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(file_inode(filp), 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(NULL, name, len); 226 return 0; 227 } 228 229 static 230 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 231 { 232 struct nfs_cache_array *array = nfs_readdir_get_array(page); 233 struct nfs_cache_array_entry *cache_entry; 234 int ret; 235 236 if (IS_ERR(array)) 237 return PTR_ERR(array); 238 239 cache_entry = &array->array[array->size]; 240 241 /* Check that this entry lies within the page bounds */ 242 ret = -ENOSPC; 243 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) 244 goto out; 245 246 cache_entry->cookie = entry->prev_cookie; 247 cache_entry->ino = entry->ino; 248 cache_entry->d_type = entry->d_type; 249 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); 250 if (ret) 251 goto out; 252 array->last_cookie = entry->cookie; 253 array->size++; 254 if (entry->eof != 0) 255 array->eof_index = array->size; 256 out: 257 nfs_readdir_release_array(page); 258 return ret; 259 } 260 261 static 262 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 263 { 264 loff_t diff = desc->ctx->pos - desc->current_index; 265 unsigned int index; 266 267 if (diff < 0) 268 goto out_eof; 269 if (diff >= array->size) { 270 if (array->eof_index >= 0) 271 goto out_eof; 272 return -EAGAIN; 273 } 274 275 index = (unsigned int)diff; 276 *desc->dir_cookie = array->array[index].cookie; 277 desc->cache_entry_index = index; 278 return 0; 279 out_eof: 280 desc->eof = 1; 281 return -EBADCOOKIE; 282 } 283 284 static bool 285 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) 286 { 287 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 288 return false; 289 smp_rmb(); 290 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); 291 } 292 293 static 294 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 295 { 296 int i; 297 loff_t new_pos; 298 int status = -EAGAIN; 299 300 for (i = 0; i < array->size; i++) { 301 if (array->array[i].cookie == *desc->dir_cookie) { 302 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); 303 struct nfs_open_dir_context *ctx = desc->file->private_data; 304 305 new_pos = desc->current_index + i; 306 if (ctx->attr_gencount != nfsi->attr_gencount || 307 !nfs_readdir_inode_mapping_valid(nfsi)) { 308 ctx->duped = 0; 309 ctx->attr_gencount = nfsi->attr_gencount; 310 } else if (new_pos < desc->ctx->pos) { 311 if (ctx->duped > 0 312 && ctx->dup_cookie == *desc->dir_cookie) { 313 if (printk_ratelimit()) { 314 pr_notice("NFS: directory %pD2 contains a readdir loop." 315 "Please contact your server vendor. " 316 "The file: %.*s has duplicate cookie %llu\n", 317 desc->file, array->array[i].string.len, 318 array->array[i].string.name, *desc->dir_cookie); 319 } 320 status = -ELOOP; 321 goto out; 322 } 323 ctx->dup_cookie = *desc->dir_cookie; 324 ctx->duped = -1; 325 } 326 desc->ctx->pos = new_pos; 327 desc->cache_entry_index = i; 328 return 0; 329 } 330 } 331 if (array->eof_index >= 0) { 332 status = -EBADCOOKIE; 333 if (*desc->dir_cookie == array->last_cookie) 334 desc->eof = 1; 335 } 336 out: 337 return status; 338 } 339 340 static 341 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) 342 { 343 struct nfs_cache_array *array; 344 int status; 345 346 array = nfs_readdir_get_array(desc->page); 347 if (IS_ERR(array)) { 348 status = PTR_ERR(array); 349 goto out; 350 } 351 352 if (*desc->dir_cookie == 0) 353 status = nfs_readdir_search_for_pos(array, desc); 354 else 355 status = nfs_readdir_search_for_cookie(array, desc); 356 357 if (status == -EAGAIN) { 358 desc->last_cookie = array->last_cookie; 359 desc->current_index += array->size; 360 desc->page_index++; 361 } 362 nfs_readdir_release_array(desc->page); 363 out: 364 return status; 365 } 366 367 /* Fill a page with xdr information before transferring to the cache page */ 368 static 369 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, 370 struct nfs_entry *entry, struct file *file, struct inode *inode) 371 { 372 struct nfs_open_dir_context *ctx = file->private_data; 373 struct rpc_cred *cred = ctx->cred; 374 unsigned long timestamp, gencount; 375 int error; 376 377 again: 378 timestamp = jiffies; 379 gencount = nfs_inc_attr_generation_counter(); 380 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages, 381 NFS_SERVER(inode)->dtsize, desc->plus); 382 if (error < 0) { 383 /* We requested READDIRPLUS, but the server doesn't grok it */ 384 if (error == -ENOTSUPP && desc->plus) { 385 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 386 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 387 desc->plus = 0; 388 goto again; 389 } 390 goto error; 391 } 392 desc->timestamp = timestamp; 393 desc->gencount = gencount; 394 error: 395 return error; 396 } 397 398 static int xdr_decode(nfs_readdir_descriptor_t *desc, 399 struct nfs_entry *entry, struct xdr_stream *xdr) 400 { 401 int error; 402 403 error = desc->decode(xdr, entry, desc->plus); 404 if (error) 405 return error; 406 entry->fattr->time_start = desc->timestamp; 407 entry->fattr->gencount = desc->gencount; 408 return 0; 409 } 410 411 /* Match file and dirent using either filehandle or fileid 412 * Note: caller is responsible for checking the fsid 413 */ 414 static 415 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 416 { 417 struct inode *inode; 418 struct nfs_inode *nfsi; 419 420 if (d_really_is_negative(dentry)) 421 return 0; 422 423 inode = d_inode(dentry); 424 if (is_bad_inode(inode) || NFS_STALE(inode)) 425 return 0; 426 427 nfsi = NFS_I(inode); 428 if (entry->fattr->fileid != nfsi->fileid) 429 return 0; 430 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) 431 return 0; 432 return 1; 433 } 434 435 static 436 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) 437 { 438 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 439 return false; 440 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 441 return true; 442 if (ctx->pos == 0) 443 return true; 444 return false; 445 } 446 447 /* 448 * This function is called by the lookup and getattr code to request the 449 * use of readdirplus to accelerate any future lookups in the same 450 * directory. 451 */ 452 void nfs_advise_use_readdirplus(struct inode *dir) 453 { 454 struct nfs_inode *nfsi = NFS_I(dir); 455 456 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 457 !list_empty(&nfsi->open_files)) 458 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 459 } 460 461 /* 462 * This function is mainly for use by nfs_getattr(). 463 * 464 * If this is an 'ls -l', we want to force use of readdirplus. 465 * Do this by checking if there is an active file descriptor 466 * and calling nfs_advise_use_readdirplus, then forcing a 467 * cache flush. 468 */ 469 void nfs_force_use_readdirplus(struct inode *dir) 470 { 471 struct nfs_inode *nfsi = NFS_I(dir); 472 473 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 474 !list_empty(&nfsi->open_files)) { 475 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 476 invalidate_mapping_pages(dir->i_mapping, 0, -1); 477 } 478 } 479 480 static 481 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) 482 { 483 struct qstr filename = QSTR_INIT(entry->name, entry->len); 484 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 485 struct dentry *dentry; 486 struct dentry *alias; 487 struct inode *dir = d_inode(parent); 488 struct inode *inode; 489 int status; 490 491 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 492 return; 493 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 494 return; 495 if (filename.len == 0) 496 return; 497 /* Validate that the name doesn't contain any illegal '\0' */ 498 if (strnlen(filename.name, filename.len) != filename.len) 499 return; 500 /* ...or '/' */ 501 if (strnchr(filename.name, filename.len, '/')) 502 return; 503 if (filename.name[0] == '.') { 504 if (filename.len == 1) 505 return; 506 if (filename.len == 2 && filename.name[1] == '.') 507 return; 508 } 509 filename.hash = full_name_hash(parent, filename.name, filename.len); 510 511 dentry = d_lookup(parent, &filename); 512 again: 513 if (!dentry) { 514 dentry = d_alloc_parallel(parent, &filename, &wq); 515 if (IS_ERR(dentry)) 516 return; 517 } 518 if (!d_in_lookup(dentry)) { 519 /* Is there a mountpoint here? If so, just exit */ 520 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 521 &entry->fattr->fsid)) 522 goto out; 523 if (nfs_same_file(dentry, entry)) { 524 if (!entry->fh->size) 525 goto out; 526 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 527 status = nfs_refresh_inode(d_inode(dentry), entry->fattr); 528 if (!status) 529 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); 530 goto out; 531 } else { 532 d_invalidate(dentry); 533 dput(dentry); 534 dentry = NULL; 535 goto again; 536 } 537 } 538 if (!entry->fh->size) { 539 d_lookup_done(dentry); 540 goto out; 541 } 542 543 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 544 alias = d_splice_alias(inode, dentry); 545 d_lookup_done(dentry); 546 if (alias) { 547 if (IS_ERR(alias)) 548 goto out; 549 dput(dentry); 550 dentry = alias; 551 } 552 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 553 out: 554 dput(dentry); 555 } 556 557 /* Perform conversion from xdr to cache array */ 558 static 559 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 560 struct page **xdr_pages, struct page *page, unsigned int buflen) 561 { 562 struct xdr_stream stream; 563 struct xdr_buf buf; 564 struct page *scratch; 565 struct nfs_cache_array *array; 566 unsigned int count = 0; 567 int status; 568 569 scratch = alloc_page(GFP_KERNEL); 570 if (scratch == NULL) 571 return -ENOMEM; 572 573 if (buflen == 0) 574 goto out_nopages; 575 576 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 577 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 578 579 do { 580 status = xdr_decode(desc, entry, &stream); 581 if (status != 0) { 582 if (status == -EAGAIN) 583 status = 0; 584 break; 585 } 586 587 count++; 588 589 if (desc->plus != 0) 590 nfs_prime_dcache(file_dentry(desc->file), entry); 591 592 status = nfs_readdir_add_to_array(entry, page); 593 if (status != 0) 594 break; 595 } while (!entry->eof); 596 597 out_nopages: 598 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { 599 array = nfs_readdir_get_array(page); 600 if (!IS_ERR(array)) { 601 array->eof_index = array->size; 602 status = 0; 603 nfs_readdir_release_array(page); 604 } else 605 status = PTR_ERR(array); 606 } 607 608 put_page(scratch); 609 return status; 610 } 611 612 static 613 void nfs_readdir_free_pages(struct page **pages, unsigned int npages) 614 { 615 unsigned int i; 616 for (i = 0; i < npages; i++) 617 put_page(pages[i]); 618 } 619 620 /* 621 * nfs_readdir_large_page will allocate pages that must be freed with a call 622 * to nfs_readdir_free_pagearray 623 */ 624 static 625 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages) 626 { 627 unsigned int i; 628 629 for (i = 0; i < npages; i++) { 630 struct page *page = alloc_page(GFP_KERNEL); 631 if (page == NULL) 632 goto out_freepages; 633 pages[i] = page; 634 } 635 return 0; 636 637 out_freepages: 638 nfs_readdir_free_pages(pages, i); 639 return -ENOMEM; 640 } 641 642 static 643 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) 644 { 645 struct page *pages[NFS_MAX_READDIR_PAGES]; 646 struct nfs_entry entry; 647 struct file *file = desc->file; 648 struct nfs_cache_array *array; 649 int status = -ENOMEM; 650 unsigned int array_size = ARRAY_SIZE(pages); 651 652 entry.prev_cookie = 0; 653 entry.cookie = desc->last_cookie; 654 entry.eof = 0; 655 entry.fh = nfs_alloc_fhandle(); 656 entry.fattr = nfs_alloc_fattr(); 657 entry.server = NFS_SERVER(inode); 658 if (entry.fh == NULL || entry.fattr == NULL) 659 goto out; 660 661 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 662 if (IS_ERR(entry.label)) { 663 status = PTR_ERR(entry.label); 664 goto out; 665 } 666 667 array = nfs_readdir_get_array(page); 668 if (IS_ERR(array)) { 669 status = PTR_ERR(array); 670 goto out_label_free; 671 } 672 memset(array, 0, sizeof(struct nfs_cache_array)); 673 array->eof_index = -1; 674 675 status = nfs_readdir_alloc_pages(pages, array_size); 676 if (status < 0) 677 goto out_release_array; 678 do { 679 unsigned int pglen; 680 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 681 682 if (status < 0) 683 break; 684 pglen = status; 685 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 686 if (status < 0) { 687 if (status == -ENOSPC) 688 status = 0; 689 break; 690 } 691 } while (array->eof_index < 0); 692 693 nfs_readdir_free_pages(pages, array_size); 694 out_release_array: 695 nfs_readdir_release_array(page); 696 out_label_free: 697 nfs4_label_free(entry.label); 698 out: 699 nfs_free_fattr(entry.fattr); 700 nfs_free_fhandle(entry.fh); 701 return status; 702 } 703 704 /* 705 * Now we cache directories properly, by converting xdr information 706 * to an array that can be used for lookups later. This results in 707 * fewer cache pages, since we can store more information on each page. 708 * We only need to convert from xdr once so future lookups are much simpler 709 */ 710 static 711 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) 712 { 713 struct inode *inode = file_inode(desc->file); 714 int ret; 715 716 ret = nfs_readdir_xdr_to_array(desc, page, inode); 717 if (ret < 0) 718 goto error; 719 SetPageUptodate(page); 720 721 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 722 /* Should never happen */ 723 nfs_zap_mapping(inode, inode->i_mapping); 724 } 725 unlock_page(page); 726 return 0; 727 error: 728 unlock_page(page); 729 return ret; 730 } 731 732 static 733 void cache_page_release(nfs_readdir_descriptor_t *desc) 734 { 735 if (!desc->page->mapping) 736 nfs_readdir_clear_array(desc->page); 737 put_page(desc->page); 738 desc->page = NULL; 739 } 740 741 static 742 struct page *get_cache_page(nfs_readdir_descriptor_t *desc) 743 { 744 return read_cache_page(desc->file->f_mapping, 745 desc->page_index, (filler_t *)nfs_readdir_filler, desc); 746 } 747 748 /* 749 * Returns 0 if desc->dir_cookie was found on page desc->page_index 750 */ 751 static 752 int find_cache_page(nfs_readdir_descriptor_t *desc) 753 { 754 int res; 755 756 desc->page = get_cache_page(desc); 757 if (IS_ERR(desc->page)) 758 return PTR_ERR(desc->page); 759 760 res = nfs_readdir_search_array(desc); 761 if (res != 0) 762 cache_page_release(desc); 763 return res; 764 } 765 766 /* Search for desc->dir_cookie from the beginning of the page cache */ 767 static inline 768 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 769 { 770 int res; 771 772 if (desc->page_index == 0) { 773 desc->current_index = 0; 774 desc->last_cookie = 0; 775 } 776 do { 777 res = find_cache_page(desc); 778 } while (res == -EAGAIN); 779 return res; 780 } 781 782 /* 783 * Once we've found the start of the dirent within a page: fill 'er up... 784 */ 785 static 786 int nfs_do_filldir(nfs_readdir_descriptor_t *desc) 787 { 788 struct file *file = desc->file; 789 int i = 0; 790 int res = 0; 791 struct nfs_cache_array *array = NULL; 792 struct nfs_open_dir_context *ctx = file->private_data; 793 794 array = nfs_readdir_get_array(desc->page); 795 if (IS_ERR(array)) { 796 res = PTR_ERR(array); 797 goto out; 798 } 799 800 for (i = desc->cache_entry_index; i < array->size; i++) { 801 struct nfs_cache_array_entry *ent; 802 803 ent = &array->array[i]; 804 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, 805 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 806 desc->eof = 1; 807 break; 808 } 809 desc->ctx->pos++; 810 if (i < (array->size-1)) 811 *desc->dir_cookie = array->array[i+1].cookie; 812 else 813 *desc->dir_cookie = array->last_cookie; 814 if (ctx->duped != 0) 815 ctx->duped = 1; 816 } 817 if (array->eof_index >= 0) 818 desc->eof = 1; 819 820 nfs_readdir_release_array(desc->page); 821 out: 822 cache_page_release(desc); 823 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 824 (unsigned long long)*desc->dir_cookie, res); 825 return res; 826 } 827 828 /* 829 * If we cannot find a cookie in our cache, we suspect that this is 830 * because it points to a deleted file, so we ask the server to return 831 * whatever it thinks is the next entry. We then feed this to filldir. 832 * If all goes well, we should then be able to find our way round the 833 * cache on the next call to readdir_search_pagecache(); 834 * 835 * NOTE: we cannot add the anonymous page to the pagecache because 836 * the data it contains might not be page aligned. Besides, 837 * we should already have a complete representation of the 838 * directory in the page cache by the time we get here. 839 */ 840 static inline 841 int uncached_readdir(nfs_readdir_descriptor_t *desc) 842 { 843 struct page *page = NULL; 844 int status; 845 struct inode *inode = file_inode(desc->file); 846 struct nfs_open_dir_context *ctx = desc->file->private_data; 847 848 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 849 (unsigned long long)*desc->dir_cookie); 850 851 page = alloc_page(GFP_HIGHUSER); 852 if (!page) { 853 status = -ENOMEM; 854 goto out; 855 } 856 857 desc->page_index = 0; 858 desc->last_cookie = *desc->dir_cookie; 859 desc->page = page; 860 ctx->duped = 0; 861 862 status = nfs_readdir_xdr_to_array(desc, page, inode); 863 if (status < 0) 864 goto out_release; 865 866 status = nfs_do_filldir(desc); 867 868 out: 869 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 870 __func__, status); 871 return status; 872 out_release: 873 cache_page_release(desc); 874 goto out; 875 } 876 877 /* The file offset position represents the dirent entry number. A 878 last cookie cache takes care of the common case of reading the 879 whole directory. 880 */ 881 static int nfs_readdir(struct file *file, struct dir_context *ctx) 882 { 883 struct dentry *dentry = file_dentry(file); 884 struct inode *inode = d_inode(dentry); 885 nfs_readdir_descriptor_t my_desc, 886 *desc = &my_desc; 887 struct nfs_open_dir_context *dir_ctx = file->private_data; 888 int res = 0; 889 890 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 891 file, (long long)ctx->pos); 892 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 893 894 /* 895 * ctx->pos points to the dirent entry number. 896 * *desc->dir_cookie has the cookie for the next entry. We have 897 * to either find the entry with the appropriate number or 898 * revalidate the cookie. 899 */ 900 memset(desc, 0, sizeof(*desc)); 901 902 desc->file = file; 903 desc->ctx = ctx; 904 desc->dir_cookie = &dir_ctx->dir_cookie; 905 desc->decode = NFS_PROTO(inode)->decode_dirent; 906 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0; 907 908 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) 909 res = nfs_revalidate_mapping(inode, file->f_mapping); 910 if (res < 0) 911 goto out; 912 913 do { 914 res = readdir_search_pagecache(desc); 915 916 if (res == -EBADCOOKIE) { 917 res = 0; 918 /* This means either end of directory */ 919 if (*desc->dir_cookie && desc->eof == 0) { 920 /* Or that the server has 'lost' a cookie */ 921 res = uncached_readdir(desc); 922 if (res == 0) 923 continue; 924 } 925 break; 926 } 927 if (res == -ETOOSMALL && desc->plus) { 928 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 929 nfs_zap_caches(inode); 930 desc->page_index = 0; 931 desc->plus = 0; 932 desc->eof = 0; 933 continue; 934 } 935 if (res < 0) 936 break; 937 938 res = nfs_do_filldir(desc); 939 if (res < 0) 940 break; 941 } while (!desc->eof); 942 out: 943 if (res > 0) 944 res = 0; 945 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 946 return res; 947 } 948 949 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 950 { 951 struct inode *inode = file_inode(filp); 952 struct nfs_open_dir_context *dir_ctx = filp->private_data; 953 954 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 955 filp, offset, whence); 956 957 inode_lock(inode); 958 switch (whence) { 959 case 1: 960 offset += filp->f_pos; 961 case 0: 962 if (offset >= 0) 963 break; 964 default: 965 offset = -EINVAL; 966 goto out; 967 } 968 if (offset != filp->f_pos) { 969 filp->f_pos = offset; 970 dir_ctx->dir_cookie = 0; 971 dir_ctx->duped = 0; 972 } 973 out: 974 inode_unlock(inode); 975 return offset; 976 } 977 978 /* 979 * All directory operations under NFS are synchronous, so fsync() 980 * is a dummy operation. 981 */ 982 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 983 int datasync) 984 { 985 struct inode *inode = file_inode(filp); 986 987 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 988 989 inode_lock(inode); 990 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 991 inode_unlock(inode); 992 return 0; 993 } 994 995 /** 996 * nfs_force_lookup_revalidate - Mark the directory as having changed 997 * @dir - pointer to directory inode 998 * 999 * This forces the revalidation code in nfs_lookup_revalidate() to do a 1000 * full lookup on all child dentries of 'dir' whenever a change occurs 1001 * on the server that might have invalidated our dcache. 1002 * 1003 * The caller should be holding dir->i_lock 1004 */ 1005 void nfs_force_lookup_revalidate(struct inode *dir) 1006 { 1007 NFS_I(dir)->cache_change_attribute++; 1008 } 1009 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1010 1011 /* 1012 * A check for whether or not the parent directory has changed. 1013 * In the case it has, we assume that the dentries are untrustworthy 1014 * and may need to be looked up again. 1015 * If rcu_walk prevents us from performing a full check, return 0. 1016 */ 1017 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1018 int rcu_walk) 1019 { 1020 if (IS_ROOT(dentry)) 1021 return 1; 1022 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1023 return 0; 1024 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1025 return 0; 1026 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1027 if (nfs_mapping_need_revalidate_inode(dir)) { 1028 if (rcu_walk) 1029 return 0; 1030 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 1031 return 0; 1032 } 1033 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1034 return 0; 1035 return 1; 1036 } 1037 1038 /* 1039 * Use intent information to check whether or not we're going to do 1040 * an O_EXCL create using this path component. 1041 */ 1042 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1043 { 1044 if (NFS_PROTO(dir)->version == 2) 1045 return 0; 1046 return flags & LOOKUP_EXCL; 1047 } 1048 1049 /* 1050 * Inode and filehandle revalidation for lookups. 1051 * 1052 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1053 * or if the intent information indicates that we're about to open this 1054 * particular file and the "nocto" mount flag is not set. 1055 * 1056 */ 1057 static 1058 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1059 { 1060 struct nfs_server *server = NFS_SERVER(inode); 1061 int ret; 1062 1063 if (IS_AUTOMOUNT(inode)) 1064 return 0; 1065 /* VFS wants an on-the-wire revalidation */ 1066 if (flags & LOOKUP_REVAL) 1067 goto out_force; 1068 /* This is an open(2) */ 1069 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && 1070 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) 1071 goto out_force; 1072 out: 1073 return (inode->i_nlink == 0) ? -ENOENT : 0; 1074 out_force: 1075 if (flags & LOOKUP_RCU) 1076 return -ECHILD; 1077 ret = __nfs_revalidate_inode(server, inode); 1078 if (ret != 0) 1079 return ret; 1080 goto out; 1081 } 1082 1083 /* 1084 * We judge how long we want to trust negative 1085 * dentries by looking at the parent inode mtime. 1086 * 1087 * If parent mtime has changed, we revalidate, else we wait for a 1088 * period corresponding to the parent's attribute cache timeout value. 1089 * 1090 * If LOOKUP_RCU prevents us from performing a full check, return 1 1091 * suggesting a reval is needed. 1092 */ 1093 static inline 1094 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1095 unsigned int flags) 1096 { 1097 /* Don't revalidate a negative dentry if we're creating a new file */ 1098 if (flags & LOOKUP_CREATE) 1099 return 0; 1100 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1101 return 1; 1102 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1103 } 1104 1105 /* 1106 * This is called every time the dcache has a lookup hit, 1107 * and we should check whether we can really trust that 1108 * lookup. 1109 * 1110 * NOTE! The hit can be a negative hit too, don't assume 1111 * we have an inode! 1112 * 1113 * If the parent directory is seen to have changed, we throw out the 1114 * cached dentry and do a new lookup. 1115 */ 1116 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1117 { 1118 struct inode *dir; 1119 struct inode *inode; 1120 struct dentry *parent; 1121 struct nfs_fh *fhandle = NULL; 1122 struct nfs_fattr *fattr = NULL; 1123 struct nfs4_label *label = NULL; 1124 int error; 1125 1126 if (flags & LOOKUP_RCU) { 1127 parent = ACCESS_ONCE(dentry->d_parent); 1128 dir = d_inode_rcu(parent); 1129 if (!dir) 1130 return -ECHILD; 1131 } else { 1132 parent = dget_parent(dentry); 1133 dir = d_inode(parent); 1134 } 1135 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1136 inode = d_inode(dentry); 1137 1138 if (!inode) { 1139 if (nfs_neg_need_reval(dir, dentry, flags)) { 1140 if (flags & LOOKUP_RCU) 1141 return -ECHILD; 1142 goto out_bad; 1143 } 1144 goto out_valid; 1145 } 1146 1147 if (is_bad_inode(inode)) { 1148 if (flags & LOOKUP_RCU) 1149 return -ECHILD; 1150 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1151 __func__, dentry); 1152 goto out_bad; 1153 } 1154 1155 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) 1156 goto out_set_verifier; 1157 1158 /* Force a full look up iff the parent directory has changed */ 1159 if (!nfs_is_exclusive_create(dir, flags) && 1160 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1161 1162 if (nfs_lookup_verify_inode(inode, flags)) { 1163 if (flags & LOOKUP_RCU) 1164 return -ECHILD; 1165 goto out_zap_parent; 1166 } 1167 nfs_advise_use_readdirplus(dir); 1168 goto out_valid; 1169 } 1170 1171 if (flags & LOOKUP_RCU) 1172 return -ECHILD; 1173 1174 if (NFS_STALE(inode)) 1175 goto out_bad; 1176 1177 error = -ENOMEM; 1178 fhandle = nfs_alloc_fhandle(); 1179 fattr = nfs_alloc_fattr(); 1180 if (fhandle == NULL || fattr == NULL) 1181 goto out_error; 1182 1183 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 1184 if (IS_ERR(label)) 1185 goto out_error; 1186 1187 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1188 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1189 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1190 if (error) 1191 goto out_bad; 1192 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1193 goto out_bad; 1194 if ((error = nfs_refresh_inode(inode, fattr)) != 0) 1195 goto out_bad; 1196 1197 nfs_setsecurity(inode, fattr, label); 1198 1199 nfs_free_fattr(fattr); 1200 nfs_free_fhandle(fhandle); 1201 nfs4_label_free(label); 1202 1203 /* set a readdirplus hint that we had a cache miss */ 1204 nfs_force_use_readdirplus(dir); 1205 1206 out_set_verifier: 1207 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1208 out_valid: 1209 if (flags & LOOKUP_RCU) { 1210 if (parent != ACCESS_ONCE(dentry->d_parent)) 1211 return -ECHILD; 1212 } else 1213 dput(parent); 1214 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1215 __func__, dentry); 1216 return 1; 1217 out_zap_parent: 1218 nfs_zap_caches(dir); 1219 out_bad: 1220 WARN_ON(flags & LOOKUP_RCU); 1221 nfs_free_fattr(fattr); 1222 nfs_free_fhandle(fhandle); 1223 nfs4_label_free(label); 1224 nfs_mark_for_revalidate(dir); 1225 if (inode && S_ISDIR(inode->i_mode)) { 1226 /* Purge readdir caches. */ 1227 nfs_zap_caches(inode); 1228 /* 1229 * We can't d_drop the root of a disconnected tree: 1230 * its d_hash is on the s_anon list and d_drop() would hide 1231 * it from shrink_dcache_for_unmount(), leading to busy 1232 * inodes on unmount and further oopses. 1233 */ 1234 if (IS_ROOT(dentry)) 1235 goto out_valid; 1236 } 1237 dput(parent); 1238 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1239 __func__, dentry); 1240 return 0; 1241 out_error: 1242 WARN_ON(flags & LOOKUP_RCU); 1243 nfs_free_fattr(fattr); 1244 nfs_free_fhandle(fhandle); 1245 nfs4_label_free(label); 1246 dput(parent); 1247 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1248 __func__, dentry, error); 1249 return error; 1250 } 1251 1252 /* 1253 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1254 * when we don't really care about the dentry name. This is called when a 1255 * pathwalk ends on a dentry that was not found via a normal lookup in the 1256 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1257 * 1258 * In this situation, we just want to verify that the inode itself is OK 1259 * since the dentry might have changed on the server. 1260 */ 1261 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1262 { 1263 struct inode *inode = d_inode(dentry); 1264 int error = 0; 1265 1266 /* 1267 * I believe we can only get a negative dentry here in the case of a 1268 * procfs-style symlink. Just assume it's correct for now, but we may 1269 * eventually need to do something more here. 1270 */ 1271 if (!inode) { 1272 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1273 __func__, dentry); 1274 return 1; 1275 } 1276 1277 if (is_bad_inode(inode)) { 1278 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1279 __func__, dentry); 1280 return 0; 1281 } 1282 1283 if (nfs_mapping_need_revalidate_inode(inode)) 1284 error = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 1285 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1286 __func__, inode->i_ino, error ? "invalid" : "valid"); 1287 return !error; 1288 } 1289 1290 /* 1291 * This is called from dput() when d_count is going to 0. 1292 */ 1293 static int nfs_dentry_delete(const struct dentry *dentry) 1294 { 1295 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1296 dentry, dentry->d_flags); 1297 1298 /* Unhash any dentry with a stale inode */ 1299 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1300 return 1; 1301 1302 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1303 /* Unhash it, so that ->d_iput() would be called */ 1304 return 1; 1305 } 1306 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 1307 /* Unhash it, so that ancestors of killed async unlink 1308 * files will be cleaned up during umount */ 1309 return 1; 1310 } 1311 return 0; 1312 1313 } 1314 1315 /* Ensure that we revalidate inode->i_nlink */ 1316 static void nfs_drop_nlink(struct inode *inode) 1317 { 1318 spin_lock(&inode->i_lock); 1319 /* drop the inode if we're reasonably sure this is the last link */ 1320 if (inode->i_nlink == 1) 1321 clear_nlink(inode); 1322 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; 1323 spin_unlock(&inode->i_lock); 1324 } 1325 1326 /* 1327 * Called when the dentry loses inode. 1328 * We use it to clean up silly-renamed files. 1329 */ 1330 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1331 { 1332 if (S_ISDIR(inode->i_mode)) 1333 /* drop any readdir cache as it could easily be old */ 1334 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1335 1336 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1337 nfs_complete_unlink(dentry, inode); 1338 nfs_drop_nlink(inode); 1339 } 1340 iput(inode); 1341 } 1342 1343 static void nfs_d_release(struct dentry *dentry) 1344 { 1345 /* free cached devname value, if it survived that far */ 1346 if (unlikely(dentry->d_fsdata)) { 1347 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1348 WARN_ON(1); 1349 else 1350 kfree(dentry->d_fsdata); 1351 } 1352 } 1353 1354 const struct dentry_operations nfs_dentry_operations = { 1355 .d_revalidate = nfs_lookup_revalidate, 1356 .d_weak_revalidate = nfs_weak_revalidate, 1357 .d_delete = nfs_dentry_delete, 1358 .d_iput = nfs_dentry_iput, 1359 .d_automount = nfs_d_automount, 1360 .d_release = nfs_d_release, 1361 }; 1362 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1363 1364 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1365 { 1366 struct dentry *res; 1367 struct inode *inode = NULL; 1368 struct nfs_fh *fhandle = NULL; 1369 struct nfs_fattr *fattr = NULL; 1370 struct nfs4_label *label = NULL; 1371 int error; 1372 1373 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1374 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1375 1376 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1377 return ERR_PTR(-ENAMETOOLONG); 1378 1379 /* 1380 * If we're doing an exclusive create, optimize away the lookup 1381 * but don't hash the dentry. 1382 */ 1383 if (nfs_is_exclusive_create(dir, flags)) 1384 return NULL; 1385 1386 res = ERR_PTR(-ENOMEM); 1387 fhandle = nfs_alloc_fhandle(); 1388 fattr = nfs_alloc_fattr(); 1389 if (fhandle == NULL || fattr == NULL) 1390 goto out; 1391 1392 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1393 if (IS_ERR(label)) 1394 goto out; 1395 1396 trace_nfs_lookup_enter(dir, dentry, flags); 1397 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1398 if (error == -ENOENT) 1399 goto no_entry; 1400 if (error < 0) { 1401 res = ERR_PTR(error); 1402 goto out_label; 1403 } 1404 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1405 res = ERR_CAST(inode); 1406 if (IS_ERR(res)) 1407 goto out_label; 1408 1409 /* Notify readdir to use READDIRPLUS */ 1410 nfs_force_use_readdirplus(dir); 1411 1412 no_entry: 1413 res = d_splice_alias(inode, dentry); 1414 if (res != NULL) { 1415 if (IS_ERR(res)) 1416 goto out_label; 1417 dentry = res; 1418 } 1419 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1420 out_label: 1421 trace_nfs_lookup_exit(dir, dentry, flags, error); 1422 nfs4_label_free(label); 1423 out: 1424 nfs_free_fattr(fattr); 1425 nfs_free_fhandle(fhandle); 1426 return res; 1427 } 1428 EXPORT_SYMBOL_GPL(nfs_lookup); 1429 1430 #if IS_ENABLED(CONFIG_NFS_V4) 1431 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1432 1433 const struct dentry_operations nfs4_dentry_operations = { 1434 .d_revalidate = nfs4_lookup_revalidate, 1435 .d_delete = nfs_dentry_delete, 1436 .d_iput = nfs_dentry_iput, 1437 .d_automount = nfs_d_automount, 1438 .d_release = nfs_d_release, 1439 }; 1440 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1441 1442 static fmode_t flags_to_mode(int flags) 1443 { 1444 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1445 if ((flags & O_ACCMODE) != O_WRONLY) 1446 res |= FMODE_READ; 1447 if ((flags & O_ACCMODE) != O_RDONLY) 1448 res |= FMODE_WRITE; 1449 return res; 1450 } 1451 1452 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) 1453 { 1454 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); 1455 } 1456 1457 static int do_open(struct inode *inode, struct file *filp) 1458 { 1459 nfs_fscache_open_file(inode, filp); 1460 return 0; 1461 } 1462 1463 static int nfs_finish_open(struct nfs_open_context *ctx, 1464 struct dentry *dentry, 1465 struct file *file, unsigned open_flags, 1466 int *opened) 1467 { 1468 int err; 1469 1470 err = finish_open(file, dentry, do_open, opened); 1471 if (err) 1472 goto out; 1473 nfs_file_set_open_context(file, ctx); 1474 1475 out: 1476 return err; 1477 } 1478 1479 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1480 struct file *file, unsigned open_flags, 1481 umode_t mode, int *opened) 1482 { 1483 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1484 struct nfs_open_context *ctx; 1485 struct dentry *res; 1486 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1487 struct inode *inode; 1488 unsigned int lookup_flags = 0; 1489 bool switched = false; 1490 int err; 1491 1492 /* Expect a negative dentry */ 1493 BUG_ON(d_inode(dentry)); 1494 1495 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1496 dir->i_sb->s_id, dir->i_ino, dentry); 1497 1498 err = nfs_check_flags(open_flags); 1499 if (err) 1500 return err; 1501 1502 /* NFS only supports OPEN on regular files */ 1503 if ((open_flags & O_DIRECTORY)) { 1504 if (!d_in_lookup(dentry)) { 1505 /* 1506 * Hashed negative dentry with O_DIRECTORY: dentry was 1507 * revalidated and is fine, no need to perform lookup 1508 * again 1509 */ 1510 return -ENOENT; 1511 } 1512 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1513 goto no_open; 1514 } 1515 1516 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1517 return -ENAMETOOLONG; 1518 1519 if (open_flags & O_CREAT) { 1520 struct nfs_server *server = NFS_SERVER(dir); 1521 1522 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) 1523 mode &= ~current_umask(); 1524 1525 attr.ia_valid |= ATTR_MODE; 1526 attr.ia_mode = mode; 1527 } 1528 if (open_flags & O_TRUNC) { 1529 attr.ia_valid |= ATTR_SIZE; 1530 attr.ia_size = 0; 1531 } 1532 1533 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 1534 d_drop(dentry); 1535 switched = true; 1536 dentry = d_alloc_parallel(dentry->d_parent, 1537 &dentry->d_name, &wq); 1538 if (IS_ERR(dentry)) 1539 return PTR_ERR(dentry); 1540 if (unlikely(!d_in_lookup(dentry))) 1541 return finish_no_open(file, dentry); 1542 } 1543 1544 ctx = create_nfs_open_context(dentry, open_flags, file); 1545 err = PTR_ERR(ctx); 1546 if (IS_ERR(ctx)) 1547 goto out; 1548 1549 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1550 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened); 1551 if (IS_ERR(inode)) { 1552 err = PTR_ERR(inode); 1553 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1554 put_nfs_open_context(ctx); 1555 d_drop(dentry); 1556 switch (err) { 1557 case -ENOENT: 1558 d_add(dentry, NULL); 1559 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1560 break; 1561 case -EISDIR: 1562 case -ENOTDIR: 1563 goto no_open; 1564 case -ELOOP: 1565 if (!(open_flags & O_NOFOLLOW)) 1566 goto no_open; 1567 break; 1568 /* case -EINVAL: */ 1569 default: 1570 break; 1571 } 1572 goto out; 1573 } 1574 1575 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened); 1576 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1577 put_nfs_open_context(ctx); 1578 out: 1579 if (unlikely(switched)) { 1580 d_lookup_done(dentry); 1581 dput(dentry); 1582 } 1583 return err; 1584 1585 no_open: 1586 res = nfs_lookup(dir, dentry, lookup_flags); 1587 if (switched) { 1588 d_lookup_done(dentry); 1589 if (!res) 1590 res = dentry; 1591 else 1592 dput(dentry); 1593 } 1594 if (IS_ERR(res)) 1595 return PTR_ERR(res); 1596 return finish_no_open(file, res); 1597 } 1598 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1599 1600 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1601 { 1602 struct inode *inode; 1603 int ret = 0; 1604 1605 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1606 goto no_open; 1607 if (d_mountpoint(dentry)) 1608 goto no_open; 1609 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) 1610 goto no_open; 1611 1612 inode = d_inode(dentry); 1613 1614 /* We can't create new files in nfs_open_revalidate(), so we 1615 * optimize away revalidation of negative dentries. 1616 */ 1617 if (inode == NULL) { 1618 struct dentry *parent; 1619 struct inode *dir; 1620 1621 if (flags & LOOKUP_RCU) { 1622 parent = ACCESS_ONCE(dentry->d_parent); 1623 dir = d_inode_rcu(parent); 1624 if (!dir) 1625 return -ECHILD; 1626 } else { 1627 parent = dget_parent(dentry); 1628 dir = d_inode(parent); 1629 } 1630 if (!nfs_neg_need_reval(dir, dentry, flags)) 1631 ret = 1; 1632 else if (flags & LOOKUP_RCU) 1633 ret = -ECHILD; 1634 if (!(flags & LOOKUP_RCU)) 1635 dput(parent); 1636 else if (parent != ACCESS_ONCE(dentry->d_parent)) 1637 return -ECHILD; 1638 goto out; 1639 } 1640 1641 /* NFS only supports OPEN on regular files */ 1642 if (!S_ISREG(inode->i_mode)) 1643 goto no_open; 1644 /* We cannot do exclusive creation on a positive dentry */ 1645 if (flags & LOOKUP_EXCL) 1646 goto no_open; 1647 1648 /* Let f_op->open() actually open (and revalidate) the file */ 1649 ret = 1; 1650 1651 out: 1652 return ret; 1653 1654 no_open: 1655 return nfs_lookup_revalidate(dentry, flags); 1656 } 1657 1658 #endif /* CONFIG_NFSV4 */ 1659 1660 /* 1661 * Code common to create, mkdir, and mknod. 1662 */ 1663 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1664 struct nfs_fattr *fattr, 1665 struct nfs4_label *label) 1666 { 1667 struct dentry *parent = dget_parent(dentry); 1668 struct inode *dir = d_inode(parent); 1669 struct inode *inode; 1670 int error = -EACCES; 1671 1672 d_drop(dentry); 1673 1674 /* We may have been initialized further down */ 1675 if (d_really_is_positive(dentry)) 1676 goto out; 1677 if (fhandle->size == 0) { 1678 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); 1679 if (error) 1680 goto out_error; 1681 } 1682 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1683 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1684 struct nfs_server *server = NFS_SB(dentry->d_sb); 1685 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); 1686 if (error < 0) 1687 goto out_error; 1688 } 1689 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1690 error = PTR_ERR(inode); 1691 if (IS_ERR(inode)) 1692 goto out_error; 1693 d_add(dentry, inode); 1694 out: 1695 dput(parent); 1696 return 0; 1697 out_error: 1698 nfs_mark_for_revalidate(dir); 1699 dput(parent); 1700 return error; 1701 } 1702 EXPORT_SYMBOL_GPL(nfs_instantiate); 1703 1704 /* 1705 * Following a failed create operation, we drop the dentry rather 1706 * than retain a negative dentry. This avoids a problem in the event 1707 * that the operation succeeded on the server, but an error in the 1708 * reply path made it appear to have failed. 1709 */ 1710 int nfs_create(struct inode *dir, struct dentry *dentry, 1711 umode_t mode, bool excl) 1712 { 1713 struct iattr attr; 1714 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 1715 int error; 1716 1717 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 1718 dir->i_sb->s_id, dir->i_ino, dentry); 1719 1720 attr.ia_mode = mode; 1721 attr.ia_valid = ATTR_MODE; 1722 1723 trace_nfs_create_enter(dir, dentry, open_flags); 1724 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1725 trace_nfs_create_exit(dir, dentry, open_flags, error); 1726 if (error != 0) 1727 goto out_err; 1728 return 0; 1729 out_err: 1730 d_drop(dentry); 1731 return error; 1732 } 1733 EXPORT_SYMBOL_GPL(nfs_create); 1734 1735 /* 1736 * See comments for nfs_proc_create regarding failed operations. 1737 */ 1738 int 1739 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) 1740 { 1741 struct iattr attr; 1742 int status; 1743 1744 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 1745 dir->i_sb->s_id, dir->i_ino, dentry); 1746 1747 attr.ia_mode = mode; 1748 attr.ia_valid = ATTR_MODE; 1749 1750 trace_nfs_mknod_enter(dir, dentry); 1751 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1752 trace_nfs_mknod_exit(dir, dentry, status); 1753 if (status != 0) 1754 goto out_err; 1755 return 0; 1756 out_err: 1757 d_drop(dentry); 1758 return status; 1759 } 1760 EXPORT_SYMBOL_GPL(nfs_mknod); 1761 1762 /* 1763 * See comments for nfs_proc_create regarding failed operations. 1764 */ 1765 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 1766 { 1767 struct iattr attr; 1768 int error; 1769 1770 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 1771 dir->i_sb->s_id, dir->i_ino, dentry); 1772 1773 attr.ia_valid = ATTR_MODE; 1774 attr.ia_mode = mode | S_IFDIR; 1775 1776 trace_nfs_mkdir_enter(dir, dentry); 1777 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1778 trace_nfs_mkdir_exit(dir, dentry, error); 1779 if (error != 0) 1780 goto out_err; 1781 return 0; 1782 out_err: 1783 d_drop(dentry); 1784 return error; 1785 } 1786 EXPORT_SYMBOL_GPL(nfs_mkdir); 1787 1788 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1789 { 1790 if (simple_positive(dentry)) 1791 d_delete(dentry); 1792 } 1793 1794 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1795 { 1796 int error; 1797 1798 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 1799 dir->i_sb->s_id, dir->i_ino, dentry); 1800 1801 trace_nfs_rmdir_enter(dir, dentry); 1802 if (d_really_is_positive(dentry)) { 1803 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 1804 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1805 /* Ensure the VFS deletes this inode */ 1806 switch (error) { 1807 case 0: 1808 clear_nlink(d_inode(dentry)); 1809 break; 1810 case -ENOENT: 1811 nfs_dentry_handle_enoent(dentry); 1812 } 1813 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 1814 } else 1815 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1816 trace_nfs_rmdir_exit(dir, dentry, error); 1817 1818 return error; 1819 } 1820 EXPORT_SYMBOL_GPL(nfs_rmdir); 1821 1822 /* 1823 * Remove a file after making sure there are no pending writes, 1824 * and after checking that the file has only one user. 1825 * 1826 * We invalidate the attribute cache and free the inode prior to the operation 1827 * to avoid possible races if the server reuses the inode. 1828 */ 1829 static int nfs_safe_remove(struct dentry *dentry) 1830 { 1831 struct inode *dir = d_inode(dentry->d_parent); 1832 struct inode *inode = d_inode(dentry); 1833 int error = -EBUSY; 1834 1835 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 1836 1837 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1838 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1839 error = 0; 1840 goto out; 1841 } 1842 1843 trace_nfs_remove_enter(dir, dentry); 1844 if (inode != NULL) { 1845 NFS_PROTO(inode)->return_delegation(inode); 1846 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1847 if (error == 0) 1848 nfs_drop_nlink(inode); 1849 } else 1850 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1851 if (error == -ENOENT) 1852 nfs_dentry_handle_enoent(dentry); 1853 trace_nfs_remove_exit(dir, dentry, error); 1854 out: 1855 return error; 1856 } 1857 1858 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1859 * belongs to an active ".nfs..." file and we return -EBUSY. 1860 * 1861 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1862 */ 1863 int nfs_unlink(struct inode *dir, struct dentry *dentry) 1864 { 1865 int error; 1866 int need_rehash = 0; 1867 1868 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 1869 dir->i_ino, dentry); 1870 1871 trace_nfs_unlink_enter(dir, dentry); 1872 spin_lock(&dentry->d_lock); 1873 if (d_count(dentry) > 1) { 1874 spin_unlock(&dentry->d_lock); 1875 /* Start asynchronous writeout of the inode */ 1876 write_inode_now(d_inode(dentry), 0); 1877 error = nfs_sillyrename(dir, dentry); 1878 goto out; 1879 } 1880 if (!d_unhashed(dentry)) { 1881 __d_drop(dentry); 1882 need_rehash = 1; 1883 } 1884 spin_unlock(&dentry->d_lock); 1885 error = nfs_safe_remove(dentry); 1886 if (!error || error == -ENOENT) { 1887 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1888 } else if (need_rehash) 1889 d_rehash(dentry); 1890 out: 1891 trace_nfs_unlink_exit(dir, dentry, error); 1892 return error; 1893 } 1894 EXPORT_SYMBOL_GPL(nfs_unlink); 1895 1896 /* 1897 * To create a symbolic link, most file systems instantiate a new inode, 1898 * add a page to it containing the path, then write it out to the disk 1899 * using prepare_write/commit_write. 1900 * 1901 * Unfortunately the NFS client can't create the in-core inode first 1902 * because it needs a file handle to create an in-core inode (see 1903 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1904 * symlink request has completed on the server. 1905 * 1906 * So instead we allocate a raw page, copy the symname into it, then do 1907 * the SYMLINK request with the page as the buffer. If it succeeds, we 1908 * now have a new file handle and can instantiate an in-core NFS inode 1909 * and move the raw page into its mapping. 1910 */ 1911 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1912 { 1913 struct page *page; 1914 char *kaddr; 1915 struct iattr attr; 1916 unsigned int pathlen = strlen(symname); 1917 int error; 1918 1919 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 1920 dir->i_ino, dentry, symname); 1921 1922 if (pathlen > PAGE_SIZE) 1923 return -ENAMETOOLONG; 1924 1925 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1926 attr.ia_valid = ATTR_MODE; 1927 1928 page = alloc_page(GFP_USER); 1929 if (!page) 1930 return -ENOMEM; 1931 1932 kaddr = page_address(page); 1933 memcpy(kaddr, symname, pathlen); 1934 if (pathlen < PAGE_SIZE) 1935 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1936 1937 trace_nfs_symlink_enter(dir, dentry); 1938 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1939 trace_nfs_symlink_exit(dir, dentry, error); 1940 if (error != 0) { 1941 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 1942 dir->i_sb->s_id, dir->i_ino, 1943 dentry, symname, error); 1944 d_drop(dentry); 1945 __free_page(page); 1946 return error; 1947 } 1948 1949 /* 1950 * No big deal if we can't add this page to the page cache here. 1951 * READLINK will get the missing page from the server if needed. 1952 */ 1953 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 1954 GFP_KERNEL)) { 1955 SetPageUptodate(page); 1956 unlock_page(page); 1957 /* 1958 * add_to_page_cache_lru() grabs an extra page refcount. 1959 * Drop it here to avoid leaking this page later. 1960 */ 1961 put_page(page); 1962 } else 1963 __free_page(page); 1964 1965 return 0; 1966 } 1967 EXPORT_SYMBOL_GPL(nfs_symlink); 1968 1969 int 1970 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1971 { 1972 struct inode *inode = d_inode(old_dentry); 1973 int error; 1974 1975 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 1976 old_dentry, dentry); 1977 1978 trace_nfs_link_enter(inode, dir, dentry); 1979 NFS_PROTO(inode)->return_delegation(inode); 1980 1981 d_drop(dentry); 1982 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1983 if (error == 0) { 1984 ihold(inode); 1985 d_add(dentry, inode); 1986 } 1987 trace_nfs_link_exit(inode, dir, dentry, error); 1988 return error; 1989 } 1990 EXPORT_SYMBOL_GPL(nfs_link); 1991 1992 static void 1993 nfs_complete_rename(struct rpc_task *task, struct nfs_renamedata *data) 1994 { 1995 struct dentry *old_dentry = data->old_dentry; 1996 struct dentry *new_dentry = data->new_dentry; 1997 struct inode *old_inode = d_inode(old_dentry); 1998 struct inode *new_inode = d_inode(new_dentry); 1999 2000 nfs_mark_for_revalidate(old_inode); 2001 2002 switch (task->tk_status) { 2003 case 0: 2004 if (new_inode != NULL) 2005 nfs_drop_nlink(new_inode); 2006 d_move(old_dentry, new_dentry); 2007 nfs_set_verifier(new_dentry, 2008 nfs_save_change_attribute(data->new_dir)); 2009 break; 2010 case -ENOENT: 2011 nfs_dentry_handle_enoent(old_dentry); 2012 } 2013 } 2014 2015 /* 2016 * RENAME 2017 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2018 * different file handle for the same inode after a rename (e.g. when 2019 * moving to a different directory). A fail-safe method to do so would 2020 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2021 * rename the old file using the sillyrename stuff. This way, the original 2022 * file in old_dir will go away when the last process iput()s the inode. 2023 * 2024 * FIXED. 2025 * 2026 * It actually works quite well. One needs to have the possibility for 2027 * at least one ".nfs..." file in each directory the file ever gets 2028 * moved or linked to which happens automagically with the new 2029 * implementation that only depends on the dcache stuff instead of 2030 * using the inode layer 2031 * 2032 * Unfortunately, things are a little more complicated than indicated 2033 * above. For a cross-directory move, we want to make sure we can get 2034 * rid of the old inode after the operation. This means there must be 2035 * no pending writes (if it's a file), and the use count must be 1. 2036 * If these conditions are met, we can drop the dentries before doing 2037 * the rename. 2038 */ 2039 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 2040 struct inode *new_dir, struct dentry *new_dentry, 2041 unsigned int flags) 2042 { 2043 struct inode *old_inode = d_inode(old_dentry); 2044 struct inode *new_inode = d_inode(new_dentry); 2045 struct dentry *dentry = NULL; 2046 struct rpc_task *task; 2047 int error = -EBUSY; 2048 2049 if (flags) 2050 return -EINVAL; 2051 2052 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2053 old_dentry, new_dentry, 2054 d_count(new_dentry)); 2055 2056 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2057 /* 2058 * For non-directories, check whether the target is busy and if so, 2059 * make a copy of the dentry and then do a silly-rename. If the 2060 * silly-rename succeeds, the copied dentry is hashed and becomes 2061 * the new target. 2062 */ 2063 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2064 /* 2065 * To prevent any new references to the target during the 2066 * rename, we unhash the dentry in advance. 2067 */ 2068 if (!d_unhashed(new_dentry)) 2069 d_drop(new_dentry); 2070 2071 if (d_count(new_dentry) > 2) { 2072 int err; 2073 2074 /* copy the target dentry's name */ 2075 dentry = d_alloc(new_dentry->d_parent, 2076 &new_dentry->d_name); 2077 if (!dentry) 2078 goto out; 2079 2080 /* silly-rename the existing target ... */ 2081 err = nfs_sillyrename(new_dir, new_dentry); 2082 if (err) 2083 goto out; 2084 2085 new_dentry = dentry; 2086 new_inode = NULL; 2087 } 2088 } 2089 2090 NFS_PROTO(old_inode)->return_delegation(old_inode); 2091 if (new_inode != NULL) 2092 NFS_PROTO(new_inode)->return_delegation(new_inode); 2093 2094 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, 2095 nfs_complete_rename); 2096 if (IS_ERR(task)) { 2097 error = PTR_ERR(task); 2098 goto out; 2099 } 2100 2101 error = rpc_wait_for_completion_task(task); 2102 if (error == 0) 2103 error = task->tk_status; 2104 rpc_put_task(task); 2105 out: 2106 trace_nfs_rename_exit(old_dir, old_dentry, 2107 new_dir, new_dentry, error); 2108 /* new dentry created? */ 2109 if (dentry) 2110 dput(dentry); 2111 return error; 2112 } 2113 EXPORT_SYMBOL_GPL(nfs_rename); 2114 2115 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2116 static LIST_HEAD(nfs_access_lru_list); 2117 static atomic_long_t nfs_access_nr_entries; 2118 2119 static unsigned long nfs_access_max_cachesize = ULONG_MAX; 2120 module_param(nfs_access_max_cachesize, ulong, 0644); 2121 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2122 2123 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2124 { 2125 put_rpccred(entry->cred); 2126 kfree_rcu(entry, rcu_head); 2127 smp_mb__before_atomic(); 2128 atomic_long_dec(&nfs_access_nr_entries); 2129 smp_mb__after_atomic(); 2130 } 2131 2132 static void nfs_access_free_list(struct list_head *head) 2133 { 2134 struct nfs_access_entry *cache; 2135 2136 while (!list_empty(head)) { 2137 cache = list_entry(head->next, struct nfs_access_entry, lru); 2138 list_del(&cache->lru); 2139 nfs_access_free_entry(cache); 2140 } 2141 } 2142 2143 static unsigned long 2144 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2145 { 2146 LIST_HEAD(head); 2147 struct nfs_inode *nfsi, *next; 2148 struct nfs_access_entry *cache; 2149 long freed = 0; 2150 2151 spin_lock(&nfs_access_lru_lock); 2152 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2153 struct inode *inode; 2154 2155 if (nr_to_scan-- == 0) 2156 break; 2157 inode = &nfsi->vfs_inode; 2158 spin_lock(&inode->i_lock); 2159 if (list_empty(&nfsi->access_cache_entry_lru)) 2160 goto remove_lru_entry; 2161 cache = list_entry(nfsi->access_cache_entry_lru.next, 2162 struct nfs_access_entry, lru); 2163 list_move(&cache->lru, &head); 2164 rb_erase(&cache->rb_node, &nfsi->access_cache); 2165 freed++; 2166 if (!list_empty(&nfsi->access_cache_entry_lru)) 2167 list_move_tail(&nfsi->access_cache_inode_lru, 2168 &nfs_access_lru_list); 2169 else { 2170 remove_lru_entry: 2171 list_del_init(&nfsi->access_cache_inode_lru); 2172 smp_mb__before_atomic(); 2173 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2174 smp_mb__after_atomic(); 2175 } 2176 spin_unlock(&inode->i_lock); 2177 } 2178 spin_unlock(&nfs_access_lru_lock); 2179 nfs_access_free_list(&head); 2180 return freed; 2181 } 2182 2183 unsigned long 2184 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2185 { 2186 int nr_to_scan = sc->nr_to_scan; 2187 gfp_t gfp_mask = sc->gfp_mask; 2188 2189 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2190 return SHRINK_STOP; 2191 return nfs_do_access_cache_scan(nr_to_scan); 2192 } 2193 2194 2195 unsigned long 2196 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2197 { 2198 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2199 } 2200 2201 static void 2202 nfs_access_cache_enforce_limit(void) 2203 { 2204 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2205 unsigned long diff; 2206 unsigned int nr_to_scan; 2207 2208 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2209 return; 2210 nr_to_scan = 100; 2211 diff = nr_entries - nfs_access_max_cachesize; 2212 if (diff < nr_to_scan) 2213 nr_to_scan = diff; 2214 nfs_do_access_cache_scan(nr_to_scan); 2215 } 2216 2217 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2218 { 2219 struct rb_root *root_node = &nfsi->access_cache; 2220 struct rb_node *n; 2221 struct nfs_access_entry *entry; 2222 2223 /* Unhook entries from the cache */ 2224 while ((n = rb_first(root_node)) != NULL) { 2225 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2226 rb_erase(n, root_node); 2227 list_move(&entry->lru, head); 2228 } 2229 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2230 } 2231 2232 void nfs_access_zap_cache(struct inode *inode) 2233 { 2234 LIST_HEAD(head); 2235 2236 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2237 return; 2238 /* Remove from global LRU init */ 2239 spin_lock(&nfs_access_lru_lock); 2240 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2241 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2242 2243 spin_lock(&inode->i_lock); 2244 __nfs_access_zap_cache(NFS_I(inode), &head); 2245 spin_unlock(&inode->i_lock); 2246 spin_unlock(&nfs_access_lru_lock); 2247 nfs_access_free_list(&head); 2248 } 2249 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2250 2251 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 2252 { 2253 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2254 struct nfs_access_entry *entry; 2255 2256 while (n != NULL) { 2257 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2258 2259 if (cred < entry->cred) 2260 n = n->rb_left; 2261 else if (cred > entry->cred) 2262 n = n->rb_right; 2263 else 2264 return entry; 2265 } 2266 return NULL; 2267 } 2268 2269 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block) 2270 { 2271 struct nfs_inode *nfsi = NFS_I(inode); 2272 struct nfs_access_entry *cache; 2273 bool retry = true; 2274 int err; 2275 2276 spin_lock(&inode->i_lock); 2277 for(;;) { 2278 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2279 goto out_zap; 2280 cache = nfs_access_search_rbtree(inode, cred); 2281 err = -ENOENT; 2282 if (cache == NULL) 2283 goto out; 2284 /* Found an entry, is our attribute cache valid? */ 2285 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2286 break; 2287 err = -ECHILD; 2288 if (!may_block) 2289 goto out; 2290 if (!retry) 2291 goto out_zap; 2292 spin_unlock(&inode->i_lock); 2293 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2294 if (err) 2295 return err; 2296 spin_lock(&inode->i_lock); 2297 retry = false; 2298 } 2299 res->jiffies = cache->jiffies; 2300 res->cred = cache->cred; 2301 res->mask = cache->mask; 2302 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2303 err = 0; 2304 out: 2305 spin_unlock(&inode->i_lock); 2306 return err; 2307 out_zap: 2308 spin_unlock(&inode->i_lock); 2309 nfs_access_zap_cache(inode); 2310 return -ENOENT; 2311 } 2312 2313 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2314 { 2315 /* Only check the most recently returned cache entry, 2316 * but do it without locking. 2317 */ 2318 struct nfs_inode *nfsi = NFS_I(inode); 2319 struct nfs_access_entry *cache; 2320 int err = -ECHILD; 2321 struct list_head *lh; 2322 2323 rcu_read_lock(); 2324 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2325 goto out; 2326 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); 2327 cache = list_entry(lh, struct nfs_access_entry, lru); 2328 if (lh == &nfsi->access_cache_entry_lru || 2329 cred != cache->cred) 2330 cache = NULL; 2331 if (cache == NULL) 2332 goto out; 2333 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2334 goto out; 2335 res->jiffies = cache->jiffies; 2336 res->cred = cache->cred; 2337 res->mask = cache->mask; 2338 err = 0; 2339 out: 2340 rcu_read_unlock(); 2341 return err; 2342 } 2343 2344 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2345 { 2346 struct nfs_inode *nfsi = NFS_I(inode); 2347 struct rb_root *root_node = &nfsi->access_cache; 2348 struct rb_node **p = &root_node->rb_node; 2349 struct rb_node *parent = NULL; 2350 struct nfs_access_entry *entry; 2351 2352 spin_lock(&inode->i_lock); 2353 while (*p != NULL) { 2354 parent = *p; 2355 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2356 2357 if (set->cred < entry->cred) 2358 p = &parent->rb_left; 2359 else if (set->cred > entry->cred) 2360 p = &parent->rb_right; 2361 else 2362 goto found; 2363 } 2364 rb_link_node(&set->rb_node, parent, p); 2365 rb_insert_color(&set->rb_node, root_node); 2366 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2367 spin_unlock(&inode->i_lock); 2368 return; 2369 found: 2370 rb_replace_node(parent, &set->rb_node, root_node); 2371 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2372 list_del(&entry->lru); 2373 spin_unlock(&inode->i_lock); 2374 nfs_access_free_entry(entry); 2375 } 2376 2377 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2378 { 2379 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2380 if (cache == NULL) 2381 return; 2382 RB_CLEAR_NODE(&cache->rb_node); 2383 cache->jiffies = set->jiffies; 2384 cache->cred = get_rpccred(set->cred); 2385 cache->mask = set->mask; 2386 2387 /* The above field assignments must be visible 2388 * before this item appears on the lru. We cannot easily 2389 * use rcu_assign_pointer, so just force the memory barrier. 2390 */ 2391 smp_wmb(); 2392 nfs_access_add_rbtree(inode, cache); 2393 2394 /* Update accounting */ 2395 smp_mb__before_atomic(); 2396 atomic_long_inc(&nfs_access_nr_entries); 2397 smp_mb__after_atomic(); 2398 2399 /* Add inode to global LRU list */ 2400 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2401 spin_lock(&nfs_access_lru_lock); 2402 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2403 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2404 &nfs_access_lru_list); 2405 spin_unlock(&nfs_access_lru_lock); 2406 } 2407 nfs_access_cache_enforce_limit(); 2408 } 2409 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2410 2411 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2412 { 2413 entry->mask = 0; 2414 if (access_result & NFS4_ACCESS_READ) 2415 entry->mask |= MAY_READ; 2416 if (access_result & 2417 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) 2418 entry->mask |= MAY_WRITE; 2419 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) 2420 entry->mask |= MAY_EXEC; 2421 } 2422 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2423 2424 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 2425 { 2426 struct nfs_access_entry cache; 2427 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2428 int status; 2429 2430 trace_nfs_access_enter(inode); 2431 2432 status = nfs_access_get_cached_rcu(inode, cred, &cache); 2433 if (status != 0) 2434 status = nfs_access_get_cached(inode, cred, &cache, may_block); 2435 if (status == 0) 2436 goto out_cached; 2437 2438 status = -ECHILD; 2439 if (!may_block) 2440 goto out; 2441 2442 /* Be clever: ask server to check for all possible rights */ 2443 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 2444 cache.cred = cred; 2445 cache.jiffies = jiffies; 2446 status = NFS_PROTO(inode)->access(inode, &cache); 2447 if (status != 0) { 2448 if (status == -ESTALE) { 2449 nfs_zap_caches(inode); 2450 if (!S_ISDIR(inode->i_mode)) 2451 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 2452 } 2453 goto out; 2454 } 2455 nfs_access_add_cache(inode, &cache); 2456 out_cached: 2457 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2458 status = -EACCES; 2459 out: 2460 trace_nfs_access_exit(inode, status); 2461 return status; 2462 } 2463 2464 static int nfs_open_permission_mask(int openflags) 2465 { 2466 int mask = 0; 2467 2468 if (openflags & __FMODE_EXEC) { 2469 /* ONLY check exec rights */ 2470 mask = MAY_EXEC; 2471 } else { 2472 if ((openflags & O_ACCMODE) != O_WRONLY) 2473 mask |= MAY_READ; 2474 if ((openflags & O_ACCMODE) != O_RDONLY) 2475 mask |= MAY_WRITE; 2476 } 2477 2478 return mask; 2479 } 2480 2481 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 2482 { 2483 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2484 } 2485 EXPORT_SYMBOL_GPL(nfs_may_open); 2486 2487 static int nfs_execute_ok(struct inode *inode, int mask) 2488 { 2489 struct nfs_server *server = NFS_SERVER(inode); 2490 int ret = 0; 2491 2492 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) { 2493 if (mask & MAY_NOT_BLOCK) 2494 return -ECHILD; 2495 ret = __nfs_revalidate_inode(server, inode); 2496 } 2497 if (ret == 0 && !execute_ok(inode)) 2498 ret = -EACCES; 2499 return ret; 2500 } 2501 2502 int nfs_permission(struct inode *inode, int mask) 2503 { 2504 struct rpc_cred *cred; 2505 int res = 0; 2506 2507 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2508 2509 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2510 goto out; 2511 /* Is this sys_access() ? */ 2512 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2513 goto force_lookup; 2514 2515 switch (inode->i_mode & S_IFMT) { 2516 case S_IFLNK: 2517 goto out; 2518 case S_IFREG: 2519 if ((mask & MAY_OPEN) && 2520 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 2521 return 0; 2522 break; 2523 case S_IFDIR: 2524 /* 2525 * Optimize away all write operations, since the server 2526 * will check permissions when we perform the op. 2527 */ 2528 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2529 goto out; 2530 } 2531 2532 force_lookup: 2533 if (!NFS_PROTO(inode)->access) 2534 goto out_notsup; 2535 2536 /* Always try fast lookups first */ 2537 rcu_read_lock(); 2538 cred = rpc_lookup_cred_nonblock(); 2539 if (!IS_ERR(cred)) 2540 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); 2541 else 2542 res = PTR_ERR(cred); 2543 rcu_read_unlock(); 2544 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { 2545 /* Fast lookup failed, try the slow way */ 2546 cred = rpc_lookup_cred(); 2547 if (!IS_ERR(cred)) { 2548 res = nfs_do_access(inode, cred, mask); 2549 put_rpccred(cred); 2550 } else 2551 res = PTR_ERR(cred); 2552 } 2553 out: 2554 if (!res && (mask & MAY_EXEC)) 2555 res = nfs_execute_ok(inode, mask); 2556 2557 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2558 inode->i_sb->s_id, inode->i_ino, mask, res); 2559 return res; 2560 out_notsup: 2561 if (mask & MAY_NOT_BLOCK) 2562 return -ECHILD; 2563 2564 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2565 if (res == 0) 2566 res = generic_permission(inode, mask); 2567 goto out; 2568 } 2569 EXPORT_SYMBOL_GPL(nfs_permission); 2570 2571 /* 2572 * Local variables: 2573 * version-control: t 2574 * kept-new-versions: 5 2575 * End: 2576 */ 2577