fs/nfs/dir.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/nfs/dir.c
   4  *
   5  *  Copyright (C) 1992  Rick Sladkey
   6  *
   7  *  nfs directory handling functions
   8  *
   9  * 10 Apr 1996  Added silly rename for unlink   --okir
  10  * 28 Sep 1996  Improved directory cache --okir
  11  * 23 Aug 1997  Claus Heine [email protected]
  12  *              Re-implemented silly rename for unlink, newly implemented
  13  *              silly rename for nfs_rename() following the suggestions
  14  *              of Olaf Kirch (okir) found in this file.
  15  *              Following Linus comments on my original hack, this version
  16  *              depends only on the dcache stuff and doesn't touch the inode
  17  *              layer (iput() and friends).
  18  *  6 Jun 1999  Cache readdir lookups in the page cache. -DaveM
  19  */
  20
  21 #include <linux/compat.h>
  22 #include <linux/module.h>
  23 #include <linux/time.h>
  24 #include <linux/errno.h>
  25 #include <linux/stat.h>
  26 #include <linux/fcntl.h>
  27 #include <linux/string.h>
  28 #include <linux/kernel.h>
  29 #include <linux/slab.h>
  30 #include <linux/mm.h>
  31 #include <linux/sunrpc/clnt.h>
  32 #include <linux/nfs_fs.h>
  33 #include <linux/nfs_mount.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/pagevec.h>
  36 #include <linux/namei.h>
  37 #include <linux/mount.h>
  38 #include <linux/swap.h>
  39 #include <linux/sched.h>
  40 #include <linux/kmemleak.h>
  41 #include <linux/xattr.h>
  42 #include <linux/hash.h>
  43
  44 #include "delegation.h"
  45 #include "iostat.h"
  46 #include "internal.h"
  47 #include "fscache.h"
  48
  49 #include "nfstrace.h"
  50
  51 /* #define NFS_DEBUG_VERBOSE 1 */
  52
  53 static int nfs_opendir(struct inode *, struct file *);
  54 static int nfs_closedir(struct inode *, struct file *);
  55 static int nfs_readdir(struct file *, struct dir_context *);
  56 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
  57 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
  58 static void nfs_readdir_free_folio(struct folio *);
  59
  60 const struct file_operations nfs_dir_operations = {
  61         .llseek         = nfs_llseek_dir,
  62         .read           = generic_read_dir,
  63         .iterate_shared = nfs_readdir,
  64         .open           = nfs_opendir,
  65         .release        = nfs_closedir,
  66         .fsync          = nfs_fsync_dir,
  67 };
  68
  69 const struct address_space_operations nfs_dir_aops = {
  70         .free_folio = nfs_readdir_free_folio,
  71 };
  72
  73 #define NFS_INIT_DTSIZE PAGE_SIZE
  74
  75 static struct nfs_open_dir_context *
  76 alloc_nfs_open_dir_context(struct inode *dir)
  77 {
  78         struct nfs_inode *nfsi = NFS_I(dir);
  79         struct nfs_open_dir_context *ctx;
  80
  81         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
  82         if (ctx != NULL) {
  83                 ctx->attr_gencount = nfsi->attr_gencount;
  84                 ctx->dtsize = NFS_INIT_DTSIZE;
  85                 spin_lock(&dir->i_lock);
  86                 if (list_empty(&nfsi->open_files) &&
  87                     (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
  88                         nfs_set_cache_invalid(dir,
  89                                               NFS_INO_INVALID_DATA |
  90                                                       NFS_INO_REVAL_FORCED);
  91                 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
  92                 memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
  93                 spin_unlock(&dir->i_lock);
  94                 return ctx;
  95         }
  96         return  ERR_PTR(-ENOMEM);
  97 }
  98
  99 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
 100 {
 101         spin_lock(&dir->i_lock);
 102         list_del_rcu(&ctx->list);
 103         spin_unlock(&dir->i_lock);
 104         kfree_rcu(ctx, rcu_head);
 105 }
 106
 107 /*
 108  * Open file
 109  */
 110 static int
 111 nfs_opendir(struct inode *inode, struct file *filp)
 112 {
 113         int res = 0;
 114         struct nfs_open_dir_context *ctx;
 115
 116         dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
 117
 118         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 119
 120         ctx = alloc_nfs_open_dir_context(inode);
 121         if (IS_ERR(ctx)) {
 122                 res = PTR_ERR(ctx);
 123                 goto out;
 124         }
 125         filp->private_data = ctx;
 126 out:
 127         return res;
 128 }
 129
 130 static int
 131 nfs_closedir(struct inode *inode, struct file *filp)
 132 {
 133         put_nfs_open_dir_context(file_inode(filp), filp->private_data);
 134         return 0;
 135 }
 136
 137 struct nfs_cache_array_entry {
 138         u64 cookie;
 139         u64 ino;
 140         const char *name;
 141         unsigned int name_len;
 142         unsigned char d_type;
 143 };
 144
 145 struct nfs_cache_array {
 146         u64 change_attr;
 147         u64 last_cookie;
 148         unsigned int size;
 149         unsigned char page_full : 1,
 150                       page_is_eof : 1,
 151                       cookies_are_ordered : 1;
 152         struct nfs_cache_array_entry array[];
 153 };
 154
 155 struct nfs_readdir_descriptor {
 156         struct file     *file;
 157         struct page     *page;
 158         struct dir_context *ctx;
 159         pgoff_t         page_index;
 160         pgoff_t         page_index_max;
 161         u64             dir_cookie;
 162         u64             last_cookie;
 163         loff_t          current_index;
 164
 165         __be32          verf[NFS_DIR_VERIFIER_SIZE];
 166         unsigned long   dir_verifier;
 167         unsigned long   timestamp;
 168         unsigned long   gencount;
 169         unsigned long   attr_gencount;
 170         unsigned int    cache_entry_index;
 171         unsigned int    buffer_fills;
 172         unsigned int    dtsize;
 173         bool clear_cache;
 174         bool plus;
 175         bool eob;
 176         bool eof;
 177 };
 178
 179 static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz)
 180 {
 181         struct nfs_server *server = NFS_SERVER(file_inode(desc->file));
 182         unsigned int maxsize = server->dtsize;
 183
 184         if (sz > maxsize)
 185                 sz = maxsize;
 186         if (sz < NFS_MIN_FILE_IO_SIZE)
 187                 sz = NFS_MIN_FILE_IO_SIZE;
 188         desc->dtsize = sz;
 189 }
 190
 191 static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
 192 {
 193         nfs_set_dtsize(desc, desc->dtsize >> 1);
 194 }
 195
 196 static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
 197 {
 198         nfs_set_dtsize(desc, desc->dtsize << 1);
 199 }
 200
 201 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie,
 202                                         u64 change_attr)
 203 {
 204         struct nfs_cache_array *array;
 205
 206         array = kmap_atomic(page);
 207         array->change_attr = change_attr;
 208         array->last_cookie = last_cookie;
 209         array->size = 0;
 210         array->page_full = 0;
 211         array->page_is_eof = 0;
 212         array->cookies_are_ordered = 1;
 213         kunmap_atomic(array);
 214 }
 215
 216 /*
 217  * we are freeing strings created by nfs_add_to_readdir_array()
 218  */
 219 static void nfs_readdir_clear_array(struct page *page)
 220 {
 221         struct nfs_cache_array *array;
 222         unsigned int i;
 223
 224         array = kmap_atomic(page);
 225         for (i = 0; i < array->size; i++)
 226                 kfree(array->array[i].name);
 227         array->size = 0;
 228         kunmap_atomic(array);
 229 }
 230
 231 static void nfs_readdir_free_folio(struct folio *folio)
 232 {
 233         nfs_readdir_clear_array(&folio->page);
 234 }
 235
 236 static void nfs_readdir_page_reinit_array(struct page *page, u64 last_cookie,
 237                                           u64 change_attr)
 238 {
 239         nfs_readdir_clear_array(page);
 240         nfs_readdir_page_init_array(page, last_cookie, change_attr);
 241 }
 242
 243 static struct page *
 244 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
 245 {
 246         struct page *page = alloc_page(gfp_flags);
 247         if (page)
 248                 nfs_readdir_page_init_array(page, last_cookie, 0);
 249         return page;
 250 }
 251
 252 static void nfs_readdir_page_array_free(struct page *page)
 253 {
 254         if (page) {
 255                 nfs_readdir_clear_array(page);
 256                 put_page(page);
 257         }
 258 }
 259
 260 static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
 261 {
 262         return array->size == 0 ? array->last_cookie : array->array[0].cookie;
 263 }
 264
 265 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
 266 {
 267         array->page_is_eof = 1;
 268         array->page_full = 1;
 269 }
 270
 271 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
 272 {
 273         return array->page_full;
 274 }
 275
 276 /*
 277  * the caller is responsible for freeing qstr.name
 278  * when called by nfs_readdir_add_to_array, the strings will be freed in
 279  * nfs_clear_readdir_array()
 280  */
 281 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
 282 {
 283         const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
 284
 285         /*
 286          * Avoid a kmemleak false positive. The pointer to the name is stored
 287          * in a page cache page which kmemleak does not scan.
 288          */
 289         if (ret != NULL)
 290                 kmemleak_not_leak(ret);
 291         return ret;
 292 }
 293
 294 static size_t nfs_readdir_array_maxentries(void)
 295 {
 296         return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
 297                sizeof(struct nfs_cache_array_entry);
 298 }
 299
 300 /*
 301  * Check that the next array entry lies entirely within the page bounds
 302  */
 303 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
 304 {
 305         if (array->page_full)
 306                 return -ENOSPC;
 307         if (array->size == nfs_readdir_array_maxentries()) {
 308                 array->page_full = 1;
 309                 return -ENOSPC;
 310         }
 311         return 0;
 312 }
 313
 314 static int nfs_readdir_page_array_append(struct page *page,
 315                                          const struct nfs_entry *entry,
 316                                          u64 *cookie)
 317 {
 318         struct nfs_cache_array *array;
 319         struct nfs_cache_array_entry *cache_entry;
 320         const char *name;
 321         int ret = -ENOMEM;
 322
 323         name = nfs_readdir_copy_name(entry->name, entry->len);
 324
 325         array = kmap_atomic(page);
 326         if (!name)
 327                 goto out;
 328         ret = nfs_readdir_array_can_expand(array);
 329         if (ret) {
 330                 kfree(name);
 331                 goto out;
 332         }
 333
 334         cache_entry = &array->array[array->size];
 335         cache_entry->cookie = array->last_cookie;
 336         cache_entry->ino = entry->ino;
 337         cache_entry->d_type = entry->d_type;
 338         cache_entry->name_len = entry->len;
 339         cache_entry->name = name;
 340         array->last_cookie = entry->cookie;
 341         if (array->last_cookie <= cache_entry->cookie)
 342                 array->cookies_are_ordered = 0;
 343         array->size++;
 344         if (entry->eof != 0)
 345                 nfs_readdir_array_set_eof(array);
 346 out:
 347         *cookie = array->last_cookie;
 348         kunmap_atomic(array);
 349         return ret;
 350 }
 351
 352 #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
 353 /*
 354  * Hash algorithm allowing content addressible access to sequences
 355  * of directory cookies. Content is addressed by the value of the
 356  * cookie index of the first readdir entry in a page.
 357  *
 358  * We select only the first 18 bits to avoid issues with excessive
 359  * memory use for the page cache XArray. 18 bits should allow the caching
 360  * of 262144 pages of sequences of readdir entries. Since each page holds
 361  * 127 readdir entries for a typical 64-bit system, that works out to a
 362  * cache of ~ 33 million entries per directory.
 363  */
 364 static pgoff_t nfs_readdir_page_cookie_hash(u64 cookie)
 365 {
 366         if (cookie == 0)
 367                 return 0;
 368         return hash_64(cookie, 18);
 369 }
 370
 371 static bool nfs_readdir_page_validate(struct page *page, u64 last_cookie,
 372                                       u64 change_attr)
 373 {
 374         struct nfs_cache_array *array = kmap_atomic(page);
 375         int ret = true;
 376
 377         if (array->change_attr != change_attr)
 378                 ret = false;
 379         if (nfs_readdir_array_index_cookie(array) != last_cookie)
 380                 ret = false;
 381         kunmap_atomic(array);
 382         return ret;
 383 }
 384
 385 static void nfs_readdir_page_unlock_and_put(struct page *page)
 386 {
 387         unlock_page(page);
 388         put_page(page);
 389 }
 390
 391 static void nfs_readdir_page_init_and_validate(struct page *page, u64 cookie,
 392                                                u64 change_attr)
 393 {
 394         if (PageUptodate(page)) {
 395                 if (nfs_readdir_page_validate(page, cookie, change_attr))
 396                         return;
 397                 nfs_readdir_clear_array(page);
 398         }
 399         nfs_readdir_page_init_array(page, cookie, change_attr);
 400         SetPageUptodate(page);
 401 }
 402
 403 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
 404                                                 u64 cookie, u64 change_attr)
 405 {
 406         pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
 407         struct page *page;
 408
 409         page = grab_cache_page(mapping, index);
 410         if (!page)
 411                 return NULL;
 412         nfs_readdir_page_init_and_validate(page, cookie, change_attr);
 413         return page;
 414 }
 415
 416 static u64 nfs_readdir_page_last_cookie(struct page *page)
 417 {
 418         struct nfs_cache_array *array;
 419         u64 ret;
 420
 421         array = kmap_atomic(page);
 422         ret = array->last_cookie;
 423         kunmap_atomic(array);
 424         return ret;
 425 }
 426
 427 static bool nfs_readdir_page_needs_filling(struct page *page)
 428 {
 429         struct nfs_cache_array *array;
 430         bool ret;
 431
 432         array = kmap_atomic(page);
 433         ret = !nfs_readdir_array_is_full(array);
 434         kunmap_atomic(array);
 435         return ret;
 436 }
 437
 438 static void nfs_readdir_page_set_eof(struct page *page)
 439 {
 440         struct nfs_cache_array *array;
 441
 442         array = kmap_atomic(page);
 443         nfs_readdir_array_set_eof(array);
 444         kunmap_atomic(array);
 445 }
 446
 447 static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
 448                                               u64 cookie, u64 change_attr)
 449 {
 450         pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
 451         struct page *page;
 452
 453         page = grab_cache_page_nowait(mapping, index);
 454         if (!page)
 455                 return NULL;
 456         nfs_readdir_page_init_and_validate(page, cookie, change_attr);
 457         if (nfs_readdir_page_last_cookie(page) != cookie)
 458                 nfs_readdir_page_reinit_array(page, cookie, change_attr);
 459         return page;
 460 }
 461
 462 static inline
 463 int is_32bit_api(void)
 464 {
 465 #ifdef CONFIG_COMPAT
 466         return in_compat_syscall();
 467 #else
 468         return (BITS_PER_LONG == 32);
 469 #endif
 470 }
 471
 472 static
 473 bool nfs_readdir_use_cookie(const struct file *filp)
 474 {
 475         if ((filp->f_mode & FMODE_32BITHASH) ||
 476             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 477                 return false;
 478         return true;
 479 }
 480
 481 static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
 482                                         struct nfs_readdir_descriptor *desc)
 483 {
 484         if (array->page_full) {
 485                 desc->last_cookie = array->last_cookie;
 486                 desc->current_index += array->size;
 487                 desc->cache_entry_index = 0;
 488                 desc->page_index++;
 489         } else
 490                 desc->last_cookie = nfs_readdir_array_index_cookie(array);
 491 }
 492
 493 static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
 494 {
 495         desc->current_index = 0;
 496         desc->last_cookie = 0;
 497         desc->page_index = 0;
 498 }
 499
 500 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
 501                                       struct nfs_readdir_descriptor *desc)
 502 {
 503         loff_t diff = desc->ctx->pos - desc->current_index;
 504         unsigned int index;
 505
 506         if (diff < 0)
 507                 goto out_eof;
 508         if (diff >= array->size) {
 509                 if (array->page_is_eof)
 510                         goto out_eof;
 511                 nfs_readdir_seek_next_array(array, desc);
 512                 return -EAGAIN;
 513         }
 514
 515         index = (unsigned int)diff;
 516         desc->dir_cookie = array->array[index].cookie;
 517         desc->cache_entry_index = index;
 518         return 0;
 519 out_eof:
 520         desc->eof = true;
 521         return -EBADCOOKIE;
 522 }
 523
 524 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
 525                                               u64 cookie)
 526 {
 527         if (!array->cookies_are_ordered)
 528                 return true;
 529         /* Optimisation for monotonically increasing cookies */
 530         if (cookie >= array->last_cookie)
 531                 return false;
 532         if (array->size && cookie < array->array[0].cookie)
 533                 return false;
 534         return true;
 535 }
 536
 537 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
 538                                          struct nfs_readdir_descriptor *desc)
 539 {
 540         unsigned int i;
 541         int status = -EAGAIN;
 542
 543         if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
 544                 goto check_eof;
 545
 546         for (i = 0; i < array->size; i++) {
 547                 if (array->array[i].cookie == desc->dir_cookie) {
 548                         if (nfs_readdir_use_cookie(desc->file))
 549                                 desc->ctx->pos = desc->dir_cookie;
 550                         else
 551                                 desc->ctx->pos = desc->current_index + i;
 552                         desc->cache_entry_index = i;
 553                         return 0;
 554                 }
 555         }
 556 check_eof:
 557         if (array->page_is_eof) {
 558                 status = -EBADCOOKIE;
 559                 if (desc->dir_cookie == array->last_cookie)
 560                         desc->eof = true;
 561         } else
 562                 nfs_readdir_seek_next_array(array, desc);
 563         return status;
 564 }
 565
 566 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
 567 {
 568         struct nfs_cache_array *array;
 569         int status;
 570
 571         array = kmap_atomic(desc->page);
 572
 573         if (desc->dir_cookie == 0)
 574                 status = nfs_readdir_search_for_pos(array, desc);
 575         else
 576                 status = nfs_readdir_search_for_cookie(array, desc);
 577
 578         kunmap_atomic(array);
 579         return status;
 580 }
 581
 582 /* Fill a page with xdr information before transferring to the cache page */
 583 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
 584                                   __be32 *verf, u64 cookie,
 585                                   struct page **pages, size_t bufsize,
 586                                   __be32 *verf_res)
 587 {
 588         struct inode *inode = file_inode(desc->file);
 589         struct nfs_readdir_arg arg = {
 590                 .dentry = file_dentry(desc->file),
 591                 .cred = desc->file->f_cred,
 592                 .verf = verf,
 593                 .cookie = cookie,
 594                 .pages = pages,
 595                 .page_len = bufsize,
 596                 .plus = desc->plus,
 597         };
 598         struct nfs_readdir_res res = {
 599                 .verf = verf_res,
 600         };
 601         unsigned long   timestamp, gencount;
 602         int             error;
 603
 604  again:
 605         timestamp = jiffies;
 606         gencount = nfs_inc_attr_generation_counter();
 607         desc->dir_verifier = nfs_save_change_attribute(inode);
 608         error = NFS_PROTO(inode)->readdir(&arg, &res);
 609         if (error < 0) {
 610                 /* We requested READDIRPLUS, but the server doesn't grok it */
 611                 if (error == -ENOTSUPP && desc->plus) {
 612                         NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
 613                         desc->plus = arg.plus = false;
 614                         goto again;
 615                 }
 616                 goto error;
 617         }
 618         desc->timestamp = timestamp;
 619         desc->gencount = gencount;
 620 error:
 621         return error;
 622 }
 623
 624 static int xdr_decode(struct nfs_readdir_descriptor *desc,
 625                       struct nfs_entry *entry, struct xdr_stream *xdr)
 626 {
 627         struct inode *inode = file_inode(desc->file);
 628         int error;
 629
 630         error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
 631         if (error)
 632                 return error;
 633         entry->fattr->time_start = desc->timestamp;
 634         entry->fattr->gencount = desc->gencount;
 635         return 0;
 636 }
 637
 638 /* Match file and dirent using either filehandle or fileid
 639  * Note: caller is responsible for checking the fsid
 640  */
 641 static
 642 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
 643 {
 644         struct inode *inode;
 645         struct nfs_inode *nfsi;
 646
 647         if (d_really_is_negative(dentry))
 648                 return 0;
 649
 650         inode = d_inode(dentry);
 651         if (is_bad_inode(inode) || NFS_STALE(inode))
 652                 return 0;
 653
 654         nfsi = NFS_I(inode);
 655         if (entry->fattr->fileid != nfsi->fileid)
 656                 return 0;
 657         if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
 658                 return 0;
 659         return 1;
 660 }
 661
 662 #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
 663
 664 static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
 665                                 unsigned int cache_hits,
 666                                 unsigned int cache_misses)
 667 {
 668         if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
 669                 return false;
 670         if (ctx->pos == 0 ||
 671             cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
 672                 return true;
 673         return false;
 674 }
 675
 676 /*
 677  * This function is called by the getattr code to request the
 678  * use of readdirplus to accelerate any future lookups in the same
 679  * directory.
 680  */
 681 void nfs_readdir_record_entry_cache_hit(struct inode *dir)
 682 {
 683         struct nfs_inode *nfsi = NFS_I(dir);
 684         struct nfs_open_dir_context *ctx;
 685
 686         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 687             S_ISDIR(dir->i_mode)) {
 688                 rcu_read_lock();
 689                 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
 690                         atomic_inc(&ctx->cache_hits);
 691                 rcu_read_unlock();
 692         }
 693 }
 694
 695 /*
 696  * This function is mainly for use by nfs_getattr().
 697  *
 698  * If this is an 'ls -l', we want to force use of readdirplus.
 699  */
 700 void nfs_readdir_record_entry_cache_miss(struct inode *dir)
 701 {
 702         struct nfs_inode *nfsi = NFS_I(dir);
 703         struct nfs_open_dir_context *ctx;
 704
 705         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 706             S_ISDIR(dir->i_mode)) {
 707                 rcu_read_lock();
 708                 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
 709                         atomic_inc(&ctx->cache_misses);
 710                 rcu_read_unlock();
 711         }
 712 }
 713
 714 static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
 715                                                 unsigned int flags)
 716 {
 717         if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
 718                 return;
 719         if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
 720                 return;
 721         nfs_readdir_record_entry_cache_miss(dir);
 722 }
 723
 724 static
 725 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
 726                 unsigned long dir_verifier)
 727 {
 728         struct qstr filename = QSTR_INIT(entry->name, entry->len);
 729         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 730         struct dentry *dentry;
 731         struct dentry *alias;
 732         struct inode *inode;
 733         int status;
 734
 735         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
 736                 return;
 737         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
 738                 return;
 739         if (filename.len == 0)
 740                 return;
 741         /* Validate that the name doesn't contain any illegal '\0' */
 742         if (strnlen(filename.name, filename.len) != filename.len)
 743                 return;
 744         /* ...or '/' */
 745         if (strnchr(filename.name, filename.len, '/'))
 746                 return;
 747         if (filename.name[0] == '.') {
 748                 if (filename.len == 1)
 749                         return;
 750                 if (filename.len == 2 && filename.name[1] == '.')
 751                         return;
 752         }
 753         filename.hash = full_name_hash(parent, filename.name, filename.len);
 754
 755         dentry = d_lookup(parent, &filename);
 756 again:
 757         if (!dentry) {
 758                 dentry = d_alloc_parallel(parent, &filename, &wq);
 759                 if (IS_ERR(dentry))
 760                         return;
 761         }
 762         if (!d_in_lookup(dentry)) {
 763                 /* Is there a mountpoint here? If so, just exit */
 764                 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
 765                                         &entry->fattr->fsid))
 766                         goto out;
 767                 if (nfs_same_file(dentry, entry)) {
 768                         if (!entry->fh->size)
 769                                 goto out;
 770                         nfs_set_verifier(dentry, dir_verifier);
 771                         status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
 772                         if (!status)
 773                                 nfs_setsecurity(d_inode(dentry), entry->fattr);
 774                         trace_nfs_readdir_lookup_revalidate(d_inode(parent),
 775                                                             dentry, 0, status);
 776                         goto out;
 777                 } else {
 778                         trace_nfs_readdir_lookup_revalidate_failed(
 779                                 d_inode(parent), dentry, 0);
 780                         d_invalidate(dentry);
 781                         dput(dentry);
 782                         dentry = NULL;
 783                         goto again;
 784                 }
 785         }
 786         if (!entry->fh->size) {
 787                 d_lookup_done(dentry);
 788                 goto out;
 789         }
 790
 791         inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
 792         alias = d_splice_alias(inode, dentry);
 793         d_lookup_done(dentry);
 794         if (alias) {
 795                 if (IS_ERR(alias))
 796                         goto out;
 797                 dput(dentry);
 798                 dentry = alias;
 799         }
 800         nfs_set_verifier(dentry, dir_verifier);
 801         trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
 802 out:
 803         dput(dentry);
 804 }
 805
 806 static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
 807                                     struct nfs_entry *entry,
 808                                     struct xdr_stream *stream)
 809 {
 810         int ret;
 811
 812         if (entry->fattr->label)
 813                 entry->fattr->label->len = NFS4_MAXLABELLEN;
 814         ret = xdr_decode(desc, entry, stream);
 815         if (ret || !desc->plus)
 816                 return ret;
 817         nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
 818         return 0;
 819 }
 820
 821 /* Perform conversion from xdr to cache array */
 822 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
 823                                    struct nfs_entry *entry,
 824                                    struct page **xdr_pages, unsigned int buflen,
 825                                    struct page **arrays, size_t narrays,
 826                                    u64 change_attr)
 827 {
 828         struct address_space *mapping = desc->file->f_mapping;
 829         struct xdr_stream stream;
 830         struct xdr_buf buf;
 831         struct page *scratch, *new, *page = *arrays;
 832         u64 cookie;
 833         int status;
 834
 835         scratch = alloc_page(GFP_KERNEL);
 836         if (scratch == NULL)
 837                 return -ENOMEM;
 838
 839         xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
 840         xdr_set_scratch_page(&stream, scratch);
 841
 842         do {
 843                 status = nfs_readdir_entry_decode(desc, entry, &stream);
 844                 if (status != 0)
 845                         break;
 846
 847                 status = nfs_readdir_page_array_append(page, entry, &cookie);
 848                 if (status != -ENOSPC)
 849                         continue;
 850
 851                 if (page->mapping != mapping) {
 852                         if (!--narrays)
 853                                 break;
 854                         new = nfs_readdir_page_array_alloc(cookie, GFP_KERNEL);
 855                         if (!new)
 856                                 break;
 857                         arrays++;
 858                         *arrays = page = new;
 859                 } else {
 860                         new = nfs_readdir_page_get_next(mapping, cookie,
 861                                                         change_attr);
 862                         if (!new)
 863                                 break;
 864                         if (page != *arrays)
 865                                 nfs_readdir_page_unlock_and_put(page);
 866                         page = new;
 867                 }
 868                 desc->page_index_max++;
 869                 status = nfs_readdir_page_array_append(page, entry, &cookie);
 870         } while (!status && !entry->eof);
 871
 872         switch (status) {
 873         case -EBADCOOKIE:
 874                 if (!entry->eof)
 875                         break;
 876                 nfs_readdir_page_set_eof(page);
 877                 fallthrough;
 878         case -EAGAIN:
 879                 status = 0;
 880                 break;
 881         case -ENOSPC:
 882                 status = 0;
 883                 if (!desc->plus)
 884                         break;
 885                 while (!nfs_readdir_entry_decode(desc, entry, &stream))
 886                         ;
 887         }
 888
 889         if (page != *arrays)
 890                 nfs_readdir_page_unlock_and_put(page);
 891
 892         put_page(scratch);
 893         return status;
 894 }
 895
 896 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
 897 {
 898         while (npages--)
 899                 put_page(pages[npages]);
 900         kfree(pages);
 901 }
 902
 903 /*
 904  * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
 905  * to nfs_readdir_free_pages()
 906  */
 907 static struct page **nfs_readdir_alloc_pages(size_t npages)
 908 {
 909         struct page **pages;
 910         size_t i;
 911
 912         pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
 913         if (!pages)
 914                 return NULL;
 915         for (i = 0; i < npages; i++) {
 916                 struct page *page = alloc_page(GFP_KERNEL);
 917                 if (page == NULL)
 918                         goto out_freepages;
 919                 pages[i] = page;
 920         }
 921         return pages;
 922
 923 out_freepages:
 924         nfs_readdir_free_pages(pages, i);
 925         return NULL;
 926 }
 927
 928 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
 929                                     __be32 *verf_arg, __be32 *verf_res,
 930                                     struct page **arrays, size_t narrays)
 931 {
 932         u64 change_attr;
 933         struct page **pages;
 934         struct page *page = *arrays;
 935         struct nfs_entry *entry;
 936         size_t array_size;
 937         struct inode *inode = file_inode(desc->file);
 938         unsigned int dtsize = desc->dtsize;
 939         unsigned int pglen;
 940         int status = -ENOMEM;
 941
 942         entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 943         if (!entry)
 944                 return -ENOMEM;
 945         entry->cookie = nfs_readdir_page_last_cookie(page);
 946         entry->fh = nfs_alloc_fhandle();
 947         entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
 948         entry->server = NFS_SERVER(inode);
 949         if (entry->fh == NULL || entry->fattr == NULL)
 950                 goto out;
 951
 952         array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
 953         pages = nfs_readdir_alloc_pages(array_size);
 954         if (!pages)
 955                 goto out;
 956
 957         change_attr = inode_peek_iversion_raw(inode);
 958         status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
 959                                         dtsize, verf_res);
 960         if (status < 0)
 961                 goto free_pages;
 962
 963         pglen = status;
 964         if (pglen != 0)
 965                 status = nfs_readdir_page_filler(desc, entry, pages, pglen,
 966                                                  arrays, narrays, change_attr);
 967         else
 968                 nfs_readdir_page_set_eof(page);
 969         desc->buffer_fills++;
 970
 971 free_pages:
 972         nfs_readdir_free_pages(pages, array_size);
 973 out:
 974         nfs_free_fattr(entry->fattr);
 975         nfs_free_fhandle(entry->fh);
 976         kfree(entry);
 977         return status;
 978 }
 979
 980 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
 981 {
 982         put_page(desc->page);
 983         desc->page = NULL;
 984 }
 985
 986 static void
 987 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
 988 {
 989         unlock_page(desc->page);
 990         nfs_readdir_page_put(desc);
 991 }
 992
 993 static struct page *
 994 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
 995 {
 996         struct address_space *mapping = desc->file->f_mapping;
 997         u64 change_attr = inode_peek_iversion_raw(mapping->host);
 998         u64 cookie = desc->last_cookie;
 999         struct page *page;
1000
1001         page = nfs_readdir_page_get_locked(mapping, cookie, change_attr);
1002         if (!page)
1003                 return NULL;
1004         if (desc->clear_cache && !nfs_readdir_page_needs_filling(page))
1005                 nfs_readdir_page_reinit_array(page, cookie, change_attr);
1006         return page;
1007 }
1008
1009 /*
1010  * Returns 0 if desc->dir_cookie was found on page desc->page_index
1011  * and locks the page to prevent removal from the page cache.
1012  */
1013 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1014 {
1015         struct inode *inode = file_inode(desc->file);
1016         struct nfs_inode *nfsi = NFS_I(inode);
1017         __be32 verf[NFS_DIR_VERIFIER_SIZE];
1018         int res;
1019
1020         desc->page = nfs_readdir_page_get_cached(desc);
1021         if (!desc->page)
1022                 return -ENOMEM;
1023         if (nfs_readdir_page_needs_filling(desc->page)) {
1024                 /* Grow the dtsize if we had to go back for more pages */
1025                 if (desc->page_index == desc->page_index_max)
1026                         nfs_grow_dtsize(desc);
1027                 desc->page_index_max = desc->page_index;
1028                 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1029                                              desc->last_cookie,
1030                                              desc->page->index, desc->dtsize);
1031                 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1032                                                &desc->page, 1);
1033                 if (res < 0) {
1034                         nfs_readdir_page_unlock_and_put_cached(desc);
1035                         trace_nfs_readdir_cache_fill_done(inode, res);
1036                         if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1037                                 invalidate_inode_pages2(desc->file->f_mapping);
1038                                 nfs_readdir_rewind_search(desc);
1039                                 trace_nfs_readdir_invalidate_cache_range(
1040                                         inode, 0, MAX_LFS_FILESIZE);
1041                                 return -EAGAIN;
1042                         }
1043                         return res;
1044                 }
1045                 /*
1046                  * Set the cookie verifier if the page cache was empty
1047                  */
1048                 if (desc->last_cookie == 0 &&
1049                     memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1050                         memcpy(nfsi->cookieverf, verf,
1051                                sizeof(nfsi->cookieverf));
1052                         invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1053                                                       -1);
1054                         trace_nfs_readdir_invalidate_cache_range(
1055                                 inode, 1, MAX_LFS_FILESIZE);
1056                 }
1057                 desc->clear_cache = false;
1058         }
1059         res = nfs_readdir_search_array(desc);
1060         if (res == 0)
1061                 return 0;
1062         nfs_readdir_page_unlock_and_put_cached(desc);
1063         return res;
1064 }
1065
1066 /* Search for desc->dir_cookie from the beginning of the page cache */
1067 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1068 {
1069         int res;
1070
1071         do {
1072                 res = find_and_lock_cache_page(desc);
1073         } while (res == -EAGAIN);
1074         return res;
1075 }
1076
1077 #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1078
1079 /*
1080  * Once we've found the start of the dirent within a page: fill 'er up...
1081  */
1082 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1083                            const __be32 *verf)
1084 {
1085         struct file     *file = desc->file;
1086         struct nfs_cache_array *array;
1087         unsigned int i;
1088         bool first_emit = !desc->dir_cookie;
1089
1090         array = kmap_local_page(desc->page);
1091         for (i = desc->cache_entry_index; i < array->size; i++) {
1092                 struct nfs_cache_array_entry *ent;
1093
1094                 ent = &array->array[i];
1095                 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1096                     nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1097                         desc->eob = true;
1098                         break;
1099                 }
1100                 memcpy(desc->verf, verf, sizeof(desc->verf));
1101                 if (i == array->size - 1) {
1102                         desc->dir_cookie = array->last_cookie;
1103                         nfs_readdir_seek_next_array(array, desc);
1104                 } else {
1105                         desc->dir_cookie = array->array[i + 1].cookie;
1106                         desc->last_cookie = array->array[0].cookie;
1107                 }
1108                 if (nfs_readdir_use_cookie(file))
1109                         desc->ctx->pos = desc->dir_cookie;
1110                 else
1111                         desc->ctx->pos++;
1112                 if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 1) {
1113                         desc->eob = true;
1114                         break;
1115                 }
1116         }
1117         if (array->page_is_eof)
1118                 desc->eof = !desc->eob;
1119
1120         kunmap_local(array);
1121         dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1122                         (unsigned long long)desc->dir_cookie);
1123 }
1124
1125 /*
1126  * If we cannot find a cookie in our cache, we suspect that this is
1127  * because it points to a deleted file, so we ask the server to return
1128  * whatever it thinks is the next entry. We then feed this to filldir.
1129  * If all goes well, we should then be able to find our way round the
1130  * cache on the next call to readdir_search_pagecache();
1131  *
1132  * NOTE: we cannot add the anonymous page to the pagecache because
1133  *       the data it contains might not be page aligned. Besides,
1134  *       we should already have a complete representation of the
1135  *       directory in the page cache by the time we get here.
1136  */
1137 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1138 {
1139         struct page     **arrays;
1140         size_t          i, sz = 512;
1141         __be32          verf[NFS_DIR_VERIFIER_SIZE];
1142         int             status = -ENOMEM;
1143
1144         dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1145                         (unsigned long long)desc->dir_cookie);
1146
1147         arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1148         if (!arrays)
1149                 goto out;
1150         arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1151         if (!arrays[0])
1152                 goto out;
1153
1154         desc->page_index = 0;
1155         desc->cache_entry_index = 0;
1156         desc->last_cookie = desc->dir_cookie;
1157         desc->page_index_max = 0;
1158
1159         trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1160                                    -1, desc->dtsize);
1161
1162         status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1163         if (status < 0) {
1164                 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1165                 goto out_free;
1166         }
1167
1168         for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1169                 desc->page = arrays[i];
1170                 nfs_do_filldir(desc, verf);
1171         }
1172         desc->page = NULL;
1173
1174         /*
1175          * Grow the dtsize if we have to go back for more pages,
1176          * or shrink it if we're reading too many.
1177          */
1178         if (!desc->eof) {
1179                 if (!desc->eob)
1180                         nfs_grow_dtsize(desc);
1181                 else if (desc->buffer_fills == 1 &&
1182                          i < (desc->page_index_max >> 1))
1183                         nfs_shrink_dtsize(desc);
1184         }
1185 out_free:
1186         for (i = 0; i < sz && arrays[i]; i++)
1187                 nfs_readdir_page_array_free(arrays[i]);
1188 out:
1189         if (!nfs_readdir_use_cookie(desc->file))
1190                 nfs_readdir_rewind_search(desc);
1191         desc->page_index_max = -1;
1192         kfree(arrays);
1193         dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1194         return status;
1195 }
1196
1197 static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1198                                             struct nfs_readdir_descriptor *desc,
1199                                             unsigned int cache_misses,
1200                                             bool force_clear)
1201 {
1202         if (desc->ctx->pos == 0 || !desc->plus)
1203                 return false;
1204         if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1205                 return false;
1206         trace_nfs_readdir_force_readdirplus(inode);
1207         return true;
1208 }
1209
1210 /* The file offset position represents the dirent entry number.  A
1211    last cookie cache takes care of the common case of reading the
1212    whole directory.
1213  */
1214 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1215 {
1216         struct dentry   *dentry = file_dentry(file);
1217         struct inode    *inode = d_inode(dentry);
1218         struct nfs_inode *nfsi = NFS_I(inode);
1219         struct nfs_open_dir_context *dir_ctx = file->private_data;
1220         struct nfs_readdir_descriptor *desc;
1221         unsigned int cache_hits, cache_misses;
1222         bool force_clear;
1223         int res;
1224
1225         dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1226                         file, (long long)ctx->pos);
1227         nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1228
1229         /*
1230          * ctx->pos points to the dirent entry number.
1231          * *desc->dir_cookie has the cookie for the next entry. We have
1232          * to either find the entry with the appropriate number or
1233          * revalidate the cookie.
1234          */
1235         nfs_revalidate_mapping(inode, file->f_mapping);
1236
1237         res = -ENOMEM;
1238         desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1239         if (!desc)
1240                 goto out;
1241         desc->file = file;
1242         desc->ctx = ctx;
1243         desc->page_index_max = -1;
1244
1245         spin_lock(&file->f_lock);
1246         desc->dir_cookie = dir_ctx->dir_cookie;
1247         desc->page_index = dir_ctx->page_index;
1248         desc->last_cookie = dir_ctx->last_cookie;
1249         desc->attr_gencount = dir_ctx->attr_gencount;
1250         desc->eof = dir_ctx->eof;
1251         nfs_set_dtsize(desc, dir_ctx->dtsize);
1252         memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1253         cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1254         cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1255         force_clear = dir_ctx->force_clear;
1256         spin_unlock(&file->f_lock);
1257
1258         if (desc->eof) {
1259                 res = 0;
1260                 goto out_free;
1261         }
1262
1263         desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1264         force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1265                                                       force_clear);
1266         desc->clear_cache = force_clear;
1267
1268         do {
1269                 res = readdir_search_pagecache(desc);
1270
1271                 if (res == -EBADCOOKIE) {
1272                         res = 0;
1273                         /* This means either end of directory */
1274                         if (desc->dir_cookie && !desc->eof) {
1275                                 /* Or that the server has 'lost' a cookie */
1276                                 res = uncached_readdir(desc);
1277                                 if (res == 0)
1278                                         continue;
1279                                 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1280                                         res = 0;
1281                         }
1282                         break;
1283                 }
1284                 if (res == -ETOOSMALL && desc->plus) {
1285                         nfs_zap_caches(inode);
1286                         desc->plus = false;
1287                         desc->eof = false;
1288                         continue;
1289                 }
1290                 if (res < 0)
1291                         break;
1292
1293                 nfs_do_filldir(desc, nfsi->cookieverf);
1294                 nfs_readdir_page_unlock_and_put_cached(desc);
1295                 if (desc->page_index == desc->page_index_max)
1296                         desc->clear_cache = force_clear;
1297         } while (!desc->eob && !desc->eof);
1298
1299         spin_lock(&file->f_lock);
1300         dir_ctx->dir_cookie = desc->dir_cookie;
1301         dir_ctx->last_cookie = desc->last_cookie;
1302         dir_ctx->attr_gencount = desc->attr_gencount;
1303         dir_ctx->page_index = desc->page_index;
1304         dir_ctx->force_clear = force_clear;
1305         dir_ctx->eof = desc->eof;
1306         dir_ctx->dtsize = desc->dtsize;
1307         memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1308         spin_unlock(&file->f_lock);
1309 out_free:
1310         kfree(desc);
1311
1312 out:
1313         dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1314         return res;
1315 }
1316
1317 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1318 {
1319         struct nfs_open_dir_context *dir_ctx = filp->private_data;
1320
1321         dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1322                         filp, offset, whence);
1323
1324         switch (whence) {
1325         default:
1326                 return -EINVAL;
1327         case SEEK_SET:
1328                 if (offset < 0)
1329                         return -EINVAL;
1330                 spin_lock(&filp->f_lock);
1331                 break;
1332         case SEEK_CUR:
1333                 if (offset == 0)
1334                         return filp->f_pos;
1335                 spin_lock(&filp->f_lock);
1336                 offset += filp->f_pos;
1337                 if (offset < 0) {
1338                         spin_unlock(&filp->f_lock);
1339                         return -EINVAL;
1340                 }
1341         }
1342         if (offset != filp->f_pos) {
1343                 filp->f_pos = offset;
1344                 dir_ctx->page_index = 0;
1345                 if (!nfs_readdir_use_cookie(filp)) {
1346                         dir_ctx->dir_cookie = 0;
1347                         dir_ctx->last_cookie = 0;
1348                 } else {
1349                         dir_ctx->dir_cookie = offset;
1350                         dir_ctx->last_cookie = offset;
1351                 }
1352                 dir_ctx->eof = false;
1353         }
1354         spin_unlock(&filp->f_lock);
1355         return offset;
1356 }
1357
1358 /*
1359  * All directory operations under NFS are synchronous, so fsync()
1360  * is a dummy operation.
1361  */
1362 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1363                          int datasync)
1364 {
1365         dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1366
1367         nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1368         return 0;
1369 }
1370
1371 /**
1372  * nfs_force_lookup_revalidate - Mark the directory as having changed
1373  * @dir: pointer to directory inode
1374  *
1375  * This forces the revalidation code in nfs_lookup_revalidate() to do a
1376  * full lookup on all child dentries of 'dir' whenever a change occurs
1377  * on the server that might have invalidated our dcache.
1378  *
1379  * Note that we reserve bit '0' as a tag to let us know when a dentry
1380  * was revalidated while holding a delegation on its inode.
1381  *
1382  * The caller should be holding dir->i_lock
1383  */
1384 void nfs_force_lookup_revalidate(struct inode *dir)
1385 {
1386         NFS_I(dir)->cache_change_attribute += 2;
1387 }
1388 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1389
1390 /**
1391  * nfs_verify_change_attribute - Detects NFS remote directory changes
1392  * @dir: pointer to parent directory inode
1393  * @verf: previously saved change attribute
1394  *
1395  * Return "false" if the verifiers doesn't match the change attribute.
1396  * This would usually indicate that the directory contents have changed on
1397  * the server, and that any dentries need revalidating.
1398  */
1399 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1400 {
1401         return (verf & ~1UL) == nfs_save_change_attribute(dir);
1402 }
1403
1404 static void nfs_set_verifier_delegated(unsigned long *verf)
1405 {
1406         *verf |= 1UL;
1407 }
1408
1409 #if IS_ENABLED(CONFIG_NFS_V4)
1410 static void nfs_unset_verifier_delegated(unsigned long *verf)
1411 {
1412         *verf &= ~1UL;
1413 }
1414 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1415
1416 static bool nfs_test_verifier_delegated(unsigned long verf)
1417 {
1418         return verf & 1;
1419 }
1420
1421 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1422 {
1423         return nfs_test_verifier_delegated(dentry->d_time);
1424 }
1425
1426 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1427 {
1428         struct inode *inode = d_inode(dentry);
1429         struct inode *dir = d_inode(dentry->d_parent);
1430
1431         if (!nfs_verify_change_attribute(dir, verf))
1432                 return;
1433         if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1434                 nfs_set_verifier_delegated(&verf);
1435         dentry->d_time = verf;
1436 }
1437
1438 /**
1439  * nfs_set_verifier - save a parent directory verifier in the dentry
1440  * @dentry: pointer to dentry
1441  * @verf: verifier to save
1442  *
1443  * Saves the parent directory verifier in @dentry. If the inode has
1444  * a delegation, we also tag the dentry as having been revalidated
1445  * while holding a delegation so that we know we don't have to
1446  * look it up again after a directory change.
1447  */
1448 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1449 {
1450
1451         spin_lock(&dentry->d_lock);
1452         nfs_set_verifier_locked(dentry, verf);
1453         spin_unlock(&dentry->d_lock);
1454 }
1455 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1456
1457 #if IS_ENABLED(CONFIG_NFS_V4)
1458 /**
1459  * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1460  * @inode: pointer to inode
1461  *
1462  * Iterates through the dentries in the inode alias list and clears
1463  * the tag used to indicate that the dentry has been revalidated
1464  * while holding a delegation.
1465  * This function is intended for use when the delegation is being
1466  * returned or revoked.
1467  */
1468 void nfs_clear_verifier_delegated(struct inode *inode)
1469 {
1470         struct dentry *alias;
1471
1472         if (!inode)
1473                 return;
1474         spin_lock(&inode->i_lock);
1475         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1476                 spin_lock(&alias->d_lock);
1477                 nfs_unset_verifier_delegated(&alias->d_time);
1478                 spin_unlock(&alias->d_lock);
1479         }
1480         spin_unlock(&inode->i_lock);
1481 }
1482 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1483 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1484
1485 static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1486 {
1487         if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1488             d_really_is_negative(dentry))
1489                 return dentry->d_time == inode_peek_iversion_raw(dir);
1490         return nfs_verify_change_attribute(dir, dentry->d_time);
1491 }
1492
1493 /*
1494  * A check for whether or not the parent directory has changed.
1495  * In the case it has, we assume that the dentries are untrustworthy
1496  * and may need to be looked up again.
1497  * If rcu_walk prevents us from performing a full check, return 0.
1498  */
1499 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1500                               int rcu_walk)
1501 {
1502         if (IS_ROOT(dentry))
1503                 return 1;
1504         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1505                 return 0;
1506         if (!nfs_dentry_verify_change(dir, dentry))
1507                 return 0;
1508         /* Revalidate nfsi->cache_change_attribute before we declare a match */
1509         if (nfs_mapping_need_revalidate_inode(dir)) {
1510                 if (rcu_walk)
1511                         return 0;
1512                 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1513                         return 0;
1514         }
1515         if (!nfs_dentry_verify_change(dir, dentry))
1516                 return 0;
1517         return 1;
1518 }
1519
1520 /*
1521  * Use intent information to check whether or not we're going to do
1522  * an O_EXCL create using this path component.
1523  */
1524 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1525 {
1526         if (NFS_PROTO(dir)->version == 2)
1527                 return 0;
1528         return flags & LOOKUP_EXCL;
1529 }
1530
1531 /*
1532  * Inode and filehandle revalidation for lookups.
1533  *
1534  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1535  * or if the intent information indicates that we're about to open this
1536  * particular file and the "nocto" mount flag is not set.
1537  *
1538  */
1539 static
1540 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1541 {
1542         struct nfs_server *server = NFS_SERVER(inode);
1543         int ret;
1544
1545         if (IS_AUTOMOUNT(inode))
1546                 return 0;
1547
1548         if (flags & LOOKUP_OPEN) {
1549                 switch (inode->i_mode & S_IFMT) {
1550                 case S_IFREG:
1551                         /* A NFSv4 OPEN will revalidate later */
1552                         if (server->caps & NFS_CAP_ATOMIC_OPEN)
1553                                 goto out;
1554                         fallthrough;
1555                 case S_IFDIR:
1556                         if (server->flags & NFS_MOUNT_NOCTO)
1557                                 break;
1558                         /* NFS close-to-open cache consistency validation */
1559                         goto out_force;
1560                 }
1561         }
1562
1563         /* VFS wants an on-the-wire revalidation */
1564         if (flags & LOOKUP_REVAL)
1565                 goto out_force;
1566 out:
1567         if (inode->i_nlink > 0 ||
1568             (inode->i_nlink == 0 &&
1569              test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1570                 return 0;
1571         else
1572                 return -ESTALE;
1573 out_force:
1574         if (flags & LOOKUP_RCU)
1575                 return -ECHILD;
1576         ret = __nfs_revalidate_inode(server, inode);
1577         if (ret != 0)
1578                 return ret;
1579         goto out;
1580 }
1581
1582 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1583 {
1584         spin_lock(&inode->i_lock);
1585         nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1586         spin_unlock(&inode->i_lock);
1587 }
1588
1589 /*
1590  * We judge how long we want to trust negative
1591  * dentries by looking at the parent inode mtime.
1592  *
1593  * If parent mtime has changed, we revalidate, else we wait for a
1594  * period corresponding to the parent's attribute cache timeout value.
1595  *
1596  * If LOOKUP_RCU prevents us from performing a full check, return 1
1597  * suggesting a reval is needed.
1598  *
1599  * Note that when creating a new file, or looking up a rename target,
1600  * then it shouldn't be necessary to revalidate a negative dentry.
1601  */
1602 static inline
1603 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1604                        unsigned int flags)
1605 {
1606         if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1607                 return 0;
1608         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1609                 return 1;
1610         /* Case insensitive server? Revalidate negative dentries */
1611         if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1612                 return 1;
1613         return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1614 }
1615
1616 static int
1617 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1618                            struct inode *inode, int error)
1619 {
1620         switch (error) {
1621         case 1:
1622                 break;
1623         case 0:
1624                 /*
1625                  * We can't d_drop the root of a disconnected tree:
1626                  * its d_hash is on the s_anon list and d_drop() would hide
1627                  * it from shrink_dcache_for_unmount(), leading to busy
1628                  * inodes on unmount and further oopses.
1629                  */
1630                 if (inode && IS_ROOT(dentry))
1631                         error = 1;
1632                 break;
1633         }
1634         trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1635         return error;
1636 }
1637
1638 static int
1639 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1640                                unsigned int flags)
1641 {
1642         int ret = 1;
1643         if (nfs_neg_need_reval(dir, dentry, flags)) {
1644                 if (flags & LOOKUP_RCU)
1645                         return -ECHILD;
1646                 ret = 0;
1647         }
1648         return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1649 }
1650
1651 static int
1652 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1653                                 struct inode *inode)
1654 {
1655         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1656         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1657 }
1658
1659 static int nfs_lookup_revalidate_dentry(struct inode *dir,
1660                                         struct dentry *dentry,
1661                                         struct inode *inode, unsigned int flags)
1662 {
1663         struct nfs_fh *fhandle;
1664         struct nfs_fattr *fattr;
1665         unsigned long dir_verifier;
1666         int ret;
1667
1668         trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1669
1670         ret = -ENOMEM;
1671         fhandle = nfs_alloc_fhandle();
1672         fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1673         if (fhandle == NULL || fattr == NULL)
1674                 goto out;
1675
1676         dir_verifier = nfs_save_change_attribute(dir);
1677         ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1678         if (ret < 0) {
1679                 switch (ret) {
1680                 case -ESTALE:
1681                 case -ENOENT:
1682                         ret = 0;
1683                         break;
1684                 case -ETIMEDOUT:
1685                         if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1686                                 ret = 1;
1687                 }
1688                 goto out;
1689         }
1690
1691         /* Request help from readdirplus */
1692         nfs_lookup_advise_force_readdirplus(dir, flags);
1693
1694         ret = 0;
1695         if (nfs_compare_fh(NFS_FH(inode), fhandle))
1696                 goto out;
1697         if (nfs_refresh_inode(inode, fattr) < 0)
1698                 goto out;
1699
1700         nfs_setsecurity(inode, fattr);
1701         nfs_set_verifier(dentry, dir_verifier);
1702
1703         ret = 1;
1704 out:
1705         nfs_free_fattr(fattr);
1706         nfs_free_fhandle(fhandle);
1707
1708         /*
1709          * If the lookup failed despite the dentry change attribute being
1710          * a match, then we should revalidate the directory cache.
1711          */
1712         if (!ret && nfs_dentry_verify_change(dir, dentry))
1713                 nfs_mark_dir_for_revalidate(dir);
1714         return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1715 }
1716
1717 /*
1718  * This is called every time the dcache has a lookup hit,
1719  * and we should check whether we can really trust that
1720  * lookup.
1721  *
1722  * NOTE! The hit can be a negative hit too, don't assume
1723  * we have an inode!
1724  *
1725  * If the parent directory is seen to have changed, we throw out the
1726  * cached dentry and do a new lookup.
1727  */
1728 static int
1729 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1730                          unsigned int flags)
1731 {
1732         struct inode *inode;
1733         int error;
1734
1735         nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1736         inode = d_inode(dentry);
1737
1738         if (!inode)
1739                 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1740
1741         if (is_bad_inode(inode)) {
1742                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1743                                 __func__, dentry);
1744                 goto out_bad;
1745         }
1746
1747         if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1748             nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1749                 goto out_bad;
1750
1751         if (nfs_verifier_is_delegated(dentry))
1752                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1753
1754         /* Force a full look up iff the parent directory has changed */
1755         if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1756             nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1757                 error = nfs_lookup_verify_inode(inode, flags);
1758                 if (error) {
1759                         if (error == -ESTALE)
1760                                 nfs_mark_dir_for_revalidate(dir);
1761                         goto out_bad;
1762                 }
1763                 goto out_valid;
1764         }
1765
1766         if (flags & LOOKUP_RCU)
1767                 return -ECHILD;
1768
1769         if (NFS_STALE(inode))
1770                 goto out_bad;
1771
1772         return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1773 out_valid:
1774         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1775 out_bad:
1776         if (flags & LOOKUP_RCU)
1777                 return -ECHILD;
1778         return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1779 }
1780
1781 static int
1782 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1783                         int (*reval)(struct inode *, struct dentry *, unsigned int))
1784 {
1785         struct dentry *parent;
1786         struct inode *dir;
1787         int ret;
1788
1789         if (flags & LOOKUP_RCU) {
1790                 if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1791                         return -ECHILD;
1792                 parent = READ_ONCE(dentry->d_parent);
1793                 dir = d_inode_rcu(parent);
1794                 if (!dir)
1795                         return -ECHILD;
1796                 ret = reval(dir, dentry, flags);
1797                 if (parent != READ_ONCE(dentry->d_parent))
1798                         return -ECHILD;
1799         } else {
1800                 /* Wait for unlink to complete */
1801                 wait_var_event(&dentry->d_fsdata,
1802                                dentry->d_fsdata != NFS_FSDATA_BLOCKED);
1803                 parent = dget_parent(dentry);
1804                 ret = reval(d_inode(parent), dentry, flags);
1805                 dput(parent);
1806         }
1807         return ret;
1808 }
1809
1810 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1811 {
1812         return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1813 }
1814
1815 /*
1816  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1817  * when we don't really care about the dentry name. This is called when a
1818  * pathwalk ends on a dentry that was not found via a normal lookup in the
1819  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1820  *
1821  * In this situation, we just want to verify that the inode itself is OK
1822  * since the dentry might have changed on the server.
1823  */
1824 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1825 {
1826         struct inode *inode = d_inode(dentry);
1827         int error = 0;
1828
1829         /*
1830          * I believe we can only get a negative dentry here in the case of a
1831          * procfs-style symlink. Just assume it's correct for now, but we may
1832          * eventually need to do something more here.
1833          */
1834         if (!inode) {
1835                 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1836                                 __func__, dentry);
1837                 return 1;
1838         }
1839
1840         if (is_bad_inode(inode)) {
1841                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1842                                 __func__, dentry);
1843                 return 0;
1844         }
1845
1846         error = nfs_lookup_verify_inode(inode, flags);
1847         dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1848                         __func__, inode->i_ino, error ? "invalid" : "valid");
1849         return !error;
1850 }
1851
1852 /*
1853  * This is called from dput() when d_count is going to 0.
1854  */
1855 static int nfs_dentry_delete(const struct dentry *dentry)
1856 {
1857         dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1858                 dentry, dentry->d_flags);
1859
1860         /* Unhash any dentry with a stale inode */
1861         if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1862                 return 1;
1863
1864         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1865                 /* Unhash it, so that ->d_iput() would be called */
1866                 return 1;
1867         }
1868         if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1869                 /* Unhash it, so that ancestors of killed async unlink
1870                  * files will be cleaned up during umount */
1871                 return 1;
1872         }
1873         return 0;
1874
1875 }
1876
1877 /* Ensure that we revalidate inode->i_nlink */
1878 static void nfs_drop_nlink(struct inode *inode)
1879 {
1880         spin_lock(&inode->i_lock);
1881         /* drop the inode if we're reasonably sure this is the last link */
1882         if (inode->i_nlink > 0)
1883                 drop_nlink(inode);
1884         NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1885         nfs_set_cache_invalid(
1886                 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1887                                NFS_INO_INVALID_NLINK);
1888         spin_unlock(&inode->i_lock);
1889 }
1890
1891 /*
1892  * Called when the dentry loses inode.
1893  * We use it to clean up silly-renamed files.
1894  */
1895 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1896 {
1897         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1898                 nfs_complete_unlink(dentry, inode);
1899                 nfs_drop_nlink(inode);
1900         }
1901         iput(inode);
1902 }
1903
1904 static void nfs_d_release(struct dentry *dentry)
1905 {
1906         /* free cached devname value, if it survived that far */
1907         if (unlikely(dentry->d_fsdata)) {
1908                 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1909                         WARN_ON(1);
1910                 else
1911                         kfree(dentry->d_fsdata);
1912         }
1913 }
1914
1915 const struct dentry_operations nfs_dentry_operations = {
1916         .d_revalidate   = nfs_lookup_revalidate,
1917         .d_weak_revalidate      = nfs_weak_revalidate,
1918         .d_delete       = nfs_dentry_delete,
1919         .d_iput         = nfs_dentry_iput,
1920         .d_automount    = nfs_d_automount,
1921         .d_release      = nfs_d_release,
1922 };
1923 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1924
1925 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1926 {
1927         struct dentry *res;
1928         struct inode *inode = NULL;
1929         struct nfs_fh *fhandle = NULL;
1930         struct nfs_fattr *fattr = NULL;
1931         unsigned long dir_verifier;
1932         int error;
1933
1934         dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1935         nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1936
1937         if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1938                 return ERR_PTR(-ENAMETOOLONG);
1939
1940         /*
1941          * If we're doing an exclusive create, optimize away the lookup
1942          * but don't hash the dentry.
1943          */
1944         if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1945                 return NULL;
1946
1947         res = ERR_PTR(-ENOMEM);
1948         fhandle = nfs_alloc_fhandle();
1949         fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1950         if (fhandle == NULL || fattr == NULL)
1951                 goto out;
1952
1953         dir_verifier = nfs_save_change_attribute(dir);
1954         trace_nfs_lookup_enter(dir, dentry, flags);
1955         error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1956         if (error == -ENOENT) {
1957                 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1958                         dir_verifier = inode_peek_iversion_raw(dir);
1959                 goto no_entry;
1960         }
1961         if (error < 0) {
1962                 res = ERR_PTR(error);
1963                 goto out;
1964         }
1965         inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1966         res = ERR_CAST(inode);
1967         if (IS_ERR(res))
1968                 goto out;
1969
1970         /* Notify readdir to use READDIRPLUS */
1971         nfs_lookup_advise_force_readdirplus(dir, flags);
1972
1973 no_entry:
1974         res = d_splice_alias(inode, dentry);
1975         if (res != NULL) {
1976                 if (IS_ERR(res))
1977                         goto out;
1978                 dentry = res;
1979         }
1980         nfs_set_verifier(dentry, dir_verifier);
1981 out:
1982         trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1983         nfs_free_fattr(fattr);
1984         nfs_free_fhandle(fhandle);
1985         return res;
1986 }
1987 EXPORT_SYMBOL_GPL(nfs_lookup);
1988
1989 void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1990 {
1991         /* Case insensitive server? Revalidate dentries */
1992         if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1993                 d_prune_aliases(inode);
1994 }
1995 EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
1996
1997 #if IS_ENABLED(CONFIG_NFS_V4)
1998 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1999
2000 const struct dentry_operations nfs4_dentry_operations = {
2001         .d_revalidate   = nfs4_lookup_revalidate,
2002         .d_weak_revalidate      = nfs_weak_revalidate,
2003         .d_delete       = nfs_dentry_delete,
2004         .d_iput         = nfs_dentry_iput,
2005         .d_automount    = nfs_d_automount,
2006         .d_release      = nfs_d_release,
2007 };
2008 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2009
2010 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2011 {
2012         return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2013 }
2014
2015 static int do_open(struct inode *inode, struct file *filp)
2016 {
2017         nfs_fscache_open_file(inode, filp);
2018         return 0;
2019 }
2020
2021 static int nfs_finish_open(struct nfs_open_context *ctx,
2022                            struct dentry *dentry,
2023                            struct file *file, unsigned open_flags)
2024 {
2025         int err;
2026
2027         err = finish_open(file, dentry, do_open);
2028         if (err)
2029                 goto out;
2030         if (S_ISREG(file_inode(file)->i_mode))
2031                 nfs_file_set_open_context(file, ctx);
2032         else
2033                 err = -EOPENSTALE;
2034 out:
2035         return err;
2036 }
2037
2038 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2039                     struct file *file, unsigned open_flags,
2040                     umode_t mode)
2041 {
2042         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2043         struct nfs_open_context *ctx;
2044         struct dentry *res;
2045         struct iattr attr = { .ia_valid = ATTR_OPEN };
2046         struct inode *inode;
2047         unsigned int lookup_flags = 0;
2048         unsigned long dir_verifier;
2049         bool switched = false;
2050         int created = 0;
2051         int err;
2052
2053         /* Expect a negative dentry */
2054         BUG_ON(d_inode(dentry));
2055
2056         dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2057                         dir->i_sb->s_id, dir->i_ino, dentry);
2058
2059         err = nfs_check_flags(open_flags);
2060         if (err)
2061                 return err;
2062
2063         /* NFS only supports OPEN on regular files */
2064         if ((open_flags & O_DIRECTORY)) {
2065                 if (!d_in_lookup(dentry)) {
2066                         /*
2067                          * Hashed negative dentry with O_DIRECTORY: dentry was
2068                          * revalidated and is fine, no need to perform lookup
2069                          * again
2070                          */
2071                         return -ENOENT;
2072                 }
2073                 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2074                 goto no_open;
2075         }
2076
2077         if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2078                 return -ENAMETOOLONG;
2079
2080         if (open_flags & O_CREAT) {
2081                 struct nfs_server *server = NFS_SERVER(dir);
2082
2083                 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2084                         mode &= ~current_umask();
2085
2086                 attr.ia_valid |= ATTR_MODE;
2087                 attr.ia_mode = mode;
2088         }
2089         if (open_flags & O_TRUNC) {
2090                 attr.ia_valid |= ATTR_SIZE;
2091                 attr.ia_size = 0;
2092         }
2093
2094         if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2095                 d_drop(dentry);
2096                 switched = true;
2097                 dentry = d_alloc_parallel(dentry->d_parent,
2098                                           &dentry->d_name, &wq);
2099                 if (IS_ERR(dentry))
2100                         return PTR_ERR(dentry);
2101                 if (unlikely(!d_in_lookup(dentry)))
2102                         return finish_no_open(file, dentry);
2103         }
2104
2105         ctx = create_nfs_open_context(dentry, open_flags, file);
2106         err = PTR_ERR(ctx);
2107         if (IS_ERR(ctx))
2108                 goto out;
2109
2110         trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2111         inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2112         if (created)
2113                 file->f_mode |= FMODE_CREATED;
2114         if (IS_ERR(inode)) {
2115                 err = PTR_ERR(inode);
2116                 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2117                 put_nfs_open_context(ctx);
2118                 d_drop(dentry);
2119                 switch (err) {
2120                 case -ENOENT:
2121                         d_splice_alias(NULL, dentry);
2122                         if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2123                                 dir_verifier = inode_peek_iversion_raw(dir);
2124                         else
2125                                 dir_verifier = nfs_save_change_attribute(dir);
2126                         nfs_set_verifier(dentry, dir_verifier);
2127                         break;
2128                 case -EISDIR:
2129                 case -ENOTDIR:
2130                         goto no_open;
2131                 case -ELOOP:
2132                         if (!(open_flags & O_NOFOLLOW))
2133                                 goto no_open;
2134                         break;
2135                         /* case -EINVAL: */
2136                 default:
2137                         break;
2138                 }
2139                 goto out;
2140         }
2141         file->f_mode |= FMODE_CAN_ODIRECT;
2142
2143         err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2144         trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2145         put_nfs_open_context(ctx);
2146 out:
2147         if (unlikely(switched)) {
2148                 d_lookup_done(dentry);
2149                 dput(dentry);
2150         }
2151         return err;
2152
2153 no_open:
2154         res = nfs_lookup(dir, dentry, lookup_flags);
2155         if (!res) {
2156                 inode = d_inode(dentry);
2157                 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2158                     !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2159                         res = ERR_PTR(-ENOTDIR);
2160                 else if (inode && S_ISREG(inode->i_mode))
2161                         res = ERR_PTR(-EOPENSTALE);
2162         } else if (!IS_ERR(res)) {
2163                 inode = d_inode(res);
2164                 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2165                     !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2166                         dput(res);
2167                         res = ERR_PTR(-ENOTDIR);
2168                 } else if (inode && S_ISREG(inode->i_mode)) {
2169                         dput(res);
2170                         res = ERR_PTR(-EOPENSTALE);
2171                 }
2172         }
2173         if (switched) {
2174                 d_lookup_done(dentry);
2175                 if (!res)
2176                         res = dentry;
2177                 else
2178                         dput(dentry);
2179         }
2180         if (IS_ERR(res))
2181                 return PTR_ERR(res);
2182         return finish_no_open(file, res);
2183 }
2184 EXPORT_SYMBOL_GPL(nfs_atomic_open);
2185
2186 static int
2187 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2188                           unsigned int flags)
2189 {
2190         struct inode *inode;
2191
2192         if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2193                 goto full_reval;
2194         if (d_mountpoint(dentry))
2195                 goto full_reval;
2196
2197         inode = d_inode(dentry);
2198
2199         /* We can't create new files in nfs_open_revalidate(), so we
2200          * optimize away revalidation of negative dentries.
2201          */
2202         if (inode == NULL)
2203                 goto full_reval;
2204
2205         if (nfs_verifier_is_delegated(dentry))
2206                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2207
2208         /* NFS only supports OPEN on regular files */
2209         if (!S_ISREG(inode->i_mode))
2210                 goto full_reval;
2211
2212         /* We cannot do exclusive creation on a positive dentry */
2213         if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2214                 goto reval_dentry;
2215
2216         /* Check if the directory changed */
2217         if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2218                 goto reval_dentry;
2219
2220         /* Let f_op->open() actually open (and revalidate) the file */
2221         return 1;
2222 reval_dentry:
2223         if (flags & LOOKUP_RCU)
2224                 return -ECHILD;
2225         return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2226
2227 full_reval:
2228         return nfs_do_lookup_revalidate(dir, dentry, flags);
2229 }
2230
2231 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2232 {
2233         return __nfs_lookup_revalidate(dentry, flags,
2234                         nfs4_do_lookup_revalidate);
2235 }
2236
2237 #endif /* CONFIG_NFSV4 */
2238
2239 struct dentry *
2240 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2241                                 struct nfs_fattr *fattr)
2242 {
2243         struct dentry *parent = dget_parent(dentry);
2244         struct inode *dir = d_inode(parent);
2245         struct inode *inode;
2246         struct dentry *d;
2247         int error;
2248
2249         d_drop(dentry);
2250
2251         if (fhandle->size == 0) {
2252                 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2253                 if (error)
2254                         goto out_error;
2255         }
2256         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2257         if (!(fattr->valid & NFS_ATTR_FATTR)) {
2258                 struct nfs_server *server = NFS_SB(dentry->d_sb);
2259                 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2260                                 fattr, NULL);
2261                 if (error < 0)
2262                         goto out_error;
2263         }
2264         inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2265         d = d_splice_alias(inode, dentry);
2266 out:
2267         dput(parent);
2268         return d;
2269 out_error:
2270         d = ERR_PTR(error);
2271         goto out;
2272 }
2273 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2274
2275 /*
2276  * Code common to create, mkdir, and mknod.
2277  */
2278 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2279                                 struct nfs_fattr *fattr)
2280 {
2281         struct dentry *d;
2282
2283         d = nfs_add_or_obtain(dentry, fhandle, fattr);
2284         if (IS_ERR(d))
2285                 return PTR_ERR(d);
2286
2287         /* Callers don't care */
2288         dput(d);
2289         return 0;
2290 }
2291 EXPORT_SYMBOL_GPL(nfs_instantiate);
2292
2293 /*
2294  * Following a failed create operation, we drop the dentry rather
2295  * than retain a negative dentry. This avoids a problem in the event
2296  * that the operation succeeded on the server, but an error in the
2297  * reply path made it appear to have failed.
2298  */
2299 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2300                struct dentry *dentry, umode_t mode, bool excl)
2301 {
2302         struct iattr attr;
2303         int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2304         int error;
2305
2306         dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2307                         dir->i_sb->s_id, dir->i_ino, dentry);
2308
2309         attr.ia_mode = mode;
2310         attr.ia_valid = ATTR_MODE;
2311
2312         trace_nfs_create_enter(dir, dentry, open_flags);
2313         error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2314         trace_nfs_create_exit(dir, dentry, open_flags, error);
2315         if (error != 0)
2316                 goto out_err;
2317         return 0;
2318 out_err:
2319         d_drop(dentry);
2320         return error;
2321 }
2322 EXPORT_SYMBOL_GPL(nfs_create);
2323
2324 /*
2325  * See comments for nfs_proc_create regarding failed operations.
2326  */
2327 int
2328 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2329           struct dentry *dentry, umode_t mode, dev_t rdev)
2330 {
2331         struct iattr attr;
2332         int status;
2333
2334         dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2335                         dir->i_sb->s_id, dir->i_ino, dentry);
2336
2337         attr.ia_mode = mode;
2338         attr.ia_valid = ATTR_MODE;
2339
2340         trace_nfs_mknod_enter(dir, dentry);
2341         status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2342         trace_nfs_mknod_exit(dir, dentry, status);
2343         if (status != 0)
2344                 goto out_err;
2345         return 0;
2346 out_err:
2347         d_drop(dentry);
2348         return status;
2349 }
2350 EXPORT_SYMBOL_GPL(nfs_mknod);
2351
2352 /*
2353  * See comments for nfs_proc_create regarding failed operations.
2354  */
2355 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2356               struct dentry *dentry, umode_t mode)
2357 {
2358         struct iattr attr;
2359         int error;
2360
2361         dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2362                         dir->i_sb->s_id, dir->i_ino, dentry);
2363
2364         attr.ia_valid = ATTR_MODE;
2365         attr.ia_mode = mode | S_IFDIR;
2366
2367         trace_nfs_mkdir_enter(dir, dentry);
2368         error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2369         trace_nfs_mkdir_exit(dir, dentry, error);
2370         if (error != 0)
2371                 goto out_err;
2372         return 0;
2373 out_err:
2374         d_drop(dentry);
2375         return error;
2376 }
2377 EXPORT_SYMBOL_GPL(nfs_mkdir);
2378
2379 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2380 {
2381         if (simple_positive(dentry))
2382                 d_delete(dentry);
2383 }
2384
2385 static void nfs_dentry_remove_handle_error(struct inode *dir,
2386                                            struct dentry *dentry, int error)
2387 {
2388         switch (error) {
2389         case -ENOENT:
2390                 if (d_really_is_positive(dentry))
2391                         d_delete(dentry);
2392                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2393                 break;
2394         case 0:
2395                 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2396                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2397         }
2398 }
2399
2400 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2401 {
2402         int error;
2403
2404         dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2405                         dir->i_sb->s_id, dir->i_ino, dentry);
2406
2407         trace_nfs_rmdir_enter(dir, dentry);
2408         if (d_really_is_positive(dentry)) {
2409                 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2410                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2411                 /* Ensure the VFS deletes this inode */
2412                 switch (error) {
2413                 case 0:
2414                         clear_nlink(d_inode(dentry));
2415                         break;
2416                 case -ENOENT:
2417                         nfs_dentry_handle_enoent(dentry);
2418                 }
2419                 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2420         } else
2421                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2422         nfs_dentry_remove_handle_error(dir, dentry, error);
2423         trace_nfs_rmdir_exit(dir, dentry, error);
2424
2425         return error;
2426 }
2427 EXPORT_SYMBOL_GPL(nfs_rmdir);
2428
2429 /*
2430  * Remove a file after making sure there are no pending writes,
2431  * and after checking that the file has only one user.
2432  *
2433  * We invalidate the attribute cache and free the inode prior to the operation
2434  * to avoid possible races if the server reuses the inode.
2435  */
2436 static int nfs_safe_remove(struct dentry *dentry)
2437 {
2438         struct inode *dir = d_inode(dentry->d_parent);
2439         struct inode *inode = d_inode(dentry);
2440         int error = -EBUSY;
2441
2442         dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2443
2444         /* If the dentry was sillyrenamed, we simply call d_delete() */
2445         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2446                 error = 0;
2447                 goto out;
2448         }
2449
2450         trace_nfs_remove_enter(dir, dentry);
2451         if (inode != NULL) {
2452                 error = NFS_PROTO(dir)->remove(dir, dentry);
2453                 if (error == 0)
2454                         nfs_drop_nlink(inode);
2455         } else
2456                 error = NFS_PROTO(dir)->remove(dir, dentry);
2457         if (error == -ENOENT)
2458                 nfs_dentry_handle_enoent(dentry);
2459         trace_nfs_remove_exit(dir, dentry, error);
2460 out:
2461         return error;
2462 }
2463
2464 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
2465  *  belongs to an active ".nfs..." file and we return -EBUSY.
2466  *
2467  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
2468  */
2469 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2470 {
2471         int error;
2472
2473         dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2474                 dir->i_ino, dentry);
2475
2476         trace_nfs_unlink_enter(dir, dentry);
2477         spin_lock(&dentry->d_lock);
2478         if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2479                                              &NFS_I(d_inode(dentry))->flags)) {
2480                 spin_unlock(&dentry->d_lock);
2481                 /* Start asynchronous writeout of the inode */
2482                 write_inode_now(d_inode(dentry), 0);
2483                 error = nfs_sillyrename(dir, dentry);
2484                 goto out;
2485         }
2486         /* We must prevent any concurrent open until the unlink
2487          * completes.  ->d_revalidate will wait for ->d_fsdata
2488          * to clear.  We set it here to ensure no lookup succeeds until
2489          * the unlink is complete on the server.
2490          */
2491         error = -ETXTBSY;
2492         if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2493             WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2494                 spin_unlock(&dentry->d_lock);
2495                 goto out;
2496         }
2497         /* old devname */
2498         kfree(dentry->d_fsdata);
2499         dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2500
2501         spin_unlock(&dentry->d_lock);
2502         error = nfs_safe_remove(dentry);
2503         nfs_dentry_remove_handle_error(dir, dentry, error);
2504         dentry->d_fsdata = NULL;
2505         wake_up_var(&dentry->d_fsdata);
2506 out:
2507         trace_nfs_unlink_exit(dir, dentry, error);
2508         return error;
2509 }
2510 EXPORT_SYMBOL_GPL(nfs_unlink);
2511
2512 /*
2513  * To create a symbolic link, most file systems instantiate a new inode,
2514  * add a page to it containing the path, then write it out to the disk
2515  * using prepare_write/commit_write.
2516  *
2517  * Unfortunately the NFS client can't create the in-core inode first
2518  * because it needs a file handle to create an in-core inode (see
2519  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
2520  * symlink request has completed on the server.
2521  *
2522  * So instead we allocate a raw page, copy the symname into it, then do
2523  * the SYMLINK request with the page as the buffer.  If it succeeds, we
2524  * now have a new file handle and can instantiate an in-core NFS inode
2525  * and move the raw page into its mapping.
2526  */
2527 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2528                 struct dentry *dentry, const char *symname)
2529 {
2530         struct page *page;
2531         char *kaddr;
2532         struct iattr attr;
2533         unsigned int pathlen = strlen(symname);
2534         int error;
2535
2536         dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2537                 dir->i_ino, dentry, symname);
2538
2539         if (pathlen > PAGE_SIZE)
2540                 return -ENAMETOOLONG;
2541
2542         attr.ia_mode = S_IFLNK | S_IRWXUGO;
2543         attr.ia_valid = ATTR_MODE;
2544
2545         page = alloc_page(GFP_USER);
2546         if (!page)
2547                 return -ENOMEM;
2548
2549         kaddr = page_address(page);
2550         memcpy(kaddr, symname, pathlen);
2551         if (pathlen < PAGE_SIZE)
2552                 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2553
2554         trace_nfs_symlink_enter(dir, dentry);
2555         error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2556         trace_nfs_symlink_exit(dir, dentry, error);
2557         if (error != 0) {
2558                 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2559                         dir->i_sb->s_id, dir->i_ino,
2560                         dentry, symname, error);
2561                 d_drop(dentry);
2562                 __free_page(page);
2563                 return error;
2564         }
2565
2566         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2567
2568         /*
2569          * No big deal if we can't add this page to the page cache here.
2570          * READLINK will get the missing page from the server if needed.
2571          */
2572         if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2573                                                         GFP_KERNEL)) {
2574                 SetPageUptodate(page);
2575                 unlock_page(page);
2576                 /*
2577                  * add_to_page_cache_lru() grabs an extra page refcount.
2578                  * Drop it here to avoid leaking this page later.
2579                  */
2580                 put_page(page);
2581         } else
2582                 __free_page(page);
2583
2584         return 0;
2585 }
2586 EXPORT_SYMBOL_GPL(nfs_symlink);
2587
2588 int
2589 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2590 {
2591         struct inode *inode = d_inode(old_dentry);
2592         int error;
2593
2594         dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2595                 old_dentry, dentry);
2596
2597         trace_nfs_link_enter(inode, dir, dentry);
2598         d_drop(dentry);
2599         if (S_ISREG(inode->i_mode))
2600                 nfs_sync_inode(inode);
2601         error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2602         if (error == 0) {
2603                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2604                 ihold(inode);
2605                 d_add(dentry, inode);
2606         }
2607         trace_nfs_link_exit(inode, dir, dentry, error);
2608         return error;
2609 }
2610 EXPORT_SYMBOL_GPL(nfs_link);
2611
2612 static void
2613 nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2614 {
2615         struct dentry *new_dentry = data->new_dentry;
2616
2617         new_dentry->d_fsdata = NULL;
2618         wake_up_var(&new_dentry->d_fsdata);
2619 }
2620
2621 /*
2622  * RENAME
2623  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2624  * different file handle for the same inode after a rename (e.g. when
2625  * moving to a different directory). A fail-safe method to do so would
2626  * be to look up old_dir/old_name, create a link to new_dir/new_name and
2627  * rename the old file using the sillyrename stuff. This way, the original
2628  * file in old_dir will go away when the last process iput()s the inode.
2629  *
2630  * FIXED.
2631  *
2632  * It actually works quite well. One needs to have the possibility for
2633  * at least one ".nfs..." file in each directory the file ever gets
2634  * moved or linked to which happens automagically with the new
2635  * implementation that only depends on the dcache stuff instead of
2636  * using the inode layer
2637  *
2638  * Unfortunately, things are a little more complicated than indicated
2639  * above. For a cross-directory move, we want to make sure we can get
2640  * rid of the old inode after the operation.  This means there must be
2641  * no pending writes (if it's a file), and the use count must be 1.
2642  * If these conditions are met, we can drop the dentries before doing
2643  * the rename.
2644  */
2645 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2646                struct dentry *old_dentry, struct inode *new_dir,
2647                struct dentry *new_dentry, unsigned int flags)
2648 {
2649         struct inode *old_inode = d_inode(old_dentry);
2650         struct inode *new_inode = d_inode(new_dentry);
2651         struct dentry *dentry = NULL;
2652         struct rpc_task *task;
2653         bool must_unblock = false;
2654         int error = -EBUSY;
2655
2656         if (flags)
2657                 return -EINVAL;
2658
2659         dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2660                  old_dentry, new_dentry,
2661                  d_count(new_dentry));
2662
2663         trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2664         /*
2665          * For non-directories, check whether the target is busy and if so,
2666          * make a copy of the dentry and then do a silly-rename. If the
2667          * silly-rename succeeds, the copied dentry is hashed and becomes
2668          * the new target.
2669          */
2670         if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2671                 /* We must prevent any concurrent open until the unlink
2672                  * completes.  ->d_revalidate will wait for ->d_fsdata
2673                  * to clear.  We set it here to ensure no lookup succeeds until
2674                  * the unlink is complete on the server.
2675                  */
2676                 error = -ETXTBSY;
2677                 if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2678                     WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2679                         goto out;
2680                 if (new_dentry->d_fsdata) {
2681                         /* old devname */
2682                         kfree(new_dentry->d_fsdata);
2683                         new_dentry->d_fsdata = NULL;
2684                 }
2685
2686                 spin_lock(&new_dentry->d_lock);
2687                 if (d_count(new_dentry) > 2) {
2688                         int err;
2689
2690                         spin_unlock(&new_dentry->d_lock);
2691
2692                         /* copy the target dentry's name */
2693                         dentry = d_alloc(new_dentry->d_parent,
2694                                          &new_dentry->d_name);
2695                         if (!dentry)
2696                                 goto out;
2697
2698                         /* silly-rename the existing target ... */
2699                         err = nfs_sillyrename(new_dir, new_dentry);
2700                         if (err)
2701                                 goto out;
2702
2703                         new_dentry = dentry;
2704                         new_inode = NULL;
2705                 } else {
2706                         new_dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2707                         must_unblock = true;
2708                         spin_unlock(&new_dentry->d_lock);
2709                 }
2710
2711         }
2712
2713         if (S_ISREG(old_inode->i_mode))
2714                 nfs_sync_inode(old_inode);
2715         task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2716                                 must_unblock ? nfs_unblock_rename : NULL);
2717         if (IS_ERR(task)) {
2718                 error = PTR_ERR(task);
2719                 goto out;
2720         }
2721
2722         error = rpc_wait_for_completion_task(task);
2723         if (error != 0) {
2724                 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2725                 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2726                 smp_wmb();
2727         } else
2728                 error = task->tk_status;
2729         rpc_put_task(task);
2730         /* Ensure the inode attributes are revalidated */
2731         if (error == 0) {
2732                 spin_lock(&old_inode->i_lock);
2733                 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2734                 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2735                                                          NFS_INO_INVALID_CTIME |
2736                                                          NFS_INO_REVAL_FORCED);
2737                 spin_unlock(&old_inode->i_lock);
2738         }
2739 out:
2740         trace_nfs_rename_exit(old_dir, old_dentry,
2741                         new_dir, new_dentry, error);
2742         if (!error) {
2743                 if (new_inode != NULL)
2744                         nfs_drop_nlink(new_inode);
2745                 /*
2746                  * The d_move() should be here instead of in an async RPC completion
2747                  * handler because we need the proper locks to move the dentry.  If
2748                  * we're interrupted by a signal, the async RPC completion handler
2749                  * should mark the directories for revalidation.
2750                  */
2751                 d_move(old_dentry, new_dentry);
2752                 nfs_set_verifier(old_dentry,
2753                                         nfs_save_change_attribute(new_dir));
2754         } else if (error == -ENOENT)
2755                 nfs_dentry_handle_enoent(old_dentry);
2756
2757         /* new dentry created? */
2758         if (dentry)
2759                 dput(dentry);
2760         return error;
2761 }
2762 EXPORT_SYMBOL_GPL(nfs_rename);
2763
2764 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2765 static LIST_HEAD(nfs_access_lru_list);
2766 static atomic_long_t nfs_access_nr_entries;
2767
2768 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2769 module_param(nfs_access_max_cachesize, ulong, 0644);
2770 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2771
2772 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2773 {
2774         put_group_info(entry->group_info);
2775         kfree_rcu(entry, rcu_head);
2776         smp_mb__before_atomic();
2777         atomic_long_dec(&nfs_access_nr_entries);
2778         smp_mb__after_atomic();
2779 }
2780
2781 static void nfs_access_free_list(struct list_head *head)
2782 {
2783         struct nfs_access_entry *cache;
2784
2785         while (!list_empty(head)) {
2786                 cache = list_entry(head->next, struct nfs_access_entry, lru);
2787                 list_del(&cache->lru);
2788                 nfs_access_free_entry(cache);
2789         }
2790 }
2791
2792 static unsigned long
2793 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2794 {
2795         LIST_HEAD(head);
2796         struct nfs_inode *nfsi, *next;
2797         struct nfs_access_entry *cache;
2798         long freed = 0;
2799
2800         spin_lock(&nfs_access_lru_lock);
2801         list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2802                 struct inode *inode;
2803
2804                 if (nr_to_scan-- == 0)
2805                         break;
2806                 inode = &nfsi->vfs_inode;
2807                 spin_lock(&inode->i_lock);
2808                 if (list_empty(&nfsi->access_cache_entry_lru))
2809                         goto remove_lru_entry;
2810                 cache = list_entry(nfsi->access_cache_entry_lru.next,
2811                                 struct nfs_access_entry, lru);
2812                 list_move(&cache->lru, &head);
2813                 rb_erase(&cache->rb_node, &nfsi->access_cache);
2814                 freed++;
2815                 if (!list_empty(&nfsi->access_cache_entry_lru))
2816                         list_move_tail(&nfsi->access_cache_inode_lru,
2817                                         &nfs_access_lru_list);
2818                 else {
2819 remove_lru_entry:
2820                         list_del_init(&nfsi->access_cache_inode_lru);
2821                         smp_mb__before_atomic();
2822                         clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2823                         smp_mb__after_atomic();
2824                 }
2825                 spin_unlock(&inode->i_lock);
2826         }
2827         spin_unlock(&nfs_access_lru_lock);
2828         nfs_access_free_list(&head);
2829         return freed;
2830 }
2831
2832 unsigned long
2833 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2834 {
2835         int nr_to_scan = sc->nr_to_scan;
2836         gfp_t gfp_mask = sc->gfp_mask;
2837
2838         if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2839                 return SHRINK_STOP;
2840         return nfs_do_access_cache_scan(nr_to_scan);
2841 }
2842
2843
2844 unsigned long
2845 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2846 {
2847         return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2848 }
2849
2850 static void
2851 nfs_access_cache_enforce_limit(void)
2852 {
2853         long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2854         unsigned long diff;
2855         unsigned int nr_to_scan;
2856
2857         if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2858                 return;
2859         nr_to_scan = 100;
2860         diff = nr_entries - nfs_access_max_cachesize;
2861         if (diff < nr_to_scan)
2862                 nr_to_scan = diff;
2863         nfs_do_access_cache_scan(nr_to_scan);
2864 }
2865
2866 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2867 {
2868         struct rb_root *root_node = &nfsi->access_cache;
2869         struct rb_node *n;
2870         struct nfs_access_entry *entry;
2871
2872         /* Unhook entries from the cache */
2873         while ((n = rb_first(root_node)) != NULL) {
2874                 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2875                 rb_erase(n, root_node);
2876                 list_move(&entry->lru, head);
2877         }
2878         nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2879 }
2880
2881 void nfs_access_zap_cache(struct inode *inode)
2882 {
2883         LIST_HEAD(head);
2884
2885         if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2886                 return;
2887         /* Remove from global LRU init */
2888         spin_lock(&nfs_access_lru_lock);
2889         if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2890                 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2891
2892         spin_lock(&inode->i_lock);
2893         __nfs_access_zap_cache(NFS_I(inode), &head);
2894         spin_unlock(&inode->i_lock);
2895         spin_unlock(&nfs_access_lru_lock);
2896         nfs_access_free_list(&head);
2897 }
2898 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2899
2900 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2901 {
2902         struct group_info *ga, *gb;
2903         int g;
2904
2905         if (uid_lt(a->fsuid, b->fsuid))
2906                 return -1;
2907         if (uid_gt(a->fsuid, b->fsuid))
2908                 return 1;
2909
2910         if (gid_lt(a->fsgid, b->fsgid))
2911                 return -1;
2912         if (gid_gt(a->fsgid, b->fsgid))
2913                 return 1;
2914
2915         ga = a->group_info;
2916         gb = b->group_info;
2917         if (ga == gb)
2918                 return 0;
2919         if (ga == NULL)
2920                 return -1;
2921         if (gb == NULL)
2922                 return 1;
2923         if (ga->ngroups < gb->ngroups)
2924                 return -1;
2925         if (ga->ngroups > gb->ngroups)
2926                 return 1;
2927
2928         for (g = 0; g < ga->ngroups; g++) {
2929                 if (gid_lt(ga->gid[g], gb->gid[g]))
2930                         return -1;
2931                 if (gid_gt(ga->gid[g], gb->gid[g]))
2932                         return 1;
2933         }
2934         return 0;
2935 }
2936
2937 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2938 {
2939         struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2940
2941         while (n != NULL) {
2942                 struct nfs_access_entry *entry =
2943                         rb_entry(n, struct nfs_access_entry, rb_node);
2944                 int cmp = access_cmp(cred, entry);
2945
2946                 if (cmp < 0)
2947                         n = n->rb_left;
2948                 else if (cmp > 0)
2949                         n = n->rb_right;
2950                 else
2951                         return entry;
2952         }
2953         return NULL;
2954 }
2955
2956 static u64 nfs_access_login_time(const struct task_struct *task,
2957                                  const struct cred *cred)
2958 {
2959         const struct task_struct *parent;
2960         u64 ret;
2961
2962         rcu_read_lock();
2963         for (;;) {
2964                 parent = rcu_dereference(task->real_parent);
2965                 if (parent == task || cred_fscmp(parent->cred, cred) != 0)
2966                         break;
2967                 task = parent;
2968         }
2969         ret = task->start_time;
2970         rcu_read_unlock();
2971         return ret;
2972 }
2973
2974 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2975 {
2976         struct nfs_inode *nfsi = NFS_I(inode);
2977         u64 login_time = nfs_access_login_time(current, cred);
2978         struct nfs_access_entry *cache;
2979         bool retry = true;
2980         int err;
2981
2982         spin_lock(&inode->i_lock);
2983         for(;;) {
2984                 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2985                         goto out_zap;
2986                 cache = nfs_access_search_rbtree(inode, cred);
2987                 err = -ENOENT;
2988                 if (cache == NULL)
2989                         goto out;
2990                 /* Found an entry, is our attribute cache valid? */
2991                 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2992                         break;
2993                 if (!retry)
2994                         break;
2995                 err = -ECHILD;
2996                 if (!may_block)
2997                         goto out;
2998                 spin_unlock(&inode->i_lock);
2999                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3000                 if (err)
3001                         return err;
3002                 spin_lock(&inode->i_lock);
3003                 retry = false;
3004         }
3005         err = -ENOENT;
3006         if ((s64)(login_time - cache->timestamp) > 0)
3007                 goto out;
3008         *mask = cache->mask;
3009         list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
3010         err = 0;
3011 out:
3012         spin_unlock(&inode->i_lock);
3013         return err;
3014 out_zap:
3015         spin_unlock(&inode->i_lock);
3016         nfs_access_zap_cache(inode);
3017         return -ENOENT;
3018 }
3019
3020 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3021 {
3022         /* Only check the most recently returned cache entry,
3023          * but do it without locking.
3024          */
3025         struct nfs_inode *nfsi = NFS_I(inode);
3026         struct nfs_access_entry *cache;
3027         int err = -ECHILD;
3028         struct list_head *lh;
3029
3030         rcu_read_lock();
3031         if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3032                 goto out;
3033         lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3034         cache = list_entry(lh, struct nfs_access_entry, lru);
3035         if (lh == &nfsi->access_cache_entry_lru ||
3036             access_cmp(cred, cache) != 0)
3037                 cache = NULL;
3038         if (cache == NULL)
3039                 goto out;
3040         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3041                 goto out;
3042         *mask = cache->mask;
3043         err = 0;
3044 out:
3045         rcu_read_unlock();
3046         return err;
3047 }
3048
3049 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3050                           u32 *mask, bool may_block)
3051 {
3052         int status;
3053
3054         status = nfs_access_get_cached_rcu(inode, cred, mask);
3055         if (status != 0)
3056                 status = nfs_access_get_cached_locked(inode, cred, mask,
3057                     may_block);
3058
3059         return status;
3060 }
3061 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3062
3063 static void nfs_access_add_rbtree(struct inode *inode,
3064                                   struct nfs_access_entry *set,
3065                                   const struct cred *cred)
3066 {
3067         struct nfs_inode *nfsi = NFS_I(inode);
3068         struct rb_root *root_node = &nfsi->access_cache;
3069         struct rb_node **p = &root_node->rb_node;
3070         struct rb_node *parent = NULL;
3071         struct nfs_access_entry *entry;
3072         int cmp;
3073
3074         spin_lock(&inode->i_lock);
3075         while (*p != NULL) {
3076                 parent = *p;
3077                 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3078                 cmp = access_cmp(cred, entry);
3079
3080                 if (cmp < 0)
3081                         p = &parent->rb_left;
3082                 else if (cmp > 0)
3083                         p = &parent->rb_right;
3084                 else
3085                         goto found;
3086         }
3087         set->timestamp = ktime_get_ns();
3088         rb_link_node(&set->rb_node, parent, p);
3089         rb_insert_color(&set->rb_node, root_node);
3090         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3091         spin_unlock(&inode->i_lock);
3092         return;
3093 found:
3094         rb_replace_node(parent, &set->rb_node, root_node);
3095         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3096         list_del(&entry->lru);
3097         spin_unlock(&inode->i_lock);
3098         nfs_access_free_entry(entry);
3099 }
3100
3101 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3102                           const struct cred *cred)
3103 {
3104         struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3105         if (cache == NULL)
3106                 return;
3107         RB_CLEAR_NODE(&cache->rb_node);
3108         cache->fsuid = cred->fsuid;
3109         cache->fsgid = cred->fsgid;
3110         cache->group_info = get_group_info(cred->group_info);
3111         cache->mask = set->mask;
3112
3113         /* The above field assignments must be visible
3114          * before this item appears on the lru.  We cannot easily
3115          * use rcu_assign_pointer, so just force the memory barrier.
3116          */
3117         smp_wmb();
3118         nfs_access_add_rbtree(inode, cache, cred);
3119
3120         /* Update accounting */
3121         smp_mb__before_atomic();
3122         atomic_long_inc(&nfs_access_nr_entries);
3123         smp_mb__after_atomic();
3124
3125         /* Add inode to global LRU list */
3126         if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3127                 spin_lock(&nfs_access_lru_lock);
3128                 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3129                         list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3130                                         &nfs_access_lru_list);
3131                 spin_unlock(&nfs_access_lru_lock);
3132         }
3133         nfs_access_cache_enforce_limit();
3134 }
3135 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3136
3137 #define NFS_MAY_READ (NFS_ACCESS_READ)
3138 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3139                 NFS_ACCESS_EXTEND | \
3140                 NFS_ACCESS_DELETE)
3141 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3142                 NFS_ACCESS_EXTEND)
3143 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3144 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3145 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3146 static int
3147 nfs_access_calc_mask(u32 access_result, umode_t umode)
3148 {
3149         int mask = 0;
3150
3151         if (access_result & NFS_MAY_READ)
3152                 mask |= MAY_READ;
3153         if (S_ISDIR(umode)) {
3154                 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3155                         mask |= MAY_WRITE;
3156                 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3157                         mask |= MAY_EXEC;
3158         } else if (S_ISREG(umode)) {
3159                 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3160                         mask |= MAY_WRITE;
3161                 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3162                         mask |= MAY_EXEC;
3163         } else if (access_result & NFS_MAY_WRITE)
3164                         mask |= MAY_WRITE;
3165         return mask;
3166 }
3167
3168 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3169 {
3170         entry->mask = access_result;
3171 }
3172 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3173
3174 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3175 {
3176         struct nfs_access_entry cache;
3177         bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3178         int cache_mask = -1;
3179         int status;
3180
3181         trace_nfs_access_enter(inode);
3182
3183         status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3184         if (status == 0)
3185                 goto out_cached;
3186
3187         status = -ECHILD;
3188         if (!may_block)
3189                 goto out;
3190
3191         /*
3192          * Determine which access bits we want to ask for...
3193          */
3194         cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3195                      nfs_access_xattr_mask(NFS_SERVER(inode));
3196         if (S_ISDIR(inode->i_mode))
3197                 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3198         else
3199                 cache.mask |= NFS_ACCESS_EXECUTE;
3200         status = NFS_PROTO(inode)->access(inode, &cache, cred);
3201         if (status != 0) {
3202                 if (status == -ESTALE) {
3203                         if (!S_ISDIR(inode->i_mode))
3204                                 nfs_set_inode_stale(inode);
3205                         else
3206                                 nfs_zap_caches(inode);
3207                 }
3208                 goto out;
3209         }
3210         nfs_access_add_cache(inode, &cache, cred);
3211 out_cached:
3212         cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3213         if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3214                 status = -EACCES;
3215 out:
3216         trace_nfs_access_exit(inode, mask, cache_mask, status);
3217         return status;
3218 }
3219
3220 static int nfs_open_permission_mask(int openflags)
3221 {
3222         int mask = 0;
3223
3224         if (openflags & __FMODE_EXEC) {
3225                 /* ONLY check exec rights */
3226                 mask = MAY_EXEC;
3227         } else {
3228                 if ((openflags & O_ACCMODE) != O_WRONLY)
3229                         mask |= MAY_READ;
3230                 if ((openflags & O_ACCMODE) != O_RDONLY)
3231                         mask |= MAY_WRITE;
3232         }
3233
3234         return mask;
3235 }
3236
3237 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3238 {
3239         return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3240 }
3241 EXPORT_SYMBOL_GPL(nfs_may_open);
3242
3243 static int nfs_execute_ok(struct inode *inode, int mask)
3244 {
3245         struct nfs_server *server = NFS_SERVER(inode);
3246         int ret = 0;
3247
3248         if (S_ISDIR(inode->i_mode))
3249                 return 0;
3250         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3251                 if (mask & MAY_NOT_BLOCK)
3252                         return -ECHILD;
3253                 ret = __nfs_revalidate_inode(server, inode);
3254         }
3255         if (ret == 0 && !execute_ok(inode))
3256                 ret = -EACCES;
3257         return ret;
3258 }
3259
3260 int nfs_permission(struct user_namespace *mnt_userns,
3261                    struct inode *inode,
3262                    int mask)
3263 {
3264         const struct cred *cred = current_cred();
3265         int res = 0;
3266
3267         nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3268
3269         if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3270                 goto out;
3271         /* Is this sys_access() ? */
3272         if (mask & (MAY_ACCESS | MAY_CHDIR))
3273                 goto force_lookup;
3274
3275         switch (inode->i_mode & S_IFMT) {
3276                 case S_IFLNK:
3277                         goto out;
3278                 case S_IFREG:
3279                         if ((mask & MAY_OPEN) &&
3280                            nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3281                                 return 0;
3282                         break;
3283                 case S_IFDIR:
3284                         /*
3285                          * Optimize away all write operations, since the server
3286                          * will check permissions when we perform the op.
3287                          */
3288                         if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3289                                 goto out;
3290         }
3291
3292 force_lookup:
3293         if (!NFS_PROTO(inode)->access)
3294                 goto out_notsup;
3295
3296         res = nfs_do_access(inode, cred, mask);
3297 out:
3298         if (!res && (mask & MAY_EXEC))
3299                 res = nfs_execute_ok(inode, mask);
3300
3301         dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3302                 inode->i_sb->s_id, inode->i_ino, mask, res);
3303         return res;
3304 out_notsup:
3305         if (mask & MAY_NOT_BLOCK)
3306                 return -ECHILD;
3307
3308         res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3309                                                   NFS_INO_INVALID_OTHER);
3310         if (res == 0)
3311                 res = generic_permission(&init_user_ns, inode, mask);
3312         goto out;
3313 }
3314 EXPORT_SYMBOL_GPL(nfs_permission);