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                         nfs_set_cache_invalid(dir,
  85                                               NFS_INO_INVALID_DATA |
  86                                                       NFS_INO_REVAL_FORCED);
  87                 list_add(&ctx->list, &nfsi->open_files);
  88                 spin_unlock(&dir->i_lock);
  89                 return ctx;
  90         }
  91         return  ERR_PTR(-ENOMEM);
  92 }
  93
  94 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
  95 {
  96         spin_lock(&dir->i_lock);
  97         list_del(&ctx->list);
  98         spin_unlock(&dir->i_lock);
  99         kfree(ctx);
 100 }
 101
 102 /*
 103  * Open file
 104  */
 105 static int
 106 nfs_opendir(struct inode *inode, struct file *filp)
 107 {
 108         int res = 0;
 109         struct nfs_open_dir_context *ctx;
 110
 111         dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
 112
 113         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 114
 115         ctx = alloc_nfs_open_dir_context(inode);
 116         if (IS_ERR(ctx)) {
 117                 res = PTR_ERR(ctx);
 118                 goto out;
 119         }
 120         filp->private_data = ctx;
 121 out:
 122         return res;
 123 }
 124
 125 static int
 126 nfs_closedir(struct inode *inode, struct file *filp)
 127 {
 128         put_nfs_open_dir_context(file_inode(filp), filp->private_data);
 129         return 0;
 130 }
 131
 132 struct nfs_cache_array_entry {
 133         u64 cookie;
 134         u64 ino;
 135         const char *name;
 136         unsigned int name_len;
 137         unsigned char d_type;
 138 };
 139
 140 struct nfs_cache_array {
 141         u64 last_cookie;
 142         unsigned int size;
 143         unsigned char page_full : 1,
 144                       page_is_eof : 1,
 145                       cookies_are_ordered : 1;
 146         struct nfs_cache_array_entry array[];
 147 };
 148
 149 struct nfs_readdir_descriptor {
 150         struct file     *file;
 151         struct page     *page;
 152         struct dir_context *ctx;
 153         pgoff_t         page_index;
 154         u64             dir_cookie;
 155         u64             last_cookie;
 156         u64             dup_cookie;
 157         loff_t          current_index;
 158         loff_t          prev_index;
 159
 160         __be32          verf[NFS_DIR_VERIFIER_SIZE];
 161         unsigned long   dir_verifier;
 162         unsigned long   timestamp;
 163         unsigned long   gencount;
 164         unsigned long   attr_gencount;
 165         unsigned int    cache_entry_index;
 166         signed char duped;
 167         bool plus;
 168         bool eof;
 169 };
 170
 171 static void nfs_readdir_array_init(struct nfs_cache_array *array)
 172 {
 173         memset(array, 0, sizeof(struct nfs_cache_array));
 174 }
 175
 176 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie)
 177 {
 178         struct nfs_cache_array *array;
 179
 180         array = kmap_atomic(page);
 181         nfs_readdir_array_init(array);
 182         array->last_cookie = last_cookie;
 183         array->cookies_are_ordered = 1;
 184         kunmap_atomic(array);
 185 }
 186
 187 /*
 188  * we are freeing strings created by nfs_add_to_readdir_array()
 189  */
 190 static
 191 void nfs_readdir_clear_array(struct page *page)
 192 {
 193         struct nfs_cache_array *array;
 194         int i;
 195
 196         array = kmap_atomic(page);
 197         for (i = 0; i < array->size; i++)
 198                 kfree(array->array[i].name);
 199         nfs_readdir_array_init(array);
 200         kunmap_atomic(array);
 201 }
 202
 203 static struct page *
 204 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
 205 {
 206         struct page *page = alloc_page(gfp_flags);
 207         if (page)
 208                 nfs_readdir_page_init_array(page, last_cookie);
 209         return page;
 210 }
 211
 212 static void nfs_readdir_page_array_free(struct page *page)
 213 {
 214         if (page) {
 215                 nfs_readdir_clear_array(page);
 216                 put_page(page);
 217         }
 218 }
 219
 220 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
 221 {
 222         array->page_is_eof = 1;
 223         array->page_full = 1;
 224 }
 225
 226 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
 227 {
 228         return array->page_full;
 229 }
 230
 231 /*
 232  * the caller is responsible for freeing qstr.name
 233  * when called by nfs_readdir_add_to_array, the strings will be freed in
 234  * nfs_clear_readdir_array()
 235  */
 236 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
 237 {
 238         const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
 239
 240         /*
 241          * Avoid a kmemleak false positive. The pointer to the name is stored
 242          * in a page cache page which kmemleak does not scan.
 243          */
 244         if (ret != NULL)
 245                 kmemleak_not_leak(ret);
 246         return ret;
 247 }
 248
 249 /*
 250  * Check that the next array entry lies entirely within the page bounds
 251  */
 252 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
 253 {
 254         struct nfs_cache_array_entry *cache_entry;
 255
 256         if (array->page_full)
 257                 return -ENOSPC;
 258         cache_entry = &array->array[array->size + 1];
 259         if ((char *)cache_entry - (char *)array > PAGE_SIZE) {
 260                 array->page_full = 1;
 261                 return -ENOSPC;
 262         }
 263         return 0;
 264 }
 265
 266 static
 267 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
 268 {
 269         struct nfs_cache_array *array;
 270         struct nfs_cache_array_entry *cache_entry;
 271         const char *name;
 272         int ret;
 273
 274         name = nfs_readdir_copy_name(entry->name, entry->len);
 275         if (!name)
 276                 return -ENOMEM;
 277
 278         array = kmap_atomic(page);
 279         ret = nfs_readdir_array_can_expand(array);
 280         if (ret) {
 281                 kfree(name);
 282                 goto out;
 283         }
 284
 285         cache_entry = &array->array[array->size];
 286         cache_entry->cookie = entry->prev_cookie;
 287         cache_entry->ino = entry->ino;
 288         cache_entry->d_type = entry->d_type;
 289         cache_entry->name_len = entry->len;
 290         cache_entry->name = name;
 291         array->last_cookie = entry->cookie;
 292         if (array->last_cookie <= cache_entry->cookie)
 293                 array->cookies_are_ordered = 0;
 294         array->size++;
 295         if (entry->eof != 0)
 296                 nfs_readdir_array_set_eof(array);
 297 out:
 298         kunmap_atomic(array);
 299         return ret;
 300 }
 301
 302 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
 303                                                 pgoff_t index, u64 last_cookie)
 304 {
 305         struct page *page;
 306
 307         page = grab_cache_page(mapping, index);
 308         if (page && !PageUptodate(page)) {
 309                 nfs_readdir_page_init_array(page, last_cookie);
 310                 if (invalidate_inode_pages2_range(mapping, index + 1, -1) < 0)
 311                         nfs_zap_mapping(mapping->host, mapping);
 312                 SetPageUptodate(page);
 313         }
 314
 315         return page;
 316 }
 317
 318 static u64 nfs_readdir_page_last_cookie(struct page *page)
 319 {
 320         struct nfs_cache_array *array;
 321         u64 ret;
 322
 323         array = kmap_atomic(page);
 324         ret = array->last_cookie;
 325         kunmap_atomic(array);
 326         return ret;
 327 }
 328
 329 static bool nfs_readdir_page_needs_filling(struct page *page)
 330 {
 331         struct nfs_cache_array *array;
 332         bool ret;
 333
 334         array = kmap_atomic(page);
 335         ret = !nfs_readdir_array_is_full(array);
 336         kunmap_atomic(array);
 337         return ret;
 338 }
 339
 340 static void nfs_readdir_page_set_eof(struct page *page)
 341 {
 342         struct nfs_cache_array *array;
 343
 344         array = kmap_atomic(page);
 345         nfs_readdir_array_set_eof(array);
 346         kunmap_atomic(array);
 347 }
 348
 349 static void nfs_readdir_page_unlock_and_put(struct page *page)
 350 {
 351         unlock_page(page);
 352         put_page(page);
 353 }
 354
 355 static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
 356                                               pgoff_t index, u64 cookie)
 357 {
 358         struct page *page;
 359
 360         page = nfs_readdir_page_get_locked(mapping, index, cookie);
 361         if (page) {
 362                 if (nfs_readdir_page_last_cookie(page) == cookie)
 363                         return page;
 364                 nfs_readdir_page_unlock_and_put(page);
 365         }
 366         return NULL;
 367 }
 368
 369 static inline
 370 int is_32bit_api(void)
 371 {
 372 #ifdef CONFIG_COMPAT
 373         return in_compat_syscall();
 374 #else
 375         return (BITS_PER_LONG == 32);
 376 #endif
 377 }
 378
 379 static
 380 bool nfs_readdir_use_cookie(const struct file *filp)
 381 {
 382         if ((filp->f_mode & FMODE_32BITHASH) ||
 383             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 384                 return false;
 385         return true;
 386 }
 387
 388 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
 389                                       struct nfs_readdir_descriptor *desc)
 390 {
 391         loff_t diff = desc->ctx->pos - desc->current_index;
 392         unsigned int index;
 393
 394         if (diff < 0)
 395                 goto out_eof;
 396         if (diff >= array->size) {
 397                 if (array->page_is_eof)
 398                         goto out_eof;
 399                 return -EAGAIN;
 400         }
 401
 402         index = (unsigned int)diff;
 403         desc->dir_cookie = array->array[index].cookie;
 404         desc->cache_entry_index = index;
 405         return 0;
 406 out_eof:
 407         desc->eof = true;
 408         return -EBADCOOKIE;
 409 }
 410
 411 static bool
 412 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
 413 {
 414         if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
 415                 return false;
 416         smp_rmb();
 417         return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
 418 }
 419
 420 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
 421                                               u64 cookie)
 422 {
 423         if (!array->cookies_are_ordered)
 424                 return true;
 425         /* Optimisation for monotonically increasing cookies */
 426         if (cookie >= array->last_cookie)
 427                 return false;
 428         if (array->size && cookie < array->array[0].cookie)
 429                 return false;
 430         return true;
 431 }
 432
 433 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
 434                                          struct nfs_readdir_descriptor *desc)
 435 {
 436         int i;
 437         loff_t new_pos;
 438         int status = -EAGAIN;
 439
 440         if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
 441                 goto check_eof;
 442
 443         for (i = 0; i < array->size; i++) {
 444                 if (array->array[i].cookie == desc->dir_cookie) {
 445                         struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
 446
 447                         new_pos = desc->current_index + i;
 448                         if (desc->attr_gencount != nfsi->attr_gencount ||
 449                             !nfs_readdir_inode_mapping_valid(nfsi)) {
 450                                 desc->duped = 0;
 451                                 desc->attr_gencount = nfsi->attr_gencount;
 452                         } else if (new_pos < desc->prev_index) {
 453                                 if (desc->duped > 0
 454                                     && desc->dup_cookie == desc->dir_cookie) {
 455                                         if (printk_ratelimit()) {
 456                                                 pr_notice("NFS: directory %pD2 contains a readdir loop."
 457                                                                 "Please contact your server vendor.  "
 458                                                                 "The file: %s has duplicate cookie %llu\n",
 459                                                                 desc->file, array->array[i].name, desc->dir_cookie);
 460                                         }
 461                                         status = -ELOOP;
 462                                         goto out;
 463                                 }
 464                                 desc->dup_cookie = desc->dir_cookie;
 465                                 desc->duped = -1;
 466                         }
 467                         if (nfs_readdir_use_cookie(desc->file))
 468                                 desc->ctx->pos = desc->dir_cookie;
 469                         else
 470                                 desc->ctx->pos = new_pos;
 471                         desc->prev_index = new_pos;
 472                         desc->cache_entry_index = i;
 473                         return 0;
 474                 }
 475         }
 476 check_eof:
 477         if (array->page_is_eof) {
 478                 status = -EBADCOOKIE;
 479                 if (desc->dir_cookie == array->last_cookie)
 480                         desc->eof = true;
 481         }
 482 out:
 483         return status;
 484 }
 485
 486 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
 487 {
 488         struct nfs_cache_array *array;
 489         int status;
 490
 491         array = kmap_atomic(desc->page);
 492
 493         if (desc->dir_cookie == 0)
 494                 status = nfs_readdir_search_for_pos(array, desc);
 495         else
 496                 status = nfs_readdir_search_for_cookie(array, desc);
 497
 498         if (status == -EAGAIN) {
 499                 desc->last_cookie = array->last_cookie;
 500                 desc->current_index += array->size;
 501                 desc->page_index++;
 502         }
 503         kunmap_atomic(array);
 504         return status;
 505 }
 506
 507 /* Fill a page with xdr information before transferring to the cache page */
 508 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
 509                                   __be32 *verf, u64 cookie,
 510                                   struct page **pages, size_t bufsize,
 511                                   __be32 *verf_res)
 512 {
 513         struct inode *inode = file_inode(desc->file);
 514         struct nfs_readdir_arg arg = {
 515                 .dentry = file_dentry(desc->file),
 516                 .cred = desc->file->f_cred,
 517                 .verf = verf,
 518                 .cookie = cookie,
 519                 .pages = pages,
 520                 .page_len = bufsize,
 521                 .plus = desc->plus,
 522         };
 523         struct nfs_readdir_res res = {
 524                 .verf = verf_res,
 525         };
 526         unsigned long   timestamp, gencount;
 527         int             error;
 528
 529  again:
 530         timestamp = jiffies;
 531         gencount = nfs_inc_attr_generation_counter();
 532         desc->dir_verifier = nfs_save_change_attribute(inode);
 533         error = NFS_PROTO(inode)->readdir(&arg, &res);
 534         if (error < 0) {
 535                 /* We requested READDIRPLUS, but the server doesn't grok it */
 536                 if (error == -ENOTSUPP && desc->plus) {
 537                         NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
 538                         clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 539                         desc->plus = arg.plus = false;
 540                         goto again;
 541                 }
 542                 goto error;
 543         }
 544         desc->timestamp = timestamp;
 545         desc->gencount = gencount;
 546 error:
 547         return error;
 548 }
 549
 550 static int xdr_decode(struct nfs_readdir_descriptor *desc,
 551                       struct nfs_entry *entry, struct xdr_stream *xdr)
 552 {
 553         struct inode *inode = file_inode(desc->file);
 554         int error;
 555
 556         error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
 557         if (error)
 558                 return error;
 559         entry->fattr->time_start = desc->timestamp;
 560         entry->fattr->gencount = desc->gencount;
 561         return 0;
 562 }
 563
 564 /* Match file and dirent using either filehandle or fileid
 565  * Note: caller is responsible for checking the fsid
 566  */
 567 static
 568 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
 569 {
 570         struct inode *inode;
 571         struct nfs_inode *nfsi;
 572
 573         if (d_really_is_negative(dentry))
 574                 return 0;
 575
 576         inode = d_inode(dentry);
 577         if (is_bad_inode(inode) || NFS_STALE(inode))
 578                 return 0;
 579
 580         nfsi = NFS_I(inode);
 581         if (entry->fattr->fileid != nfsi->fileid)
 582                 return 0;
 583         if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
 584                 return 0;
 585         return 1;
 586 }
 587
 588 static
 589 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
 590 {
 591         if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
 592                 return false;
 593         if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
 594                 return true;
 595         if (ctx->pos == 0)
 596                 return true;
 597         return false;
 598 }
 599
 600 /*
 601  * This function is called by the lookup and getattr code to request the
 602  * use of readdirplus to accelerate any future lookups in the same
 603  * directory.
 604  */
 605 void nfs_advise_use_readdirplus(struct inode *dir)
 606 {
 607         struct nfs_inode *nfsi = NFS_I(dir);
 608
 609         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 610             !list_empty(&nfsi->open_files))
 611                 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 612 }
 613
 614 /*
 615  * This function is mainly for use by nfs_getattr().
 616  *
 617  * If this is an 'ls -l', we want to force use of readdirplus.
 618  * Do this by checking if there is an active file descriptor
 619  * and calling nfs_advise_use_readdirplus, then forcing a
 620  * cache flush.
 621  */
 622 void nfs_force_use_readdirplus(struct inode *dir)
 623 {
 624         struct nfs_inode *nfsi = NFS_I(dir);
 625
 626         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 627             !list_empty(&nfsi->open_files)) {
 628                 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 629                 invalidate_mapping_pages(dir->i_mapping,
 630                         nfsi->page_index + 1, -1);
 631         }
 632 }
 633
 634 static
 635 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
 636                 unsigned long dir_verifier)
 637 {
 638         struct qstr filename = QSTR_INIT(entry->name, entry->len);
 639         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 640         struct dentry *dentry;
 641         struct dentry *alias;
 642         struct inode *inode;
 643         int status;
 644
 645         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
 646                 return;
 647         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
 648                 return;
 649         if (filename.len == 0)
 650                 return;
 651         /* Validate that the name doesn't contain any illegal '\0' */
 652         if (strnlen(filename.name, filename.len) != filename.len)
 653                 return;
 654         /* ...or '/' */
 655         if (strnchr(filename.name, filename.len, '/'))
 656                 return;
 657         if (filename.name[0] == '.') {
 658                 if (filename.len == 1)
 659                         return;
 660                 if (filename.len == 2 && filename.name[1] == '.')
 661                         return;
 662         }
 663         filename.hash = full_name_hash(parent, filename.name, filename.len);
 664
 665         dentry = d_lookup(parent, &filename);
 666 again:
 667         if (!dentry) {
 668                 dentry = d_alloc_parallel(parent, &filename, &wq);
 669                 if (IS_ERR(dentry))
 670                         return;
 671         }
 672         if (!d_in_lookup(dentry)) {
 673                 /* Is there a mountpoint here? If so, just exit */
 674                 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
 675                                         &entry->fattr->fsid))
 676                         goto out;
 677                 if (nfs_same_file(dentry, entry)) {
 678                         if (!entry->fh->size)
 679                                 goto out;
 680                         nfs_set_verifier(dentry, dir_verifier);
 681                         status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
 682                         if (!status)
 683                                 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
 684                         goto out;
 685                 } else {
 686                         d_invalidate(dentry);
 687                         dput(dentry);
 688                         dentry = NULL;
 689                         goto again;
 690                 }
 691         }
 692         if (!entry->fh->size) {
 693                 d_lookup_done(dentry);
 694                 goto out;
 695         }
 696
 697         inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
 698         alias = d_splice_alias(inode, dentry);
 699         d_lookup_done(dentry);
 700         if (alias) {
 701                 if (IS_ERR(alias))
 702                         goto out;
 703                 dput(dentry);
 704                 dentry = alias;
 705         }
 706         nfs_set_verifier(dentry, dir_verifier);
 707 out:
 708         dput(dentry);
 709 }
 710
 711 /* Perform conversion from xdr to cache array */
 712 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
 713                                    struct nfs_entry *entry,
 714                                    struct page **xdr_pages,
 715                                    unsigned int buflen,
 716                                    struct page **arrays,
 717                                    size_t narrays)
 718 {
 719         struct address_space *mapping = desc->file->f_mapping;
 720         struct xdr_stream stream;
 721         struct xdr_buf buf;
 722         struct page *scratch, *new, *page = *arrays;
 723         int status;
 724
 725         scratch = alloc_page(GFP_KERNEL);
 726         if (scratch == NULL)
 727                 return -ENOMEM;
 728
 729         xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
 730         xdr_set_scratch_page(&stream, scratch);
 731
 732         do {
 733                 if (entry->label)
 734                         entry->label->len = NFS4_MAXLABELLEN;
 735
 736                 status = xdr_decode(desc, entry, &stream);
 737                 if (status != 0)
 738                         break;
 739
 740                 if (desc->plus)
 741                         nfs_prime_dcache(file_dentry(desc->file), entry,
 742                                         desc->dir_verifier);
 743
 744                 status = nfs_readdir_add_to_array(entry, page);
 745                 if (status != -ENOSPC)
 746                         continue;
 747
 748                 if (page->mapping != mapping) {
 749                         if (!--narrays)
 750                                 break;
 751                         new = nfs_readdir_page_array_alloc(entry->prev_cookie,
 752                                                            GFP_KERNEL);
 753                         if (!new)
 754                                 break;
 755                         arrays++;
 756                         *arrays = page = new;
 757                 } else {
 758                         new = nfs_readdir_page_get_next(mapping,
 759                                                         page->index + 1,
 760                                                         entry->prev_cookie);
 761                         if (!new)
 762                                 break;
 763                         if (page != *arrays)
 764                                 nfs_readdir_page_unlock_and_put(page);
 765                         page = new;
 766                 }
 767                 status = nfs_readdir_add_to_array(entry, page);
 768         } while (!status && !entry->eof);
 769
 770         switch (status) {
 771         case -EBADCOOKIE:
 772                 if (entry->eof) {
 773                         nfs_readdir_page_set_eof(page);
 774                         status = 0;
 775                 }
 776                 break;
 777         case -ENOSPC:
 778         case -EAGAIN:
 779                 status = 0;
 780                 break;
 781         }
 782
 783         if (page != *arrays)
 784                 nfs_readdir_page_unlock_and_put(page);
 785
 786         put_page(scratch);
 787         return status;
 788 }
 789
 790 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
 791 {
 792         while (npages--)
 793                 put_page(pages[npages]);
 794         kfree(pages);
 795 }
 796
 797 /*
 798  * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
 799  * to nfs_readdir_free_pages()
 800  */
 801 static struct page **nfs_readdir_alloc_pages(size_t npages)
 802 {
 803         struct page **pages;
 804         size_t i;
 805
 806         pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
 807         if (!pages)
 808                 return NULL;
 809         for (i = 0; i < npages; i++) {
 810                 struct page *page = alloc_page(GFP_KERNEL);
 811                 if (page == NULL)
 812                         goto out_freepages;
 813                 pages[i] = page;
 814         }
 815         return pages;
 816
 817 out_freepages:
 818         nfs_readdir_free_pages(pages, i);
 819         return NULL;
 820 }
 821
 822 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
 823                                     __be32 *verf_arg, __be32 *verf_res,
 824                                     struct page **arrays, size_t narrays)
 825 {
 826         struct page **pages;
 827         struct page *page = *arrays;
 828         struct nfs_entry *entry;
 829         size_t array_size;
 830         struct inode *inode = file_inode(desc->file);
 831         size_t dtsize = NFS_SERVER(inode)->dtsize;
 832         int status = -ENOMEM;
 833
 834         entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 835         if (!entry)
 836                 return -ENOMEM;
 837         entry->cookie = nfs_readdir_page_last_cookie(page);
 838         entry->fh = nfs_alloc_fhandle();
 839         entry->fattr = nfs_alloc_fattr();
 840         entry->server = NFS_SERVER(inode);
 841         if (entry->fh == NULL || entry->fattr == NULL)
 842                 goto out;
 843
 844         entry->label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
 845         if (IS_ERR(entry->label)) {
 846                 status = PTR_ERR(entry->label);
 847                 goto out;
 848         }
 849
 850         array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
 851         pages = nfs_readdir_alloc_pages(array_size);
 852         if (!pages)
 853                 goto out_release_label;
 854
 855         do {
 856                 unsigned int pglen;
 857                 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie,
 858                                                 pages, dtsize,
 859                                                 verf_res);
 860                 if (status < 0)
 861                         break;
 862
 863                 pglen = status;
 864                 if (pglen == 0) {
 865                         nfs_readdir_page_set_eof(page);
 866                         break;
 867                 }
 868
 869                 status = nfs_readdir_page_filler(desc, entry, pages, pglen,
 870                                                  arrays, narrays);
 871         } while (!status && nfs_readdir_page_needs_filling(page));
 872
 873         nfs_readdir_free_pages(pages, array_size);
 874 out_release_label:
 875         nfs4_label_free(entry->label);
 876 out:
 877         nfs_free_fattr(entry->fattr);
 878         nfs_free_fhandle(entry->fh);
 879         kfree(entry);
 880         return status;
 881 }
 882
 883 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
 884 {
 885         put_page(desc->page);
 886         desc->page = NULL;
 887 }
 888
 889 static void
 890 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
 891 {
 892         unlock_page(desc->page);
 893         nfs_readdir_page_put(desc);
 894 }
 895
 896 static struct page *
 897 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
 898 {
 899         return nfs_readdir_page_get_locked(desc->file->f_mapping,
 900                                            desc->page_index,
 901                                            desc->last_cookie);
 902 }
 903
 904 /*
 905  * Returns 0 if desc->dir_cookie was found on page desc->page_index
 906  * and locks the page to prevent removal from the page cache.
 907  */
 908 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
 909 {
 910         struct inode *inode = file_inode(desc->file);
 911         struct nfs_inode *nfsi = NFS_I(inode);
 912         __be32 verf[NFS_DIR_VERIFIER_SIZE];
 913         int res;
 914
 915         desc->page = nfs_readdir_page_get_cached(desc);
 916         if (!desc->page)
 917                 return -ENOMEM;
 918         if (nfs_readdir_page_needs_filling(desc->page)) {
 919                 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
 920                                                &desc->page, 1);
 921                 if (res < 0) {
 922                         nfs_readdir_page_unlock_and_put_cached(desc);
 923                         if (res == -EBADCOOKIE || res == -ENOTSYNC) {
 924                                 invalidate_inode_pages2(desc->file->f_mapping);
 925                                 desc->page_index = 0;
 926                                 return -EAGAIN;
 927                         }
 928                         return res;
 929                 }
 930                 memcpy(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf));
 931         }
 932         res = nfs_readdir_search_array(desc);
 933         if (res == 0) {
 934                 nfsi->page_index = desc->page_index;
 935                 return 0;
 936         }
 937         nfs_readdir_page_unlock_and_put_cached(desc);
 938         return res;
 939 }
 940
 941 static bool nfs_readdir_dont_search_cache(struct nfs_readdir_descriptor *desc)
 942 {
 943         struct address_space *mapping = desc->file->f_mapping;
 944         struct inode *dir = file_inode(desc->file);
 945         unsigned int dtsize = NFS_SERVER(dir)->dtsize;
 946         loff_t size = i_size_read(dir);
 947
 948         /*
 949          * Default to uncached readdir if the page cache is empty, and
 950          * we're looking for a non-zero cookie in a large directory.
 951          */
 952         return desc->dir_cookie != 0 && mapping->nrpages == 0 && size > dtsize;
 953 }
 954
 955 /* Search for desc->dir_cookie from the beginning of the page cache */
 956 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
 957 {
 958         int res;
 959
 960         if (nfs_readdir_dont_search_cache(desc))
 961                 return -EBADCOOKIE;
 962
 963         do {
 964                 if (desc->page_index == 0) {
 965                         desc->current_index = 0;
 966                         desc->prev_index = 0;
 967                         desc->last_cookie = 0;
 968                 }
 969                 res = find_and_lock_cache_page(desc);
 970         } while (res == -EAGAIN);
 971         return res;
 972 }
 973
 974 /*
 975  * Once we've found the start of the dirent within a page: fill 'er up...
 976  */
 977 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc)
 978 {
 979         struct file     *file = desc->file;
 980         struct nfs_inode *nfsi = NFS_I(file_inode(file));
 981         struct nfs_cache_array *array;
 982         unsigned int i = 0;
 983
 984         array = kmap(desc->page);
 985         for (i = desc->cache_entry_index; i < array->size; i++) {
 986                 struct nfs_cache_array_entry *ent;
 987
 988                 ent = &array->array[i];
 989                 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
 990                     nfs_compat_user_ino64(ent->ino), ent->d_type)) {
 991                         desc->eof = true;
 992                         break;
 993                 }
 994                 memcpy(desc->verf, nfsi->cookieverf, sizeof(desc->verf));
 995                 if (i < (array->size-1))
 996                         desc->dir_cookie = array->array[i+1].cookie;
 997                 else
 998                         desc->dir_cookie = array->last_cookie;
 999                 if (nfs_readdir_use_cookie(file))
1000                         desc->ctx->pos = desc->dir_cookie;
1001                 else
1002                         desc->ctx->pos++;
1003                 if (desc->duped != 0)
1004                         desc->duped = 1;
1005         }
1006         if (array->page_is_eof)
1007                 desc->eof = true;
1008
1009         kunmap(desc->page);
1010         dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1011                         (unsigned long long)desc->dir_cookie);
1012 }
1013
1014 /*
1015  * If we cannot find a cookie in our cache, we suspect that this is
1016  * because it points to a deleted file, so we ask the server to return
1017  * whatever it thinks is the next entry. We then feed this to filldir.
1018  * If all goes well, we should then be able to find our way round the
1019  * cache on the next call to readdir_search_pagecache();
1020  *
1021  * NOTE: we cannot add the anonymous page to the pagecache because
1022  *       the data it contains might not be page aligned. Besides,
1023  *       we should already have a complete representation of the
1024  *       directory in the page cache by the time we get here.
1025  */
1026 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1027 {
1028         struct page     **arrays;
1029         size_t          i, sz = 512;
1030         __be32          verf[NFS_DIR_VERIFIER_SIZE];
1031         int             status = -ENOMEM;
1032
1033         dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1034                         (unsigned long long)desc->dir_cookie);
1035
1036         arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1037         if (!arrays)
1038                 goto out;
1039         arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1040         if (!arrays[0])
1041                 goto out;
1042
1043         desc->page_index = 0;
1044         desc->last_cookie = desc->dir_cookie;
1045         desc->duped = 0;
1046
1047         status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1048
1049         for (i = 0; !desc->eof && i < sz && arrays[i]; i++) {
1050                 desc->page = arrays[i];
1051                 nfs_do_filldir(desc);
1052         }
1053         desc->page = NULL;
1054
1055
1056         for (i = 0; i < sz && arrays[i]; i++)
1057                 nfs_readdir_page_array_free(arrays[i]);
1058 out:
1059         kfree(arrays);
1060         dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1061         return status;
1062 }
1063
1064 /* The file offset position represents the dirent entry number.  A
1065    last cookie cache takes care of the common case of reading the
1066    whole directory.
1067  */
1068 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1069 {
1070         struct dentry   *dentry = file_dentry(file);
1071         struct inode    *inode = d_inode(dentry);
1072         struct nfs_open_dir_context *dir_ctx = file->private_data;
1073         struct nfs_readdir_descriptor *desc;
1074         int res;
1075
1076         dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1077                         file, (long long)ctx->pos);
1078         nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1079
1080         /*
1081          * ctx->pos points to the dirent entry number.
1082          * *desc->dir_cookie has the cookie for the next entry. We have
1083          * to either find the entry with the appropriate number or
1084          * revalidate the cookie.
1085          */
1086         if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) {
1087                 res = nfs_revalidate_mapping(inode, file->f_mapping);
1088                 if (res < 0)
1089                         goto out;
1090         }
1091
1092         res = -ENOMEM;
1093         desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1094         if (!desc)
1095                 goto out;
1096         desc->file = file;
1097         desc->ctx = ctx;
1098         desc->plus = nfs_use_readdirplus(inode, ctx);
1099
1100         spin_lock(&file->f_lock);
1101         desc->dir_cookie = dir_ctx->dir_cookie;
1102         desc->dup_cookie = dir_ctx->dup_cookie;
1103         desc->duped = dir_ctx->duped;
1104         desc->attr_gencount = dir_ctx->attr_gencount;
1105         memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1106         spin_unlock(&file->f_lock);
1107
1108         do {
1109                 res = readdir_search_pagecache(desc);
1110
1111                 if (res == -EBADCOOKIE) {
1112                         res = 0;
1113                         /* This means either end of directory */
1114                         if (desc->dir_cookie && !desc->eof) {
1115                                 /* Or that the server has 'lost' a cookie */
1116                                 res = uncached_readdir(desc);
1117                                 if (res == 0)
1118                                         continue;
1119                                 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1120                                         res = 0;
1121                         }
1122                         break;
1123                 }
1124                 if (res == -ETOOSMALL && desc->plus) {
1125                         clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
1126                         nfs_zap_caches(inode);
1127                         desc->page_index = 0;
1128                         desc->plus = false;
1129                         desc->eof = false;
1130                         continue;
1131                 }
1132                 if (res < 0)
1133                         break;
1134
1135                 nfs_do_filldir(desc);
1136                 nfs_readdir_page_unlock_and_put_cached(desc);
1137         } while (!desc->eof);
1138
1139         spin_lock(&file->f_lock);
1140         dir_ctx->dir_cookie = desc->dir_cookie;
1141         dir_ctx->dup_cookie = desc->dup_cookie;
1142         dir_ctx->duped = desc->duped;
1143         dir_ctx->attr_gencount = desc->attr_gencount;
1144         memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1145         spin_unlock(&file->f_lock);
1146
1147         kfree(desc);
1148
1149 out:
1150         dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1151         return res;
1152 }
1153
1154 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1155 {
1156         struct nfs_open_dir_context *dir_ctx = filp->private_data;
1157
1158         dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1159                         filp, offset, whence);
1160
1161         switch (whence) {
1162         default:
1163                 return -EINVAL;
1164         case SEEK_SET:
1165                 if (offset < 0)
1166                         return -EINVAL;
1167                 spin_lock(&filp->f_lock);
1168                 break;
1169         case SEEK_CUR:
1170                 if (offset == 0)
1171                         return filp->f_pos;
1172                 spin_lock(&filp->f_lock);
1173                 offset += filp->f_pos;
1174                 if (offset < 0) {
1175                         spin_unlock(&filp->f_lock);
1176                         return -EINVAL;
1177                 }
1178         }
1179         if (offset != filp->f_pos) {
1180                 filp->f_pos = offset;
1181                 if (nfs_readdir_use_cookie(filp))
1182                         dir_ctx->dir_cookie = offset;
1183                 else
1184                         dir_ctx->dir_cookie = 0;
1185                 if (offset == 0)
1186                         memset(dir_ctx->verf, 0, sizeof(dir_ctx->verf));
1187                 dir_ctx->duped = 0;
1188         }
1189         spin_unlock(&filp->f_lock);
1190         return offset;
1191 }
1192
1193 /*
1194  * All directory operations under NFS are synchronous, so fsync()
1195  * is a dummy operation.
1196  */
1197 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1198                          int datasync)
1199 {
1200         dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1201
1202         nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1203         return 0;
1204 }
1205
1206 /**
1207  * nfs_force_lookup_revalidate - Mark the directory as having changed
1208  * @dir: pointer to directory inode
1209  *
1210  * This forces the revalidation code in nfs_lookup_revalidate() to do a
1211  * full lookup on all child dentries of 'dir' whenever a change occurs
1212  * on the server that might have invalidated our dcache.
1213  *
1214  * Note that we reserve bit '0' as a tag to let us know when a dentry
1215  * was revalidated while holding a delegation on its inode.
1216  *
1217  * The caller should be holding dir->i_lock
1218  */
1219 void nfs_force_lookup_revalidate(struct inode *dir)
1220 {
1221         NFS_I(dir)->cache_change_attribute += 2;
1222 }
1223 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1224
1225 /**
1226  * nfs_verify_change_attribute - Detects NFS remote directory changes
1227  * @dir: pointer to parent directory inode
1228  * @verf: previously saved change attribute
1229  *
1230  * Return "false" if the verifiers doesn't match the change attribute.
1231  * This would usually indicate that the directory contents have changed on
1232  * the server, and that any dentries need revalidating.
1233  */
1234 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1235 {
1236         return (verf & ~1UL) == nfs_save_change_attribute(dir);
1237 }
1238
1239 static void nfs_set_verifier_delegated(unsigned long *verf)
1240 {
1241         *verf |= 1UL;
1242 }
1243
1244 #if IS_ENABLED(CONFIG_NFS_V4)
1245 static void nfs_unset_verifier_delegated(unsigned long *verf)
1246 {
1247         *verf &= ~1UL;
1248 }
1249 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1250
1251 static bool nfs_test_verifier_delegated(unsigned long verf)
1252 {
1253         return verf & 1;
1254 }
1255
1256 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1257 {
1258         return nfs_test_verifier_delegated(dentry->d_time);
1259 }
1260
1261 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1262 {
1263         struct inode *inode = d_inode(dentry);
1264
1265         if (!nfs_verifier_is_delegated(dentry) &&
1266             !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf))
1267                 goto out;
1268         if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1269                 nfs_set_verifier_delegated(&verf);
1270 out:
1271         dentry->d_time = verf;
1272 }
1273
1274 /**
1275  * nfs_set_verifier - save a parent directory verifier in the dentry
1276  * @dentry: pointer to dentry
1277  * @verf: verifier to save
1278  *
1279  * Saves the parent directory verifier in @dentry. If the inode has
1280  * a delegation, we also tag the dentry as having been revalidated
1281  * while holding a delegation so that we know we don't have to
1282  * look it up again after a directory change.
1283  */
1284 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1285 {
1286
1287         spin_lock(&dentry->d_lock);
1288         nfs_set_verifier_locked(dentry, verf);
1289         spin_unlock(&dentry->d_lock);
1290 }
1291 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1292
1293 #if IS_ENABLED(CONFIG_NFS_V4)
1294 /**
1295  * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1296  * @inode: pointer to inode
1297  *
1298  * Iterates through the dentries in the inode alias list and clears
1299  * the tag used to indicate that the dentry has been revalidated
1300  * while holding a delegation.
1301  * This function is intended for use when the delegation is being
1302  * returned or revoked.
1303  */
1304 void nfs_clear_verifier_delegated(struct inode *inode)
1305 {
1306         struct dentry *alias;
1307
1308         if (!inode)
1309                 return;
1310         spin_lock(&inode->i_lock);
1311         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1312                 spin_lock(&alias->d_lock);
1313                 nfs_unset_verifier_delegated(&alias->d_time);
1314                 spin_unlock(&alias->d_lock);
1315         }
1316         spin_unlock(&inode->i_lock);
1317 }
1318 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1319 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1320
1321 /*
1322  * A check for whether or not the parent directory has changed.
1323  * In the case it has, we assume that the dentries are untrustworthy
1324  * and may need to be looked up again.
1325  * If rcu_walk prevents us from performing a full check, return 0.
1326  */
1327 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1328                               int rcu_walk)
1329 {
1330         if (IS_ROOT(dentry))
1331                 return 1;
1332         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1333                 return 0;
1334         if (!nfs_verify_change_attribute(dir, dentry->d_time))
1335                 return 0;
1336         /* Revalidate nfsi->cache_change_attribute before we declare a match */
1337         if (nfs_mapping_need_revalidate_inode(dir)) {
1338                 if (rcu_walk)
1339                         return 0;
1340                 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1341                         return 0;
1342         }
1343         if (!nfs_verify_change_attribute(dir, dentry->d_time))
1344                 return 0;
1345         return 1;
1346 }
1347
1348 /*
1349  * Use intent information to check whether or not we're going to do
1350  * an O_EXCL create using this path component.
1351  */
1352 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1353 {
1354         if (NFS_PROTO(dir)->version == 2)
1355                 return 0;
1356         return flags & LOOKUP_EXCL;
1357 }
1358
1359 /*
1360  * Inode and filehandle revalidation for lookups.
1361  *
1362  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1363  * or if the intent information indicates that we're about to open this
1364  * particular file and the "nocto" mount flag is not set.
1365  *
1366  */
1367 static
1368 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1369 {
1370         struct nfs_server *server = NFS_SERVER(inode);
1371         int ret;
1372
1373         if (IS_AUTOMOUNT(inode))
1374                 return 0;
1375
1376         if (flags & LOOKUP_OPEN) {
1377                 switch (inode->i_mode & S_IFMT) {
1378                 case S_IFREG:
1379                         /* A NFSv4 OPEN will revalidate later */
1380                         if (server->caps & NFS_CAP_ATOMIC_OPEN)
1381                                 goto out;
1382                         fallthrough;
1383                 case S_IFDIR:
1384                         if (server->flags & NFS_MOUNT_NOCTO)
1385                                 break;
1386                         /* NFS close-to-open cache consistency validation */
1387                         goto out_force;
1388                 }
1389         }
1390
1391         /* VFS wants an on-the-wire revalidation */
1392         if (flags & LOOKUP_REVAL)
1393                 goto out_force;
1394 out:
1395         return (inode->i_nlink == 0) ? -ESTALE : 0;
1396 out_force:
1397         if (flags & LOOKUP_RCU)
1398                 return -ECHILD;
1399         ret = __nfs_revalidate_inode(server, inode);
1400         if (ret != 0)
1401                 return ret;
1402         goto out;
1403 }
1404
1405 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1406 {
1407         spin_lock(&inode->i_lock);
1408         nfs_set_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE);
1409         spin_unlock(&inode->i_lock);
1410 }
1411
1412 /*
1413  * We judge how long we want to trust negative
1414  * dentries by looking at the parent inode mtime.
1415  *
1416  * If parent mtime has changed, we revalidate, else we wait for a
1417  * period corresponding to the parent's attribute cache timeout value.
1418  *
1419  * If LOOKUP_RCU prevents us from performing a full check, return 1
1420  * suggesting a reval is needed.
1421  *
1422  * Note that when creating a new file, or looking up a rename target,
1423  * then it shouldn't be necessary to revalidate a negative dentry.
1424  */
1425 static inline
1426 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1427                        unsigned int flags)
1428 {
1429         if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1430                 return 0;
1431         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1432                 return 1;
1433         return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1434 }
1435
1436 static int
1437 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1438                            struct inode *inode, int error)
1439 {
1440         switch (error) {
1441         case 1:
1442                 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1443                         __func__, dentry);
1444                 return 1;
1445         case 0:
1446                 /*
1447                  * We can't d_drop the root of a disconnected tree:
1448                  * its d_hash is on the s_anon list and d_drop() would hide
1449                  * it from shrink_dcache_for_unmount(), leading to busy
1450                  * inodes on unmount and further oopses.
1451                  */
1452                 if (inode && IS_ROOT(dentry))
1453                         return 1;
1454                 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1455                                 __func__, dentry);
1456                 return 0;
1457         }
1458         dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1459                                 __func__, dentry, error);
1460         return error;
1461 }
1462
1463 static int
1464 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1465                                unsigned int flags)
1466 {
1467         int ret = 1;
1468         if (nfs_neg_need_reval(dir, dentry, flags)) {
1469                 if (flags & LOOKUP_RCU)
1470                         return -ECHILD;
1471                 ret = 0;
1472         }
1473         return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1474 }
1475
1476 static int
1477 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1478                                 struct inode *inode)
1479 {
1480         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1481         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1482 }
1483
1484 static int
1485 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1486                              struct inode *inode)
1487 {
1488         struct nfs_fh *fhandle;
1489         struct nfs_fattr *fattr;
1490         struct nfs4_label *label;
1491         unsigned long dir_verifier;
1492         int ret;
1493
1494         ret = -ENOMEM;
1495         fhandle = nfs_alloc_fhandle();
1496         fattr = nfs_alloc_fattr();
1497         label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1498         if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1499                 goto out;
1500
1501         dir_verifier = nfs_save_change_attribute(dir);
1502         ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1503         if (ret < 0) {
1504                 switch (ret) {
1505                 case -ESTALE:
1506                 case -ENOENT:
1507                         ret = 0;
1508                         break;
1509                 case -ETIMEDOUT:
1510                         if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1511                                 ret = 1;
1512                 }
1513                 goto out;
1514         }
1515         ret = 0;
1516         if (nfs_compare_fh(NFS_FH(inode), fhandle))
1517                 goto out;
1518         if (nfs_refresh_inode(inode, fattr) < 0)
1519                 goto out;
1520
1521         nfs_setsecurity(inode, fattr, label);
1522         nfs_set_verifier(dentry, dir_verifier);
1523
1524         /* set a readdirplus hint that we had a cache miss */
1525         nfs_force_use_readdirplus(dir);
1526         ret = 1;
1527 out:
1528         nfs_free_fattr(fattr);
1529         nfs_free_fhandle(fhandle);
1530         nfs4_label_free(label);
1531
1532         /*
1533          * If the lookup failed despite the dentry change attribute being
1534          * a match, then we should revalidate the directory cache.
1535          */
1536         if (!ret && nfs_verify_change_attribute(dir, dentry->d_time))
1537                 nfs_mark_dir_for_revalidate(dir);
1538         return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1539 }
1540
1541 /*
1542  * This is called every time the dcache has a lookup hit,
1543  * and we should check whether we can really trust that
1544  * lookup.
1545  *
1546  * NOTE! The hit can be a negative hit too, don't assume
1547  * we have an inode!
1548  *
1549  * If the parent directory is seen to have changed, we throw out the
1550  * cached dentry and do a new lookup.
1551  */
1552 static int
1553 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1554                          unsigned int flags)
1555 {
1556         struct inode *inode;
1557         int error;
1558
1559         nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1560         inode = d_inode(dentry);
1561
1562         if (!inode)
1563                 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1564
1565         if (is_bad_inode(inode)) {
1566                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1567                                 __func__, dentry);
1568                 goto out_bad;
1569         }
1570
1571         if (nfs_verifier_is_delegated(dentry))
1572                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1573
1574         /* Force a full look up iff the parent directory has changed */
1575         if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1576             nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1577                 error = nfs_lookup_verify_inode(inode, flags);
1578                 if (error) {
1579                         if (error == -ESTALE)
1580                                 nfs_mark_dir_for_revalidate(dir);
1581                         goto out_bad;
1582                 }
1583                 nfs_advise_use_readdirplus(dir);
1584                 goto out_valid;
1585         }
1586
1587         if (flags & LOOKUP_RCU)
1588                 return -ECHILD;
1589
1590         if (NFS_STALE(inode))
1591                 goto out_bad;
1592
1593         trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1594         error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1595         trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1596         return error;
1597 out_valid:
1598         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1599 out_bad:
1600         if (flags & LOOKUP_RCU)
1601                 return -ECHILD;
1602         return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1603 }
1604
1605 static int
1606 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1607                         int (*reval)(struct inode *, struct dentry *, unsigned int))
1608 {
1609         struct dentry *parent;
1610         struct inode *dir;
1611         int ret;
1612
1613         if (flags & LOOKUP_RCU) {
1614                 parent = READ_ONCE(dentry->d_parent);
1615                 dir = d_inode_rcu(parent);
1616                 if (!dir)
1617                         return -ECHILD;
1618                 ret = reval(dir, dentry, flags);
1619                 if (parent != READ_ONCE(dentry->d_parent))
1620                         return -ECHILD;
1621         } else {
1622                 parent = dget_parent(dentry);
1623                 ret = reval(d_inode(parent), dentry, flags);
1624                 dput(parent);
1625         }
1626         return ret;
1627 }
1628
1629 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1630 {
1631         return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1632 }
1633
1634 /*
1635  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1636  * when we don't really care about the dentry name. This is called when a
1637  * pathwalk ends on a dentry that was not found via a normal lookup in the
1638  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1639  *
1640  * In this situation, we just want to verify that the inode itself is OK
1641  * since the dentry might have changed on the server.
1642  */
1643 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1644 {
1645         struct inode *inode = d_inode(dentry);
1646         int error = 0;
1647
1648         /*
1649          * I believe we can only get a negative dentry here in the case of a
1650          * procfs-style symlink. Just assume it's correct for now, but we may
1651          * eventually need to do something more here.
1652          */
1653         if (!inode) {
1654                 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1655                                 __func__, dentry);
1656                 return 1;
1657         }
1658
1659         if (is_bad_inode(inode)) {
1660                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1661                                 __func__, dentry);
1662                 return 0;
1663         }
1664
1665         error = nfs_lookup_verify_inode(inode, flags);
1666         dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1667                         __func__, inode->i_ino, error ? "invalid" : "valid");
1668         return !error;
1669 }
1670
1671 /*
1672  * This is called from dput() when d_count is going to 0.
1673  */
1674 static int nfs_dentry_delete(const struct dentry *dentry)
1675 {
1676         dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1677                 dentry, dentry->d_flags);
1678
1679         /* Unhash any dentry with a stale inode */
1680         if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1681                 return 1;
1682
1683         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1684                 /* Unhash it, so that ->d_iput() would be called */
1685                 return 1;
1686         }
1687         if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1688                 /* Unhash it, so that ancestors of killed async unlink
1689                  * files will be cleaned up during umount */
1690                 return 1;
1691         }
1692         return 0;
1693
1694 }
1695
1696 /* Ensure that we revalidate inode->i_nlink */
1697 static void nfs_drop_nlink(struct inode *inode)
1698 {
1699         spin_lock(&inode->i_lock);
1700         /* drop the inode if we're reasonably sure this is the last link */
1701         if (inode->i_nlink > 0)
1702                 drop_nlink(inode);
1703         NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1704         nfs_set_cache_invalid(
1705                 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1706                                NFS_INO_INVALID_OTHER | NFS_INO_REVAL_FORCED);
1707         spin_unlock(&inode->i_lock);
1708 }
1709
1710 /*
1711  * Called when the dentry loses inode.
1712  * We use it to clean up silly-renamed files.
1713  */
1714 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1715 {
1716         if (S_ISDIR(inode->i_mode))
1717                 /* drop any readdir cache as it could easily be old */
1718                 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1719
1720         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1721                 nfs_complete_unlink(dentry, inode);
1722                 nfs_drop_nlink(inode);
1723         }
1724         iput(inode);
1725 }
1726
1727 static void nfs_d_release(struct dentry *dentry)
1728 {
1729         /* free cached devname value, if it survived that far */
1730         if (unlikely(dentry->d_fsdata)) {
1731                 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1732                         WARN_ON(1);
1733                 else
1734                         kfree(dentry->d_fsdata);
1735         }
1736 }
1737
1738 const struct dentry_operations nfs_dentry_operations = {
1739         .d_revalidate   = nfs_lookup_revalidate,
1740         .d_weak_revalidate      = nfs_weak_revalidate,
1741         .d_delete       = nfs_dentry_delete,
1742         .d_iput         = nfs_dentry_iput,
1743         .d_automount    = nfs_d_automount,
1744         .d_release      = nfs_d_release,
1745 };
1746 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1747
1748 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1749 {
1750         struct dentry *res;
1751         struct inode *inode = NULL;
1752         struct nfs_fh *fhandle = NULL;
1753         struct nfs_fattr *fattr = NULL;
1754         struct nfs4_label *label = NULL;
1755         unsigned long dir_verifier;
1756         int error;
1757
1758         dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1759         nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1760
1761         if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1762                 return ERR_PTR(-ENAMETOOLONG);
1763
1764         /*
1765          * If we're doing an exclusive create, optimize away the lookup
1766          * but don't hash the dentry.
1767          */
1768         if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1769                 return NULL;
1770
1771         res = ERR_PTR(-ENOMEM);
1772         fhandle = nfs_alloc_fhandle();
1773         fattr = nfs_alloc_fattr();
1774         if (fhandle == NULL || fattr == NULL)
1775                 goto out;
1776
1777         label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1778         if (IS_ERR(label))
1779                 goto out;
1780
1781         dir_verifier = nfs_save_change_attribute(dir);
1782         trace_nfs_lookup_enter(dir, dentry, flags);
1783         error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1784         if (error == -ENOENT)
1785                 goto no_entry;
1786         if (error < 0) {
1787                 res = ERR_PTR(error);
1788                 goto out_label;
1789         }
1790         inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1791         res = ERR_CAST(inode);
1792         if (IS_ERR(res))
1793                 goto out_label;
1794
1795         /* Notify readdir to use READDIRPLUS */
1796         nfs_force_use_readdirplus(dir);
1797
1798 no_entry:
1799         res = d_splice_alias(inode, dentry);
1800         if (res != NULL) {
1801                 if (IS_ERR(res))
1802                         goto out_label;
1803                 dentry = res;
1804         }
1805         nfs_set_verifier(dentry, dir_verifier);
1806 out_label:
1807         trace_nfs_lookup_exit(dir, dentry, flags, error);
1808         nfs4_label_free(label);
1809 out:
1810         nfs_free_fattr(fattr);
1811         nfs_free_fhandle(fhandle);
1812         return res;
1813 }
1814 EXPORT_SYMBOL_GPL(nfs_lookup);
1815
1816 #if IS_ENABLED(CONFIG_NFS_V4)
1817 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1818
1819 const struct dentry_operations nfs4_dentry_operations = {
1820         .d_revalidate   = nfs4_lookup_revalidate,
1821         .d_weak_revalidate      = nfs_weak_revalidate,
1822         .d_delete       = nfs_dentry_delete,
1823         .d_iput         = nfs_dentry_iput,
1824         .d_automount    = nfs_d_automount,
1825         .d_release      = nfs_d_release,
1826 };
1827 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1828
1829 static fmode_t flags_to_mode(int flags)
1830 {
1831         fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1832         if ((flags & O_ACCMODE) != O_WRONLY)
1833                 res |= FMODE_READ;
1834         if ((flags & O_ACCMODE) != O_RDONLY)
1835                 res |= FMODE_WRITE;
1836         return res;
1837 }
1838
1839 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1840 {
1841         return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1842 }
1843
1844 static int do_open(struct inode *inode, struct file *filp)
1845 {
1846         nfs_fscache_open_file(inode, filp);
1847         return 0;
1848 }
1849
1850 static int nfs_finish_open(struct nfs_open_context *ctx,
1851                            struct dentry *dentry,
1852                            struct file *file, unsigned open_flags)
1853 {
1854         int err;
1855
1856         err = finish_open(file, dentry, do_open);
1857         if (err)
1858                 goto out;
1859         if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1860                 nfs_file_set_open_context(file, ctx);
1861         else
1862                 err = -EOPENSTALE;
1863 out:
1864         return err;
1865 }
1866
1867 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1868                     struct file *file, unsigned open_flags,
1869                     umode_t mode)
1870 {
1871         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1872         struct nfs_open_context *ctx;
1873         struct dentry *res;
1874         struct iattr attr = { .ia_valid = ATTR_OPEN };
1875         struct inode *inode;
1876         unsigned int lookup_flags = 0;
1877         bool switched = false;
1878         int created = 0;
1879         int err;
1880
1881         /* Expect a negative dentry */
1882         BUG_ON(d_inode(dentry));
1883
1884         dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1885                         dir->i_sb->s_id, dir->i_ino, dentry);
1886
1887         err = nfs_check_flags(open_flags);
1888         if (err)
1889                 return err;
1890
1891         /* NFS only supports OPEN on regular files */
1892         if ((open_flags & O_DIRECTORY)) {
1893                 if (!d_in_lookup(dentry)) {
1894                         /*
1895                          * Hashed negative dentry with O_DIRECTORY: dentry was
1896                          * revalidated and is fine, no need to perform lookup
1897                          * again
1898                          */
1899                         return -ENOENT;
1900                 }
1901                 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1902                 goto no_open;
1903         }
1904
1905         if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1906                 return -ENAMETOOLONG;
1907
1908         if (open_flags & O_CREAT) {
1909                 struct nfs_server *server = NFS_SERVER(dir);
1910
1911                 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1912                         mode &= ~current_umask();
1913
1914                 attr.ia_valid |= ATTR_MODE;
1915                 attr.ia_mode = mode;
1916         }
1917         if (open_flags & O_TRUNC) {
1918                 attr.ia_valid |= ATTR_SIZE;
1919                 attr.ia_size = 0;
1920         }
1921
1922         if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1923                 d_drop(dentry);
1924                 switched = true;
1925                 dentry = d_alloc_parallel(dentry->d_parent,
1926                                           &dentry->d_name, &wq);
1927                 if (IS_ERR(dentry))
1928                         return PTR_ERR(dentry);
1929                 if (unlikely(!d_in_lookup(dentry)))
1930                         return finish_no_open(file, dentry);
1931         }
1932
1933         ctx = create_nfs_open_context(dentry, open_flags, file);
1934         err = PTR_ERR(ctx);
1935         if (IS_ERR(ctx))
1936                 goto out;
1937
1938         trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1939         inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1940         if (created)
1941                 file->f_mode |= FMODE_CREATED;
1942         if (IS_ERR(inode)) {
1943                 err = PTR_ERR(inode);
1944                 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1945                 put_nfs_open_context(ctx);
1946                 d_drop(dentry);
1947                 switch (err) {
1948                 case -ENOENT:
1949                         d_splice_alias(NULL, dentry);
1950                         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1951                         break;
1952                 case -EISDIR:
1953                 case -ENOTDIR:
1954                         goto no_open;
1955                 case -ELOOP:
1956                         if (!(open_flags & O_NOFOLLOW))
1957                                 goto no_open;
1958                         break;
1959                         /* case -EINVAL: */
1960                 default:
1961                         break;
1962                 }
1963                 goto out;
1964         }
1965
1966         err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1967         trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1968         put_nfs_open_context(ctx);
1969 out:
1970         if (unlikely(switched)) {
1971                 d_lookup_done(dentry);
1972                 dput(dentry);
1973         }
1974         return err;
1975
1976 no_open:
1977         res = nfs_lookup(dir, dentry, lookup_flags);
1978         if (switched) {
1979                 d_lookup_done(dentry);
1980                 if (!res)
1981                         res = dentry;
1982                 else
1983                         dput(dentry);
1984         }
1985         if (IS_ERR(res))
1986                 return PTR_ERR(res);
1987         return finish_no_open(file, res);
1988 }
1989 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1990
1991 static int
1992 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1993                           unsigned int flags)
1994 {
1995         struct inode *inode;
1996
1997         if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1998                 goto full_reval;
1999         if (d_mountpoint(dentry))
2000                 goto full_reval;
2001
2002         inode = d_inode(dentry);
2003
2004         /* We can't create new files in nfs_open_revalidate(), so we
2005          * optimize away revalidation of negative dentries.
2006          */
2007         if (inode == NULL)
2008                 goto full_reval;
2009
2010         if (nfs_verifier_is_delegated(dentry))
2011                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2012
2013         /* NFS only supports OPEN on regular files */
2014         if (!S_ISREG(inode->i_mode))
2015                 goto full_reval;
2016
2017         /* We cannot do exclusive creation on a positive dentry */
2018         if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2019                 goto reval_dentry;
2020
2021         /* Check if the directory changed */
2022         if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2023                 goto reval_dentry;
2024
2025         /* Let f_op->open() actually open (and revalidate) the file */
2026         return 1;
2027 reval_dentry:
2028         if (flags & LOOKUP_RCU)
2029                 return -ECHILD;
2030         return nfs_lookup_revalidate_dentry(dir, dentry, inode);
2031
2032 full_reval:
2033         return nfs_do_lookup_revalidate(dir, dentry, flags);
2034 }
2035
2036 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2037 {
2038         return __nfs_lookup_revalidate(dentry, flags,
2039                         nfs4_do_lookup_revalidate);
2040 }
2041
2042 #endif /* CONFIG_NFSV4 */
2043
2044 struct dentry *
2045 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2046                                 struct nfs_fattr *fattr,
2047                                 struct nfs4_label *label)
2048 {
2049         struct dentry *parent = dget_parent(dentry);
2050         struct inode *dir = d_inode(parent);
2051         struct inode *inode;
2052         struct dentry *d;
2053         int error;
2054
2055         d_drop(dentry);
2056
2057         if (fhandle->size == 0) {
2058                 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
2059                 if (error)
2060                         goto out_error;
2061         }
2062         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2063         if (!(fattr->valid & NFS_ATTR_FATTR)) {
2064                 struct nfs_server *server = NFS_SB(dentry->d_sb);
2065                 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2066                                 fattr, NULL, NULL);
2067                 if (error < 0)
2068                         goto out_error;
2069         }
2070         inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
2071         d = d_splice_alias(inode, dentry);
2072 out:
2073         dput(parent);
2074         return d;
2075 out_error:
2076         d = ERR_PTR(error);
2077         goto out;
2078 }
2079 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2080
2081 /*
2082  * Code common to create, mkdir, and mknod.
2083  */
2084 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2085                                 struct nfs_fattr *fattr,
2086                                 struct nfs4_label *label)
2087 {
2088         struct dentry *d;
2089
2090         d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
2091         if (IS_ERR(d))
2092                 return PTR_ERR(d);
2093
2094         /* Callers don't care */
2095         dput(d);
2096         return 0;
2097 }
2098 EXPORT_SYMBOL_GPL(nfs_instantiate);
2099
2100 /*
2101  * Following a failed create operation, we drop the dentry rather
2102  * than retain a negative dentry. This avoids a problem in the event
2103  * that the operation succeeded on the server, but an error in the
2104  * reply path made it appear to have failed.
2105  */
2106 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2107                struct dentry *dentry, umode_t mode, bool excl)
2108 {
2109         struct iattr attr;
2110         int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2111         int error;
2112
2113         dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2114                         dir->i_sb->s_id, dir->i_ino, dentry);
2115
2116         attr.ia_mode = mode;
2117         attr.ia_valid = ATTR_MODE;
2118
2119         trace_nfs_create_enter(dir, dentry, open_flags);
2120         error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2121         trace_nfs_create_exit(dir, dentry, open_flags, error);
2122         if (error != 0)
2123                 goto out_err;
2124         return 0;
2125 out_err:
2126         d_drop(dentry);
2127         return error;
2128 }
2129 EXPORT_SYMBOL_GPL(nfs_create);
2130
2131 /*
2132  * See comments for nfs_proc_create regarding failed operations.
2133  */
2134 int
2135 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2136           struct dentry *dentry, umode_t mode, dev_t rdev)
2137 {
2138         struct iattr attr;
2139         int status;
2140
2141         dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2142                         dir->i_sb->s_id, dir->i_ino, dentry);
2143
2144         attr.ia_mode = mode;
2145         attr.ia_valid = ATTR_MODE;
2146
2147         trace_nfs_mknod_enter(dir, dentry);
2148         status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2149         trace_nfs_mknod_exit(dir, dentry, status);
2150         if (status != 0)
2151                 goto out_err;
2152         return 0;
2153 out_err:
2154         d_drop(dentry);
2155         return status;
2156 }
2157 EXPORT_SYMBOL_GPL(nfs_mknod);
2158
2159 /*
2160  * See comments for nfs_proc_create regarding failed operations.
2161  */
2162 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2163               struct dentry *dentry, umode_t mode)
2164 {
2165         struct iattr attr;
2166         int error;
2167
2168         dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2169                         dir->i_sb->s_id, dir->i_ino, dentry);
2170
2171         attr.ia_valid = ATTR_MODE;
2172         attr.ia_mode = mode | S_IFDIR;
2173
2174         trace_nfs_mkdir_enter(dir, dentry);
2175         error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2176         trace_nfs_mkdir_exit(dir, dentry, error);
2177         if (error != 0)
2178                 goto out_err;
2179         return 0;
2180 out_err:
2181         d_drop(dentry);
2182         return error;
2183 }
2184 EXPORT_SYMBOL_GPL(nfs_mkdir);
2185
2186 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2187 {
2188         if (simple_positive(dentry))
2189                 d_delete(dentry);
2190 }
2191
2192 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2193 {
2194         int error;
2195
2196         dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2197                         dir->i_sb->s_id, dir->i_ino, dentry);
2198
2199         trace_nfs_rmdir_enter(dir, dentry);
2200         if (d_really_is_positive(dentry)) {
2201                 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2202                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2203                 /* Ensure the VFS deletes this inode */
2204                 switch (error) {
2205                 case 0:
2206                         clear_nlink(d_inode(dentry));
2207                         break;
2208                 case -ENOENT:
2209                         nfs_dentry_handle_enoent(dentry);
2210                 }
2211                 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2212         } else
2213                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2214         trace_nfs_rmdir_exit(dir, dentry, error);
2215
2216         return error;
2217 }
2218 EXPORT_SYMBOL_GPL(nfs_rmdir);
2219
2220 /*
2221  * Remove a file after making sure there are no pending writes,
2222  * and after checking that the file has only one user.
2223  *
2224  * We invalidate the attribute cache and free the inode prior to the operation
2225  * to avoid possible races if the server reuses the inode.
2226  */
2227 static int nfs_safe_remove(struct dentry *dentry)
2228 {
2229         struct inode *dir = d_inode(dentry->d_parent);
2230         struct inode *inode = d_inode(dentry);
2231         int error = -EBUSY;
2232
2233         dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2234
2235         /* If the dentry was sillyrenamed, we simply call d_delete() */
2236         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2237                 error = 0;
2238                 goto out;
2239         }
2240
2241         trace_nfs_remove_enter(dir, dentry);
2242         if (inode != NULL) {
2243                 error = NFS_PROTO(dir)->remove(dir, dentry);
2244                 if (error == 0)
2245                         nfs_drop_nlink(inode);
2246         } else
2247                 error = NFS_PROTO(dir)->remove(dir, dentry);
2248         if (error == -ENOENT)
2249                 nfs_dentry_handle_enoent(dentry);
2250         trace_nfs_remove_exit(dir, dentry, error);
2251 out:
2252         return error;
2253 }
2254
2255 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
2256  *  belongs to an active ".nfs..." file and we return -EBUSY.
2257  *
2258  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
2259  */
2260 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2261 {
2262         int error;
2263         int need_rehash = 0;
2264
2265         dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2266                 dir->i_ino, dentry);
2267
2268         trace_nfs_unlink_enter(dir, dentry);
2269         spin_lock(&dentry->d_lock);
2270         if (d_count(dentry) > 1) {
2271                 spin_unlock(&dentry->d_lock);
2272                 /* Start asynchronous writeout of the inode */
2273                 write_inode_now(d_inode(dentry), 0);
2274                 error = nfs_sillyrename(dir, dentry);
2275                 goto out;
2276         }
2277         if (!d_unhashed(dentry)) {
2278                 __d_drop(dentry);
2279                 need_rehash = 1;
2280         }
2281         spin_unlock(&dentry->d_lock);
2282         error = nfs_safe_remove(dentry);
2283         if (!error || error == -ENOENT) {
2284                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2285         } else if (need_rehash)
2286                 d_rehash(dentry);
2287 out:
2288         trace_nfs_unlink_exit(dir, dentry, error);
2289         return error;
2290 }
2291 EXPORT_SYMBOL_GPL(nfs_unlink);
2292
2293 /*
2294  * To create a symbolic link, most file systems instantiate a new inode,
2295  * add a page to it containing the path, then write it out to the disk
2296  * using prepare_write/commit_write.
2297  *
2298  * Unfortunately the NFS client can't create the in-core inode first
2299  * because it needs a file handle to create an in-core inode (see
2300  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
2301  * symlink request has completed on the server.
2302  *
2303  * So instead we allocate a raw page, copy the symname into it, then do
2304  * the SYMLINK request with the page as the buffer.  If it succeeds, we
2305  * now have a new file handle and can instantiate an in-core NFS inode
2306  * and move the raw page into its mapping.
2307  */
2308 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2309                 struct dentry *dentry, const char *symname)
2310 {
2311         struct page *page;
2312         char *kaddr;
2313         struct iattr attr;
2314         unsigned int pathlen = strlen(symname);
2315         int error;
2316
2317         dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2318                 dir->i_ino, dentry, symname);
2319
2320         if (pathlen > PAGE_SIZE)
2321                 return -ENAMETOOLONG;
2322
2323         attr.ia_mode = S_IFLNK | S_IRWXUGO;
2324         attr.ia_valid = ATTR_MODE;
2325
2326         page = alloc_page(GFP_USER);
2327         if (!page)
2328                 return -ENOMEM;
2329
2330         kaddr = page_address(page);
2331         memcpy(kaddr, symname, pathlen);
2332         if (pathlen < PAGE_SIZE)
2333                 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2334
2335         trace_nfs_symlink_enter(dir, dentry);
2336         error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2337         trace_nfs_symlink_exit(dir, dentry, error);
2338         if (error != 0) {
2339                 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2340                         dir->i_sb->s_id, dir->i_ino,
2341                         dentry, symname, error);
2342                 d_drop(dentry);
2343                 __free_page(page);
2344                 return error;
2345         }
2346
2347         /*
2348          * No big deal if we can't add this page to the page cache here.
2349          * READLINK will get the missing page from the server if needed.
2350          */
2351         if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2352                                                         GFP_KERNEL)) {
2353                 SetPageUptodate(page);
2354                 unlock_page(page);
2355                 /*
2356                  * add_to_page_cache_lru() grabs an extra page refcount.
2357                  * Drop it here to avoid leaking this page later.
2358                  */
2359                 put_page(page);
2360         } else
2361                 __free_page(page);
2362
2363         return 0;
2364 }
2365 EXPORT_SYMBOL_GPL(nfs_symlink);
2366
2367 int
2368 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2369 {
2370         struct inode *inode = d_inode(old_dentry);
2371         int error;
2372
2373         dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2374                 old_dentry, dentry);
2375
2376         trace_nfs_link_enter(inode, dir, dentry);
2377         d_drop(dentry);
2378         error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2379         if (error == 0) {
2380                 ihold(inode);
2381                 d_add(dentry, inode);
2382         }
2383         trace_nfs_link_exit(inode, dir, dentry, error);
2384         return error;
2385 }
2386 EXPORT_SYMBOL_GPL(nfs_link);
2387
2388 /*
2389  * RENAME
2390  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2391  * different file handle for the same inode after a rename (e.g. when
2392  * moving to a different directory). A fail-safe method to do so would
2393  * be to look up old_dir/old_name, create a link to new_dir/new_name and
2394  * rename the old file using the sillyrename stuff. This way, the original
2395  * file in old_dir will go away when the last process iput()s the inode.
2396  *
2397  * FIXED.
2398  *
2399  * It actually works quite well. One needs to have the possibility for
2400  * at least one ".nfs..." file in each directory the file ever gets
2401  * moved or linked to which happens automagically with the new
2402  * implementation that only depends on the dcache stuff instead of
2403  * using the inode layer
2404  *
2405  * Unfortunately, things are a little more complicated than indicated
2406  * above. For a cross-directory move, we want to make sure we can get
2407  * rid of the old inode after the operation.  This means there must be
2408  * no pending writes (if it's a file), and the use count must be 1.
2409  * If these conditions are met, we can drop the dentries before doing
2410  * the rename.
2411  */
2412 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2413                struct dentry *old_dentry, struct inode *new_dir,
2414                struct dentry *new_dentry, unsigned int flags)
2415 {
2416         struct inode *old_inode = d_inode(old_dentry);
2417         struct inode *new_inode = d_inode(new_dentry);
2418         struct dentry *dentry = NULL, *rehash = NULL;
2419         struct rpc_task *task;
2420         int error = -EBUSY;
2421
2422         if (flags)
2423                 return -EINVAL;
2424
2425         dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2426                  old_dentry, new_dentry,
2427                  d_count(new_dentry));
2428
2429         trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2430         /*
2431          * For non-directories, check whether the target is busy and if so,
2432          * make a copy of the dentry and then do a silly-rename. If the
2433          * silly-rename succeeds, the copied dentry is hashed and becomes
2434          * the new target.
2435          */
2436         if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2437                 /*
2438                  * To prevent any new references to the target during the
2439                  * rename, we unhash the dentry in advance.
2440                  */
2441                 if (!d_unhashed(new_dentry)) {
2442                         d_drop(new_dentry);
2443                         rehash = new_dentry;
2444                 }
2445
2446                 if (d_count(new_dentry) > 2) {
2447                         int err;
2448
2449                         /* copy the target dentry's name */
2450                         dentry = d_alloc(new_dentry->d_parent,
2451                                          &new_dentry->d_name);
2452                         if (!dentry)
2453                                 goto out;
2454
2455                         /* silly-rename the existing target ... */
2456                         err = nfs_sillyrename(new_dir, new_dentry);
2457                         if (err)
2458                                 goto out;
2459
2460                         new_dentry = dentry;
2461                         rehash = NULL;
2462                         new_inode = NULL;
2463                 }
2464         }
2465
2466         task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2467         if (IS_ERR(task)) {
2468                 error = PTR_ERR(task);
2469                 goto out;
2470         }
2471
2472         error = rpc_wait_for_completion_task(task);
2473         if (error != 0) {
2474                 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2475                 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2476                 smp_wmb();
2477         } else
2478                 error = task->tk_status;
2479         rpc_put_task(task);
2480         /* Ensure the inode attributes are revalidated */
2481         if (error == 0) {
2482                 spin_lock(&old_inode->i_lock);
2483                 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2484                 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2485                                                          NFS_INO_INVALID_CTIME |
2486                                                          NFS_INO_REVAL_FORCED);
2487                 spin_unlock(&old_inode->i_lock);
2488         }
2489 out:
2490         if (rehash)
2491                 d_rehash(rehash);
2492         trace_nfs_rename_exit(old_dir, old_dentry,
2493                         new_dir, new_dentry, error);
2494         if (!error) {
2495                 if (new_inode != NULL)
2496                         nfs_drop_nlink(new_inode);
2497                 /*
2498                  * The d_move() should be here instead of in an async RPC completion
2499                  * handler because we need the proper locks to move the dentry.  If
2500                  * we're interrupted by a signal, the async RPC completion handler
2501                  * should mark the directories for revalidation.
2502                  */
2503                 d_move(old_dentry, new_dentry);
2504                 nfs_set_verifier(old_dentry,
2505                                         nfs_save_change_attribute(new_dir));
2506         } else if (error == -ENOENT)
2507                 nfs_dentry_handle_enoent(old_dentry);
2508
2509         /* new dentry created? */
2510         if (dentry)
2511                 dput(dentry);
2512         return error;
2513 }
2514 EXPORT_SYMBOL_GPL(nfs_rename);
2515
2516 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2517 static LIST_HEAD(nfs_access_lru_list);
2518 static atomic_long_t nfs_access_nr_entries;
2519
2520 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2521 module_param(nfs_access_max_cachesize, ulong, 0644);
2522 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2523
2524 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2525 {
2526         put_cred(entry->cred);
2527         kfree_rcu(entry, rcu_head);
2528         smp_mb__before_atomic();
2529         atomic_long_dec(&nfs_access_nr_entries);
2530         smp_mb__after_atomic();
2531 }
2532
2533 static void nfs_access_free_list(struct list_head *head)
2534 {
2535         struct nfs_access_entry *cache;
2536
2537         while (!list_empty(head)) {
2538                 cache = list_entry(head->next, struct nfs_access_entry, lru);
2539                 list_del(&cache->lru);
2540                 nfs_access_free_entry(cache);
2541         }
2542 }
2543
2544 static unsigned long
2545 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2546 {
2547         LIST_HEAD(head);
2548         struct nfs_inode *nfsi, *next;
2549         struct nfs_access_entry *cache;
2550         long freed = 0;
2551
2552         spin_lock(&nfs_access_lru_lock);
2553         list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2554                 struct inode *inode;
2555
2556                 if (nr_to_scan-- == 0)
2557                         break;
2558                 inode = &nfsi->vfs_inode;
2559                 spin_lock(&inode->i_lock);
2560                 if (list_empty(&nfsi->access_cache_entry_lru))
2561                         goto remove_lru_entry;
2562                 cache = list_entry(nfsi->access_cache_entry_lru.next,
2563                                 struct nfs_access_entry, lru);
2564                 list_move(&cache->lru, &head);
2565                 rb_erase(&cache->rb_node, &nfsi->access_cache);
2566                 freed++;
2567                 if (!list_empty(&nfsi->access_cache_entry_lru))
2568                         list_move_tail(&nfsi->access_cache_inode_lru,
2569                                         &nfs_access_lru_list);
2570                 else {
2571 remove_lru_entry:
2572                         list_del_init(&nfsi->access_cache_inode_lru);
2573                         smp_mb__before_atomic();
2574                         clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2575                         smp_mb__after_atomic();
2576                 }
2577                 spin_unlock(&inode->i_lock);
2578         }
2579         spin_unlock(&nfs_access_lru_lock);
2580         nfs_access_free_list(&head);
2581         return freed;
2582 }
2583
2584 unsigned long
2585 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2586 {
2587         int nr_to_scan = sc->nr_to_scan;
2588         gfp_t gfp_mask = sc->gfp_mask;
2589
2590         if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2591                 return SHRINK_STOP;
2592         return nfs_do_access_cache_scan(nr_to_scan);
2593 }
2594
2595
2596 unsigned long
2597 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2598 {
2599         return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2600 }
2601
2602 static void
2603 nfs_access_cache_enforce_limit(void)
2604 {
2605         long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2606         unsigned long diff;
2607         unsigned int nr_to_scan;
2608
2609         if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2610                 return;
2611         nr_to_scan = 100;
2612         diff = nr_entries - nfs_access_max_cachesize;
2613         if (diff < nr_to_scan)
2614                 nr_to_scan = diff;
2615         nfs_do_access_cache_scan(nr_to_scan);
2616 }
2617
2618 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2619 {
2620         struct rb_root *root_node = &nfsi->access_cache;
2621         struct rb_node *n;
2622         struct nfs_access_entry *entry;
2623
2624         /* Unhook entries from the cache */
2625         while ((n = rb_first(root_node)) != NULL) {
2626                 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2627                 rb_erase(n, root_node);
2628                 list_move(&entry->lru, head);
2629         }
2630         nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2631 }
2632
2633 void nfs_access_zap_cache(struct inode *inode)
2634 {
2635         LIST_HEAD(head);
2636
2637         if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2638                 return;
2639         /* Remove from global LRU init */
2640         spin_lock(&nfs_access_lru_lock);
2641         if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2642                 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2643
2644         spin_lock(&inode->i_lock);
2645         __nfs_access_zap_cache(NFS_I(inode), &head);
2646         spin_unlock(&inode->i_lock);
2647         spin_unlock(&nfs_access_lru_lock);
2648         nfs_access_free_list(&head);
2649 }
2650 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2651
2652 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2653 {
2654         struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2655
2656         while (n != NULL) {
2657                 struct nfs_access_entry *entry =
2658                         rb_entry(n, struct nfs_access_entry, rb_node);
2659                 int cmp = cred_fscmp(cred, entry->cred);
2660
2661                 if (cmp < 0)
2662                         n = n->rb_left;
2663                 else if (cmp > 0)
2664                         n = n->rb_right;
2665                 else
2666                         return entry;
2667         }
2668         return NULL;
2669 }
2670
2671 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2672 {
2673         struct nfs_inode *nfsi = NFS_I(inode);
2674         struct nfs_access_entry *cache;
2675         bool retry = true;
2676         int err;
2677
2678         spin_lock(&inode->i_lock);
2679         for(;;) {
2680                 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2681                         goto out_zap;
2682                 cache = nfs_access_search_rbtree(inode, cred);
2683                 err = -ENOENT;
2684                 if (cache == NULL)
2685                         goto out;
2686                 /* Found an entry, is our attribute cache valid? */
2687                 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2688                         break;
2689                 if (!retry)
2690                         break;
2691                 err = -ECHILD;
2692                 if (!may_block)
2693                         goto out;
2694                 spin_unlock(&inode->i_lock);
2695                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2696                 if (err)
2697                         return err;
2698                 spin_lock(&inode->i_lock);
2699                 retry = false;
2700         }
2701         res->cred = cache->cred;
2702         res->mask = cache->mask;
2703         list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2704         err = 0;
2705 out:
2706         spin_unlock(&inode->i_lock);
2707         return err;
2708 out_zap:
2709         spin_unlock(&inode->i_lock);
2710         nfs_access_zap_cache(inode);
2711         return -ENOENT;
2712 }
2713
2714 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2715 {
2716         /* Only check the most recently returned cache entry,
2717          * but do it without locking.
2718          */
2719         struct nfs_inode *nfsi = NFS_I(inode);
2720         struct nfs_access_entry *cache;
2721         int err = -ECHILD;
2722         struct list_head *lh;
2723
2724         rcu_read_lock();
2725         if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2726                 goto out;
2727         lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2728         cache = list_entry(lh, struct nfs_access_entry, lru);
2729         if (lh == &nfsi->access_cache_entry_lru ||
2730             cred_fscmp(cred, cache->cred) != 0)
2731                 cache = NULL;
2732         if (cache == NULL)
2733                 goto out;
2734         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2735                 goto out;
2736         res->cred = cache->cred;
2737         res->mask = cache->mask;
2738         err = 0;
2739 out:
2740         rcu_read_unlock();
2741         return err;
2742 }
2743
2744 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct
2745 nfs_access_entry *res, bool may_block)
2746 {
2747         int status;
2748
2749         status = nfs_access_get_cached_rcu(inode, cred, res);
2750         if (status != 0)
2751                 status = nfs_access_get_cached_locked(inode, cred, res,
2752                     may_block);
2753
2754         return status;
2755 }
2756 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
2757
2758 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2759 {
2760         struct nfs_inode *nfsi = NFS_I(inode);
2761         struct rb_root *root_node = &nfsi->access_cache;
2762         struct rb_node **p = &root_node->rb_node;
2763         struct rb_node *parent = NULL;
2764         struct nfs_access_entry *entry;
2765         int cmp;
2766
2767         spin_lock(&inode->i_lock);
2768         while (*p != NULL) {
2769                 parent = *p;
2770                 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2771                 cmp = cred_fscmp(set->cred, entry->cred);
2772
2773                 if (cmp < 0)
2774                         p = &parent->rb_left;
2775                 else if (cmp > 0)
2776                         p = &parent->rb_right;
2777                 else
2778                         goto found;
2779         }
2780         rb_link_node(&set->rb_node, parent, p);
2781         rb_insert_color(&set->rb_node, root_node);
2782         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2783         spin_unlock(&inode->i_lock);
2784         return;
2785 found:
2786         rb_replace_node(parent, &set->rb_node, root_node);
2787         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2788         list_del(&entry->lru);
2789         spin_unlock(&inode->i_lock);
2790         nfs_access_free_entry(entry);
2791 }
2792
2793 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2794 {
2795         struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2796         if (cache == NULL)
2797                 return;
2798         RB_CLEAR_NODE(&cache->rb_node);
2799         cache->cred = get_cred(set->cred);
2800         cache->mask = set->mask;
2801
2802         /* The above field assignments must be visible
2803          * before this item appears on the lru.  We cannot easily
2804          * use rcu_assign_pointer, so just force the memory barrier.
2805          */
2806         smp_wmb();
2807         nfs_access_add_rbtree(inode, cache);
2808
2809         /* Update accounting */
2810         smp_mb__before_atomic();
2811         atomic_long_inc(&nfs_access_nr_entries);
2812         smp_mb__after_atomic();
2813
2814         /* Add inode to global LRU list */
2815         if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2816                 spin_lock(&nfs_access_lru_lock);
2817                 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2818                         list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2819                                         &nfs_access_lru_list);
2820                 spin_unlock(&nfs_access_lru_lock);
2821         }
2822         nfs_access_cache_enforce_limit();
2823 }
2824 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2825
2826 #define NFS_MAY_READ (NFS_ACCESS_READ)
2827 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2828                 NFS_ACCESS_EXTEND | \
2829                 NFS_ACCESS_DELETE)
2830 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2831                 NFS_ACCESS_EXTEND)
2832 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2833 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2834 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2835 static int
2836 nfs_access_calc_mask(u32 access_result, umode_t umode)
2837 {
2838         int mask = 0;
2839
2840         if (access_result & NFS_MAY_READ)
2841                 mask |= MAY_READ;
2842         if (S_ISDIR(umode)) {
2843                 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2844                         mask |= MAY_WRITE;
2845                 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2846                         mask |= MAY_EXEC;
2847         } else if (S_ISREG(umode)) {
2848                 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2849                         mask |= MAY_WRITE;
2850                 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2851                         mask |= MAY_EXEC;
2852         } else if (access_result & NFS_MAY_WRITE)
2853                         mask |= MAY_WRITE;
2854         return mask;
2855 }
2856
2857 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2858 {
2859         entry->mask = access_result;
2860 }
2861 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2862
2863 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2864 {
2865         struct nfs_access_entry cache;
2866         bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2867         int cache_mask = -1;
2868         int status;
2869
2870         trace_nfs_access_enter(inode);
2871
2872         status = nfs_access_get_cached(inode, cred, &cache, may_block);
2873         if (status == 0)
2874                 goto out_cached;
2875
2876         status = -ECHILD;
2877         if (!may_block)
2878                 goto out;
2879
2880         /*
2881          * Determine which access bits we want to ask for...
2882          */
2883         cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2884         if (nfs_server_capable(inode, NFS_CAP_XATTR)) {
2885                 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE |
2886                     NFS_ACCESS_XALIST;
2887         }
2888         if (S_ISDIR(inode->i_mode))
2889                 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2890         else
2891                 cache.mask |= NFS_ACCESS_EXECUTE;
2892         cache.cred = cred;
2893         status = NFS_PROTO(inode)->access(inode, &cache);
2894         if (status != 0) {
2895                 if (status == -ESTALE) {
2896                         if (!S_ISDIR(inode->i_mode))
2897                                 nfs_set_inode_stale(inode);
2898                         else
2899                                 nfs_zap_caches(inode);
2900                 }
2901                 goto out;
2902         }
2903         nfs_access_add_cache(inode, &cache);
2904 out_cached:
2905         cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2906         if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2907                 status = -EACCES;
2908 out:
2909         trace_nfs_access_exit(inode, mask, cache_mask, status);
2910         return status;
2911 }
2912
2913 static int nfs_open_permission_mask(int openflags)
2914 {
2915         int mask = 0;
2916
2917         if (openflags & __FMODE_EXEC) {
2918                 /* ONLY check exec rights */
2919                 mask = MAY_EXEC;
2920         } else {
2921                 if ((openflags & O_ACCMODE) != O_WRONLY)
2922                         mask |= MAY_READ;
2923                 if ((openflags & O_ACCMODE) != O_RDONLY)
2924                         mask |= MAY_WRITE;
2925         }
2926
2927         return mask;
2928 }
2929
2930 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2931 {
2932         return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2933 }
2934 EXPORT_SYMBOL_GPL(nfs_may_open);
2935
2936 static int nfs_execute_ok(struct inode *inode, int mask)
2937 {
2938         struct nfs_server *server = NFS_SERVER(inode);
2939         int ret = 0;
2940
2941         if (S_ISDIR(inode->i_mode))
2942                 return 0;
2943         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2944                 if (mask & MAY_NOT_BLOCK)
2945                         return -ECHILD;
2946                 ret = __nfs_revalidate_inode(server, inode);
2947         }
2948         if (ret == 0 && !execute_ok(inode))
2949                 ret = -EACCES;
2950         return ret;
2951 }
2952
2953 int nfs_permission(struct user_namespace *mnt_userns,
2954                    struct inode *inode,
2955                    int mask)
2956 {
2957         const struct cred *cred = current_cred();
2958         int res = 0;
2959
2960         nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2961
2962         if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2963                 goto out;
2964         /* Is this sys_access() ? */
2965         if (mask & (MAY_ACCESS | MAY_CHDIR))
2966                 goto force_lookup;
2967
2968         switch (inode->i_mode & S_IFMT) {
2969                 case S_IFLNK:
2970                         goto out;
2971                 case S_IFREG:
2972                         if ((mask & MAY_OPEN) &&
2973                            nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2974                                 return 0;
2975                         break;
2976                 case S_IFDIR:
2977                         /*
2978                          * Optimize away all write operations, since the server
2979                          * will check permissions when we perform the op.
2980                          */
2981                         if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2982                                 goto out;
2983         }
2984
2985 force_lookup:
2986         if (!NFS_PROTO(inode)->access)
2987                 goto out_notsup;
2988
2989         res = nfs_do_access(inode, cred, mask);
2990 out:
2991         if (!res && (mask & MAY_EXEC))
2992                 res = nfs_execute_ok(inode, mask);
2993
2994         dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2995                 inode->i_sb->s_id, inode->i_ino, mask, res);
2996         return res;
2997 out_notsup:
2998         if (mask & MAY_NOT_BLOCK)
2999                 return -ECHILD;
3000
3001         res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
3002         if (res == 0)
3003                 res = generic_permission(&init_user_ns, inode, mask);
3004         goto out;
3005 }
3006 EXPORT_SYMBOL_GPL(nfs_permission);
3007
3008 /*
3009  * Local variables:
3010  *  version-control: t
3011  *  kept-new-versions: 5
3012  * End:
3013  */