fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/version.h>
  38 #include <linux/xattr.h>
  39 #include <linux/posix_acl.h>
  40 #include <linux/falloc.h>
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "compat.h"
  51 #include "tree-log.h"
  52 #include "ref-cache.h"
  53 #include "compression.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static struct inode_operations btrfs_dir_inode_operations;
  61 static struct inode_operations btrfs_symlink_inode_operations;
  62 static struct inode_operations btrfs_dir_ro_inode_operations;
  63 static struct inode_operations btrfs_special_inode_operations;
  64 static struct inode_operations btrfs_file_inode_operations;
  65 static struct address_space_operations btrfs_aops;
  66 static struct address_space_operations btrfs_symlink_aops;
  67 static struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_bit_radix_cachep;
  74 struct kmem_cache *btrfs_path_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static void btrfs_truncate(struct inode *inode);
  88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  89 static noinline int cow_file_range(struct inode *inode,
  90                                    struct page *locked_page,
  91                                    u64 start, u64 end, int *page_started,
  92                                    unsigned long *nr_written, int unlock);
  93
  94 /*
  95  * a very lame attempt at stopping writes when the FS is 85% full.  There
  96  * are countless ways this is incorrect, but it is better than nothing.
  97  */
  98 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  99                            int for_del)
 100 {
 101         u64 total;
 102         u64 used;
 103         u64 thresh;
 104         unsigned long flags;
 105         int ret = 0;
 106
 107         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 108         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
 109         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
 110         if (for_del)
 111                 thresh = total * 90;
 112         else
 113                 thresh = total * 85;
 114
 115         do_div(thresh, 100);
 116
 117         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
 118                 ret = -ENOSPC;
 119         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 120         return ret;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_root *root, struct inode *inode,
 130                                 u64 start, size_t size, size_t compressed_size,
 131                                 struct page **compressed_pages)
 132 {
 133         struct btrfs_key key;
 134         struct btrfs_path *path;
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         size_t datasize;
 144         unsigned long offset;
 145         int use_compress = 0;
 146
 147         if (compressed_size && compressed_pages) {
 148                 use_compress = 1;
 149                 cur_size = compressed_size;
 150         }
 151
 152         path = btrfs_alloc_path(); if (!path)
 153                 return -ENOMEM;
 154
 155         btrfs_set_trans_block_group(trans, inode);
 156
 157         key.objectid = inode->i_ino;
 158         key.offset = start;
 159         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 160         inode_add_bytes(inode, size);
 161         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 162
 163         inode_add_bytes(inode, size);
 164         ret = btrfs_insert_empty_item(trans, root, path, &key,
 165                                       datasize);
 166         BUG_ON(ret);
 167         if (ret) {
 168                 err = ret;
 169                 printk("got bad ret %d\n", ret);
 170                 goto fail;
 171         }
 172         leaf = path->nodes[0];
 173         ei = btrfs_item_ptr(leaf, path->slots[0],
 174                             struct btrfs_file_extent_item);
 175         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 176         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 177         btrfs_set_file_extent_encryption(leaf, ei, 0);
 178         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 179         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 180         ptr = btrfs_file_extent_inline_start(ei);
 181
 182         if (use_compress) {
 183                 struct page *cpage;
 184                 int i = 0;
 185                 while(compressed_size > 0) {
 186                         cpage = compressed_pages[i];
 187                         cur_size = min_t(unsigned long, compressed_size,
 188                                        PAGE_CACHE_SIZE);
 189
 190                         kaddr = kmap(cpage);
 191                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 192                         kunmap(cpage);
 193
 194                         i++;
 195                         ptr += cur_size;
 196                         compressed_size -= cur_size;
 197                 }
 198                 btrfs_set_file_extent_compression(leaf, ei,
 199                                                   BTRFS_COMPRESS_ZLIB);
 200         } else {
 201                 page = find_get_page(inode->i_mapping,
 202                                      start >> PAGE_CACHE_SHIFT);
 203                 btrfs_set_file_extent_compression(leaf, ei, 0);
 204                 kaddr = kmap_atomic(page, KM_USER0);
 205                 offset = start & (PAGE_CACHE_SIZE - 1);
 206                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 207                 kunmap_atomic(kaddr, KM_USER0);
 208                 page_cache_release(page);
 209         }
 210         btrfs_mark_buffer_dirty(leaf);
 211         btrfs_free_path(path);
 212
 213         BTRFS_I(inode)->disk_i_size = inode->i_size;
 214         btrfs_update_inode(trans, root, inode);
 215         return 0;
 216 fail:
 217         btrfs_free_path(path);
 218         return err;
 219 }
 220
 221
 222 /*
 223  * conditionally insert an inline extent into the file.  This
 224  * does the checks required to make sure the data is small enough
 225  * to fit as an inline extent.
 226  */
 227 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
 228                                  struct btrfs_root *root,
 229                                  struct inode *inode, u64 start, u64 end,
 230                                  size_t compressed_size,
 231                                  struct page **compressed_pages)
 232 {
 233         u64 isize = i_size_read(inode);
 234         u64 actual_end = min(end + 1, isize);
 235         u64 inline_len = actual_end - start;
 236         u64 aligned_end = (end + root->sectorsize - 1) &
 237                         ~((u64)root->sectorsize - 1);
 238         u64 hint_byte;
 239         u64 data_len = inline_len;
 240         int ret;
 241
 242         if (compressed_size)
 243                 data_len = compressed_size;
 244
 245         if (start > 0 ||
 246             actual_end >= PAGE_CACHE_SIZE ||
 247             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 248             (!compressed_size &&
 249             (actual_end & (root->sectorsize - 1)) == 0) ||
 250             end + 1 < isize ||
 251             data_len > root->fs_info->max_inline) {
 252                 return 1;
 253         }
 254
 255         ret = btrfs_drop_extents(trans, root, inode, start,
 256                                  aligned_end, start, &hint_byte);
 257         BUG_ON(ret);
 258
 259         if (isize > actual_end)
 260                 inline_len = min_t(u64, isize, actual_end);
 261         ret = insert_inline_extent(trans, root, inode, start,
 262                                    inline_len, compressed_size,
 263                                    compressed_pages);
 264         BUG_ON(ret);
 265         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
 266         return 0;
 267 }
 268
 269 struct async_extent {
 270         u64 start;
 271         u64 ram_size;
 272         u64 compressed_size;
 273         struct page **pages;
 274         unsigned long nr_pages;
 275         struct list_head list;
 276 };
 277
 278 struct async_cow {
 279         struct inode *inode;
 280         struct btrfs_root *root;
 281         struct page *locked_page;
 282         u64 start;
 283         u64 end;
 284         struct list_head extents;
 285         struct btrfs_work work;
 286 };
 287
 288 static noinline int add_async_extent(struct async_cow *cow,
 289                                      u64 start, u64 ram_size,
 290                                      u64 compressed_size,
 291                                      struct page **pages,
 292                                      unsigned long nr_pages)
 293 {
 294         struct async_extent *async_extent;
 295
 296         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 297         async_extent->start = start;
 298         async_extent->ram_size = ram_size;
 299         async_extent->compressed_size = compressed_size;
 300         async_extent->pages = pages;
 301         async_extent->nr_pages = nr_pages;
 302         list_add_tail(&async_extent->list, &cow->extents);
 303         return 0;
 304 }
 305
 306 /*
 307  * we create compressed extents in two phases.  The first
 308  * phase compresses a range of pages that have already been
 309  * locked (both pages and state bits are locked).
 310  *
 311  * This is done inside an ordered work queue, and the compression
 312  * is spread across many cpus.  The actual IO submission is step
 313  * two, and the ordered work queue takes care of making sure that
 314  * happens in the same order things were put onto the queue by
 315  * writepages and friends.
 316  *
 317  * If this code finds it can't get good compression, it puts an
 318  * entry onto the work queue to write the uncompressed bytes.  This
 319  * makes sure that both compressed inodes and uncompressed inodes
 320  * are written in the same order that pdflush sent them down.
 321  */
 322 static noinline int compress_file_range(struct inode *inode,
 323                                         struct page *locked_page,
 324                                         u64 start, u64 end,
 325                                         struct async_cow *async_cow,
 326                                         int *num_added)
 327 {
 328         struct btrfs_root *root = BTRFS_I(inode)->root;
 329         struct btrfs_trans_handle *trans;
 330         u64 num_bytes;
 331         u64 orig_start;
 332         u64 disk_num_bytes;
 333         u64 blocksize = root->sectorsize;
 334         u64 actual_end;
 335         int ret = 0;
 336         struct page **pages = NULL;
 337         unsigned long nr_pages;
 338         unsigned long nr_pages_ret = 0;
 339         unsigned long total_compressed = 0;
 340         unsigned long total_in = 0;
 341         unsigned long max_compressed = 128 * 1024;
 342         unsigned long max_uncompressed = 128 * 1024;
 343         int i;
 344         int will_compress;
 345
 346         orig_start = start;
 347
 348 again:
 349         will_compress = 0;
 350         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 351         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 352
 353         actual_end = min_t(u64, i_size_read(inode), end + 1);
 354         total_compressed = actual_end - start;
 355
 356         /* we want to make sure that amount of ram required to uncompress
 357          * an extent is reasonable, so we limit the total size in ram
 358          * of a compressed extent to 128k.  This is a crucial number
 359          * because it also controls how easily we can spread reads across
 360          * cpus for decompression.
 361          *
 362          * We also want to make sure the amount of IO required to do
 363          * a random read is reasonably small, so we limit the size of
 364          * a compressed extent to 128k.
 365          */
 366         total_compressed = min(total_compressed, max_uncompressed);
 367         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 368         num_bytes = max(blocksize,  num_bytes);
 369         disk_num_bytes = num_bytes;
 370         total_in = 0;
 371         ret = 0;
 372
 373         /*
 374          * we do compression for mount -o compress and when the
 375          * inode has not been flagged as nocompress.  This flag can
 376          * change at any time if we discover bad compression ratios.
 377          */
 378         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
 379             btrfs_test_opt(root, COMPRESS)) {
 380                 WARN_ON(pages);
 381                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 382
 383                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
 384                                                 total_compressed, pages,
 385                                                 nr_pages, &nr_pages_ret,
 386                                                 &total_in,
 387                                                 &total_compressed,
 388                                                 max_compressed);
 389
 390                 if (!ret) {
 391                         unsigned long offset = total_compressed &
 392                                 (PAGE_CACHE_SIZE - 1);
 393                         struct page *page = pages[nr_pages_ret - 1];
 394                         char *kaddr;
 395
 396                         /* zero the tail end of the last page, we might be
 397                          * sending it down to disk
 398                          */
 399                         if (offset) {
 400                                 kaddr = kmap_atomic(page, KM_USER0);
 401                                 memset(kaddr + offset, 0,
 402                                        PAGE_CACHE_SIZE - offset);
 403                                 kunmap_atomic(kaddr, KM_USER0);
 404                         }
 405                         will_compress = 1;
 406                 }
 407         }
 408         if (start == 0) {
 409                 trans = btrfs_join_transaction(root, 1);
 410                 BUG_ON(!trans);
 411                 btrfs_set_trans_block_group(trans, inode);
 412
 413                 /* lets try to make an inline extent */
 414                 if (ret || total_in < (actual_end - start)) {
 415                         /* we didn't compress the entire range, try
 416                          * to make an uncompressed inline extent.
 417                          */
 418                         ret = cow_file_range_inline(trans, root, inode,
 419                                                     start, end, 0, NULL);
 420                 } else {
 421                         /* try making a compressed inline extent */
 422                         ret = cow_file_range_inline(trans, root, inode,
 423                                                     start, end,
 424                                                     total_compressed, pages);
 425                 }
 426                 btrfs_end_transaction(trans, root);
 427                 if (ret == 0) {
 428                         /*
 429                          * inline extent creation worked, we don't need
 430                          * to create any more async work items.  Unlock
 431                          * and free up our temp pages.
 432                          */
 433                         extent_clear_unlock_delalloc(inode,
 434                                                      &BTRFS_I(inode)->io_tree,
 435                                                      start, end, NULL, 1, 0,
 436                                                      0, 1, 1, 1);
 437                         ret = 0;
 438                         goto free_pages_out;
 439                 }
 440         }
 441
 442         if (will_compress) {
 443                 /*
 444                  * we aren't doing an inline extent round the compressed size
 445                  * up to a block size boundary so the allocator does sane
 446                  * things
 447                  */
 448                 total_compressed = (total_compressed + blocksize - 1) &
 449                         ~(blocksize - 1);
 450
 451                 /*
 452                  * one last check to make sure the compression is really a
 453                  * win, compare the page count read with the blocks on disk
 454                  */
 455                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 456                         ~(PAGE_CACHE_SIZE - 1);
 457                 if (total_compressed >= total_in) {
 458                         will_compress = 0;
 459                 } else {
 460                         disk_num_bytes = total_compressed;
 461                         num_bytes = total_in;
 462                 }
 463         }
 464         if (!will_compress && pages) {
 465                 /*
 466                  * the compression code ran but failed to make things smaller,
 467                  * free any pages it allocated and our page pointer array
 468                  */
 469                 for (i = 0; i < nr_pages_ret; i++) {
 470                         WARN_ON(pages[i]->mapping);
 471                         page_cache_release(pages[i]);
 472                 }
 473                 kfree(pages);
 474                 pages = NULL;
 475                 total_compressed = 0;
 476                 nr_pages_ret = 0;
 477
 478                 /* flag the file so we don't compress in the future */
 479                 btrfs_set_flag(inode, NOCOMPRESS);
 480         }
 481         if (will_compress) {
 482                 *num_added += 1;
 483
 484                 /* the async work queues will take care of doing actual
 485                  * allocation on disk for these compressed pages,
 486                  * and will submit them to the elevator.
 487                  */
 488                 add_async_extent(async_cow, start, num_bytes,
 489                                  total_compressed, pages, nr_pages_ret);
 490
 491                 if (start + num_bytes < end) {
 492                         start += num_bytes;
 493                         pages = NULL;
 494                         cond_resched();
 495                         goto again;
 496                 }
 497         } else {
 498                 /*
 499                  * No compression, but we still need to write the pages in
 500                  * the file we've been given so far.  redirty the locked
 501                  * page if it corresponds to our extent and set things up
 502                  * for the async work queue to run cow_file_range to do
 503                  * the normal delalloc dance
 504                  */
 505                 if (page_offset(locked_page) >= start &&
 506                     page_offset(locked_page) <= end) {
 507                         __set_page_dirty_nobuffers(locked_page);
 508                         /* unlocked later on in the async handlers */
 509                 }
 510                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
 511                 *num_added += 1;
 512         }
 513
 514 out:
 515         return 0;
 516
 517 free_pages_out:
 518         for (i = 0; i < nr_pages_ret; i++) {
 519                 WARN_ON(pages[i]->mapping);
 520                 page_cache_release(pages[i]);
 521         }
 522         if (pages)
 523                 kfree(pages);
 524
 525         goto out;
 526 }
 527
 528 /*
 529  * phase two of compressed writeback.  This is the ordered portion
 530  * of the code, which only gets called in the order the work was
 531  * queued.  We walk all the async extents created by compress_file_range
 532  * and send them down to the disk.
 533  */
 534 static noinline int submit_compressed_extents(struct inode *inode,
 535                                               struct async_cow *async_cow)
 536 {
 537         struct async_extent *async_extent;
 538         u64 alloc_hint = 0;
 539         struct btrfs_trans_handle *trans;
 540         struct btrfs_key ins;
 541         struct extent_map *em;
 542         struct btrfs_root *root = BTRFS_I(inode)->root;
 543         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 544         struct extent_io_tree *io_tree;
 545         int ret;
 546
 547         if (list_empty(&async_cow->extents))
 548                 return 0;
 549
 550         trans = btrfs_join_transaction(root, 1);
 551
 552         while(!list_empty(&async_cow->extents)) {
 553                 async_extent = list_entry(async_cow->extents.next,
 554                                           struct async_extent, list);
 555                 list_del(&async_extent->list);
 556
 557                 io_tree = &BTRFS_I(inode)->io_tree;
 558
 559                 /* did the compression code fall back to uncompressed IO? */
 560                 if (!async_extent->pages) {
 561                         int page_started = 0;
 562                         unsigned long nr_written = 0;
 563
 564                         lock_extent(io_tree, async_extent->start,
 565                                     async_extent->start + async_extent->ram_size - 1,
 566                                     GFP_NOFS);
 567
 568                         /* allocate blocks */
 569                         cow_file_range(inode, async_cow->locked_page,
 570                                        async_extent->start,
 571                                        async_extent->start +
 572                                        async_extent->ram_size - 1,
 573                                        &page_started, &nr_written, 0);
 574
 575                         /*
 576                          * if page_started, cow_file_range inserted an
 577                          * inline extent and took care of all the unlocking
 578                          * and IO for us.  Otherwise, we need to submit
 579                          * all those pages down to the drive.
 580                          */
 581                         if (!page_started)
 582                                 extent_write_locked_range(io_tree,
 583                                                   inode, async_extent->start,
 584                                                   async_extent->start +
 585                                                   async_extent->ram_size - 1,
 586                                                   btrfs_get_extent,
 587                                                   WB_SYNC_ALL);
 588                         kfree(async_extent);
 589                         cond_resched();
 590                         continue;
 591                 }
 592
 593                 lock_extent(io_tree, async_extent->start,
 594                             async_extent->start + async_extent->ram_size - 1,
 595                             GFP_NOFS);
 596                 /*
 597                  * here we're doing allocation and writeback of the
 598                  * compressed pages
 599                  */
 600                 btrfs_drop_extent_cache(inode, async_extent->start,
 601                                         async_extent->start +
 602                                         async_extent->ram_size - 1, 0);
 603
 604                 ret = btrfs_reserve_extent(trans, root,
 605                                            async_extent->compressed_size,
 606                                            async_extent->compressed_size,
 607                                            0, alloc_hint,
 608                                            (u64)-1, &ins, 1);
 609                 BUG_ON(ret);
 610                 em = alloc_extent_map(GFP_NOFS);
 611                 em->start = async_extent->start;
 612                 em->len = async_extent->ram_size;
 613                 em->orig_start = em->start;
 614
 615                 em->block_start = ins.objectid;
 616                 em->block_len = ins.offset;
 617                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 618                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 619                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 620
 621                 while(1) {
 622                         spin_lock(&em_tree->lock);
 623                         ret = add_extent_mapping(em_tree, em);
 624                         spin_unlock(&em_tree->lock);
 625                         if (ret != -EEXIST) {
 626                                 free_extent_map(em);
 627                                 break;
 628                         }
 629                         btrfs_drop_extent_cache(inode, async_extent->start,
 630                                                 async_extent->start +
 631                                                 async_extent->ram_size - 1, 0);
 632                 }
 633
 634                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
 635                                                ins.objectid,
 636                                                async_extent->ram_size,
 637                                                ins.offset,
 638                                                BTRFS_ORDERED_COMPRESSED);
 639                 BUG_ON(ret);
 640
 641                 btrfs_end_transaction(trans, root);
 642
 643                 /*
 644                  * clear dirty, set writeback and unlock the pages.
 645                  */
 646                 extent_clear_unlock_delalloc(inode,
 647                                              &BTRFS_I(inode)->io_tree,
 648                                              async_extent->start,
 649                                              async_extent->start +
 650                                              async_extent->ram_size - 1,
 651                                              NULL, 1, 1, 0, 1, 1, 0);
 652
 653                 ret = btrfs_submit_compressed_write(inode,
 654                                          async_extent->start,
 655                                          async_extent->ram_size,
 656                                          ins.objectid,
 657                                          ins.offset, async_extent->pages,
 658                                          async_extent->nr_pages);
 659
 660                 BUG_ON(ret);
 661                 trans = btrfs_join_transaction(root, 1);
 662                 alloc_hint = ins.objectid + ins.offset;
 663                 kfree(async_extent);
 664                 cond_resched();
 665         }
 666
 667         btrfs_end_transaction(trans, root);
 668         return 0;
 669 }
 670
 671 /*
 672  * when extent_io.c finds a delayed allocation range in the file,
 673  * the call backs end up in this code.  The basic idea is to
 674  * allocate extents on disk for the range, and create ordered data structs
 675  * in ram to track those extents.
 676  *
 677  * locked_page is the page that writepage had locked already.  We use
 678  * it to make sure we don't do extra locks or unlocks.
 679  *
 680  * *page_started is set to one if we unlock locked_page and do everything
 681  * required to start IO on it.  It may be clean and already done with
 682  * IO when we return.
 683  */
 684 static noinline int cow_file_range(struct inode *inode,
 685                                    struct page *locked_page,
 686                                    u64 start, u64 end, int *page_started,
 687                                    unsigned long *nr_written,
 688                                    int unlock)
 689 {
 690         struct btrfs_root *root = BTRFS_I(inode)->root;
 691         struct btrfs_trans_handle *trans;
 692         u64 alloc_hint = 0;
 693         u64 num_bytes;
 694         unsigned long ram_size;
 695         u64 disk_num_bytes;
 696         u64 cur_alloc_size;
 697         u64 blocksize = root->sectorsize;
 698         u64 actual_end;
 699         struct btrfs_key ins;
 700         struct extent_map *em;
 701         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 702         int ret = 0;
 703
 704         trans = btrfs_join_transaction(root, 1);
 705         BUG_ON(!trans);
 706         btrfs_set_trans_block_group(trans, inode);
 707
 708         actual_end = min_t(u64, i_size_read(inode), end + 1);
 709
 710         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 711         num_bytes = max(blocksize,  num_bytes);
 712         disk_num_bytes = num_bytes;
 713         ret = 0;
 714
 715         if (start == 0) {
 716                 /* lets try to make an inline extent */
 717                 ret = cow_file_range_inline(trans, root, inode,
 718                                             start, end, 0, NULL);
 719                 if (ret == 0) {
 720                         extent_clear_unlock_delalloc(inode,
 721                                                      &BTRFS_I(inode)->io_tree,
 722                                                      start, end, NULL, 1, 1,
 723                                                      1, 1, 1, 1);
 724                         *nr_written = *nr_written +
 725                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 726                         *page_started = 1;
 727                         ret = 0;
 728                         goto out;
 729                 }
 730         }
 731
 732         BUG_ON(disk_num_bytes >
 733                btrfs_super_total_bytes(&root->fs_info->super_copy));
 734
 735         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 736
 737         while(disk_num_bytes > 0) {
 738                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
 739                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 740                                            root->sectorsize, 0, alloc_hint,
 741                                            (u64)-1, &ins, 1);
 742                 if (ret) {
 743                         BUG();
 744                 }
 745                 em = alloc_extent_map(GFP_NOFS);
 746                 em->start = start;
 747                 em->orig_start = em->start;
 748
 749                 ram_size = ins.offset;
 750                 em->len = ins.offset;
 751
 752                 em->block_start = ins.objectid;
 753                 em->block_len = ins.offset;
 754                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 755                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 756
 757                 while(1) {
 758                         spin_lock(&em_tree->lock);
 759                         ret = add_extent_mapping(em_tree, em);
 760                         spin_unlock(&em_tree->lock);
 761                         if (ret != -EEXIST) {
 762                                 free_extent_map(em);
 763                                 break;
 764                         }
 765                         btrfs_drop_extent_cache(inode, start,
 766                                                 start + ram_size - 1, 0);
 767                 }
 768
 769                 cur_alloc_size = ins.offset;
 770                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 771                                                ram_size, cur_alloc_size, 0);
 772                 BUG_ON(ret);
 773
 774                 if (disk_num_bytes < cur_alloc_size) {
 775                         printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
 776                                cur_alloc_size);
 777                         break;
 778                 }
 779                 /* we're not doing compressed IO, don't unlock the first
 780                  * page (which the caller expects to stay locked), don't
 781                  * clear any dirty bits and don't set any writeback bits
 782                  */
 783                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 784                                              start, start + ram_size - 1,
 785                                              locked_page, unlock, 1,
 786                                              1, 0, 0, 0);
 787                 disk_num_bytes -= cur_alloc_size;
 788                 num_bytes -= cur_alloc_size;
 789                 alloc_hint = ins.objectid + ins.offset;
 790                 start += cur_alloc_size;
 791         }
 792 out:
 793         ret = 0;
 794         btrfs_end_transaction(trans, root);
 795
 796         return ret;
 797 }
 798
 799 /*
 800  * work queue call back to started compression on a file and pages
 801  */
 802 static noinline void async_cow_start(struct btrfs_work *work)
 803 {
 804         struct async_cow *async_cow;
 805         int num_added = 0;
 806         async_cow = container_of(work, struct async_cow, work);
 807
 808         compress_file_range(async_cow->inode, async_cow->locked_page,
 809                             async_cow->start, async_cow->end, async_cow,
 810                             &num_added);
 811         if (num_added == 0)
 812                 async_cow->inode = NULL;
 813 }
 814
 815 /*
 816  * work queue call back to submit previously compressed pages
 817  */
 818 static noinline void async_cow_submit(struct btrfs_work *work)
 819 {
 820         struct async_cow *async_cow;
 821         struct btrfs_root *root;
 822         unsigned long nr_pages;
 823
 824         async_cow = container_of(work, struct async_cow, work);
 825
 826         root = async_cow->root;
 827         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 828                 PAGE_CACHE_SHIFT;
 829
 830         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 831
 832         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 833             5 * 1042 * 1024 &&
 834             waitqueue_active(&root->fs_info->async_submit_wait))
 835                 wake_up(&root->fs_info->async_submit_wait);
 836
 837         if (async_cow->inode) {
 838                 submit_compressed_extents(async_cow->inode, async_cow);
 839         }
 840 }
 841
 842 static noinline void async_cow_free(struct btrfs_work *work)
 843 {
 844         struct async_cow *async_cow;
 845         async_cow = container_of(work, struct async_cow, work);
 846         kfree(async_cow);
 847 }
 848
 849 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 850                                 u64 start, u64 end, int *page_started,
 851                                 unsigned long *nr_written)
 852 {
 853         struct async_cow *async_cow;
 854         struct btrfs_root *root = BTRFS_I(inode)->root;
 855         unsigned long nr_pages;
 856         u64 cur_end;
 857         int limit = 10 * 1024 * 1042;
 858
 859         if (!btrfs_test_opt(root, COMPRESS)) {
 860                 return cow_file_range(inode, locked_page, start, end,
 861                                       page_started, nr_written, 1);
 862         }
 863
 864         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
 865                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
 866         while(start < end) {
 867                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 868                 async_cow->inode = inode;
 869                 async_cow->root = root;
 870                 async_cow->locked_page = locked_page;
 871                 async_cow->start = start;
 872
 873                 if (btrfs_test_flag(inode, NOCOMPRESS))
 874                         cur_end = end;
 875                 else
 876                         cur_end = min(end, start + 512 * 1024 - 1);
 877
 878                 async_cow->end = cur_end;
 879                 INIT_LIST_HEAD(&async_cow->extents);
 880
 881                 async_cow->work.func = async_cow_start;
 882                 async_cow->work.ordered_func = async_cow_submit;
 883                 async_cow->work.ordered_free = async_cow_free;
 884                 async_cow->work.flags = 0;
 885
 886                 while(atomic_read(&root->fs_info->async_submit_draining) &&
 887                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 888                         wait_event(root->fs_info->async_submit_wait,
 889                              (atomic_read(&root->fs_info->async_delalloc_pages)
 890                               == 0));
 891                 }
 892
 893                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 894                         PAGE_CACHE_SHIFT;
 895                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 896
 897                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 898                                    &async_cow->work);
 899
 900                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 901                         wait_event(root->fs_info->async_submit_wait,
 902                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 903                             limit));
 904                 }
 905
 906                 while(atomic_read(&root->fs_info->async_submit_draining) &&
 907                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 908                         wait_event(root->fs_info->async_submit_wait,
 909                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 910                            0));
 911                 }
 912
 913                 *nr_written += nr_pages;
 914                 start = cur_end + 1;
 915         }
 916         *page_started = 1;
 917         return 0;
 918 }
 919
 920 /*
 921  * when nowcow writeback call back.  This checks for snapshots or COW copies
 922  * of the extents that exist in the file, and COWs the file as required.
 923  *
 924  * If no cow copies or snapshots exist, we write directly to the existing
 925  * blocks on disk
 926  */
 927 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
 928                               u64 start, u64 end, int *page_started, int force,
 929                               unsigned long *nr_written)
 930 {
 931         struct btrfs_root *root = BTRFS_I(inode)->root;
 932         struct btrfs_trans_handle *trans;
 933         struct extent_buffer *leaf;
 934         struct btrfs_path *path;
 935         struct btrfs_file_extent_item *fi;
 936         struct btrfs_key found_key;
 937         u64 cow_start;
 938         u64 cur_offset;
 939         u64 extent_end;
 940         u64 disk_bytenr;
 941         u64 num_bytes;
 942         int extent_type;
 943         int ret;
 944         int type;
 945         int nocow;
 946         int check_prev = 1;
 947
 948         path = btrfs_alloc_path();
 949         BUG_ON(!path);
 950         trans = btrfs_join_transaction(root, 1);
 951         BUG_ON(!trans);
 952
 953         cow_start = (u64)-1;
 954         cur_offset = start;
 955         while (1) {
 956                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 957                                                cur_offset, 0);
 958                 BUG_ON(ret < 0);
 959                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
 960                         leaf = path->nodes[0];
 961                         btrfs_item_key_to_cpu(leaf, &found_key,
 962                                               path->slots[0] - 1);
 963                         if (found_key.objectid == inode->i_ino &&
 964                             found_key.type == BTRFS_EXTENT_DATA_KEY)
 965                                 path->slots[0]--;
 966                 }
 967                 check_prev = 0;
 968 next_slot:
 969                 leaf = path->nodes[0];
 970                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 971                         ret = btrfs_next_leaf(root, path);
 972                         if (ret < 0)
 973                                 BUG_ON(1);
 974                         if (ret > 0)
 975                                 break;
 976                         leaf = path->nodes[0];
 977                 }
 978
 979                 nocow = 0;
 980                 disk_bytenr = 0;
 981                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 982
 983                 if (found_key.objectid > inode->i_ino ||
 984                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
 985                     found_key.offset > end)
 986                         break;
 987
 988                 if (found_key.offset > cur_offset) {
 989                         extent_end = found_key.offset;
 990                         goto out_check;
 991                 }
 992
 993                 fi = btrfs_item_ptr(leaf, path->slots[0],
 994                                     struct btrfs_file_extent_item);
 995                 extent_type = btrfs_file_extent_type(leaf, fi);
 996
 997                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 998                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 999                         struct btrfs_block_group_cache *block_group;
1000                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1001                         extent_end = found_key.offset +
1002                                 btrfs_file_extent_num_bytes(leaf, fi);
1003                         if (extent_end <= start) {
1004                                 path->slots[0]++;
1005                                 goto next_slot;
1006                         }
1007                         if (btrfs_file_extent_compression(leaf, fi) ||
1008                             btrfs_file_extent_encryption(leaf, fi) ||
1009                             btrfs_file_extent_other_encoding(leaf, fi))
1010                                 goto out_check;
1011                         if (disk_bytenr == 0)
1012                                 goto out_check;
1013                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1014                                 goto out_check;
1015                         if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
1016                                 goto out_check;
1017                         block_group = btrfs_lookup_block_group(root->fs_info,
1018                                                                disk_bytenr);
1019                         if (!block_group || block_group->ro)
1020                                 goto out_check;
1021                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1022                         nocow = 1;
1023                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1024                         extent_end = found_key.offset +
1025                                 btrfs_file_extent_inline_len(leaf, fi);
1026                         extent_end = ALIGN(extent_end, root->sectorsize);
1027                 } else {
1028                         BUG_ON(1);
1029                 }
1030 out_check:
1031                 if (extent_end <= start) {
1032                         path->slots[0]++;
1033                         goto next_slot;
1034                 }
1035                 if (!nocow) {
1036                         if (cow_start == (u64)-1)
1037                                 cow_start = cur_offset;
1038                         cur_offset = extent_end;
1039                         if (cur_offset > end)
1040                                 break;
1041                         path->slots[0]++;
1042                         goto next_slot;
1043                 }
1044
1045                 btrfs_release_path(root, path);
1046                 if (cow_start != (u64)-1) {
1047                         ret = cow_file_range(inode, locked_page, cow_start,
1048                                         found_key.offset - 1, page_started,
1049                                         nr_written, 1);
1050                         BUG_ON(ret);
1051                         cow_start = (u64)-1;
1052                 }
1053
1054                 disk_bytenr += cur_offset - found_key.offset;
1055                 num_bytes = min(end + 1, extent_end) - cur_offset;
1056                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1057                         struct extent_map *em;
1058                         struct extent_map_tree *em_tree;
1059                         em_tree = &BTRFS_I(inode)->extent_tree;
1060                         em = alloc_extent_map(GFP_NOFS);
1061                         em->start = cur_offset;
1062                         em->orig_start = em->start;
1063                         em->len = num_bytes;
1064                         em->block_len = num_bytes;
1065                         em->block_start = disk_bytenr;
1066                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1067                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1068                         while (1) {
1069                                 spin_lock(&em_tree->lock);
1070                                 ret = add_extent_mapping(em_tree, em);
1071                                 spin_unlock(&em_tree->lock);
1072                                 if (ret != -EEXIST) {
1073                                         free_extent_map(em);
1074                                         break;
1075                                 }
1076                                 btrfs_drop_extent_cache(inode, em->start,
1077                                                 em->start + em->len - 1, 0);
1078                         }
1079                         type = BTRFS_ORDERED_PREALLOC;
1080                 } else {
1081                         type = BTRFS_ORDERED_NOCOW;
1082                 }
1083
1084                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1085                                                num_bytes, num_bytes, type);
1086                 BUG_ON(ret);
1087
1088                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1089                                         cur_offset, cur_offset + num_bytes - 1,
1090                                         locked_page, 1, 1, 1, 0, 0, 0);
1091                 cur_offset = extent_end;
1092                 if (cur_offset > end)
1093                         break;
1094         }
1095         btrfs_release_path(root, path);
1096
1097         if (cur_offset <= end && cow_start == (u64)-1)
1098                 cow_start = cur_offset;
1099         if (cow_start != (u64)-1) {
1100                 ret = cow_file_range(inode, locked_page, cow_start, end,
1101                                      page_started, nr_written, 1);
1102                 BUG_ON(ret);
1103         }
1104
1105         ret = btrfs_end_transaction(trans, root);
1106         BUG_ON(ret);
1107         btrfs_free_path(path);
1108         return 0;
1109 }
1110
1111 /*
1112  * extent_io.c call back to do delayed allocation processing
1113  */
1114 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1115                               u64 start, u64 end, int *page_started,
1116                               unsigned long *nr_written)
1117 {
1118         struct btrfs_root *root = BTRFS_I(inode)->root;
1119         int ret;
1120
1121         if (btrfs_test_opt(root, NODATACOW) ||
1122             btrfs_test_flag(inode, NODATACOW))
1123                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1124                                          page_started, 0, nr_written);
1125         else if (btrfs_test_flag(inode, PREALLOC))
1126                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1127                                          page_started, 1, nr_written);
1128         else
1129                 ret = cow_file_range_async(inode, locked_page, start, end,
1130                                      page_started, nr_written);
1131
1132         return ret;
1133 }
1134
1135 /*
1136  * extent_io.c set_bit_hook, used to track delayed allocation
1137  * bytes in this file, and to maintain the list of inodes that
1138  * have pending delalloc work to be done.
1139  */
1140 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1141                        unsigned long old, unsigned long bits)
1142 {
1143         unsigned long flags;
1144         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1145                 struct btrfs_root *root = BTRFS_I(inode)->root;
1146                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1147                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1148                 root->fs_info->delalloc_bytes += end - start + 1;
1149                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1150                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1151                                       &root->fs_info->delalloc_inodes);
1152                 }
1153                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1154         }
1155         return 0;
1156 }
1157
1158 /*
1159  * extent_io.c clear_bit_hook, see set_bit_hook for why
1160  */
1161 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1162                          unsigned long old, unsigned long bits)
1163 {
1164         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1165                 struct btrfs_root *root = BTRFS_I(inode)->root;
1166                 unsigned long flags;
1167
1168                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1169                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1170                         printk("warning: delalloc account %Lu %Lu\n",
1171                                end - start + 1, root->fs_info->delalloc_bytes);
1172                         root->fs_info->delalloc_bytes = 0;
1173                         BTRFS_I(inode)->delalloc_bytes = 0;
1174                 } else {
1175                         root->fs_info->delalloc_bytes -= end - start + 1;
1176                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1177                 }
1178                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1179                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1180                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1181                 }
1182                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1183         }
1184         return 0;
1185 }
1186
1187 /*
1188  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1189  * we don't create bios that span stripes or chunks
1190  */
1191 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1192                          size_t size, struct bio *bio,
1193                          unsigned long bio_flags)
1194 {
1195         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1196         struct btrfs_mapping_tree *map_tree;
1197         u64 logical = (u64)bio->bi_sector << 9;
1198         u64 length = 0;
1199         u64 map_length;
1200         int ret;
1201
1202         if (bio_flags & EXTENT_BIO_COMPRESSED)
1203                 return 0;
1204
1205         length = bio->bi_size;
1206         map_tree = &root->fs_info->mapping_tree;
1207         map_length = length;
1208         ret = btrfs_map_block(map_tree, READ, logical,
1209                               &map_length, NULL, 0);
1210
1211         if (map_length < length + size) {
1212                 return 1;
1213         }
1214         return 0;
1215 }
1216
1217 /*
1218  * in order to insert checksums into the metadata in large chunks,
1219  * we wait until bio submission time.   All the pages in the bio are
1220  * checksummed and sums are attached onto the ordered extent record.
1221  *
1222  * At IO completion time the cums attached on the ordered extent record
1223  * are inserted into the btree
1224  */
1225 int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
1226                           int mirror_num, unsigned long bio_flags)
1227 {
1228         struct btrfs_root *root = BTRFS_I(inode)->root;
1229         int ret = 0;
1230
1231         ret = btrfs_csum_one_bio(root, inode, bio);
1232         BUG_ON(ret);
1233         return 0;
1234 }
1235
1236 /*
1237  * in order to insert checksums into the metadata in large chunks,
1238  * we wait until bio submission time.   All the pages in the bio are
1239  * checksummed and sums are attached onto the ordered extent record.
1240  *
1241  * At IO completion time the cums attached on the ordered extent record
1242  * are inserted into the btree
1243  */
1244 int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1245                           int mirror_num, unsigned long bio_flags)
1246 {
1247         struct btrfs_root *root = BTRFS_I(inode)->root;
1248         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1249 }
1250
1251 /*
1252  * extent_io.c submission hook. This does the right thing for csum calculation on write,
1253  * or reading the csums from the tree before a read
1254  */
1255 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1256                           int mirror_num, unsigned long bio_flags)
1257 {
1258         struct btrfs_root *root = BTRFS_I(inode)->root;
1259         int ret = 0;
1260         int skip_sum;
1261
1262         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1263         BUG_ON(ret);
1264
1265         skip_sum = btrfs_test_opt(root, NODATASUM) ||
1266                 btrfs_test_flag(inode, NODATASUM);
1267
1268         if (!(rw & (1 << BIO_RW))) {
1269
1270                 if (bio_flags & EXTENT_BIO_COMPRESSED)
1271                         return btrfs_submit_compressed_read(inode, bio,
1272                                                     mirror_num, bio_flags);
1273                 else if (!skip_sum)
1274                         btrfs_lookup_bio_sums(root, inode, bio);
1275                 goto mapit;
1276         } else if (!skip_sum) {
1277                 /* we're doing a write, do the async checksumming */
1278                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1279                                    inode, rw, bio, mirror_num,
1280                                    bio_flags, __btrfs_submit_bio_start,
1281                                    __btrfs_submit_bio_done);
1282         }
1283
1284 mapit:
1285         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1286 }
1287
1288 /*
1289  * given a list of ordered sums record them in the inode.  This happens
1290  * at IO completion time based on sums calculated at bio submission time.
1291  */
1292 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1293                              struct inode *inode, u64 file_offset,
1294                              struct list_head *list)
1295 {
1296         struct list_head *cur;
1297         struct btrfs_ordered_sum *sum;
1298
1299         btrfs_set_trans_block_group(trans, inode);
1300         list_for_each(cur, list) {
1301                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
1302                 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
1303                                        inode, sum);
1304         }
1305         return 0;
1306 }
1307
1308 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1309 {
1310         if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1311                 WARN_ON(1);
1312         }
1313         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1314                                    GFP_NOFS);
1315 }
1316
1317 /* see btrfs_writepage_start_hook for details on why this is required */
1318 struct btrfs_writepage_fixup {
1319         struct page *page;
1320         struct btrfs_work work;
1321 };
1322
1323 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1324 {
1325         struct btrfs_writepage_fixup *fixup;
1326         struct btrfs_ordered_extent *ordered;
1327         struct page *page;
1328         struct inode *inode;
1329         u64 page_start;
1330         u64 page_end;
1331
1332         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1333         page = fixup->page;
1334 again:
1335         lock_page(page);
1336         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1337                 ClearPageChecked(page);
1338                 goto out_page;
1339         }
1340
1341         inode = page->mapping->host;
1342         page_start = page_offset(page);
1343         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1344
1345         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1346
1347         /* already ordered? We're done */
1348         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1349                              EXTENT_ORDERED, 0)) {
1350                 goto out;
1351         }
1352
1353         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1354         if (ordered) {
1355                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1356                               page_end, GFP_NOFS);
1357                 unlock_page(page);
1358                 btrfs_start_ordered_extent(inode, ordered, 1);
1359                 goto again;
1360         }
1361
1362         btrfs_set_extent_delalloc(inode, page_start, page_end);
1363         ClearPageChecked(page);
1364 out:
1365         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1366 out_page:
1367         unlock_page(page);
1368         page_cache_release(page);
1369 }
1370
1371 /*
1372  * There are a few paths in the higher layers of the kernel that directly
1373  * set the page dirty bit without asking the filesystem if it is a
1374  * good idea.  This causes problems because we want to make sure COW
1375  * properly happens and the data=ordered rules are followed.
1376  *
1377  * In our case any range that doesn't have the ORDERED bit set
1378  * hasn't been properly setup for IO.  We kick off an async process
1379  * to fix it up.  The async helper will wait for ordered extents, set
1380  * the delalloc bit and make it safe to write the page.
1381  */
1382 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1383 {
1384         struct inode *inode = page->mapping->host;
1385         struct btrfs_writepage_fixup *fixup;
1386         struct btrfs_root *root = BTRFS_I(inode)->root;
1387         int ret;
1388
1389         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1390                              EXTENT_ORDERED, 0);
1391         if (ret)
1392                 return 0;
1393
1394         if (PageChecked(page))
1395                 return -EAGAIN;
1396
1397         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1398         if (!fixup)
1399                 return -EAGAIN;
1400
1401         SetPageChecked(page);
1402         page_cache_get(page);
1403         fixup->work.func = btrfs_writepage_fixup_worker;
1404         fixup->page = page;
1405         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1406         return -EAGAIN;
1407 }
1408
1409 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1410                                        struct inode *inode, u64 file_pos,
1411                                        u64 disk_bytenr, u64 disk_num_bytes,
1412                                        u64 num_bytes, u64 ram_bytes,
1413                                        u8 compression, u8 encryption,
1414                                        u16 other_encoding, int extent_type)
1415 {
1416         struct btrfs_root *root = BTRFS_I(inode)->root;
1417         struct btrfs_file_extent_item *fi;
1418         struct btrfs_path *path;
1419         struct extent_buffer *leaf;
1420         struct btrfs_key ins;
1421         u64 hint;
1422         int ret;
1423
1424         path = btrfs_alloc_path();
1425         BUG_ON(!path);
1426
1427         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1428                                  file_pos + num_bytes, file_pos, &hint);
1429         BUG_ON(ret);
1430
1431         ins.objectid = inode->i_ino;
1432         ins.offset = file_pos;
1433         ins.type = BTRFS_EXTENT_DATA_KEY;
1434         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1435         BUG_ON(ret);
1436         leaf = path->nodes[0];
1437         fi = btrfs_item_ptr(leaf, path->slots[0],
1438                             struct btrfs_file_extent_item);
1439         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1440         btrfs_set_file_extent_type(leaf, fi, extent_type);
1441         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1442         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1443         btrfs_set_file_extent_offset(leaf, fi, 0);
1444         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1445         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1446         btrfs_set_file_extent_compression(leaf, fi, compression);
1447         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1448         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1449         btrfs_mark_buffer_dirty(leaf);
1450
1451         inode_add_bytes(inode, num_bytes);
1452         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1453
1454         ins.objectid = disk_bytenr;
1455         ins.offset = disk_num_bytes;
1456         ins.type = BTRFS_EXTENT_ITEM_KEY;
1457         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1458                                           root->root_key.objectid,
1459                                           trans->transid, inode->i_ino, &ins);
1460         BUG_ON(ret);
1461
1462         btrfs_free_path(path);
1463         return 0;
1464 }
1465
1466 /* as ordered data IO finishes, this gets called so we can finish
1467  * an ordered extent if the range of bytes in the file it covers are
1468  * fully written.
1469  */
1470 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1471 {
1472         struct btrfs_root *root = BTRFS_I(inode)->root;
1473         struct btrfs_trans_handle *trans;
1474         struct btrfs_ordered_extent *ordered_extent;
1475         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1476         int compressed = 0;
1477         int ret;
1478
1479         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1480         if (!ret)
1481                 return 0;
1482
1483         trans = btrfs_join_transaction(root, 1);
1484
1485         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1486         BUG_ON(!ordered_extent);
1487         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1488                 goto nocow;
1489
1490         lock_extent(io_tree, ordered_extent->file_offset,
1491                     ordered_extent->file_offset + ordered_extent->len - 1,
1492                     GFP_NOFS);
1493
1494         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1495                 compressed = 1;
1496         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1497                 BUG_ON(compressed);
1498                 ret = btrfs_mark_extent_written(trans, root, inode,
1499                                                 ordered_extent->file_offset,
1500                                                 ordered_extent->file_offset +
1501                                                 ordered_extent->len);
1502                 BUG_ON(ret);
1503         } else {
1504                 ret = insert_reserved_file_extent(trans, inode,
1505                                                 ordered_extent->file_offset,
1506                                                 ordered_extent->start,
1507                                                 ordered_extent->disk_len,
1508                                                 ordered_extent->len,
1509                                                 ordered_extent->len,
1510                                                 compressed, 0, 0,
1511                                                 BTRFS_FILE_EXTENT_REG);
1512                 BUG_ON(ret);
1513         }
1514         unlock_extent(io_tree, ordered_extent->file_offset,
1515                     ordered_extent->file_offset + ordered_extent->len - 1,
1516                     GFP_NOFS);
1517 nocow:
1518         add_pending_csums(trans, inode, ordered_extent->file_offset,
1519                           &ordered_extent->list);
1520
1521         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1522         btrfs_ordered_update_i_size(inode, ordered_extent);
1523         btrfs_update_inode(trans, root, inode);
1524         btrfs_remove_ordered_extent(inode, ordered_extent);
1525         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1526
1527         /* once for us */
1528         btrfs_put_ordered_extent(ordered_extent);
1529         /* once for the tree */
1530         btrfs_put_ordered_extent(ordered_extent);
1531
1532         btrfs_end_transaction(trans, root);
1533         return 0;
1534 }
1535
1536 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1537                                 struct extent_state *state, int uptodate)
1538 {
1539         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1540 }
1541
1542 /*
1543  * When IO fails, either with EIO or csum verification fails, we
1544  * try other mirrors that might have a good copy of the data.  This
1545  * io_failure_record is used to record state as we go through all the
1546  * mirrors.  If another mirror has good data, the page is set up to date
1547  * and things continue.  If a good mirror can't be found, the original
1548  * bio end_io callback is called to indicate things have failed.
1549  */
1550 struct io_failure_record {
1551         struct page *page;
1552         u64 start;
1553         u64 len;
1554         u64 logical;
1555         int last_mirror;
1556 };
1557
1558 int btrfs_io_failed_hook(struct bio *failed_bio,
1559                          struct page *page, u64 start, u64 end,
1560                          struct extent_state *state)
1561 {
1562         struct io_failure_record *failrec = NULL;
1563         u64 private;
1564         struct extent_map *em;
1565         struct inode *inode = page->mapping->host;
1566         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1567         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1568         struct bio *bio;
1569         int num_copies;
1570         int ret;
1571         int rw;
1572         u64 logical;
1573         unsigned long bio_flags = 0;
1574
1575         ret = get_state_private(failure_tree, start, &private);
1576         if (ret) {
1577                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1578                 if (!failrec)
1579                         return -ENOMEM;
1580                 failrec->start = start;
1581                 failrec->len = end - start + 1;
1582                 failrec->last_mirror = 0;
1583
1584                 spin_lock(&em_tree->lock);
1585                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1586                 if (em->start > start || em->start + em->len < start) {
1587                         free_extent_map(em);
1588                         em = NULL;
1589                 }
1590                 spin_unlock(&em_tree->lock);
1591
1592                 if (!em || IS_ERR(em)) {
1593                         kfree(failrec);
1594                         return -EIO;
1595                 }
1596                 logical = start - em->start;
1597                 logical = em->block_start + logical;
1598                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags))
1599                         bio_flags = EXTENT_BIO_COMPRESSED;
1600                 failrec->logical = logical;
1601                 free_extent_map(em);
1602                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1603                                 EXTENT_DIRTY, GFP_NOFS);
1604                 set_state_private(failure_tree, start,
1605                                  (u64)(unsigned long)failrec);
1606         } else {
1607                 failrec = (struct io_failure_record *)(unsigned long)private;
1608         }
1609         num_copies = btrfs_num_copies(
1610                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1611                               failrec->logical, failrec->len);
1612         failrec->last_mirror++;
1613         if (!state) {
1614                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1615                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1616                                                     failrec->start,
1617                                                     EXTENT_LOCKED);
1618                 if (state && state->start != failrec->start)
1619                         state = NULL;
1620                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1621         }
1622         if (!state || failrec->last_mirror > num_copies) {
1623                 set_state_private(failure_tree, failrec->start, 0);
1624                 clear_extent_bits(failure_tree, failrec->start,
1625                                   failrec->start + failrec->len - 1,
1626                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1627                 kfree(failrec);
1628                 return -EIO;
1629         }
1630         bio = bio_alloc(GFP_NOFS, 1);
1631         bio->bi_private = state;
1632         bio->bi_end_io = failed_bio->bi_end_io;
1633         bio->bi_sector = failrec->logical >> 9;
1634         bio->bi_bdev = failed_bio->bi_bdev;
1635         bio->bi_size = 0;
1636         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1637         if (failed_bio->bi_rw & (1 << BIO_RW))
1638                 rw = WRITE;
1639         else
1640                 rw = READ;
1641
1642         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1643                                                       failrec->last_mirror,
1644                                                       bio_flags);
1645         return 0;
1646 }
1647
1648 /*
1649  * each time an IO finishes, we do a fast check in the IO failure tree
1650  * to see if we need to process or clean up an io_failure_record
1651  */
1652 int btrfs_clean_io_failures(struct inode *inode, u64 start)
1653 {
1654         u64 private;
1655         u64 private_failure;
1656         struct io_failure_record *failure;
1657         int ret;
1658
1659         private = 0;
1660         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1661                              (u64)-1, 1, EXTENT_DIRTY)) {
1662                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1663                                         start, &private_failure);
1664                 if (ret == 0) {
1665                         failure = (struct io_failure_record *)(unsigned long)
1666                                    private_failure;
1667                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1668                                           failure->start, 0);
1669                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1670                                           failure->start,
1671                                           failure->start + failure->len - 1,
1672                                           EXTENT_DIRTY | EXTENT_LOCKED,
1673                                           GFP_NOFS);
1674                         kfree(failure);
1675                 }
1676         }
1677         return 0;
1678 }
1679
1680 /*
1681  * when reads are done, we need to check csums to verify the data is correct
1682  * if there's a match, we allow the bio to finish.  If not, we go through
1683  * the io_failure_record routines to find good copies
1684  */
1685 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1686                                struct extent_state *state)
1687 {
1688         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1689         struct inode *inode = page->mapping->host;
1690         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1691         char *kaddr;
1692         u64 private = ~(u32)0;
1693         int ret;
1694         struct btrfs_root *root = BTRFS_I(inode)->root;
1695         u32 csum = ~(u32)0;
1696         unsigned long flags;
1697
1698         if (btrfs_test_opt(root, NODATASUM) ||
1699             btrfs_test_flag(inode, NODATASUM))
1700                 return 0;
1701         if (state && state->start == start) {
1702                 private = state->private;
1703                 ret = 0;
1704         } else {
1705                 ret = get_state_private(io_tree, start, &private);
1706         }
1707         local_irq_save(flags);
1708         kaddr = kmap_atomic(page, KM_IRQ0);
1709         if (ret) {
1710                 goto zeroit;
1711         }
1712         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1713         btrfs_csum_final(csum, (char *)&csum);
1714         if (csum != private) {
1715                 goto zeroit;
1716         }
1717         kunmap_atomic(kaddr, KM_IRQ0);
1718         local_irq_restore(flags);
1719
1720         /* if the io failure tree for this inode is non-empty,
1721          * check to see if we've recovered from a failed IO
1722          */
1723         btrfs_clean_io_failures(inode, start);
1724         return 0;
1725
1726 zeroit:
1727         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1728                page->mapping->host->i_ino, (unsigned long long)start, csum,
1729                private);
1730         memset(kaddr + offset, 1, end - start + 1);
1731         flush_dcache_page(page);
1732         kunmap_atomic(kaddr, KM_IRQ0);
1733         local_irq_restore(flags);
1734         if (private == 0)
1735                 return 0;
1736         return -EIO;
1737 }
1738
1739 /*
1740  * This creates an orphan entry for the given inode in case something goes
1741  * wrong in the middle of an unlink/truncate.
1742  */
1743 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1744 {
1745         struct btrfs_root *root = BTRFS_I(inode)->root;
1746         int ret = 0;
1747
1748         spin_lock(&root->list_lock);
1749
1750         /* already on the orphan list, we're good */
1751         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1752                 spin_unlock(&root->list_lock);
1753                 return 0;
1754         }
1755
1756         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1757
1758         spin_unlock(&root->list_lock);
1759
1760         /*
1761          * insert an orphan item to track this unlinked/truncated file
1762          */
1763         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1764
1765         return ret;
1766 }
1767
1768 /*
1769  * We have done the truncate/delete so we can go ahead and remove the orphan
1770  * item for this particular inode.
1771  */
1772 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1773 {
1774         struct btrfs_root *root = BTRFS_I(inode)->root;
1775         int ret = 0;
1776
1777         spin_lock(&root->list_lock);
1778
1779         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1780                 spin_unlock(&root->list_lock);
1781                 return 0;
1782         }
1783
1784         list_del_init(&BTRFS_I(inode)->i_orphan);
1785         if (!trans) {
1786                 spin_unlock(&root->list_lock);
1787                 return 0;
1788         }
1789
1790         spin_unlock(&root->list_lock);
1791
1792         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1793
1794         return ret;
1795 }
1796
1797 /*
1798  * this cleans up any orphans that may be left on the list from the last use
1799  * of this root.
1800  */
1801 void btrfs_orphan_cleanup(struct btrfs_root *root)
1802 {
1803         struct btrfs_path *path;
1804         struct extent_buffer *leaf;
1805         struct btrfs_item *item;
1806         struct btrfs_key key, found_key;
1807         struct btrfs_trans_handle *trans;
1808         struct inode *inode;
1809         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1810
1811         path = btrfs_alloc_path();
1812         if (!path)
1813                 return;
1814         path->reada = -1;
1815
1816         key.objectid = BTRFS_ORPHAN_OBJECTID;
1817         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1818         key.offset = (u64)-1;
1819
1820
1821         while (1) {
1822                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1823                 if (ret < 0) {
1824                         printk(KERN_ERR "Error searching slot for orphan: %d"
1825                                "\n", ret);
1826                         break;
1827                 }
1828
1829                 /*
1830                  * if ret == 0 means we found what we were searching for, which
1831                  * is weird, but possible, so only screw with path if we didnt
1832                  * find the key and see if we have stuff that matches
1833                  */
1834                 if (ret > 0) {
1835                         if (path->slots[0] == 0)
1836                                 break;
1837                         path->slots[0]--;
1838                 }
1839
1840                 /* pull out the item */
1841                 leaf = path->nodes[0];
1842                 item = btrfs_item_nr(leaf, path->slots[0]);
1843                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1844
1845                 /* make sure the item matches what we want */
1846                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1847                         break;
1848                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1849                         break;
1850
1851                 /* release the path since we're done with it */
1852                 btrfs_release_path(root, path);
1853
1854                 /*
1855                  * this is where we are basically btrfs_lookup, without the
1856                  * crossing root thing.  we store the inode number in the
1857                  * offset of the orphan item.
1858                  */
1859                 inode = btrfs_iget_locked(root->fs_info->sb,
1860                                           found_key.offset, root);
1861                 if (!inode)
1862                         break;
1863
1864                 if (inode->i_state & I_NEW) {
1865                         BTRFS_I(inode)->root = root;
1866
1867                         /* have to set the location manually */
1868                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1869                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1870                         BTRFS_I(inode)->location.offset = 0;
1871
1872                         btrfs_read_locked_inode(inode);
1873                         unlock_new_inode(inode);
1874                 }
1875
1876                 /*
1877                  * add this inode to the orphan list so btrfs_orphan_del does
1878                  * the proper thing when we hit it
1879                  */
1880                 spin_lock(&root->list_lock);
1881                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1882                 spin_unlock(&root->list_lock);
1883
1884                 /*
1885                  * if this is a bad inode, means we actually succeeded in
1886                  * removing the inode, but not the orphan record, which means
1887                  * we need to manually delete the orphan since iput will just
1888                  * do a destroy_inode
1889                  */
1890                 if (is_bad_inode(inode)) {
1891                         trans = btrfs_start_transaction(root, 1);
1892                         btrfs_orphan_del(trans, inode);
1893                         btrfs_end_transaction(trans, root);
1894                         iput(inode);
1895                         continue;
1896                 }
1897
1898                 /* if we have links, this was a truncate, lets do that */
1899                 if (inode->i_nlink) {
1900                         nr_truncate++;
1901                         btrfs_truncate(inode);
1902                 } else {
1903                         nr_unlink++;
1904                 }
1905
1906                 /* this will do delete_inode and everything for us */
1907                 iput(inode);
1908         }
1909
1910         if (nr_unlink)
1911                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1912         if (nr_truncate)
1913                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1914
1915         btrfs_free_path(path);
1916 }
1917
1918 /*
1919  * read an inode from the btree into the in-memory inode
1920  */
1921 void btrfs_read_locked_inode(struct inode *inode)
1922 {
1923         struct btrfs_path *path;
1924         struct extent_buffer *leaf;
1925         struct btrfs_inode_item *inode_item;
1926         struct btrfs_timespec *tspec;
1927         struct btrfs_root *root = BTRFS_I(inode)->root;
1928         struct btrfs_key location;
1929         u64 alloc_group_block;
1930         u32 rdev;
1931         int ret;
1932
1933         path = btrfs_alloc_path();
1934         BUG_ON(!path);
1935         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1936
1937         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1938         if (ret)
1939                 goto make_bad;
1940
1941         leaf = path->nodes[0];
1942         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1943                                     struct btrfs_inode_item);
1944
1945         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1946         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1947         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1948         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1949         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1950
1951         tspec = btrfs_inode_atime(inode_item);
1952         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1953         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1954
1955         tspec = btrfs_inode_mtime(inode_item);
1956         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1957         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1958
1959         tspec = btrfs_inode_ctime(inode_item);
1960         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1961         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1962
1963         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
1964         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1965         inode->i_generation = BTRFS_I(inode)->generation;
1966         inode->i_rdev = 0;
1967         rdev = btrfs_inode_rdev(leaf, inode_item);
1968
1969         BTRFS_I(inode)->index_cnt = (u64)-1;
1970
1971         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1972         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1973                                                        alloc_group_block);
1974         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1975         if (!BTRFS_I(inode)->block_group) {
1976                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1977                                                  NULL, 0,
1978                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1979         }
1980         btrfs_free_path(path);
1981         inode_item = NULL;
1982
1983         switch (inode->i_mode & S_IFMT) {
1984         case S_IFREG:
1985                 inode->i_mapping->a_ops = &btrfs_aops;
1986                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1987                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1988                 inode->i_fop = &btrfs_file_operations;
1989                 inode->i_op = &btrfs_file_inode_operations;
1990                 break;
1991         case S_IFDIR:
1992                 inode->i_fop = &btrfs_dir_file_operations;
1993                 if (root == root->fs_info->tree_root)
1994                         inode->i_op = &btrfs_dir_ro_inode_operations;
1995                 else
1996                         inode->i_op = &btrfs_dir_inode_operations;
1997                 break;
1998         case S_IFLNK:
1999                 inode->i_op = &btrfs_symlink_inode_operations;
2000                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2001                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2002                 break;
2003         default:
2004                 init_special_inode(inode, inode->i_mode, rdev);
2005                 break;
2006         }
2007         return;
2008
2009 make_bad:
2010         btrfs_free_path(path);
2011         make_bad_inode(inode);
2012 }
2013
2014 /*
2015  * given a leaf and an inode, copy the inode fields into the leaf
2016  */
2017 static void fill_inode_item(struct btrfs_trans_handle *trans,
2018                             struct extent_buffer *leaf,
2019                             struct btrfs_inode_item *item,
2020                             struct inode *inode)
2021 {
2022         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2023         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2024         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2025         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2026         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2027
2028         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2029                                inode->i_atime.tv_sec);
2030         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2031                                 inode->i_atime.tv_nsec);
2032
2033         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2034                                inode->i_mtime.tv_sec);
2035         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2036                                 inode->i_mtime.tv_nsec);
2037
2038         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2039                                inode->i_ctime.tv_sec);
2040         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2041                                 inode->i_ctime.tv_nsec);
2042
2043         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2044         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2045         btrfs_set_inode_transid(leaf, item, trans->transid);
2046         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2047         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2048         btrfs_set_inode_block_group(leaf, item,
2049                                     BTRFS_I(inode)->block_group->key.objectid);
2050 }
2051
2052 /*
2053  * copy everything in the in-memory inode into the btree.
2054  */
2055 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
2056                               struct btrfs_root *root,
2057                               struct inode *inode)
2058 {
2059         struct btrfs_inode_item *inode_item;
2060         struct btrfs_path *path;
2061         struct extent_buffer *leaf;
2062         int ret;
2063
2064         path = btrfs_alloc_path();
2065         BUG_ON(!path);
2066         ret = btrfs_lookup_inode(trans, root, path,
2067                                  &BTRFS_I(inode)->location, 1);
2068         if (ret) {
2069                 if (ret > 0)
2070                         ret = -ENOENT;
2071                 goto failed;
2072         }
2073
2074         leaf = path->nodes[0];
2075         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2076                                   struct btrfs_inode_item);
2077
2078         fill_inode_item(trans, leaf, inode_item, inode);
2079         btrfs_mark_buffer_dirty(leaf);
2080         btrfs_set_inode_last_trans(trans, inode);
2081         ret = 0;
2082 failed:
2083         btrfs_free_path(path);
2084         return ret;
2085 }
2086
2087
2088 /*
2089  * unlink helper that gets used here in inode.c and in the tree logging
2090  * recovery code.  It remove a link in a directory with a given name, and
2091  * also drops the back refs in the inode to the directory
2092  */
2093 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2094                        struct btrfs_root *root,
2095                        struct inode *dir, struct inode *inode,
2096                        const char *name, int name_len)
2097 {
2098         struct btrfs_path *path;
2099         int ret = 0;
2100         struct extent_buffer *leaf;
2101         struct btrfs_dir_item *di;
2102         struct btrfs_key key;
2103         u64 index;
2104
2105         path = btrfs_alloc_path();
2106         if (!path) {
2107                 ret = -ENOMEM;
2108                 goto err;
2109         }
2110
2111         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2112                                     name, name_len, -1);
2113         if (IS_ERR(di)) {
2114                 ret = PTR_ERR(di);
2115                 goto err;
2116         }
2117         if (!di) {
2118                 ret = -ENOENT;
2119                 goto err;
2120         }
2121         leaf = path->nodes[0];
2122         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2123         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2124         if (ret)
2125                 goto err;
2126         btrfs_release_path(root, path);
2127
2128         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2129                                   inode->i_ino,
2130                                   dir->i_ino, &index);
2131         if (ret) {
2132                 printk("failed to delete reference to %.*s, "
2133                        "inode %lu parent %lu\n", name_len, name,
2134                        inode->i_ino, dir->i_ino);
2135                 goto err;
2136         }
2137
2138         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2139                                          index, name, name_len, -1);
2140         if (IS_ERR(di)) {
2141                 ret = PTR_ERR(di);
2142                 goto err;
2143         }
2144         if (!di) {
2145                 ret = -ENOENT;
2146                 goto err;
2147         }
2148         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2149         btrfs_release_path(root, path);
2150
2151         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2152                                          inode, dir->i_ino);
2153         BUG_ON(ret != 0 && ret != -ENOENT);
2154         if (ret != -ENOENT)
2155                 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2156
2157         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2158                                            dir, index);
2159         BUG_ON(ret);
2160 err:
2161         btrfs_free_path(path);
2162         if (ret)
2163                 goto out;
2164
2165         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2166         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2167         btrfs_update_inode(trans, root, dir);
2168         btrfs_drop_nlink(inode);
2169         ret = btrfs_update_inode(trans, root, inode);
2170         dir->i_sb->s_dirt = 1;
2171 out:
2172         return ret;
2173 }
2174
2175 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2176 {
2177         struct btrfs_root *root;
2178         struct btrfs_trans_handle *trans;
2179         struct inode *inode = dentry->d_inode;
2180         int ret;
2181         unsigned long nr = 0;
2182
2183         root = BTRFS_I(dir)->root;
2184
2185         ret = btrfs_check_free_space(root, 1, 1);
2186         if (ret)
2187                 goto fail;
2188
2189         trans = btrfs_start_transaction(root, 1);
2190
2191         btrfs_set_trans_block_group(trans, dir);
2192         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2193                                  dentry->d_name.name, dentry->d_name.len);
2194
2195         if (inode->i_nlink == 0)
2196                 ret = btrfs_orphan_add(trans, inode);
2197
2198         nr = trans->blocks_used;
2199
2200         btrfs_end_transaction_throttle(trans, root);
2201 fail:
2202         btrfs_btree_balance_dirty(root, nr);
2203         return ret;
2204 }
2205
2206 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2207 {
2208         struct inode *inode = dentry->d_inode;
2209         int err = 0;
2210         int ret;
2211         struct btrfs_root *root = BTRFS_I(dir)->root;
2212         struct btrfs_trans_handle *trans;
2213         unsigned long nr = 0;
2214
2215         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2216                 return -ENOTEMPTY;
2217         }
2218
2219         ret = btrfs_check_free_space(root, 1, 1);
2220         if (ret)
2221                 goto fail;
2222
2223         trans = btrfs_start_transaction(root, 1);
2224         btrfs_set_trans_block_group(trans, dir);
2225
2226         err = btrfs_orphan_add(trans, inode);
2227         if (err)
2228                 goto fail_trans;
2229
2230         /* now the directory is empty */
2231         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2232                                  dentry->d_name.name, dentry->d_name.len);
2233         if (!err) {
2234                 btrfs_i_size_write(inode, 0);
2235         }
2236
2237 fail_trans:
2238         nr = trans->blocks_used;
2239         ret = btrfs_end_transaction_throttle(trans, root);
2240 fail:
2241         btrfs_btree_balance_dirty(root, nr);
2242
2243         if (ret && !err)
2244                 err = ret;
2245         return err;
2246 }
2247
2248 /*
2249  * when truncating bytes in a file, it is possible to avoid reading
2250  * the leaves that contain only checksum items.  This can be the
2251  * majority of the IO required to delete a large file, but it must
2252  * be done carefully.
2253  *
2254  * The keys in the level just above the leaves are checked to make sure
2255  * the lowest key in a given leaf is a csum key, and starts at an offset
2256  * after the new  size.
2257  *
2258  * Then the key for the next leaf is checked to make sure it also has
2259  * a checksum item for the same file.  If it does, we know our target leaf
2260  * contains only checksum items, and it can be safely freed without reading
2261  * it.
2262  *
2263  * This is just an optimization targeted at large files.  It may do
2264  * nothing.  It will return 0 unless things went badly.
2265  */
2266 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2267                                      struct btrfs_root *root,
2268                                      struct btrfs_path *path,
2269                                      struct inode *inode, u64 new_size)
2270 {
2271         struct btrfs_key key;
2272         int ret;
2273         int nritems;
2274         struct btrfs_key found_key;
2275         struct btrfs_key other_key;
2276         struct btrfs_leaf_ref *ref;
2277         u64 leaf_gen;
2278         u64 leaf_start;
2279
2280         path->lowest_level = 1;
2281         key.objectid = inode->i_ino;
2282         key.type = BTRFS_CSUM_ITEM_KEY;
2283         key.offset = new_size;
2284 again:
2285         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2286         if (ret < 0)
2287                 goto out;
2288
2289         if (path->nodes[1] == NULL) {
2290                 ret = 0;
2291                 goto out;
2292         }
2293         ret = 0;
2294         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2295         nritems = btrfs_header_nritems(path->nodes[1]);
2296
2297         if (!nritems)
2298                 goto out;
2299
2300         if (path->slots[1] >= nritems)
2301                 goto next_node;
2302
2303         /* did we find a key greater than anything we want to delete? */
2304         if (found_key.objectid > inode->i_ino ||
2305            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2306                 goto out;
2307
2308         /* we check the next key in the node to make sure the leave contains
2309          * only checksum items.  This comparison doesn't work if our
2310          * leaf is the last one in the node
2311          */
2312         if (path->slots[1] + 1 >= nritems) {
2313 next_node:
2314                 /* search forward from the last key in the node, this
2315                  * will bring us into the next node in the tree
2316                  */
2317                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2318
2319                 /* unlikely, but we inc below, so check to be safe */
2320                 if (found_key.offset == (u64)-1)
2321                         goto out;
2322
2323                 /* search_forward needs a path with locks held, do the
2324                  * search again for the original key.  It is possible
2325                  * this will race with a balance and return a path that
2326                  * we could modify, but this drop is just an optimization
2327                  * and is allowed to miss some leaves.
2328                  */
2329                 btrfs_release_path(root, path);
2330                 found_key.offset++;
2331
2332                 /* setup a max key for search_forward */
2333                 other_key.offset = (u64)-1;
2334                 other_key.type = key.type;
2335                 other_key.objectid = key.objectid;
2336
2337                 path->keep_locks = 1;
2338                 ret = btrfs_search_forward(root, &found_key, &other_key,
2339                                            path, 0, 0);
2340                 path->keep_locks = 0;
2341                 if (ret || found_key.objectid != key.objectid ||
2342                     found_key.type != key.type) {
2343                         ret = 0;
2344                         goto out;
2345                 }
2346
2347                 key.offset = found_key.offset;
2348                 btrfs_release_path(root, path);
2349                 cond_resched();
2350                 goto again;
2351         }
2352
2353         /* we know there's one more slot after us in the tree,
2354          * read that key so we can verify it is also a checksum item
2355          */
2356         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2357
2358         if (found_key.objectid < inode->i_ino)
2359                 goto next_key;
2360
2361         if (found_key.type != key.type || found_key.offset < new_size)
2362                 goto next_key;
2363
2364         /*
2365          * if the key for the next leaf isn't a csum key from this objectid,
2366          * we can't be sure there aren't good items inside this leaf.
2367          * Bail out
2368          */
2369         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2370                 goto out;
2371
2372         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2373         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2374         /*
2375          * it is safe to delete this leaf, it contains only
2376          * csum items from this inode at an offset >= new_size
2377          */
2378         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2379         BUG_ON(ret);
2380
2381         if (root->ref_cows && leaf_gen < trans->transid) {
2382                 ref = btrfs_alloc_leaf_ref(root, 0);
2383                 if (ref) {
2384                         ref->root_gen = root->root_key.offset;
2385                         ref->bytenr = leaf_start;
2386                         ref->owner = 0;
2387                         ref->generation = leaf_gen;
2388                         ref->nritems = 0;
2389
2390                         ret = btrfs_add_leaf_ref(root, ref, 0);
2391                         WARN_ON(ret);
2392                         btrfs_free_leaf_ref(root, ref);
2393                 } else {
2394                         WARN_ON(1);
2395                 }
2396         }
2397 next_key:
2398         btrfs_release_path(root, path);
2399
2400         if (other_key.objectid == inode->i_ino &&
2401             other_key.type == key.type && other_key.offset > key.offset) {
2402                 key.offset = other_key.offset;
2403                 cond_resched();
2404                 goto again;
2405         }
2406         ret = 0;
2407 out:
2408         /* fixup any changes we've made to the path */
2409         path->lowest_level = 0;
2410         path->keep_locks = 0;
2411         btrfs_release_path(root, path);
2412         return ret;
2413 }
2414
2415 /*
2416  * this can truncate away extent items, csum items and directory items.
2417  * It starts at a high offset and removes keys until it can't find
2418  * any higher than new_size
2419  *
2420  * csum items that cross the new i_size are truncated to the new size
2421  * as well.
2422  *
2423  * min_type is the minimum key type to truncate down to.  If set to 0, this
2424  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2425  */
2426 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2427                                         struct btrfs_root *root,
2428                                         struct inode *inode,
2429                                         u64 new_size, u32 min_type)
2430 {
2431         int ret;
2432         struct btrfs_path *path;
2433         struct btrfs_key key;
2434         struct btrfs_key found_key;
2435         u32 found_type;
2436         struct extent_buffer *leaf;
2437         struct btrfs_file_extent_item *fi;
2438         u64 extent_start = 0;
2439         u64 extent_num_bytes = 0;
2440         u64 item_end = 0;
2441         u64 root_gen = 0;
2442         u64 root_owner = 0;
2443         int found_extent;
2444         int del_item;
2445         int pending_del_nr = 0;
2446         int pending_del_slot = 0;
2447         int extent_type = -1;
2448         int encoding;
2449         u64 mask = root->sectorsize - 1;
2450
2451         if (root->ref_cows)
2452                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2453         path = btrfs_alloc_path();
2454         path->reada = -1;
2455         BUG_ON(!path);
2456
2457         /* FIXME, add redo link to tree so we don't leak on crash */
2458         key.objectid = inode->i_ino;
2459         key.offset = (u64)-1;
2460         key.type = (u8)-1;
2461
2462         btrfs_init_path(path);
2463
2464         ret = drop_csum_leaves(trans, root, path, inode, new_size);
2465         BUG_ON(ret);
2466
2467 search_again:
2468         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2469         if (ret < 0) {
2470                 goto error;
2471         }
2472         if (ret > 0) {
2473                 /* there are no items in the tree for us to truncate, we're
2474                  * done
2475                  */
2476                 if (path->slots[0] == 0) {
2477                         ret = 0;
2478                         goto error;
2479                 }
2480                 path->slots[0]--;
2481         }
2482
2483         while(1) {
2484                 fi = NULL;
2485                 leaf = path->nodes[0];
2486                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2487                 found_type = btrfs_key_type(&found_key);
2488                 encoding = 0;
2489
2490                 if (found_key.objectid != inode->i_ino)
2491                         break;
2492
2493                 if (found_type < min_type)
2494                         break;
2495
2496                 item_end = found_key.offset;
2497                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2498                         fi = btrfs_item_ptr(leaf, path->slots[0],
2499                                             struct btrfs_file_extent_item);
2500                         extent_type = btrfs_file_extent_type(leaf, fi);
2501                         encoding = btrfs_file_extent_compression(leaf, fi);
2502                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2503                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2504
2505                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2506                                 item_end +=
2507                                     btrfs_file_extent_num_bytes(leaf, fi);
2508                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2509                                 item_end += btrfs_file_extent_inline_len(leaf,
2510                                                                          fi);
2511                         }
2512                         item_end--;
2513                 }
2514                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
2515                         ret = btrfs_csum_truncate(trans, root, path,
2516                                                   new_size);
2517                         BUG_ON(ret);
2518                 }
2519                 if (item_end < new_size) {
2520                         if (found_type == BTRFS_DIR_ITEM_KEY) {
2521                                 found_type = BTRFS_INODE_ITEM_KEY;
2522                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2523                                 found_type = BTRFS_CSUM_ITEM_KEY;
2524                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2525                                 found_type = BTRFS_XATTR_ITEM_KEY;
2526                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2527                                 found_type = BTRFS_INODE_REF_KEY;
2528                         } else if (found_type) {
2529                                 found_type--;
2530                         } else {
2531                                 break;
2532                         }
2533                         btrfs_set_key_type(&key, found_type);
2534                         goto next;
2535                 }
2536                 if (found_key.offset >= new_size)
2537                         del_item = 1;
2538                 else
2539                         del_item = 0;
2540                 found_extent = 0;
2541
2542                 /* FIXME, shrink the extent if the ref count is only 1 */
2543                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2544                         goto delete;
2545
2546                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2547                         u64 num_dec;
2548                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2549                         if (!del_item && !encoding) {
2550                                 u64 orig_num_bytes =
2551                                         btrfs_file_extent_num_bytes(leaf, fi);
2552                                 extent_num_bytes = new_size -
2553                                         found_key.offset + root->sectorsize - 1;
2554                                 extent_num_bytes = extent_num_bytes &
2555                                         ~((u64)root->sectorsize - 1);
2556                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2557                                                          extent_num_bytes);
2558                                 num_dec = (orig_num_bytes -
2559                                            extent_num_bytes);
2560                                 if (root->ref_cows && extent_start != 0)
2561                                         inode_sub_bytes(inode, num_dec);
2562                                 btrfs_mark_buffer_dirty(leaf);
2563                         } else {
2564                                 extent_num_bytes =
2565                                         btrfs_file_extent_disk_num_bytes(leaf,
2566                                                                          fi);
2567                                 /* FIXME blocksize != 4096 */
2568                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2569                                 if (extent_start != 0) {
2570                                         found_extent = 1;
2571                                         if (root->ref_cows)
2572                                                 inode_sub_bytes(inode, num_dec);
2573                                 }
2574                                 root_gen = btrfs_header_generation(leaf);
2575                                 root_owner = btrfs_header_owner(leaf);
2576                         }
2577                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2578                         /*
2579                          * we can't truncate inline items that have had
2580                          * special encodings
2581                          */
2582                         if (!del_item &&
2583                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2584                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2585                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2586                                 u32 size = new_size - found_key.offset;
2587
2588                                 if (root->ref_cows) {
2589                                         inode_sub_bytes(inode, item_end + 1 -
2590                                                         new_size);
2591                                 }
2592                                 size =
2593                                     btrfs_file_extent_calc_inline_size(size);
2594                                 ret = btrfs_truncate_item(trans, root, path,
2595                                                           size, 1);
2596                                 BUG_ON(ret);
2597                         } else if (root->ref_cows) {
2598                                 inode_sub_bytes(inode, item_end + 1 -
2599                                                 found_key.offset);
2600                         }
2601                 }
2602 delete:
2603                 if (del_item) {
2604                         if (!pending_del_nr) {
2605                                 /* no pending yet, add ourselves */
2606                                 pending_del_slot = path->slots[0];
2607                                 pending_del_nr = 1;
2608                         } else if (pending_del_nr &&
2609                                    path->slots[0] + 1 == pending_del_slot) {
2610                                 /* hop on the pending chunk */
2611                                 pending_del_nr++;
2612                                 pending_del_slot = path->slots[0];
2613                         } else {
2614                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2615                         }
2616                 } else {
2617                         break;
2618                 }
2619                 if (found_extent) {
2620                         ret = btrfs_free_extent(trans, root, extent_start,
2621                                                 extent_num_bytes,
2622                                                 leaf->start, root_owner,
2623                                                 root_gen, inode->i_ino, 0);
2624                         BUG_ON(ret);
2625                 }
2626 next:
2627                 if (path->slots[0] == 0) {
2628                         if (pending_del_nr)
2629                                 goto del_pending;
2630                         btrfs_release_path(root, path);
2631                         goto search_again;
2632                 }
2633
2634                 path->slots[0]--;
2635                 if (pending_del_nr &&
2636                     path->slots[0] + 1 != pending_del_slot) {
2637                         struct btrfs_key debug;
2638 del_pending:
2639                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2640                                               pending_del_slot);
2641                         ret = btrfs_del_items(trans, root, path,
2642                                               pending_del_slot,
2643                                               pending_del_nr);
2644                         BUG_ON(ret);
2645                         pending_del_nr = 0;
2646                         btrfs_release_path(root, path);
2647                         goto search_again;
2648                 }
2649         }
2650         ret = 0;
2651 error:
2652         if (pending_del_nr) {
2653                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2654                                       pending_del_nr);
2655         }
2656         btrfs_free_path(path);
2657         inode->i_sb->s_dirt = 1;
2658         return ret;
2659 }
2660
2661 /*
2662  * taken from block_truncate_page, but does cow as it zeros out
2663  * any bytes left in the last page in the file.
2664  */
2665 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2666 {
2667         struct inode *inode = mapping->host;
2668         struct btrfs_root *root = BTRFS_I(inode)->root;
2669         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2670         struct btrfs_ordered_extent *ordered;
2671         char *kaddr;
2672         u32 blocksize = root->sectorsize;
2673         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2674         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2675         struct page *page;
2676         int ret = 0;
2677         u64 page_start;
2678         u64 page_end;
2679
2680         if ((offset & (blocksize - 1)) == 0)
2681                 goto out;
2682
2683         ret = -ENOMEM;
2684 again:
2685         page = grab_cache_page(mapping, index);
2686         if (!page)
2687                 goto out;
2688
2689         page_start = page_offset(page);
2690         page_end = page_start + PAGE_CACHE_SIZE - 1;
2691
2692         if (!PageUptodate(page)) {
2693                 ret = btrfs_readpage(NULL, page);
2694                 lock_page(page);
2695                 if (page->mapping != mapping) {
2696                         unlock_page(page);
2697                         page_cache_release(page);
2698                         goto again;
2699                 }
2700                 if (!PageUptodate(page)) {
2701                         ret = -EIO;
2702                         goto out_unlock;
2703                 }
2704         }
2705         wait_on_page_writeback(page);
2706
2707         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2708         set_page_extent_mapped(page);
2709
2710         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2711         if (ordered) {
2712                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2713                 unlock_page(page);
2714                 page_cache_release(page);
2715                 btrfs_start_ordered_extent(inode, ordered, 1);
2716                 btrfs_put_ordered_extent(ordered);
2717                 goto again;
2718         }
2719
2720         btrfs_set_extent_delalloc(inode, page_start, page_end);
2721         ret = 0;
2722         if (offset != PAGE_CACHE_SIZE) {
2723                 kaddr = kmap(page);
2724                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2725                 flush_dcache_page(page);
2726                 kunmap(page);
2727         }
2728         ClearPageChecked(page);
2729         set_page_dirty(page);
2730         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2731
2732 out_unlock:
2733         unlock_page(page);
2734         page_cache_release(page);
2735 out:
2736         return ret;
2737 }
2738
2739 int btrfs_cont_expand(struct inode *inode, loff_t size)
2740 {
2741         struct btrfs_trans_handle *trans;
2742         struct btrfs_root *root = BTRFS_I(inode)->root;
2743         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2744         struct extent_map *em;
2745         u64 mask = root->sectorsize - 1;
2746         u64 hole_start = (inode->i_size + mask) & ~mask;
2747         u64 block_end = (size + mask) & ~mask;
2748         u64 last_byte;
2749         u64 cur_offset;
2750         u64 hole_size;
2751         int err;
2752
2753         if (size <= hole_start)
2754                 return 0;
2755
2756         err = btrfs_check_free_space(root, 1, 0);
2757         if (err)
2758                 return err;
2759
2760         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2761
2762         while (1) {
2763                 struct btrfs_ordered_extent *ordered;
2764                 btrfs_wait_ordered_range(inode, hole_start,
2765                                          block_end - hole_start);
2766                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2767                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2768                 if (!ordered)
2769                         break;
2770                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2771                 btrfs_put_ordered_extent(ordered);
2772         }
2773
2774         trans = btrfs_start_transaction(root, 1);
2775         btrfs_set_trans_block_group(trans, inode);
2776
2777         cur_offset = hole_start;
2778         while (1) {
2779                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2780                                 block_end - cur_offset, 0);
2781                 BUG_ON(IS_ERR(em) || !em);
2782                 last_byte = min(extent_map_end(em), block_end);
2783                 last_byte = (last_byte + mask) & ~mask;
2784                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2785                         u64 hint_byte = 0;
2786                         hole_size = last_byte - cur_offset;
2787                         err = btrfs_drop_extents(trans, root, inode,
2788                                                  cur_offset,
2789                                                  cur_offset + hole_size,
2790                                                  cur_offset, &hint_byte);
2791                         if (err)
2792                                 break;
2793                         err = btrfs_insert_file_extent(trans, root,
2794                                         inode->i_ino, cur_offset, 0,
2795                                         0, hole_size, 0, hole_size,
2796                                         0, 0, 0);
2797                         btrfs_drop_extent_cache(inode, hole_start,
2798                                         last_byte - 1, 0);
2799                 }
2800                 free_extent_map(em);
2801                 cur_offset = last_byte;
2802                 if (err || cur_offset >= block_end)
2803                         break;
2804         }
2805
2806         btrfs_end_transaction(trans, root);
2807         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2808         return err;
2809 }
2810
2811 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2812 {
2813         struct inode *inode = dentry->d_inode;
2814         int err;
2815
2816         err = inode_change_ok(inode, attr);
2817         if (err)
2818                 return err;
2819
2820         if (S_ISREG(inode->i_mode) &&
2821             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2822                 err = btrfs_cont_expand(inode, attr->ia_size);
2823                 if (err)
2824                         return err;
2825         }
2826
2827         err = inode_setattr(inode, attr);
2828
2829         if (!err && ((attr->ia_valid & ATTR_MODE)))
2830                 err = btrfs_acl_chmod(inode);
2831         return err;
2832 }
2833
2834 void btrfs_delete_inode(struct inode *inode)
2835 {
2836         struct btrfs_trans_handle *trans;
2837         struct btrfs_root *root = BTRFS_I(inode)->root;
2838         unsigned long nr;
2839         int ret;
2840
2841         truncate_inode_pages(&inode->i_data, 0);
2842         if (is_bad_inode(inode)) {
2843                 btrfs_orphan_del(NULL, inode);
2844                 goto no_delete;
2845         }
2846         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2847
2848         btrfs_i_size_write(inode, 0);
2849         trans = btrfs_start_transaction(root, 1);
2850
2851         btrfs_set_trans_block_group(trans, inode);
2852         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2853         if (ret) {
2854                 btrfs_orphan_del(NULL, inode);
2855                 goto no_delete_lock;
2856         }
2857
2858         btrfs_orphan_del(trans, inode);
2859
2860         nr = trans->blocks_used;
2861         clear_inode(inode);
2862
2863         btrfs_end_transaction(trans, root);
2864         btrfs_btree_balance_dirty(root, nr);
2865         return;
2866
2867 no_delete_lock:
2868         nr = trans->blocks_used;
2869         btrfs_end_transaction(trans, root);
2870         btrfs_btree_balance_dirty(root, nr);
2871 no_delete:
2872         clear_inode(inode);
2873 }
2874
2875 /*
2876  * this returns the key found in the dir entry in the location pointer.
2877  * If no dir entries were found, location->objectid is 0.
2878  */
2879 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2880                                struct btrfs_key *location)
2881 {
2882         const char *name = dentry->d_name.name;
2883         int namelen = dentry->d_name.len;
2884         struct btrfs_dir_item *di;
2885         struct btrfs_path *path;
2886         struct btrfs_root *root = BTRFS_I(dir)->root;
2887         int ret = 0;
2888
2889         path = btrfs_alloc_path();
2890         BUG_ON(!path);
2891
2892         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2893                                     namelen, 0);
2894         if (IS_ERR(di))
2895                 ret = PTR_ERR(di);
2896         if (!di || IS_ERR(di)) {
2897                 goto out_err;
2898         }
2899         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2900 out:
2901         btrfs_free_path(path);
2902         return ret;
2903 out_err:
2904         location->objectid = 0;
2905         goto out;
2906 }
2907
2908 /*
2909  * when we hit a tree root in a directory, the btrfs part of the inode
2910  * needs to be changed to reflect the root directory of the tree root.  This
2911  * is kind of like crossing a mount point.
2912  */
2913 static int fixup_tree_root_location(struct btrfs_root *root,
2914                              struct btrfs_key *location,
2915                              struct btrfs_root **sub_root,
2916                              struct dentry *dentry)
2917 {
2918         struct btrfs_root_item *ri;
2919
2920         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2921                 return 0;
2922         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2923                 return 0;
2924
2925         *sub_root = btrfs_read_fs_root(root->fs_info, location,
2926                                         dentry->d_name.name,
2927                                         dentry->d_name.len);
2928         if (IS_ERR(*sub_root))
2929                 return PTR_ERR(*sub_root);
2930
2931         ri = &(*sub_root)->root_item;
2932         location->objectid = btrfs_root_dirid(ri);
2933         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2934         location->offset = 0;
2935
2936         return 0;
2937 }
2938
2939 static noinline void init_btrfs_i(struct inode *inode)
2940 {
2941         struct btrfs_inode *bi = BTRFS_I(inode);
2942
2943         bi->i_acl = NULL;
2944         bi->i_default_acl = NULL;
2945
2946         bi->generation = 0;
2947         bi->last_trans = 0;
2948         bi->logged_trans = 0;
2949         bi->delalloc_bytes = 0;
2950         bi->disk_i_size = 0;
2951         bi->flags = 0;
2952         bi->index_cnt = (u64)-1;
2953         bi->log_dirty_trans = 0;
2954         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2955         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2956                              inode->i_mapping, GFP_NOFS);
2957         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2958                              inode->i_mapping, GFP_NOFS);
2959         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2960         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2961         mutex_init(&BTRFS_I(inode)->csum_mutex);
2962         mutex_init(&BTRFS_I(inode)->extent_mutex);
2963         mutex_init(&BTRFS_I(inode)->log_mutex);
2964 }
2965
2966 static int btrfs_init_locked_inode(struct inode *inode, void *p)
2967 {
2968         struct btrfs_iget_args *args = p;
2969         inode->i_ino = args->ino;
2970         init_btrfs_i(inode);
2971         BTRFS_I(inode)->root = args->root;
2972         return 0;
2973 }
2974
2975 static int btrfs_find_actor(struct inode *inode, void *opaque)
2976 {
2977         struct btrfs_iget_args *args = opaque;
2978         return (args->ino == inode->i_ino &&
2979                 args->root == BTRFS_I(inode)->root);
2980 }
2981
2982 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2983                             struct btrfs_root *root, int wait)
2984 {
2985         struct inode *inode;
2986         struct btrfs_iget_args args;
2987         args.ino = objectid;
2988         args.root = root;
2989
2990         if (wait) {
2991                 inode = ilookup5(s, objectid, btrfs_find_actor,
2992                                  (void *)&args);
2993         } else {
2994                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
2995                                         (void *)&args);
2996         }
2997         return inode;
2998 }
2999
3000 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3001                                 struct btrfs_root *root)
3002 {
3003         struct inode *inode;
3004         struct btrfs_iget_args args;
3005         args.ino = objectid;
3006         args.root = root;
3007
3008         inode = iget5_locked(s, objectid, btrfs_find_actor,
3009                              btrfs_init_locked_inode,
3010                              (void *)&args);
3011         return inode;
3012 }
3013
3014 /* Get an inode object given its location and corresponding root.
3015  * Returns in *is_new if the inode was read from disk
3016  */
3017 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3018                          struct btrfs_root *root, int *is_new)
3019 {
3020         struct inode *inode;
3021
3022         inode = btrfs_iget_locked(s, location->objectid, root);
3023         if (!inode)
3024                 return ERR_PTR(-EACCES);
3025
3026         if (inode->i_state & I_NEW) {
3027                 BTRFS_I(inode)->root = root;
3028                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3029                 btrfs_read_locked_inode(inode);
3030                 unlock_new_inode(inode);
3031                 if (is_new)
3032                         *is_new = 1;
3033         } else {
3034                 if (is_new)
3035                         *is_new = 0;
3036         }
3037
3038         return inode;
3039 }
3040
3041 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3042                                    struct nameidata *nd)
3043 {
3044         struct inode * inode;
3045         struct btrfs_inode *bi = BTRFS_I(dir);
3046         struct btrfs_root *root = bi->root;
3047         struct btrfs_root *sub_root = root;
3048         struct btrfs_key location;
3049         int ret, new;
3050
3051         if (dentry->d_name.len > BTRFS_NAME_LEN)
3052                 return ERR_PTR(-ENAMETOOLONG);
3053
3054         ret = btrfs_inode_by_name(dir, dentry, &location);
3055
3056         if (ret < 0)
3057                 return ERR_PTR(ret);
3058
3059         inode = NULL;
3060         if (location.objectid) {
3061                 ret = fixup_tree_root_location(root, &location, &sub_root,
3062                                                 dentry);
3063                 if (ret < 0)
3064                         return ERR_PTR(ret);
3065                 if (ret > 0)
3066                         return ERR_PTR(-ENOENT);
3067                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3068                 if (IS_ERR(inode))
3069                         return ERR_CAST(inode);
3070
3071                 /* the inode and parent dir are two different roots */
3072                 if (new && root != sub_root) {
3073                         igrab(inode);
3074                         sub_root->inode = inode;
3075                 }
3076         }
3077
3078         return d_splice_alias(inode, dentry);
3079 }
3080
3081 static unsigned char btrfs_filetype_table[] = {
3082         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3083 };
3084
3085 static int btrfs_real_readdir(struct file *filp, void *dirent,
3086                               filldir_t filldir)
3087 {
3088         struct inode *inode = filp->f_dentry->d_inode;
3089         struct btrfs_root *root = BTRFS_I(inode)->root;
3090         struct btrfs_item *item;
3091         struct btrfs_dir_item *di;
3092         struct btrfs_key key;
3093         struct btrfs_key found_key;
3094         struct btrfs_path *path;
3095         int ret;
3096         u32 nritems;
3097         struct extent_buffer *leaf;
3098         int slot;
3099         int advance;
3100         unsigned char d_type;
3101         int over = 0;
3102         u32 di_cur;
3103         u32 di_total;
3104         u32 di_len;
3105         int key_type = BTRFS_DIR_INDEX_KEY;
3106         char tmp_name[32];
3107         char *name_ptr;
3108         int name_len;
3109
3110         /* FIXME, use a real flag for deciding about the key type */
3111         if (root->fs_info->tree_root == root)
3112                 key_type = BTRFS_DIR_ITEM_KEY;
3113
3114         /* special case for "." */
3115         if (filp->f_pos == 0) {
3116                 over = filldir(dirent, ".", 1,
3117                                1, inode->i_ino,
3118                                DT_DIR);
3119                 if (over)
3120                         return 0;
3121                 filp->f_pos = 1;
3122         }
3123         /* special case for .., just use the back ref */
3124         if (filp->f_pos == 1) {
3125                 u64 pino = parent_ino(filp->f_path.dentry);
3126                 over = filldir(dirent, "..", 2,
3127                                2, pino, DT_DIR);
3128                 if (over)
3129                         return 0;
3130                 filp->f_pos = 2;
3131         }
3132
3133         path = btrfs_alloc_path();
3134         path->reada = 2;
3135
3136         btrfs_set_key_type(&key, key_type);
3137         key.offset = filp->f_pos;
3138         key.objectid = inode->i_ino;
3139
3140         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3141         if (ret < 0)
3142                 goto err;
3143         advance = 0;
3144
3145         while (1) {
3146                 leaf = path->nodes[0];
3147                 nritems = btrfs_header_nritems(leaf);
3148                 slot = path->slots[0];
3149                 if (advance || slot >= nritems) {
3150                         if (slot >= nritems - 1) {
3151                                 ret = btrfs_next_leaf(root, path);
3152                                 if (ret)
3153                                         break;
3154                                 leaf = path->nodes[0];
3155                                 nritems = btrfs_header_nritems(leaf);
3156                                 slot = path->slots[0];
3157                         } else {
3158                                 slot++;
3159                                 path->slots[0]++;
3160                         }
3161                 }
3162                 advance = 1;
3163                 item = btrfs_item_nr(leaf, slot);
3164                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3165
3166                 if (found_key.objectid != key.objectid)
3167                         break;
3168                 if (btrfs_key_type(&found_key) != key_type)
3169                         break;
3170                 if (found_key.offset < filp->f_pos)
3171                         continue;
3172
3173                 filp->f_pos = found_key.offset;
3174
3175                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3176                 di_cur = 0;
3177                 di_total = btrfs_item_size(leaf, item);
3178
3179                 while (di_cur < di_total) {
3180                         struct btrfs_key location;
3181
3182                         name_len = btrfs_dir_name_len(leaf, di);
3183                         if (name_len <= sizeof(tmp_name)) {
3184                                 name_ptr = tmp_name;
3185                         } else {
3186                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3187                                 if (!name_ptr) {
3188                                         ret = -ENOMEM;
3189                                         goto err;
3190                                 }
3191                         }
3192                         read_extent_buffer(leaf, name_ptr,
3193                                            (unsigned long)(di + 1), name_len);
3194
3195                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3196                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3197                         over = filldir(dirent, name_ptr, name_len,
3198                                        found_key.offset, location.objectid,
3199                                        d_type);
3200
3201                         if (name_ptr != tmp_name)
3202                                 kfree(name_ptr);
3203
3204                         if (over)
3205                                 goto nopos;
3206
3207                         di_len = btrfs_dir_name_len(leaf, di) +
3208                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3209                         di_cur += di_len;
3210                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3211                 }
3212         }
3213
3214         /* Reached end of directory/root. Bump pos past the last item. */
3215         if (key_type == BTRFS_DIR_INDEX_KEY)
3216                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3217         else
3218                 filp->f_pos++;
3219 nopos:
3220         ret = 0;
3221 err:
3222         btrfs_free_path(path);
3223         return ret;
3224 }
3225
3226 int btrfs_write_inode(struct inode *inode, int wait)
3227 {
3228         struct btrfs_root *root = BTRFS_I(inode)->root;
3229         struct btrfs_trans_handle *trans;
3230         int ret = 0;
3231
3232         if (root->fs_info->btree_inode == inode)
3233                 return 0;
3234
3235         if (wait) {
3236                 trans = btrfs_join_transaction(root, 1);
3237                 btrfs_set_trans_block_group(trans, inode);
3238                 ret = btrfs_commit_transaction(trans, root);
3239         }
3240         return ret;
3241 }
3242
3243 /*
3244  * This is somewhat expensive, updating the tree every time the
3245  * inode changes.  But, it is most likely to find the inode in cache.
3246  * FIXME, needs more benchmarking...there are no reasons other than performance
3247  * to keep or drop this code.
3248  */
3249 void btrfs_dirty_inode(struct inode *inode)
3250 {
3251         struct btrfs_root *root = BTRFS_I(inode)->root;
3252         struct btrfs_trans_handle *trans;
3253
3254         trans = btrfs_join_transaction(root, 1);
3255         btrfs_set_trans_block_group(trans, inode);
3256         btrfs_update_inode(trans, root, inode);
3257         btrfs_end_transaction(trans, root);
3258 }
3259
3260 /*
3261  * find the highest existing sequence number in a directory
3262  * and then set the in-memory index_cnt variable to reflect
3263  * free sequence numbers
3264  */
3265 static int btrfs_set_inode_index_count(struct inode *inode)
3266 {
3267         struct btrfs_root *root = BTRFS_I(inode)->root;
3268         struct btrfs_key key, found_key;
3269         struct btrfs_path *path;
3270         struct extent_buffer *leaf;
3271         int ret;
3272
3273         key.objectid = inode->i_ino;
3274         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3275         key.offset = (u64)-1;
3276
3277         path = btrfs_alloc_path();
3278         if (!path)
3279                 return -ENOMEM;
3280
3281         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3282         if (ret < 0)
3283                 goto out;
3284         /* FIXME: we should be able to handle this */
3285         if (ret == 0)
3286                 goto out;
3287         ret = 0;
3288
3289         /*
3290          * MAGIC NUMBER EXPLANATION:
3291          * since we search a directory based on f_pos we have to start at 2
3292          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3293          * else has to start at 2
3294          */
3295         if (path->slots[0] == 0) {
3296                 BTRFS_I(inode)->index_cnt = 2;
3297                 goto out;
3298         }
3299
3300         path->slots[0]--;
3301
3302         leaf = path->nodes[0];
3303         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3304
3305         if (found_key.objectid != inode->i_ino ||
3306             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3307                 BTRFS_I(inode)->index_cnt = 2;
3308                 goto out;
3309         }
3310
3311         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3312 out:
3313         btrfs_free_path(path);
3314         return ret;
3315 }
3316
3317 /*
3318  * helper to find a free sequence number in a given directory.  This current
3319  * code is very simple, later versions will do smarter things in the btree
3320  */
3321 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
3322                                  u64 *index)
3323 {
3324         int ret = 0;
3325
3326         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3327                 ret = btrfs_set_inode_index_count(dir);
3328                 if (ret) {
3329                         return ret;
3330                 }
3331         }
3332
3333         *index = BTRFS_I(dir)->index_cnt;
3334         BTRFS_I(dir)->index_cnt++;
3335
3336         return ret;
3337 }
3338
3339 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3340                                      struct btrfs_root *root,
3341                                      struct inode *dir,
3342                                      const char *name, int name_len,
3343                                      u64 ref_objectid,
3344                                      u64 objectid,
3345                                      struct btrfs_block_group_cache *group,
3346                                      int mode, u64 *index)
3347 {
3348         struct inode *inode;
3349         struct btrfs_inode_item *inode_item;
3350         struct btrfs_block_group_cache *new_inode_group;
3351         struct btrfs_key *location;
3352         struct btrfs_path *path;
3353         struct btrfs_inode_ref *ref;
3354         struct btrfs_key key[2];
3355         u32 sizes[2];
3356         unsigned long ptr;
3357         int ret;
3358         int owner;
3359
3360         path = btrfs_alloc_path();
3361         BUG_ON(!path);
3362
3363         inode = new_inode(root->fs_info->sb);
3364         if (!inode)
3365                 return ERR_PTR(-ENOMEM);
3366
3367         if (dir) {
3368                 ret = btrfs_set_inode_index(dir, inode, index);
3369                 if (ret)
3370                         return ERR_PTR(ret);
3371         }
3372         /*
3373          * index_cnt is ignored for everything but a dir,
3374          * btrfs_get_inode_index_count has an explanation for the magic
3375          * number
3376          */
3377         init_btrfs_i(inode);
3378         BTRFS_I(inode)->index_cnt = 2;
3379         BTRFS_I(inode)->root = root;
3380         BTRFS_I(inode)->generation = trans->transid;
3381
3382         if (mode & S_IFDIR)
3383                 owner = 0;
3384         else
3385                 owner = 1;
3386         new_inode_group = btrfs_find_block_group(root, group, 0,
3387                                        BTRFS_BLOCK_GROUP_METADATA, owner);
3388         if (!new_inode_group) {
3389                 printk("find_block group failed\n");
3390                 new_inode_group = group;
3391         }
3392         BTRFS_I(inode)->block_group = new_inode_group;
3393
3394         key[0].objectid = objectid;
3395         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3396         key[0].offset = 0;
3397
3398         key[1].objectid = objectid;
3399         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3400         key[1].offset = ref_objectid;
3401
3402         sizes[0] = sizeof(struct btrfs_inode_item);
3403         sizes[1] = name_len + sizeof(*ref);
3404
3405         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3406         if (ret != 0)
3407                 goto fail;
3408
3409         if (objectid > root->highest_inode)
3410                 root->highest_inode = objectid;
3411
3412         inode->i_uid = current->fsuid;
3413         inode->i_gid = current->fsgid;
3414         inode->i_mode = mode;
3415         inode->i_ino = objectid;
3416         inode_set_bytes(inode, 0);
3417         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3418         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3419                                   struct btrfs_inode_item);
3420         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3421
3422         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3423                              struct btrfs_inode_ref);
3424         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3425         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3426         ptr = (unsigned long)(ref + 1);
3427         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3428
3429         btrfs_mark_buffer_dirty(path->nodes[0]);
3430         btrfs_free_path(path);
3431
3432         location = &BTRFS_I(inode)->location;
3433         location->objectid = objectid;
3434         location->offset = 0;
3435         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3436
3437         insert_inode_hash(inode);
3438         return inode;
3439 fail:
3440         if (dir)
3441                 BTRFS_I(dir)->index_cnt--;
3442         btrfs_free_path(path);
3443         return ERR_PTR(ret);
3444 }
3445
3446 static inline u8 btrfs_inode_type(struct inode *inode)
3447 {
3448         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3449 }
3450
3451 /*
3452  * utility function to add 'inode' into 'parent_inode' with
3453  * a give name and a given sequence number.
3454  * if 'add_backref' is true, also insert a backref from the
3455  * inode to the parent directory.
3456  */
3457 int btrfs_add_link(struct btrfs_trans_handle *trans,
3458                    struct inode *parent_inode, struct inode *inode,
3459                    const char *name, int name_len, int add_backref, u64 index)
3460 {
3461         int ret;
3462         struct btrfs_key key;
3463         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3464
3465         key.objectid = inode->i_ino;
3466         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3467         key.offset = 0;
3468
3469         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3470                                     parent_inode->i_ino,
3471                                     &key, btrfs_inode_type(inode),
3472                                     index);
3473         if (ret == 0) {
3474                 if (add_backref) {
3475                         ret = btrfs_insert_inode_ref(trans, root,
3476                                                      name, name_len,
3477                                                      inode->i_ino,
3478                                                      parent_inode->i_ino,
3479                                                      index);
3480                 }
3481                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3482                                    name_len * 2);
3483                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3484                 ret = btrfs_update_inode(trans, root, parent_inode);
3485         }
3486         return ret;
3487 }
3488
3489 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3490                             struct dentry *dentry, struct inode *inode,
3491                             int backref, u64 index)
3492 {
3493         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3494                                  inode, dentry->d_name.name,
3495                                  dentry->d_name.len, backref, index);
3496         if (!err) {
3497                 d_instantiate(dentry, inode);
3498                 return 0;
3499         }
3500         if (err > 0)
3501                 err = -EEXIST;
3502         return err;
3503 }
3504
3505 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3506                         int mode, dev_t rdev)
3507 {
3508         struct btrfs_trans_handle *trans;
3509         struct btrfs_root *root = BTRFS_I(dir)->root;
3510         struct inode *inode = NULL;
3511         int err;
3512         int drop_inode = 0;
3513         u64 objectid;
3514         unsigned long nr = 0;
3515         u64 index = 0;
3516
3517         if (!new_valid_dev(rdev))
3518                 return -EINVAL;
3519
3520         err = btrfs_check_free_space(root, 1, 0);
3521         if (err)
3522                 goto fail;
3523
3524         trans = btrfs_start_transaction(root, 1);
3525         btrfs_set_trans_block_group(trans, dir);
3526
3527         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3528         if (err) {
3529                 err = -ENOSPC;
3530                 goto out_unlock;
3531         }
3532
3533         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3534                                 dentry->d_name.len,
3535                                 dentry->d_parent->d_inode->i_ino, objectid,
3536                                 BTRFS_I(dir)->block_group, mode, &index);
3537         err = PTR_ERR(inode);
3538         if (IS_ERR(inode))
3539                 goto out_unlock;
3540
3541         err = btrfs_init_acl(inode, dir);
3542         if (err) {
3543                 drop_inode = 1;
3544                 goto out_unlock;
3545         }
3546
3547         btrfs_set_trans_block_group(trans, inode);
3548         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3549         if (err)
3550                 drop_inode = 1;
3551         else {
3552                 inode->i_op = &btrfs_special_inode_operations;
3553                 init_special_inode(inode, inode->i_mode, rdev);
3554                 btrfs_update_inode(trans, root, inode);
3555         }
3556         dir->i_sb->s_dirt = 1;
3557         btrfs_update_inode_block_group(trans, inode);
3558         btrfs_update_inode_block_group(trans, dir);
3559 out_unlock:
3560         nr = trans->blocks_used;
3561         btrfs_end_transaction_throttle(trans, root);
3562 fail:
3563         if (drop_inode) {
3564                 inode_dec_link_count(inode);
3565                 iput(inode);
3566         }
3567         btrfs_btree_balance_dirty(root, nr);
3568         return err;
3569 }
3570
3571 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3572                         int mode, struct nameidata *nd)
3573 {
3574         struct btrfs_trans_handle *trans;
3575         struct btrfs_root *root = BTRFS_I(dir)->root;
3576         struct inode *inode = NULL;
3577         int err;
3578         int drop_inode = 0;
3579         unsigned long nr = 0;
3580         u64 objectid;
3581         u64 index = 0;
3582
3583         err = btrfs_check_free_space(root, 1, 0);
3584         if (err)
3585                 goto fail;
3586         trans = btrfs_start_transaction(root, 1);
3587         btrfs_set_trans_block_group(trans, dir);
3588
3589         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3590         if (err) {
3591                 err = -ENOSPC;
3592                 goto out_unlock;
3593         }
3594
3595         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3596                                 dentry->d_name.len,
3597                                 dentry->d_parent->d_inode->i_ino,
3598                                 objectid, BTRFS_I(dir)->block_group, mode,
3599                                 &index);
3600         err = PTR_ERR(inode);
3601         if (IS_ERR(inode))
3602                 goto out_unlock;
3603
3604         err = btrfs_init_acl(inode, dir);
3605         if (err) {
3606                 drop_inode = 1;
3607                 goto out_unlock;
3608         }
3609
3610         btrfs_set_trans_block_group(trans, inode);
3611         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3612         if (err)
3613                 drop_inode = 1;
3614         else {
3615                 inode->i_mapping->a_ops = &btrfs_aops;
3616                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3617                 inode->i_fop = &btrfs_file_operations;
3618                 inode->i_op = &btrfs_file_inode_operations;
3619                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3620         }
3621         dir->i_sb->s_dirt = 1;
3622         btrfs_update_inode_block_group(trans, inode);
3623         btrfs_update_inode_block_group(trans, dir);
3624 out_unlock:
3625         nr = trans->blocks_used;
3626         btrfs_end_transaction_throttle(trans, root);
3627 fail:
3628         if (drop_inode) {
3629                 inode_dec_link_count(inode);
3630                 iput(inode);
3631         }
3632         btrfs_btree_balance_dirty(root, nr);
3633         return err;
3634 }
3635
3636 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3637                       struct dentry *dentry)
3638 {
3639         struct btrfs_trans_handle *trans;
3640         struct btrfs_root *root = BTRFS_I(dir)->root;
3641         struct inode *inode = old_dentry->d_inode;
3642         u64 index;
3643         unsigned long nr = 0;
3644         int err;
3645         int drop_inode = 0;
3646
3647         if (inode->i_nlink == 0)
3648                 return -ENOENT;
3649
3650         btrfs_inc_nlink(inode);
3651         err = btrfs_check_free_space(root, 1, 0);
3652         if (err)
3653                 goto fail;
3654         err = btrfs_set_inode_index(dir, inode, &index);
3655         if (err)
3656                 goto fail;
3657
3658         trans = btrfs_start_transaction(root, 1);
3659
3660         btrfs_set_trans_block_group(trans, dir);
3661         atomic_inc(&inode->i_count);
3662
3663         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3664
3665         if (err)
3666                 drop_inode = 1;
3667
3668         dir->i_sb->s_dirt = 1;
3669         btrfs_update_inode_block_group(trans, dir);
3670         err = btrfs_update_inode(trans, root, inode);
3671
3672         if (err)
3673                 drop_inode = 1;
3674
3675         nr = trans->blocks_used;
3676         btrfs_end_transaction_throttle(trans, root);
3677 fail:
3678         if (drop_inode) {
3679                 inode_dec_link_count(inode);
3680                 iput(inode);
3681         }
3682         btrfs_btree_balance_dirty(root, nr);
3683         return err;
3684 }
3685
3686 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3687 {
3688         struct inode *inode = NULL;
3689         struct btrfs_trans_handle *trans;
3690         struct btrfs_root *root = BTRFS_I(dir)->root;
3691         int err = 0;
3692         int drop_on_err = 0;
3693         u64 objectid = 0;
3694         u64 index = 0;
3695         unsigned long nr = 1;
3696
3697         err = btrfs_check_free_space(root, 1, 0);
3698         if (err)
3699                 goto out_unlock;
3700
3701         trans = btrfs_start_transaction(root, 1);
3702         btrfs_set_trans_block_group(trans, dir);
3703
3704         if (IS_ERR(trans)) {
3705                 err = PTR_ERR(trans);
3706                 goto out_unlock;
3707         }
3708
3709         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3710         if (err) {
3711                 err = -ENOSPC;
3712                 goto out_unlock;
3713         }
3714
3715         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3716                                 dentry->d_name.len,
3717                                 dentry->d_parent->d_inode->i_ino, objectid,
3718                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3719                                 &index);
3720         if (IS_ERR(inode)) {
3721                 err = PTR_ERR(inode);
3722                 goto out_fail;
3723         }
3724
3725         drop_on_err = 1;
3726
3727         err = btrfs_init_acl(inode, dir);
3728         if (err)
3729                 goto out_fail;
3730
3731         inode->i_op = &btrfs_dir_inode_operations;
3732         inode->i_fop = &btrfs_dir_file_operations;
3733         btrfs_set_trans_block_group(trans, inode);
3734
3735         btrfs_i_size_write(inode, 0);
3736         err = btrfs_update_inode(trans, root, inode);
3737         if (err)
3738                 goto out_fail;
3739
3740         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3741                                  inode, dentry->d_name.name,
3742                                  dentry->d_name.len, 0, index);
3743         if (err)
3744                 goto out_fail;
3745
3746         d_instantiate(dentry, inode);
3747         drop_on_err = 0;
3748         dir->i_sb->s_dirt = 1;
3749         btrfs_update_inode_block_group(trans, inode);
3750         btrfs_update_inode_block_group(trans, dir);
3751
3752 out_fail:
3753         nr = trans->blocks_used;
3754         btrfs_end_transaction_throttle(trans, root);
3755
3756 out_unlock:
3757         if (drop_on_err)
3758                 iput(inode);
3759         btrfs_btree_balance_dirty(root, nr);
3760         return err;
3761 }
3762
3763 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3764  * and an extent that you want to insert, deal with overlap and insert
3765  * the new extent into the tree.
3766  */
3767 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3768                                 struct extent_map *existing,
3769                                 struct extent_map *em,
3770                                 u64 map_start, u64 map_len)
3771 {
3772         u64 start_diff;
3773
3774         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3775         start_diff = map_start - em->start;
3776         em->start = map_start;
3777         em->len = map_len;
3778         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3779             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3780                 em->block_start += start_diff;
3781                 em->block_len -= start_diff;
3782         }
3783         return add_extent_mapping(em_tree, em);
3784 }
3785
3786 static noinline int uncompress_inline(struct btrfs_path *path,
3787                                       struct inode *inode, struct page *page,
3788                                       size_t pg_offset, u64 extent_offset,
3789                                       struct btrfs_file_extent_item *item)
3790 {
3791         int ret;
3792         struct extent_buffer *leaf = path->nodes[0];
3793         char *tmp;
3794         size_t max_size;
3795         unsigned long inline_size;
3796         unsigned long ptr;
3797
3798         WARN_ON(pg_offset != 0);
3799         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3800         inline_size = btrfs_file_extent_inline_item_len(leaf,
3801                                         btrfs_item_nr(leaf, path->slots[0]));
3802         tmp = kmalloc(inline_size, GFP_NOFS);
3803         ptr = btrfs_file_extent_inline_start(item);
3804
3805         read_extent_buffer(leaf, tmp, ptr, inline_size);
3806
3807         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3808         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3809                                     inline_size, max_size);
3810         if (ret) {
3811                 char *kaddr = kmap_atomic(page, KM_USER0);
3812                 unsigned long copy_size = min_t(u64,
3813                                   PAGE_CACHE_SIZE - pg_offset,
3814                                   max_size - extent_offset);
3815                 memset(kaddr + pg_offset, 0, copy_size);
3816                 kunmap_atomic(kaddr, KM_USER0);
3817         }
3818         kfree(tmp);
3819         return 0;
3820 }
3821
3822 /*
3823  * a bit scary, this does extent mapping from logical file offset to the disk.
3824  * the ugly parts come from merging extents from the disk with the
3825  * in-ram representation.  This gets more complex because of the data=ordered code,
3826  * where the in-ram extents might be locked pending data=ordered completion.
3827  *
3828  * This also copies inline extents directly into the page.
3829  */
3830 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3831                                     size_t pg_offset, u64 start, u64 len,
3832                                     int create)
3833 {
3834         int ret;
3835         int err = 0;
3836         u64 bytenr;
3837         u64 extent_start = 0;
3838         u64 extent_end = 0;
3839         u64 objectid = inode->i_ino;
3840         u32 found_type;
3841         struct btrfs_path *path = NULL;
3842         struct btrfs_root *root = BTRFS_I(inode)->root;
3843         struct btrfs_file_extent_item *item;
3844         struct extent_buffer *leaf;
3845         struct btrfs_key found_key;
3846         struct extent_map *em = NULL;
3847         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3848         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3849         struct btrfs_trans_handle *trans = NULL;
3850         int compressed;
3851
3852 again:
3853         spin_lock(&em_tree->lock);
3854         em = lookup_extent_mapping(em_tree, start, len);
3855         if (em)
3856                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3857         spin_unlock(&em_tree->lock);
3858
3859         if (em) {
3860                 if (em->start > start || em->start + em->len <= start)
3861                         free_extent_map(em);
3862                 else if (em->block_start == EXTENT_MAP_INLINE && page)
3863                         free_extent_map(em);
3864                 else
3865                         goto out;
3866         }
3867         em = alloc_extent_map(GFP_NOFS);
3868         if (!em) {
3869                 err = -ENOMEM;
3870                 goto out;
3871         }
3872         em->bdev = root->fs_info->fs_devices->latest_bdev;
3873         em->start = EXTENT_MAP_HOLE;
3874         em->orig_start = EXTENT_MAP_HOLE;
3875         em->len = (u64)-1;
3876         em->block_len = (u64)-1;
3877
3878         if (!path) {
3879                 path = btrfs_alloc_path();
3880                 BUG_ON(!path);
3881         }
3882
3883         ret = btrfs_lookup_file_extent(trans, root, path,
3884                                        objectid, start, trans != NULL);
3885         if (ret < 0) {
3886                 err = ret;
3887                 goto out;
3888         }
3889
3890         if (ret != 0) {
3891                 if (path->slots[0] == 0)
3892                         goto not_found;
3893                 path->slots[0]--;
3894         }
3895
3896         leaf = path->nodes[0];
3897         item = btrfs_item_ptr(leaf, path->slots[0],
3898                               struct btrfs_file_extent_item);
3899         /* are we inside the extent that was found? */
3900         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3901         found_type = btrfs_key_type(&found_key);
3902         if (found_key.objectid != objectid ||
3903             found_type != BTRFS_EXTENT_DATA_KEY) {
3904                 goto not_found;
3905         }
3906
3907         found_type = btrfs_file_extent_type(leaf, item);
3908         extent_start = found_key.offset;
3909         compressed = btrfs_file_extent_compression(leaf, item);
3910         if (found_type == BTRFS_FILE_EXTENT_REG ||
3911             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3912                 extent_end = extent_start +
3913                        btrfs_file_extent_num_bytes(leaf, item);
3914         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3915                 size_t size;
3916                 size = btrfs_file_extent_inline_len(leaf, item);
3917                 extent_end = (extent_start + size + root->sectorsize - 1) &
3918                         ~((u64)root->sectorsize - 1);
3919         }
3920
3921         if (start >= extent_end) {
3922                 path->slots[0]++;
3923                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3924                         ret = btrfs_next_leaf(root, path);
3925                         if (ret < 0) {
3926                                 err = ret;
3927                                 goto out;
3928                         }
3929                         if (ret > 0)
3930                                 goto not_found;
3931                         leaf = path->nodes[0];
3932                 }
3933                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3934                 if (found_key.objectid != objectid ||
3935                     found_key.type != BTRFS_EXTENT_DATA_KEY)
3936                         goto not_found;
3937                 if (start + len <= found_key.offset)
3938                         goto not_found;
3939                 em->start = start;
3940                 em->len = found_key.offset - start;
3941                 goto not_found_em;
3942         }
3943
3944         if (found_type == BTRFS_FILE_EXTENT_REG ||
3945             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3946                 em->start = extent_start;
3947                 em->len = extent_end - extent_start;
3948                 em->orig_start = extent_start -
3949                                  btrfs_file_extent_offset(leaf, item);
3950                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3951                 if (bytenr == 0) {
3952                         em->block_start = EXTENT_MAP_HOLE;
3953                         goto insert;
3954                 }
3955                 if (compressed) {
3956                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3957                         em->block_start = bytenr;
3958                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3959                                                                          item);
3960                 } else {
3961                         bytenr += btrfs_file_extent_offset(leaf, item);
3962                         em->block_start = bytenr;
3963                         em->block_len = em->len;
3964                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3965                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
3966                 }
3967                 goto insert;
3968         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3969                 unsigned long ptr;
3970                 char *map;
3971                 size_t size;
3972                 size_t extent_offset;
3973                 size_t copy_size;
3974
3975                 em->block_start = EXTENT_MAP_INLINE;
3976                 if (!page || create) {
3977                         em->start = extent_start;
3978                         em->len = extent_end - extent_start;
3979                         goto out;
3980                 }
3981
3982                 size = btrfs_file_extent_inline_len(leaf, item);
3983                 extent_offset = page_offset(page) + pg_offset - extent_start;
3984                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
3985                                 size - extent_offset);
3986                 em->start = extent_start + extent_offset;
3987                 em->len = (copy_size + root->sectorsize - 1) &
3988                         ~((u64)root->sectorsize - 1);
3989                 em->orig_start = EXTENT_MAP_INLINE;
3990                 if (compressed)
3991                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3992                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
3993                 if (create == 0 && !PageUptodate(page)) {
3994                         if (btrfs_file_extent_compression(leaf, item) ==
3995                             BTRFS_COMPRESS_ZLIB) {
3996                                 ret = uncompress_inline(path, inode, page,
3997                                                         pg_offset,
3998                                                         extent_offset, item);
3999                                 BUG_ON(ret);
4000                         } else {
4001                                 map = kmap(page);
4002                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4003                                                    copy_size);
4004                                 kunmap(page);
4005                         }
4006                         flush_dcache_page(page);
4007                 } else if (create && PageUptodate(page)) {
4008                         if (!trans) {
4009                                 kunmap(page);
4010                                 free_extent_map(em);
4011                                 em = NULL;
4012                                 btrfs_release_path(root, path);
4013                                 trans = btrfs_join_transaction(root, 1);
4014                                 goto again;
4015                         }
4016                         map = kmap(page);
4017                         write_extent_buffer(leaf, map + pg_offset, ptr,
4018                                             copy_size);
4019                         kunmap(page);
4020                         btrfs_mark_buffer_dirty(leaf);
4021                 }
4022                 set_extent_uptodate(io_tree, em->start,
4023                                     extent_map_end(em) - 1, GFP_NOFS);
4024                 goto insert;
4025         } else {
4026                 printk("unkknown found_type %d\n", found_type);
4027                 WARN_ON(1);
4028         }
4029 not_found:
4030         em->start = start;
4031         em->len = len;
4032 not_found_em:
4033         em->block_start = EXTENT_MAP_HOLE;
4034         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4035 insert:
4036         btrfs_release_path(root, path);
4037         if (em->start > start || extent_map_end(em) <= start) {
4038                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
4039                 err = -EIO;
4040                 goto out;
4041         }
4042
4043         err = 0;
4044         spin_lock(&em_tree->lock);
4045         ret = add_extent_mapping(em_tree, em);
4046         /* it is possible that someone inserted the extent into the tree
4047          * while we had the lock dropped.  It is also possible that
4048          * an overlapping map exists in the tree
4049          */
4050         if (ret == -EEXIST) {
4051                 struct extent_map *existing;
4052
4053                 ret = 0;
4054
4055                 existing = lookup_extent_mapping(em_tree, start, len);
4056                 if (existing && (existing->start > start ||
4057                     existing->start + existing->len <= start)) {
4058                         free_extent_map(existing);
4059                         existing = NULL;
4060                 }
4061                 if (!existing) {
4062                         existing = lookup_extent_mapping(em_tree, em->start,
4063                                                          em->len);
4064                         if (existing) {
4065                                 err = merge_extent_mapping(em_tree, existing,
4066                                                            em, start,
4067                                                            root->sectorsize);
4068                                 free_extent_map(existing);
4069                                 if (err) {
4070                                         free_extent_map(em);
4071                                         em = NULL;
4072                                 }
4073                         } else {
4074                                 err = -EIO;
4075                                 printk("failing to insert %Lu %Lu\n",
4076                                        start, len);
4077                                 free_extent_map(em);
4078                                 em = NULL;
4079                         }
4080                 } else {
4081                         free_extent_map(em);
4082                         em = existing;
4083                         err = 0;
4084                 }
4085         }
4086         spin_unlock(&em_tree->lock);
4087 out:
4088         if (path)
4089                 btrfs_free_path(path);
4090         if (trans) {
4091                 ret = btrfs_end_transaction(trans, root);
4092                 if (!err) {
4093                         err = ret;
4094                 }
4095         }
4096         if (err) {
4097                 free_extent_map(em);
4098                 WARN_ON(1);
4099                 return ERR_PTR(err);
4100         }
4101         return em;
4102 }
4103
4104 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4105                         const struct iovec *iov, loff_t offset,
4106                         unsigned long nr_segs)
4107 {
4108         return -EINVAL;
4109 }
4110
4111 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4112 {
4113         return extent_bmap(mapping, iblock, btrfs_get_extent);
4114 }
4115
4116 int btrfs_readpage(struct file *file, struct page *page)
4117 {
4118         struct extent_io_tree *tree;
4119         tree = &BTRFS_I(page->mapping->host)->io_tree;
4120         return extent_read_full_page(tree, page, btrfs_get_extent);
4121 }
4122
4123 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4124 {
4125         struct extent_io_tree *tree;
4126
4127
4128         if (current->flags & PF_MEMALLOC) {
4129                 redirty_page_for_writepage(wbc, page);
4130                 unlock_page(page);
4131                 return 0;
4132         }
4133         tree = &BTRFS_I(page->mapping->host)->io_tree;
4134         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4135 }
4136
4137 int btrfs_writepages(struct address_space *mapping,
4138                      struct writeback_control *wbc)
4139 {
4140         struct extent_io_tree *tree;
4141
4142         tree = &BTRFS_I(mapping->host)->io_tree;
4143         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4144 }
4145
4146 static int
4147 btrfs_readpages(struct file *file, struct address_space *mapping,
4148                 struct list_head *pages, unsigned nr_pages)
4149 {
4150         struct extent_io_tree *tree;
4151         tree = &BTRFS_I(mapping->host)->io_tree;
4152         return extent_readpages(tree, mapping, pages, nr_pages,
4153                                 btrfs_get_extent);
4154 }
4155 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4156 {
4157         struct extent_io_tree *tree;
4158         struct extent_map_tree *map;
4159         int ret;
4160
4161         tree = &BTRFS_I(page->mapping->host)->io_tree;
4162         map = &BTRFS_I(page->mapping->host)->extent_tree;
4163         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4164         if (ret == 1) {
4165                 ClearPagePrivate(page);
4166                 set_page_private(page, 0);
4167                 page_cache_release(page);
4168         }
4169         return ret;
4170 }
4171
4172 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4173 {
4174         if (PageWriteback(page) || PageDirty(page))
4175                 return 0;
4176         return __btrfs_releasepage(page, gfp_flags);
4177 }
4178
4179 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4180 {
4181         struct extent_io_tree *tree;
4182         struct btrfs_ordered_extent *ordered;
4183         u64 page_start = page_offset(page);
4184         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4185
4186         wait_on_page_writeback(page);
4187         tree = &BTRFS_I(page->mapping->host)->io_tree;
4188         if (offset) {
4189                 btrfs_releasepage(page, GFP_NOFS);
4190                 return;
4191         }
4192
4193         lock_extent(tree, page_start, page_end, GFP_NOFS);
4194         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4195                                            page_offset(page));
4196         if (ordered) {
4197                 /*
4198                  * IO on this page will never be started, so we need
4199                  * to account for any ordered extents now
4200                  */
4201                 clear_extent_bit(tree, page_start, page_end,
4202                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4203                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4204                 btrfs_finish_ordered_io(page->mapping->host,
4205                                         page_start, page_end);
4206                 btrfs_put_ordered_extent(ordered);
4207                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4208         }
4209         clear_extent_bit(tree, page_start, page_end,
4210                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4211                  EXTENT_ORDERED,
4212                  1, 1, GFP_NOFS);
4213         __btrfs_releasepage(page, GFP_NOFS);
4214
4215         ClearPageChecked(page);
4216         if (PagePrivate(page)) {
4217                 ClearPagePrivate(page);
4218                 set_page_private(page, 0);
4219                 page_cache_release(page);
4220         }
4221 }
4222
4223 /*
4224  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4225  * called from a page fault handler when a page is first dirtied. Hence we must
4226  * be careful to check for EOF conditions here. We set the page up correctly
4227  * for a written page which means we get ENOSPC checking when writing into
4228  * holes and correct delalloc and unwritten extent mapping on filesystems that
4229  * support these features.
4230  *
4231  * We are not allowed to take the i_mutex here so we have to play games to
4232  * protect against truncate races as the page could now be beyond EOF.  Because
4233  * vmtruncate() writes the inode size before removing pages, once we have the
4234  * page lock we can determine safely if the page is beyond EOF. If it is not
4235  * beyond EOF, then the page is guaranteed safe against truncation until we
4236  * unlock the page.
4237  */
4238 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4239 {
4240         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4241         struct btrfs_root *root = BTRFS_I(inode)->root;
4242         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4243         struct btrfs_ordered_extent *ordered;
4244         char *kaddr;
4245         unsigned long zero_start;
4246         loff_t size;
4247         int ret;
4248         u64 page_start;
4249         u64 page_end;
4250
4251         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4252         if (ret)
4253                 goto out;
4254
4255         ret = -EINVAL;
4256 again:
4257         lock_page(page);
4258         size = i_size_read(inode);
4259         page_start = page_offset(page);
4260         page_end = page_start + PAGE_CACHE_SIZE - 1;
4261
4262         if ((page->mapping != inode->i_mapping) ||
4263             (page_start >= size)) {
4264                 /* page got truncated out from underneath us */
4265                 goto out_unlock;
4266         }
4267         wait_on_page_writeback(page);
4268
4269         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4270         set_page_extent_mapped(page);
4271
4272         /*
4273          * we can't set the delalloc bits if there are pending ordered
4274          * extents.  Drop our locks and wait for them to finish
4275          */
4276         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4277         if (ordered) {
4278                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4279                 unlock_page(page);
4280                 btrfs_start_ordered_extent(inode, ordered, 1);
4281                 btrfs_put_ordered_extent(ordered);
4282                 goto again;
4283         }
4284
4285         btrfs_set_extent_delalloc(inode, page_start, page_end);
4286         ret = 0;
4287
4288         /* page is wholly or partially inside EOF */
4289         if (page_start + PAGE_CACHE_SIZE > size)
4290                 zero_start = size & ~PAGE_CACHE_MASK;
4291         else
4292                 zero_start = PAGE_CACHE_SIZE;
4293
4294         if (zero_start != PAGE_CACHE_SIZE) {
4295                 kaddr = kmap(page);
4296                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4297                 flush_dcache_page(page);
4298                 kunmap(page);
4299         }
4300         ClearPageChecked(page);
4301         set_page_dirty(page);
4302         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4303
4304 out_unlock:
4305         unlock_page(page);
4306 out:
4307         return ret;
4308 }
4309
4310 static void btrfs_truncate(struct inode *inode)
4311 {
4312         struct btrfs_root *root = BTRFS_I(inode)->root;
4313         int ret;
4314         struct btrfs_trans_handle *trans;
4315         unsigned long nr;
4316         u64 mask = root->sectorsize - 1;
4317
4318         if (!S_ISREG(inode->i_mode))
4319                 return;
4320         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4321                 return;
4322
4323         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4324         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4325
4326         trans = btrfs_start_transaction(root, 1);
4327         btrfs_set_trans_block_group(trans, inode);
4328         btrfs_i_size_write(inode, inode->i_size);
4329
4330         ret = btrfs_orphan_add(trans, inode);
4331         if (ret)
4332                 goto out;
4333         /* FIXME, add redo link to tree so we don't leak on crash */
4334         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4335                                       BTRFS_EXTENT_DATA_KEY);
4336         btrfs_update_inode(trans, root, inode);
4337
4338         ret = btrfs_orphan_del(trans, inode);
4339         BUG_ON(ret);
4340
4341 out:
4342         nr = trans->blocks_used;
4343         ret = btrfs_end_transaction_throttle(trans, root);
4344         BUG_ON(ret);
4345         btrfs_btree_balance_dirty(root, nr);
4346 }
4347
4348 /*
4349  * Invalidate a single dcache entry at the root of the filesystem.
4350  * Needed after creation of snapshot or subvolume.
4351  */
4352 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
4353                                   int namelen)
4354 {
4355         struct dentry *alias, *entry;
4356         struct qstr qstr;
4357
4358         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
4359         if (alias) {
4360                 qstr.name = name;
4361                 qstr.len = namelen;
4362                 /* change me if btrfs ever gets a d_hash operation */
4363                 qstr.hash = full_name_hash(qstr.name, qstr.len);
4364                 entry = d_lookup(alias, &qstr);
4365                 dput(alias);
4366                 if (entry) {
4367                         d_invalidate(entry);
4368                         dput(entry);
4369                 }
4370         }
4371 }
4372
4373 /*
4374  * create a new subvolume directory/inode (helper for the ioctl).
4375  */
4376 int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
4377                 struct btrfs_trans_handle *trans, u64 new_dirid,
4378                 struct btrfs_block_group_cache *block_group)
4379 {
4380         struct inode *inode;
4381         int error;
4382         u64 index = 0;
4383
4384         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4385                                 new_dirid, block_group, S_IFDIR | 0700, &index);
4386         if (IS_ERR(inode))
4387                 return PTR_ERR(inode);
4388         inode->i_op = &btrfs_dir_inode_operations;
4389         inode->i_fop = &btrfs_dir_file_operations;
4390         new_root->inode = inode;
4391
4392         inode->i_nlink = 1;
4393         btrfs_i_size_write(inode, 0);
4394
4395         error = btrfs_update_inode(trans, new_root, inode);
4396         if (error)
4397                 return error;
4398
4399         atomic_inc(&inode->i_count);
4400         d_instantiate(dentry, inode);
4401         return 0;
4402 }
4403
4404 /* helper function for file defrag and space balancing.  This
4405  * forces readahead on a given range of bytes in an inode
4406  */
4407 unsigned long btrfs_force_ra(struct address_space *mapping,
4408                               struct file_ra_state *ra, struct file *file,
4409                               pgoff_t offset, pgoff_t last_index)
4410 {
4411         pgoff_t req_size = last_index - offset + 1;
4412
4413         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4414         return offset + req_size;
4415 }
4416
4417 struct inode *btrfs_alloc_inode(struct super_block *sb)
4418 {
4419         struct btrfs_inode *ei;
4420
4421         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4422         if (!ei)
4423                 return NULL;
4424         ei->last_trans = 0;
4425         ei->logged_trans = 0;
4426         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4427         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4428         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4429         INIT_LIST_HEAD(&ei->i_orphan);
4430         return &ei->vfs_inode;
4431 }
4432
4433 void btrfs_destroy_inode(struct inode *inode)
4434 {
4435         struct btrfs_ordered_extent *ordered;
4436         WARN_ON(!list_empty(&inode->i_dentry));
4437         WARN_ON(inode->i_data.nrpages);
4438
4439         if (BTRFS_I(inode)->i_acl &&
4440             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4441                 posix_acl_release(BTRFS_I(inode)->i_acl);
4442         if (BTRFS_I(inode)->i_default_acl &&
4443             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4444                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4445
4446         spin_lock(&BTRFS_I(inode)->root->list_lock);
4447         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4448                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4449                        " list\n", inode->i_ino);
4450                 dump_stack();
4451         }
4452         spin_unlock(&BTRFS_I(inode)->root->list_lock);
4453
4454         while(1) {
4455                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4456                 if (!ordered)
4457                         break;
4458                 else {
4459                         printk("found ordered extent %Lu %Lu\n",
4460                                ordered->file_offset, ordered->len);
4461                         btrfs_remove_ordered_extent(inode, ordered);
4462                         btrfs_put_ordered_extent(ordered);
4463                         btrfs_put_ordered_extent(ordered);
4464                 }
4465         }
4466         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4467         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4468 }
4469
4470 static void init_once(void *foo)
4471 {
4472         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4473
4474         inode_init_once(&ei->vfs_inode);
4475 }
4476
4477 void btrfs_destroy_cachep(void)
4478 {
4479         if (btrfs_inode_cachep)
4480                 kmem_cache_destroy(btrfs_inode_cachep);
4481         if (btrfs_trans_handle_cachep)
4482                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4483         if (btrfs_transaction_cachep)
4484                 kmem_cache_destroy(btrfs_transaction_cachep);
4485         if (btrfs_bit_radix_cachep)
4486                 kmem_cache_destroy(btrfs_bit_radix_cachep);
4487         if (btrfs_path_cachep)
4488                 kmem_cache_destroy(btrfs_path_cachep);
4489 }
4490
4491 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4492                                        unsigned long extra_flags,
4493                                        void (*ctor)(void *))
4494 {
4495         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4496                                  SLAB_MEM_SPREAD | extra_flags), ctor);
4497 }
4498
4499 int btrfs_init_cachep(void)
4500 {
4501         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4502                                           sizeof(struct btrfs_inode),
4503                                           0, init_once);
4504         if (!btrfs_inode_cachep)
4505                 goto fail;
4506         btrfs_trans_handle_cachep =
4507                         btrfs_cache_create("btrfs_trans_handle_cache",
4508                                            sizeof(struct btrfs_trans_handle),
4509                                            0, NULL);
4510         if (!btrfs_trans_handle_cachep)
4511                 goto fail;
4512         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4513                                              sizeof(struct btrfs_transaction),
4514                                              0, NULL);
4515         if (!btrfs_transaction_cachep)
4516                 goto fail;
4517         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4518                                          sizeof(struct btrfs_path),
4519                                          0, NULL);
4520         if (!btrfs_path_cachep)
4521                 goto fail;
4522         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4523                                               SLAB_DESTROY_BY_RCU, NULL);
4524         if (!btrfs_bit_radix_cachep)
4525                 goto fail;
4526         return 0;
4527 fail:
4528         btrfs_destroy_cachep();
4529         return -ENOMEM;
4530 }
4531
4532 static int btrfs_getattr(struct vfsmount *mnt,
4533                          struct dentry *dentry, struct kstat *stat)
4534 {
4535         struct inode *inode = dentry->d_inode;
4536         generic_fillattr(inode, stat);
4537         stat->blksize = PAGE_CACHE_SIZE;
4538         stat->blocks = (inode_get_bytes(inode) +
4539                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4540         return 0;
4541 }
4542
4543 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4544                            struct inode * new_dir,struct dentry *new_dentry)
4545 {
4546         struct btrfs_trans_handle *trans;
4547         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4548         struct inode *new_inode = new_dentry->d_inode;
4549         struct inode *old_inode = old_dentry->d_inode;
4550         struct timespec ctime = CURRENT_TIME;
4551         u64 index = 0;
4552         int ret;
4553
4554         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4555             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4556                 return -ENOTEMPTY;
4557         }
4558
4559         ret = btrfs_check_free_space(root, 1, 0);
4560         if (ret)
4561                 goto out_unlock;
4562
4563         trans = btrfs_start_transaction(root, 1);
4564
4565         btrfs_set_trans_block_group(trans, new_dir);
4566
4567         btrfs_inc_nlink(old_dentry->d_inode);
4568         old_dir->i_ctime = old_dir->i_mtime = ctime;
4569         new_dir->i_ctime = new_dir->i_mtime = ctime;
4570         old_inode->i_ctime = ctime;
4571
4572         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4573                                  old_dentry->d_name.name,
4574                                  old_dentry->d_name.len);
4575         if (ret)
4576                 goto out_fail;
4577
4578         if (new_inode) {
4579                 new_inode->i_ctime = CURRENT_TIME;
4580                 ret = btrfs_unlink_inode(trans, root, new_dir,
4581                                          new_dentry->d_inode,
4582                                          new_dentry->d_name.name,
4583                                          new_dentry->d_name.len);
4584                 if (ret)
4585                         goto out_fail;
4586                 if (new_inode->i_nlink == 0) {
4587                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4588                         if (ret)
4589                                 goto out_fail;
4590                 }
4591
4592         }
4593         ret = btrfs_set_inode_index(new_dir, old_inode, &index);
4594         if (ret)
4595                 goto out_fail;
4596
4597         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4598                              old_inode, new_dentry->d_name.name,
4599                              new_dentry->d_name.len, 1, index);
4600         if (ret)
4601                 goto out_fail;
4602
4603 out_fail:
4604         btrfs_end_transaction_throttle(trans, root);
4605 out_unlock:
4606         return ret;
4607 }
4608
4609 /*
4610  * some fairly slow code that needs optimization. This walks the list
4611  * of all the inodes with pending delalloc and forces them to disk.
4612  */
4613 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4614 {
4615         struct list_head *head = &root->fs_info->delalloc_inodes;
4616         struct btrfs_inode *binode;
4617         struct inode *inode;
4618         unsigned long flags;
4619
4620         if (root->fs_info->sb->s_flags & MS_RDONLY)
4621                 return -EROFS;
4622
4623         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4624         while(!list_empty(head)) {
4625                 binode = list_entry(head->next, struct btrfs_inode,
4626                                     delalloc_inodes);
4627                 inode = igrab(&binode->vfs_inode);
4628                 if (!inode)
4629                         list_del_init(&binode->delalloc_inodes);
4630                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4631                 if (inode) {
4632                         filemap_flush(inode->i_mapping);
4633                         iput(inode);
4634                 }
4635                 cond_resched();
4636                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4637         }
4638         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4639
4640         /* the filemap_flush will queue IO into the worker threads, but
4641          * we have to make sure the IO is actually started and that
4642          * ordered extents get created before we return
4643          */
4644         atomic_inc(&root->fs_info->async_submit_draining);
4645         while(atomic_read(&root->fs_info->nr_async_submits) ||
4646               atomic_read(&root->fs_info->async_delalloc_pages)) {
4647                 wait_event(root->fs_info->async_submit_wait,
4648                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4649                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4650         }
4651         atomic_dec(&root->fs_info->async_submit_draining);
4652         return 0;
4653 }
4654
4655 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4656                          const char *symname)
4657 {
4658         struct btrfs_trans_handle *trans;
4659         struct btrfs_root *root = BTRFS_I(dir)->root;
4660         struct btrfs_path *path;
4661         struct btrfs_key key;
4662         struct inode *inode = NULL;
4663         int err;
4664         int drop_inode = 0;
4665         u64 objectid;
4666         u64 index = 0 ;
4667         int name_len;
4668         int datasize;
4669         unsigned long ptr;
4670         struct btrfs_file_extent_item *ei;
4671         struct extent_buffer *leaf;
4672         unsigned long nr = 0;
4673
4674         name_len = strlen(symname) + 1;
4675         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4676                 return -ENAMETOOLONG;
4677
4678         err = btrfs_check_free_space(root, 1, 0);
4679         if (err)
4680                 goto out_fail;
4681
4682         trans = btrfs_start_transaction(root, 1);
4683         btrfs_set_trans_block_group(trans, dir);
4684
4685         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4686         if (err) {
4687                 err = -ENOSPC;
4688                 goto out_unlock;
4689         }
4690
4691         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4692                                 dentry->d_name.len,
4693                                 dentry->d_parent->d_inode->i_ino, objectid,
4694                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4695                                 &index);
4696         err = PTR_ERR(inode);
4697         if (IS_ERR(inode))
4698                 goto out_unlock;
4699
4700         err = btrfs_init_acl(inode, dir);
4701         if (err) {
4702                 drop_inode = 1;
4703                 goto out_unlock;
4704         }
4705
4706         btrfs_set_trans_block_group(trans, inode);
4707         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4708         if (err)
4709                 drop_inode = 1;
4710         else {
4711                 inode->i_mapping->a_ops = &btrfs_aops;
4712                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4713                 inode->i_fop = &btrfs_file_operations;
4714                 inode->i_op = &btrfs_file_inode_operations;
4715                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4716         }
4717         dir->i_sb->s_dirt = 1;
4718         btrfs_update_inode_block_group(trans, inode);
4719         btrfs_update_inode_block_group(trans, dir);
4720         if (drop_inode)
4721                 goto out_unlock;
4722
4723         path = btrfs_alloc_path();
4724         BUG_ON(!path);
4725         key.objectid = inode->i_ino;
4726         key.offset = 0;
4727         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4728         datasize = btrfs_file_extent_calc_inline_size(name_len);
4729         err = btrfs_insert_empty_item(trans, root, path, &key,
4730                                       datasize);
4731         if (err) {
4732                 drop_inode = 1;
4733                 goto out_unlock;
4734         }
4735         leaf = path->nodes[0];
4736         ei = btrfs_item_ptr(leaf, path->slots[0],
4737                             struct btrfs_file_extent_item);
4738         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4739         btrfs_set_file_extent_type(leaf, ei,
4740                                    BTRFS_FILE_EXTENT_INLINE);
4741         btrfs_set_file_extent_encryption(leaf, ei, 0);
4742         btrfs_set_file_extent_compression(leaf, ei, 0);
4743         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4744         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4745
4746         ptr = btrfs_file_extent_inline_start(ei);
4747         write_extent_buffer(leaf, symname, ptr, name_len);
4748         btrfs_mark_buffer_dirty(leaf);
4749         btrfs_free_path(path);
4750
4751         inode->i_op = &btrfs_symlink_inode_operations;
4752         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4753         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4754         inode_set_bytes(inode, name_len);
4755         btrfs_i_size_write(inode, name_len - 1);
4756         err = btrfs_update_inode(trans, root, inode);
4757         if (err)
4758                 drop_inode = 1;
4759
4760 out_unlock:
4761         nr = trans->blocks_used;
4762         btrfs_end_transaction_throttle(trans, root);
4763 out_fail:
4764         if (drop_inode) {
4765                 inode_dec_link_count(inode);
4766                 iput(inode);
4767         }
4768         btrfs_btree_balance_dirty(root, nr);
4769         return err;
4770 }
4771
4772 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4773                                u64 alloc_hint, int mode)
4774 {
4775         struct btrfs_trans_handle *trans;
4776         struct btrfs_root *root = BTRFS_I(inode)->root;
4777         struct btrfs_key ins;
4778         u64 alloc_size;
4779         u64 cur_offset = start;
4780         u64 num_bytes = end - start;
4781         int ret = 0;
4782
4783         trans = btrfs_join_transaction(root, 1);
4784         BUG_ON(!trans);
4785         btrfs_set_trans_block_group(trans, inode);
4786
4787         while (num_bytes > 0) {
4788                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4789                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4790                                            root->sectorsize, 0, alloc_hint,
4791                                            (u64)-1, &ins, 1);
4792                 if (ret) {
4793                         WARN_ON(1);
4794                         goto out;
4795                 }
4796                 ret = insert_reserved_file_extent(trans, inode,
4797                                                   cur_offset, ins.objectid,
4798                                                   ins.offset, ins.offset,
4799                                                   ins.offset, 0, 0, 0,
4800                                                   BTRFS_FILE_EXTENT_PREALLOC);
4801                 BUG_ON(ret);
4802                 num_bytes -= ins.offset;
4803                 cur_offset += ins.offset;
4804                 alloc_hint = ins.objectid + ins.offset;
4805         }
4806 out:
4807         if (cur_offset > start) {
4808                 inode->i_ctime = CURRENT_TIME;
4809                 btrfs_set_flag(inode, PREALLOC);
4810                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4811                     cur_offset > i_size_read(inode))
4812                         btrfs_i_size_write(inode, cur_offset);
4813                 ret = btrfs_update_inode(trans, root, inode);
4814                 BUG_ON(ret);
4815         }
4816
4817         btrfs_end_transaction(trans, root);
4818         return ret;
4819 }
4820
4821 static long btrfs_fallocate(struct inode *inode, int mode,
4822                             loff_t offset, loff_t len)
4823 {
4824         u64 cur_offset;
4825         u64 last_byte;
4826         u64 alloc_start;
4827         u64 alloc_end;
4828         u64 alloc_hint = 0;
4829         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4830         struct extent_map *em;
4831         int ret;
4832
4833         alloc_start = offset & ~mask;
4834         alloc_end =  (offset + len + mask) & ~mask;
4835
4836         mutex_lock(&inode->i_mutex);
4837         if (alloc_start > inode->i_size) {
4838                 ret = btrfs_cont_expand(inode, alloc_start);
4839                 if (ret)
4840                         goto out;
4841         }
4842
4843         while (1) {
4844                 struct btrfs_ordered_extent *ordered;
4845                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4846                             alloc_end - 1, GFP_NOFS);
4847                 ordered = btrfs_lookup_first_ordered_extent(inode,
4848                                                             alloc_end - 1);
4849                 if (ordered &&
4850                     ordered->file_offset + ordered->len > alloc_start &&
4851                     ordered->file_offset < alloc_end) {
4852                         btrfs_put_ordered_extent(ordered);
4853                         unlock_extent(&BTRFS_I(inode)->io_tree,
4854                                       alloc_start, alloc_end - 1, GFP_NOFS);
4855                         btrfs_wait_ordered_range(inode, alloc_start,
4856                                                  alloc_end - alloc_start);
4857                 } else {
4858                         if (ordered)
4859                                 btrfs_put_ordered_extent(ordered);
4860                         break;
4861                 }
4862         }
4863
4864         cur_offset = alloc_start;
4865         while (1) {
4866                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4867                                       alloc_end - cur_offset, 0);
4868                 BUG_ON(IS_ERR(em) || !em);
4869                 last_byte = min(extent_map_end(em), alloc_end);
4870                 last_byte = (last_byte + mask) & ~mask;
4871                 if (em->block_start == EXTENT_MAP_HOLE) {
4872                         ret = prealloc_file_range(inode, cur_offset,
4873                                         last_byte, alloc_hint, mode);
4874                         if (ret < 0) {
4875                                 free_extent_map(em);
4876                                 break;
4877                         }
4878                 }
4879                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4880                         alloc_hint = em->block_start;
4881                 free_extent_map(em);
4882
4883                 cur_offset = last_byte;
4884                 if (cur_offset >= alloc_end) {
4885                         ret = 0;
4886                         break;
4887                 }
4888         }
4889         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4890                       GFP_NOFS);
4891 out:
4892         mutex_unlock(&inode->i_mutex);
4893         return ret;
4894 }
4895
4896 static int btrfs_set_page_dirty(struct page *page)
4897 {
4898         return __set_page_dirty_nobuffers(page);
4899 }
4900
4901 static int btrfs_permission(struct inode *inode, int mask)
4902 {
4903         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4904                 return -EACCES;
4905         return generic_permission(inode, mask, btrfs_check_acl);
4906 }
4907
4908 static struct inode_operations btrfs_dir_inode_operations = {
4909         .lookup         = btrfs_lookup,
4910         .create         = btrfs_create,
4911         .unlink         = btrfs_unlink,
4912         .link           = btrfs_link,
4913         .mkdir          = btrfs_mkdir,
4914         .rmdir          = btrfs_rmdir,
4915         .rename         = btrfs_rename,
4916         .symlink        = btrfs_symlink,
4917         .setattr        = btrfs_setattr,
4918         .mknod          = btrfs_mknod,
4919         .setxattr       = btrfs_setxattr,
4920         .getxattr       = btrfs_getxattr,
4921         .listxattr      = btrfs_listxattr,
4922         .removexattr    = btrfs_removexattr,
4923         .permission     = btrfs_permission,
4924 };
4925 static struct inode_operations btrfs_dir_ro_inode_operations = {
4926         .lookup         = btrfs_lookup,
4927         .permission     = btrfs_permission,
4928 };
4929 static struct file_operations btrfs_dir_file_operations = {
4930         .llseek         = generic_file_llseek,
4931         .read           = generic_read_dir,
4932         .readdir        = btrfs_real_readdir,
4933         .unlocked_ioctl = btrfs_ioctl,
4934 #ifdef CONFIG_COMPAT
4935         .compat_ioctl   = btrfs_ioctl,
4936 #endif
4937         .release        = btrfs_release_file,
4938         .fsync          = btrfs_sync_file,
4939 };
4940
4941 static struct extent_io_ops btrfs_extent_io_ops = {
4942         .fill_delalloc = run_delalloc_range,
4943         .submit_bio_hook = btrfs_submit_bio_hook,
4944         .merge_bio_hook = btrfs_merge_bio_hook,
4945         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4946         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4947         .writepage_start_hook = btrfs_writepage_start_hook,
4948         .readpage_io_failed_hook = btrfs_io_failed_hook,
4949         .set_bit_hook = btrfs_set_bit_hook,
4950         .clear_bit_hook = btrfs_clear_bit_hook,
4951 };
4952
4953 static struct address_space_operations btrfs_aops = {
4954         .readpage       = btrfs_readpage,
4955         .writepage      = btrfs_writepage,
4956         .writepages     = btrfs_writepages,
4957         .readpages      = btrfs_readpages,
4958         .sync_page      = block_sync_page,
4959         .bmap           = btrfs_bmap,
4960         .direct_IO      = btrfs_direct_IO,
4961         .invalidatepage = btrfs_invalidatepage,
4962         .releasepage    = btrfs_releasepage,
4963         .set_page_dirty = btrfs_set_page_dirty,
4964 };
4965
4966 static struct address_space_operations btrfs_symlink_aops = {
4967         .readpage       = btrfs_readpage,
4968         .writepage      = btrfs_writepage,
4969         .invalidatepage = btrfs_invalidatepage,
4970         .releasepage    = btrfs_releasepage,
4971 };
4972
4973 static struct inode_operations btrfs_file_inode_operations = {
4974         .truncate       = btrfs_truncate,
4975         .getattr        = btrfs_getattr,
4976         .setattr        = btrfs_setattr,
4977         .setxattr       = btrfs_setxattr,
4978         .getxattr       = btrfs_getxattr,
4979         .listxattr      = btrfs_listxattr,
4980         .removexattr    = btrfs_removexattr,
4981         .permission     = btrfs_permission,
4982         .fallocate      = btrfs_fallocate,
4983 };
4984 static struct inode_operations btrfs_special_inode_operations = {
4985         .getattr        = btrfs_getattr,
4986         .setattr        = btrfs_setattr,
4987         .permission     = btrfs_permission,
4988         .setxattr       = btrfs_setxattr,
4989         .getxattr       = btrfs_getxattr,
4990         .listxattr      = btrfs_listxattr,
4991         .removexattr    = btrfs_removexattr,
4992 };
4993 static struct inode_operations btrfs_symlink_inode_operations = {
4994         .readlink       = generic_readlink,
4995         .follow_link    = page_follow_link_light,
4996         .put_link       = page_put_link,
4997         .permission     = btrfs_permission,
4998 };