nfs_open_dir_context *ctx = file->private_data; 749 750 array = kmap(desc->page); 751 for (i = desc->cache_entry_index; i < array->size; i++) { 752 struct nfs_cache_array_entry *ent; 753 754 ent = &array->array[i]; 755 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, 756 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 757 desc->eof = true; 758 break; 759 } 760 desc->ctx->pos++; 761 if (i < (array->size-1)) 762 *desc->dir_cookie = array->array[i+1].cookie; 763 else 764 *desc->dir_cookie = array->last_cookie; 765 if (ctx->duped != 0) 766 ctx->duped = 1; 767 } 768 if (array->eof_index >= 0) 769 desc->eof = true; 770 771 kunmap(desc->page); 772 cache_page_release(desc); 773 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 774 (unsigned long long)*desc->dir_cookie, res); 775 return res; 776 } 777 778 /* 779 * If we cannot find a cookie in our cache, we suspect that this is 780 * because it points to a deleted file, so we ask the server to return 781 * whatever it thinks is the next entry. We then feed this to filldir. 782 * If all goes well, we should then be able to find our way round the 783 * cache on the next call to readdir_search_pagecache(); 784 * 785 * NOTE: we cannot add the anonymous page to the pagecache because 786 * the data it contains might not be page aligned. Besides, 787 * we should already have a complete representation of the 788 * directory in the page cache by the time we get here. 789 */ 790 static inline 791 int uncached_readdir(nfs_readdir_descriptor_t *desc) 792 { 793 struct page *page = NULL; 794 int status; 795 struct inode *inode = file_inode(desc->file); 796 struct nfs_open_dir_context *ctx = desc->file->private_data; 797 798 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 799 (unsigned long long)*desc->dir_cookie); 800 801 page = alloc_page(GFP_HIGHUSER); 802 if (!page) { 803 status = -ENOMEM; 804 goto out; 805 } 806 807 desc->page_index = 0; 808 desc->last_cookie = *desc->dir_cookie; 809 desc->page = page; 810 ctx->duped = 0; 811 812 status = nfs_readdir_xdr_to_array(desc, page, inode); 813 if (status < 0) 814 goto out_release; 815 816 status = nfs_do_filldir(desc); 817 818 out: 819 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 820 __func__, status); 821 return status; 822 out_release: 823 cache_page_release(desc); 824 goto out; 825 } 826 827 /* The file offset position represents the dirent entry number. A 828 last cookie cache takes care of the common case of reading the 829 whole directory. 830 */ 831 static int nfs_readdir(struct file *file, struct dir_context *ctx) 832 { 833 struct dentry *dentry = file_dentry(file); 834 struct inode *inode = d_inode(dentry); 835 nfs_readdir_descriptor_t my_desc, 836 *desc = &my_desc; 837 struct nfs_open_dir_context *dir_ctx = file->private_data; 838 int res = 0; 839 840 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 841 file, (long long)ctx->pos); 842 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 843 844 /* 845 * ctx->pos points to the dirent entry number. 846 * *desc->dir_cookie has the cookie for the next entry. We have 847 * to either find the entry with the appropriate number or 848 * revalidate the cookie. 849 */ 850 memset(desc, 0, sizeof(*desc)); 851 852 desc->file = file; 853 desc->ctx = ctx; 854 desc->dir_cookie = &dir_ctx->dir_cookie; 855 desc->decode = NFS_PROTO(inode)->decode_dirent; 856 desc->plus = nfs_use_readdirplus(inode, ctx); 857 858 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) 859 res = nfs_revalidate_mapping(inode, file->f_mapping); 860 if (res < 0) 861 goto out; 862 863 do { 864 res = readdir_search_pagecache(desc); 865 866 if (res == -EBADCOOKIE) { 867 res = 0; 868 /* This means either end of directory */ 869 if (*desc->dir_cookie && !desc->eof) { 870 /* Or that the server has 'lost' a cookie */ 871 res = uncached_readdir(desc); 872 if (res == 0) 873 continue; 874 } 875 break; 876 } 877 if (res == -ETOOSMALL && desc->plus) { 878 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 879 nfs_zap_caches(inode); 880 desc->page_index = 0; 881 desc->plus = false; 882 desc->eof = false; 883 continue; 884 } 885 if (res < 0) 886 break; 887 888 res = nfs_do_filldir(desc); 889 if (res < 0) 890 break; 891 } while (!desc->eof); 892 out: 893 if (res > 0) 894 res = 0; 895 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 896 return res; 897 } 898 899 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 900 { 901 struct inode *inode = file_inode(filp); 902 struct nfs_open_dir_context *dir_ctx = filp->private_data; 903 904 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 905 filp, offset, whence); 906 907 inode_lock(inode); 908 switch (whence) { 909 case 1: 910 offset += filp->f_pos; 911 case 0: 912 if (offset >= 0) 913 break; 914 default: 915 offset = -EINVAL; 916 goto out; 917 } 918 if (offset != filp->f_pos) { 919 filp->f_pos = offset; 920 dir_ctx->dir_cookie = 0; 921 dir_ctx->duped = 0; 922 } 923 out: 924 inode_unlock(inode); 925 return offset; 926 } 927 928 /* 929 * All directory operations under NFS are synchronous, so fsync() 930 * is a dummy operation. 931 */ 932 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 933 int datasync) 934 { 935 struct inode *inode = file_inode(filp); 936 937 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 938 939 inode_lock(inode); 940 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 941 inode_unlock(inode); 942 return 0; 943 } 944 945 /** 946 * nfs_force_lookup_revalidate - Mark the directory as having changed 947 * @dir - pointer to directory inode 948 * 949 * This forces the revalidation code in nfs_lookup_revalidate() to do a 950 * full lookup on all child dentries of 'dir' whenever a change occurs 951 * on the server that might have invalidated our dcache. 952 * 953 * The caller should be holding dir->i_lock 954 */ 955 void nfs_force_lookup_revalidate(struct inode *dir) 956 { 957 NFS_I(dir)->cache_change_attribute++; 958 } 959 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 960 961 /* 962 * A check for whether or not the parent directory has changed. 963 * In the case it has, we assume that the dentries are untrustworthy 964 * and may need to be looked up again. 965 * If rcu_walk prevents us from performing a full check, return 0. 966 */ 967 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 968 int rcu_walk) 969 { 970 if (IS_ROOT(dentry)) 971 return 1; 972 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 973 return 0; 974 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 975 return 0; 976 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 977 if (nfs_mapping_need_revalidate_inode(dir)) { 978 if (rcu_walk) 979 return 0; 980 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 981 return 0; 982 } 983 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 984 return 0; 985 return 1; 986 } 987 988 /* 989 * Use intent information to check whether or not we're going to do 990 * an O_EXCL create using this path component. 991 */ 992 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 993 { 994 if (NFS_PROTO(dir)->version == 2) 995 return 0; 996 return flags & LOOKUP_EXCL; 997 } 998 999 /* 1000 * Inode and filehandle revalidation for lookups. 1001 * 1002 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1003 * or if the intent information indicates that we're about to open this 1004 * particular file and the "nocto" mount flag is not set. 1005 * 1006 */ 1007 static 1008 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1009 { 1010 struct nfs_server *server = NFS_SERVER(inode); 1011 int ret; 1012 1013 if (IS_AUTOMOUNT(inode)) 1014 return 0; 1015 1016 if (flags & LOOKUP_OPEN) { 1017 switch (inode->i_mode & S_IFMT) { 1018 case S_IFREG: 1019 /* A NFSv4 OPEN will revalidate later */ 1020 if (server->caps & NFS_CAP_ATOMIC_OPEN) 1021 goto out; 1022 /* Fallthrough */ 1023 case S_IFDIR: 1024 if (server->flags & NFS_MOUNT_NOCTO) 1025 break; 1026 /* NFS close-to-open cache consistency validation */ 1027 goto out_force; 1028 } 1029 } 1030 1031 /* VFS wants an on-the-wire revalidation */ 1032 if (flags & LOOKUP_REVAL) 1033 goto out_force; 1034 out: 1035 return (inode->i_nlink == 0) ? -ENOENT : 0; 1036 out_force: 1037 if (flags & LOOKUP_RCU) 1038 return -ECHILD; 1039 ret = __nfs_revalidate_inode(server, inode); 1040 if (ret != 0) 1041 return ret; 1042 goto out; 1043 } 1044 1045 /* 1046 * We judge how long we want to trust negative 1047 * dentries by looking at the parent inode mtime. 1048 * 1049 * If parent mtime has changed, we revalidate, else we wait for a 1050 * period corresponding to the parent's attribute cache timeout value. 1051 * 1052 * If LOOKUP_RCU prevents us from performing a full check, return 1 1053 * suggesting a reval is needed. 1054 * 1055 * Note that when creating a new file, or looking up a rename target, 1056 * then it shouldn't be necessary to revalidate a negative dentry. 1057 */ 1058 static inline 1059 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1060 unsigned int flags) 1061 { 1062 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET)) 1063 return 0; 1064 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1065 return 1; 1066 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1067 } 1068 1069 /* 1070 * This is called every time the dcache has a lookup hit, 1071 * and we should check whether we can really trust that 1072 * lookup. 1073 * 1074 * NOTE! The hit can be a negative hit too, don't assume 1075 * we have an inode! 1076 * 1077 * If the parent directory is seen to have changed, we throw out the 1078 * cached dentry and do a new lookup. 1079 */ 1080 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1081 { 1082 struct inode *dir; 1083 struct inode *inode; 1084 struct dentry *parent; 1085 struct nfs_fh *fhandle = NULL; 1086 struct nfs_fattr *fattr = NULL; 1087 struct nfs4_label *label = NULL; 1088 int error; 1089 1090 if (flags & LOOKUP_RCU) { 1091 parent = READ_ONCE(dentry->d_parent); 1092 dir = d_inode_rcu(parent); 1093 if (!dir) 1094 return -ECHILD; 1095 } else { 1096 parent = dget_parent(dentry); 1097 dir = d_inode(parent); 1098 } 1099 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1100 inode = d_inode(dentry); 1101 1102 if (!inode) { 1103 if (nfs_neg_need_reval(dir, dentry, flags)) { 1104 if (flags & LOOKUP_RCU) 1105 return -ECHILD; 1106 goto out_bad; 1107 } 1108 goto out_valid; 1109 } 1110 1111 if (is_bad_inode(inode)) { 1112 if (flags & LOOKUP_RCU) 1113 return -ECHILD; 1114 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1115 __func__, dentry); 1116 goto out_bad; 1117 } 1118 1119 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) 1120 goto out_set_verifier; 1121 1122 /* Force a full look up iff the parent directory has changed */ 1123 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) && 1124 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1125 error = nfs_lookup_verify_inode(inode, flags); 1126 if (error) { 1127 if (flags & LOOKUP_RCU) 1128 return -ECHILD; 1129 if (error == -ESTALE) 1130 goto out_zap_parent; 1131 goto out_error; 1132 } 1133 nfs_advise_use_readdirplus(dir); 1134 goto out_valid; 1135 } 1136 1137 if (flags & LOOKUP_RCU) 1138 return -ECHILD; 1139 1140 if (NFS_STALE(inode)) 1141 goto out_bad; 1142 1143 error = -ENOMEM; 1144 fhandle = nfs_alloc_fhandle(); 1145 fattr = nfs_alloc_fattr(); 1146 if (fhandle == NULL || fattr == NULL) 1147 goto out_error; 1148 1149 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 1150 if (IS_ERR(label)) 1151 goto out_error; 1152 1153 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1154 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1155 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1156 if (error == -ESTALE || error == -ENOENT) 1157 goto out_bad; 1158 if (error) 1159 goto out_error; 1160 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1161 goto out_bad; 1162 if ((error = nfs_refresh_inode(inode, fattr)) != 0) 1163 goto out_bad; 1164 1165 nfs_setsecurity(inode, fattr, label); 1166 1167 nfs_free_fattr(fattr); 1168 nfs_free_fhandle(fhandle); 1169 nfs4_label_free(label); 1170 1171 /* set a readdirplus hint that we had a cache miss */ 1172 nfs_force_use_readdirplus(dir); 1173 1174 out_set_verifier: 1175 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1176 out_valid: 1177 if (flags & LOOKUP_RCU) { 1178 if (parent != READ_ONCE(dentry->d_parent)) 1179 return -ECHILD; 1180 } else 1181 dput(parent); 1182 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1183 __func__, dentry); 1184 return 1; 1185 out_zap_parent: 1186 nfs_zap_caches(dir); 1187 out_bad: 1188 WARN_ON(flags & LOOKUP_RCU); 1189 nfs_free_fattr(fattr); 1190 nfs_free_fhandle(fhandle); 1191 nfs4_label_free(label); 1192 nfs_mark_for_revalidate(dir); 1193 if (inode && S_ISDIR(inode->i_mode)) { 1194 /* Purge readdir caches. */ 1195 nfs_zap_caches(inode); 1196 /* 1197 * We can't d_drop the root of a disconnected tree: 1198 * its d_hash is on the s_anon list and d_drop() would hide 1199 * it from shrink_dcache_for_unmount(), leading to busy 1200 * inodes on unmount and further oopses. 1201 */ 1202 if (IS_ROOT(dentry)) 1203 goto out_valid; 1204 } 1205 dput(parent); 1206 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1207 __func__, dentry); 1208 return 0; 1209 out_error: 1210 WARN_ON(flags & LOOKUP_RCU); 1211 nfs_free_fattr(fattr); 1212 nfs_free_fhandle(fhandle); 1213 nfs4_label_free(label); 1214 dput(parent); 1215 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1216 __func__, dentry, error); 1217 return error; 1218 } 1219 1220 /* 1221 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1222 * when we don't really care about the dentry name. This is called when a 1223 * pathwalk ends on a dentry that was not found via a normal lookup in the 1224 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1225 * 1226 * In this situation, we just want to verify that the inode itself is OK 1227 * since the dentry might have changed on the server. 1228 */ 1229 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1230 { 1231 struct inode *inode = d_inode(dentry); 1232 int error = 0; 1233 1234 /* 1235 * I believe we can only get a negative dentry here in the case of a 1236 * procfs-style symlink. Just assume it's correct for now, but we may 1237 * eventually need to do something more here. 1238 */ 1239 if (!inode) { 1240 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1241 __func__, dentry); 1242 return 1; 1243 } 1244 1245 if (is_bad_inode(inode)) { 1246 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1247 __func__, dentry); 1248 return 0; 1249 } 1250 1251 error = nfs_lookup_verify_inode(inode, flags); 1252 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1253 __func__, inode->i_ino, error ? "invalid" : "valid"); 1254 return !error; 1255 } 1256 1257 /* 1258 * This is called from dput() when d_count is going to 0. 1259 */ 1260 static int nfs_dentry_delete(const struct dentry *dentry) 1261 { 1262 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1263 dentry, dentry->d_flags); 1264 1265 /* Unhash any dentry with a stale inode */ 1266 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1267 return 1; 1268 1269 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1270 /* Unhash it, so that ->d_iput() would be called */ 1271 return 1; 1272 } 1273 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { 1274 /* Unhash it, so that ancestors of killed async unlink 1275 * files will be cleaned up during umount */ 1276 return 1; 1277 } 1278 return 0; 1279 1280 } 1281 1282 /* Ensure that we revalidate inode->i_nlink */ 1283 static void nfs_drop_nlink(struct inode *inode) 1284 { 1285 spin_lock(&inode->i_lock); 1286 /* drop the inode if we're reasonably sure this is the last link */ 1287 if (inode->i_nlink == 1) 1288 clear_nlink(inode); 1289 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE 1290 | NFS_INO_INVALID_CTIME 1291 | NFS_INO_INVALID_OTHER; 1292 spin_unlock(&inode->i_lock); 1293 } 1294 1295 /* 1296 * Called when the dentry loses inode. 1297 * We use it to clean up silly-renamed files. 1298 */ 1299 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1300 { 1301 if (S_ISDIR(inode->i_mode)) 1302 /* drop any readdir cache as it could easily be old */ 1303 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1304 1305 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1306 nfs_complete_unlink(dentry, inode); 1307 nfs_drop_nlink(inode); 1308 } 1309 iput(inode); 1310 } 1311 1312 static void nfs_d_release(struct dentry *dentry) 1313 { 1314 /* free cached devname value, if it survived that far */ 1315 if (unlikely(dentry->d_fsdata)) { 1316 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1317 WARN_ON(1); 1318 else 1319 kfree(dentry->d_fsdata); 1320 } 1321 } 1322 1323 const struct dentry_operations nfs_dentry_operations = { 1324 .d_revalidate = nfs_lookup_revalidate, 1325 .d_weak_revalidate = nfs_weak_revalidate, 1326 .d_delete = nfs_dentry_delete, 1327 .d_iput = nfs_dentry_iput, 1328 .d_automount = nfs_d_automount, 1329 .d_release = nfs_d_release, 1330 }; 1331 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1332 1333 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1334 { 1335 struct dentry *res; 1336 struct inode *inode = NULL; 1337 struct nfs_fh *fhandle = NULL; 1338 struct nfs_fattr *fattr = NULL; 1339 struct nfs4_label *label = NULL; 1340 int error; 1341 1342 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1343 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1344 1345 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1346 return ERR_PTR(-ENAMETOOLONG); 1347 1348 /* 1349 * If we're doing an exclusive create, optimize away the lookup 1350 * but don't hash the dentry. 1351 */ 1352 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET) 1353 return NULL; 1354 1355 res = ERR_PTR(-ENOMEM); 1356 fhandle = nfs_alloc_fhandle(); 1357 fattr = nfs_alloc_fattr(); 1358 if (fhandle == NULL || fattr == NULL) 1359 goto out; 1360 1361 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1362 if (IS_ERR(label)) 1363 goto out; 1364 1365 trace_nfs_lookup_enter(dir, dentry, flags); 1366 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1367 if (error == -ENOENT) 1368 goto no_entry; 1369 if (error < 0) { 1370 res = ERR_PTR(error); 1371 goto out_label; 1372 } 1373 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1374 res = ERR_CAST(inode); 1375 if (IS_ERR(res)) 1376 goto out_label; 1377 1378 /* Notify readdir to use READDIRPLUS */ 1379 nfs_force_use_readdirplus(dir); 1380 1381 no_entry: 1382 res = d_splice_alias(inode, dentry); 1383 if (res != NULL) { 1384 if (IS_ERR(res)) 1385 goto out_label; 1386 dentry = res; 1387 } 1388 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1389 out_label: 1390 trace_nfs_lookup_exit(dir, dentry, flags, error); 1391 nfs4_label_free(label); 1392 out: 1393 nfs_free_fattr(fattr); 1394 nfs_free_fhandle(fhandle); 1395 return res; 1396 } 1397 EXPORT_SYMBOL_GPL(nfs_lookup); 1398 1399 #if IS_ENABLED(CONFIG_NFS_V4) 1400 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1401 1402 const struct dentry_operations nfs4_dentry_operations = { 1403 .d_revalidate = nfs4_lookup_revalidate, 1404 .d_weak_revalidate = nfs_weak_revalidate, 1405 .d_delete = nfs_dentry_delete, 1406 .d_iput = nfs_dentry_iput, 1407 .d_automount = nfs_d_automount, 1408 .d_release = nfs_d_release, 1409 }; 1410 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1411 1412 static fmode_t flags_to_mode(int flags) 1413 { 1414 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1415 if ((flags & O_ACCMODE) != O_WRONLY) 1416 res |= FMODE_READ; 1417 if ((flags & O_ACCMODE) != O_RDONLY) 1418 res |= FMODE_WRITE; 1419 return res; 1420 } 1421 1422 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) 1423 { 1424 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); 1425 } 1426 1427 static int do_open(struct inode *inode, struct file *filp) 1428 { 1429 nfs_fscache_open_file(inode, filp); 1430 return 0; 1431 } 1432 1433 static int nfs_finish_open(struct nfs_open_context *ctx, 1434 struct dentry *dentry, 1435 struct file *file, unsigned open_flags, 1436 int *opened) 1437 { 1438 int err; 1439 1440 err = finish_open(file, dentry, do_open, opened); 1441 if (err) 1442 goto out; 1443 if (S_ISREG(file->f_path.dentry->d_inode->i_mode)) 1444 nfs_file_set_open_context(file, ctx); 1445 else 1446 err = -ESTALE; 1447 out: 1448 return err; 1449 } 1450 1451 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1452 struct file *file, unsigned open_flags, 1453 umode_t mode, int *opened) 1454 { 1455 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1456 struct nfs_open_context *ctx; 1457 struct dentry *res; 1458 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1459 struct inode *inode; 1460 unsigned int lookup_flags = 0; 1461 bool switched = false; 1462 int err; 1463 1464 /* Expect a negative dentry */ 1465 BUG_ON(d_inode(dentry)); 1466 1467 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1468 dir->i_sb->s_id, dir->i_ino, dentry); 1469 1470 err = nfs_check_flags(open_flags); 1471 if (err) 1472 return err; 1473 1474 /* NFS only supports OPEN on regular files */ 1475 if ((open_flags & O_DIRECTORY)) { 1476 if (!d_in_lookup(dentry)) { 1477 /* 1478 * Hashed negative dentry with O_DIRECTORY: dentry was 1479 * revalidated and is fine, no need to perform lookup 1480 * again 1481 */ 1482 return -ENOENT; 1483 } 1484 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1485 goto no_open; 1486 } 1487 1488 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1489 return -ENAMETOOLONG; 1490 1491 if (open_flags & O_CREAT) { 1492 struct nfs_server *server = NFS_SERVER(dir); 1493 1494 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) 1495 mode &= ~current_umask(); 1496 1497 attr.ia_valid |= ATTR_MODE; 1498 attr.ia_mode = mode; 1499 } 1500 if (open_flags & O_TRUNC) { 1501 attr.ia_valid |= ATTR_SIZE; 1502 attr.ia_size = 0; 1503 } 1504 1505 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 1506 d_drop(dentry); 1507 switched = true; 1508 dentry = d_alloc_parallel(dentry->d_parent, 1509 &dentry->d_name, &wq); 1510 if (IS_ERR(dentry)) 1511 return PTR_ERR(dentry); 1512 if (unlikely(!d_in_lookup(dentry))) 1513 return finish_no_open(file, dentry); 1514 } 1515 1516 ctx = create_nfs_open_context(dentry, open_flags, file); 1517 err = PTR_ERR(ctx); 1518 if (IS_ERR(ctx)) 1519 goto out; 1520 1521 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1522 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened); 1523 if (IS_ERR(inode)) { 1524 err = PTR_ERR(inode); 1525 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1526 put_nfs_open_context(ctx); 1527 d_drop(dentry); 1528 switch (err) { 1529 case -ENOENT: 1530 d_splice_alias(NULL, dentry); 1531 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1532 break; 1533 case -EISDIR: 1534 case -ENOTDIR: 1535 goto no_open; 1536 case -ELOOP: 1537 if (!(open_flags & O_NOFOLLOW)) 1538 goto no_open; 1539 break; 1540 /* case -EINVAL: */ 1541 default: 1542 break; 1543 } 1544 goto out; 1545 } 1546 1547 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened); 1548 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1549 put_nfs_open_context(ctx); 1550 out: 1551 if (unlikely(switched)) { 1552 d_lookup_done(dentry); 1553 dput(dentry); 1554 } 1555 return err; 1556 1557 no_open: 1558 res = nfs_lookup(dir, dentry, lookup_flags); 1559 if (switched) { 1560 d_lookup_done(dentry); 1561 if (!res) 1562 res = dentry; 1563 else 1564 dput(dentry); 1565 } 1566 if (IS_ERR(res)) 1567 return PTR_ERR(res); 1568 return finish_no_open(file, res); 1569 } 1570 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1571 1572 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1573 { 1574 struct inode *inode; 1575 int ret = 0; 1576 1577 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1578 goto no_open; 1579 if (d_mountpoint(dentry)) 1580 goto no_open; 1581 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) 1582 goto no_open; 1583 1584 inode = d_inode(dentry); 1585 1586 /* We can't create new files in nfs_open_revalidate(), so we 1587 * optimize away revalidation of negative dentries. 1588 */ 1589 if (inode == NULL) { 1590 struct dentry *parent; 1591 struct inode *dir; 1592 1593 if (flags & LOOKUP_RCU) { 1594 parent = READ_ONCE(dentry->d_parent); 1595 dir = d_inode_rcu(parent); 1596 if (!dir) 1597 return -ECHILD; 1598 } else { 1599 parent = dget_parent(dentry); 1600 dir = d_inode(parent); 1601 } 1602 if (!nfs_neg_need_reval(dir, dentry, flags)) 1603 ret = 1; 1604 else if (flags & LOOKUP_RCU) 1605 ret = -ECHILD; 1606 if (!(flags & LOOKUP_RCU)) 1607 dput(parent); 1608 else if (parent != READ_ONCE(dentry->d_parent)) 1609 return -ECHILD; 1610 goto out; 1611 } 1612 1613 /* NFS only supports OPEN on regular files */ 1614 if (!S_ISREG(inode->i_mode)) 1615 goto no_open; 1616 /* We cannot do exclusive creation on a positive dentry */ 1617 if (flags & LOOKUP_EXCL) 1618 goto no_open; 1619 1620 /* Let f_op->open() actually open (and revalidate) the file */ 1621 ret = 1; 1622 1623 out: 1624 return ret; 1625 1626 no_open: 1627 return nfs_lookup_revalidate(dentry, flags); 1628 } 1629 1630 #endif /* CONFIG_NFSV4 */ 1631 1632 /* 1633 * Code common to create, mkdir, and mknod. 1634 */ 1635 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1636 struct nfs_fattr *fattr, 1637 struct nfs4_label *label) 1638 { 1639 struct dentry *parent = dget_parent(dentry); 1640 struct inode *dir = d_inode(parent); 1641 struct inode *inode; 1642 int error = -EACCES; 1643 1644 d_drop(dentry); 1645 1646 /* We may have been initialized further down */ 1647 if (d_really_is_positive(dentry)) 1648 goto out; 1649 if (fhandle->size == 0) { 1650 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); 1651 if (error) 1652 goto out_error; 1653 } 1654 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1655 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1656 struct nfs_server *server = NFS_SB(dentry->d_sb); 1657 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); 1658 if (error < 0) 1659 goto out_error; 1660 } 1661 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1662 error = PTR_ERR(inode); 1663 if (IS_ERR(inode)) 1664 goto out_error; 1665 d_add(dentry, inode); 1666 out: 1667 dput(parent); 1668 return 0; 1669 out_error: 1670 nfs_mark_for_revalidate(dir); 1671 dput(parent); 1672 return error; 1673 } 1674 EXPORT_SYMBOL_GPL(nfs_instantiate); 1675 1676 /* 1677 * Following a failed create operation, we drop the dentry rather 1678 * than retain a negative dentry. This avoids a problem in the event 1679 * that the operation succeeded on the server, but an error in the 1680 * reply path made it appear to have failed. 1681 */ 1682 int nfs_create(struct inode *dir, struct dentry *dentry, 1683 umode_t mode, bool excl) 1684 { 1685 struct iattr attr; 1686 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 1687 int error; 1688 1689 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 1690 dir->i_sb->s_id, dir->i_ino, dentry); 1691 1692 attr.ia_mode = mode; 1693 attr.ia_valid = ATTR_MODE; 1694 1695 trace_nfs_create_enter(dir, dentry, open_flags); 1696 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1697 trace_nfs_create_exit(dir, dentry, open_flags, error); 1698 if (error != 0) 1699 goto out_err; 1700 return 0; 1701 out_err: 1702 d_drop(dentry); 1703 return error; 1704 } 1705 EXPORT_SYMBOL_GPL(nfs_create); 1706 1707 /* 1708 * See comments for nfs_proc_create regarding failed operations. 1709 */ 1710 int 1711 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) 1712 { 1713 struct iattr attr; 1714 int status; 1715 1716 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 1717 dir->i_sb->s_id, dir->i_ino, dentry); 1718 1719 attr.ia_mode = mode; 1720 attr.ia_valid = ATTR_MODE; 1721 1722 trace_nfs_mknod_enter(dir, dentry); 1723 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1724 trace_nfs_mknod_exit(dir, dentry, status); 1725 if (status != 0) 1726 goto out_err; 1727 return 0; 1728 out_err: 1729 d_drop(dentry); 1730 return status; 1731 } 1732 EXPORT_SYMBOL_GPL(nfs_mknod); 1733 1734 /* 1735 * See comments for nfs_proc_create regarding failed operations. 1736 */ 1737 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 1738 { 1739 struct iattr attr; 1740 int error; 1741 1742 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 1743 dir->i_sb->s_id, dir->i_ino, dentry); 1744 1745 attr.ia_valid = ATTR_MODE; 1746 attr.ia_mode = mode | S_IFDIR; 1747 1748 trace_nfs_mkdir_enter(dir, dentry); 1749 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1750 trace_nfs_mkdir_exit(dir, dentry, error); 1751 if (error != 0) 1752 goto out_err; 1753 return 0; 1754 out_err: 1755 d_drop(dentry); 1756 return error; 1757 } 1758 EXPORT_SYMBOL_GPL(nfs_mkdir); 1759 1760 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1761 { 1762 if (simple_positive(dentry)) 1763 d_delete(dentry); 1764 } 1765 1766 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1767 { 1768 int error; 1769 1770 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 1771 dir->i_sb->s_id, dir->i_ino, dentry); 1772 1773 trace_nfs_rmdir_enter(dir, dentry); 1774 if (d_really_is_positive(dentry)) { 1775 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 1776 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1777 /* Ensure the VFS deletes this inode */ 1778 switch (error) { 1779 case 0: 1780 clear_nlink(d_inode(dentry)); 1781 break; 1782 case -ENOENT: 1783 nfs_dentry_handle_enoent(dentry); 1784 } 1785 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 1786 } else 1787 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1788 trace_nfs_rmdir_exit(dir, dentry, error); 1789 1790 return error; 1791 } 1792 EXPORT_SYMBOL_GPL(nfs_rmdir); 1793 1794 /* 1795 * Remove a file after making sure there are no pending writes, 1796 * and after checking that the file has only one user. 1797 * 1798 * We invalidate the attribute cache and free the inode prior to the operation 1799 * to avoid possible races if the server reuses the inode. 1800 */ 1801 static int nfs_safe_remove(struct dentry *dentry) 1802 { 1803 struct inode *dir = d_inode(dentry->d_parent); 1804 struct inode *inode = d_inode(dentry); 1805 int error = -EBUSY; 1806 1807 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 1808 1809 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1810 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1811 error = 0; 1812 goto out; 1813 } 1814 1815 trace_nfs_remove_enter(dir, dentry); 1816 if (inode != NULL) { 1817 error = NFS_PROTO(dir)->remove(dir, dentry); 1818 if (error == 0) 1819 nfs_drop_nlink(inode); 1820 } else 1821 error = NFS_PROTO(dir)->remove(dir, dentry); 1822 if (error == -ENOENT) 1823 nfs_dentry_handle_enoent(dentry); 1824 trace_nfs_remove_exit(dir, dentry, error); 1825 out: 1826 return error; 1827 } 1828 1829 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1830 * belongs to an active ".nfs..." file and we return -EBUSY. 1831 * 1832 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1833 */ 1834 int nfs_unlink(struct inode *dir, struct dentry *dentry) 1835 { 1836 int error; 1837 int need_rehash = 0; 1838 1839 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 1840 dir->i_ino, dentry); 1841 1842 trace_nfs_unlink_enter(dir, dentry); 1843 spin_lock(&dentry->d_lock); 1844 if (d_count(dentry) > 1) { 1845 spin_unlock(&dentry->d_lock); 1846 /* Start asynchronous writeout of the inode */ 1847 write_inode_now(d_inode(dentry), 0); 1848 error = nfs_sillyrename(dir, dentry); 1849 goto out; 1850 } 1851 if (!d_unhashed(dentry)) { 1852 __d_drop(dentry); 1853 need_rehash = 1; 1854 } 1855 spin_unlock(&dentry->d_lock); 1856 error = nfs_safe_remove(dentry); 1857 if (!error || error == -ENOENT) { 1858 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1859 } else if (need_rehash) 1860 d_rehash(dentry); 1861 out: 1862 trace_nfs_unlink_exit(dir, dentry, error); 1863 return error; 1864 } 1865 EXPORT_SYMBOL_GPL(nfs_unlink); 1866 1867 /* 1868 * To create a symbolic link, most file systems instantiate a new inode, 1869 * add a page to it containing the path, then write it out to the disk 1870 * using prepare_write/commit_write. 1871 * 1872 * Unfortunately the NFS client can't create the in-core inode first 1873 * because it needs a file handle to create an in-core inode (see 1874 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1875 * symlink request has completed on the server. 1876 * 1877 * So instead we allocate a raw page, copy the symname into it, then do 1878 * the SYMLINK request with the page as the buffer. If it succeeds, we 1879 * now have a new file handle and can instantiate an in-core NFS inode 1880 * and move the raw page into its mapping. 1881 */ 1882 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1883 { 1884 struct page *page; 1885 char *kaddr; 1886 struct iattr attr; 1887 unsigned int pathlen = strlen(symname); 1888 int error; 1889 1890 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 1891 dir->i_ino, dentry, symname); 1892 1893 if (pathlen > PAGE_SIZE) 1894 return -ENAMETOOLONG; 1895 1896 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1897 attr.ia_valid = ATTR_MODE; 1898 1899 page = alloc_page(GFP_USER); 1900 if (!page) 1901 return -ENOMEM; 1902 1903 kaddr = page_address(page); 1904 memcpy(kaddr, symname, pathlen); 1905 if (pathlen < PAGE_SIZE) 1906 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1907 1908 trace_nfs_symlink_enter(dir, dentry); 1909 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1910 trace_nfs_symlink_exit(dir, dentry, error); 1911 if (error != 0) { 1912 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 1913 dir->i_sb->s_id, dir->i_ino, 1914 dentry, symname, error); 1915 d_drop(dentry); 1916 __free_page(page); 1917 return error; 1918 } 1919 1920 /* 1921 * No big deal if we can't add this page to the page cache here. 1922 * READLINK will get the missing page from the server if needed. 1923 */ 1924 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 1925 GFP_KERNEL)) { 1926 SetPageUptodate(page); 1927 unlock_page(page); 1928 /* 1929 * add_to_page_cache_lru() grabs an extra page refcount. 1930 * Drop it here to avoid leaking this page later. 1931 */ 1932 put_page(page); 1933 } else 1934 __free_page(page); 1935 1936 return 0; 1937 } 1938 EXPORT_SYMBOL_GPL(nfs_symlink); 1939 1940 int 1941 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1942 { 1943 struct inode *inode = d_inode(old_dentry); 1944 int error; 1945 1946 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 1947 old_dentry, dentry); 1948 1949 trace_nfs_link_enter(inode, dir, dentry); 1950 d_drop(dentry); 1951 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1952 if (error == 0) { 1953 ihold(inode); 1954 d_add(dentry, inode); 1955 } 1956 trace_nfs_link_exit(inode, dir, dentry, error); 1957 return error; 1958 } 1959 EXPORT_SYMBOL_GPL(nfs_link); 1960 1961 /* 1962 * RENAME 1963 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1964 * different file handle for the same inode after a rename (e.g. when 1965 * moving to a different directory). A fail-safe method to do so would 1966 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1967 * rename the old file using the sillyrename stuff. This way, the original 1968 * file in old_dir will go away when the last process iput()s the inode. 1969 * 1970 * FIXED. 1971 * 1972 * It actually works quite well. One needs to have the possibility for 1973 * at least one ".nfs..." file in each directory the file ever gets 1974 * moved or linked to which happens automagically with the new 1975 * implementation that only depends on the dcache stuff instead of 1976 * using the inode layer 1977 * 1978 * Unfortunately, things are a little more complicated than indicated 1979 * above. For a cross-directory move, we want to make sure we can get 1980 * rid of the old inode after the operation. This means there must be 1981 * no pending writes (if it's a file), and the use count must be 1. 1982 * If these conditions are met, we can drop the dentries before doing 1983 * the rename. 1984 */ 1985 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 1986 struct inode *new_dir, struct dentry *new_dentry, 1987 unsigned int flags) 1988 { 1989 struct inode *old_inode = d_inode(old_dentry); 1990 struct inode *new_inode = d_inode(new_dentry); 1991 struct dentry *dentry = NULL, *rehash = NULL; 1992 struct rpc_task *task; 1993 int error = -EBUSY; 1994 1995 if (flags) 1996 return -EINVAL; 1997 1998 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 1999 old_dentry, new_dentry, 2000 d_count(new_dentry)); 2001 2002 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2003 /* 2004 * For non-directories, check whether the target is busy and if so, 2005 * make a copy of the dentry and then do a silly-rename. If the 2006 * silly-rename succeeds, the copied dentry is hashed and becomes 2007 * the new target. 2008 */ 2009 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2010 /* 2011 * To prevent any new references to the target during the 2012 * rename, we unhash the dentry in advance. 2013 */ 2014 if (!d_unhashed(new_dentry)) { 2015 d_drop(new_dentry); 2016 rehash = new_dentry; 2017 } 2018 2019 if (d_count(new_dentry) > 2) { 2020 int err; 2021 2022 /* copy the target dentry's name */ 2023 dentry = d_alloc(new_dentry->d_parent, 2024 &new_dentry->d_name); 2025 if (!dentry) 2026 goto out; 2027 2028 /* silly-rename the existing target ... */ 2029 err = nfs_sillyrename(new_dir, new_dentry); 2030 if (err) 2031 goto out; 2032 2033 new_dentry = dentry; 2034 rehash = NULL; 2035 new_inode = NULL; 2036 } 2037 } 2038 2039 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2040 if (IS_ERR(task)) { 2041 error = PTR_ERR(task); 2042 goto out; 2043 } 2044 2045 error = rpc_wait_for_completion_task(task); 2046 if (error != 0) { 2047 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; 2048 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ 2049 smp_wmb(); 2050 } else 2051 error = task->tk_status; 2052 rpc_put_task(task); 2053 nfs_mark_for_revalidate(old_inode); 2054 out: 2055 if (rehash) 2056 d_rehash(rehash); 2057 trace_nfs_rename_exit(old_dir, old_dentry, 2058 new_dir, new_dentry, error); 2059 if (!error) { 2060 if (new_inode != NULL) 2061 nfs_drop_nlink(new_inode); 2062 /* 2063 * The d_move() should be here instead of in an async RPC completion 2064 * handler because we need the proper locks to move the dentry. If 2065 * we're interrupted by a signal, the async RPC completion handler 2066 * should mark the directories for revalidation. 2067 */ 2068 d_move(old_dentry, new_dentry); 2069 nfs_set_verifier(old_dentry, 2070 nfs_save_change_attribute(new_dir)); 2071 } else if (error == -ENOENT) 2072 nfs_dentry_handle_enoent(old_dentry); 2073 2074 /* new dentry created? */ 2075 if (dentry) 2076 dput(dentry); 2077 return error; 2078 } 2079 EXPORT_SYMBOL_GPL(nfs_rename); 2080 2081 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2082 static LIST_HEAD(nfs_access_lru_list); 2083 static atomic_long_t nfs_access_nr_entries; 2084 2085 static unsigned long nfs_access_max_cachesize = ULONG_MAX; 2086 module_param(nfs_access_max_cachesize, ulong, 0644); 2087 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2088 2089 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2090 { 2091 put_rpccred(entry->cred); 2092 kfree_rcu(entry, rcu_head); 2093 smp_mb__before_atomic(); 2094 atomic_long_dec(&nfs_access_nr_entries); 2095 smp_mb__after_atomic(); 2096 } 2097 2098 static void nfs_access_free_list(struct list_head *head) 2099 { 2100 struct nfs_access_entry *cache; 2101 2102 while (!list_empty(head)) { 2103 cache = list_entry(head->next, struct nfs_access_entry, lru); 2104 list_del(&cache->lru); 2105 nfs_access_free_entry(cache); 2106 } 2107 } 2108 2109 static unsigned long 2110 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2111 { 2112 LIST_HEAD(head); 2113 struct nfs_inode *nfsi, *next; 2114 struct nfs_access_entry *cache; 2115 long freed = 0; 2116 2117 spin_lock(&nfs_access_lru_lock); 2118 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2119 struct inode *inode; 2120 2121 if (nr_to_scan-- == 0) 2122 break; 2123 inode = &nfsi->vfs_inode; 2124 spin_lock(&inode->i_lock); 2125 if (list_empty(&nfsi->access_cache_entry_lru)) 2126 goto remove_lru_entry; 2127 cache = list_entry(nfsi->access_cache_entry_lru.next, 2128 struct nfs_access_entry, lru); 2129 list_move(&cache->lru, &head); 2130 rb_erase(&cache->rb_node, &nfsi->access_cache); 2131 freed++; 2132 if (!list_empty(&nfsi->access_cache_entry_lru)) 2133 list_move_tail(&nfsi->access_cache_inode_lru, 2134 &nfs_access_lru_list); 2135 else { 2136 remove_lru_entry: 2137 list_del_init(&nfsi->access_cache_inode_lru); 2138 smp_mb__before_atomic(); 2139 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2140 smp_mb__after_atomic(); 2141 } 2142 spin_unlock(&inode->i_lock); 2143 } 2144 spin_unlock(&nfs_access_lru_lock); 2145 nfs_access_free_list(&head); 2146 return freed; 2147 } 2148 2149 unsigned long 2150 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2151 { 2152 int nr_to_scan = sc->nr_to_scan; 2153 gfp_t gfp_mask = sc->gfp_mask; 2154 2155 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2156 return SHRINK_STOP; 2157 return nfs_do_access_cache_scan(nr_to_scan); 2158 } 2159 2160 2161 unsigned long 2162 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2163 { 2164 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2165 } 2166 2167 static void 2168 nfs_access_cache_enforce_limit(void) 2169 { 2170 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2171 unsigned long diff; 2172 unsigned int nr_to_scan; 2173 2174 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2175 return; 2176 nr_to_scan = 100; 2177 diff = nr_entries - nfs_access_max_cachesize; 2178 if (diff < nr_to_scan) 2179 nr_to_scan = diff; 2180 nfs_do_access_cache_scan(nr_to_scan); 2181 } 2182 2183 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2184 { 2185 struct rb_root *root_node = &nfsi->access_cache; 2186 struct rb_node *n; 2187 struct nfs_access_entry *entry; 2188 2189 /* Unhook entries from the cache */ 2190 while ((n = rb_first(root_node)) != NULL) { 2191 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2192 rb_erase(n, root_node); 2193 list_move(&entry->lru, head); 2194 } 2195 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2196 } 2197 2198 void nfs_access_zap_cache(struct inode *inode) 2199 { 2200 LIST_HEAD(head); 2201 2202 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2203 return; 2204 /* Remove from global LRU init */ 2205 spin_lock(&nfs_access_lru_lock); 2206 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2207 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2208 2209 spin_lock(&inode->i_lock); 2210 __nfs_access_zap_cache(NFS_I(inode), &head); 2211 spin_unlock(&inode->i_lock); 2212 spin_unlock(&nfs_access_lru_lock); 2213 nfs_access_free_list(&head); 2214 } 2215 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2216 2217 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 2218 { 2219 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2220 struct nfs_access_entry *entry; 2221 2222 while (n != NULL) { 2223 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2224 2225 if (cred < entry->cred) 2226 n = n->rb_left; 2227 else if (cred > entry->cred) 2228 n = n->rb_right; 2229 else 2230 return entry; 2231 } 2232 return NULL; 2233 } 2234 2235 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block) 2236 { 2237 struct nfs_inode *nfsi = NFS_I(inode); 2238 struct nfs_access_entry *cache; 2239 bool retry = true; 2240 int err; 2241 2242 spin_lock(&inode->i_lock); 2243 for(;;) { 2244 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2245 goto out_zap; 2246 cache = nfs_access_search_rbtree(inode, cred); 2247 err = -ENOENT; 2248 if (cache == NULL) 2249 goto out; 2250 /* Found an entry, is our attribute cache valid? */ 2251 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2252 break; 2253 err = -ECHILD; 2254 if (!may_block) 2255 goto out; 2256 if (!retry) 2257 goto out_zap; 2258 spin_unlock(&inode->i_lock); 2259 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2260 if (err) 2261 return err; 2262 spin_lock(&inode->i_lock); 2263 retry = false; 2264 } 2265 res->cred = cache->cred; 2266 res->mask = cache->mask; 2267 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2268 err = 0; 2269 out: 2270 spin_unlock(&inode->i_lock); 2271 return err; 2272 out_zap: 2273 spin_unlock(&inode->i_lock); 2274 nfs_access_zap_cache(inode); 2275 return -ENOENT; 2276 } 2277 2278 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2279 { 2280 /* Only check the most recently returned cache entry, 2281 * but do it without locking. 2282 */ 2283 struct nfs_inode *nfsi = NFS_I(inode); 2284 struct nfs_access_entry *cache; 2285 int err = -ECHILD; 2286 struct list_head *lh; 2287 2288 rcu_read_lock(); 2289 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2290 goto out; 2291 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); 2292 cache = list_entry(lh, struct nfs_access_entry, lru); 2293 if (lh == &nfsi->access_cache_entry_lru || 2294 cred != cache->cred) 2295 cache = NULL; 2296 if (cache == NULL) 2297 goto out; 2298 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2299 goto out; 2300 res->cred = cache->cred; 2301 res->mask = cache->mask; 2302 err = 0; 2303 out: 2304 rcu_read_unlock(); 2305 return err; 2306 } 2307 2308 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2309 { 2310 struct nfs_inode *nfsi = NFS_I(inode); 2311 struct rb_root *root_node = &nfsi->access_cache; 2312 struct rb_node **p = &root_node->rb_node; 2313 struct rb_node *parent = NULL; 2314 struct nfs_access_entry *entry; 2315 2316 spin_lock(&inode->i_lock); 2317 while (*p != NULL) { 2318 parent = *p; 2319 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2320 2321 if (set->cred < entry->cred) 2322 p = &parent->rb_left; 2323 else if (set->cred > entry->cred) 2324 p = &parent->rb_right; 2325 else 2326 goto found; 2327 } 2328 rb_link_node(&set->rb_node, parent, p); 2329 rb_insert_color(&set->rb_node, root_node); 2330 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2331 spin_unlock(&inode->i_lock); 2332 return; 2333 found: 2334 rb_replace_node(parent, &set->rb_node, root_node); 2335 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2336 list_del(&entry->lru); 2337 spin_unlock(&inode->i_lock); 2338 nfs_access_free_entry(entry); 2339 } 2340 2341 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2342 { 2343 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2344 if (cache == NULL) 2345 return; 2346 RB_CLEAR_NODE(&cache->rb_node); 2347 cache->cred = get_rpccred(set->cred); 2348 cache->mask = set->mask; 2349 2350 /* The above field assignments must be visible 2351 * before this item appears on the lru. We cannot easily 2352 * use rcu_assign_pointer, so just force the memory barrier. 2353 */ 2354 smp_wmb(); 2355 nfs_access_add_rbtree(inode, cache); 2356 2357 /* Update accounting */ 2358 smp_mb__before_atomic(); 2359 atomic_long_inc(&nfs_access_nr_entries); 2360 smp_mb__after_atomic(); 2361 2362 /* Add inode to global LRU list */ 2363 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2364 spin_lock(&nfs_access_lru_lock); 2365 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2366 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2367 &nfs_access_lru_list); 2368 spin_unlock(&nfs_access_lru_lock); 2369 } 2370 nfs_access_cache_enforce_limit(); 2371 } 2372 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2373 2374 #define NFS_MAY_READ (NFS_ACCESS_READ) 2375 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2376 NFS_ACCESS_EXTEND | \ 2377 NFS_ACCESS_DELETE) 2378 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2379 NFS_ACCESS_EXTEND) 2380 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE 2381 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) 2382 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) 2383 static int 2384 nfs_access_calc_mask(u32 access_result, umode_t umode) 2385 { 2386 int mask = 0; 2387 2388 if (access_result & NFS_MAY_READ) 2389 mask |= MAY_READ; 2390 if (S_ISDIR(umode)) { 2391 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) 2392 mask |= MAY_WRITE; 2393 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) 2394 mask |= MAY_EXEC; 2395 } else if (S_ISREG(umode)) { 2396 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) 2397 mask |= MAY_WRITE; 2398 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) 2399 mask |= MAY_EXEC; 2400 } else if (access_result & NFS_MAY_WRITE) 2401 mask |= MAY_WRITE; 2402 return mask; 2403 } 2404 2405 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2406 { 2407 entry->mask = access_result; 2408 } 2409 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2410 2411 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 2412 { 2413 struct nfs_access_entry cache; 2414 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2415 int cache_mask; 2416 int status; 2417 2418 trace_nfs_access_enter(inode); 2419 2420 status = nfs_access_get_cached_rcu(inode, cred, &cache); 2421 if (status != 0) 2422 status = nfs_access_get_cached(inode, cred, &cache, may_block); 2423 if (status == 0) 2424 goto out_cached; 2425 2426 status = -ECHILD; 2427 if (!may_block) 2428 goto out; 2429 2430 /* 2431 * Determine which access bits we want to ask for... 2432 */ 2433 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; 2434 if (S_ISDIR(inode->i_mode)) 2435 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; 2436 else 2437 cache.mask |= NFS_ACCESS_EXECUTE; 2438 cache.cred = cred; 2439 status = NFS_PROTO(inode)->access(inode, &cache); 2440 if (status != 0) { 2441 if (status == -ESTALE) { 2442 nfs_zap_caches(inode); 2443 if (!S_ISDIR(inode->i_mode)) 2444 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 2445 } 2446 goto out; 2447 } 2448 nfs_access_add_cache(inode, &cache); 2449 out_cached: 2450 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); 2451 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2452 status = -EACCES; 2453 out: 2454 trace_nfs_access_exit(inode, status); 2455 return status; 2456 } 2457 2458 static int nfs_open_permission_mask(int openflags) 2459 { 2460 int mask = 0; 2461 2462 if (openflags & __FMODE_EXEC) { 2463 /* ONLY check exec rights */ 2464 mask = MAY_EXEC; 2465 } else { 2466 if ((openflags & O_ACCMODE) != O_WRONLY) 2467 mask |= MAY_READ; 2468 if ((openflags & O_ACCMODE) != O_RDONLY) 2469 mask |= MAY_WRITE; 2470 } 2471 2472 return mask; 2473 } 2474 2475 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 2476 { 2477 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2478 } 2479 EXPORT_SYMBOL_GPL(nfs_may_open); 2480 2481 static int nfs_execute_ok(struct inode *inode, int mask) 2482 { 2483 struct nfs_server *server = NFS_SERVER(inode); 2484 int ret = 0; 2485 2486 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) { 2487 if (mask & MAY_NOT_BLOCK) 2488 return -ECHILD; 2489 ret = __nfs_revalidate_inode(server, inode); 2490 } 2491 if (ret == 0 && !execute_ok(inode)) 2492 ret = -EACCES; 2493 return ret; 2494 } 2495 2496 int nfs_permission(struct inode *inode, int mask) 2497 { 2498 struct rpc_cred *cred; 2499 int res = 0; 2500 2501 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2502 2503 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2504 goto out; 2505 /* Is this sys_access() ? */ 2506 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2507 goto force_lookup; 2508 2509 switch (inode->i_mode & S_IFMT) { 2510 case S_IFLNK: 2511 goto out; 2512 case S_IFREG: 2513 if ((mask & MAY_OPEN) && 2514 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 2515 return 0; 2516 break; 2517 case S_IFDIR: 2518 /* 2519 * Optimize away all write operations, since the server 2520 * will check permissions when we perform the op. 2521 */ 2522 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2523 goto out; 2524 } 2525 2526 force_lookup: 2527 if (!NFS_PROTO(inode)->access) 2528 goto out_notsup; 2529 2530 /* Always try fast lookups first */ 2531 rcu_read_lock(); 2532 cred = rpc_lookup_cred_nonblock(); 2533 if (!IS_ERR(cred)) 2534 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); 2535 else 2536 res = PTR_ERR(cred); 2537 rcu_read_unlock(); 2538 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { 2539 /* Fast lookup failed, try the slow way */ 2540 cred = rpc_lookup_cred(); 2541 if (!IS_ERR(cred)) { 2542 res = nfs_do_access(inode, cred, mask); 2543 put_rpccred(cred); 2544 } else 2545 res = PTR_ERR(cred); 2546 } 2547 out: 2548 if (!res && (mask & MAY_EXEC)) 2549 res = nfs_execute_ok(inode, mask); 2550 2551 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2552 inode->i_sb->s_id, inode->i_ino, mask, res); 2553 return res; 2554 out_notsup: 2555 if (mask & MAY_NOT_BLOCK) 2556 return -ECHILD; 2557 2558 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2559 if (res == 0) 2560 res = generic_permission(inode, mask); 2561 goto out; 2562 } 2563 EXPORT_SYMBOL_GPL(nfs_permission); 2564 2565 /* 2566 * Local variables: 2567 * version-control: t 2568 * kept-new-versions: 5 2569 * End: 2570 */ 2571