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 claus@momo.math.rwth-aachen.de
  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/module.h>
  22 #include <linux/time.h>
  23 #include <linux/errno.h>
  24 #include <linux/stat.h>
  25 #include <linux/fcntl.h>
  26 #include <linux/string.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/mm.h>
  30 #include <linux/sunrpc/clnt.h>
  31 #include <linux/nfs_fs.h>
  32 #include <linux/nfs_mount.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/pagevec.h>
  35 #include <linux/namei.h>
  36 #include <linux/mount.h>
  37 #include <linux/swap.h>
  38 #include <linux/sched.h>
  39 #include <linux/kmemleak.h>
  40 #include <linux/xattr.h>
  41
  42 #include "delegation.h"
  43 #include "iostat.h"
  44 #include "internal.h"
  45 #include "fscache.h"
  46
  47 #include "nfstrace.h"
  48
  49 /* #define NFS_DEBUG_VERBOSE 1 */
  50
  51 static int nfs_opendir(struct inode *, struct file *);
  52 static int nfs_closedir(struct inode *, struct file *);
  53 static int nfs_readdir(struct file *, struct dir_context *);
  54 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
  55 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
  56 static void nfs_readdir_clear_array(struct page*);
  57
  58 const struct file_operations nfs_dir_operations = {
  59         .llseek         = nfs_llseek_dir,
  60         .read           = generic_read_dir,
  61         .iterate_shared = nfs_readdir,
  62         .open           = nfs_opendir,
  63         .release        = nfs_closedir,
  64         .fsync          = nfs_fsync_dir,
  65 };
  66
  67 const struct address_space_operations nfs_dir_aops = {
  68         .freepage = nfs_readdir_clear_array,
  69 };
  70
  71 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir)
  72 {
  73         struct nfs_inode *nfsi = NFS_I(dir);
  74         struct nfs_open_dir_context *ctx;
  75         ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
  76         if (ctx != NULL) {
  77                 ctx->duped = 0;
  78                 ctx->attr_gencount = nfsi->attr_gencount;
  79                 ctx->dir_cookie = 0;
  80                 ctx->dup_cookie = 0;
  81                 spin_lock(&dir->i_lock);
  82                 if (list_empty(&nfsi->open_files) &&
  83                     (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
  84                         nfsi->cache_validity |= NFS_INO_INVALID_DATA |
  85                                 NFS_INO_REVAL_FORCED;
  86                 list_add(&ctx->list, &nfsi->open_files);
  87                 spin_unlock(&dir->i_lock);
  88                 return ctx;
  89         }
  90         return  ERR_PTR(-ENOMEM);
  91 }
  92
  93 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
  94 {
  95         spin_lock(&dir->i_lock);
  96         list_del(&ctx->list);
  97         spin_unlock(&dir->i_lock);
  98         kfree(ctx);
  99 }
 100
 101 /*
 102  * Open file
 103  */
 104 static int
 105 nfs_opendir(struct inode *inode, struct file *filp)
 106 {
 107         int res = 0;
 108         struct nfs_open_dir_context *ctx;
 109
 110         dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
 111
 112         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 113
 114         ctx = alloc_nfs_open_dir_context(inode);
 115         if (IS_ERR(ctx)) {
 116                 res = PTR_ERR(ctx);
 117                 goto out;
 118         }
 119         filp->private_data = ctx;
 120 out:
 121         return res;
 122 }
 123
 124 static int
 125 nfs_closedir(struct inode *inode, struct file *filp)
 126 {
 127         put_nfs_open_dir_context(file_inode(filp), filp->private_data);
 128         return 0;
 129 }
 130
 131 struct nfs_cache_array_entry {
 132         u64 cookie;
 133         u64 ino;
 134         const char *name;
 135         unsigned int name_len;
 136         unsigned char d_type;
 137 };
 138
 139 struct nfs_cache_array {
 140         u64 last_cookie;
 141         unsigned int size;
 142         unsigned char page_full : 1,
 143                       page_is_eof : 1,
 144                       cookies_are_ordered : 1;
 145         struct nfs_cache_array_entry array[];
 146 };
 147
 148 struct nfs_readdir_descriptor {
 149         struct file     *file;
 150         struct page     *page;
 151         struct dir_context *ctx;
 152         pgoff_t         page_index;
 153         u64             dir_cookie;
 154         u64             last_cookie;
 155         u64             dup_cookie;
 156         loff_t          current_index;
 157         loff_t          prev_index;
 158
 159         __be32          verf[NFS_DIR_VERIFIER_SIZE];
 160         unsigned long   dir_verifier;
 161         unsigned long   timestamp;
 162         unsigned long   gencount;
 163         unsigned long   attr_gencount;
 164         unsigned int    cache_entry_index;
 165         signed char duped;
 166         bool plus;
 167         bool eof;
 168 };
 169
 170 static void nfs_readdir_array_init(struct nfs_cache_array *array)
 171 {
 172         memset(array, 0, sizeof(struct nfs_cache_array));
 173 }
 174
 175 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie)
 176 {
 177         struct nfs_cache_array *array;
 178
 179         array = kmap_atomic(page);
 180         nfs_readdir_array_init(array);
 181         array->last_cookie = last_cookie;
 182         array->cookies_are_ordered = 1;
 183         kunmap_atomic(array);
 184 }
 185
 186 /*
 187  * we are freeing strings created by nfs_add_to_readdir_array()
 188  */
 189 static
 190 void nfs_readdir_clear_array(struct page *page)
 191 {
 192         struct nfs_cache_array *array;
 193         int i;
 194
 195         array = kmap_atomic(page);
 196         for (i = 0; i < array->size; i++)
 197                 kfree(array->array[i].name);
 198         nfs_readdir_array_init(array);
 199         kunmap_atomic(array);
 200 }
 201
 202 static struct page *
 203 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
 204 {
 205         struct page *page = alloc_page(gfp_flags);
 206         if (page)
 207                 nfs_readdir_page_init_array(page, last_cookie);
 208         return page;
 209 }
 210
 211 static void nfs_readdir_page_array_free(struct page *page)
 212 {
 213         if (page) {
 214                 nfs_readdir_clear_array(page);
 215                 put_page(page);
 216         }
 217 }
 218
 219 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
 220 {
 221         array->page_is_eof = 1;
 222         array->page_full = 1;
 223 }
 224
 225 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
 226 {
 227         return array->page_full;
 228 }
 229
 230 /*
 231  * the caller is responsible for freeing qstr.name
 232  * when called by nfs_readdir_add_to_array, the strings will be freed in
 233  * nfs_clear_readdir_array()
 234  */
 235 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
 236 {
 237         const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
 238
 239         /*
 240          * Avoid a kmemleak false positive. The pointer to the name is stored
 241          * in a page cache page which kmemleak does not scan.
 242          */
 243         if (ret != NULL)
 244                 kmemleak_not_leak(ret);
 245         return ret;
 246 }
 247
 248 /*
 249  * Check that the next array entry lies entirely within the page bounds
 250  */
 251 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
 252 {
 253         struct nfs_cache_array_entry *cache_entry;
 254
 255         if (array->page_full)
 256                 return -ENOSPC;
 257         cache_entry = &array->array[array->size + 1];
 258         if ((char *)cache_entry - (char *)array > PAGE_SIZE) {
 259                 array->page_full = 1;
 260                 return -ENOSPC;
 261         }
 262         return 0;
 263 }
 264
 265 static
 266 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
 267 {
 268         struct nfs_cache_array *array;
 269         struct nfs_cache_array_entry *cache_entry;
 270         const char *name;
 271         int ret;
 272
 273         name = nfs_readdir_copy_name(entry->name, entry->len);
 274         if (!name)
 275                 return -ENOMEM;
 276
 277         array = kmap_atomic(page);
 278         ret = nfs_readdir_array_can_expand(array);
 279         if (ret) {
 280                 kfree(name);
 281                 goto out;
 282         }
 283
 284         cache_entry = &array->array[array->size];
 285         cache_entry->cookie = entry->prev_cookie;
 286         cache_entry->ino = entry->ino;
 287         cache_entry->d_type = entry->d_type;
 288         cache_entry->name_len = entry->len;
 289         cache_entry->name = name;
 290         array->last_cookie = entry->cookie;
 291         if (array->last_cookie <= cache_entry->cookie)
 292                 array->cookies_are_ordered = 0;
 293         array->size++;
 294         if (entry->eof != 0)
 295                 nfs_readdir_array_set_eof(array);
 296 out:
 297         kunmap_atomic(array);
 298         return ret;
 299 }
 300
 301 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
 302                                                 pgoff_t index, u64 last_cookie)
 303 {
 304         struct page *page;
 305
 306         page = grab_cache_page(mapping, index);
 307         if (page && !PageUptodate(page)) {
 308                 nfs_readdir_page_init_array(page, last_cookie);
 309                 if (invalidate_inode_pages2_range(mapping, index + 1, -1) < 0)
 310                         nfs_zap_mapping(mapping->host, mapping);
 311                 SetPageUptodate(page);
 312         }
 313
 314         return page;
 315 }
 316
 317 static u64 nfs_readdir_page_last_cookie(struct page *page)
 318 {
 319         struct nfs_cache_array *array;
 320         u64 ret;
 321
 322         array = kmap_atomic(page);
 323         ret = array->last_cookie;
 324         kunmap_atomic(array);
 325         return ret;
 326 }
 327
 328 static bool nfs_readdir_page_needs_filling(struct page *page)
 329 {
 330         struct nfs_cache_array *array;
 331         bool ret;
 332
 333         array = kmap_atomic(page);
 334         ret = !nfs_readdir_array_is_full(array);
 335         kunmap_atomic(array);
 336         return ret;
 337 }
 338
 339 static void nfs_readdir_page_set_eof(struct page *page)
 340 {
 341         struct nfs_cache_array *array;
 342
 343         array = kmap_atomic(page);
 344         nfs_readdir_array_set_eof(array);
 345         kunmap_atomic(array);
 346 }
 347
 348 static void nfs_readdir_page_unlock_and_put(struct page *page)
 349 {
 350         unlock_page(page);
 351         put_page(page);
 352 }
 353
 354 static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
 355                                               pgoff_t index, u64 cookie)
 356 {
 357         struct page *page;
 358
 359         page = nfs_readdir_page_get_locked(mapping, index, cookie);
 360         if (page) {
 361                 if (nfs_readdir_page_last_cookie(page) == cookie)
 362                         return page;
 363                 nfs_readdir_page_unlock_and_put(page);
 364         }
 365         return NULL;
 366 }
 367
 368 static inline
 369 int is_32bit_api(void)
 370 {
 371 #ifdef CONFIG_COMPAT
 372         return in_compat_syscall();
 373 #else
 374         return (BITS_PER_LONG == 32);
 375 #endif
 376 }
 377
 378 static
 379 bool nfs_readdir_use_cookie(const struct file *filp)
 380 {
 381         if ((filp->f_mode & FMODE_32BITHASH) ||
 382             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 383                 return false;
 384         return true;
 385 }
 386
 387 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
 388                                       struct nfs_readdir_descriptor *desc)
 389 {
 390         loff_t diff = desc->ctx->pos - desc->current_index;
 391         unsigned int index;
 392
 393         if (diff < 0)
 394                 goto out_eof;
 395         if (diff >= array->size) {
 396                 if (array->page_is_eof)
 397                         goto out_eof;
 398                 return -EAGAIN;
 399         }
 400
 401         index = (unsigned int)diff;
 402         desc->dir_cookie = array->array[index].cookie;
 403         desc->cache_entry_index = index;
 404         return 0;
 405 out_eof:
 406         desc->eof = true;
 407         return -EBADCOOKIE;
 408 }
 409
 410 static bool
 411 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
 412 {
 413         if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
 414                 return false;
 415         smp_rmb();
 416         return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
 417 }
 418
 419 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
 420                                               u64 cookie)
 421 {
 422         if (!array->cookies_are_ordered)
 423                 return true;
 424         /* Optimisation for monotonically increasing cookies */
 425         if (cookie >= array->last_cookie)
 426                 return false;
 427         if (array->size && cookie < array->array[0].cookie)
 428                 return false;
 429         return true;
 430 }
 431
 432 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
 433                                          struct nfs_readdir_descriptor *desc)
 434 {
 435         int i;
 436         loff_t new_pos;
 437         int status = -EAGAIN;
 438
 439         if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
 440                 goto check_eof;
 441
 442         for (i = 0; i < array->size; i++) {
 443                 if (array->array[i].cookie == desc->dir_cookie) {
 444                         struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
 445
 446                         new_pos = desc->current_index + i;
 447                         if (desc->attr_gencount != nfsi->attr_gencount ||
 448                             !nfs_readdir_inode_mapping_valid(nfsi)) {
 449                                 desc->duped = 0;
 450                                 desc->attr_gencount = nfsi->attr_gencount;
 451                         } else if (new_pos < desc->prev_index) {
 452                                 if (desc->duped > 0
 453                                     && desc->dup_cookie == desc->dir_cookie) {
 454                                         if (printk_ratelimit()) {
 455                                                 pr_notice("NFS: directory %pD2 contains a readdir loop."
 456                                                                 "Please contact your server vendor.  "
 457                                                                 "The file: %s has duplicate cookie %llu\n",
 458                                                                 desc->file, array->array[i].name, desc->dir_cookie);
 459                                         }
 460                                         status = -ELOOP;
 461                                         goto out;
 462                                 }
 463                                 desc->dup_cookie = desc->dir_cookie;
 464                                 desc->duped = -1;
 465                         }
 466                         if (nfs_readdir_use_cookie(desc->file))
 467                                 desc->ctx->pos = desc->dir_cookie;
 468                         else
 469                                 desc->ctx->pos = new_pos;
 470                         desc->prev_index = new_pos;
 471                         desc->cache_entry_index = i;
 472                         return 0;
 473                 }
 474         }
 475 check_eof:
 476         if (array->page_is_eof) {
 477                 status = -EBADCOOKIE;
 478                 if (desc->dir_cookie == array->last_cookie)
 479                         desc->eof = true;
 480         }
 481 out:
 482         return status;
 483 }
 484
 485 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
 486 {
 487         struct nfs_cache_array *array;
 488         int status;
 489
 490         array = kmap_atomic(desc->page);
 491
 492         if (desc->dir_cookie == 0)
 493                 status = nfs_readdir_search_for_pos(array, desc);
 494         else
 495                 status = nfs_readdir_search_for_cookie(array, desc);
 496
 497         if (status == -EAGAIN) {
 498                 desc->last_cookie = array->last_cookie;
 499                 desc->current_index += array->size;
 500                 desc->page_index++;
 501         }
 502         kunmap_atomic(array);
 503         return status;
 504 }
 505
 506 /* Fill a page with xdr information before transferring to the cache page */
 507 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
 508                                   __be32 *verf, u64 cookie,
 509                                   struct page **pages, size_t bufsize,
 510                                   __be32 *verf_res)
 511 {
 512         struct inode *inode = file_inode(desc->file);
 513         struct nfs_readdir_arg arg = {
 514                 .dentry = file_dentry(desc->file),
 515                 .cred = desc->file->f_cred,
 516                 .verf = verf,
 517                 .cookie = cookie,
 518                 .pages = pages,
 519                 .page_len = bufsize,
 520                 .plus = desc->plus,
 521         };
 522         struct nfs_readdir_res res = {
 523                 .verf = verf_res,
 524         };
 525         unsigned long   timestamp, gencount;
 526         int             error;
 527
 528  again:
 529         timestamp = jiffies;
 530         gencount = nfs_inc_attr_generation_counter();
 531         desc->dir_verifier = nfs_save_change_attribute(inode);
 532         error = NFS_PROTO(inode)->readdir(&arg, &res);
 533         if (error < 0) {
 534                 /* We requested READDIRPLUS, but the server doesn't grok it */
 535                 if (error == -ENOTSUPP && desc->plus) {
 536                         NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
 537                         clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 538                         desc->plus = arg.plus = false;
 539                         goto again;
 540                 }
 541                 goto error;
 542         }
 543         desc->timestamp = timestamp;
 544         desc->gencount = gencount;
 545 error:
 546         return error;
 547 }
 548
 549 static int xdr_decode(struct nfs_readdir_descriptor *desc,
 550                       struct nfs_entry *entry, struct xdr_stream *xdr)
 551 {
 552         struct inode *inode = file_inode(desc->file);
 553         int error;
 554
 555         error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
 556         if (error)
 557                 return error;
 558         entry->fattr->time_start = desc->timestamp;
 559         entry->fattr->gencount = desc->gencount;
 560         return 0;
 561 }
 562
 563 /* Match file and dirent using either filehandle or fileid
 564  * Note: caller is responsible for checking the fsid
 565  */
 566 static
 567 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
 568 {
 569         struct inode *inode;
 570         struct nfs_inode *nfsi;
 571
 572         if (d_really_is_negative(dentry))
 573                 return 0;
 574
 575         inode = d_inode(dentry);
 576         if (is_bad_inode(inode) || NFS_STALE(inode))
 577                 return 0;
 578
 579         nfsi = NFS_I(inode);
 580         if (entry->fattr->fileid != nfsi->fileid)
 581                 return 0;
 582         if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
 583                 return 0;
 584         return 1;
 585 }
 586
 587 static
 588 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
 589 {
 590         if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
 591                 return false;
 592         if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
 593                 return true;
 594         if (ctx->pos == 0)
 595                 return true;
 596         return false;
 597 }
 598
 599 /*
 600  * This function is called by the lookup and getattr code to request the
 601  * use of readdirplus to accelerate any future lookups in the same
 602  * directory.
 603  */
 604 void nfs_advise_use_readdirplus(struct inode *dir)
 605 {
 606         struct nfs_inode *nfsi = NFS_I(dir);
 607
 608         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 609             !list_empty(&nfsi->open_files))
 610                 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 611 }
 612
 613 /*
 614  * This function is mainly for use by nfs_getattr().
 615  *
 616  * If this is an 'ls -l', we want to force use of readdirplus.
 617  * Do this by checking if there is an active file descriptor
 618  * and calling nfs_advise_use_readdirplus, then forcing a
 619  * cache flush.
 620  */
 621 void nfs_force_use_readdirplus(struct inode *dir)
 622 {
 623         struct nfs_inode *nfsi = NFS_I(dir);
 624
 625         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 626             !list_empty(&nfsi->open_files)) {
 627                 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 628                 invalidate_mapping_pages(dir->i_mapping,
 629                         nfsi->page_index + 1, -1);
 630         }
 631 }
 632
 633 static
 634 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
 635                 unsigned long dir_verifier)
 636 {
 637         struct qstr filename = QSTR_INIT(entry->name, entry->len);
 638         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 639         struct dentry *dentry;
 640         struct dentry *alias;
 641         struct inode *inode;
 642         int status;
 643
 644         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
 645                 return;
 646         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
 647                 return;
 648         if (filename.len == 0)
 649                 return;
 650         /* Validate that the name doesn't contain any illegal '\0' */
 651         if (strnlen(filename.name, filename.len) != filename.len)
 652                 return;
 653         /* ...or '/' */
 654         if (strnchr(filename.name, filename.len, '/'))
 655                 return;
 656         if (filename.name[0] == '.') {
 657                 if (filename.len == 1)
 658                         return;
 659                 if (filename.len == 2 && filename.name[1] == '.')
 660                         return;
 661         }
 662         filename.hash = full_name_hash(parent, filename.name, filename.len);
 663
 664         dentry = d_lookup(parent, &filename);
 665 again:
 666         if (!dentry) {
 667                 dentry = d_alloc_parallel(parent, &filename, &wq);
 668                 if (IS_ERR(dentry))
 669                         return;
 670         }
 671         if (!d_in_lookup(dentry)) {
 672                 /* Is there a mountpoint here? If so, just exit */
 673                 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
 674                                         &entry->fattr->fsid))
 675                         goto out;
 676                 if (nfs_same_file(dentry, entry)) {
 677                         if (!entry->fh->size)
 678                                 goto out;
 679                         nfs_set_verifier(dentry, dir_verifier);
 680                         status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
 681                         if (!status)
 682                                 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
 683                         goto out;
 684                 } else {
 685                         d_invalidate(dentry);
 686                         dput(dentry);
 687                         dentry = NULL;
 688                         goto again;
 689                 }
 690         }
 691         if (!entry->fh->size) {
 692                 d_lookup_done(dentry);
 693                 goto out;
 694         }
 695
 696         inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
 697         alias = d_splice_alias(inode, dentry);
 698         d_lookup_done(dentry);
 699         if (alias) {
 700                 if (IS_ERR(alias))
 701                         goto out;
 702                 dput(dentry);
 703                 dentry = alias;
 704         }
 705         nfs_set_verifier(dentry, dir_verifier);
 706 out:
 707         dput(dentry);
 708 }
 709
 710 /* Perform conversion from xdr to cache array */
 711 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
 712                                    struct nfs_entry *entry,
 713                                    struct page **xdr_pages,
 714                                    unsigned int buflen,
 715                                    struct page **arrays,
 716                                    size_t narrays)
 717 {
 718         struct address_space *mapping = desc->file->f_mapping;
 719         struct xdr_stream stream;
 720         struct xdr_buf buf;
 721         struct page *scratch, *new, *page = *arrays;
 722         int status;
 723
 724         scratch = alloc_page(GFP_KERNEL);
 725         if (scratch == NULL)
 726                 return -ENOMEM;
 727
 728         xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
 729         xdr_set_scratch_page(&stream, scratch);
 730
 731         do {
 732                 if (entry->label)
 733                         entry->label->len = NFS4_MAXLABELLEN;
 734
 735                 status = xdr_decode(desc, entry, &stream);
 736                 if (status != 0)
 737                         break;
 738
 739                 if (desc->plus)
 740                         nfs_prime_dcache(file_dentry(desc->file), entry,
 741                                         desc->dir_verifier);
 742
 743                 status = nfs_readdir_add_to_array(entry, page);
 744                 if (status != -ENOSPC)
 745                         continue;
 746
 747                 if (page->mapping != mapping) {
 748                         if (!--narrays)
 749                                 break;
 750                         new = nfs_readdir_page_array_alloc(entry->prev_cookie,
 751                                                            GFP_KERNEL);
 752                         if (!new)
 753                                 break;
 754                         arrays++;
 755                         *arrays = page = new;
 756                 } else {
 757                         new = nfs_readdir_page_get_next(mapping,
 758                                                         page->index + 1,
 759                                                         entry->prev_cookie);
 760                         if (!new)
 761                                 break;
 762                         if (page != *arrays)
 763                                 nfs_readdir_page_unlock_and_put(page);
 764                         page = new;
 765                 }
 766                 status = nfs_readdir_add_to_array(entry, page);
 767         } while (!status && !entry->eof);
 768
 769         switch (status) {
 770         case -EBADCOOKIE:
 771                 if (entry->eof) {
 772                         nfs_readdir_page_set_eof(page);
 773                         status = 0;
 774                 }
 775                 break;
 776         case -ENOSPC:
 777         case -EAGAIN:
 778                 status = 0;
 779                 break;
 780         }
 781
 782         if (page != *arrays)
 783                 nfs_readdir_page_unlock_and_put(page);
 784
 785         put_page(scratch);
 786         return status;
 787 }
 788
 789 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
 790 {
 791         while (npages--)
 792                 put_page(pages[npages]);
 793         kfree(pages);
 794 }
 795
 796 /*
 797  * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
 798  * to nfs_readdir_free_pages()
 799  */
 800 static struct page **nfs_readdir_alloc_pages(size_t npages)
 801 {
 802         struct page **pages;
 803         size_t i;
 804
 805         pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
 806         if (!pages)
 807                 return NULL;
 808         for (i = 0; i < npages; i++) {
 809                 struct page *page = alloc_page(GFP_KERNEL);
 810                 if (page == NULL)
 811                         goto out_freepages;
 812                 pages[i] = page;
 813         }
 814         return pages;
 815
 816 out_freepages:
 817         nfs_readdir_free_pages(pages, i);
 818         return NULL;
 819 }
 820
 821 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
 822                                     __be32 *verf_arg, __be32 *verf_res,
 823                                     struct page **arrays, size_t narrays)
 824 {
 825         struct page **pages;
 826         struct page *page = *arrays;
 827         struct nfs_entry *entry;
 828         size_t array_size;
 829         struct inode *inode = file_inode(desc->file);
 830         size_t dtsize = NFS_SERVER(inode)->dtsize;
 831         int status = -ENOMEM;
 832
 833         entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 834         if (!entry)
 835                 return -ENOMEM;
 836         entry->cookie = nfs_readdir_page_last_cookie(page);
 837         entry->fh = nfs_alloc_fhandle();
 838         entry->fattr = nfs_alloc_fattr();
 839         entry->server = NFS_SERVER(inode);
 840         if (entry->fh == NULL || entry->fattr == NULL)
 841                 goto out;
 842
 843         entry->label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
 844         if (IS_ERR(entry->label)) {
 845                 status = PTR_ERR(entry->label);
 846                 goto out;
 847         }
 848
 849         array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
 850         pages = nfs_readdir_alloc_pages(array_size);
 851         if (!pages)
 852                 goto out_release_label;
 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                 status = nfs_readdir_page_filler(desc, entry, pages, pglen,
 869                                                  arrays, narrays);
 870         } while (!status && nfs_readdir_page_needs_filling(page));
 871
 872         nfs_readdir_free_pages(pages, array_size);
 873 out_release_label:
 874         nfs4_label_free(entry->label);
 875 out:
 876         nfs_free_fattr(entry->fattr);
 877         nfs_free_fhandle(entry->fh);
 878         kfree(entry);
 879         return status;
 880 }
 881
 882 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
 883 {
 884         put_page(desc->page);
 885         desc->page = NULL;
 886 }
 887
 888 static void
 889 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
 890 {
 891         unlock_page(desc->page);
 892         nfs_readdir_page_put(desc);
 893 }
 894
 895 static struct page *
 896 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
 897 {
 898         return nfs_readdir_page_get_locked(desc->file->f_mapping,
 899                                            desc->page_index,
 900                                            desc->last_cookie);
 901 }
 902
 903 /*
 904  * Returns 0 if desc->dir_cookie was found on page desc->page_index
 905  * and locks the page to prevent removal from the page cache.
 906  */
 907 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
 908 {
 909         struct inode *inode = file_inode(desc->file);
 910         struct nfs_inode *nfsi = NFS_I(inode);
 911         __be32 verf[NFS_DIR_VERIFIER_SIZE];
 912         int res;
 913
 914         desc->page = nfs_readdir_page_get_cached(desc);
 915         if (!desc->page)
 916                 return -ENOMEM;
 917         if (nfs_readdir_page_needs_filling(desc->page)) {
 918                 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
 919                                                &desc->page, 1);
 920                 if (res < 0) {
 921                         nfs_readdir_page_unlock_and_put_cached(desc);
 922                         if (res == -EBADCOOKIE || res == -ENOTSYNC) {
 923                                 invalidate_inode_pages2(desc->file->f_mapping);
 924                                 desc->page_index = 0;
 925                                 return -EAGAIN;
 926                         }
 927                         return res;
 928                 }
 929                 memcpy(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf));
 930         }
 931         res = nfs_readdir_search_array(desc);
 932         if (res == 0) {
 933                 nfsi->page_index = desc->page_index;
 934                 return 0;
 935         }
 936         nfs_readdir_page_unlock_and_put_cached(desc);
 937         return res;
 938 }
 939
 940 static bool nfs_readdir_dont_search_cache(struct nfs_readdir_descriptor *desc)
 941 {
 942         struct address_space *mapping = desc->file->f_mapping;
 943         struct inode *dir = file_inode(desc->file);
 944         unsigned int dtsize = NFS_SERVER(dir)->dtsize;
 945         loff_t size = i_size_read(dir);
 946
 947         /*
 948          * Default to uncached readdir if the page cache is empty, and
 949          * we're looking for a non-zero cookie in a large directory.
 950          */
 951         return desc->dir_cookie != 0 && mapping->nrpages == 0 && size > dtsize;
 952 }
 953
 954 /* Search for desc->dir_cookie from the beginning of the page cache */
 955 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
 956 {
 957         int res;
 958
 959         if (nfs_readdir_dont_search_cache(desc))
 960                 return -EBADCOOKIE;
 961
 962         do {
 963                 if (desc->page_index == 0) {
 964                         desc->current_index = 0;
 965                         desc->prev_index = 0;
 966                         desc->last_cookie = 0;
 967                 }
 968                 res = find_and_lock_cache_page(desc);
 969         } while (res == -EAGAIN);
 970         return res;
 971 }
 972
 973 /*
 974  * Once we've found the start of the dirent within a page: fill 'er up...
 975  */
 976 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc)
 977 {
 978         struct file     *file = desc->file;
 979         struct nfs_inode *nfsi = NFS_I(file_inode(file));
 980         struct nfs_cache_array *array;
 981         unsigned int i = 0;
 982
 983         array = kmap(desc->page);
 984         for (i = desc->cache_entry_index; i < array->size; i++) {
 985                 struct nfs_cache_array_entry *ent;
 986
 987                 ent = &array->array[i];
 988                 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
 989                     nfs_compat_user_ino64(ent->ino), ent->d_type)) {
 990                         desc->eof = true;
 991                         break;
 992                 }
 993                 memcpy(desc->verf, nfsi->cookieverf, sizeof(desc->verf));
 994                 if (i < (array->size-1))
 995                         desc->dir_cookie = array->array[i+1].cookie;
 996                 else
 997                         desc->dir_cookie = array->last_cookie;
 998                 if (nfs_readdir_use_cookie(file))
 999                         desc->ctx->pos = desc->dir_cookie;
1000                 else
1001                         desc->ctx->pos++;
1002                 if (desc->duped != 0)
1003                         desc->duped = 1;
1004         }
1005         if (array->page_is_eof)
1006                 desc->eof = true;
1007
1008         kunmap(desc->page);
1009         dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1010                         (unsigned long long)desc->dir_cookie);
1011 }
1012
1013 /*
1014  * If we cannot find a cookie in our cache, we suspect that this is
1015  * because it points to a deleted file, so we ask the server to return
1016  * whatever it thinks is the next entry. We then feed this to filldir.
1017  * If all goes well, we should then be able to find our way round the
1018  * cache on the next call to readdir_search_pagecache();
1019  *
1020  * NOTE: we cannot add the anonymous page to the pagecache because
1021  *       the data it contains might not be page aligned. Besides,
1022  *       we should already have a complete representation of the
1023  *       directory in the page cache by the time we get here.
1024  */
1025 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1026 {
1027         struct page     **arrays;
1028         size_t          i, sz = 512;
1029         __be32          verf[NFS_DIR_VERIFIER_SIZE];
1030         int             status = -ENOMEM;
1031
1032         dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1033                         (unsigned long long)desc->dir_cookie);
1034
1035         arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1036         if (!arrays)
1037                 goto out;
1038         arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1039         if (!arrays[0])
1040                 goto out;
1041
1042         desc->page_index = 0;
1043         desc->last_cookie = desc->dir_cookie;
1044         desc->duped = 0;
1045
1046         status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1047
1048         for (i = 0; !desc->eof && i < sz && arrays[i]; i++) {
1049                 desc->page = arrays[i];
1050                 nfs_do_filldir(desc);
1051         }
1052         desc->page = NULL;
1053
1054
1055         for (i = 0; i < sz && arrays[i]; i++)
1056                 nfs_readdir_page_array_free(arrays[i]);
1057 out:
1058         kfree(arrays);
1059         dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1060         return status;
1061 }
1062
1063 /* The file offset position represents the dirent entry number.  A
1064    last cookie cache takes care of the common case of reading the
1065    whole directory.
1066  */
1067 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1068 {
1069         struct dentry   *dentry = file_dentry(file);
1070         struct inode    *inode = d_inode(dentry);
1071         struct nfs_open_dir_context *dir_ctx = file->private_data;
1072         struct nfs_readdir_descriptor *desc;
1073         int res;
1074
1075         dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1076                         file, (long long)ctx->pos);
1077         nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1078
1079         /*
1080          * ctx->pos points to the dirent entry number.
1081          * *desc->dir_cookie has the cookie for the next entry. We have
1082          * to either find the entry with the appropriate number or
1083          * revalidate the cookie.
1084          */
1085         if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) {
1086                 res = nfs_revalidate_mapping(inode, file->f_mapping);
1087                 if (res < 0)
1088                         goto out;
1089         }
1090
1091         res = -ENOMEM;
1092         desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1093         if (!desc)
1094                 goto out;
1095         desc->file = file;
1096         desc->ctx = ctx;
1097         desc->plus = nfs_use_readdirplus(inode, ctx);
1098
1099         spin_lock(&file->f_lock);
1100         desc->dir_cookie = dir_ctx->dir_cookie;
1101         desc->dup_cookie = dir_ctx->dup_cookie;
1102         desc->duped = dir_ctx->duped;
1103         desc->attr_gencount = dir_ctx->attr_gencount;
1104         memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1105         spin_unlock(&file->f_lock);
1106
1107         do {
1108                 res = readdir_search_pagecache(desc);
1109
1110                 if (res == -EBADCOOKIE) {
1111                         res = 0;
1112                         /* This means either end of directory */
1113                         if (desc->dir_cookie && !desc->eof) {
1114                                 /* Or that the server has 'lost' a cookie */
1115                                 res = uncached_readdir(desc);
1116                                 if (res == 0)
1117                                         continue;
1118                                 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1119                                         res = 0;
1120                         }
1121                         break;
1122                 }
1123                 if (res == -ETOOSMALL && desc->plus) {
1124                         clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
1125                         nfs_zap_caches(inode);
1126                         desc->page_index = 0;
1127                         desc->plus = false;
1128                         desc->eof = false;
1129                         continue;
1130                 }
1131                 if (res < 0)
1132                         break;
1133
1134                 nfs_do_filldir(desc);
1135                 nfs_readdir_page_unlock_and_put_cached(desc);
1136         } while (!desc->eof);
1137
1138         spin_lock(&file->f_lock);
1139         dir_ctx->dir_cookie = desc->dir_cookie;
1140         dir_ctx->dup_cookie = desc->dup_cookie;
1141         dir_ctx->duped = desc->duped;
1142         dir_ctx->attr_gencount = desc->attr_gencount;
1143         memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1144         spin_unlock(&file->f_lock);
1145
1146         kfree(desc);
1147
1148 out:
1149         dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1150         return res;
1151 }
1152
1153 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1154 {
1155         struct nfs_open_dir_context *dir_ctx = filp->private_data;
1156
1157         dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1158                         filp, offset, whence);
1159
1160         switch (whence) {
1161         default:
1162                 return -EINVAL;
1163         case SEEK_SET:
1164                 if (offset < 0)
1165                         return -EINVAL;
1166                 spin_lock(&filp->f_lock);
1167                 break;
1168         case SEEK_CUR:
1169                 if (offset == 0)
1170                         return filp->f_pos;
1171                 spin_lock(&filp->f_lock);
1172                 offset += filp->f_pos;
1173                 if (offset < 0) {
1174                         spin_unlock(&filp->f_lock);
1175                         return -EINVAL;
1176                 }
1177         }
1178         if (offset != filp->f_pos) {
1179                 filp->f_pos = offset;
1180                 if (nfs_readdir_use_cookie(filp))
1181                         dir_ctx->dir_cookie = offset;
1182                 else
1183                         dir_ctx->dir_cookie = 0;
1184                 if (offset == 0)
1185                         memset(dir_ctx->verf, 0, sizeof(dir_ctx->verf));
1186                 dir_ctx->duped = 0;
1187         }
1188         spin_unlock(&filp->f_lock);
1189         return offset;
1190 }
1191
1192 /*
1193  * All directory operations under NFS are synchronous, so fsync()
1194  * is a dummy operation.
1195  */
1196 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1197                          int datasync)
1198 {
1199         dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1200
1201         nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1202         return 0;
1203 }
1204
1205 /**
1206  * nfs_force_lookup_revalidate - Mark the directory as having changed
1207  * @dir: pointer to directory inode
1208  *
1209  * This forces the revalidation code in nfs_lookup_revalidate() to do a
1210  * full lookup on all child dentries of 'dir' whenever a change occurs
1211  * on the server that might have invalidated our dcache.
1212  *
1213  * Note that we reserve bit '0' as a tag to let us know when a dentry
1214  * was revalidated while holding a delegation on its inode.
1215  *
1216  * The caller should be holding dir->i_lock
1217  */
1218 void nfs_force_lookup_revalidate(struct inode *dir)
1219 {
1220         NFS_I(dir)->cache_change_attribute += 2;
1221 }
1222 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1223
1224 /**
1225  * nfs_verify_change_attribute - Detects NFS remote directory changes
1226  * @dir: pointer to parent directory inode
1227  * @verf: previously saved change attribute
1228  *
1229  * Return "false" if the verifiers doesn't match the change attribute.
1230  * This would usually indicate that the directory contents have changed on
1231  * the server, and that any dentries need revalidating.
1232  */
1233 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1234 {
1235         return (verf & ~1UL) == nfs_save_change_attribute(dir);
1236 }
1237
1238 static void nfs_set_verifier_delegated(unsigned long *verf)
1239 {
1240         *verf |= 1UL;
1241 }
1242
1243 #if IS_ENABLED(CONFIG_NFS_V4)
1244 static void nfs_unset_verifier_delegated(unsigned long *verf)
1245 {
1246         *verf &= ~1UL;
1247 }
1248 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1249
1250 static bool nfs_test_verifier_delegated(unsigned long verf)
1251 {
1252         return verf & 1;
1253 }
1254
1255 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1256 {
1257         return nfs_test_verifier_delegated(dentry->d_time);
1258 }
1259
1260 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1261 {
1262         struct inode *inode = d_inode(dentry);
1263
1264         if (!nfs_verifier_is_delegated(dentry) &&
1265             !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf))
1266                 goto out;
1267         if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1268                 nfs_set_verifier_delegated(&verf);
1269 out:
1270         dentry->d_time = verf;
1271 }
1272
1273 /**
1274  * nfs_set_verifier - save a parent directory verifier in the dentry
1275  * @dentry: pointer to dentry
1276  * @verf: verifier to save
1277  *
1278  * Saves the parent directory verifier in @dentry. If the inode has
1279  * a delegation, we also tag the dentry as having been revalidated
1280  * while holding a delegation so that we know we don't have to
1281  * look it up again after a directory change.
1282  */
1283 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1284 {
1285
1286         spin_lock(&dentry->d_lock);
1287         nfs_set_verifier_locked(dentry, verf);
1288         spin_unlock(&dentry->d_lock);
1289 }
1290 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1291
1292 #if IS_ENABLED(CONFIG_NFS_V4)
1293 /**
1294  * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1295  * @inode: pointer to inode
1296  *
1297  * Iterates through the dentries in the inode alias list and clears
1298  * the tag used to indicate that the dentry has been revalidated
1299  * while holding a delegation.
1300  * This function is intended for use when the delegation is being
1301  * returned or revoked.
1302  */
1303 void nfs_clear_verifier_delegated(struct inode *inode)
1304 {
1305         struct dentry *alias;
1306
1307         if (!inode)
1308                 return;
1309         spin_lock(&inode->i_lock);
1310         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1311                 spin_lock(&alias->d_lock);
1312                 nfs_unset_verifier_delegated(&alias->d_time);
1313                 spin_unlock(&alias->d_lock);
1314         }
1315         spin_unlock(&inode->i_lock);
1316 }
1317 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1318 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1319
1320 /*
1321  * A check for whether or not the parent directory has changed.
1322  * In the case it has, we assume that the dentries are untrustworthy
1323  * and may need to be looked up again.
1324  * If rcu_walk prevents us from performing a full check, return 0.
1325  */
1326 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1327                               int rcu_walk)
1328 {
1329         if (IS_ROOT(dentry))
1330                 return 1;
1331         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1332                 return 0;
1333         if (!nfs_verify_change_attribute(dir, dentry->d_time))
1334                 return 0;
1335         /* Revalidate nfsi->cache_change_attribute before we declare a match */
1336         if (nfs_mapping_need_revalidate_inode(dir)) {
1337                 if (rcu_walk)
1338                         return 0;
1339                 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1340                         return 0;
1341         }
1342         if (!nfs_verify_change_attribute(dir, dentry->d_time))
1343                 return 0;
1344         return 1;
1345 }
1346
1347 /*
1348  * Use intent information to check whether or not we're going to do
1349  * an O_EXCL create using this path component.
1350  */
1351 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1352 {
1353         if (NFS_PROTO(dir)->version == 2)
1354                 return 0;
1355         return flags & LOOKUP_EXCL;
1356 }
1357
1358 /*
1359  * Inode and filehandle revalidation for lookups.
1360  *
1361  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1362  * or if the intent information indicates that we're about to open this
1363  * particular file and the "nocto" mount flag is not set.
1364  *
1365  */
1366 static
1367 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1368 {
1369         struct nfs_server *server = NFS_SERVER(inode);
1370         int ret;
1371
1372         if (IS_AUTOMOUNT(inode))
1373                 return 0;
1374
1375         if (flags & LOOKUP_OPEN) {
1376                 switch (inode->i_mode & S_IFMT) {
1377                 case S_IFREG:
1378                         /* A NFSv4 OPEN will revalidate later */
1379                         if (server->caps & NFS_CAP_ATOMIC_OPEN)
1380                                 goto out;
1381                         fallthrough;
1382                 case S_IFDIR:
1383                         if (server->flags & NFS_MOUNT_NOCTO)
1384                                 break;
1385                         /* NFS close-to-open cache consistency validation */
1386                         goto out_force;
1387                 }
1388         }
1389
1390         /* VFS wants an on-the-wire revalidation */
1391         if (flags & LOOKUP_REVAL)
1392                 goto out_force;
1393 out:
1394         return (inode->i_nlink == 0) ? -ESTALE : 0;
1395 out_force:
1396         if (flags & LOOKUP_RCU)
1397                 return -ECHILD;
1398         ret = __nfs_revalidate_inode(server, inode);
1399         if (ret != 0)
1400                 return ret;
1401         goto out;
1402 }
1403
1404 /*
1405  * We judge how long we want to trust negative
1406  * dentries by looking at the parent inode mtime.
1407  *
1408  * If parent mtime has changed, we revalidate, else we wait for a
1409  * period corresponding to the parent's attribute cache timeout value.
1410  *
1411  * If LOOKUP_RCU prevents us from performing a full check, return 1
1412  * suggesting a reval is needed.
1413  *
1414  * Note that when creating a new file, or looking up a rename target,
1415  * then it shouldn't be necessary to revalidate a negative dentry.
1416  */
1417 static inline
1418 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1419                        unsigned int flags)
1420 {
1421         if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1422                 return 0;
1423         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1424                 return 1;
1425         return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1426 }
1427
1428 static int
1429 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1430                            struct inode *inode, int error)
1431 {
1432         switch (error) {
1433         case 1:
1434                 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1435                         __func__, dentry);
1436                 return 1;
1437         case 0:
1438                 nfs_mark_for_revalidate(dir);
1439                 if (inode && S_ISDIR(inode->i_mode)) {
1440                         /* Purge readdir caches. */
1441                         nfs_zap_caches(inode);
1442                         /*
1443                          * We can't d_drop the root of a disconnected tree:
1444                          * its d_hash is on the s_anon list and d_drop() would hide
1445                          * it from shrink_dcache_for_unmount(), leading to busy
1446                          * inodes on unmount and further oopses.
1447                          */
1448                         if (IS_ROOT(dentry))
1449                                 return 1;
1450                 }
1451                 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1452                                 __func__, dentry);
1453                 return 0;
1454         }
1455         dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1456                                 __func__, dentry, error);
1457         return error;
1458 }
1459
1460 static int
1461 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1462                                unsigned int flags)
1463 {
1464         int ret = 1;
1465         if (nfs_neg_need_reval(dir, dentry, flags)) {
1466                 if (flags & LOOKUP_RCU)
1467                         return -ECHILD;
1468                 ret = 0;
1469         }
1470         return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1471 }
1472
1473 static int
1474 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1475                                 struct inode *inode)
1476 {
1477         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1478         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1479 }
1480
1481 static int
1482 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1483                              struct inode *inode)
1484 {
1485         struct nfs_fh *fhandle;
1486         struct nfs_fattr *fattr;
1487         struct nfs4_label *label;
1488         unsigned long dir_verifier;
1489         int ret;
1490
1491         ret = -ENOMEM;
1492         fhandle = nfs_alloc_fhandle();
1493         fattr = nfs_alloc_fattr();
1494         label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1495         if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1496                 goto out;
1497
1498         dir_verifier = nfs_save_change_attribute(dir);
1499         ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1500         if (ret < 0) {
1501                 switch (ret) {
1502                 case -ESTALE:
1503                 case -ENOENT:
1504                         ret = 0;
1505                         break;
1506                 case -ETIMEDOUT:
1507                         if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1508                                 ret = 1;
1509                 }
1510                 goto out;
1511         }
1512         ret = 0;
1513         if (nfs_compare_fh(NFS_FH(inode), fhandle))
1514                 goto out;
1515         if (nfs_refresh_inode(inode, fattr) < 0)
1516                 goto out;
1517
1518         nfs_setsecurity(inode, fattr, label);
1519         nfs_set_verifier(dentry, dir_verifier);
1520
1521         /* set a readdirplus hint that we had a cache miss */
1522         nfs_force_use_readdirplus(dir);
1523         ret = 1;
1524 out:
1525         nfs_free_fattr(fattr);
1526         nfs_free_fhandle(fhandle);
1527         nfs4_label_free(label);
1528         return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1529 }
1530
1531 /*
1532  * This is called every time the dcache has a lookup hit,
1533  * and we should check whether we can really trust that
1534  * lookup.
1535  *
1536  * NOTE! The hit can be a negative hit too, don't assume
1537  * we have an inode!
1538  *
1539  * If the parent directory is seen to have changed, we throw out the
1540  * cached dentry and do a new lookup.
1541  */
1542 static int
1543 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1544                          unsigned int flags)
1545 {
1546         struct inode *inode;
1547         int error;
1548
1549         nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1550         inode = d_inode(dentry);
1551
1552         if (!inode)
1553                 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1554
1555         if (is_bad_inode(inode)) {
1556                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1557                                 __func__, dentry);
1558                 goto out_bad;
1559         }
1560
1561         if (nfs_verifier_is_delegated(dentry))
1562                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1563
1564         /* Force a full look up iff the parent directory has changed */
1565         if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1566             nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1567                 error = nfs_lookup_verify_inode(inode, flags);
1568                 if (error) {
1569                         if (error == -ESTALE)
1570                                 nfs_zap_caches(dir);
1571                         goto out_bad;
1572                 }
1573                 nfs_advise_use_readdirplus(dir);
1574                 goto out_valid;
1575         }
1576
1577         if (flags & LOOKUP_RCU)
1578                 return -ECHILD;
1579
1580         if (NFS_STALE(inode))
1581                 goto out_bad;
1582
1583         trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1584         error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1585         trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1586         return error;
1587 out_valid:
1588         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1589 out_bad:
1590         if (flags & LOOKUP_RCU)
1591                 return -ECHILD;
1592         return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1593 }
1594
1595 static int
1596 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1597                         int (*reval)(struct inode *, struct dentry *, unsigned int))
1598 {
1599         struct dentry *parent;
1600         struct inode *dir;
1601         int ret;
1602
1603         if (flags & LOOKUP_RCU) {
1604                 parent = READ_ONCE(dentry->d_parent);
1605                 dir = d_inode_rcu(parent);
1606                 if (!dir)
1607                         return -ECHILD;
1608                 ret = reval(dir, dentry, flags);
1609                 if (parent != READ_ONCE(dentry->d_parent))
1610                         return -ECHILD;
1611         } else {
1612                 parent = dget_parent(dentry);
1613                 ret = reval(d_inode(parent), dentry, flags);
1614                 dput(parent);
1615         }
1616         return ret;
1617 }
1618
1619 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1620 {
1621         return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1622 }
1623
1624 /*
1625  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1626  * when we don't really care about the dentry name. This is called when a
1627  * pathwalk ends on a dentry that was not found via a normal lookup in the
1628  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1629  *
1630  * In this situation, we just want to verify that the inode itself is OK
1631  * since the dentry might have changed on the server.
1632  */
1633 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1634 {
1635         struct inode *inode = d_inode(dentry);
1636         int error = 0;
1637
1638         /*
1639          * I believe we can only get a negative dentry here in the case of a
1640          * procfs-style symlink. Just assume it's correct for now, but we may
1641          * eventually need to do something more here.
1642          */
1643         if (!inode) {
1644                 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1645                                 __func__, dentry);
1646                 return 1;
1647         }
1648
1649         if (is_bad_inode(inode)) {
1650                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1651                                 __func__, dentry);
1652                 return 0;
1653         }
1654
1655         error = nfs_lookup_verify_inode(inode, flags);
1656         dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1657                         __func__, inode->i_ino, error ? "invalid" : "valid");
1658         return !error;
1659 }
1660
1661 /*
1662  * This is called from dput() when d_count is going to 0.
1663  */
1664 static int nfs_dentry_delete(const struct dentry *dentry)
1665 {
1666         dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1667                 dentry, dentry->d_flags);
1668
1669         /* Unhash any dentry with a stale inode */
1670         if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1671                 return 1;
1672
1673         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1674                 /* Unhash it, so that ->d_iput() would be called */
1675                 return 1;
1676         }
1677         if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1678                 /* Unhash it, so that ancestors of killed async unlink
1679                  * files will be cleaned up during umount */
1680                 return 1;
1681         }
1682         return 0;
1683
1684 }
1685
1686 /* Ensure that we revalidate inode->i_nlink */
1687 static void nfs_drop_nlink(struct inode *inode)
1688 {
1689         spin_lock(&inode->i_lock);
1690         /* drop the inode if we're reasonably sure this is the last link */
1691         if (inode->i_nlink > 0)
1692                 drop_nlink(inode);
1693         NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1694         NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1695                 | NFS_INO_INVALID_CTIME
1696                 | NFS_INO_INVALID_OTHER
1697                 | NFS_INO_REVAL_FORCED;
1698         spin_unlock(&inode->i_lock);
1699 }
1700
1701 /*
1702  * Called when the dentry loses inode.
1703  * We use it to clean up silly-renamed files.
1704  */
1705 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1706 {
1707         if (S_ISDIR(inode->i_mode))
1708                 /* drop any readdir cache as it could easily be old */
1709                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1710
1711         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1712                 nfs_complete_unlink(dentry, inode);
1713                 nfs_drop_nlink(inode);
1714         }
1715         iput(inode);
1716 }
1717
1718 static void nfs_d_release(struct dentry *dentry)
1719 {
1720         /* free cached devname value, if it survived that far */
1721         if (unlikely(dentry->d_fsdata)) {
1722                 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1723                         WARN_ON(1);
1724                 else
1725                         kfree(dentry->d_fsdata);
1726         }
1727 }
1728
1729 const struct dentry_operations nfs_dentry_operations = {
1730         .d_revalidate   = nfs_lookup_revalidate,
1731         .d_weak_revalidate      = nfs_weak_revalidate,
1732         .d_delete       = nfs_dentry_delete,
1733         .d_iput         = nfs_dentry_iput,
1734         .d_automount    = nfs_d_automount,
1735         .d_release      = nfs_d_release,
1736 };
1737 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1738
1739 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1740 {
1741         struct dentry *res;
1742         struct inode *inode = NULL;
1743         struct nfs_fh *fhandle = NULL;
1744         struct nfs_fattr *fattr = NULL;
1745         struct nfs4_label *label = NULL;
1746         unsigned long dir_verifier;
1747         int error;
1748
1749         dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1750         nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1751
1752         if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1753                 return ERR_PTR(-ENAMETOOLONG);
1754
1755         /*
1756          * If we're doing an exclusive create, optimize away the lookup
1757          * but don't hash the dentry.
1758          */
1759         if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1760                 return NULL;
1761
1762         res = ERR_PTR(-ENOMEM);
1763         fhandle = nfs_alloc_fhandle();
1764         fattr = nfs_alloc_fattr();
1765         if (fhandle == NULL || fattr == NULL)
1766                 goto out;
1767
1768         label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1769         if (IS_ERR(label))
1770                 goto out;
1771
1772         dir_verifier = nfs_save_change_attribute(dir);
1773         trace_nfs_lookup_enter(dir, dentry, flags);
1774         error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1775         if (error == -ENOENT)
1776                 goto no_entry;
1777         if (error < 0) {
1778                 res = ERR_PTR(error);
1779                 goto out_label;
1780         }
1781         inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1782         res = ERR_CAST(inode);
1783         if (IS_ERR(res))
1784                 goto out_label;
1785
1786         /* Notify readdir to use READDIRPLUS */
1787         nfs_force_use_readdirplus(dir);
1788
1789 no_entry:
1790         res = d_splice_alias(inode, dentry);
1791         if (res != NULL) {
1792                 if (IS_ERR(res))
1793                         goto out_label;
1794                 dentry = res;
1795         }
1796         nfs_set_verifier(dentry, dir_verifier);
1797 out_label:
1798         trace_nfs_lookup_exit(dir, dentry, flags, error);
1799         nfs4_label_free(label);
1800 out:
1801         nfs_free_fattr(fattr);
1802         nfs_free_fhandle(fhandle);
1803         return res;
1804 }
1805 EXPORT_SYMBOL_GPL(nfs_lookup);
1806
1807 #if IS_ENABLED(CONFIG_NFS_V4)
1808 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1809
1810 const struct dentry_operations nfs4_dentry_operations = {
1811         .d_revalidate   = nfs4_lookup_revalidate,
1812         .d_weak_revalidate      = nfs_weak_revalidate,
1813         .d_delete       = nfs_dentry_delete,
1814         .d_iput         = nfs_dentry_iput,
1815         .d_automount    = nfs_d_automount,
1816         .d_release      = nfs_d_release,
1817 };
1818 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1819
1820 static fmode_t flags_to_mode(int flags)
1821 {
1822         fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1823         if ((flags & O_ACCMODE) != O_WRONLY)
1824                 res |= FMODE_READ;
1825         if ((flags & O_ACCMODE) != O_RDONLY)
1826                 res |= FMODE_WRITE;
1827         return res;
1828 }
1829
1830 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1831 {
1832         return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1833 }
1834
1835 static int do_open(struct inode *inode, struct file *filp)
1836 {
1837         nfs_fscache_open_file(inode, filp);
1838         return 0;
1839 }
1840
1841 static int nfs_finish_open(struct nfs_open_context *ctx,
1842                            struct dentry *dentry,
1843                            struct file *file, unsigned open_flags)
1844 {
1845         int err;
1846
1847         err = finish_open(file, dentry, do_open);
1848         if (err)
1849                 goto out;
1850         if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1851                 nfs_file_set_open_context(file, ctx);
1852         else
1853                 err = -EOPENSTALE;
1854 out:
1855         return err;
1856 }
1857
1858 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1859                     struct file *file, unsigned open_flags,
1860                     umode_t mode)
1861 {
1862         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1863         struct nfs_open_context *ctx;
1864         struct dentry *res;
1865         struct iattr attr = { .ia_valid = ATTR_OPEN };
1866         struct inode *inode;
1867         unsigned int lookup_flags = 0;
1868         bool switched = false;
1869         int created = 0;
1870         int err;
1871
1872         /* Expect a negative dentry */
1873         BUG_ON(d_inode(dentry));
1874
1875         dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1876                         dir->i_sb->s_id, dir->i_ino, dentry);
1877
1878         err = nfs_check_flags(open_flags);
1879         if (err)
1880                 return err;
1881
1882         /* NFS only supports OPEN on regular files */
1883         if ((open_flags & O_DIRECTORY)) {
1884                 if (!d_in_lookup(dentry)) {
1885                         /*
1886                          * Hashed negative dentry with O_DIRECTORY: dentry was
1887                          * revalidated and is fine, no need to perform lookup
1888                          * again
1889                          */
1890                         return -ENOENT;
1891                 }
1892                 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1893                 goto no_open;
1894         }
1895
1896         if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1897                 return -ENAMETOOLONG;
1898
1899         if (open_flags & O_CREAT) {
1900                 struct nfs_server *server = NFS_SERVER(dir);
1901
1902                 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1903                         mode &= ~current_umask();
1904
1905                 attr.ia_valid |= ATTR_MODE;
1906                 attr.ia_mode = mode;
1907         }
1908         if (open_flags & O_TRUNC) {
1909                 attr.ia_valid |= ATTR_SIZE;
1910                 attr.ia_size = 0;
1911         }
1912
1913         if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1914                 d_drop(dentry);
1915                 switched = true;
1916                 dentry = d_alloc_parallel(dentry->d_parent,
1917                                           &dentry->d_name, &wq);
1918                 if (IS_ERR(dentry))
1919                         return PTR_ERR(dentry);
1920                 if (unlikely(!d_in_lookup(dentry)))
1921                         return finish_no_open(file, dentry);
1922         }
1923
1924         ctx = create_nfs_open_context(dentry, open_flags, file);
1925         err = PTR_ERR(ctx);
1926         if (IS_ERR(ctx))
1927                 goto out;
1928
1929         trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1930         inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1931         if (created)
1932                 file->f_mode |= FMODE_CREATED;
1933         if (IS_ERR(inode)) {
1934                 err = PTR_ERR(inode);
1935                 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1936                 put_nfs_open_context(ctx);
1937                 d_drop(dentry);
1938                 switch (err) {
1939                 case -ENOENT:
1940                         d_splice_alias(NULL, dentry);
1941                         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1942                         break;
1943                 case -EISDIR:
1944                 case -ENOTDIR:
1945                         goto no_open;
1946                 case -ELOOP:
1947                         if (!(open_flags & O_NOFOLLOW))
1948                                 goto no_open;
1949                         break;
1950                         /* case -EINVAL: */
1951                 default:
1952                         break;
1953                 }
1954                 goto out;
1955         }
1956
1957         err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1958         trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1959         put_nfs_open_context(ctx);
1960 out:
1961         if (unlikely(switched)) {
1962                 d_lookup_done(dentry);
1963                 dput(dentry);
1964         }
1965         return err;
1966
1967 no_open:
1968         res = nfs_lookup(dir, dentry, lookup_flags);
1969         if (switched) {
1970                 d_lookup_done(dentry);
1971                 if (!res)
1972                         res = dentry;
1973                 else
1974                         dput(dentry);
1975         }
1976         if (IS_ERR(res))
1977                 return PTR_ERR(res);
1978         return finish_no_open(file, res);
1979 }
1980 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1981
1982 static int
1983 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1984                           unsigned int flags)
1985 {
1986         struct inode *inode;
1987
1988         if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1989                 goto full_reval;
1990         if (d_mountpoint(dentry))
1991                 goto full_reval;
1992
1993         inode = d_inode(dentry);
1994
1995         /* We can't create new files in nfs_open_revalidate(), so we
1996          * optimize away revalidation of negative dentries.
1997          */
1998         if (inode == NULL)
1999                 goto full_reval;
2000
2001         if (nfs_verifier_is_delegated(dentry))
2002                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2003
2004         /* NFS only supports OPEN on regular files */
2005         if (!S_ISREG(inode->i_mode))
2006                 goto full_reval;
2007
2008         /* We cannot do exclusive creation on a positive dentry */
2009         if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2010                 goto reval_dentry;
2011
2012         /* Check if the directory changed */
2013         if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2014                 goto reval_dentry;
2015
2016         /* Let f_op->open() actually open (and revalidate) the file */
2017         return 1;
2018 reval_dentry:
2019         if (flags & LOOKUP_RCU)
2020                 return -ECHILD;
2021         return nfs_lookup_revalidate_dentry(dir, dentry, inode);
2022
2023 full_reval:
2024         return nfs_do_lookup_revalidate(dir, dentry, flags);
2025 }
2026
2027 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2028 {
2029         return __nfs_lookup_revalidate(dentry, flags,
2030                         nfs4_do_lookup_revalidate);
2031 }
2032
2033 #endif /* CONFIG_NFSV4 */
2034
2035 struct dentry *
2036 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2037                                 struct nfs_fattr *fattr,
2038                                 struct nfs4_label *label)
2039 {
2040         struct dentry *parent = dget_parent(dentry);
2041         struct inode *dir = d_inode(parent);
2042         struct inode *inode;
2043         struct dentry *d;
2044         int error;
2045
2046         d_drop(dentry);
2047
2048         if (fhandle->size == 0) {
2049                 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
2050                 if (error)
2051                         goto out_error;
2052         }
2053         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2054         if (!(fattr->valid & NFS_ATTR_FATTR)) {
2055                 struct nfs_server *server = NFS_SB(dentry->d_sb);
2056                 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2057                                 fattr, NULL, NULL);
2058                 if (error < 0)
2059                         goto out_error;
2060         }
2061         inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
2062         d = d_splice_alias(inode, dentry);
2063 out:
2064         dput(parent);
2065         return d;
2066 out_error:
2067         nfs_mark_for_revalidate(dir);
2068         d = ERR_PTR(error);
2069         goto out;
2070 }
2071 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2072
2073 /*
2074  * Code common to create, mkdir, and mknod.
2075  */
2076 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2077                                 struct nfs_fattr *fattr,
2078                                 struct nfs4_label *label)
2079 {
2080         struct dentry *d;
2081
2082         d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
2083         if (IS_ERR(d))
2084                 return PTR_ERR(d);
2085
2086         /* Callers don't care */
2087         dput(d);
2088         return 0;
2089 }
2090 EXPORT_SYMBOL_GPL(nfs_instantiate);
2091
2092 /*
2093  * Following a failed create operation, we drop the dentry rather
2094  * than retain a negative dentry. This avoids a problem in the event
2095  * that the operation succeeded on the server, but an error in the
2096  * reply path made it appear to have failed.
2097  */
2098 int nfs_create(struct inode *dir, struct dentry *dentry,
2099                 umode_t mode, bool excl)
2100 {
2101         struct iattr attr;
2102         int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2103         int error;
2104
2105         dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2106                         dir->i_sb->s_id, dir->i_ino, dentry);
2107
2108         attr.ia_mode = mode;
2109         attr.ia_valid = ATTR_MODE;
2110
2111         trace_nfs_create_enter(dir, dentry, open_flags);
2112         error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2113         trace_nfs_create_exit(dir, dentry, open_flags, error);
2114         if (error != 0)
2115                 goto out_err;
2116         return 0;
2117 out_err:
2118         d_drop(dentry);
2119         return error;
2120 }
2121 EXPORT_SYMBOL_GPL(nfs_create);
2122
2123 /*
2124  * See comments for nfs_proc_create regarding failed operations.
2125  */
2126 int
2127 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
2128 {
2129         struct iattr attr;
2130         int status;
2131
2132         dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2133                         dir->i_sb->s_id, dir->i_ino, dentry);
2134
2135         attr.ia_mode = mode;
2136         attr.ia_valid = ATTR_MODE;
2137
2138         trace_nfs_mknod_enter(dir, dentry);
2139         status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2140         trace_nfs_mknod_exit(dir, dentry, status);
2141         if (status != 0)
2142                 goto out_err;
2143         return 0;
2144 out_err:
2145         d_drop(dentry);
2146         return status;
2147 }
2148 EXPORT_SYMBOL_GPL(nfs_mknod);
2149
2150 /*
2151  * See comments for nfs_proc_create regarding failed operations.
2152  */
2153 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2154 {
2155         struct iattr attr;
2156         int error;
2157
2158         dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2159                         dir->i_sb->s_id, dir->i_ino, dentry);
2160
2161         attr.ia_valid = ATTR_MODE;
2162         attr.ia_mode = mode | S_IFDIR;
2163
2164         trace_nfs_mkdir_enter(dir, dentry);
2165         error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2166         trace_nfs_mkdir_exit(dir, dentry, error);
2167         if (error != 0)
2168                 goto out_err;
2169         return 0;
2170 out_err:
2171         d_drop(dentry);
2172         return error;
2173 }
2174 EXPORT_SYMBOL_GPL(nfs_mkdir);
2175
2176 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2177 {
2178         if (simple_positive(dentry))
2179                 d_delete(dentry);
2180 }
2181
2182 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2183 {
2184         int error;
2185
2186         dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2187                         dir->i_sb->s_id, dir->i_ino, dentry);
2188
2189         trace_nfs_rmdir_enter(dir, dentry);
2190         if (d_really_is_positive(dentry)) {
2191                 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2192                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2193                 /* Ensure the VFS deletes this inode */
2194                 switch (error) {
2195                 case 0:
2196                         clear_nlink(d_inode(dentry));
2197                         break;
2198                 case -ENOENT:
2199                         nfs_dentry_handle_enoent(dentry);
2200                 }
2201                 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2202         } else
2203                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2204         trace_nfs_rmdir_exit(dir, dentry, error);
2205
2206         return error;
2207 }
2208 EXPORT_SYMBOL_GPL(nfs_rmdir);
2209
2210 /*
2211  * Remove a file after making sure there are no pending writes,
2212  * and after checking that the file has only one user.
2213  *
2214  * We invalidate the attribute cache and free the inode prior to the operation
2215  * to avoid possible races if the server reuses the inode.
2216  */
2217 static int nfs_safe_remove(struct dentry *dentry)
2218 {
2219         struct inode *dir = d_inode(dentry->d_parent);
2220         struct inode *inode = d_inode(dentry);
2221         int error = -EBUSY;
2222
2223         dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2224
2225         /* If the dentry was sillyrenamed, we simply call d_delete() */
2226         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2227                 error = 0;
2228                 goto out;
2229         }
2230
2231         trace_nfs_remove_enter(dir, dentry);
2232         if (inode != NULL) {
2233                 error = NFS_PROTO(dir)->remove(dir, dentry);
2234                 if (error == 0)
2235                         nfs_drop_nlink(inode);
2236         } else
2237                 error = NFS_PROTO(dir)->remove(dir, dentry);
2238         if (error == -ENOENT)
2239                 nfs_dentry_handle_enoent(dentry);
2240         trace_nfs_remove_exit(dir, dentry, error);
2241 out:
2242         return error;
2243 }
2244
2245 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
2246  *  belongs to an active ".nfs..." file and we return -EBUSY.
2247  *
2248  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
2249  */
2250 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2251 {
2252         int error;
2253         int need_rehash = 0;
2254
2255         dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2256                 dir->i_ino, dentry);
2257
2258         trace_nfs_unlink_enter(dir, dentry);
2259         spin_lock(&dentry->d_lock);
2260         if (d_count(dentry) > 1) {
2261                 spin_unlock(&dentry->d_lock);
2262                 /* Start asynchronous writeout of the inode */
2263                 write_inode_now(d_inode(dentry), 0);
2264                 error = nfs_sillyrename(dir, dentry);
2265                 goto out;
2266         }
2267         if (!d_unhashed(dentry)) {
2268                 __d_drop(dentry);
2269                 need_rehash = 1;
2270         }
2271         spin_unlock(&dentry->d_lock);
2272         error = nfs_safe_remove(dentry);
2273         if (!error || error == -ENOENT) {
2274                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2275         } else if (need_rehash)
2276                 d_rehash(dentry);
2277 out:
2278         trace_nfs_unlink_exit(dir, dentry, error);
2279         return error;
2280 }
2281 EXPORT_SYMBOL_GPL(nfs_unlink);
2282
2283 /*
2284  * To create a symbolic link, most file systems instantiate a new inode,
2285  * add a page to it containing the path, then write it out to the disk
2286  * using prepare_write/commit_write.
2287  *
2288  * Unfortunately the NFS client can't create the in-core inode first
2289  * because it needs a file handle to create an in-core inode (see
2290  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
2291  * symlink request has completed on the server.
2292  *
2293  * So instead we allocate a raw page, copy the symname into it, then do
2294  * the SYMLINK request with the page as the buffer.  If it succeeds, we
2295  * now have a new file handle and can instantiate an in-core NFS inode
2296  * and move the raw page into its mapping.
2297  */
2298 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2299 {
2300         struct page *page;
2301         char *kaddr;
2302         struct iattr attr;
2303         unsigned int pathlen = strlen(symname);
2304         int error;
2305
2306         dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2307                 dir->i_ino, dentry, symname);
2308
2309         if (pathlen > PAGE_SIZE)
2310                 return -ENAMETOOLONG;
2311
2312         attr.ia_mode = S_IFLNK | S_IRWXUGO;
2313         attr.ia_valid = ATTR_MODE;
2314
2315         page = alloc_page(GFP_USER);
2316         if (!page)
2317                 return -ENOMEM;
2318
2319         kaddr = page_address(page);
2320         memcpy(kaddr, symname, pathlen);
2321         if (pathlen < PAGE_SIZE)
2322                 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2323
2324         trace_nfs_symlink_enter(dir, dentry);
2325         error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2326         trace_nfs_symlink_exit(dir, dentry, error);
2327         if (error != 0) {
2328                 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2329                         dir->i_sb->s_id, dir->i_ino,
2330                         dentry, symname, error);
2331                 d_drop(dentry);
2332                 __free_page(page);
2333                 return error;
2334         }
2335
2336         /*
2337          * No big deal if we can't add this page to the page cache here.
2338          * READLINK will get the missing page from the server if needed.
2339          */
2340         if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2341                                                         GFP_KERNEL)) {
2342                 SetPageUptodate(page);
2343                 unlock_page(page);
2344                 /*
2345                  * add_to_page_cache_lru() grabs an extra page refcount.
2346                  * Drop it here to avoid leaking this page later.
2347                  */
2348                 put_page(page);
2349         } else
2350                 __free_page(page);
2351
2352         return 0;
2353 }
2354 EXPORT_SYMBOL_GPL(nfs_symlink);
2355
2356 int
2357 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2358 {
2359         struct inode *inode = d_inode(old_dentry);
2360         int error;
2361
2362         dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2363                 old_dentry, dentry);
2364
2365         trace_nfs_link_enter(inode, dir, dentry);
2366         d_drop(dentry);
2367         error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2368         if (error == 0) {
2369                 ihold(inode);
2370                 d_add(dentry, inode);
2371         }
2372         trace_nfs_link_exit(inode, dir, dentry, error);
2373         return error;
2374 }
2375 EXPORT_SYMBOL_GPL(nfs_link);
2376
2377 /*
2378  * RENAME
2379  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2380  * different file handle for the same inode after a rename (e.g. when
2381  * moving to a different directory). A fail-safe method to do so would
2382  * be to look up old_dir/old_name, create a link to new_dir/new_name and
2383  * rename the old file using the sillyrename stuff. This way, the original
2384  * file in old_dir will go away when the last process iput()s the inode.
2385  *
2386  * FIXED.
2387  *
2388  * It actually works quite well. One needs to have the possibility for
2389  * at least one ".nfs..." file in each directory the file ever gets
2390  * moved or linked to which happens automagically with the new
2391  * implementation that only depends on the dcache stuff instead of
2392  * using the inode layer
2393  *
2394  * Unfortunately, things are a little more complicated than indicated
2395  * above. For a cross-directory move, we want to make sure we can get
2396  * rid of the old inode after the operation.  This means there must be
2397  * no pending writes (if it's a file), and the use count must be 1.
2398  * If these conditions are met, we can drop the dentries before doing
2399  * the rename.
2400  */
2401 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2402                struct inode *new_dir, struct dentry *new_dentry,
2403                unsigned int flags)
2404 {
2405         struct inode *old_inode = d_inode(old_dentry);
2406         struct inode *new_inode = d_inode(new_dentry);
2407         struct dentry *dentry = NULL, *rehash = NULL;
2408         struct rpc_task *task;
2409         int error = -EBUSY;
2410
2411         if (flags)
2412                 return -EINVAL;
2413
2414         dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2415                  old_dentry, new_dentry,
2416                  d_count(new_dentry));
2417
2418         trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2419         /*
2420          * For non-directories, check whether the target is busy and if so,
2421          * make a copy of the dentry and then do a silly-rename. If the
2422          * silly-rename succeeds, the copied dentry is hashed and becomes
2423          * the new target.
2424          */
2425         if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2426                 /*
2427                  * To prevent any new references to the target during the
2428                  * rename, we unhash the dentry in advance.
2429                  */
2430                 if (!d_unhashed(new_dentry)) {
2431                         d_drop(new_dentry);
2432                         rehash = new_dentry;
2433                 }
2434
2435                 if (d_count(new_dentry) > 2) {
2436                         int err;
2437
2438                         /* copy the target dentry's name */
2439                         dentry = d_alloc(new_dentry->d_parent,
2440                                          &new_dentry->d_name);
2441                         if (!dentry)
2442                                 goto out;
2443
2444                         /* silly-rename the existing target ... */
2445                         err = nfs_sillyrename(new_dir, new_dentry);
2446                         if (err)
2447                                 goto out;
2448
2449                         new_dentry = dentry;
2450                         rehash = NULL;
2451                         new_inode = NULL;
2452                 }
2453         }
2454
2455         task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2456         if (IS_ERR(task)) {
2457                 error = PTR_ERR(task);
2458                 goto out;
2459         }
2460
2461         error = rpc_wait_for_completion_task(task);
2462         if (error != 0) {
2463                 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2464                 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2465                 smp_wmb();
2466         } else
2467                 error = task->tk_status;
2468         rpc_put_task(task);
2469         /* Ensure the inode attributes are revalidated */
2470         if (error == 0) {
2471                 spin_lock(&old_inode->i_lock);
2472                 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2473                 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2474                         | NFS_INO_INVALID_CTIME
2475                         | NFS_INO_REVAL_FORCED;
2476                 spin_unlock(&old_inode->i_lock);
2477         }
2478 out:
2479         if (rehash)
2480                 d_rehash(rehash);
2481         trace_nfs_rename_exit(old_dir, old_dentry,
2482                         new_dir, new_dentry, error);
2483         if (!error) {
2484                 if (new_inode != NULL)
2485                         nfs_drop_nlink(new_inode);
2486                 /*
2487                  * The d_move() should be here instead of in an async RPC completion
2488                  * handler because we need the proper locks to move the dentry.  If
2489                  * we're interrupted by a signal, the async RPC completion handler
2490                  * should mark the directories for revalidation.
2491                  */
2492                 d_move(old_dentry, new_dentry);
2493                 nfs_set_verifier(old_dentry,
2494                                         nfs_save_change_attribute(new_dir));
2495         } else if (error == -ENOENT)
2496                 nfs_dentry_handle_enoent(old_dentry);
2497
2498         /* new dentry created? */
2499         if (dentry)
2500                 dput(dentry);
2501         return error;
2502 }
2503 EXPORT_SYMBOL_GPL(nfs_rename);
2504
2505 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2506 static LIST_HEAD(nfs_access_lru_list);
2507 static atomic_long_t nfs_access_nr_entries;
2508
2509 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2510 module_param(nfs_access_max_cachesize, ulong, 0644);
2511 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2512
2513 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2514 {
2515         put_cred(entry->cred);
2516         kfree_rcu(entry, rcu_head);
2517         smp_mb__before_atomic();
2518         atomic_long_dec(&nfs_access_nr_entries);
2519         smp_mb__after_atomic();
2520 }
2521
2522 static void nfs_access_free_list(struct list_head *head)
2523 {
2524         struct nfs_access_entry *cache;
2525
2526         while (!list_empty(head)) {
2527                 cache = list_entry(head->next, struct nfs_access_entry, lru);
2528                 list_del(&cache->lru);
2529                 nfs_access_free_entry(cache);
2530         }
2531 }
2532
2533 static unsigned long
2534 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2535 {
2536         LIST_HEAD(head);
2537         struct nfs_inode *nfsi, *next;
2538         struct nfs_access_entry *cache;
2539         long freed = 0;
2540
2541         spin_lock(&nfs_access_lru_lock);
2542         list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2543                 struct inode *inode;
2544
2545                 if (nr_to_scan-- == 0)
2546                         break;
2547                 inode = &nfsi->vfs_inode;
2548                 spin_lock(&inode->i_lock);
2549                 if (list_empty(&nfsi->access_cache_entry_lru))
2550                         goto remove_lru_entry;
2551                 cache = list_entry(nfsi->access_cache_entry_lru.next,
2552                                 struct nfs_access_entry, lru);
2553                 list_move(&cache->lru, &head);
2554                 rb_erase(&cache->rb_node, &nfsi->access_cache);
2555                 freed++;
2556                 if (!list_empty(&nfsi->access_cache_entry_lru))
2557                         list_move_tail(&nfsi->access_cache_inode_lru,
2558                                         &nfs_access_lru_list);
2559                 else {
2560 remove_lru_entry:
2561                         list_del_init(&nfsi->access_cache_inode_lru);
2562                         smp_mb__before_atomic();
2563                         clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2564                         smp_mb__after_atomic();
2565                 }
2566                 spin_unlock(&inode->i_lock);
2567         }
2568         spin_unlock(&nfs_access_lru_lock);
2569         nfs_access_free_list(&head);
2570         return freed;
2571 }
2572
2573 unsigned long
2574 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2575 {
2576         int nr_to_scan = sc->nr_to_scan;
2577         gfp_t gfp_mask = sc->gfp_mask;
2578
2579         if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2580                 return SHRINK_STOP;
2581         return nfs_do_access_cache_scan(nr_to_scan);
2582 }
2583
2584
2585 unsigned long
2586 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2587 {
2588         return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2589 }
2590
2591 static void
2592 nfs_access_cache_enforce_limit(void)
2593 {
2594         long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2595         unsigned long diff;
2596         unsigned int nr_to_scan;
2597
2598         if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2599                 return;
2600         nr_to_scan = 100;
2601         diff = nr_entries - nfs_access_max_cachesize;
2602         if (diff < nr_to_scan)
2603                 nr_to_scan = diff;
2604         nfs_do_access_cache_scan(nr_to_scan);
2605 }
2606
2607 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2608 {
2609         struct rb_root *root_node = &nfsi->access_cache;
2610         struct rb_node *n;
2611         struct nfs_access_entry *entry;
2612
2613         /* Unhook entries from the cache */
2614         while ((n = rb_first(root_node)) != NULL) {
2615                 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2616                 rb_erase(n, root_node);
2617                 list_move(&entry->lru, head);
2618         }
2619         nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2620 }
2621
2622 void nfs_access_zap_cache(struct inode *inode)
2623 {
2624         LIST_HEAD(head);
2625
2626         if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2627                 return;
2628         /* Remove from global LRU init */
2629         spin_lock(&nfs_access_lru_lock);
2630         if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2631                 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2632
2633         spin_lock(&inode->i_lock);
2634         __nfs_access_zap_cache(NFS_I(inode), &head);
2635         spin_unlock(&inode->i_lock);
2636         spin_unlock(&nfs_access_lru_lock);
2637         nfs_access_free_list(&head);
2638 }
2639 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2640
2641 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2642 {
2643         struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2644
2645         while (n != NULL) {
2646                 struct nfs_access_entry *entry =
2647                         rb_entry(n, struct nfs_access_entry, rb_node);
2648                 int cmp = cred_fscmp(cred, entry->cred);
2649
2650                 if (cmp < 0)
2651                         n = n->rb_left;
2652                 else if (cmp > 0)
2653                         n = n->rb_right;
2654                 else
2655                         return entry;
2656         }
2657         return NULL;
2658 }
2659
2660 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2661 {
2662         struct nfs_inode *nfsi = NFS_I(inode);
2663         struct nfs_access_entry *cache;
2664         bool retry = true;
2665         int err;
2666
2667         spin_lock(&inode->i_lock);
2668         for(;;) {
2669                 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2670                         goto out_zap;
2671                 cache = nfs_access_search_rbtree(inode, cred);
2672                 err = -ENOENT;
2673                 if (cache == NULL)
2674                         goto out;
2675                 /* Found an entry, is our attribute cache valid? */
2676                 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2677                         break;
2678                 if (!retry)
2679                         break;
2680                 err = -ECHILD;
2681                 if (!may_block)
2682                         goto out;
2683                 spin_unlock(&inode->i_lock);
2684                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2685                 if (err)
2686                         return err;
2687                 spin_lock(&inode->i_lock);
2688                 retry = false;
2689         }
2690         res->cred = cache->cred;
2691         res->mask = cache->mask;
2692         list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2693         err = 0;
2694 out:
2695         spin_unlock(&inode->i_lock);
2696         return err;
2697 out_zap:
2698         spin_unlock(&inode->i_lock);
2699         nfs_access_zap_cache(inode);
2700         return -ENOENT;
2701 }
2702
2703 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2704 {
2705         /* Only check the most recently returned cache entry,
2706          * but do it without locking.
2707          */
2708         struct nfs_inode *nfsi = NFS_I(inode);
2709         struct nfs_access_entry *cache;
2710         int err = -ECHILD;
2711         struct list_head *lh;
2712
2713         rcu_read_lock();
2714         if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2715                 goto out;
2716         lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2717         cache = list_entry(lh, struct nfs_access_entry, lru);
2718         if (lh == &nfsi->access_cache_entry_lru ||
2719             cred_fscmp(cred, cache->cred) != 0)
2720                 cache = NULL;
2721         if (cache == NULL)
2722                 goto out;
2723         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2724                 goto out;
2725         res->cred = cache->cred;
2726         res->mask = cache->mask;
2727         err = 0;
2728 out:
2729         rcu_read_unlock();
2730         return err;
2731 }
2732
2733 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct
2734 nfs_access_entry *res, bool may_block)
2735 {
2736         int status;
2737
2738         status = nfs_access_get_cached_rcu(inode, cred, res);
2739         if (status != 0)
2740                 status = nfs_access_get_cached_locked(inode, cred, res,
2741                     may_block);
2742
2743         return status;
2744 }
2745 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
2746
2747 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2748 {
2749         struct nfs_inode *nfsi = NFS_I(inode);
2750         struct rb_root *root_node = &nfsi->access_cache;
2751         struct rb_node **p = &root_node->rb_node;
2752         struct rb_node *parent = NULL;
2753         struct nfs_access_entry *entry;
2754         int cmp;
2755
2756         spin_lock(&inode->i_lock);
2757         while (*p != NULL) {
2758                 parent = *p;
2759                 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2760                 cmp = cred_fscmp(set->cred, entry->cred);
2761
2762                 if (cmp < 0)
2763                         p = &parent->rb_left;
2764                 else if (cmp > 0)
2765                         p = &parent->rb_right;
2766                 else
2767                         goto found;
2768         }
2769         rb_link_node(&set->rb_node, parent, p);
2770         rb_insert_color(&set->rb_node, root_node);
2771         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2772         spin_unlock(&inode->i_lock);
2773         return;
2774 found:
2775         rb_replace_node(parent, &set->rb_node, root_node);
2776         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2777         list_del(&entry->lru);
2778         spin_unlock(&inode->i_lock);
2779         nfs_access_free_entry(entry);
2780 }
2781
2782 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2783 {
2784         struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2785         if (cache == NULL)
2786                 return;
2787         RB_CLEAR_NODE(&cache->rb_node);
2788         cache->cred = get_cred(set->cred);
2789         cache->mask = set->mask;
2790
2791         /* The above field assignments must be visible
2792          * before this item appears on the lru.  We cannot easily
2793          * use rcu_assign_pointer, so just force the memory barrier.
2794          */
2795         smp_wmb();
2796         nfs_access_add_rbtree(inode, cache);
2797
2798         /* Update accounting */
2799         smp_mb__before_atomic();
2800         atomic_long_inc(&nfs_access_nr_entries);
2801         smp_mb__after_atomic();
2802
2803         /* Add inode to global LRU list */
2804         if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2805                 spin_lock(&nfs_access_lru_lock);
2806                 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2807                         list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2808                                         &nfs_access_lru_list);
2809                 spin_unlock(&nfs_access_lru_lock);
2810         }
2811         nfs_access_cache_enforce_limit();
2812 }
2813 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2814
2815 #define NFS_MAY_READ (NFS_ACCESS_READ)
2816 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2817                 NFS_ACCESS_EXTEND | \
2818                 NFS_ACCESS_DELETE)
2819 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2820                 NFS_ACCESS_EXTEND)
2821 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2822 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2823 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2824 static int
2825 nfs_access_calc_mask(u32 access_result, umode_t umode)
2826 {
2827         int mask = 0;
2828
2829         if (access_result & NFS_MAY_READ)
2830                 mask |= MAY_READ;
2831         if (S_ISDIR(umode)) {
2832                 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2833                         mask |= MAY_WRITE;
2834                 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2835                         mask |= MAY_EXEC;
2836         } else if (S_ISREG(umode)) {
2837                 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2838                         mask |= MAY_WRITE;
2839                 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2840                         mask |= MAY_EXEC;
2841         } else if (access_result & NFS_MAY_WRITE)
2842                         mask |= MAY_WRITE;
2843         return mask;
2844 }
2845
2846 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2847 {
2848         entry->mask = access_result;
2849 }
2850 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2851
2852 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2853 {
2854         struct nfs_access_entry cache;
2855         bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2856         int cache_mask = -1;
2857         int status;
2858
2859         trace_nfs_access_enter(inode);
2860
2861         status = nfs_access_get_cached(inode, cred, &cache, may_block);
2862         if (status == 0)
2863                 goto out_cached;
2864
2865         status = -ECHILD;
2866         if (!may_block)
2867                 goto out;
2868
2869         /*
2870          * Determine which access bits we want to ask for...
2871          */
2872         cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2873         if (nfs_server_capable(inode, NFS_CAP_XATTR)) {
2874                 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE |
2875                     NFS_ACCESS_XALIST;
2876         }
2877         if (S_ISDIR(inode->i_mode))
2878                 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2879         else
2880                 cache.mask |= NFS_ACCESS_EXECUTE;
2881         cache.cred = cred;
2882         status = NFS_PROTO(inode)->access(inode, &cache);
2883         if (status != 0) {
2884                 if (status == -ESTALE) {
2885                         if (!S_ISDIR(inode->i_mode))
2886                                 nfs_set_inode_stale(inode);
2887                         else
2888                                 nfs_zap_caches(inode);
2889                 }
2890                 goto out;
2891         }
2892         nfs_access_add_cache(inode, &cache);
2893 out_cached:
2894         cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2895         if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2896                 status = -EACCES;
2897 out:
2898         trace_nfs_access_exit(inode, mask, cache_mask, status);
2899         return status;
2900 }
2901
2902 static int nfs_open_permission_mask(int openflags)
2903 {
2904         int mask = 0;
2905
2906         if (openflags & __FMODE_EXEC) {
2907                 /* ONLY check exec rights */
2908                 mask = MAY_EXEC;
2909         } else {
2910                 if ((openflags & O_ACCMODE) != O_WRONLY)
2911                         mask |= MAY_READ;
2912                 if ((openflags & O_ACCMODE) != O_RDONLY)
2913                         mask |= MAY_WRITE;
2914         }
2915
2916         return mask;
2917 }
2918
2919 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2920 {
2921         return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2922 }
2923 EXPORT_SYMBOL_GPL(nfs_may_open);
2924
2925 static int nfs_execute_ok(struct inode *inode, int mask)
2926 {
2927         struct nfs_server *server = NFS_SERVER(inode);
2928         int ret = 0;
2929
2930         if (S_ISDIR(inode->i_mode))
2931                 return 0;
2932         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2933                 if (mask & MAY_NOT_BLOCK)
2934                         return -ECHILD;
2935                 ret = __nfs_revalidate_inode(server, inode);
2936         }
2937         if (ret == 0 && !execute_ok(inode))
2938                 ret = -EACCES;
2939         return ret;
2940 }
2941
2942 int nfs_permission(struct inode *inode, int mask)
2943 {
2944         const struct cred *cred = current_cred();
2945         int res = 0;
2946
2947         nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2948
2949         if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2950                 goto out;
2951         /* Is this sys_access() ? */
2952         if (mask & (MAY_ACCESS | MAY_CHDIR))
2953                 goto force_lookup;
2954
2955         switch (inode->i_mode & S_IFMT) {
2956                 case S_IFLNK:
2957                         goto out;
2958                 case S_IFREG:
2959                         if ((mask & MAY_OPEN) &&
2960                            nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2961                                 return 0;
2962                         break;
2963                 case S_IFDIR:
2964                         /*
2965                          * Optimize away all write operations, since the server
2966                          * will check permissions when we perform the op.
2967                          */
2968                         if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2969                                 goto out;
2970         }
2971
2972 force_lookup:
2973         if (!NFS_PROTO(inode)->access)
2974                 goto out_notsup;
2975
2976         res = nfs_do_access(inode, cred, mask);
2977 out:
2978         if (!res && (mask & MAY_EXEC))
2979                 res = nfs_execute_ok(inode, mask);
2980
2981         dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2982                 inode->i_sb->s_id, inode->i_ino, mask, res);
2983         return res;
2984 out_notsup:
2985         if (mask & MAY_NOT_BLOCK)
2986                 return -ECHILD;
2987
2988         res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2989         if (res == 0)
2990                 res = generic_permission(inode, mask);
2991         goto out;
2992 }
2993 EXPORT_SYMBOL_GPL(nfs_permission);
2994
2995 /*
2996  * Local variables:
2997  *  version-control: t
2998  *  kept-new-versions: 5
2999  * End:
3000  */