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