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_shared = 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 atomic_t refcount; 149 int size; 150 int eof_index; 151 u64 last_cookie; 152 struct nfs_cache_array_entry array[0]; 153 }; 154 155 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int); 156 typedef struct { 157 struct file *file; 158 struct page *page; 159 struct dir_context *ctx; 160 unsigned long page_index; 161 u64 *dir_cookie; 162 u64 last_cookie; 163 loff_t current_index; 164 decode_dirent_t decode; 165 166 unsigned long timestamp; 167 unsigned long gencount; 168 unsigned int cache_entry_index; 169 unsigned int plus:1; 170 unsigned int eof:1; 171 } nfs_readdir_descriptor_t; 172 173 /* 174 * The caller is responsible for calling nfs_readdir_release_array(page) 175 */ 176 static 177 struct nfs_cache_array *nfs_readdir_get_array(struct page *page) 178 { 179 void *ptr; 180 if (page == NULL) 181 return ERR_PTR(-EIO); 182 ptr = kmap(page); 183 if (ptr == NULL) 184 return ERR_PTR(-ENOMEM); 185 return ptr; 186 } 187 188 static 189 void nfs_readdir_release_array(struct page *page) 190 { 191 kunmap(page); 192 } 193 194 /* 195 * we are freeing strings created by nfs_add_to_readdir_array() 196 */ 197 static 198 void nfs_readdir_clear_array(struct page *page) 199 { 200 struct nfs_cache_array *array; 201 int i; 202 203 array = kmap_atomic(page); 204 if (atomic_dec_and_test(&array->refcount)) 205 for (i = 0; i < array->size; i++) 206 kfree(array->array[i].string.name); 207 kunmap_atomic(array); 208 } 209 210 static bool grab_page(struct page *page) 211 { 212 struct nfs_cache_array *array = kmap_atomic(page); 213 bool res = atomic_inc_not_zero(&array->refcount); 214 kunmap_atomic(array); 215 return res; 216 } 217 218 /* 219 * the caller is responsible for freeing qstr.name 220 * when called by nfs_readdir_add_to_array, the strings will be freed in 221 * nfs_clear_readdir_array() 222 */ 223 static 224 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) 225 { 226 string->len = len; 227 string->name = kmemdup(name, len, GFP_KERNEL); 228 if (string->name == NULL) 229 return -ENOMEM; 230 /* 231 * Avoid a kmemleak false positive. The pointer to the name is stored 232 * in a page cache page which kmemleak does not scan. 233 */ 234 kmemleak_not_leak(string->name); 235 string->hash = full_name_hash(name, len); 236 return 0; 237 } 238 239 static 240 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 241 { 242 struct nfs_cache_array *array = nfs_readdir_get_array(page); 243 struct nfs_cache_array_entry *cache_entry; 244 int ret; 245 246 if (IS_ERR(array)) 247 return PTR_ERR(array); 248 249 cache_entry = &array->array[array->size]; 250 251 /* Check that this entry lies within the page bounds */ 252 ret = -ENOSPC; 253 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) 254 goto out; 255 256 cache_entry->cookie = entry->prev_cookie; 257 cache_entry->ino = entry->ino; 258 cache_entry->d_type = entry->d_type; 259 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); 260 if (ret) 261 goto out; 262 array->last_cookie = entry->cookie; 263 array->size++; 264 if (entry->eof != 0) 265 array->eof_index = array->size; 266 out: 267 nfs_readdir_release_array(page); 268 return ret; 269 } 270 271 static 272 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 273 { 274 loff_t diff = desc->ctx->pos - desc->current_index; 275 unsigned int index; 276 277 if (diff < 0) 278 goto out_eof; 279 if (diff >= array->size) { 280 if (array->eof_index >= 0) 281 goto out_eof; 282 return -EAGAIN; 283 } 284 285 index = (unsigned int)diff; 286 *desc->dir_cookie = array->array[index].cookie; 287 desc->cache_entry_index = index; 288 return 0; 289 out_eof: 290 desc->eof = 1; 291 return -EBADCOOKIE; 292 } 293 294 static bool 295 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) 296 { 297 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 298 return false; 299 smp_rmb(); 300 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); 301 } 302 303 static 304 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 305 { 306 int i; 307 loff_t new_pos; 308 int status = -EAGAIN; 309 310 for (i = 0; i < array->size; i++) { 311 if (array->array[i].cookie == *desc->dir_cookie) { 312 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); 313 struct nfs_open_dir_context *ctx = desc->file->private_data; 314 315 new_pos = desc->current_index + i; 316 if (ctx->attr_gencount != nfsi->attr_gencount || 317 !nfs_readdir_inode_mapping_valid(nfsi)) { 318 ctx->duped = 0; 319 ctx->attr_gencount = nfsi->attr_gencount; 320 } else if (new_pos < desc->ctx->pos) { 321 if (ctx->duped > 0 322 && ctx->dup_cookie == *desc->dir_cookie) { 323 if (printk_ratelimit()) { 324 pr_notice("NFS: directory %pD2 contains a readdir loop." 325 "Please contact your server vendor. " 326 "The file: %.*s has duplicate cookie %llu\n", 327 desc->file, array->array[i].string.len, 328 array->array[i].string.name, *desc->dir_cookie); 329 } 330 status = -ELOOP; 331 goto out; 332 } 333 ctx->dup_cookie = *desc->dir_cookie; 334 ctx->duped = -1; 335 } 336 desc->ctx->pos = new_pos; 337 desc->cache_entry_index = i; 338 return 0; 339 } 340 } 341 if (array->eof_index >= 0) { 342 status = -EBADCOOKIE; 343 if (*desc->dir_cookie == array->last_cookie) 344 desc->eof = 1; 345 } 346 out: 347 return status; 348 } 349 350 static 351 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) 352 { 353 struct nfs_cache_array *array; 354 int status; 355 356 array = nfs_readdir_get_array(desc->page); 357 if (IS_ERR(array)) { 358 status = PTR_ERR(array); 359 goto out; 360 } 361 362 if (*desc->dir_cookie == 0) 363 status = nfs_readdir_search_for_pos(array, desc); 364 else 365 status = nfs_readdir_search_for_cookie(array, desc); 366 367 if (status == -EAGAIN) { 368 desc->last_cookie = array->last_cookie; 369 desc->current_index += array->size; 370 desc->page_index++; 371 } 372 nfs_readdir_release_array(desc->page); 373 out: 374 return status; 375 } 376 377 /* Fill a page with xdr information before transferring to the cache page */ 378 static 379 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, 380 struct nfs_entry *entry, struct file *file, struct inode *inode) 381 { 382 struct nfs_open_dir_context *ctx = file->private_data; 383 struct rpc_cred *cred = ctx->cred; 384 unsigned long timestamp, gencount; 385 int error; 386 387 again: 388 timestamp = jiffies; 389 gencount = nfs_inc_attr_generation_counter(); 390 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages, 391 NFS_SERVER(inode)->dtsize, desc->plus); 392 if (error < 0) { 393 /* We requested READDIRPLUS, but the server doesn't grok it */ 394 if (error == -ENOTSUPP && desc->plus) { 395 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 396 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 397 desc->plus = 0; 398 goto again; 399 } 400 goto error; 401 } 402 desc->timestamp = timestamp; 403 desc->gencount = gencount; 404 error: 405 return error; 406 } 407 408 static int xdr_decode(nfs_readdir_descriptor_t *desc, 409 struct nfs_entry *entry, struct xdr_stream *xdr) 410 { 411 int error; 412 413 error = desc->decode(xdr, entry, desc->plus); 414 if (error) 415 return error; 416 entry->fattr->time_start = desc->timestamp; 417 entry->fattr->gencount = desc->gencount; 418 return 0; 419 } 420 421 /* Match file and dirent using either filehandle or fileid 422 * Note: caller is responsible for checking the fsid 423 */ 424 static 425 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 426 { 427 struct inode *inode; 428 struct nfs_inode *nfsi; 429 430 if (d_really_is_negative(dentry)) 431 return 0; 432 433 inode = d_inode(dentry); 434 if (is_bad_inode(inode) || NFS_STALE(inode)) 435 return 0; 436 437 nfsi = NFS_I(inode); 438 if (entry->fattr->fileid == nfsi->fileid) 439 return 1; 440 if (nfs_compare_fh(entry->fh, &nfsi->fh) == 0) 441 return 1; 442 return 0; 443 } 444 445 static 446 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) 447 { 448 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 449 return false; 450 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 451 return true; 452 if (ctx->pos == 0) 453 return true; 454 return false; 455 } 456 457 /* 458 * This function is called by the lookup code to request the use of 459 * readdirplus to accelerate any future lookups in the same 460 * directory. 461 */ 462 static 463 void nfs_advise_use_readdirplus(struct inode *dir) 464 { 465 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags); 466 } 467 468 /* 469 * This function is mainly for use by nfs_getattr(). 470 * 471 * If this is an 'ls -l', we want to force use of readdirplus. 472 * Do this by checking if there is an active file descriptor 473 * and calling nfs_advise_use_readdirplus, then forcing a 474 * cache flush. 475 */ 476 void nfs_force_use_readdirplus(struct inode *dir) 477 { 478 if (!list_empty(&NFS_I(dir)->open_files)) { 479 nfs_advise_use_readdirplus(dir); 480 nfs_zap_mapping(dir, dir->i_mapping); 481 } 482 } 483 484 static 485 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) 486 { 487 struct qstr filename = QSTR_INIT(entry->name, entry->len); 488 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 489 struct dentry *dentry; 490 struct dentry *alias; 491 struct inode *dir = d_inode(parent); 492 struct inode *inode; 493 int status; 494 495 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 496 return; 497 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 498 return; 499 if (filename.name[0] == '.') { 500 if (filename.len == 1) 501 return; 502 if (filename.len == 2 && filename.name[1] == '.') 503 return; 504 } 505 filename.hash = full_name_hash(filename.name, filename.len); 506 507 dentry = d_lookup(parent, &filename); 508 again: 509 if (!dentry) { 510 dentry = d_alloc_parallel(parent, &filename, &wq); 511 if (IS_ERR(dentry)) 512 return; 513 } 514 if (!d_in_lookup(dentry)) { 515 /* Is there a mountpoint here? If so, just exit */ 516 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 517 &entry->fattr->fsid)) 518 goto out; 519 if (nfs_same_file(dentry, entry)) { 520 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 521 status = nfs_refresh_inode(d_inode(dentry), entry->fattr); 522 if (!status) 523 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); 524 goto out; 525 } else { 526 d_invalidate(dentry); 527 dput(dentry); 528 dentry = NULL; 529 goto again; 530 } 531 } 532 533 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 534 alias = d_splice_alias(inode, dentry); 535 d_lookup_done(dentry); 536 if (alias) { 537 if (IS_ERR(alias)) 538 goto out; 539 dput(dentry); 540 dentry = alias; 541 } 542 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 543 out: 544 dput(dentry); 545 } 546 547 /* Perform conversion from xdr to cache array */ 548 static 549 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 550 struct page **xdr_pages, struct page *page, unsigned int buflen) 551 { 552 struct xdr_stream stream; 553 struct xdr_buf buf; 554 struct page *scratch; 555 struct nfs_cache_array *array; 556 unsigned int count = 0; 557 int status; 558 559 scratch = alloc_page(GFP_KERNEL); 560 if (scratch == NULL) 561 return -ENOMEM; 562 563 if (buflen == 0) 564 goto out_nopages; 565 566 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 567 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 568 569 do { 570 status = xdr_decode(desc, entry, &stream); 571 if (status != 0) { 572 if (status == -EAGAIN) 573 status = 0; 574 break; 575 } 576 577 count++; 578 579 if (desc->plus != 0) 580 nfs_prime_dcache(file_dentry(desc->file), entry); 581 582 status = nfs_readdir_add_to_array(entry, page); 583 if (status != 0) 584 break; 585 } while (!entry->eof); 586 587 out_nopages: 588 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { 589 array = nfs_readdir_get_array(page); 590 if (!IS_ERR(array)) { 591 array->eof_index = array->size; 592 status = 0; 593 nfs_readdir_release_array(page); 594 } else 595 status = PTR_ERR(array); 596 } 597 598 put_page(scratch); 599 return status; 600 } 601 602 static 603 void nfs_readdir_free_pages(struct page **pages, unsigned int npages) 604 { 605 unsigned int i; 606 for (i = 0; i < npages; i++) 607 put_page(pages[i]); 608 } 609 610 /* 611 * nfs_readdir_large_page will allocate pages that must be freed with a call 612 * to nfs_readdir_free_pagearray 613 */ 614 static 615 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages) 616 { 617 unsigned int i; 618 619 for (i = 0; i < npages; i++) { 620 struct page *page = alloc_page(GFP_KERNEL); 621 if (page == NULL) 622 goto out_freepages; 623 pages[i] = page; 624 } 625 return 0; 626 627 out_freepages: 628 nfs_readdir_free_pages(pages, i); 629 return -ENOMEM; 630 } 631 632 static 633 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) 634 { 635 struct page *pages[NFS_MAX_READDIR_PAGES]; 636 struct nfs_entry entry; 637 struct file *file = desc->file; 638 struct nfs_cache_array *array; 639 int status = -ENOMEM; 640 unsigned int array_size = ARRAY_SIZE(pages); 641 642 entry.prev_cookie = 0; 643 entry.cookie = desc->last_cookie; 644 entry.eof = 0; 645 entry.fh = nfs_alloc_fhandle(); 646 entry.fattr = nfs_alloc_fattr(); 647 entry.server = NFS_SERVER(inode); 648 if (entry.fh == NULL || entry.fattr == NULL) 649 goto out; 650 651 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 652 if (IS_ERR(entry.label)) { 653 status = PTR_ERR(entry.label); 654 goto out; 655 } 656 657 array = nfs_readdir_get_array(page); 658 if (IS_ERR(array)) { 659 status = PTR_ERR(array); 660 goto out_label_free; 661 } 662 memset(array, 0, sizeof(struct nfs_cache_array)); 663 atomic_set(&array->refcount, 1); 664 array->eof_index = -1; 665 666 status = nfs_readdir_alloc_pages(pages, array_size); 667 if (status < 0) 668 goto out_release_array; 669 do { 670 unsigned int pglen; 671 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 672 673 if (status < 0) 674 break; 675 pglen = status; 676 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 677 if (status < 0) { 678 if (status == -ENOSPC) 679 status = 0; 680 break; 681 } 682 } while (array->eof_index < 0); 683 684 nfs_readdir_free_pages(pages, array_size); 685 out_release_array: 686 nfs_readdir_release_array(page); 687 out_label_free: 688 nfs4_label_free(entry.label); 689 out: 690 nfs_free_fattr(entry.fattr); 691 nfs_free_fhandle(entry.fh); 692 return status; 693 } 694 695 /* 696 * Now we cache directories properly, by converting xdr information 697 * to an array that can be used for lookups later. This results in 698 * fewer cache pages, since we can store more information on each page. 699 * We only need to convert from xdr once so future lookups are much simpler 700 */ 701 static 702 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) 703 { 704 struct inode *inode = file_inode(desc->file); 705 int ret; 706 707 ret = nfs_readdir_xdr_to_array(desc, page, inode); 708 if (ret < 0) 709 goto error; 710 SetPageUptodate(page); 711 712 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 713 /* Should never happen */ 714 nfs_zap_mapping(inode, inode->i_mapping); 715 } 716 unlock_page(page); 717 return 0; 718 error: 719 unlock_page(page); 720 return ret; 721 } 722 723 static 724 void cache_page_release(nfs_readdir_descriptor_t *desc) 725 { 726 nfs_readdir_clear_array(desc->page); 727 put_page(desc->page); 728 desc->page = NULL; 729 } 730 731 static 732 struct page *get_cache_page(nfs_readdir_descriptor_t *desc) 733 { 734 struct page *page; 735 736 for (;;) { 737 page = read_cache_page(file_inode(desc->file)->i_mapping, 738 desc->page_index, (filler_t *)nfs_readdir_filler, desc); 739 if (IS_ERR(page) || grab_page(page)) 740 break; 741 put_page(page); 742 } 743 return page; 744 } 745 746 /* 747 * Returns 0 if desc->dir_cookie was found on page desc->page_index 748 */ 749 static 750 int find_cache_page(nfs_readdir_descriptor_t *desc) 751 { 752 int res; 753 754 desc->page = get_cache_page(desc); 755 if (IS_ERR(desc->page)) 756 return PTR_ERR(desc->page); 757 758 res = nfs_readdir_search_array(desc); 759 if (res != 0) 760 cache_page_release(desc); 761 return res; 762 } 763 764 /* Search for desc->dir_cookie from the beginning of the page cache */ 765 static inline 766 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 767 { 768 int res; 769 770 if (desc->page_index == 0) { 771 desc->current_index = 0; 772 desc->last_cookie = 0; 773 } 774 do { 775 res = find_cache_page(desc); 776 } while (res == -EAGAIN); 777 return res; 778 } 779 780 /* 781 * Once we've found the start of the dirent within a page: fill 'er up... 782 */ 783 static 784 int nfs_do_filldir(nfs_readdir_descriptor_t *desc) 785 { 786 struct file *file = desc->file; 787 int i = 0; 788 int res = 0; 789 struct nfs_cache_array *array = NULL; 790 struct nfs_open_dir_context *ctx = file->private_data; 791 792 array = nfs_readdir_get_array(desc->page); 793 if (IS_ERR(array)) { 794 res = PTR_ERR(array); 795 goto out; 796 } 797 798 for (i = desc->cache_entry_index; i < array->size; i++) { 799 struct nfs_cache_array_entry *ent; 800 801 ent = &array->array[i]; 802 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, 803 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 804 desc->eof = 1; 805 break; 806 } 807 desc->ctx->pos++; 808 if (i < (array->size-1)) 809 *desc->dir_cookie = array->array[i+1].cookie; 810 else 811 *desc->dir_cookie = array->last_cookie; 812 if (ctx->duped != 0) 813 ctx->duped = 1; 814 } 815 if (array->eof_index >= 0) 816 desc->eof = 1; 817 818 nfs_readdir_release_array(desc->page); 819 out: 820 cache_page_release(desc); 821 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 822 (unsigned long long)*desc->dir_cookie, res); 823 return res; 824 } 825 826 /* 827 * If we cannot find a cookie in our cache, we suspect that this is 828 * because it points to a deleted file, so we ask the server to return 829 * whatever it thinks is the next entry. We then feed this to filldir. 830 * If all goes well, we should then be able to find our way round the 831 * cache on the next call to readdir_search_pagecache(); 832 * 833 * NOTE: we cannot add the anonymous page to the pagecache because 834 * the data it contains might not be page aligned. Besides, 835 * we should already have a complete representation of the 836 * directory in the page cache by the time we get here. 837 */ 838 static inline 839 int uncached_readdir(nfs_readdir_descriptor_t *desc) 840 { 841 struct page *page = NULL; 842 int status; 843 struct inode *inode = file_inode(desc->file); 844 struct nfs_open_dir_context *ctx = desc->file->private_data; 845 846 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 847 (unsigned long long)*desc->dir_cookie); 848 849 page = alloc_page(GFP_HIGHUSER); 850 if (!page) { 851 status = -ENOMEM; 852 goto out; 853 } 854 855 desc->page_index = 0; 856 desc->last_cookie = *desc->dir_cookie; 857 desc->page = page; 858 ctx->duped = 0; 859 860 status = nfs_readdir_xdr_to_array(desc, page, inode); 861 if (status < 0) 862 goto out_release; 863 864 status = nfs_do_filldir(desc); 865 866 out: 867 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 868 __func__, status); 869 return status; 870 out_release: 871 cache_page_release(desc); 872 goto out; 873 } 874 875 static bool nfs_dir_mapping_need_revalidate(struct inode *dir) 876 { 877 struct nfs_inode *nfsi = NFS_I(dir); 878 879 if (nfs_attribute_cache_expired(dir)) 880 return true; 881 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 882 return true; 883 return false; 884 } 885 886 /* The file offset position represents the dirent entry number. A 887 last cookie cache takes care of the common case of reading the 888 whole directory. 889 */ 890 static int nfs_readdir(struct file *file, struct dir_context *ctx) 891 { 892 struct dentry *dentry = file_dentry(file); 893 struct inode *inode = d_inode(dentry); 894 nfs_readdir_descriptor_t my_desc, 895 *desc = &my_desc; 896 struct nfs_open_dir_context *dir_ctx = file->private_data; 897 int res = 0; 898 899 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 900 file, (long long)ctx->pos); 901 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 902 903 /* 904 * ctx->pos points to the dirent entry number. 905 * *desc->dir_cookie has the cookie for the next entry. We have 906 * to either find the entry with the appropriate number or 907 * revalidate the cookie. 908 */ 909 memset(desc, 0, sizeof(*desc)); 910 911 desc->file = file; 912 desc->ctx = ctx; 913 desc->dir_cookie = &dir_ctx->dir_cookie; 914 desc->decode = NFS_PROTO(inode)->decode_dirent; 915 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0; 916 917 if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode)) 918 res = nfs_revalidate_mapping(inode, file->f_mapping); 919 if (res < 0) 920 goto out; 921 922 do { 923 res = readdir_search_pagecache(desc); 924 925 if (res == -EBADCOOKIE) { 926 res = 0; 927 /* This means either end of directory */ 928 if (*desc->dir_cookie && desc->eof == 0) { 929 /* Or that the server has 'lost' a cookie */ 930 res = uncached_readdir(desc); 931 if (res == 0) 932 continue; 933 } 934 break; 935 } 936 if (res == -ETOOSMALL && desc->plus) { 937 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 938 nfs_zap_caches(inode); 939 desc->page_index = 0; 940 desc->plus = 0; 941 desc->eof = 0; 942 continue; 943 } 944 if (res < 0) 945 break; 946 947 res = nfs_do_filldir(desc); 948 if (res < 0) 949 break; 950 } while (!desc->eof); 951 out: 952 if (res > 0) 953 res = 0; 954 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 955 return res; 956 } 957 958 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 959 { 960 struct nfs_open_dir_context *dir_ctx = filp->private_data; 961 962 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 963 filp, offset, whence); 964 965 switch (whence) { 966 case 1: 967 offset += filp->f_pos; 968 case 0: 969 if (offset >= 0) 970 break; 971 default: 972 return -EINVAL; 973 } 974 if (offset != filp->f_pos) { 975 filp->f_pos = offset; 976 dir_ctx->dir_cookie = 0; 977 dir_ctx->duped = 0; 978 } 979 return offset; 980 } 981 982 /* 983 * All directory operations under NFS are synchronous, so fsync() 984 * is a dummy operation. 985 */ 986 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 987 int datasync) 988 { 989 struct inode *inode = file_inode(filp); 990 991 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 992 993 inode_lock(inode); 994 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 995 inode_unlock(inode); 996 return 0; 997 } 998 999 /** 1000 * nfs_force_lookup_revalidate - Mark the directory as having changed 1001 * @dir - pointer to directory inode 1002 * 1003 * This forces the revalidation code in nfs_lookup_revalidate() to do a 1004 * full lookup on all child dentries of 'dir' whenever a change occurs 1005 * on the server that might have invalidated our dcache. 1006 * 1007 * The caller should be holding dir->i_lock 1008 */ 1009 void nfs_force_lookup_revalidate(struct inode *dir) 1010 { 1011 NFS_I(dir)->cache_change_attribute++; 1012 } 1013 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1014 1015 /* 1016 * A check for whether or not the parent directory has changed. 1017 * In the case it has, we assume that the dentries are untrustworthy 1018 * and may need to be looked up again. 1019 * If rcu_walk prevents us from performing a full check, return 0. 1020 */ 1021 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1022 int rcu_walk) 1023 { 1024 int ret; 1025 1026 if (IS_ROOT(dentry)) 1027 return 1; 1028 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1029 return 0; 1030 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1031 return 0; 1032 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1033 if (rcu_walk) 1034 ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir); 1035 else 1036 ret = nfs_revalidate_inode(NFS_SERVER(dir), dir); 1037 if (ret < 0) 1038 return 0; 1039 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1040 return 0; 1041 return 1; 1042 } 1043 1044 /* 1045 * Use intent information to check whether or not we're going to do 1046 * an O_EXCL create using this path component. 1047 */ 1048 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1049 { 1050 if (NFS_PROTO(dir)->version == 2) 1051 return 0; 1052 return flags & LOOKUP_EXCL; 1053 } 1054 1055 /* 1056 * Inode and filehandle revalidation for lookups. 1057 * 1058 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1059 * or if the intent information indicates that we're about to open this 1060 * particular file and the "nocto" mount flag is not set. 1061 * 1062 */ 1063 static 1064 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1065 { 1066 struct nfs_server *server = NFS_SERVER(inode); 1067 int ret; 1068 1069 if (IS_AUTOMOUNT(inode)) 1070 return 0; 1071 /* VFS wants an on-the-wire revalidation */ 1072 if (flags & LOOKUP_REVAL) 1073 goto out_force; 1074 /* This is an open(2) */ 1075 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && 1076 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) 1077 goto out_force; 1078 out: 1079 return (inode->i_nlink == 0) ? -ENOENT : 0; 1080 out_force: 1081 if (flags & LOOKUP_RCU) 1082 return -ECHILD; 1083 ret = __nfs_revalidate_inode(server, inode); 1084 if (ret != 0) 1085 return ret; 1086 goto out; 1087 } 1088 1089 /* 1090 * We judge how long we want to trust negative 1091 * dentries by looking at the parent inode mtime. 1092 * 1093 * If parent mtime has changed, we revalidate, else we wait for a 1094 * period corresponding to the parent's attribute cache timeout value. 1095 * 1096 * If LOOKUP_RCU prevents us from performing a full check, return 1 1097 * suggesting a reval is needed. 1098 */ 1099 static inline 1100 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1101 unsigned int flags) 1102 { 1103 /* Don't revalidate a negative dentry if we're creating a new file */ 1104 if (flags & LOOKUP_CREATE) 1105 return 0; 1106 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1107 return 1; 1108 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1109 } 1110 1111 /* 1112 * This is called every time the dcache has a lookup hit, 1113 * and we should check whether we can really trust that 1114 * lookup. 1115 * 1116 * NOTE! The hit can be a negative hit too, don't assume 1117 * we have an inode! 1118 * 1119 * If the parent directory is seen to have changed, we throw out the 1120 * cached dentry and do a new lookup. 1121 */ 1122 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1123 { 1124 struct inode *dir; 1125 struct inode *inode; 1126 struct dentry *parent; 1127 struct nfs_fh *fhandle = NULL; 1128 struct nfs_fattr *fattr = NULL; 1129 struct nfs4_label *label = NULL; 1130 int error; 1131 1132 if (flags & LOOKUP_RCU) { 1133 parent = ACCESS_ONCE(dentry->d_parent); 1134 dir = d_inode_rcu(parent); 1135 if (!dir) 1136 return -ECHILD; 1137 } else { 1138 parent = dget_parent(dentry); 1139 dir = d_inode(parent); 1140 } 1141 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1142 inode = d_inode(dentry); 1143 1144 if (!inode) { 1145 if (nfs_neg_need_reval(dir, dentry, flags)) { 1146 if (flags & LOOKUP_RCU) 1147 return -ECHILD; 1148 goto out_bad; 1149 } 1150 goto out_valid_noent; 1151 } 1152 1153 if (is_bad_inode(inode)) { 1154 if (flags & LOOKUP_RCU) 1155 return -ECHILD; 1156 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1157 __func__, dentry); 1158 goto out_bad; 1159 } 1160 1161 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) 1162 goto out_set_verifier; 1163 1164 /* Force a full look up iff the parent directory has changed */ 1165 if (!nfs_is_exclusive_create(dir, flags) && 1166 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1167 1168 if (nfs_lookup_verify_inode(inode, flags)) { 1169 if (flags & LOOKUP_RCU) 1170 return -ECHILD; 1171 goto out_zap_parent; 1172 } 1173 goto out_valid; 1174 } 1175 1176 if (flags & LOOKUP_RCU) 1177 return -ECHILD; 1178 1179 if (NFS_STALE(inode)) 1180 goto out_bad; 1181 1182 error = -ENOMEM; 1183 fhandle = nfs_alloc_fhandle(); 1184 fattr = nfs_alloc_fattr(); 1185 if (fhandle == NULL || fattr == NULL) 1186 goto out_error; 1187 1188 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 1189 if (IS_ERR(label)) 1190 goto out_error; 1191 1192 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1193 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1194 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1195 if (error) 1196 goto out_bad; 1197 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1198 goto out_bad; 1199 if ((error = nfs_refresh_inode(inode, fattr)) != 0) 1200 goto out_bad; 1201 1202 nfs_setsecurity(inode, fattr, label); 1203 1204 nfs_free_fattr(fattr); 1205 nfs_free_fhandle(fhandle); 1206 nfs4_label_free(label); 1207 1208 out_set_verifier: 1209 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1210 out_valid: 1211 /* Success: notify readdir to use READDIRPLUS */ 1212 nfs_advise_use_readdirplus(dir); 1213 out_valid_noent: 1214 if (flags & LOOKUP_RCU) { 1215 if (parent != ACCESS_ONCE(dentry->d_parent)) 1216 return -ECHILD; 1217 } else 1218 dput(parent); 1219 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1220 __func__, dentry); 1221 return 1; 1222 out_zap_parent: 1223 nfs_zap_caches(dir); 1224 out_bad: 1225 WARN_ON(flags & LOOKUP_RCU); 1226 nfs_free_fattr(fattr); 1227 nfs_free_fhandle(fhandle); 1228 nfs4_label_free(label); 1229 nfs_mark_for_revalidate(dir); 1230 if (inode && S_ISDIR(inode->i_mode)) { 1231 /* Purge readdir caches. */ 1232 nfs_zap_caches(inode); 1233 /* 1234 * We can't d_drop the root of a disconnected tree: 1235 * its d_hash is on the s_anon list and d_drop() would hide 1236 * it from shrink_dcache_for_unmount(), leading to busy 1237 * inodes on unmount and further oopses. 1238 */ 1239 if (IS_ROOT(dentry)) 1240 goto out_valid; 1241 } 1242 dput(parent); 1243 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1244 __func__, dentry); 1245 return 0; 1246 out_error: 1247 WARN_ON(flags & LOOKUP_RCU); 1248 nfs_free_fattr(fattr); 1249 nfs_free_fhandle(fhandle); 1250 nfs4_label_free(label); 1251 dput(parent); 1252 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1253 __func__, dentry, error); 1254 return error; 1255 } 1256 1257 /* 1258 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1259 * when we don't really care about the dentry name. This is called when a 1260 * pathwalk ends on a dentry that was not found via a normal lookup in the 1261 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1262 * 1263 * In this situation, we just want to verify that the inode itself is OK 1264 * since the dentry might have changed on the server. 1265 */ 1266 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1267 { 1268 int error; 1269 struct inode *inode = d_inode(dentry); 1270 1271 /* 1272 * I believe we can only get a negative dentry here in the case of a 1273 * procfs-style symlink. Just assume it's correct for now, but we may 1274 * eventually need to do something more here. 1275 */ 1276 if (!inode) { 1277 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1278 __func__, dentry); 1279 return 1; 1280 } 1281 1282 if (is_bad_inode(inode)) { 1283 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1284 __func__, dentry); 1285 return 0; 1286 } 1287 1288 error = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1289 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1290 __func__, inode->i_ino, error ? "invalid" : "valid"); 1291 return !error; 1292 } 1293 1294 /* 1295 * This is called from dput() when d_count is going to 0. 1296 */ 1297 static int nfs_dentry_delete(const struct dentry *dentry) 1298 { 1299 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1300 dentry, dentry->d_flags); 1301 1302 /* Unhash any dentry with a stale inode */ 1303 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1304 return 1; 1305 1306 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1307 /* Unhash it, so that ->d_iput() would be called */ 1308 return 1; 1309 } 1310 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 1311 /* Unhash it, so that ancestors of killed async unlink 1312 * files will be cleaned up during umount */ 1313 return 1; 1314 } 1315 return 0; 1316 1317 } 1318 1319 /* Ensure that we revalidate inode->i_nlink */ 1320 static void nfs_drop_nlink(struct inode *inode) 1321 { 1322 spin_lock(&inode->i_lock); 1323 /* drop the inode if we're reasonably sure this is the last link */ 1324 if (inode->i_nlink == 1) 1325 clear_nlink(inode); 1326 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; 1327 spin_unlock(&inode->i_lock); 1328 } 1329 1330 /* 1331 * Called when the dentry loses inode. 1332 * We use it to clean up silly-renamed files. 1333 */ 1334 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1335 { 1336 if (S_ISDIR(inode->i_mode)) 1337 /* drop any readdir cache as it could easily be old */ 1338 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1339 1340 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1341 nfs_complete_unlink(dentry, inode); 1342 nfs_drop_nlink(inode); 1343 } 1344 iput(inode); 1345 } 1346 1347 static void nfs_d_release(struct dentry *dentry) 1348 { 1349 /* free cached devname value, if it survived that far */ 1350 if (unlikely(dentry->d_fsdata)) { 1351 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1352 WARN_ON(1); 1353 else 1354 kfree(dentry->d_fsdata); 1355 } 1356 } 1357 1358 const struct dentry_operations nfs_dentry_operations = { 1359 .d_revalidate = nfs_lookup_revalidate, 1360 .d_weak_revalidate = nfs_weak_revalidate, 1361 .d_delete = nfs_dentry_delete, 1362 .d_iput = nfs_dentry_iput, 1363 .d_automount = nfs_d_automount, 1364 .d_release = nfs_d_release, 1365 }; 1366 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1367 1368 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1369 { 1370 struct dentry *res; 1371 struct inode *inode = NULL; 1372 struct nfs_fh *fhandle = NULL; 1373 struct nfs_fattr *fattr = NULL; 1374 struct nfs4_label *label = NULL; 1375 int error; 1376 1377 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1378 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1379 1380 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1381 return ERR_PTR(-ENAMETOOLONG); 1382 1383 /* 1384 * If we're doing an exclusive create, optimize away the lookup 1385 * but don't hash the dentry. 1386 */ 1387 if (nfs_is_exclusive_create(dir, flags)) 1388 return NULL; 1389 1390 res = ERR_PTR(-ENOMEM); 1391 fhandle = nfs_alloc_fhandle(); 1392 fattr = nfs_alloc_fattr(); 1393 if (fhandle == NULL || fattr == NULL) 1394 goto out; 1395 1396 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1397 if (IS_ERR(label)) 1398 goto out; 1399 1400 /* Protect against concurrent sillydeletes */ 1401 trace_nfs_lookup_enter(dir, dentry, flags); 1402 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1403 if (error == -ENOENT) 1404 goto no_entry; 1405 if (error < 0) { 1406 res = ERR_PTR(error); 1407 goto out_unblock_sillyrename; 1408 } 1409 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1410 res = ERR_CAST(inode); 1411 if (IS_ERR(res)) 1412 goto out_unblock_sillyrename; 1413 1414 /* Success: notify readdir to use READDIRPLUS */ 1415 nfs_advise_use_readdirplus(dir); 1416 1417 no_entry: 1418 res = d_splice_alias(inode, dentry); 1419 if (res != NULL) { 1420 if (IS_ERR(res)) 1421 goto out_unblock_sillyrename; 1422 dentry = res; 1423 } 1424 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1425 out_unblock_sillyrename: 1426 trace_nfs_lookup_exit(dir, dentry, flags, error); 1427 nfs4_label_free(label); 1428 out: 1429 nfs_free_fattr(fattr); 1430 nfs_free_fhandle(fhandle); 1431 return res; 1432 } 1433 EXPORT_SYMBOL_GPL(nfs_lookup); 1434 1435 #if IS_ENABLED(CONFIG_NFS_V4) 1436 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1437 1438 const struct dentry_operations nfs4_dentry_operations = { 1439 .d_revalidate = nfs4_lookup_revalidate, 1440 .d_delete = nfs_dentry_delete, 1441 .d_iput = nfs_dentry_iput, 1442 .d_automount = nfs_d_automount, 1443 .d_release = nfs_d_release, 1444 }; 1445 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1446 1447 static fmode_t flags_to_mode(int flags) 1448 { 1449 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1450 if ((flags & O_ACCMODE) != O_WRONLY) 1451 res |= FMODE_READ; 1452 if ((flags & O_ACCMODE) != O_RDONLY) 1453 res |= FMODE_WRITE; 1454 return res; 1455 } 1456 1457 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags) 1458 { 1459 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags)); 1460 } 1461 1462 static int do_open(struct inode *inode, struct file *filp) 1463 { 1464 nfs_fscache_open_file(inode, filp); 1465 return 0; 1466 } 1467 1468 static int nfs_finish_open(struct nfs_open_context *ctx, 1469 struct dentry *dentry, 1470 struct file *file, unsigned open_flags, 1471 int *opened) 1472 { 1473 int err; 1474 1475 err = finish_open(file, dentry, do_open, opened); 1476 if (err) 1477 goto out; 1478 nfs_file_set_open_context(file, ctx); 1479 1480 out: 1481 return err; 1482 } 1483 1484 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1485 struct file *file, unsigned open_flags, 1486 umode_t mode, int *opened) 1487 { 1488 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1489 struct nfs_open_context *ctx; 1490 struct dentry *res; 1491 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1492 struct inode *inode; 1493 unsigned int lookup_flags = 0; 1494 bool switched = false; 1495 int err; 1496 1497 /* Expect a negative dentry */ 1498 BUG_ON(d_inode(dentry)); 1499 1500 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1501 dir->i_sb->s_id, dir->i_ino, dentry); 1502 1503 err = nfs_check_flags(open_flags); 1504 if (err) 1505 return err; 1506 1507 /* NFS only supports OPEN on regular files */ 1508 if ((open_flags & O_DIRECTORY)) { 1509 if (!d_in_lookup(dentry)) { 1510 /* 1511 * Hashed negative dentry with O_DIRECTORY: dentry was 1512 * revalidated and is fine, no need to perform lookup 1513 * again 1514 */ 1515 return -ENOENT; 1516 } 1517 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1518 goto no_open; 1519 } 1520 1521 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1522 return -ENAMETOOLONG; 1523 1524 if (open_flags & O_CREAT) { 1525 attr.ia_valid |= ATTR_MODE; 1526 attr.ia_mode = mode & ~current_umask(); 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); 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 /* 1993 * RENAME 1994 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1995 * different file handle for the same inode after a rename (e.g. when 1996 * moving to a different directory). A fail-safe method to do so would 1997 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1998 * rename the old file using the sillyrename stuff. This way, the original 1999 * file in old_dir will go away when the last process iput()s the inode. 2000 * 2001 * FIXED. 2002 * 2003 * It actually works quite well. One needs to have the possibility for 2004 * at least one ".nfs..." file in each directory the file ever gets 2005 * moved or linked to which happens automagically with the new 2006 * implementation that only depends on the dcache stuff instead of 2007 * using the inode layer 2008 * 2009 * Unfortunately, things are a little more complicated than indicated 2010 * above. For a cross-directory move, we want to make sure we can get 2011 * rid of the old inode after the operation. This means there must be 2012 * no pending writes (if it's a file), and the use count must be 1. 2013 * If these conditions are met, we can drop the dentries before doing 2014 * the rename. 2015 */ 2016 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 2017 struct inode *new_dir, struct dentry *new_dentry) 2018 { 2019 struct inode *old_inode = d_inode(old_dentry); 2020 struct inode *new_inode = d_inode(new_dentry); 2021 struct dentry *dentry = NULL, *rehash = NULL; 2022 struct rpc_task *task; 2023 int error = -EBUSY; 2024 2025 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2026 old_dentry, new_dentry, 2027 d_count(new_dentry)); 2028 2029 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2030 /* 2031 * For non-directories, check whether the target is busy and if so, 2032 * make a copy of the dentry and then do a silly-rename. If the 2033 * silly-rename succeeds, the copied dentry is hashed and becomes 2034 * the new target. 2035 */ 2036 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2037 /* 2038 * To prevent any new references to the target during the 2039 * rename, we unhash the dentry in advance. 2040 */ 2041 if (!d_unhashed(new_dentry)) { 2042 d_drop(new_dentry); 2043 rehash = new_dentry; 2044 } 2045 2046 if (d_count(new_dentry) > 2) { 2047 int err; 2048 2049 /* copy the target dentry's name */ 2050 dentry = d_alloc(new_dentry->d_parent, 2051 &new_dentry->d_name); 2052 if (!dentry) 2053 goto out; 2054 2055 /* silly-rename the existing target ... */ 2056 err = nfs_sillyrename(new_dir, new_dentry); 2057 if (err) 2058 goto out; 2059 2060 new_dentry = dentry; 2061 rehash = NULL; 2062 new_inode = NULL; 2063 } 2064 } 2065 2066 NFS_PROTO(old_inode)->return_delegation(old_inode); 2067 if (new_inode != NULL) 2068 NFS_PROTO(new_inode)->return_delegation(new_inode); 2069 2070 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2071 if (IS_ERR(task)) { 2072 error = PTR_ERR(task); 2073 goto out; 2074 } 2075 2076 error = rpc_wait_for_completion_task(task); 2077 if (error == 0) 2078 error = task->tk_status; 2079 rpc_put_task(task); 2080 nfs_mark_for_revalidate(old_inode); 2081 out: 2082 if (rehash) 2083 d_rehash(rehash); 2084 trace_nfs_rename_exit(old_dir, old_dentry, 2085 new_dir, new_dentry, error); 2086 if (!error) { 2087 if (new_inode != NULL) 2088 nfs_drop_nlink(new_inode); 2089 d_move(old_dentry, new_dentry); 2090 nfs_set_verifier(new_dentry, 2091 nfs_save_change_attribute(new_dir)); 2092 } else if (error == -ENOENT) 2093 nfs_dentry_handle_enoent(old_dentry); 2094 2095 /* new dentry created? */ 2096 if (dentry) 2097 dput(dentry); 2098 return error; 2099 } 2100 EXPORT_SYMBOL_GPL(nfs_rename); 2101 2102 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2103 static LIST_HEAD(nfs_access_lru_list); 2104 static atomic_long_t nfs_access_nr_entries; 2105 2106 static unsigned long nfs_access_max_cachesize = ULONG_MAX; 2107 module_param(nfs_access_max_cachesize, ulong, 0644); 2108 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2109 2110 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2111 { 2112 put_rpccred(entry->cred); 2113 kfree_rcu(entry, rcu_head); 2114 smp_mb__before_atomic(); 2115 atomic_long_dec(&nfs_access_nr_entries); 2116 smp_mb__after_atomic(); 2117 } 2118 2119 static void nfs_access_free_list(struct list_head *head) 2120 { 2121 struct nfs_access_entry *cache; 2122 2123 while (!list_empty(head)) { 2124 cache = list_entry(head->next, struct nfs_access_entry, lru); 2125 list_del(&cache->lru); 2126 nfs_access_free_entry(cache); 2127 } 2128 } 2129 2130 static unsigned long 2131 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2132 { 2133 LIST_HEAD(head); 2134 struct nfs_inode *nfsi, *next; 2135 struct nfs_access_entry *cache; 2136 long freed = 0; 2137 2138 spin_lock(&nfs_access_lru_lock); 2139 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2140 struct inode *inode; 2141 2142 if (nr_to_scan-- == 0) 2143 break; 2144 inode = &nfsi->vfs_inode; 2145 spin_lock(&inode->i_lock); 2146 if (list_empty(&nfsi->access_cache_entry_lru)) 2147 goto remove_lru_entry; 2148 cache = list_entry(nfsi->access_cache_entry_lru.next, 2149 struct nfs_access_entry, lru); 2150 list_move(&cache->lru, &head); 2151 rb_erase(&cache->rb_node, &nfsi->access_cache); 2152 freed++; 2153 if (!list_empty(&nfsi->access_cache_entry_lru)) 2154 list_move_tail(&nfsi->access_cache_inode_lru, 2155 &nfs_access_lru_list); 2156 else { 2157 remove_lru_entry: 2158 list_del_init(&nfsi->access_cache_inode_lru); 2159 smp_mb__before_atomic(); 2160 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2161 smp_mb__after_atomic(); 2162 } 2163 spin_unlock(&inode->i_lock); 2164 } 2165 spin_unlock(&nfs_access_lru_lock); 2166 nfs_access_free_list(&head); 2167 return freed; 2168 } 2169 2170 unsigned long 2171 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2172 { 2173 int nr_to_scan = sc->nr_to_scan; 2174 gfp_t gfp_mask = sc->gfp_mask; 2175 2176 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2177 return SHRINK_STOP; 2178 return nfs_do_access_cache_scan(nr_to_scan); 2179 } 2180 2181 2182 unsigned long 2183 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2184 { 2185 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2186 } 2187 2188 static void 2189 nfs_access_cache_enforce_limit(void) 2190 { 2191 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2192 unsigned long diff; 2193 unsigned int nr_to_scan; 2194 2195 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2196 return; 2197 nr_to_scan = 100; 2198 diff = nr_entries - nfs_access_max_cachesize; 2199 if (diff < nr_to_scan) 2200 nr_to_scan = diff; 2201 nfs_do_access_cache_scan(nr_to_scan); 2202 } 2203 2204 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2205 { 2206 struct rb_root *root_node = &nfsi->access_cache; 2207 struct rb_node *n; 2208 struct nfs_access_entry *entry; 2209 2210 /* Unhook entries from the cache */ 2211 while ((n = rb_first(root_node)) != NULL) { 2212 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2213 rb_erase(n, root_node); 2214 list_move(&entry->lru, head); 2215 } 2216 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2217 } 2218 2219 void nfs_access_zap_cache(struct inode *inode) 2220 { 2221 LIST_HEAD(head); 2222 2223 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2224 return; 2225 /* Remove from global LRU init */ 2226 spin_lock(&nfs_access_lru_lock); 2227 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2228 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2229 2230 spin_lock(&inode->i_lock); 2231 __nfs_access_zap_cache(NFS_I(inode), &head); 2232 spin_unlock(&inode->i_lock); 2233 spin_unlock(&nfs_access_lru_lock); 2234 nfs_access_free_list(&head); 2235 } 2236 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2237 2238 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 2239 { 2240 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2241 struct nfs_access_entry *entry; 2242 2243 while (n != NULL) { 2244 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2245 2246 if (cred < entry->cred) 2247 n = n->rb_left; 2248 else if (cred > entry->cred) 2249 n = n->rb_right; 2250 else 2251 return entry; 2252 } 2253 return NULL; 2254 } 2255 2256 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2257 { 2258 struct nfs_inode *nfsi = NFS_I(inode); 2259 struct nfs_access_entry *cache; 2260 int err = -ENOENT; 2261 2262 spin_lock(&inode->i_lock); 2263 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2264 goto out_zap; 2265 cache = nfs_access_search_rbtree(inode, cred); 2266 if (cache == NULL) 2267 goto out; 2268 if (!nfs_have_delegated_attributes(inode) && 2269 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 2270 goto out_stale; 2271 res->jiffies = cache->jiffies; 2272 res->cred = cache->cred; 2273 res->mask = cache->mask; 2274 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2275 err = 0; 2276 out: 2277 spin_unlock(&inode->i_lock); 2278 return err; 2279 out_stale: 2280 rb_erase(&cache->rb_node, &nfsi->access_cache); 2281 list_del(&cache->lru); 2282 spin_unlock(&inode->i_lock); 2283 nfs_access_free_entry(cache); 2284 return -ENOENT; 2285 out_zap: 2286 spin_unlock(&inode->i_lock); 2287 nfs_access_zap_cache(inode); 2288 return -ENOENT; 2289 } 2290 2291 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2292 { 2293 /* Only check the most recently returned cache entry, 2294 * but do it without locking. 2295 */ 2296 struct nfs_inode *nfsi = NFS_I(inode); 2297 struct nfs_access_entry *cache; 2298 int err = -ECHILD; 2299 struct list_head *lh; 2300 2301 rcu_read_lock(); 2302 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2303 goto out; 2304 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); 2305 cache = list_entry(lh, struct nfs_access_entry, lru); 2306 if (lh == &nfsi->access_cache_entry_lru || 2307 cred != cache->cred) 2308 cache = NULL; 2309 if (cache == NULL) 2310 goto out; 2311 if (!nfs_have_delegated_attributes(inode) && 2312 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 2313 goto out; 2314 res->jiffies = cache->jiffies; 2315 res->cred = cache->cred; 2316 res->mask = cache->mask; 2317 err = 0; 2318 out: 2319 rcu_read_unlock(); 2320 return err; 2321 } 2322 2323 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2324 { 2325 struct nfs_inode *nfsi = NFS_I(inode); 2326 struct rb_root *root_node = &nfsi->access_cache; 2327 struct rb_node **p = &root_node->rb_node; 2328 struct rb_node *parent = NULL; 2329 struct nfs_access_entry *entry; 2330 2331 spin_lock(&inode->i_lock); 2332 while (*p != NULL) { 2333 parent = *p; 2334 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2335 2336 if (set->cred < entry->cred) 2337 p = &parent->rb_left; 2338 else if (set->cred > entry->cred) 2339 p = &parent->rb_right; 2340 else 2341 goto found; 2342 } 2343 rb_link_node(&set->rb_node, parent, p); 2344 rb_insert_color(&set->rb_node, root_node); 2345 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2346 spin_unlock(&inode->i_lock); 2347 return; 2348 found: 2349 rb_replace_node(parent, &set->rb_node, root_node); 2350 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2351 list_del(&entry->lru); 2352 spin_unlock(&inode->i_lock); 2353 nfs_access_free_entry(entry); 2354 } 2355 2356 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2357 { 2358 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2359 if (cache == NULL) 2360 return; 2361 RB_CLEAR_NODE(&cache->rb_node); 2362 cache->jiffies = set->jiffies; 2363 cache->cred = get_rpccred(set->cred); 2364 cache->mask = set->mask; 2365 2366 /* The above field assignments must be visible 2367 * before this item appears on the lru. We cannot easily 2368 * use rcu_assign_pointer, so just force the memory barrier. 2369 */ 2370 smp_wmb(); 2371 nfs_access_add_rbtree(inode, cache); 2372 2373 /* Update accounting */ 2374 smp_mb__before_atomic(); 2375 atomic_long_inc(&nfs_access_nr_entries); 2376 smp_mb__after_atomic(); 2377 2378 /* Add inode to global LRU list */ 2379 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2380 spin_lock(&nfs_access_lru_lock); 2381 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2382 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2383 &nfs_access_lru_list); 2384 spin_unlock(&nfs_access_lru_lock); 2385 } 2386 nfs_access_cache_enforce_limit(); 2387 } 2388 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2389 2390 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2391 { 2392 entry->mask = 0; 2393 if (access_result & NFS4_ACCESS_READ) 2394 entry->mask |= MAY_READ; 2395 if (access_result & 2396 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) 2397 entry->mask |= MAY_WRITE; 2398 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) 2399 entry->mask |= MAY_EXEC; 2400 } 2401 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2402 2403 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 2404 { 2405 struct nfs_access_entry cache; 2406 int status; 2407 2408 trace_nfs_access_enter(inode); 2409 2410 status = nfs_access_get_cached_rcu(inode, cred, &cache); 2411 if (status != 0) 2412 status = nfs_access_get_cached(inode, cred, &cache); 2413 if (status == 0) 2414 goto out_cached; 2415 2416 status = -ECHILD; 2417 if (mask & MAY_NOT_BLOCK) 2418 goto out; 2419 2420 /* Be clever: ask server to check for all possible rights */ 2421 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 2422 cache.cred = cred; 2423 cache.jiffies = jiffies; 2424 status = NFS_PROTO(inode)->access(inode, &cache); 2425 if (status != 0) { 2426 if (status == -ESTALE) { 2427 nfs_zap_caches(inode); 2428 if (!S_ISDIR(inode->i_mode)) 2429 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 2430 } 2431 goto out; 2432 } 2433 nfs_access_add_cache(inode, &cache); 2434 out_cached: 2435 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2436 status = -EACCES; 2437 out: 2438 trace_nfs_access_exit(inode, status); 2439 return status; 2440 } 2441 2442 static int nfs_open_permission_mask(int openflags) 2443 { 2444 int mask = 0; 2445 2446 if (openflags & __FMODE_EXEC) { 2447 /* ONLY check exec rights */ 2448 mask = MAY_EXEC; 2449 } else { 2450 if ((openflags & O_ACCMODE) != O_WRONLY) 2451 mask |= MAY_READ; 2452 if ((openflags & O_ACCMODE) != O_RDONLY) 2453 mask |= MAY_WRITE; 2454 } 2455 2456 return mask; 2457 } 2458 2459 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 2460 { 2461 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2462 } 2463 EXPORT_SYMBOL_GPL(nfs_may_open); 2464 2465 static int nfs_execute_ok(struct inode *inode, int mask) 2466 { 2467 struct nfs_server *server = NFS_SERVER(inode); 2468 int ret; 2469 2470 if (mask & MAY_NOT_BLOCK) 2471 ret = nfs_revalidate_inode_rcu(server, inode); 2472 else 2473 ret = nfs_revalidate_inode(server, inode); 2474 if (ret == 0 && !execute_ok(inode)) 2475 ret = -EACCES; 2476 return ret; 2477 } 2478 2479 int nfs_permission(struct inode *inode, int mask) 2480 { 2481 struct rpc_cred *cred; 2482 int res = 0; 2483 2484 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2485 2486 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2487 goto out; 2488 /* Is this sys_access() ? */ 2489 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2490 goto force_lookup; 2491 2492 switch (inode->i_mode & S_IFMT) { 2493 case S_IFLNK: 2494 goto out; 2495 case S_IFREG: 2496 if ((mask & MAY_OPEN) && 2497 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 2498 return 0; 2499 break; 2500 case S_IFDIR: 2501 /* 2502 * Optimize away all write operations, since the server 2503 * will check permissions when we perform the op. 2504 */ 2505 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2506 goto out; 2507 } 2508 2509 force_lookup: 2510 if (!NFS_PROTO(inode)->access) 2511 goto out_notsup; 2512 2513 /* Always try fast lookups first */ 2514 rcu_read_lock(); 2515 cred = rpc_lookup_cred_nonblock(); 2516 if (!IS_ERR(cred)) 2517 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); 2518 else 2519 res = PTR_ERR(cred); 2520 rcu_read_unlock(); 2521 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { 2522 /* Fast lookup failed, try the slow way */ 2523 cred = rpc_lookup_cred(); 2524 if (!IS_ERR(cred)) { 2525 res = nfs_do_access(inode, cred, mask); 2526 put_rpccred(cred); 2527 } else 2528 res = PTR_ERR(cred); 2529 } 2530 out: 2531 if (!res && (mask & MAY_EXEC)) 2532 res = nfs_execute_ok(inode, mask); 2533 2534 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2535 inode->i_sb->s_id, inode->i_ino, mask, res); 2536 return res; 2537 out_notsup: 2538 if (mask & MAY_NOT_BLOCK) 2539 return -ECHILD; 2540 2541 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2542 if (res == 0) 2543 res = generic_permission(inode, mask); 2544 goto out; 2545 } 2546 EXPORT_SYMBOL_GPL(nfs_permission); 2547 2548 /* 2549 * Local variables: 2550 * version-control: t 2551 * kept-new-versions: 5 2552 * End: 2553 */ 2554