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/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "volumes.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
  97                                 struct btrfs_root *root, struct inode *inode);
  98
  99 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 100                                      struct inode *inode,  struct inode *dir,
 101                                      const struct qstr *qstr)
 102 {
 103         int err;
 104
 105         err = btrfs_init_acl(trans, inode, dir);
 106         if (!err)
 107                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 108         return err;
 109 }
 110
 111 /*
 112  * this does all the hard work for inserting an inline extent into
 113  * the btree.  The caller should have done a btrfs_drop_extents so that
 114  * no overlapping inline items exist in the btree
 115  */
 116 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 117                                 struct btrfs_root *root, struct inode *inode,
 118                                 u64 start, size_t size, size_t compressed_size,
 119                                 int compress_type,
 120                                 struct page **compressed_pages)
 121 {
 122         struct btrfs_key key;
 123         struct btrfs_path *path;
 124         struct extent_buffer *leaf;
 125         struct page *page = NULL;
 126         char *kaddr;
 127         unsigned long ptr;
 128         struct btrfs_file_extent_item *ei;
 129         int err = 0;
 130         int ret;
 131         size_t cur_size = size;
 132         size_t datasize;
 133         unsigned long offset;
 134
 135         if (compressed_size && compressed_pages)
 136                 cur_size = compressed_size;
 137
 138         path = btrfs_alloc_path();
 139         if (!path)
 140                 return -ENOMEM;
 141
 142         path->leave_spinning = 1;
 143
 144         key.objectid = btrfs_ino(inode);
 145         key.offset = start;
 146         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 147         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 148
 149         inode_add_bytes(inode, size);
 150         ret = btrfs_insert_empty_item(trans, root, path, &key,
 151                                       datasize);
 152         BUG_ON(ret);
 153         if (ret) {
 154                 err = ret;
 155                 goto fail;
 156         }
 157         leaf = path->nodes[0];
 158         ei = btrfs_item_ptr(leaf, path->slots[0],
 159                             struct btrfs_file_extent_item);
 160         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 161         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 162         btrfs_set_file_extent_encryption(leaf, ei, 0);
 163         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 164         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 165         ptr = btrfs_file_extent_inline_start(ei);
 166
 167         if (compress_type != BTRFS_COMPRESS_NONE) {
 168                 struct page *cpage;
 169                 int i = 0;
 170                 while (compressed_size > 0) {
 171                         cpage = compressed_pages[i];
 172                         cur_size = min_t(unsigned long, compressed_size,
 173                                        PAGE_CACHE_SIZE);
 174
 175                         kaddr = kmap_atomic(cpage, KM_USER0);
 176                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 177                         kunmap_atomic(kaddr, KM_USER0);
 178
 179                         i++;
 180                         ptr += cur_size;
 181                         compressed_size -= cur_size;
 182                 }
 183                 btrfs_set_file_extent_compression(leaf, ei,
 184                                                   compress_type);
 185         } else {
 186                 page = find_get_page(inode->i_mapping,
 187                                      start >> PAGE_CACHE_SHIFT);
 188                 btrfs_set_file_extent_compression(leaf, ei, 0);
 189                 kaddr = kmap_atomic(page, KM_USER0);
 190                 offset = start & (PAGE_CACHE_SIZE - 1);
 191                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 192                 kunmap_atomic(kaddr, KM_USER0);
 193                 page_cache_release(page);
 194         }
 195         btrfs_mark_buffer_dirty(leaf);
 196         btrfs_free_path(path);
 197
 198         /*
 199          * we're an inline extent, so nobody can
 200          * extend the file past i_size without locking
 201          * a page we already have locked.
 202          *
 203          * We must do any isize and inode updates
 204          * before we unlock the pages.  Otherwise we
 205          * could end up racing with unlink.
 206          */
 207         BTRFS_I(inode)->disk_i_size = inode->i_size;
 208         btrfs_update_inode(trans, root, inode);
 209
 210         return 0;
 211 fail:
 212         btrfs_free_path(path);
 213         return err;
 214 }
 215
 216
 217 /*
 218  * conditionally insert an inline extent into the file.  This
 219  * does the checks required to make sure the data is small enough
 220  * to fit as an inline extent.
 221  */
 222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 223                                  struct btrfs_root *root,
 224                                  struct inode *inode, u64 start, u64 end,
 225                                  size_t compressed_size, int compress_type,
 226                                  struct page **compressed_pages)
 227 {
 228         u64 isize = i_size_read(inode);
 229         u64 actual_end = min(end + 1, isize);
 230         u64 inline_len = actual_end - start;
 231         u64 aligned_end = (end + root->sectorsize - 1) &
 232                         ~((u64)root->sectorsize - 1);
 233         u64 hint_byte;
 234         u64 data_len = inline_len;
 235         int ret;
 236
 237         if (compressed_size)
 238                 data_len = compressed_size;
 239
 240         if (start > 0 ||
 241             actual_end >= PAGE_CACHE_SIZE ||
 242             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 243             (!compressed_size &&
 244             (actual_end & (root->sectorsize - 1)) == 0) ||
 245             end + 1 < isize ||
 246             data_len > root->fs_info->max_inline) {
 247                 return 1;
 248         }
 249
 250         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 251                                  &hint_byte, 1);
 252         BUG_ON(ret);
 253
 254         if (isize > actual_end)
 255                 inline_len = min_t(u64, isize, actual_end);
 256         ret = insert_inline_extent(trans, root, inode, start,
 257                                    inline_len, compressed_size,
 258                                    compress_type, compressed_pages);
 259         BUG_ON(ret);
 260         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 261         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 262         return 0;
 263 }
 264
 265 struct async_extent {
 266         u64 start;
 267         u64 ram_size;
 268         u64 compressed_size;
 269         struct page **pages;
 270         unsigned long nr_pages;
 271         int compress_type;
 272         struct list_head list;
 273 };
 274
 275 struct async_cow {
 276         struct inode *inode;
 277         struct btrfs_root *root;
 278         struct page *locked_page;
 279         u64 start;
 280         u64 end;
 281         struct list_head extents;
 282         struct btrfs_work work;
 283 };
 284
 285 static noinline int add_async_extent(struct async_cow *cow,
 286                                      u64 start, u64 ram_size,
 287                                      u64 compressed_size,
 288                                      struct page **pages,
 289                                      unsigned long nr_pages,
 290                                      int compress_type)
 291 {
 292         struct async_extent *async_extent;
 293
 294         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 295         BUG_ON(!async_extent);
 296         async_extent->start = start;
 297         async_extent->ram_size = ram_size;
 298         async_extent->compressed_size = compressed_size;
 299         async_extent->pages = pages;
 300         async_extent->nr_pages = nr_pages;
 301         async_extent->compress_type = compress_type;
 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 blocksize = root->sectorsize;
 332         u64 actual_end;
 333         u64 isize = i_size_read(inode);
 334         int ret = 0;
 335         struct page **pages = NULL;
 336         unsigned long nr_pages;
 337         unsigned long nr_pages_ret = 0;
 338         unsigned long total_compressed = 0;
 339         unsigned long total_in = 0;
 340         unsigned long max_compressed = 128 * 1024;
 341         unsigned long max_uncompressed = 128 * 1024;
 342         int i;
 343         int will_compress;
 344         int compress_type = root->fs_info->compress_type;
 345
 346         /* if this is a small write inside eof, kick off a defragbot */
 347         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 348                 btrfs_add_inode_defrag(NULL, inode);
 349
 350         actual_end = min_t(u64, isize, end + 1);
 351 again:
 352         will_compress = 0;
 353         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 354         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 355
 356         /*
 357          * we don't want to send crud past the end of i_size through
 358          * compression, that's just a waste of CPU time.  So, if the
 359          * end of the file is before the start of our current
 360          * requested range of bytes, we bail out to the uncompressed
 361          * cleanup code that can deal with all of this.
 362          *
 363          * It isn't really the fastest way to fix things, but this is a
 364          * very uncommon corner.
 365          */
 366         if (actual_end <= start)
 367                 goto cleanup_and_bail_uncompressed;
 368
 369         total_compressed = actual_end - start;
 370
 371         /* we want to make sure that amount of ram required to uncompress
 372          * an extent is reasonable, so we limit the total size in ram
 373          * of a compressed extent to 128k.  This is a crucial number
 374          * because it also controls how easily we can spread reads across
 375          * cpus for decompression.
 376          *
 377          * We also want to make sure the amount of IO required to do
 378          * a random read is reasonably small, so we limit the size of
 379          * a compressed extent to 128k.
 380          */
 381         total_compressed = min(total_compressed, max_uncompressed);
 382         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 383         num_bytes = max(blocksize,  num_bytes);
 384         total_in = 0;
 385         ret = 0;
 386
 387         /*
 388          * we do compression for mount -o compress and when the
 389          * inode has not been flagged as nocompress.  This flag can
 390          * change at any time if we discover bad compression ratios.
 391          */
 392         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 393             (btrfs_test_opt(root, COMPRESS) ||
 394              (BTRFS_I(inode)->force_compress) ||
 395              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 396                 WARN_ON(pages);
 397                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 398                 if (!pages) {
 399                         /* just bail out to the uncompressed code */
 400                         goto cont;
 401                 }
 402
 403                 if (BTRFS_I(inode)->force_compress)
 404                         compress_type = BTRFS_I(inode)->force_compress;
 405
 406                 ret = btrfs_compress_pages(compress_type,
 407                                            inode->i_mapping, start,
 408                                            total_compressed, pages,
 409                                            nr_pages, &nr_pages_ret,
 410                                            &total_in,
 411                                            &total_compressed,
 412                                            max_compressed);
 413
 414                 if (!ret) {
 415                         unsigned long offset = total_compressed &
 416                                 (PAGE_CACHE_SIZE - 1);
 417                         struct page *page = pages[nr_pages_ret - 1];
 418                         char *kaddr;
 419
 420                         /* zero the tail end of the last page, we might be
 421                          * sending it down to disk
 422                          */
 423                         if (offset) {
 424                                 kaddr = kmap_atomic(page, KM_USER0);
 425                                 memset(kaddr + offset, 0,
 426                                        PAGE_CACHE_SIZE - offset);
 427                                 kunmap_atomic(kaddr, KM_USER0);
 428                         }
 429                         will_compress = 1;
 430                 }
 431         }
 432 cont:
 433         if (start == 0) {
 434                 trans = btrfs_join_transaction(root);
 435                 BUG_ON(IS_ERR(trans));
 436                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 437
 438                 /* lets try to make an inline extent */
 439                 if (ret || total_in < (actual_end - start)) {
 440                         /* we didn't compress the entire range, try
 441                          * to make an uncompressed inline extent.
 442                          */
 443                         ret = cow_file_range_inline(trans, root, inode,
 444                                                     start, end, 0, 0, NULL);
 445                 } else {
 446                         /* try making a compressed inline extent */
 447                         ret = cow_file_range_inline(trans, root, inode,
 448                                                     start, end,
 449                                                     total_compressed,
 450                                                     compress_type, pages);
 451                 }
 452                 if (ret == 0) {
 453                         /*
 454                          * inline extent creation worked, we don't need
 455                          * to create any more async work items.  Unlock
 456                          * and free up our temp pages.
 457                          */
 458                         extent_clear_unlock_delalloc(inode,
 459                              &BTRFS_I(inode)->io_tree,
 460                              start, end, NULL,
 461                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 462                              EXTENT_CLEAR_DELALLOC |
 463                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 464
 465                         btrfs_end_transaction(trans, root);
 466                         goto free_pages_out;
 467                 }
 468                 btrfs_end_transaction(trans, root);
 469         }
 470
 471         if (will_compress) {
 472                 /*
 473                  * we aren't doing an inline extent round the compressed size
 474                  * up to a block size boundary so the allocator does sane
 475                  * things
 476                  */
 477                 total_compressed = (total_compressed + blocksize - 1) &
 478                         ~(blocksize - 1);
 479
 480                 /*
 481                  * one last check to make sure the compression is really a
 482                  * win, compare the page count read with the blocks on disk
 483                  */
 484                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 485                         ~(PAGE_CACHE_SIZE - 1);
 486                 if (total_compressed >= total_in) {
 487                         will_compress = 0;
 488                 } else {
 489                         num_bytes = total_in;
 490                 }
 491         }
 492         if (!will_compress && pages) {
 493                 /*
 494                  * the compression code ran but failed to make things smaller,
 495                  * free any pages it allocated and our page pointer array
 496                  */
 497                 for (i = 0; i < nr_pages_ret; i++) {
 498                         WARN_ON(pages[i]->mapping);
 499                         page_cache_release(pages[i]);
 500                 }
 501                 kfree(pages);
 502                 pages = NULL;
 503                 total_compressed = 0;
 504                 nr_pages_ret = 0;
 505
 506                 /* flag the file so we don't compress in the future */
 507                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 508                     !(BTRFS_I(inode)->force_compress)) {
 509                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 510                 }
 511         }
 512         if (will_compress) {
 513                 *num_added += 1;
 514
 515                 /* the async work queues will take care of doing actual
 516                  * allocation on disk for these compressed pages,
 517                  * and will submit them to the elevator.
 518                  */
 519                 add_async_extent(async_cow, start, num_bytes,
 520                                  total_compressed, pages, nr_pages_ret,
 521                                  compress_type);
 522
 523                 if (start + num_bytes < end) {
 524                         start += num_bytes;
 525                         pages = NULL;
 526                         cond_resched();
 527                         goto again;
 528                 }
 529         } else {
 530 cleanup_and_bail_uncompressed:
 531                 /*
 532                  * No compression, but we still need to write the pages in
 533                  * the file we've been given so far.  redirty the locked
 534                  * page if it corresponds to our extent and set things up
 535                  * for the async work queue to run cow_file_range to do
 536                  * the normal delalloc dance
 537                  */
 538                 if (page_offset(locked_page) >= start &&
 539                     page_offset(locked_page) <= end) {
 540                         __set_page_dirty_nobuffers(locked_page);
 541                         /* unlocked later on in the async handlers */
 542                 }
 543                 add_async_extent(async_cow, start, end - start + 1,
 544                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 545                 *num_added += 1;
 546         }
 547
 548 out:
 549         return 0;
 550
 551 free_pages_out:
 552         for (i = 0; i < nr_pages_ret; i++) {
 553                 WARN_ON(pages[i]->mapping);
 554                 page_cache_release(pages[i]);
 555         }
 556         kfree(pages);
 557
 558         goto out;
 559 }
 560
 561 /*
 562  * phase two of compressed writeback.  This is the ordered portion
 563  * of the code, which only gets called in the order the work was
 564  * queued.  We walk all the async extents created by compress_file_range
 565  * and send them down to the disk.
 566  */
 567 static noinline int submit_compressed_extents(struct inode *inode,
 568                                               struct async_cow *async_cow)
 569 {
 570         struct async_extent *async_extent;
 571         u64 alloc_hint = 0;
 572         struct btrfs_trans_handle *trans;
 573         struct btrfs_key ins;
 574         struct extent_map *em;
 575         struct btrfs_root *root = BTRFS_I(inode)->root;
 576         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 577         struct extent_io_tree *io_tree;
 578         int ret = 0;
 579
 580         if (list_empty(&async_cow->extents))
 581                 return 0;
 582
 583
 584         while (!list_empty(&async_cow->extents)) {
 585                 async_extent = list_entry(async_cow->extents.next,
 586                                           struct async_extent, list);
 587                 list_del(&async_extent->list);
 588
 589                 io_tree = &BTRFS_I(inode)->io_tree;
 590
 591 retry:
 592                 /* did the compression code fall back to uncompressed IO? */
 593                 if (!async_extent->pages) {
 594                         int page_started = 0;
 595                         unsigned long nr_written = 0;
 596
 597                         lock_extent(io_tree, async_extent->start,
 598                                          async_extent->start +
 599                                          async_extent->ram_size - 1, GFP_NOFS);
 600
 601                         /* allocate blocks */
 602                         ret = cow_file_range(inode, async_cow->locked_page,
 603                                              async_extent->start,
 604                                              async_extent->start +
 605                                              async_extent->ram_size - 1,
 606                                              &page_started, &nr_written, 0);
 607
 608                         /*
 609                          * if page_started, cow_file_range inserted an
 610                          * inline extent and took care of all the unlocking
 611                          * and IO for us.  Otherwise, we need to submit
 612                          * all those pages down to the drive.
 613                          */
 614                         if (!page_started && !ret)
 615                                 extent_write_locked_range(io_tree,
 616                                                   inode, async_extent->start,
 617                                                   async_extent->start +
 618                                                   async_extent->ram_size - 1,
 619                                                   btrfs_get_extent,
 620                                                   WB_SYNC_ALL);
 621                         kfree(async_extent);
 622                         cond_resched();
 623                         continue;
 624                 }
 625
 626                 lock_extent(io_tree, async_extent->start,
 627                             async_extent->start + async_extent->ram_size - 1,
 628                             GFP_NOFS);
 629
 630                 trans = btrfs_join_transaction(root);
 631                 BUG_ON(IS_ERR(trans));
 632                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 633                 ret = btrfs_reserve_extent(trans, root,
 634                                            async_extent->compressed_size,
 635                                            async_extent->compressed_size,
 636                                            0, alloc_hint,
 637                                            (u64)-1, &ins, 1);
 638                 btrfs_end_transaction(trans, root);
 639
 640                 if (ret) {
 641                         int i;
 642                         for (i = 0; i < async_extent->nr_pages; i++) {
 643                                 WARN_ON(async_extent->pages[i]->mapping);
 644                                 page_cache_release(async_extent->pages[i]);
 645                         }
 646                         kfree(async_extent->pages);
 647                         async_extent->nr_pages = 0;
 648                         async_extent->pages = NULL;
 649                         unlock_extent(io_tree, async_extent->start,
 650                                       async_extent->start +
 651                                       async_extent->ram_size - 1, GFP_NOFS);
 652                         goto retry;
 653                 }
 654
 655                 /*
 656                  * here we're doing allocation and writeback of the
 657                  * compressed pages
 658                  */
 659                 btrfs_drop_extent_cache(inode, async_extent->start,
 660                                         async_extent->start +
 661                                         async_extent->ram_size - 1, 0);
 662
 663                 em = alloc_extent_map();
 664                 BUG_ON(!em);
 665                 em->start = async_extent->start;
 666                 em->len = async_extent->ram_size;
 667                 em->orig_start = em->start;
 668
 669                 em->block_start = ins.objectid;
 670                 em->block_len = ins.offset;
 671                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 672                 em->compress_type = async_extent->compress_type;
 673                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 674                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 675
 676                 while (1) {
 677                         write_lock(&em_tree->lock);
 678                         ret = add_extent_mapping(em_tree, em);
 679                         write_unlock(&em_tree->lock);
 680                         if (ret != -EEXIST) {
 681                                 free_extent_map(em);
 682                                 break;
 683                         }
 684                         btrfs_drop_extent_cache(inode, async_extent->start,
 685                                                 async_extent->start +
 686                                                 async_extent->ram_size - 1, 0);
 687                 }
 688
 689                 ret = btrfs_add_ordered_extent_compress(inode,
 690                                                 async_extent->start,
 691                                                 ins.objectid,
 692                                                 async_extent->ram_size,
 693                                                 ins.offset,
 694                                                 BTRFS_ORDERED_COMPRESSED,
 695                                                 async_extent->compress_type);
 696                 BUG_ON(ret);
 697
 698                 /*
 699                  * clear dirty, set writeback and unlock the pages.
 700                  */
 701                 extent_clear_unlock_delalloc(inode,
 702                                 &BTRFS_I(inode)->io_tree,
 703                                 async_extent->start,
 704                                 async_extent->start +
 705                                 async_extent->ram_size - 1,
 706                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 707                                 EXTENT_CLEAR_UNLOCK |
 708                                 EXTENT_CLEAR_DELALLOC |
 709                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 710
 711                 ret = btrfs_submit_compressed_write(inode,
 712                                     async_extent->start,
 713                                     async_extent->ram_size,
 714                                     ins.objectid,
 715                                     ins.offset, async_extent->pages,
 716                                     async_extent->nr_pages);
 717
 718                 BUG_ON(ret);
 719                 alloc_hint = ins.objectid + ins.offset;
 720                 kfree(async_extent);
 721                 cond_resched();
 722         }
 723
 724         return 0;
 725 }
 726
 727 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 728                                       u64 num_bytes)
 729 {
 730         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 731         struct extent_map *em;
 732         u64 alloc_hint = 0;
 733
 734         read_lock(&em_tree->lock);
 735         em = search_extent_mapping(em_tree, start, num_bytes);
 736         if (em) {
 737                 /*
 738                  * if block start isn't an actual block number then find the
 739                  * first block in this inode and use that as a hint.  If that
 740                  * block is also bogus then just don't worry about it.
 741                  */
 742                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 743                         free_extent_map(em);
 744                         em = search_extent_mapping(em_tree, 0, 0);
 745                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 746                                 alloc_hint = em->block_start;
 747                         if (em)
 748                                 free_extent_map(em);
 749                 } else {
 750                         alloc_hint = em->block_start;
 751                         free_extent_map(em);
 752                 }
 753         }
 754         read_unlock(&em_tree->lock);
 755
 756         return alloc_hint;
 757 }
 758
 759 /*
 760  * when extent_io.c finds a delayed allocation range in the file,
 761  * the call backs end up in this code.  The basic idea is to
 762  * allocate extents on disk for the range, and create ordered data structs
 763  * in ram to track those extents.
 764  *
 765  * locked_page is the page that writepage had locked already.  We use
 766  * it to make sure we don't do extra locks or unlocks.
 767  *
 768  * *page_started is set to one if we unlock locked_page and do everything
 769  * required to start IO on it.  It may be clean and already done with
 770  * IO when we return.
 771  */
 772 static noinline int cow_file_range(struct inode *inode,
 773                                    struct page *locked_page,
 774                                    u64 start, u64 end, int *page_started,
 775                                    unsigned long *nr_written,
 776                                    int unlock)
 777 {
 778         struct btrfs_root *root = BTRFS_I(inode)->root;
 779         struct btrfs_trans_handle *trans;
 780         u64 alloc_hint = 0;
 781         u64 num_bytes;
 782         unsigned long ram_size;
 783         u64 disk_num_bytes;
 784         u64 cur_alloc_size;
 785         u64 blocksize = root->sectorsize;
 786         struct btrfs_key ins;
 787         struct extent_map *em;
 788         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 789         int ret = 0;
 790
 791         BUG_ON(btrfs_is_free_space_inode(root, inode));
 792         trans = btrfs_join_transaction(root);
 793         BUG_ON(IS_ERR(trans));
 794         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 795
 796         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 797         num_bytes = max(blocksize,  num_bytes);
 798         disk_num_bytes = num_bytes;
 799         ret = 0;
 800
 801         /* if this is a small write inside eof, kick off defrag */
 802         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 803                 btrfs_add_inode_defrag(trans, inode);
 804
 805         if (start == 0) {
 806                 /* lets try to make an inline extent */
 807                 ret = cow_file_range_inline(trans, root, inode,
 808                                             start, end, 0, 0, NULL);
 809                 if (ret == 0) {
 810                         extent_clear_unlock_delalloc(inode,
 811                                      &BTRFS_I(inode)->io_tree,
 812                                      start, end, NULL,
 813                                      EXTENT_CLEAR_UNLOCK_PAGE |
 814                                      EXTENT_CLEAR_UNLOCK |
 815                                      EXTENT_CLEAR_DELALLOC |
 816                                      EXTENT_CLEAR_DIRTY |
 817                                      EXTENT_SET_WRITEBACK |
 818                                      EXTENT_END_WRITEBACK);
 819
 820                         *nr_written = *nr_written +
 821                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 822                         *page_started = 1;
 823                         ret = 0;
 824                         goto out;
 825                 }
 826         }
 827
 828         BUG_ON(disk_num_bytes >
 829                btrfs_super_total_bytes(root->fs_info->super_copy));
 830
 831         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 832         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 833
 834         while (disk_num_bytes > 0) {
 835                 unsigned long op;
 836
 837                 cur_alloc_size = disk_num_bytes;
 838                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 839                                            root->sectorsize, 0, alloc_hint,
 840                                            (u64)-1, &ins, 1);
 841                 BUG_ON(ret);
 842
 843                 em = alloc_extent_map();
 844                 BUG_ON(!em);
 845                 em->start = start;
 846                 em->orig_start = em->start;
 847                 ram_size = ins.offset;
 848                 em->len = ins.offset;
 849
 850                 em->block_start = ins.objectid;
 851                 em->block_len = ins.offset;
 852                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 853                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 854
 855                 while (1) {
 856                         write_lock(&em_tree->lock);
 857                         ret = add_extent_mapping(em_tree, em);
 858                         write_unlock(&em_tree->lock);
 859                         if (ret != -EEXIST) {
 860                                 free_extent_map(em);
 861                                 break;
 862                         }
 863                         btrfs_drop_extent_cache(inode, start,
 864                                                 start + ram_size - 1, 0);
 865                 }
 866
 867                 cur_alloc_size = ins.offset;
 868                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 869                                                ram_size, cur_alloc_size, 0);
 870                 BUG_ON(ret);
 871
 872                 if (root->root_key.objectid ==
 873                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 874                         ret = btrfs_reloc_clone_csums(inode, start,
 875                                                       cur_alloc_size);
 876                         BUG_ON(ret);
 877                 }
 878
 879                 if (disk_num_bytes < cur_alloc_size)
 880                         break;
 881
 882                 /* we're not doing compressed IO, don't unlock the first
 883                  * page (which the caller expects to stay locked), don't
 884                  * clear any dirty bits and don't set any writeback bits
 885                  *
 886                  * Do set the Private2 bit so we know this page was properly
 887                  * setup for writepage
 888                  */
 889                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 890                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 891                         EXTENT_SET_PRIVATE2;
 892
 893                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 894                                              start, start + ram_size - 1,
 895                                              locked_page, op);
 896                 disk_num_bytes -= cur_alloc_size;
 897                 num_bytes -= cur_alloc_size;
 898                 alloc_hint = ins.objectid + ins.offset;
 899                 start += cur_alloc_size;
 900         }
 901 out:
 902         ret = 0;
 903         btrfs_end_transaction(trans, root);
 904
 905         return ret;
 906 }
 907
 908 /*
 909  * work queue call back to started compression on a file and pages
 910  */
 911 static noinline void async_cow_start(struct btrfs_work *work)
 912 {
 913         struct async_cow *async_cow;
 914         int num_added = 0;
 915         async_cow = container_of(work, struct async_cow, work);
 916
 917         compress_file_range(async_cow->inode, async_cow->locked_page,
 918                             async_cow->start, async_cow->end, async_cow,
 919                             &num_added);
 920         if (num_added == 0)
 921                 async_cow->inode = NULL;
 922 }
 923
 924 /*
 925  * work queue call back to submit previously compressed pages
 926  */
 927 static noinline void async_cow_submit(struct btrfs_work *work)
 928 {
 929         struct async_cow *async_cow;
 930         struct btrfs_root *root;
 931         unsigned long nr_pages;
 932
 933         async_cow = container_of(work, struct async_cow, work);
 934
 935         root = async_cow->root;
 936         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 937                 PAGE_CACHE_SHIFT;
 938
 939         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 940
 941         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 942             5 * 1042 * 1024 &&
 943             waitqueue_active(&root->fs_info->async_submit_wait))
 944                 wake_up(&root->fs_info->async_submit_wait);
 945
 946         if (async_cow->inode)
 947                 submit_compressed_extents(async_cow->inode, async_cow);
 948 }
 949
 950 static noinline void async_cow_free(struct btrfs_work *work)
 951 {
 952         struct async_cow *async_cow;
 953         async_cow = container_of(work, struct async_cow, work);
 954         kfree(async_cow);
 955 }
 956
 957 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 958                                 u64 start, u64 end, int *page_started,
 959                                 unsigned long *nr_written)
 960 {
 961         struct async_cow *async_cow;
 962         struct btrfs_root *root = BTRFS_I(inode)->root;
 963         unsigned long nr_pages;
 964         u64 cur_end;
 965         int limit = 10 * 1024 * 1042;
 966
 967         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 968                          1, 0, NULL, GFP_NOFS);
 969         while (start < end) {
 970                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 971                 BUG_ON(!async_cow);
 972                 async_cow->inode = inode;
 973                 async_cow->root = root;
 974                 async_cow->locked_page = locked_page;
 975                 async_cow->start = start;
 976
 977                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 978                         cur_end = end;
 979                 else
 980                         cur_end = min(end, start + 512 * 1024 - 1);
 981
 982                 async_cow->end = cur_end;
 983                 INIT_LIST_HEAD(&async_cow->extents);
 984
 985                 async_cow->work.func = async_cow_start;
 986                 async_cow->work.ordered_func = async_cow_submit;
 987                 async_cow->work.ordered_free = async_cow_free;
 988                 async_cow->work.flags = 0;
 989
 990                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 991                         PAGE_CACHE_SHIFT;
 992                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 993
 994                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 995                                    &async_cow->work);
 996
 997                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 998                         wait_event(root->fs_info->async_submit_wait,
 999                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1000                             limit));
1001                 }
1002
1003                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1004                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1005                         wait_event(root->fs_info->async_submit_wait,
1006                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1007                            0));
1008                 }
1009
1010                 *nr_written += nr_pages;
1011                 start = cur_end + 1;
1012         }
1013         *page_started = 1;
1014         return 0;
1015 }
1016
1017 static noinline int csum_exist_in_range(struct btrfs_root *root,
1018                                         u64 bytenr, u64 num_bytes)
1019 {
1020         int ret;
1021         struct btrfs_ordered_sum *sums;
1022         LIST_HEAD(list);
1023
1024         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1025                                        bytenr + num_bytes - 1, &list, 0);
1026         if (ret == 0 && list_empty(&list))
1027                 return 0;
1028
1029         while (!list_empty(&list)) {
1030                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1031                 list_del(&sums->list);
1032                 kfree(sums);
1033         }
1034         return 1;
1035 }
1036
1037 /*
1038  * when nowcow writeback call back.  This checks for snapshots or COW copies
1039  * of the extents that exist in the file, and COWs the file as required.
1040  *
1041  * If no cow copies or snapshots exist, we write directly to the existing
1042  * blocks on disk
1043  */
1044 static noinline int run_delalloc_nocow(struct inode *inode,
1045                                        struct page *locked_page,
1046                               u64 start, u64 end, int *page_started, int force,
1047                               unsigned long *nr_written)
1048 {
1049         struct btrfs_root *root = BTRFS_I(inode)->root;
1050         struct btrfs_trans_handle *trans;
1051         struct extent_buffer *leaf;
1052         struct btrfs_path *path;
1053         struct btrfs_file_extent_item *fi;
1054         struct btrfs_key found_key;
1055         u64 cow_start;
1056         u64 cur_offset;
1057         u64 extent_end;
1058         u64 extent_offset;
1059         u64 disk_bytenr;
1060         u64 num_bytes;
1061         int extent_type;
1062         int ret;
1063         int type;
1064         int nocow;
1065         int check_prev = 1;
1066         bool nolock;
1067         u64 ino = btrfs_ino(inode);
1068
1069         path = btrfs_alloc_path();
1070         if (!path)
1071                 return -ENOMEM;
1072
1073         nolock = btrfs_is_free_space_inode(root, inode);
1074
1075         if (nolock)
1076                 trans = btrfs_join_transaction_nolock(root);
1077         else
1078                 trans = btrfs_join_transaction(root);
1079
1080         BUG_ON(IS_ERR(trans));
1081         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1082
1083         cow_start = (u64)-1;
1084         cur_offset = start;
1085         while (1) {
1086                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1087                                                cur_offset, 0);
1088                 BUG_ON(ret < 0);
1089                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1090                         leaf = path->nodes[0];
1091                         btrfs_item_key_to_cpu(leaf, &found_key,
1092                                               path->slots[0] - 1);
1093                         if (found_key.objectid == ino &&
1094                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1095                                 path->slots[0]--;
1096                 }
1097                 check_prev = 0;
1098 next_slot:
1099                 leaf = path->nodes[0];
1100                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1101                         ret = btrfs_next_leaf(root, path);
1102                         if (ret < 0)
1103                                 BUG_ON(1);
1104                         if (ret > 0)
1105                                 break;
1106                         leaf = path->nodes[0];
1107                 }
1108
1109                 nocow = 0;
1110                 disk_bytenr = 0;
1111                 num_bytes = 0;
1112                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1113
1114                 if (found_key.objectid > ino ||
1115                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1116                     found_key.offset > end)
1117                         break;
1118
1119                 if (found_key.offset > cur_offset) {
1120                         extent_end = found_key.offset;
1121                         extent_type = 0;
1122                         goto out_check;
1123                 }
1124
1125                 fi = btrfs_item_ptr(leaf, path->slots[0],
1126                                     struct btrfs_file_extent_item);
1127                 extent_type = btrfs_file_extent_type(leaf, fi);
1128
1129                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1130                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1131                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1132                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1133                         extent_end = found_key.offset +
1134                                 btrfs_file_extent_num_bytes(leaf, fi);
1135                         if (extent_end <= start) {
1136                                 path->slots[0]++;
1137                                 goto next_slot;
1138                         }
1139                         if (disk_bytenr == 0)
1140                                 goto out_check;
1141                         if (btrfs_file_extent_compression(leaf, fi) ||
1142                             btrfs_file_extent_encryption(leaf, fi) ||
1143                             btrfs_file_extent_other_encoding(leaf, fi))
1144                                 goto out_check;
1145                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1146                                 goto out_check;
1147                         if (btrfs_extent_readonly(root, disk_bytenr))
1148                                 goto out_check;
1149                         if (btrfs_cross_ref_exist(trans, root, ino,
1150                                                   found_key.offset -
1151                                                   extent_offset, disk_bytenr))
1152                                 goto out_check;
1153                         disk_bytenr += extent_offset;
1154                         disk_bytenr += cur_offset - found_key.offset;
1155                         num_bytes = min(end + 1, extent_end) - cur_offset;
1156                         /*
1157                          * force cow if csum exists in the range.
1158                          * this ensure that csum for a given extent are
1159                          * either valid or do not exist.
1160                          */
1161                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1162                                 goto out_check;
1163                         nocow = 1;
1164                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1165                         extent_end = found_key.offset +
1166                                 btrfs_file_extent_inline_len(leaf, fi);
1167                         extent_end = ALIGN(extent_end, root->sectorsize);
1168                 } else {
1169                         BUG_ON(1);
1170                 }
1171 out_check:
1172                 if (extent_end <= start) {
1173                         path->slots[0]++;
1174                         goto next_slot;
1175                 }
1176                 if (!nocow) {
1177                         if (cow_start == (u64)-1)
1178                                 cow_start = cur_offset;
1179                         cur_offset = extent_end;
1180                         if (cur_offset > end)
1181                                 break;
1182                         path->slots[0]++;
1183                         goto next_slot;
1184                 }
1185
1186                 btrfs_release_path(path);
1187                 if (cow_start != (u64)-1) {
1188                         ret = cow_file_range(inode, locked_page, cow_start,
1189                                         found_key.offset - 1, page_started,
1190                                         nr_written, 1);
1191                         BUG_ON(ret);
1192                         cow_start = (u64)-1;
1193                 }
1194
1195                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1196                         struct extent_map *em;
1197                         struct extent_map_tree *em_tree;
1198                         em_tree = &BTRFS_I(inode)->extent_tree;
1199                         em = alloc_extent_map();
1200                         BUG_ON(!em);
1201                         em->start = cur_offset;
1202                         em->orig_start = em->start;
1203                         em->len = num_bytes;
1204                         em->block_len = num_bytes;
1205                         em->block_start = disk_bytenr;
1206                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1207                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1208                         while (1) {
1209                                 write_lock(&em_tree->lock);
1210                                 ret = add_extent_mapping(em_tree, em);
1211                                 write_unlock(&em_tree->lock);
1212                                 if (ret != -EEXIST) {
1213                                         free_extent_map(em);
1214                                         break;
1215                                 }
1216                                 btrfs_drop_extent_cache(inode, em->start,
1217                                                 em->start + em->len - 1, 0);
1218                         }
1219                         type = BTRFS_ORDERED_PREALLOC;
1220                 } else {
1221                         type = BTRFS_ORDERED_NOCOW;
1222                 }
1223
1224                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1225                                                num_bytes, num_bytes, type);
1226                 BUG_ON(ret);
1227
1228                 if (root->root_key.objectid ==
1229                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1230                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1231                                                       num_bytes);
1232                         BUG_ON(ret);
1233                 }
1234
1235                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1236                                 cur_offset, cur_offset + num_bytes - 1,
1237                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1238                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1239                                 EXTENT_SET_PRIVATE2);
1240                 cur_offset = extent_end;
1241                 if (cur_offset > end)
1242                         break;
1243         }
1244         btrfs_release_path(path);
1245
1246         if (cur_offset <= end && cow_start == (u64)-1)
1247                 cow_start = cur_offset;
1248         if (cow_start != (u64)-1) {
1249                 ret = cow_file_range(inode, locked_page, cow_start, end,
1250                                      page_started, nr_written, 1);
1251                 BUG_ON(ret);
1252         }
1253
1254         if (nolock) {
1255                 ret = btrfs_end_transaction_nolock(trans, root);
1256                 BUG_ON(ret);
1257         } else {
1258                 ret = btrfs_end_transaction(trans, root);
1259                 BUG_ON(ret);
1260         }
1261         btrfs_free_path(path);
1262         return 0;
1263 }
1264
1265 /*
1266  * extent_io.c call back to do delayed allocation processing
1267  */
1268 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1269                               u64 start, u64 end, int *page_started,
1270                               unsigned long *nr_written)
1271 {
1272         int ret;
1273         struct btrfs_root *root = BTRFS_I(inode)->root;
1274
1275         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1276                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1277                                          page_started, 1, nr_written);
1278         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1279                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1280                                          page_started, 0, nr_written);
1281         else if (!btrfs_test_opt(root, COMPRESS) &&
1282                  !(BTRFS_I(inode)->force_compress) &&
1283                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1284                 ret = cow_file_range(inode, locked_page, start, end,
1285                                       page_started, nr_written, 1);
1286         else
1287                 ret = cow_file_range_async(inode, locked_page, start, end,
1288                                            page_started, nr_written);
1289         return ret;
1290 }
1291
1292 static void btrfs_split_extent_hook(struct inode *inode,
1293                                     struct extent_state *orig, u64 split)
1294 {
1295         /* not delalloc, ignore it */
1296         if (!(orig->state & EXTENT_DELALLOC))
1297                 return;
1298
1299         spin_lock(&BTRFS_I(inode)->lock);
1300         BTRFS_I(inode)->outstanding_extents++;
1301         spin_unlock(&BTRFS_I(inode)->lock);
1302 }
1303
1304 /*
1305  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1306  * extents so we can keep track of new extents that are just merged onto old
1307  * extents, such as when we are doing sequential writes, so we can properly
1308  * account for the metadata space we'll need.
1309  */
1310 static void btrfs_merge_extent_hook(struct inode *inode,
1311                                     struct extent_state *new,
1312                                     struct extent_state *other)
1313 {
1314         /* not delalloc, ignore it */
1315         if (!(other->state & EXTENT_DELALLOC))
1316                 return;
1317
1318         spin_lock(&BTRFS_I(inode)->lock);
1319         BTRFS_I(inode)->outstanding_extents--;
1320         spin_unlock(&BTRFS_I(inode)->lock);
1321 }
1322
1323 /*
1324  * extent_io.c set_bit_hook, used to track delayed allocation
1325  * bytes in this file, and to maintain the list of inodes that
1326  * have pending delalloc work to be done.
1327  */
1328 static void btrfs_set_bit_hook(struct inode *inode,
1329                                struct extent_state *state, int *bits)
1330 {
1331
1332         /*
1333          * set_bit and clear bit hooks normally require _irqsave/restore
1334          * but in this case, we are only testing for the DELALLOC
1335          * bit, which is only set or cleared with irqs on
1336          */
1337         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338                 struct btrfs_root *root = BTRFS_I(inode)->root;
1339                 u64 len = state->end + 1 - state->start;
1340                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1341
1342                 if (*bits & EXTENT_FIRST_DELALLOC) {
1343                         *bits &= ~EXTENT_FIRST_DELALLOC;
1344                 } else {
1345                         spin_lock(&BTRFS_I(inode)->lock);
1346                         BTRFS_I(inode)->outstanding_extents++;
1347                         spin_unlock(&BTRFS_I(inode)->lock);
1348                 }
1349
1350                 spin_lock(&root->fs_info->delalloc_lock);
1351                 BTRFS_I(inode)->delalloc_bytes += len;
1352                 root->fs_info->delalloc_bytes += len;
1353                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1354                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1355                                       &root->fs_info->delalloc_inodes);
1356                 }
1357                 spin_unlock(&root->fs_info->delalloc_lock);
1358         }
1359 }
1360
1361 /*
1362  * extent_io.c clear_bit_hook, see set_bit_hook for why
1363  */
1364 static void btrfs_clear_bit_hook(struct inode *inode,
1365                                  struct extent_state *state, int *bits)
1366 {
1367         /*
1368          * set_bit and clear bit hooks normally require _irqsave/restore
1369          * but in this case, we are only testing for the DELALLOC
1370          * bit, which is only set or cleared with irqs on
1371          */
1372         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1373                 struct btrfs_root *root = BTRFS_I(inode)->root;
1374                 u64 len = state->end + 1 - state->start;
1375                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1376
1377                 if (*bits & EXTENT_FIRST_DELALLOC) {
1378                         *bits &= ~EXTENT_FIRST_DELALLOC;
1379                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1380                         spin_lock(&BTRFS_I(inode)->lock);
1381                         BTRFS_I(inode)->outstanding_extents--;
1382                         spin_unlock(&BTRFS_I(inode)->lock);
1383                 }
1384
1385                 if (*bits & EXTENT_DO_ACCOUNTING)
1386                         btrfs_delalloc_release_metadata(inode, len);
1387
1388                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1389                     && do_list)
1390                         btrfs_free_reserved_data_space(inode, len);
1391
1392                 spin_lock(&root->fs_info->delalloc_lock);
1393                 root->fs_info->delalloc_bytes -= len;
1394                 BTRFS_I(inode)->delalloc_bytes -= len;
1395
1396                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1397                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1398                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1399                 }
1400                 spin_unlock(&root->fs_info->delalloc_lock);
1401         }
1402 }
1403
1404 /*
1405  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1406  * we don't create bios that span stripes or chunks
1407  */
1408 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1409                          size_t size, struct bio *bio,
1410                          unsigned long bio_flags)
1411 {
1412         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1413         struct btrfs_mapping_tree *map_tree;
1414         u64 logical = (u64)bio->bi_sector << 9;
1415         u64 length = 0;
1416         u64 map_length;
1417         int ret;
1418
1419         if (bio_flags & EXTENT_BIO_COMPRESSED)
1420                 return 0;
1421
1422         length = bio->bi_size;
1423         map_tree = &root->fs_info->mapping_tree;
1424         map_length = length;
1425         ret = btrfs_map_block(map_tree, READ, logical,
1426                               &map_length, NULL, 0);
1427
1428         if (map_length < length + size)
1429                 return 1;
1430         return ret;
1431 }
1432
1433 /*
1434  * in order to insert checksums into the metadata in large chunks,
1435  * we wait until bio submission time.   All the pages in the bio are
1436  * checksummed and sums are attached onto the ordered extent record.
1437  *
1438  * At IO completion time the cums attached on the ordered extent record
1439  * are inserted into the btree
1440  */
1441 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1442                                     struct bio *bio, int mirror_num,
1443                                     unsigned long bio_flags,
1444                                     u64 bio_offset)
1445 {
1446         struct btrfs_root *root = BTRFS_I(inode)->root;
1447         int ret = 0;
1448
1449         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1450         BUG_ON(ret);
1451         return 0;
1452 }
1453
1454 /*
1455  * in order to insert checksums into the metadata in large chunks,
1456  * we wait until bio submission time.   All the pages in the bio are
1457  * checksummed and sums are attached onto the ordered extent record.
1458  *
1459  * At IO completion time the cums attached on the ordered extent record
1460  * are inserted into the btree
1461  */
1462 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1463                           int mirror_num, unsigned long bio_flags,
1464                           u64 bio_offset)
1465 {
1466         struct btrfs_root *root = BTRFS_I(inode)->root;
1467         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1468 }
1469
1470 /*
1471  * extent_io.c submission hook. This does the right thing for csum calculation
1472  * on write, or reading the csums from the tree before a read
1473  */
1474 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1475                           int mirror_num, unsigned long bio_flags,
1476                           u64 bio_offset)
1477 {
1478         struct btrfs_root *root = BTRFS_I(inode)->root;
1479         int ret = 0;
1480         int skip_sum;
1481
1482         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1483
1484         if (btrfs_is_free_space_inode(root, inode))
1485                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1486         else
1487                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1488         BUG_ON(ret);
1489
1490         if (!(rw & REQ_WRITE)) {
1491                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1492                         return btrfs_submit_compressed_read(inode, bio,
1493                                                     mirror_num, bio_flags);
1494                 } else if (!skip_sum) {
1495                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1496                         if (ret)
1497                                 return ret;
1498                 }
1499                 goto mapit;
1500         } else if (!skip_sum) {
1501                 /* csum items have already been cloned */
1502                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1503                         goto mapit;
1504                 /* we're doing a write, do the async checksumming */
1505                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1506                                    inode, rw, bio, mirror_num,
1507                                    bio_flags, bio_offset,
1508                                    __btrfs_submit_bio_start,
1509                                    __btrfs_submit_bio_done);
1510         }
1511
1512 mapit:
1513         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1514 }
1515
1516 /*
1517  * given a list of ordered sums record them in the inode.  This happens
1518  * at IO completion time based on sums calculated at bio submission time.
1519  */
1520 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1521                              struct inode *inode, u64 file_offset,
1522                              struct list_head *list)
1523 {
1524         struct btrfs_ordered_sum *sum;
1525
1526         list_for_each_entry(sum, list, list) {
1527                 btrfs_csum_file_blocks(trans,
1528                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1529         }
1530         return 0;
1531 }
1532
1533 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1534                               struct extent_state **cached_state)
1535 {
1536         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1537                 WARN_ON(1);
1538         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1539                                    cached_state, GFP_NOFS);
1540 }
1541
1542 /* see btrfs_writepage_start_hook for details on why this is required */
1543 struct btrfs_writepage_fixup {
1544         struct page *page;
1545         struct btrfs_work work;
1546 };
1547
1548 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1549 {
1550         struct btrfs_writepage_fixup *fixup;
1551         struct btrfs_ordered_extent *ordered;
1552         struct extent_state *cached_state = NULL;
1553         struct page *page;
1554         struct inode *inode;
1555         u64 page_start;
1556         u64 page_end;
1557
1558         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1559         page = fixup->page;
1560 again:
1561         lock_page(page);
1562         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1563                 ClearPageChecked(page);
1564                 goto out_page;
1565         }
1566
1567         inode = page->mapping->host;
1568         page_start = page_offset(page);
1569         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1570
1571         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1572                          &cached_state, GFP_NOFS);
1573
1574         /* already ordered? We're done */
1575         if (PagePrivate2(page))
1576                 goto out;
1577
1578         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1579         if (ordered) {
1580                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1581                                      page_end, &cached_state, GFP_NOFS);
1582                 unlock_page(page);
1583                 btrfs_start_ordered_extent(inode, ordered, 1);
1584                 goto again;
1585         }
1586
1587         BUG();
1588         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1589         ClearPageChecked(page);
1590 out:
1591         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1592                              &cached_state, GFP_NOFS);
1593 out_page:
1594         unlock_page(page);
1595         page_cache_release(page);
1596         kfree(fixup);
1597 }
1598
1599 /*
1600  * There are a few paths in the higher layers of the kernel that directly
1601  * set the page dirty bit without asking the filesystem if it is a
1602  * good idea.  This causes problems because we want to make sure COW
1603  * properly happens and the data=ordered rules are followed.
1604  *
1605  * In our case any range that doesn't have the ORDERED bit set
1606  * hasn't been properly setup for IO.  We kick off an async process
1607  * to fix it up.  The async helper will wait for ordered extents, set
1608  * the delalloc bit and make it safe to write the page.
1609  */
1610 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1611 {
1612         struct inode *inode = page->mapping->host;
1613         struct btrfs_writepage_fixup *fixup;
1614         struct btrfs_root *root = BTRFS_I(inode)->root;
1615
1616         /* this page is properly in the ordered list */
1617         if (TestClearPagePrivate2(page))
1618                 return 0;
1619
1620         if (PageChecked(page))
1621                 return -EAGAIN;
1622
1623         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1624         if (!fixup)
1625                 return -EAGAIN;
1626
1627         SetPageChecked(page);
1628         page_cache_get(page);
1629         fixup->work.func = btrfs_writepage_fixup_worker;
1630         fixup->page = page;
1631         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1632         return -EAGAIN;
1633 }
1634
1635 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1636                                        struct inode *inode, u64 file_pos,
1637                                        u64 disk_bytenr, u64 disk_num_bytes,
1638                                        u64 num_bytes, u64 ram_bytes,
1639                                        u8 compression, u8 encryption,
1640                                        u16 other_encoding, int extent_type)
1641 {
1642         struct btrfs_root *root = BTRFS_I(inode)->root;
1643         struct btrfs_file_extent_item *fi;
1644         struct btrfs_path *path;
1645         struct extent_buffer *leaf;
1646         struct btrfs_key ins;
1647         u64 hint;
1648         int ret;
1649
1650         path = btrfs_alloc_path();
1651         if (!path)
1652                 return -ENOMEM;
1653
1654         path->leave_spinning = 1;
1655
1656         /*
1657          * we may be replacing one extent in the tree with another.
1658          * The new extent is pinned in the extent map, and we don't want
1659          * to drop it from the cache until it is completely in the btree.
1660          *
1661          * So, tell btrfs_drop_extents to leave this extent in the cache.
1662          * the caller is expected to unpin it and allow it to be merged
1663          * with the others.
1664          */
1665         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1666                                  &hint, 0);
1667         BUG_ON(ret);
1668
1669         ins.objectid = btrfs_ino(inode);
1670         ins.offset = file_pos;
1671         ins.type = BTRFS_EXTENT_DATA_KEY;
1672         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1673         BUG_ON(ret);
1674         leaf = path->nodes[0];
1675         fi = btrfs_item_ptr(leaf, path->slots[0],
1676                             struct btrfs_file_extent_item);
1677         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1678         btrfs_set_file_extent_type(leaf, fi, extent_type);
1679         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1680         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1681         btrfs_set_file_extent_offset(leaf, fi, 0);
1682         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1683         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1684         btrfs_set_file_extent_compression(leaf, fi, compression);
1685         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1686         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1687
1688         btrfs_unlock_up_safe(path, 1);
1689         btrfs_set_lock_blocking(leaf);
1690
1691         btrfs_mark_buffer_dirty(leaf);
1692
1693         inode_add_bytes(inode, num_bytes);
1694
1695         ins.objectid = disk_bytenr;
1696         ins.offset = disk_num_bytes;
1697         ins.type = BTRFS_EXTENT_ITEM_KEY;
1698         ret = btrfs_alloc_reserved_file_extent(trans, root,
1699                                         root->root_key.objectid,
1700                                         btrfs_ino(inode), file_pos, &ins);
1701         BUG_ON(ret);
1702         btrfs_free_path(path);
1703
1704         return 0;
1705 }
1706
1707 /*
1708  * helper function for btrfs_finish_ordered_io, this
1709  * just reads in some of the csum leaves to prime them into ram
1710  * before we start the transaction.  It limits the amount of btree
1711  * reads required while inside the transaction.
1712  */
1713 /* as ordered data IO finishes, this gets called so we can finish
1714  * an ordered extent if the range of bytes in the file it covers are
1715  * fully written.
1716  */
1717 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1718 {
1719         struct btrfs_root *root = BTRFS_I(inode)->root;
1720         struct btrfs_trans_handle *trans = NULL;
1721         struct btrfs_ordered_extent *ordered_extent = NULL;
1722         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1723         struct extent_state *cached_state = NULL;
1724         int compress_type = 0;
1725         int ret;
1726         bool nolock;
1727
1728         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1729                                              end - start + 1);
1730         if (!ret)
1731                 return 0;
1732         BUG_ON(!ordered_extent);
1733
1734         nolock = btrfs_is_free_space_inode(root, inode);
1735
1736         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1737                 BUG_ON(!list_empty(&ordered_extent->list));
1738                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1739                 if (!ret) {
1740                         if (nolock)
1741                                 trans = btrfs_join_transaction_nolock(root);
1742                         else
1743                                 trans = btrfs_join_transaction(root);
1744                         BUG_ON(IS_ERR(trans));
1745                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1746                         ret = btrfs_update_inode_fallback(trans, root, inode);
1747                         BUG_ON(ret);
1748                 }
1749                 goto out;
1750         }
1751
1752         lock_extent_bits(io_tree, ordered_extent->file_offset,
1753                          ordered_extent->file_offset + ordered_extent->len - 1,
1754                          0, &cached_state, GFP_NOFS);
1755
1756         if (nolock)
1757                 trans = btrfs_join_transaction_nolock(root);
1758         else
1759                 trans = btrfs_join_transaction(root);
1760         BUG_ON(IS_ERR(trans));
1761         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1762
1763         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1764                 compress_type = ordered_extent->compress_type;
1765         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1766                 BUG_ON(compress_type);
1767                 ret = btrfs_mark_extent_written(trans, inode,
1768                                                 ordered_extent->file_offset,
1769                                                 ordered_extent->file_offset +
1770                                                 ordered_extent->len);
1771                 BUG_ON(ret);
1772         } else {
1773                 BUG_ON(root == root->fs_info->tree_root);
1774                 ret = insert_reserved_file_extent(trans, inode,
1775                                                 ordered_extent->file_offset,
1776                                                 ordered_extent->start,
1777                                                 ordered_extent->disk_len,
1778                                                 ordered_extent->len,
1779                                                 ordered_extent->len,
1780                                                 compress_type, 0, 0,
1781                                                 BTRFS_FILE_EXTENT_REG);
1782                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1783                                    ordered_extent->file_offset,
1784                                    ordered_extent->len);
1785                 BUG_ON(ret);
1786         }
1787         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1788                              ordered_extent->file_offset +
1789                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1790
1791         add_pending_csums(trans, inode, ordered_extent->file_offset,
1792                           &ordered_extent->list);
1793
1794         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1795         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1796                 ret = btrfs_update_inode_fallback(trans, root, inode);
1797                 BUG_ON(ret);
1798         }
1799         ret = 0;
1800 out:
1801         if (root != root->fs_info->tree_root)
1802                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1803         if (trans) {
1804                 if (nolock)
1805                         btrfs_end_transaction_nolock(trans, root);
1806                 else
1807                         btrfs_end_transaction(trans, root);
1808         }
1809
1810         /* once for us */
1811         btrfs_put_ordered_extent(ordered_extent);
1812         /* once for the tree */
1813         btrfs_put_ordered_extent(ordered_extent);
1814
1815         return 0;
1816 }
1817
1818 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1819                                 struct extent_state *state, int uptodate)
1820 {
1821         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1822
1823         ClearPagePrivate2(page);
1824         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1825 }
1826
1827 /*
1828  * when reads are done, we need to check csums to verify the data is correct
1829  * if there's a match, we allow the bio to finish.  If not, the code in
1830  * extent_io.c will try to find good copies for us.
1831  */
1832 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1833                                struct extent_state *state)
1834 {
1835         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1836         struct inode *inode = page->mapping->host;
1837         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1838         char *kaddr;
1839         u64 private = ~(u32)0;
1840         int ret;
1841         struct btrfs_root *root = BTRFS_I(inode)->root;
1842         u32 csum = ~(u32)0;
1843
1844         if (PageChecked(page)) {
1845                 ClearPageChecked(page);
1846                 goto good;
1847         }
1848
1849         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1850                 goto good;
1851
1852         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1853             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1854                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1855                                   GFP_NOFS);
1856                 return 0;
1857         }
1858
1859         if (state && state->start == start) {
1860                 private = state->private;
1861                 ret = 0;
1862         } else {
1863                 ret = get_state_private(io_tree, start, &private);
1864         }
1865         kaddr = kmap_atomic(page, KM_USER0);
1866         if (ret)
1867                 goto zeroit;
1868
1869         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1870         btrfs_csum_final(csum, (char *)&csum);
1871         if (csum != private)
1872                 goto zeroit;
1873
1874         kunmap_atomic(kaddr, KM_USER0);
1875 good:
1876         return 0;
1877
1878 zeroit:
1879         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1880                        "private %llu\n",
1881                        (unsigned long long)btrfs_ino(page->mapping->host),
1882                        (unsigned long long)start, csum,
1883                        (unsigned long long)private);
1884         memset(kaddr + offset, 1, end - start + 1);
1885         flush_dcache_page(page);
1886         kunmap_atomic(kaddr, KM_USER0);
1887         if (private == 0)
1888                 return 0;
1889         return -EIO;
1890 }
1891
1892 struct delayed_iput {
1893         struct list_head list;
1894         struct inode *inode;
1895 };
1896
1897 void btrfs_add_delayed_iput(struct inode *inode)
1898 {
1899         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1900         struct delayed_iput *delayed;
1901
1902         if (atomic_add_unless(&inode->i_count, -1, 1))
1903                 return;
1904
1905         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1906         delayed->inode = inode;
1907
1908         spin_lock(&fs_info->delayed_iput_lock);
1909         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1910         spin_unlock(&fs_info->delayed_iput_lock);
1911 }
1912
1913 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1914 {
1915         LIST_HEAD(list);
1916         struct btrfs_fs_info *fs_info = root->fs_info;
1917         struct delayed_iput *delayed;
1918         int empty;
1919
1920         spin_lock(&fs_info->delayed_iput_lock);
1921         empty = list_empty(&fs_info->delayed_iputs);
1922         spin_unlock(&fs_info->delayed_iput_lock);
1923         if (empty)
1924                 return;
1925
1926         down_read(&root->fs_info->cleanup_work_sem);
1927         spin_lock(&fs_info->delayed_iput_lock);
1928         list_splice_init(&fs_info->delayed_iputs, &list);
1929         spin_unlock(&fs_info->delayed_iput_lock);
1930
1931         while (!list_empty(&list)) {
1932                 delayed = list_entry(list.next, struct delayed_iput, list);
1933                 list_del(&delayed->list);
1934                 iput(delayed->inode);
1935                 kfree(delayed);
1936         }
1937         up_read(&root->fs_info->cleanup_work_sem);
1938 }
1939
1940 enum btrfs_orphan_cleanup_state {
1941         ORPHAN_CLEANUP_STARTED  = 1,
1942         ORPHAN_CLEANUP_DONE     = 2,
1943 };
1944
1945 /*
1946  * This is called in transaction commmit time. If there are no orphan
1947  * files in the subvolume, it removes orphan item and frees block_rsv
1948  * structure.
1949  */
1950 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1951                               struct btrfs_root *root)
1952 {
1953         int ret;
1954
1955         if (!list_empty(&root->orphan_list) ||
1956             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1957                 return;
1958
1959         if (root->orphan_item_inserted &&
1960             btrfs_root_refs(&root->root_item) > 0) {
1961                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1962                                             root->root_key.objectid);
1963                 BUG_ON(ret);
1964                 root->orphan_item_inserted = 0;
1965         }
1966
1967         if (root->orphan_block_rsv) {
1968                 WARN_ON(root->orphan_block_rsv->size > 0);
1969                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
1970                 root->orphan_block_rsv = NULL;
1971         }
1972 }
1973
1974 /*
1975  * This creates an orphan entry for the given inode in case something goes
1976  * wrong in the middle of an unlink/truncate.
1977  *
1978  * NOTE: caller of this function should reserve 5 units of metadata for
1979  *       this function.
1980  */
1981 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1982 {
1983         struct btrfs_root *root = BTRFS_I(inode)->root;
1984         struct btrfs_block_rsv *block_rsv = NULL;
1985         int reserve = 0;
1986         int insert = 0;
1987         int ret;
1988
1989         if (!root->orphan_block_rsv) {
1990                 block_rsv = btrfs_alloc_block_rsv(root);
1991                 if (!block_rsv)
1992                         return -ENOMEM;
1993         }
1994
1995         spin_lock(&root->orphan_lock);
1996         if (!root->orphan_block_rsv) {
1997                 root->orphan_block_rsv = block_rsv;
1998         } else if (block_rsv) {
1999                 btrfs_free_block_rsv(root, block_rsv);
2000                 block_rsv = NULL;
2001         }
2002
2003         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2004                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2005 #if 0
2006                 /*
2007                  * For proper ENOSPC handling, we should do orphan
2008                  * cleanup when mounting. But this introduces backward
2009                  * compatibility issue.
2010                  */
2011                 if (!xchg(&root->orphan_item_inserted, 1))
2012                         insert = 2;
2013                 else
2014                         insert = 1;
2015 #endif
2016                 insert = 1;
2017         }
2018
2019         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2020                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2021                 reserve = 1;
2022         }
2023         spin_unlock(&root->orphan_lock);
2024
2025         /* grab metadata reservation from transaction handle */
2026         if (reserve) {
2027                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2028                 BUG_ON(ret);
2029         }
2030
2031         /* insert an orphan item to track this unlinked/truncated file */
2032         if (insert >= 1) {
2033                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2034                 BUG_ON(ret);
2035         }
2036
2037         /* insert an orphan item to track subvolume contains orphan files */
2038         if (insert >= 2) {
2039                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2040                                                root->root_key.objectid);
2041                 BUG_ON(ret);
2042         }
2043         return 0;
2044 }
2045
2046 /*
2047  * We have done the truncate/delete so we can go ahead and remove the orphan
2048  * item for this particular inode.
2049  */
2050 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2051 {
2052         struct btrfs_root *root = BTRFS_I(inode)->root;
2053         int delete_item = 0;
2054         int release_rsv = 0;
2055         int ret = 0;
2056
2057         spin_lock(&root->orphan_lock);
2058         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2059                 list_del_init(&BTRFS_I(inode)->i_orphan);
2060                 delete_item = 1;
2061         }
2062
2063         if (BTRFS_I(inode)->orphan_meta_reserved) {
2064                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2065                 release_rsv = 1;
2066         }
2067         spin_unlock(&root->orphan_lock);
2068
2069         if (trans && delete_item) {
2070                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2071                 BUG_ON(ret);
2072         }
2073
2074         if (release_rsv)
2075                 btrfs_orphan_release_metadata(inode);
2076
2077         return 0;
2078 }
2079
2080 /*
2081  * this cleans up any orphans that may be left on the list from the last use
2082  * of this root.
2083  */
2084 int btrfs_orphan_cleanup(struct btrfs_root *root)
2085 {
2086         struct btrfs_path *path;
2087         struct extent_buffer *leaf;
2088         struct btrfs_key key, found_key;
2089         struct btrfs_trans_handle *trans;
2090         struct inode *inode;
2091         u64 last_objectid = 0;
2092         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2093
2094         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2095                 return 0;
2096
2097         path = btrfs_alloc_path();
2098         if (!path) {
2099                 ret = -ENOMEM;
2100                 goto out;
2101         }
2102         path->reada = -1;
2103
2104         key.objectid = BTRFS_ORPHAN_OBJECTID;
2105         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2106         key.offset = (u64)-1;
2107
2108         while (1) {
2109                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2110                 if (ret < 0)
2111                         goto out;
2112
2113                 /*
2114                  * if ret == 0 means we found what we were searching for, which
2115                  * is weird, but possible, so only screw with path if we didn't
2116                  * find the key and see if we have stuff that matches
2117                  */
2118                 if (ret > 0) {
2119                         ret = 0;
2120                         if (path->slots[0] == 0)
2121                                 break;
2122                         path->slots[0]--;
2123                 }
2124
2125                 /* pull out the item */
2126                 leaf = path->nodes[0];
2127                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2128
2129                 /* make sure the item matches what we want */
2130                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2131                         break;
2132                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2133                         break;
2134
2135                 /* release the path since we're done with it */
2136                 btrfs_release_path(path);
2137
2138                 /*
2139                  * this is where we are basically btrfs_lookup, without the
2140                  * crossing root thing.  we store the inode number in the
2141                  * offset of the orphan item.
2142                  */
2143
2144                 if (found_key.offset == last_objectid) {
2145                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2146                                "stopping orphan cleanup\n");
2147                         ret = -EINVAL;
2148                         goto out;
2149                 }
2150
2151                 last_objectid = found_key.offset;
2152
2153                 found_key.objectid = found_key.offset;
2154                 found_key.type = BTRFS_INODE_ITEM_KEY;
2155                 found_key.offset = 0;
2156                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2157                 ret = PTR_RET(inode);
2158                 if (ret && ret != -ESTALE)
2159                         goto out;
2160
2161                 /*
2162                  * Inode is already gone but the orphan item is still there,
2163                  * kill the orphan item.
2164                  */
2165                 if (ret == -ESTALE) {
2166                         trans = btrfs_start_transaction(root, 1);
2167                         if (IS_ERR(trans)) {
2168                                 ret = PTR_ERR(trans);
2169                                 goto out;
2170                         }
2171                         ret = btrfs_del_orphan_item(trans, root,
2172                                                     found_key.objectid);
2173                         BUG_ON(ret);
2174                         btrfs_end_transaction(trans, root);
2175                         continue;
2176                 }
2177
2178                 /*
2179                  * add this inode to the orphan list so btrfs_orphan_del does
2180                  * the proper thing when we hit it
2181                  */
2182                 spin_lock(&root->orphan_lock);
2183                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2184                 spin_unlock(&root->orphan_lock);
2185
2186                 /* if we have links, this was a truncate, lets do that */
2187                 if (inode->i_nlink) {
2188                         if (!S_ISREG(inode->i_mode)) {
2189                                 WARN_ON(1);
2190                                 iput(inode);
2191                                 continue;
2192                         }
2193                         nr_truncate++;
2194                         ret = btrfs_truncate(inode);
2195                 } else {
2196                         nr_unlink++;
2197                 }
2198
2199                 /* this will do delete_inode and everything for us */
2200                 iput(inode);
2201                 if (ret)
2202                         goto out;
2203         }
2204         /* release the path since we're done with it */
2205         btrfs_release_path(path);
2206
2207         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2208
2209         if (root->orphan_block_rsv)
2210                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2211                                         (u64)-1);
2212
2213         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2214                 trans = btrfs_join_transaction(root);
2215                 if (!IS_ERR(trans))
2216                         btrfs_end_transaction(trans, root);
2217         }
2218
2219         if (nr_unlink)
2220                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2221         if (nr_truncate)
2222                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2223
2224 out:
2225         if (ret)
2226                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2227         btrfs_free_path(path);
2228         return ret;
2229 }
2230
2231 /*
2232  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2233  * don't find any xattrs, we know there can't be any acls.
2234  *
2235  * slot is the slot the inode is in, objectid is the objectid of the inode
2236  */
2237 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2238                                           int slot, u64 objectid)
2239 {
2240         u32 nritems = btrfs_header_nritems(leaf);
2241         struct btrfs_key found_key;
2242         int scanned = 0;
2243
2244         slot++;
2245         while (slot < nritems) {
2246                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2247
2248                 /* we found a different objectid, there must not be acls */
2249                 if (found_key.objectid != objectid)
2250                         return 0;
2251
2252                 /* we found an xattr, assume we've got an acl */
2253                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2254                         return 1;
2255
2256                 /*
2257                  * we found a key greater than an xattr key, there can't
2258                  * be any acls later on
2259                  */
2260                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2261                         return 0;
2262
2263                 slot++;
2264                 scanned++;
2265
2266                 /*
2267                  * it goes inode, inode backrefs, xattrs, extents,
2268                  * so if there are a ton of hard links to an inode there can
2269                  * be a lot of backrefs.  Don't waste time searching too hard,
2270                  * this is just an optimization
2271                  */
2272                 if (scanned >= 8)
2273                         break;
2274         }
2275         /* we hit the end of the leaf before we found an xattr or
2276          * something larger than an xattr.  We have to assume the inode
2277          * has acls
2278          */
2279         return 1;
2280 }
2281
2282 /*
2283  * read an inode from the btree into the in-memory inode
2284  */
2285 static void btrfs_read_locked_inode(struct inode *inode)
2286 {
2287         struct btrfs_path *path;
2288         struct extent_buffer *leaf;
2289         struct btrfs_inode_item *inode_item;
2290         struct btrfs_timespec *tspec;
2291         struct btrfs_root *root = BTRFS_I(inode)->root;
2292         struct btrfs_key location;
2293         int maybe_acls;
2294         u32 rdev;
2295         int ret;
2296         bool filled = false;
2297
2298         ret = btrfs_fill_inode(inode, &rdev);
2299         if (!ret)
2300                 filled = true;
2301
2302         path = btrfs_alloc_path();
2303         if (!path)
2304                 goto make_bad;
2305
2306         path->leave_spinning = 1;
2307         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2308
2309         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2310         if (ret)
2311                 goto make_bad;
2312
2313         leaf = path->nodes[0];
2314
2315         if (filled)
2316                 goto cache_acl;
2317
2318         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2319                                     struct btrfs_inode_item);
2320         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2321         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2322         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2323         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2324         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2325
2326         tspec = btrfs_inode_atime(inode_item);
2327         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2328         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2329
2330         tspec = btrfs_inode_mtime(inode_item);
2331         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2332         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2333
2334         tspec = btrfs_inode_ctime(inode_item);
2335         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2336         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2337
2338         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2339         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2340         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2341         inode->i_generation = BTRFS_I(inode)->generation;
2342         inode->i_rdev = 0;
2343         rdev = btrfs_inode_rdev(leaf, inode_item);
2344
2345         BTRFS_I(inode)->index_cnt = (u64)-1;
2346         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2347 cache_acl:
2348         /*
2349          * try to precache a NULL acl entry for files that don't have
2350          * any xattrs or acls
2351          */
2352         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2353                                            btrfs_ino(inode));
2354         if (!maybe_acls)
2355                 cache_no_acl(inode);
2356
2357         btrfs_free_path(path);
2358
2359         switch (inode->i_mode & S_IFMT) {
2360         case S_IFREG:
2361                 inode->i_mapping->a_ops = &btrfs_aops;
2362                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2363                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2364                 inode->i_fop = &btrfs_file_operations;
2365                 inode->i_op = &btrfs_file_inode_operations;
2366                 break;
2367         case S_IFDIR:
2368                 inode->i_fop = &btrfs_dir_file_operations;
2369                 if (root == root->fs_info->tree_root)
2370                         inode->i_op = &btrfs_dir_ro_inode_operations;
2371                 else
2372                         inode->i_op = &btrfs_dir_inode_operations;
2373                 break;
2374         case S_IFLNK:
2375                 inode->i_op = &btrfs_symlink_inode_operations;
2376                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2377                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2378                 break;
2379         default:
2380                 inode->i_op = &btrfs_special_inode_operations;
2381                 init_special_inode(inode, inode->i_mode, rdev);
2382                 break;
2383         }
2384
2385         btrfs_update_iflags(inode);
2386         return;
2387
2388 make_bad:
2389         btrfs_free_path(path);
2390         make_bad_inode(inode);
2391 }
2392
2393 /*
2394  * given a leaf and an inode, copy the inode fields into the leaf
2395  */
2396 static void fill_inode_item(struct btrfs_trans_handle *trans,
2397                             struct extent_buffer *leaf,
2398                             struct btrfs_inode_item *item,
2399                             struct inode *inode)
2400 {
2401         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2402         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2403         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2404         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2405         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2406
2407         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2408                                inode->i_atime.tv_sec);
2409         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2410                                 inode->i_atime.tv_nsec);
2411
2412         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2413                                inode->i_mtime.tv_sec);
2414         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2415                                 inode->i_mtime.tv_nsec);
2416
2417         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2418                                inode->i_ctime.tv_sec);
2419         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2420                                 inode->i_ctime.tv_nsec);
2421
2422         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2423         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2424         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2425         btrfs_set_inode_transid(leaf, item, trans->transid);
2426         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2427         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2428         btrfs_set_inode_block_group(leaf, item, 0);
2429 }
2430
2431 /*
2432  * copy everything in the in-memory inode into the btree.
2433  */
2434 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2435                                 struct btrfs_root *root, struct inode *inode)
2436 {
2437         struct btrfs_inode_item *inode_item;
2438         struct btrfs_path *path;
2439         struct extent_buffer *leaf;
2440         int ret;
2441
2442         path = btrfs_alloc_path();
2443         if (!path)
2444                 return -ENOMEM;
2445
2446         path->leave_spinning = 1;
2447         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2448                                  1);
2449         if (ret) {
2450                 if (ret > 0)
2451                         ret = -ENOENT;
2452                 goto failed;
2453         }
2454
2455         btrfs_unlock_up_safe(path, 1);
2456         leaf = path->nodes[0];
2457         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2458                                     struct btrfs_inode_item);
2459
2460         fill_inode_item(trans, leaf, inode_item, inode);
2461         btrfs_mark_buffer_dirty(leaf);
2462         btrfs_set_inode_last_trans(trans, inode);
2463         ret = 0;
2464 failed:
2465         btrfs_free_path(path);
2466         return ret;
2467 }
2468
2469 /*
2470  * copy everything in the in-memory inode into the btree.
2471  */
2472 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2473                                 struct btrfs_root *root, struct inode *inode)
2474 {
2475         int ret;
2476
2477         /*
2478          * If the inode is a free space inode, we can deadlock during commit
2479          * if we put it into the delayed code.
2480          *
2481          * The data relocation inode should also be directly updated
2482          * without delay
2483          */
2484         if (!btrfs_is_free_space_inode(root, inode)
2485             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2486                 ret = btrfs_delayed_update_inode(trans, root, inode);
2487                 if (!ret)
2488                         btrfs_set_inode_last_trans(trans, inode);
2489                 return ret;
2490         }
2491
2492         return btrfs_update_inode_item(trans, root, inode);
2493 }
2494
2495 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2496                                 struct btrfs_root *root, struct inode *inode)
2497 {
2498         int ret;
2499
2500         ret = btrfs_update_inode(trans, root, inode);
2501         if (ret == -ENOSPC)
2502                 return btrfs_update_inode_item(trans, root, inode);
2503         return ret;
2504 }
2505
2506 /*
2507  * unlink helper that gets used here in inode.c and in the tree logging
2508  * recovery code.  It remove a link in a directory with a given name, and
2509  * also drops the back refs in the inode to the directory
2510  */
2511 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2512                                 struct btrfs_root *root,
2513                                 struct inode *dir, struct inode *inode,
2514                                 const char *name, int name_len)
2515 {
2516         struct btrfs_path *path;
2517         int ret = 0;
2518         struct extent_buffer *leaf;
2519         struct btrfs_dir_item *di;
2520         struct btrfs_key key;
2521         u64 index;
2522         u64 ino = btrfs_ino(inode);
2523         u64 dir_ino = btrfs_ino(dir);
2524
2525         path = btrfs_alloc_path();
2526         if (!path) {
2527                 ret = -ENOMEM;
2528                 goto out;
2529         }
2530
2531         path->leave_spinning = 1;
2532         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2533                                     name, name_len, -1);
2534         if (IS_ERR(di)) {
2535                 ret = PTR_ERR(di);
2536                 goto err;
2537         }
2538         if (!di) {
2539                 ret = -ENOENT;
2540                 goto err;
2541         }
2542         leaf = path->nodes[0];
2543         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2544         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2545         if (ret)
2546                 goto err;
2547         btrfs_release_path(path);
2548
2549         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2550                                   dir_ino, &index);
2551         if (ret) {
2552                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2553                        "inode %llu parent %llu\n", name_len, name,
2554                        (unsigned long long)ino, (unsigned long long)dir_ino);
2555                 goto err;
2556         }
2557
2558         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2559         if (ret)
2560                 goto err;
2561
2562         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2563                                          inode, dir_ino);
2564         BUG_ON(ret != 0 && ret != -ENOENT);
2565
2566         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2567                                            dir, index);
2568         if (ret == -ENOENT)
2569                 ret = 0;
2570 err:
2571         btrfs_free_path(path);
2572         if (ret)
2573                 goto out;
2574
2575         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2576         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2577         btrfs_update_inode(trans, root, dir);
2578 out:
2579         return ret;
2580 }
2581
2582 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2583                        struct btrfs_root *root,
2584                        struct inode *dir, struct inode *inode,
2585                        const char *name, int name_len)
2586 {
2587         int ret;
2588         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2589         if (!ret) {
2590                 btrfs_drop_nlink(inode);
2591                 ret = btrfs_update_inode(trans, root, inode);
2592         }
2593         return ret;
2594 }
2595
2596
2597 /* helper to check if there is any shared block in the path */
2598 static int check_path_shared(struct btrfs_root *root,
2599                              struct btrfs_path *path)
2600 {
2601         struct extent_buffer *eb;
2602         int level;
2603         u64 refs = 1;
2604
2605         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2606                 int ret;
2607
2608                 if (!path->nodes[level])
2609                         break;
2610                 eb = path->nodes[level];
2611                 if (!btrfs_block_can_be_shared(root, eb))
2612                         continue;
2613                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2614                                                &refs, NULL);
2615                 if (refs > 1)
2616                         return 1;
2617         }
2618         return 0;
2619 }
2620
2621 /*
2622  * helper to start transaction for unlink and rmdir.
2623  *
2624  * unlink and rmdir are special in btrfs, they do not always free space.
2625  * so in enospc case, we should make sure they will free space before
2626  * allowing them to use the global metadata reservation.
2627  */
2628 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2629                                                        struct dentry *dentry)
2630 {
2631         struct btrfs_trans_handle *trans;
2632         struct btrfs_root *root = BTRFS_I(dir)->root;
2633         struct btrfs_path *path;
2634         struct btrfs_inode_ref *ref;
2635         struct btrfs_dir_item *di;
2636         struct inode *inode = dentry->d_inode;
2637         u64 index;
2638         int check_link = 1;
2639         int err = -ENOSPC;
2640         int ret;
2641         u64 ino = btrfs_ino(inode);
2642         u64 dir_ino = btrfs_ino(dir);
2643
2644         /*
2645          * 1 for the possible orphan item
2646          * 1 for the dir item
2647          * 1 for the dir index
2648          * 1 for the inode ref
2649          * 1 for the inode ref in the tree log
2650          * 2 for the dir entries in the log
2651          * 1 for the inode
2652          */
2653         trans = btrfs_start_transaction(root, 8);
2654         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2655                 return trans;
2656
2657         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2658                 return ERR_PTR(-ENOSPC);
2659
2660         /* check if there is someone else holds reference */
2661         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2662                 return ERR_PTR(-ENOSPC);
2663
2664         if (atomic_read(&inode->i_count) > 2)
2665                 return ERR_PTR(-ENOSPC);
2666
2667         if (xchg(&root->fs_info->enospc_unlink, 1))
2668                 return ERR_PTR(-ENOSPC);
2669
2670         path = btrfs_alloc_path();
2671         if (!path) {
2672                 root->fs_info->enospc_unlink = 0;
2673                 return ERR_PTR(-ENOMEM);
2674         }
2675
2676         /* 1 for the orphan item */
2677         trans = btrfs_start_transaction(root, 1);
2678         if (IS_ERR(trans)) {
2679                 btrfs_free_path(path);
2680                 root->fs_info->enospc_unlink = 0;
2681                 return trans;
2682         }
2683
2684         path->skip_locking = 1;
2685         path->search_commit_root = 1;
2686
2687         ret = btrfs_lookup_inode(trans, root, path,
2688                                 &BTRFS_I(dir)->location, 0);
2689         if (ret < 0) {
2690                 err = ret;
2691                 goto out;
2692         }
2693         if (ret == 0) {
2694                 if (check_path_shared(root, path))
2695                         goto out;
2696         } else {
2697                 check_link = 0;
2698         }
2699         btrfs_release_path(path);
2700
2701         ret = btrfs_lookup_inode(trans, root, path,
2702                                 &BTRFS_I(inode)->location, 0);
2703         if (ret < 0) {
2704                 err = ret;
2705                 goto out;
2706         }
2707         if (ret == 0) {
2708                 if (check_path_shared(root, path))
2709                         goto out;
2710         } else {
2711                 check_link = 0;
2712         }
2713         btrfs_release_path(path);
2714
2715         if (ret == 0 && S_ISREG(inode->i_mode)) {
2716                 ret = btrfs_lookup_file_extent(trans, root, path,
2717                                                ino, (u64)-1, 0);
2718                 if (ret < 0) {
2719                         err = ret;
2720                         goto out;
2721                 }
2722                 BUG_ON(ret == 0);
2723                 if (check_path_shared(root, path))
2724                         goto out;
2725                 btrfs_release_path(path);
2726         }
2727
2728         if (!check_link) {
2729                 err = 0;
2730                 goto out;
2731         }
2732
2733         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2734                                 dentry->d_name.name, dentry->d_name.len, 0);
2735         if (IS_ERR(di)) {
2736                 err = PTR_ERR(di);
2737                 goto out;
2738         }
2739         if (di) {
2740                 if (check_path_shared(root, path))
2741                         goto out;
2742         } else {
2743                 err = 0;
2744                 goto out;
2745         }
2746         btrfs_release_path(path);
2747
2748         ref = btrfs_lookup_inode_ref(trans, root, path,
2749                                 dentry->d_name.name, dentry->d_name.len,
2750                                 ino, dir_ino, 0);
2751         if (IS_ERR(ref)) {
2752                 err = PTR_ERR(ref);
2753                 goto out;
2754         }
2755         BUG_ON(!ref);
2756         if (check_path_shared(root, path))
2757                 goto out;
2758         index = btrfs_inode_ref_index(path->nodes[0], ref);
2759         btrfs_release_path(path);
2760
2761         /*
2762          * This is a commit root search, if we can lookup inode item and other
2763          * relative items in the commit root, it means the transaction of
2764          * dir/file creation has been committed, and the dir index item that we
2765          * delay to insert has also been inserted into the commit root. So
2766          * we needn't worry about the delayed insertion of the dir index item
2767          * here.
2768          */
2769         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2770                                 dentry->d_name.name, dentry->d_name.len, 0);
2771         if (IS_ERR(di)) {
2772                 err = PTR_ERR(di);
2773                 goto out;
2774         }
2775         BUG_ON(ret == -ENOENT);
2776         if (check_path_shared(root, path))
2777                 goto out;
2778
2779         err = 0;
2780 out:
2781         btrfs_free_path(path);
2782         /* Migrate the orphan reservation over */
2783         if (!err)
2784                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2785                                 &root->fs_info->global_block_rsv,
2786                                 trans->bytes_reserved);
2787
2788         if (err) {
2789                 btrfs_end_transaction(trans, root);
2790                 root->fs_info->enospc_unlink = 0;
2791                 return ERR_PTR(err);
2792         }
2793
2794         trans->block_rsv = &root->fs_info->global_block_rsv;
2795         return trans;
2796 }
2797
2798 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2799                                struct btrfs_root *root)
2800 {
2801         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2802                 btrfs_block_rsv_release(root, trans->block_rsv,
2803                                         trans->bytes_reserved);
2804                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2805                 BUG_ON(!root->fs_info->enospc_unlink);
2806                 root->fs_info->enospc_unlink = 0;
2807         }
2808         btrfs_end_transaction_throttle(trans, root);
2809 }
2810
2811 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2812 {
2813         struct btrfs_root *root = BTRFS_I(dir)->root;
2814         struct btrfs_trans_handle *trans;
2815         struct inode *inode = dentry->d_inode;
2816         int ret;
2817         unsigned long nr = 0;
2818
2819         trans = __unlink_start_trans(dir, dentry);
2820         if (IS_ERR(trans))
2821                 return PTR_ERR(trans);
2822
2823         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2824
2825         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2826                                  dentry->d_name.name, dentry->d_name.len);
2827         if (ret)
2828                 goto out;
2829
2830         if (inode->i_nlink == 0) {
2831                 ret = btrfs_orphan_add(trans, inode);
2832                 if (ret)
2833                         goto out;
2834         }
2835
2836 out:
2837         nr = trans->blocks_used;
2838         __unlink_end_trans(trans, root);
2839         btrfs_btree_balance_dirty(root, nr);
2840         return ret;
2841 }
2842
2843 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2844                         struct btrfs_root *root,
2845                         struct inode *dir, u64 objectid,
2846                         const char *name, int name_len)
2847 {
2848         struct btrfs_path *path;
2849         struct extent_buffer *leaf;
2850         struct btrfs_dir_item *di;
2851         struct btrfs_key key;
2852         u64 index;
2853         int ret;
2854         u64 dir_ino = btrfs_ino(dir);
2855
2856         path = btrfs_alloc_path();
2857         if (!path)
2858                 return -ENOMEM;
2859
2860         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2861                                    name, name_len, -1);
2862         BUG_ON(IS_ERR_OR_NULL(di));
2863
2864         leaf = path->nodes[0];
2865         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2866         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2867         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2868         BUG_ON(ret);
2869         btrfs_release_path(path);
2870
2871         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2872                                  objectid, root->root_key.objectid,
2873                                  dir_ino, &index, name, name_len);
2874         if (ret < 0) {
2875                 BUG_ON(ret != -ENOENT);
2876                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2877                                                  name, name_len);
2878                 BUG_ON(IS_ERR_OR_NULL(di));
2879
2880                 leaf = path->nodes[0];
2881                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2882                 btrfs_release_path(path);
2883                 index = key.offset;
2884         }
2885         btrfs_release_path(path);
2886
2887         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2888         BUG_ON(ret);
2889
2890         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2891         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2892         ret = btrfs_update_inode(trans, root, dir);
2893         BUG_ON(ret);
2894
2895         btrfs_free_path(path);
2896         return 0;
2897 }
2898
2899 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2900 {
2901         struct inode *inode = dentry->d_inode;
2902         int err = 0;
2903         struct btrfs_root *root = BTRFS_I(dir)->root;
2904         struct btrfs_trans_handle *trans;
2905         unsigned long nr = 0;
2906
2907         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2908             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2909                 return -ENOTEMPTY;
2910
2911         trans = __unlink_start_trans(dir, dentry);
2912         if (IS_ERR(trans))
2913                 return PTR_ERR(trans);
2914
2915         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2916                 err = btrfs_unlink_subvol(trans, root, dir,
2917                                           BTRFS_I(inode)->location.objectid,
2918                                           dentry->d_name.name,
2919                                           dentry->d_name.len);
2920                 goto out;
2921         }
2922
2923         err = btrfs_orphan_add(trans, inode);
2924         if (err)
2925                 goto out;
2926
2927         /* now the directory is empty */
2928         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2929                                  dentry->d_name.name, dentry->d_name.len);
2930         if (!err)
2931                 btrfs_i_size_write(inode, 0);
2932 out:
2933         nr = trans->blocks_used;
2934         __unlink_end_trans(trans, root);
2935         btrfs_btree_balance_dirty(root, nr);
2936
2937         return err;
2938 }
2939
2940 /*
2941  * this can truncate away extent items, csum items and directory items.
2942  * It starts at a high offset and removes keys until it can't find
2943  * any higher than new_size
2944  *
2945  * csum items that cross the new i_size are truncated to the new size
2946  * as well.
2947  *
2948  * min_type is the minimum key type to truncate down to.  If set to 0, this
2949  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2950  */
2951 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2952                                struct btrfs_root *root,
2953                                struct inode *inode,
2954                                u64 new_size, u32 min_type)
2955 {
2956         struct btrfs_path *path;
2957         struct extent_buffer *leaf;
2958         struct btrfs_file_extent_item *fi;
2959         struct btrfs_key key;
2960         struct btrfs_key found_key;
2961         u64 extent_start = 0;
2962         u64 extent_num_bytes = 0;
2963         u64 extent_offset = 0;
2964         u64 item_end = 0;
2965         u64 mask = root->sectorsize - 1;
2966         u32 found_type = (u8)-1;
2967         int found_extent;
2968         int del_item;
2969         int pending_del_nr = 0;
2970         int pending_del_slot = 0;
2971         int extent_type = -1;
2972         int encoding;
2973         int ret;
2974         int err = 0;
2975         u64 ino = btrfs_ino(inode);
2976
2977         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2978
2979         path = btrfs_alloc_path();
2980         if (!path)
2981                 return -ENOMEM;
2982         path->reada = -1;
2983
2984         if (root->ref_cows || root == root->fs_info->tree_root)
2985                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2986
2987         /*
2988          * This function is also used to drop the items in the log tree before
2989          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
2990          * it is used to drop the loged items. So we shouldn't kill the delayed
2991          * items.
2992          */
2993         if (min_type == 0 && root == BTRFS_I(inode)->root)
2994                 btrfs_kill_delayed_inode_items(inode);
2995
2996         key.objectid = ino;
2997         key.offset = (u64)-1;
2998         key.type = (u8)-1;
2999
3000 search_again:
3001         path->leave_spinning = 1;
3002         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3003         if (ret < 0) {
3004                 err = ret;
3005                 goto out;
3006         }
3007
3008         if (ret > 0) {
3009                 /* there are no items in the tree for us to truncate, we're
3010                  * done
3011                  */
3012                 if (path->slots[0] == 0)
3013                         goto out;
3014                 path->slots[0]--;
3015         }
3016
3017         while (1) {
3018                 fi = NULL;
3019                 leaf = path->nodes[0];
3020                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3021                 found_type = btrfs_key_type(&found_key);
3022                 encoding = 0;
3023
3024                 if (found_key.objectid != ino)
3025                         break;
3026
3027                 if (found_type < min_type)
3028                         break;
3029
3030                 item_end = found_key.offset;
3031                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3032                         fi = btrfs_item_ptr(leaf, path->slots[0],
3033                                             struct btrfs_file_extent_item);
3034                         extent_type = btrfs_file_extent_type(leaf, fi);
3035                         encoding = btrfs_file_extent_compression(leaf, fi);
3036                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3037                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3038
3039                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3040                                 item_end +=
3041                                     btrfs_file_extent_num_bytes(leaf, fi);
3042                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3043                                 item_end += btrfs_file_extent_inline_len(leaf,
3044                                                                          fi);
3045                         }
3046                         item_end--;
3047                 }
3048                 if (found_type > min_type) {
3049                         del_item = 1;
3050                 } else {
3051                         if (item_end < new_size)
3052                                 break;
3053                         if (found_key.offset >= new_size)
3054                                 del_item = 1;
3055                         else
3056                                 del_item = 0;
3057                 }
3058                 found_extent = 0;
3059                 /* FIXME, shrink the extent if the ref count is only 1 */
3060                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3061                         goto delete;
3062
3063                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3064                         u64 num_dec;
3065                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3066                         if (!del_item && !encoding) {
3067                                 u64 orig_num_bytes =
3068                                         btrfs_file_extent_num_bytes(leaf, fi);
3069                                 extent_num_bytes = new_size -
3070                                         found_key.offset + root->sectorsize - 1;
3071                                 extent_num_bytes = extent_num_bytes &
3072                                         ~((u64)root->sectorsize - 1);
3073                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3074                                                          extent_num_bytes);
3075                                 num_dec = (orig_num_bytes -
3076                                            extent_num_bytes);
3077                                 if (root->ref_cows && extent_start != 0)
3078                                         inode_sub_bytes(inode, num_dec);
3079                                 btrfs_mark_buffer_dirty(leaf);
3080                         } else {
3081                                 extent_num_bytes =
3082                                         btrfs_file_extent_disk_num_bytes(leaf,
3083                                                                          fi);
3084                                 extent_offset = found_key.offset -
3085                                         btrfs_file_extent_offset(leaf, fi);
3086
3087                                 /* FIXME blocksize != 4096 */
3088                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3089                                 if (extent_start != 0) {
3090                                         found_extent = 1;
3091                                         if (root->ref_cows)
3092                                                 inode_sub_bytes(inode, num_dec);
3093                                 }
3094                         }
3095                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3096                         /*
3097                          * we can't truncate inline items that have had
3098                          * special encodings
3099                          */
3100                         if (!del_item &&
3101                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3102                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3103                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3104                                 u32 size = new_size - found_key.offset;
3105
3106                                 if (root->ref_cows) {
3107                                         inode_sub_bytes(inode, item_end + 1 -
3108                                                         new_size);
3109                                 }
3110                                 size =
3111                                     btrfs_file_extent_calc_inline_size(size);
3112                                 ret = btrfs_truncate_item(trans, root, path,
3113                                                           size, 1);
3114                         } else if (root->ref_cows) {
3115                                 inode_sub_bytes(inode, item_end + 1 -
3116                                                 found_key.offset);
3117                         }
3118                 }
3119 delete:
3120                 if (del_item) {
3121                         if (!pending_del_nr) {
3122                                 /* no pending yet, add ourselves */
3123                                 pending_del_slot = path->slots[0];
3124                                 pending_del_nr = 1;
3125                         } else if (pending_del_nr &&
3126                                    path->slots[0] + 1 == pending_del_slot) {
3127                                 /* hop on the pending chunk */
3128                                 pending_del_nr++;
3129                                 pending_del_slot = path->slots[0];
3130                         } else {
3131                                 BUG();
3132                         }
3133                 } else {
3134                         break;
3135                 }
3136                 if (found_extent && (root->ref_cows ||
3137                                      root == root->fs_info->tree_root)) {
3138                         btrfs_set_path_blocking(path);
3139                         ret = btrfs_free_extent(trans, root, extent_start,
3140                                                 extent_num_bytes, 0,
3141                                                 btrfs_header_owner(leaf),
3142                                                 ino, extent_offset);
3143                         BUG_ON(ret);
3144                 }
3145
3146                 if (found_type == BTRFS_INODE_ITEM_KEY)
3147                         break;
3148
3149                 if (path->slots[0] == 0 ||
3150                     path->slots[0] != pending_del_slot) {
3151                         if (root->ref_cows &&
3152                             BTRFS_I(inode)->location.objectid !=
3153                                                 BTRFS_FREE_INO_OBJECTID) {
3154                                 err = -EAGAIN;
3155                                 goto out;
3156                         }
3157                         if (pending_del_nr) {
3158                                 ret = btrfs_del_items(trans, root, path,
3159                                                 pending_del_slot,
3160                                                 pending_del_nr);
3161                                 BUG_ON(ret);
3162                                 pending_del_nr = 0;
3163                         }
3164                         btrfs_release_path(path);
3165                         goto search_again;
3166                 } else {
3167                         path->slots[0]--;
3168                 }
3169         }
3170 out:
3171         if (pending_del_nr) {
3172                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3173                                       pending_del_nr);
3174                 BUG_ON(ret);
3175         }
3176         btrfs_free_path(path);
3177         return err;
3178 }
3179
3180 /*
3181  * taken from block_truncate_page, but does cow as it zeros out
3182  * any bytes left in the last page in the file.
3183  */
3184 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3185 {
3186         struct inode *inode = mapping->host;
3187         struct btrfs_root *root = BTRFS_I(inode)->root;
3188         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3189         struct btrfs_ordered_extent *ordered;
3190         struct extent_state *cached_state = NULL;
3191         char *kaddr;
3192         u32 blocksize = root->sectorsize;
3193         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3194         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3195         struct page *page;
3196         gfp_t mask = btrfs_alloc_write_mask(mapping);
3197         int ret = 0;
3198         u64 page_start;
3199         u64 page_end;
3200
3201         if ((offset & (blocksize - 1)) == 0)
3202                 goto out;
3203         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3204         if (ret)
3205                 goto out;
3206
3207         ret = -ENOMEM;
3208 again:
3209         page = find_or_create_page(mapping, index, mask);
3210         if (!page) {
3211                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3212                 goto out;
3213         }
3214
3215         page_start = page_offset(page);
3216         page_end = page_start + PAGE_CACHE_SIZE - 1;
3217
3218         if (!PageUptodate(page)) {
3219                 ret = btrfs_readpage(NULL, page);
3220                 lock_page(page);
3221                 if (page->mapping != mapping) {
3222                         unlock_page(page);
3223                         page_cache_release(page);
3224                         goto again;
3225                 }
3226                 if (!PageUptodate(page)) {
3227                         ret = -EIO;
3228                         goto out_unlock;
3229                 }
3230         }
3231         wait_on_page_writeback(page);
3232
3233         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3234                          GFP_NOFS);
3235         set_page_extent_mapped(page);
3236
3237         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3238         if (ordered) {
3239                 unlock_extent_cached(io_tree, page_start, page_end,
3240                                      &cached_state, GFP_NOFS);
3241                 unlock_page(page);
3242                 page_cache_release(page);
3243                 btrfs_start_ordered_extent(inode, ordered, 1);
3244                 btrfs_put_ordered_extent(ordered);
3245                 goto again;
3246         }
3247
3248         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3249                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3250                           0, 0, &cached_state, GFP_NOFS);
3251
3252         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3253                                         &cached_state);
3254         if (ret) {
3255                 unlock_extent_cached(io_tree, page_start, page_end,
3256                                      &cached_state, GFP_NOFS);
3257                 goto out_unlock;
3258         }
3259
3260         ret = 0;
3261         if (offset != PAGE_CACHE_SIZE) {
3262                 kaddr = kmap(page);
3263                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3264                 flush_dcache_page(page);
3265                 kunmap(page);
3266         }
3267         ClearPageChecked(page);
3268         set_page_dirty(page);
3269         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3270                              GFP_NOFS);
3271
3272 out_unlock:
3273         if (ret)
3274                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3275         unlock_page(page);
3276         page_cache_release(page);
3277 out:
3278         return ret;
3279 }
3280
3281 /*
3282  * This function puts in dummy file extents for the area we're creating a hole
3283  * for.  So if we are truncating this file to a larger size we need to insert
3284  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3285  * the range between oldsize and size
3286  */
3287 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3288 {
3289         struct btrfs_trans_handle *trans;
3290         struct btrfs_root *root = BTRFS_I(inode)->root;
3291         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3292         struct extent_map *em = NULL;
3293         struct extent_state *cached_state = NULL;
3294         u64 mask = root->sectorsize - 1;
3295         u64 hole_start = (oldsize + mask) & ~mask;
3296         u64 block_end = (size + mask) & ~mask;
3297         u64 last_byte;
3298         u64 cur_offset;
3299         u64 hole_size;
3300         int err = 0;
3301
3302         if (size <= hole_start)
3303                 return 0;
3304
3305         while (1) {
3306                 struct btrfs_ordered_extent *ordered;
3307                 btrfs_wait_ordered_range(inode, hole_start,
3308                                          block_end - hole_start);
3309                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3310                                  &cached_state, GFP_NOFS);
3311                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3312                 if (!ordered)
3313                         break;
3314                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3315                                      &cached_state, GFP_NOFS);
3316                 btrfs_put_ordered_extent(ordered);
3317         }
3318
3319         cur_offset = hole_start;
3320         while (1) {
3321                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3322                                 block_end - cur_offset, 0);
3323                 BUG_ON(IS_ERR_OR_NULL(em));
3324                 last_byte = min(extent_map_end(em), block_end);
3325                 last_byte = (last_byte + mask) & ~mask;
3326                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3327                         u64 hint_byte = 0;
3328                         hole_size = last_byte - cur_offset;
3329
3330                         trans = btrfs_start_transaction(root, 2);
3331                         if (IS_ERR(trans)) {
3332                                 err = PTR_ERR(trans);
3333                                 break;
3334                         }
3335
3336                         err = btrfs_drop_extents(trans, inode, cur_offset,
3337                                                  cur_offset + hole_size,
3338                                                  &hint_byte, 1);
3339                         if (err) {
3340                                 btrfs_end_transaction(trans, root);
3341                                 break;
3342                         }
3343
3344                         err = btrfs_insert_file_extent(trans, root,
3345                                         btrfs_ino(inode), cur_offset, 0,
3346                                         0, hole_size, 0, hole_size,
3347                                         0, 0, 0);
3348                         if (err) {
3349                                 btrfs_end_transaction(trans, root);
3350                                 break;
3351                         }
3352
3353                         btrfs_drop_extent_cache(inode, hole_start,
3354                                         last_byte - 1, 0);
3355
3356                         btrfs_end_transaction(trans, root);
3357                 }
3358                 free_extent_map(em);
3359                 em = NULL;
3360                 cur_offset = last_byte;
3361                 if (cur_offset >= block_end)
3362                         break;
3363         }
3364
3365         free_extent_map(em);
3366         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3367                              GFP_NOFS);
3368         return err;
3369 }
3370
3371 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3372 {
3373         loff_t oldsize = i_size_read(inode);
3374         int ret;
3375
3376         if (newsize == oldsize)
3377                 return 0;
3378
3379         if (newsize > oldsize) {
3380                 i_size_write(inode, newsize);
3381                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3382                 truncate_pagecache(inode, oldsize, newsize);
3383                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3384                 if (ret) {
3385                         btrfs_setsize(inode, oldsize);
3386                         return ret;
3387                 }
3388
3389                 mark_inode_dirty(inode);
3390         } else {
3391
3392                 /*
3393                  * We're truncating a file that used to have good data down to
3394                  * zero. Make sure it gets into the ordered flush list so that
3395                  * any new writes get down to disk quickly.
3396                  */
3397                 if (newsize == 0)
3398                         BTRFS_I(inode)->ordered_data_close = 1;
3399
3400                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3401                 truncate_setsize(inode, newsize);
3402                 ret = btrfs_truncate(inode);
3403         }
3404
3405         return ret;
3406 }
3407
3408 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3409 {
3410         struct inode *inode = dentry->d_inode;
3411         struct btrfs_root *root = BTRFS_I(inode)->root;
3412         int err;
3413
3414         if (btrfs_root_readonly(root))
3415                 return -EROFS;
3416
3417         err = inode_change_ok(inode, attr);
3418         if (err)
3419                 return err;
3420
3421         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3422                 err = btrfs_setsize(inode, attr->ia_size);
3423                 if (err)
3424                         return err;
3425         }
3426
3427         if (attr->ia_valid) {
3428                 setattr_copy(inode, attr);
3429                 mark_inode_dirty(inode);
3430
3431                 if (attr->ia_valid & ATTR_MODE)
3432                         err = btrfs_acl_chmod(inode);
3433         }
3434
3435         return err;
3436 }
3437
3438 void btrfs_evict_inode(struct inode *inode)
3439 {
3440         struct btrfs_trans_handle *trans;
3441         struct btrfs_root *root = BTRFS_I(inode)->root;
3442         struct btrfs_block_rsv *rsv, *global_rsv;
3443         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3444         unsigned long nr;
3445         int ret;
3446
3447         trace_btrfs_inode_evict(inode);
3448
3449         truncate_inode_pages(&inode->i_data, 0);
3450         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3451                                btrfs_is_free_space_inode(root, inode)))
3452                 goto no_delete;
3453
3454         if (is_bad_inode(inode)) {
3455                 btrfs_orphan_del(NULL, inode);
3456                 goto no_delete;
3457         }
3458         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3459         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3460
3461         if (root->fs_info->log_root_recovering) {
3462                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3463                 goto no_delete;
3464         }
3465
3466         if (inode->i_nlink > 0) {
3467                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3468                 goto no_delete;
3469         }
3470
3471         rsv = btrfs_alloc_block_rsv(root);
3472         if (!rsv) {
3473                 btrfs_orphan_del(NULL, inode);
3474                 goto no_delete;
3475         }
3476         rsv->size = min_size;
3477         global_rsv = &root->fs_info->global_block_rsv;
3478
3479         btrfs_i_size_write(inode, 0);
3480
3481         /*
3482          * This is a bit simpler than btrfs_truncate since
3483          *
3484          * 1) We've already reserved our space for our orphan item in the
3485          *    unlink.
3486          * 2) We're going to delete the inode item, so we don't need to update
3487          *    it at all.
3488          *
3489          * So we just need to reserve some slack space in case we add bytes when
3490          * doing the truncate.
3491          */
3492         while (1) {
3493                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
3494
3495                 /*
3496                  * Try and steal from the global reserve since we will
3497                  * likely not use this space anyway, we want to try as
3498                  * hard as possible to get this to work.
3499                  */
3500                 if (ret)
3501                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3502
3503                 if (ret) {
3504                         printk(KERN_WARNING "Could not get space for a "
3505                                "delete, will truncate on mount %d\n", ret);
3506                         btrfs_orphan_del(NULL, inode);
3507                         btrfs_free_block_rsv(root, rsv);
3508                         goto no_delete;
3509                 }
3510
3511                 trans = btrfs_start_transaction(root, 0);
3512                 if (IS_ERR(trans)) {
3513                         btrfs_orphan_del(NULL, inode);
3514                         btrfs_free_block_rsv(root, rsv);
3515                         goto no_delete;
3516                 }
3517
3518                 trans->block_rsv = rsv;
3519
3520                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3521                 if (ret != -EAGAIN)
3522                         break;
3523
3524                 nr = trans->blocks_used;
3525                 btrfs_end_transaction(trans, root);
3526                 trans = NULL;
3527                 btrfs_btree_balance_dirty(root, nr);
3528         }
3529
3530         btrfs_free_block_rsv(root, rsv);
3531
3532         if (ret == 0) {
3533                 trans->block_rsv = root->orphan_block_rsv;
3534                 ret = btrfs_orphan_del(trans, inode);
3535                 BUG_ON(ret);
3536         }
3537
3538         trans->block_rsv = &root->fs_info->trans_block_rsv;
3539         if (!(root == root->fs_info->tree_root ||
3540               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3541                 btrfs_return_ino(root, btrfs_ino(inode));
3542
3543         nr = trans->blocks_used;
3544         btrfs_end_transaction(trans, root);
3545         btrfs_btree_balance_dirty(root, nr);
3546 no_delete:
3547         end_writeback(inode);
3548         return;
3549 }
3550
3551 /*
3552  * this returns the key found in the dir entry in the location pointer.
3553  * If no dir entries were found, location->objectid is 0.
3554  */
3555 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3556                                struct btrfs_key *location)
3557 {
3558         const char *name = dentry->d_name.name;
3559         int namelen = dentry->d_name.len;
3560         struct btrfs_dir_item *di;
3561         struct btrfs_path *path;
3562         struct btrfs_root *root = BTRFS_I(dir)->root;
3563         int ret = 0;
3564
3565         path = btrfs_alloc_path();
3566         if (!path)
3567                 return -ENOMEM;
3568
3569         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3570                                     namelen, 0);
3571         if (IS_ERR(di))
3572                 ret = PTR_ERR(di);
3573
3574         if (IS_ERR_OR_NULL(di))
3575                 goto out_err;
3576
3577         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3578 out:
3579         btrfs_free_path(path);
3580         return ret;
3581 out_err:
3582         location->objectid = 0;
3583         goto out;
3584 }
3585
3586 /*
3587  * when we hit a tree root in a directory, the btrfs part of the inode
3588  * needs to be changed to reflect the root directory of the tree root.  This
3589  * is kind of like crossing a mount point.
3590  */
3591 static int fixup_tree_root_location(struct btrfs_root *root,
3592                                     struct inode *dir,
3593                                     struct dentry *dentry,
3594                                     struct btrfs_key *location,
3595                                     struct btrfs_root **sub_root)
3596 {
3597         struct btrfs_path *path;
3598         struct btrfs_root *new_root;
3599         struct btrfs_root_ref *ref;
3600         struct extent_buffer *leaf;
3601         int ret;
3602         int err = 0;
3603
3604         path = btrfs_alloc_path();
3605         if (!path) {
3606                 err = -ENOMEM;
3607                 goto out;
3608         }
3609
3610         err = -ENOENT;
3611         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3612                                   BTRFS_I(dir)->root->root_key.objectid,
3613                                   location->objectid);
3614         if (ret) {
3615                 if (ret < 0)
3616                         err = ret;
3617                 goto out;
3618         }
3619
3620         leaf = path->nodes[0];
3621         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3622         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3623             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3624                 goto out;
3625
3626         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3627                                    (unsigned long)(ref + 1),
3628                                    dentry->d_name.len);
3629         if (ret)
3630                 goto out;
3631
3632         btrfs_release_path(path);
3633
3634         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3635         if (IS_ERR(new_root)) {
3636                 err = PTR_ERR(new_root);
3637                 goto out;
3638         }
3639
3640         if (btrfs_root_refs(&new_root->root_item) == 0) {
3641                 err = -ENOENT;
3642                 goto out;
3643         }
3644
3645         *sub_root = new_root;
3646         location->objectid = btrfs_root_dirid(&new_root->root_item);
3647         location->type = BTRFS_INODE_ITEM_KEY;
3648         location->offset = 0;
3649         err = 0;
3650 out:
3651         btrfs_free_path(path);
3652         return err;
3653 }
3654
3655 static void inode_tree_add(struct inode *inode)
3656 {
3657         struct btrfs_root *root = BTRFS_I(inode)->root;
3658         struct btrfs_inode *entry;
3659         struct rb_node **p;
3660         struct rb_node *parent;
3661         u64 ino = btrfs_ino(inode);
3662 again:
3663         p = &root->inode_tree.rb_node;
3664         parent = NULL;
3665
3666         if (inode_unhashed(inode))
3667                 return;
3668
3669         spin_lock(&root->inode_lock);
3670         while (*p) {
3671                 parent = *p;
3672                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3673
3674                 if (ino < btrfs_ino(&entry->vfs_inode))
3675                         p = &parent->rb_left;
3676                 else if (ino > btrfs_ino(&entry->vfs_inode))
3677                         p = &parent->rb_right;
3678                 else {
3679                         WARN_ON(!(entry->vfs_inode.i_state &
3680                                   (I_WILL_FREE | I_FREEING)));
3681                         rb_erase(parent, &root->inode_tree);
3682                         RB_CLEAR_NODE(parent);
3683                         spin_unlock(&root->inode_lock);
3684                         goto again;
3685                 }
3686         }
3687         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3688         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3689         spin_unlock(&root->inode_lock);
3690 }
3691
3692 static void inode_tree_del(struct inode *inode)
3693 {
3694         struct btrfs_root *root = BTRFS_I(inode)->root;
3695         int empty = 0;
3696
3697         spin_lock(&root->inode_lock);
3698         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3699                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3700                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3701                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3702         }
3703         spin_unlock(&root->inode_lock);
3704
3705         /*
3706          * Free space cache has inodes in the tree root, but the tree root has a
3707          * root_refs of 0, so this could end up dropping the tree root as a
3708          * snapshot, so we need the extra !root->fs_info->tree_root check to
3709          * make sure we don't drop it.
3710          */
3711         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3712             root != root->fs_info->tree_root) {
3713                 synchronize_srcu(&root->fs_info->subvol_srcu);
3714                 spin_lock(&root->inode_lock);
3715                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3716                 spin_unlock(&root->inode_lock);
3717                 if (empty)
3718                         btrfs_add_dead_root(root);
3719         }
3720 }
3721
3722 int btrfs_invalidate_inodes(struct btrfs_root *root)
3723 {
3724         struct rb_node *node;
3725         struct rb_node *prev;
3726         struct btrfs_inode *entry;
3727         struct inode *inode;
3728         u64 objectid = 0;
3729
3730         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3731
3732         spin_lock(&root->inode_lock);
3733 again:
3734         node = root->inode_tree.rb_node;
3735         prev = NULL;
3736         while (node) {
3737                 prev = node;
3738                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3739
3740                 if (objectid < btrfs_ino(&entry->vfs_inode))
3741                         node = node->rb_left;
3742                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3743                         node = node->rb_right;
3744                 else
3745                         break;
3746         }
3747         if (!node) {
3748                 while (prev) {
3749                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3750                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3751                                 node = prev;
3752                                 break;
3753                         }
3754                         prev = rb_next(prev);
3755                 }
3756         }
3757         while (node) {
3758                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3759                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3760                 inode = igrab(&entry->vfs_inode);
3761                 if (inode) {
3762                         spin_unlock(&root->inode_lock);
3763                         if (atomic_read(&inode->i_count) > 1)
3764                                 d_prune_aliases(inode);
3765                         /*
3766                          * btrfs_drop_inode will have it removed from
3767                          * the inode cache when its usage count
3768                          * hits zero.
3769                          */
3770                         iput(inode);
3771                         cond_resched();
3772                         spin_lock(&root->inode_lock);
3773                         goto again;
3774                 }
3775
3776                 if (cond_resched_lock(&root->inode_lock))
3777                         goto again;
3778
3779                 node = rb_next(node);
3780         }
3781         spin_unlock(&root->inode_lock);
3782         return 0;
3783 }
3784
3785 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3786 {
3787         struct btrfs_iget_args *args = p;
3788         inode->i_ino = args->ino;
3789         BTRFS_I(inode)->root = args->root;
3790         btrfs_set_inode_space_info(args->root, inode);
3791         return 0;
3792 }
3793
3794 static int btrfs_find_actor(struct inode *inode, void *opaque)
3795 {
3796         struct btrfs_iget_args *args = opaque;
3797         return args->ino == btrfs_ino(inode) &&
3798                 args->root == BTRFS_I(inode)->root;
3799 }
3800
3801 static struct inode *btrfs_iget_locked(struct super_block *s,
3802                                        u64 objectid,
3803                                        struct btrfs_root *root)
3804 {
3805         struct inode *inode;
3806         struct btrfs_iget_args args;
3807         args.ino = objectid;
3808         args.root = root;
3809
3810         inode = iget5_locked(s, objectid, btrfs_find_actor,
3811                              btrfs_init_locked_inode,
3812                              (void *)&args);
3813         return inode;
3814 }
3815
3816 /* Get an inode object given its location and corresponding root.
3817  * Returns in *is_new if the inode was read from disk
3818  */
3819 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3820                          struct btrfs_root *root, int *new)
3821 {
3822         struct inode *inode;
3823
3824         inode = btrfs_iget_locked(s, location->objectid, root);
3825         if (!inode)
3826                 return ERR_PTR(-ENOMEM);
3827
3828         if (inode->i_state & I_NEW) {
3829                 BTRFS_I(inode)->root = root;
3830                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3831                 btrfs_read_locked_inode(inode);
3832                 if (!is_bad_inode(inode)) {
3833                         inode_tree_add(inode);
3834                         unlock_new_inode(inode);
3835                         if (new)
3836                                 *new = 1;
3837                 } else {
3838                         unlock_new_inode(inode);
3839                         iput(inode);
3840                         inode = ERR_PTR(-ESTALE);
3841                 }
3842         }
3843
3844         return inode;
3845 }
3846
3847 static struct inode *new_simple_dir(struct super_block *s,
3848                                     struct btrfs_key *key,
3849                                     struct btrfs_root *root)
3850 {
3851         struct inode *inode = new_inode(s);
3852
3853         if (!inode)
3854                 return ERR_PTR(-ENOMEM);
3855
3856         BTRFS_I(inode)->root = root;
3857         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3858         BTRFS_I(inode)->dummy_inode = 1;
3859
3860         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3861         inode->i_op = &simple_dir_inode_operations;
3862         inode->i_fop = &simple_dir_operations;
3863         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3864         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3865
3866         return inode;
3867 }
3868
3869 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3870 {
3871         struct inode *inode;
3872         struct btrfs_root *root = BTRFS_I(dir)->root;
3873         struct btrfs_root *sub_root = root;
3874         struct btrfs_key location;
3875         int index;
3876         int ret = 0;
3877
3878         if (dentry->d_name.len > BTRFS_NAME_LEN)
3879                 return ERR_PTR(-ENAMETOOLONG);
3880
3881         if (unlikely(d_need_lookup(dentry))) {
3882                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3883                 kfree(dentry->d_fsdata);
3884                 dentry->d_fsdata = NULL;
3885                 /* This thing is hashed, drop it for now */
3886                 d_drop(dentry);
3887         } else {
3888                 ret = btrfs_inode_by_name(dir, dentry, &location);
3889         }
3890
3891         if (ret < 0)
3892                 return ERR_PTR(ret);
3893
3894         if (location.objectid == 0)
3895                 return NULL;
3896
3897         if (location.type == BTRFS_INODE_ITEM_KEY) {
3898                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3899                 return inode;
3900         }
3901
3902         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3903
3904         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3905         ret = fixup_tree_root_location(root, dir, dentry,
3906                                        &location, &sub_root);
3907         if (ret < 0) {
3908                 if (ret != -ENOENT)
3909                         inode = ERR_PTR(ret);
3910                 else
3911                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3912         } else {
3913                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3914         }
3915         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3916
3917         if (!IS_ERR(inode) && root != sub_root) {
3918                 down_read(&root->fs_info->cleanup_work_sem);
3919                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3920                         ret = btrfs_orphan_cleanup(sub_root);
3921                 up_read(&root->fs_info->cleanup_work_sem);
3922                 if (ret)
3923                         inode = ERR_PTR(ret);
3924         }
3925
3926         return inode;
3927 }
3928
3929 static int btrfs_dentry_delete(const struct dentry *dentry)
3930 {
3931         struct btrfs_root *root;
3932
3933         if (!dentry->d_inode && !IS_ROOT(dentry))
3934                 dentry = dentry->d_parent;
3935
3936         if (dentry->d_inode) {
3937                 root = BTRFS_I(dentry->d_inode)->root;
3938                 if (btrfs_root_refs(&root->root_item) == 0)
3939                         return 1;
3940         }
3941         return 0;
3942 }
3943
3944 static void btrfs_dentry_release(struct dentry *dentry)
3945 {
3946         if (dentry->d_fsdata)
3947                 kfree(dentry->d_fsdata);
3948 }
3949
3950 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3951                                    struct nameidata *nd)
3952 {
3953         struct dentry *ret;
3954
3955         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
3956         if (unlikely(d_need_lookup(dentry))) {
3957                 spin_lock(&dentry->d_lock);
3958                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
3959                 spin_unlock(&dentry->d_lock);
3960         }
3961         return ret;
3962 }
3963
3964 unsigned char btrfs_filetype_table[] = {
3965         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3966 };
3967
3968 static int btrfs_real_readdir(struct file *filp, void *dirent,
3969                               filldir_t filldir)
3970 {
3971         struct inode *inode = filp->f_dentry->d_inode;
3972         struct btrfs_root *root = BTRFS_I(inode)->root;
3973         struct btrfs_item *item;
3974         struct btrfs_dir_item *di;
3975         struct btrfs_key key;
3976         struct btrfs_key found_key;
3977         struct btrfs_path *path;
3978         struct list_head ins_list;
3979         struct list_head del_list;
3980         struct qstr q;
3981         int ret;
3982         struct extent_buffer *leaf;
3983         int slot;
3984         unsigned char d_type;
3985         int over = 0;
3986         u32 di_cur;
3987         u32 di_total;
3988         u32 di_len;
3989         int key_type = BTRFS_DIR_INDEX_KEY;
3990         char tmp_name[32];
3991         char *name_ptr;
3992         int name_len;
3993         int is_curr = 0;        /* filp->f_pos points to the current index? */
3994
3995         /* FIXME, use a real flag for deciding about the key type */
3996         if (root->fs_info->tree_root == root)
3997                 key_type = BTRFS_DIR_ITEM_KEY;
3998
3999         /* special case for "." */
4000         if (filp->f_pos == 0) {
4001                 over = filldir(dirent, ".", 1,
4002                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4003                 if (over)
4004                         return 0;
4005                 filp->f_pos = 1;
4006         }
4007         /* special case for .., just use the back ref */
4008         if (filp->f_pos == 1) {
4009                 u64 pino = parent_ino(filp->f_path.dentry);
4010                 over = filldir(dirent, "..", 2,
4011                                filp->f_pos, pino, DT_DIR);
4012                 if (over)
4013                         return 0;
4014                 filp->f_pos = 2;
4015         }
4016         path = btrfs_alloc_path();
4017         if (!path)
4018                 return -ENOMEM;
4019
4020         path->reada = 1;
4021
4022         if (key_type == BTRFS_DIR_INDEX_KEY) {
4023                 INIT_LIST_HEAD(&ins_list);
4024                 INIT_LIST_HEAD(&del_list);
4025                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4026         }
4027
4028         btrfs_set_key_type(&key, key_type);
4029         key.offset = filp->f_pos;
4030         key.objectid = btrfs_ino(inode);
4031
4032         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4033         if (ret < 0)
4034                 goto err;
4035
4036         while (1) {
4037                 leaf = path->nodes[0];
4038                 slot = path->slots[0];
4039                 if (slot >= btrfs_header_nritems(leaf)) {
4040                         ret = btrfs_next_leaf(root, path);
4041                         if (ret < 0)
4042                                 goto err;
4043                         else if (ret > 0)
4044                                 break;
4045                         continue;
4046                 }
4047
4048                 item = btrfs_item_nr(leaf, slot);
4049                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4050
4051                 if (found_key.objectid != key.objectid)
4052                         break;
4053                 if (btrfs_key_type(&found_key) != key_type)
4054                         break;
4055                 if (found_key.offset < filp->f_pos)
4056                         goto next;
4057                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4058                     btrfs_should_delete_dir_index(&del_list,
4059                                                   found_key.offset))
4060                         goto next;
4061
4062                 filp->f_pos = found_key.offset;
4063                 is_curr = 1;
4064
4065                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4066                 di_cur = 0;
4067                 di_total = btrfs_item_size(leaf, item);
4068
4069                 while (di_cur < di_total) {
4070                         struct btrfs_key location;
4071                         struct dentry *tmp;
4072
4073                         if (verify_dir_item(root, leaf, di))
4074                                 break;
4075
4076                         name_len = btrfs_dir_name_len(leaf, di);
4077                         if (name_len <= sizeof(tmp_name)) {
4078                                 name_ptr = tmp_name;
4079                         } else {
4080                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4081                                 if (!name_ptr) {
4082                                         ret = -ENOMEM;
4083                                         goto err;
4084                                 }
4085                         }
4086                         read_extent_buffer(leaf, name_ptr,
4087                                            (unsigned long)(di + 1), name_len);
4088
4089                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4090                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4091
4092                         q.name = name_ptr;
4093                         q.len = name_len;
4094                         q.hash = full_name_hash(q.name, q.len);
4095                         tmp = d_lookup(filp->f_dentry, &q);
4096                         if (!tmp) {
4097                                 struct btrfs_key *newkey;
4098
4099                                 newkey = kzalloc(sizeof(struct btrfs_key),
4100                                                  GFP_NOFS);
4101                                 if (!newkey)
4102                                         goto no_dentry;
4103                                 tmp = d_alloc(filp->f_dentry, &q);
4104                                 if (!tmp) {
4105                                         kfree(newkey);
4106                                         dput(tmp);
4107                                         goto no_dentry;
4108                                 }
4109                                 memcpy(newkey, &location,
4110                                        sizeof(struct btrfs_key));
4111                                 tmp->d_fsdata = newkey;
4112                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4113                                 d_rehash(tmp);
4114                                 dput(tmp);
4115                         } else {
4116                                 dput(tmp);
4117                         }
4118 no_dentry:
4119                         /* is this a reference to our own snapshot? If so
4120                          * skip it
4121                          */
4122                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4123                             location.objectid == root->root_key.objectid) {
4124                                 over = 0;
4125                                 goto skip;
4126                         }
4127                         over = filldir(dirent, name_ptr, name_len,
4128                                        found_key.offset, location.objectid,
4129                                        d_type);
4130
4131 skip:
4132                         if (name_ptr != tmp_name)
4133                                 kfree(name_ptr);
4134
4135                         if (over)
4136                                 goto nopos;
4137                         di_len = btrfs_dir_name_len(leaf, di) +
4138                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4139                         di_cur += di_len;
4140                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4141                 }
4142 next:
4143                 path->slots[0]++;
4144         }
4145
4146         if (key_type == BTRFS_DIR_INDEX_KEY) {
4147                 if (is_curr)
4148                         filp->f_pos++;
4149                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4150                                                       &ins_list);
4151                 if (ret)
4152                         goto nopos;
4153         }
4154
4155         /* Reached end of directory/root. Bump pos past the last item. */
4156         if (key_type == BTRFS_DIR_INDEX_KEY)
4157                 /*
4158                  * 32-bit glibc will use getdents64, but then strtol -
4159                  * so the last number we can serve is this.
4160                  */
4161                 filp->f_pos = 0x7fffffff;
4162         else
4163                 filp->f_pos++;
4164 nopos:
4165         ret = 0;
4166 err:
4167         if (key_type == BTRFS_DIR_INDEX_KEY)
4168                 btrfs_put_delayed_items(&ins_list, &del_list);
4169         btrfs_free_path(path);
4170         return ret;
4171 }
4172
4173 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4174 {
4175         struct btrfs_root *root = BTRFS_I(inode)->root;
4176         struct btrfs_trans_handle *trans;
4177         int ret = 0;
4178         bool nolock = false;
4179
4180         if (BTRFS_I(inode)->dummy_inode)
4181                 return 0;
4182
4183         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4184                 nolock = true;
4185
4186         if (wbc->sync_mode == WB_SYNC_ALL) {
4187                 if (nolock)
4188                         trans = btrfs_join_transaction_nolock(root);
4189                 else
4190                         trans = btrfs_join_transaction(root);
4191                 if (IS_ERR(trans))
4192                         return PTR_ERR(trans);
4193                 if (nolock)
4194                         ret = btrfs_end_transaction_nolock(trans, root);
4195                 else
4196                         ret = btrfs_commit_transaction(trans, root);
4197         }
4198         return ret;
4199 }
4200
4201 /*
4202  * This is somewhat expensive, updating the tree every time the
4203  * inode changes.  But, it is most likely to find the inode in cache.
4204  * FIXME, needs more benchmarking...there are no reasons other than performance
4205  * to keep or drop this code.
4206  */
4207 void btrfs_dirty_inode(struct inode *inode, int flags)
4208 {
4209         struct btrfs_root *root = BTRFS_I(inode)->root;
4210         struct btrfs_trans_handle *trans;
4211         int ret;
4212
4213         if (BTRFS_I(inode)->dummy_inode)
4214                 return;
4215
4216         trans = btrfs_join_transaction(root);
4217         BUG_ON(IS_ERR(trans));
4218
4219         ret = btrfs_update_inode(trans, root, inode);
4220         if (ret && ret == -ENOSPC) {
4221                 /* whoops, lets try again with the full transaction */
4222                 btrfs_end_transaction(trans, root);
4223                 trans = btrfs_start_transaction(root, 1);
4224                 if (IS_ERR(trans)) {
4225                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4226                                        "dirty  inode %llu error %ld\n",
4227                                        (unsigned long long)btrfs_ino(inode),
4228                                        PTR_ERR(trans));
4229                         return;
4230                 }
4231
4232                 ret = btrfs_update_inode(trans, root, inode);
4233                 if (ret) {
4234                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4235                                        "dirty  inode %llu error %d\n",
4236                                        (unsigned long long)btrfs_ino(inode),
4237                                        ret);
4238                 }
4239         }
4240         btrfs_end_transaction(trans, root);
4241         if (BTRFS_I(inode)->delayed_node)
4242                 btrfs_balance_delayed_items(root);
4243 }
4244
4245 /*
4246  * find the highest existing sequence number in a directory
4247  * and then set the in-memory index_cnt variable to reflect
4248  * free sequence numbers
4249  */
4250 static int btrfs_set_inode_index_count(struct inode *inode)
4251 {
4252         struct btrfs_root *root = BTRFS_I(inode)->root;
4253         struct btrfs_key key, found_key;
4254         struct btrfs_path *path;
4255         struct extent_buffer *leaf;
4256         int ret;
4257
4258         key.objectid = btrfs_ino(inode);
4259         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4260         key.offset = (u64)-1;
4261
4262         path = btrfs_alloc_path();
4263         if (!path)
4264                 return -ENOMEM;
4265
4266         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4267         if (ret < 0)
4268                 goto out;
4269         /* FIXME: we should be able to handle this */
4270         if (ret == 0)
4271                 goto out;
4272         ret = 0;
4273
4274         /*
4275          * MAGIC NUMBER EXPLANATION:
4276          * since we search a directory based on f_pos we have to start at 2
4277          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4278          * else has to start at 2
4279          */
4280         if (path->slots[0] == 0) {
4281                 BTRFS_I(inode)->index_cnt = 2;
4282                 goto out;
4283         }
4284
4285         path->slots[0]--;
4286
4287         leaf = path->nodes[0];
4288         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4289
4290         if (found_key.objectid != btrfs_ino(inode) ||
4291             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4292                 BTRFS_I(inode)->index_cnt = 2;
4293                 goto out;
4294         }
4295
4296         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4297 out:
4298         btrfs_free_path(path);
4299         return ret;
4300 }
4301
4302 /*
4303  * helper to find a free sequence number in a given directory.  This current
4304  * code is very simple, later versions will do smarter things in the btree
4305  */
4306 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4307 {
4308         int ret = 0;
4309
4310         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4311                 ret = btrfs_inode_delayed_dir_index_count(dir);
4312                 if (ret) {
4313                         ret = btrfs_set_inode_index_count(dir);
4314                         if (ret)
4315                                 return ret;
4316                 }
4317         }
4318
4319         *index = BTRFS_I(dir)->index_cnt;
4320         BTRFS_I(dir)->index_cnt++;
4321
4322         return ret;
4323 }
4324
4325 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4326                                      struct btrfs_root *root,
4327                                      struct inode *dir,
4328                                      const char *name, int name_len,
4329                                      u64 ref_objectid, u64 objectid, int mode,
4330                                      u64 *index)
4331 {
4332         struct inode *inode;
4333         struct btrfs_inode_item *inode_item;
4334         struct btrfs_key *location;
4335         struct btrfs_path *path;
4336         struct btrfs_inode_ref *ref;
4337         struct btrfs_key key[2];
4338         u32 sizes[2];
4339         unsigned long ptr;
4340         int ret;
4341         int owner;
4342
4343         path = btrfs_alloc_path();
4344         if (!path)
4345                 return ERR_PTR(-ENOMEM);
4346
4347         inode = new_inode(root->fs_info->sb);
4348         if (!inode) {
4349                 btrfs_free_path(path);
4350                 return ERR_PTR(-ENOMEM);
4351         }
4352
4353         /*
4354          * we have to initialize this early, so we can reclaim the inode
4355          * number if we fail afterwards in this function.
4356          */
4357         inode->i_ino = objectid;
4358
4359         if (dir) {
4360                 trace_btrfs_inode_request(dir);
4361
4362                 ret = btrfs_set_inode_index(dir, index);
4363                 if (ret) {
4364                         btrfs_free_path(path);
4365                         iput(inode);
4366                         return ERR_PTR(ret);
4367                 }
4368         }
4369         /*
4370          * index_cnt is ignored for everything but a dir,
4371          * btrfs_get_inode_index_count has an explanation for the magic
4372          * number
4373          */
4374         BTRFS_I(inode)->index_cnt = 2;
4375         BTRFS_I(inode)->root = root;
4376         BTRFS_I(inode)->generation = trans->transid;
4377         inode->i_generation = BTRFS_I(inode)->generation;
4378         btrfs_set_inode_space_info(root, inode);
4379
4380         if (S_ISDIR(mode))
4381                 owner = 0;
4382         else
4383                 owner = 1;
4384
4385         key[0].objectid = objectid;
4386         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4387         key[0].offset = 0;
4388
4389         key[1].objectid = objectid;
4390         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4391         key[1].offset = ref_objectid;
4392
4393         sizes[0] = sizeof(struct btrfs_inode_item);
4394         sizes[1] = name_len + sizeof(*ref);
4395
4396         path->leave_spinning = 1;
4397         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4398         if (ret != 0)
4399                 goto fail;
4400
4401         inode_init_owner(inode, dir, mode);
4402         inode_set_bytes(inode, 0);
4403         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4404         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4405                                   struct btrfs_inode_item);
4406         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4407
4408         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4409                              struct btrfs_inode_ref);
4410         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4411         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4412         ptr = (unsigned long)(ref + 1);
4413         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4414
4415         btrfs_mark_buffer_dirty(path->nodes[0]);
4416         btrfs_free_path(path);
4417
4418         location = &BTRFS_I(inode)->location;
4419         location->objectid = objectid;
4420         location->offset = 0;
4421         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4422
4423         btrfs_inherit_iflags(inode, dir);
4424
4425         if (S_ISREG(mode)) {
4426                 if (btrfs_test_opt(root, NODATASUM))
4427                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4428                 if (btrfs_test_opt(root, NODATACOW) ||
4429                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4430                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4431         }
4432
4433         insert_inode_hash(inode);
4434         inode_tree_add(inode);
4435
4436         trace_btrfs_inode_new(inode);
4437         btrfs_set_inode_last_trans(trans, inode);
4438
4439         return inode;
4440 fail:
4441         if (dir)
4442                 BTRFS_I(dir)->index_cnt--;
4443         btrfs_free_path(path);
4444         iput(inode);
4445         return ERR_PTR(ret);
4446 }
4447
4448 static inline u8 btrfs_inode_type(struct inode *inode)
4449 {
4450         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4451 }
4452
4453 /*
4454  * utility function to add 'inode' into 'parent_inode' with
4455  * a give name and a given sequence number.
4456  * if 'add_backref' is true, also insert a backref from the
4457  * inode to the parent directory.
4458  */
4459 int btrfs_add_link(struct btrfs_trans_handle *trans,
4460                    struct inode *parent_inode, struct inode *inode,
4461                    const char *name, int name_len, int add_backref, u64 index)
4462 {
4463         int ret = 0;
4464         struct btrfs_key key;
4465         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4466         u64 ino = btrfs_ino(inode);
4467         u64 parent_ino = btrfs_ino(parent_inode);
4468
4469         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4470                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4471         } else {
4472                 key.objectid = ino;
4473                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4474                 key.offset = 0;
4475         }
4476
4477         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4478                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4479                                          key.objectid, root->root_key.objectid,
4480                                          parent_ino, index, name, name_len);
4481         } else if (add_backref) {
4482                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4483                                              parent_ino, index);
4484         }
4485
4486         if (ret == 0) {
4487                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4488                                             parent_inode, &key,
4489                                             btrfs_inode_type(inode), index);
4490                 BUG_ON(ret);
4491
4492                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4493                                    name_len * 2);
4494                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4495                 ret = btrfs_update_inode(trans, root, parent_inode);
4496         }
4497         return ret;
4498 }
4499
4500 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4501                             struct inode *dir, struct dentry *dentry,
4502                             struct inode *inode, int backref, u64 index)
4503 {
4504         int err = btrfs_add_link(trans, dir, inode,
4505                                  dentry->d_name.name, dentry->d_name.len,
4506                                  backref, index);
4507         if (!err) {
4508                 d_instantiate(dentry, inode);
4509                 return 0;
4510         }
4511         if (err > 0)
4512                 err = -EEXIST;
4513         return err;
4514 }
4515
4516 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4517                         int mode, dev_t rdev)
4518 {
4519         struct btrfs_trans_handle *trans;
4520         struct btrfs_root *root = BTRFS_I(dir)->root;
4521         struct inode *inode = NULL;
4522         int err;
4523         int drop_inode = 0;
4524         u64 objectid;
4525         unsigned long nr = 0;
4526         u64 index = 0;
4527
4528         if (!new_valid_dev(rdev))
4529                 return -EINVAL;
4530
4531         /*
4532          * 2 for inode item and ref
4533          * 2 for dir items
4534          * 1 for xattr if selinux is on
4535          */
4536         trans = btrfs_start_transaction(root, 5);
4537         if (IS_ERR(trans))
4538                 return PTR_ERR(trans);
4539
4540         err = btrfs_find_free_ino(root, &objectid);
4541         if (err)
4542                 goto out_unlock;
4543
4544         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4545                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4546                                 mode, &index);
4547         if (IS_ERR(inode)) {
4548                 err = PTR_ERR(inode);
4549                 goto out_unlock;
4550         }
4551
4552         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4553         if (err) {
4554                 drop_inode = 1;
4555                 goto out_unlock;
4556         }
4557
4558         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4559         if (err)
4560                 drop_inode = 1;
4561         else {
4562                 inode->i_op = &btrfs_special_inode_operations;
4563                 init_special_inode(inode, inode->i_mode, rdev);
4564                 btrfs_update_inode(trans, root, inode);
4565         }
4566 out_unlock:
4567         nr = trans->blocks_used;
4568         btrfs_end_transaction_throttle(trans, root);
4569         btrfs_btree_balance_dirty(root, nr);
4570         if (drop_inode) {
4571                 inode_dec_link_count(inode);
4572                 iput(inode);
4573         }
4574         return err;
4575 }
4576
4577 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4578                         int mode, struct nameidata *nd)
4579 {
4580         struct btrfs_trans_handle *trans;
4581         struct btrfs_root *root = BTRFS_I(dir)->root;
4582         struct inode *inode = NULL;
4583         int drop_inode = 0;
4584         int err;
4585         unsigned long nr = 0;
4586         u64 objectid;
4587         u64 index = 0;
4588
4589         /*
4590          * 2 for inode item and ref
4591          * 2 for dir items
4592          * 1 for xattr if selinux is on
4593          */
4594         trans = btrfs_start_transaction(root, 5);
4595         if (IS_ERR(trans))
4596                 return PTR_ERR(trans);
4597
4598         err = btrfs_find_free_ino(root, &objectid);
4599         if (err)
4600                 goto out_unlock;
4601
4602         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4603                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4604                                 mode, &index);
4605         if (IS_ERR(inode)) {
4606                 err = PTR_ERR(inode);
4607                 goto out_unlock;
4608         }
4609
4610         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4611         if (err) {
4612                 drop_inode = 1;
4613                 goto out_unlock;
4614         }
4615
4616         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4617         if (err)
4618                 drop_inode = 1;
4619         else {
4620                 inode->i_mapping->a_ops = &btrfs_aops;
4621                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4622                 inode->i_fop = &btrfs_file_operations;
4623                 inode->i_op = &btrfs_file_inode_operations;
4624                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4625         }
4626 out_unlock:
4627         nr = trans->blocks_used;
4628         btrfs_end_transaction_throttle(trans, root);
4629         if (drop_inode) {
4630                 inode_dec_link_count(inode);
4631                 iput(inode);
4632         }
4633         btrfs_btree_balance_dirty(root, nr);
4634         return err;
4635 }
4636
4637 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4638                       struct dentry *dentry)
4639 {
4640         struct btrfs_trans_handle *trans;
4641         struct btrfs_root *root = BTRFS_I(dir)->root;
4642         struct inode *inode = old_dentry->d_inode;
4643         u64 index;
4644         unsigned long nr = 0;
4645         int err;
4646         int drop_inode = 0;
4647
4648         /* do not allow sys_link's with other subvols of the same device */
4649         if (root->objectid != BTRFS_I(inode)->root->objectid)
4650                 return -EXDEV;
4651
4652         if (inode->i_nlink == ~0U)
4653                 return -EMLINK;
4654
4655         err = btrfs_set_inode_index(dir, &index);
4656         if (err)
4657                 goto fail;
4658
4659         /*
4660          * 2 items for inode and inode ref
4661          * 2 items for dir items
4662          * 1 item for parent inode
4663          */
4664         trans = btrfs_start_transaction(root, 5);
4665         if (IS_ERR(trans)) {
4666                 err = PTR_ERR(trans);
4667                 goto fail;
4668         }
4669
4670         btrfs_inc_nlink(inode);
4671         inode->i_ctime = CURRENT_TIME;
4672         ihold(inode);
4673
4674         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4675
4676         if (err) {
4677                 drop_inode = 1;
4678         } else {
4679                 struct dentry *parent = dentry->d_parent;
4680                 err = btrfs_update_inode(trans, root, inode);
4681                 BUG_ON(err);
4682                 btrfs_log_new_name(trans, inode, NULL, parent);
4683         }
4684
4685         nr = trans->blocks_used;
4686         btrfs_end_transaction_throttle(trans, root);
4687 fail:
4688         if (drop_inode) {
4689                 inode_dec_link_count(inode);
4690                 iput(inode);
4691         }
4692         btrfs_btree_balance_dirty(root, nr);
4693         return err;
4694 }
4695
4696 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4697 {
4698         struct inode *inode = NULL;
4699         struct btrfs_trans_handle *trans;
4700         struct btrfs_root *root = BTRFS_I(dir)->root;
4701         int err = 0;
4702         int drop_on_err = 0;
4703         u64 objectid = 0;
4704         u64 index = 0;
4705         unsigned long nr = 1;
4706
4707         /*
4708          * 2 items for inode and ref
4709          * 2 items for dir items
4710          * 1 for xattr if selinux is on
4711          */
4712         trans = btrfs_start_transaction(root, 5);
4713         if (IS_ERR(trans))
4714                 return PTR_ERR(trans);
4715
4716         err = btrfs_find_free_ino(root, &objectid);
4717         if (err)
4718                 goto out_fail;
4719
4720         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4721                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4722                                 S_IFDIR | mode, &index);
4723         if (IS_ERR(inode)) {
4724                 err = PTR_ERR(inode);
4725                 goto out_fail;
4726         }
4727
4728         drop_on_err = 1;
4729
4730         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4731         if (err)
4732                 goto out_fail;
4733
4734         inode->i_op = &btrfs_dir_inode_operations;
4735         inode->i_fop = &btrfs_dir_file_operations;
4736
4737         btrfs_i_size_write(inode, 0);
4738         err = btrfs_update_inode(trans, root, inode);
4739         if (err)
4740                 goto out_fail;
4741
4742         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4743                              dentry->d_name.len, 0, index);
4744         if (err)
4745                 goto out_fail;
4746
4747         d_instantiate(dentry, inode);
4748         drop_on_err = 0;
4749
4750 out_fail:
4751         nr = trans->blocks_used;
4752         btrfs_end_transaction_throttle(trans, root);
4753         if (drop_on_err)
4754                 iput(inode);
4755         btrfs_btree_balance_dirty(root, nr);
4756         return err;
4757 }
4758
4759 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4760  * and an extent that you want to insert, deal with overlap and insert
4761  * the new extent into the tree.
4762  */
4763 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4764                                 struct extent_map *existing,
4765                                 struct extent_map *em,
4766                                 u64 map_start, u64 map_len)
4767 {
4768         u64 start_diff;
4769
4770         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4771         start_diff = map_start - em->start;
4772         em->start = map_start;
4773         em->len = map_len;
4774         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4775             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4776                 em->block_start += start_diff;
4777                 em->block_len -= start_diff;
4778         }
4779         return add_extent_mapping(em_tree, em);
4780 }
4781
4782 static noinline int uncompress_inline(struct btrfs_path *path,
4783                                       struct inode *inode, struct page *page,
4784                                       size_t pg_offset, u64 extent_offset,
4785                                       struct btrfs_file_extent_item *item)
4786 {
4787         int ret;
4788         struct extent_buffer *leaf = path->nodes[0];
4789         char *tmp;
4790         size_t max_size;
4791         unsigned long inline_size;
4792         unsigned long ptr;
4793         int compress_type;
4794
4795         WARN_ON(pg_offset != 0);
4796         compress_type = btrfs_file_extent_compression(leaf, item);
4797         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4798         inline_size = btrfs_file_extent_inline_item_len(leaf,
4799                                         btrfs_item_nr(leaf, path->slots[0]));
4800         tmp = kmalloc(inline_size, GFP_NOFS);
4801         if (!tmp)
4802                 return -ENOMEM;
4803         ptr = btrfs_file_extent_inline_start(item);
4804
4805         read_extent_buffer(leaf, tmp, ptr, inline_size);
4806
4807         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4808         ret = btrfs_decompress(compress_type, tmp, page,
4809                                extent_offset, inline_size, max_size);
4810         if (ret) {
4811                 char *kaddr = kmap_atomic(page, KM_USER0);
4812                 unsigned long copy_size = min_t(u64,
4813                                   PAGE_CACHE_SIZE - pg_offset,
4814                                   max_size - extent_offset);
4815                 memset(kaddr + pg_offset, 0, copy_size);
4816                 kunmap_atomic(kaddr, KM_USER0);
4817         }
4818         kfree(tmp);
4819         return 0;
4820 }
4821
4822 /*
4823  * a bit scary, this does extent mapping from logical file offset to the disk.
4824  * the ugly parts come from merging extents from the disk with the in-ram
4825  * representation.  This gets more complex because of the data=ordered code,
4826  * where the in-ram extents might be locked pending data=ordered completion.
4827  *
4828  * This also copies inline extents directly into the page.
4829  */
4830
4831 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4832                                     size_t pg_offset, u64 start, u64 len,
4833                                     int create)
4834 {
4835         int ret;
4836         int err = 0;
4837         u64 bytenr;
4838         u64 extent_start = 0;
4839         u64 extent_end = 0;
4840         u64 objectid = btrfs_ino(inode);
4841         u32 found_type;
4842         struct btrfs_path *path = NULL;
4843         struct btrfs_root *root = BTRFS_I(inode)->root;
4844         struct btrfs_file_extent_item *item;
4845         struct extent_buffer *leaf;
4846         struct btrfs_key found_key;
4847         struct extent_map *em = NULL;
4848         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4849         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4850         struct btrfs_trans_handle *trans = NULL;
4851         int compress_type;
4852
4853 again:
4854         read_lock(&em_tree->lock);
4855         em = lookup_extent_mapping(em_tree, start, len);
4856         if (em)
4857                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4858         read_unlock(&em_tree->lock);
4859
4860         if (em) {
4861                 if (em->start > start || em->start + em->len <= start)
4862                         free_extent_map(em);
4863                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4864                         free_extent_map(em);
4865                 else
4866                         goto out;
4867         }
4868         em = alloc_extent_map();
4869         if (!em) {
4870                 err = -ENOMEM;
4871                 goto out;
4872         }
4873         em->bdev = root->fs_info->fs_devices->latest_bdev;
4874         em->start = EXTENT_MAP_HOLE;
4875         em->orig_start = EXTENT_MAP_HOLE;
4876         em->len = (u64)-1;
4877         em->block_len = (u64)-1;
4878
4879         if (!path) {
4880                 path = btrfs_alloc_path();
4881                 if (!path) {
4882                         err = -ENOMEM;
4883                         goto out;
4884                 }
4885                 /*
4886                  * Chances are we'll be called again, so go ahead and do
4887                  * readahead
4888                  */
4889                 path->reada = 1;
4890         }
4891
4892         ret = btrfs_lookup_file_extent(trans, root, path,
4893                                        objectid, start, trans != NULL);
4894         if (ret < 0) {
4895                 err = ret;
4896                 goto out;
4897         }
4898
4899         if (ret != 0) {
4900                 if (path->slots[0] == 0)
4901                         goto not_found;
4902                 path->slots[0]--;
4903         }
4904
4905         leaf = path->nodes[0];
4906         item = btrfs_item_ptr(leaf, path->slots[0],
4907                               struct btrfs_file_extent_item);
4908         /* are we inside the extent that was found? */
4909         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4910         found_type = btrfs_key_type(&found_key);
4911         if (found_key.objectid != objectid ||
4912             found_type != BTRFS_EXTENT_DATA_KEY) {
4913                 goto not_found;
4914         }
4915
4916         found_type = btrfs_file_extent_type(leaf, item);
4917         extent_start = found_key.offset;
4918         compress_type = btrfs_file_extent_compression(leaf, item);
4919         if (found_type == BTRFS_FILE_EXTENT_REG ||
4920             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4921                 extent_end = extent_start +
4922                        btrfs_file_extent_num_bytes(leaf, item);
4923         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4924                 size_t size;
4925                 size = btrfs_file_extent_inline_len(leaf, item);
4926                 extent_end = (extent_start + size + root->sectorsize - 1) &
4927                         ~((u64)root->sectorsize - 1);
4928         }
4929
4930         if (start >= extent_end) {
4931                 path->slots[0]++;
4932                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4933                         ret = btrfs_next_leaf(root, path);
4934                         if (ret < 0) {
4935                                 err = ret;
4936                                 goto out;
4937                         }
4938                         if (ret > 0)
4939                                 goto not_found;
4940                         leaf = path->nodes[0];
4941                 }
4942                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4943                 if (found_key.objectid != objectid ||
4944                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4945                         goto not_found;
4946                 if (start + len <= found_key.offset)
4947                         goto not_found;
4948                 em->start = start;
4949                 em->len = found_key.offset - start;
4950                 goto not_found_em;
4951         }
4952
4953         if (found_type == BTRFS_FILE_EXTENT_REG ||
4954             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4955                 em->start = extent_start;
4956                 em->len = extent_end - extent_start;
4957                 em->orig_start = extent_start -
4958                                  btrfs_file_extent_offset(leaf, item);
4959                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4960                 if (bytenr == 0) {
4961                         em->block_start = EXTENT_MAP_HOLE;
4962                         goto insert;
4963                 }
4964                 if (compress_type != BTRFS_COMPRESS_NONE) {
4965                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4966                         em->compress_type = compress_type;
4967                         em->block_start = bytenr;
4968                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4969                                                                          item);
4970                 } else {
4971                         bytenr += btrfs_file_extent_offset(leaf, item);
4972                         em->block_start = bytenr;
4973                         em->block_len = em->len;
4974                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4975                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4976                 }
4977                 goto insert;
4978         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4979                 unsigned long ptr;
4980                 char *map;
4981                 size_t size;
4982                 size_t extent_offset;
4983                 size_t copy_size;
4984
4985                 em->block_start = EXTENT_MAP_INLINE;
4986                 if (!page || create) {
4987                         em->start = extent_start;
4988                         em->len = extent_end - extent_start;
4989                         goto out;
4990                 }
4991
4992                 size = btrfs_file_extent_inline_len(leaf, item);
4993                 extent_offset = page_offset(page) + pg_offset - extent_start;
4994                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4995                                 size - extent_offset);
4996                 em->start = extent_start + extent_offset;
4997                 em->len = (copy_size + root->sectorsize - 1) &
4998                         ~((u64)root->sectorsize - 1);
4999                 em->orig_start = EXTENT_MAP_INLINE;
5000                 if (compress_type) {
5001                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5002                         em->compress_type = compress_type;
5003                 }
5004                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5005                 if (create == 0 && !PageUptodate(page)) {
5006                         if (btrfs_file_extent_compression(leaf, item) !=
5007                             BTRFS_COMPRESS_NONE) {
5008                                 ret = uncompress_inline(path, inode, page,
5009                                                         pg_offset,
5010                                                         extent_offset, item);
5011                                 BUG_ON(ret);
5012                         } else {
5013                                 map = kmap(page);
5014                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5015                                                    copy_size);
5016                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5017                                         memset(map + pg_offset + copy_size, 0,
5018                                                PAGE_CACHE_SIZE - pg_offset -
5019                                                copy_size);
5020                                 }
5021                                 kunmap(page);
5022                         }
5023                         flush_dcache_page(page);
5024                 } else if (create && PageUptodate(page)) {
5025                         WARN_ON(1);
5026                         if (!trans) {
5027                                 kunmap(page);
5028                                 free_extent_map(em);
5029                                 em = NULL;
5030
5031                                 btrfs_release_path(path);
5032                                 trans = btrfs_join_transaction(root);
5033
5034                                 if (IS_ERR(trans))
5035                                         return ERR_CAST(trans);
5036                                 goto again;
5037                         }
5038                         map = kmap(page);
5039                         write_extent_buffer(leaf, map + pg_offset, ptr,
5040                                             copy_size);
5041                         kunmap(page);
5042                         btrfs_mark_buffer_dirty(leaf);
5043                 }
5044                 set_extent_uptodate(io_tree, em->start,
5045                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5046                 goto insert;
5047         } else {
5048                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5049                 WARN_ON(1);
5050         }
5051 not_found:
5052         em->start = start;
5053         em->len = len;
5054 not_found_em:
5055         em->block_start = EXTENT_MAP_HOLE;
5056         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5057 insert:
5058         btrfs_release_path(path);
5059         if (em->start > start || extent_map_end(em) <= start) {
5060                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5061                        "[%llu %llu]\n", (unsigned long long)em->start,
5062                        (unsigned long long)em->len,
5063                        (unsigned long long)start,
5064                        (unsigned long long)len);
5065                 err = -EIO;
5066                 goto out;
5067         }
5068
5069         err = 0;
5070         write_lock(&em_tree->lock);
5071         ret = add_extent_mapping(em_tree, em);
5072         /* it is possible that someone inserted the extent into the tree
5073          * while we had the lock dropped.  It is also possible that
5074          * an overlapping map exists in the tree
5075          */
5076         if (ret == -EEXIST) {
5077                 struct extent_map *existing;
5078
5079                 ret = 0;
5080
5081                 existing = lookup_extent_mapping(em_tree, start, len);
5082                 if (existing && (existing->start > start ||
5083                     existing->start + existing->len <= start)) {
5084                         free_extent_map(existing);
5085                         existing = NULL;
5086                 }
5087                 if (!existing) {
5088                         existing = lookup_extent_mapping(em_tree, em->start,
5089                                                          em->len);
5090                         if (existing) {
5091                                 err = merge_extent_mapping(em_tree, existing,
5092                                                            em, start,
5093                                                            root->sectorsize);
5094                                 free_extent_map(existing);
5095                                 if (err) {
5096                                         free_extent_map(em);
5097                                         em = NULL;
5098                                 }
5099                         } else {
5100                                 err = -EIO;
5101                                 free_extent_map(em);
5102                                 em = NULL;
5103                         }
5104                 } else {
5105                         free_extent_map(em);
5106                         em = existing;
5107                         err = 0;
5108                 }
5109         }
5110         write_unlock(&em_tree->lock);
5111 out:
5112
5113         trace_btrfs_get_extent(root, em);
5114
5115         if (path)
5116                 btrfs_free_path(path);
5117         if (trans) {
5118                 ret = btrfs_end_transaction(trans, root);
5119                 if (!err)
5120                         err = ret;
5121         }
5122         if (err) {
5123                 free_extent_map(em);
5124                 return ERR_PTR(err);
5125         }
5126         return em;
5127 }
5128
5129 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5130                                            size_t pg_offset, u64 start, u64 len,
5131                                            int create)
5132 {
5133         struct extent_map *em;
5134         struct extent_map *hole_em = NULL;
5135         u64 range_start = start;
5136         u64 end;
5137         u64 found;
5138         u64 found_end;
5139         int err = 0;
5140
5141         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5142         if (IS_ERR(em))
5143                 return em;
5144         if (em) {
5145                 /*
5146                  * if our em maps to a hole, there might
5147                  * actually be delalloc bytes behind it
5148                  */
5149                 if (em->block_start != EXTENT_MAP_HOLE)
5150                         return em;
5151                 else
5152                         hole_em = em;
5153         }
5154
5155         /* check to see if we've wrapped (len == -1 or similar) */
5156         end = start + len;
5157         if (end < start)
5158                 end = (u64)-1;
5159         else
5160                 end -= 1;
5161
5162         em = NULL;
5163
5164         /* ok, we didn't find anything, lets look for delalloc */
5165         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5166                                  end, len, EXTENT_DELALLOC, 1);
5167         found_end = range_start + found;
5168         if (found_end < range_start)
5169                 found_end = (u64)-1;
5170
5171         /*
5172          * we didn't find anything useful, return
5173          * the original results from get_extent()
5174          */
5175         if (range_start > end || found_end <= start) {
5176                 em = hole_em;
5177                 hole_em = NULL;
5178                 goto out;
5179         }
5180
5181         /* adjust the range_start to make sure it doesn't
5182          * go backwards from the start they passed in
5183          */
5184         range_start = max(start,range_start);
5185         found = found_end - range_start;
5186
5187         if (found > 0) {
5188                 u64 hole_start = start;
5189                 u64 hole_len = len;
5190
5191                 em = alloc_extent_map();
5192                 if (!em) {
5193                         err = -ENOMEM;
5194                         goto out;
5195                 }
5196                 /*
5197                  * when btrfs_get_extent can't find anything it
5198                  * returns one huge hole
5199                  *
5200                  * make sure what it found really fits our range, and
5201                  * adjust to make sure it is based on the start from
5202                  * the caller
5203                  */
5204                 if (hole_em) {
5205                         u64 calc_end = extent_map_end(hole_em);
5206
5207                         if (calc_end <= start || (hole_em->start > end)) {
5208                                 free_extent_map(hole_em);
5209                                 hole_em = NULL;
5210                         } else {
5211                                 hole_start = max(hole_em->start, start);
5212                                 hole_len = calc_end - hole_start;
5213                         }
5214                 }
5215                 em->bdev = NULL;
5216                 if (hole_em && range_start > hole_start) {
5217                         /* our hole starts before our delalloc, so we
5218                          * have to return just the parts of the hole
5219                          * that go until  the delalloc starts
5220                          */
5221                         em->len = min(hole_len,
5222                                       range_start - hole_start);
5223                         em->start = hole_start;
5224                         em->orig_start = hole_start;
5225                         /*
5226                          * don't adjust block start at all,
5227                          * it is fixed at EXTENT_MAP_HOLE
5228                          */
5229                         em->block_start = hole_em->block_start;
5230                         em->block_len = hole_len;
5231                 } else {
5232                         em->start = range_start;
5233                         em->len = found;
5234                         em->orig_start = range_start;
5235                         em->block_start = EXTENT_MAP_DELALLOC;
5236                         em->block_len = found;
5237                 }
5238         } else if (hole_em) {
5239                 return hole_em;
5240         }
5241 out:
5242
5243         free_extent_map(hole_em);
5244         if (err) {
5245                 free_extent_map(em);
5246                 return ERR_PTR(err);
5247         }
5248         return em;
5249 }
5250
5251 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5252                                                   struct extent_map *em,
5253                                                   u64 start, u64 len)
5254 {
5255         struct btrfs_root *root = BTRFS_I(inode)->root;
5256         struct btrfs_trans_handle *trans;
5257         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5258         struct btrfs_key ins;
5259         u64 alloc_hint;
5260         int ret;
5261         bool insert = false;
5262
5263         /*
5264          * Ok if the extent map we looked up is a hole and is for the exact
5265          * range we want, there is no reason to allocate a new one, however if
5266          * it is not right then we need to free this one and drop the cache for
5267          * our range.
5268          */
5269         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5270             em->len != len) {
5271                 free_extent_map(em);
5272                 em = NULL;
5273                 insert = true;
5274                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5275         }
5276
5277         trans = btrfs_join_transaction(root);
5278         if (IS_ERR(trans))
5279                 return ERR_CAST(trans);
5280
5281         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5282                 btrfs_add_inode_defrag(trans, inode);
5283
5284         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5285
5286         alloc_hint = get_extent_allocation_hint(inode, start, len);
5287         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5288                                    alloc_hint, (u64)-1, &ins, 1);
5289         if (ret) {
5290                 em = ERR_PTR(ret);
5291                 goto out;
5292         }
5293
5294         if (!em) {
5295                 em = alloc_extent_map();
5296                 if (!em) {
5297                         em = ERR_PTR(-ENOMEM);
5298                         goto out;
5299                 }
5300         }
5301
5302         em->start = start;
5303         em->orig_start = em->start;
5304         em->len = ins.offset;
5305
5306         em->block_start = ins.objectid;
5307         em->block_len = ins.offset;
5308         em->bdev = root->fs_info->fs_devices->latest_bdev;
5309
5310         /*
5311          * We need to do this because if we're using the original em we searched
5312          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5313          */
5314         em->flags = 0;
5315         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5316
5317         while (insert) {
5318                 write_lock(&em_tree->lock);
5319                 ret = add_extent_mapping(em_tree, em);
5320                 write_unlock(&em_tree->lock);
5321                 if (ret != -EEXIST)
5322                         break;
5323                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5324         }
5325
5326         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5327                                            ins.offset, ins.offset, 0);
5328         if (ret) {
5329                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5330                 em = ERR_PTR(ret);
5331         }
5332 out:
5333         btrfs_end_transaction(trans, root);
5334         return em;
5335 }
5336
5337 /*
5338  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5339  * block must be cow'd
5340  */
5341 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5342                                       struct inode *inode, u64 offset, u64 len)
5343 {
5344         struct btrfs_path *path;
5345         int ret;
5346         struct extent_buffer *leaf;
5347         struct btrfs_root *root = BTRFS_I(inode)->root;
5348         struct btrfs_file_extent_item *fi;
5349         struct btrfs_key key;
5350         u64 disk_bytenr;
5351         u64 backref_offset;
5352         u64 extent_end;
5353         u64 num_bytes;
5354         int slot;
5355         int found_type;
5356
5357         path = btrfs_alloc_path();
5358         if (!path)
5359                 return -ENOMEM;
5360
5361         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5362                                        offset, 0);
5363         if (ret < 0)
5364                 goto out;
5365
5366         slot = path->slots[0];
5367         if (ret == 1) {
5368                 if (slot == 0) {
5369                         /* can't find the item, must cow */
5370                         ret = 0;
5371                         goto out;
5372                 }
5373                 slot--;
5374         }
5375         ret = 0;
5376         leaf = path->nodes[0];
5377         btrfs_item_key_to_cpu(leaf, &key, slot);
5378         if (key.objectid != btrfs_ino(inode) ||
5379             key.type != BTRFS_EXTENT_DATA_KEY) {
5380                 /* not our file or wrong item type, must cow */
5381                 goto out;
5382         }
5383
5384         if (key.offset > offset) {
5385                 /* Wrong offset, must cow */
5386                 goto out;
5387         }
5388
5389         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5390         found_type = btrfs_file_extent_type(leaf, fi);
5391         if (found_type != BTRFS_FILE_EXTENT_REG &&
5392             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5393                 /* not a regular extent, must cow */
5394                 goto out;
5395         }
5396         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5397         backref_offset = btrfs_file_extent_offset(leaf, fi);
5398
5399         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5400         if (extent_end < offset + len) {
5401                 /* extent doesn't include our full range, must cow */
5402                 goto out;
5403         }
5404
5405         if (btrfs_extent_readonly(root, disk_bytenr))
5406                 goto out;
5407
5408         /*
5409          * look for other files referencing this extent, if we
5410          * find any we must cow
5411          */
5412         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5413                                   key.offset - backref_offset, disk_bytenr))
5414                 goto out;
5415
5416         /*
5417          * adjust disk_bytenr and num_bytes to cover just the bytes
5418          * in this extent we are about to write.  If there
5419          * are any csums in that range we have to cow in order
5420          * to keep the csums correct
5421          */
5422         disk_bytenr += backref_offset;
5423         disk_bytenr += offset - key.offset;
5424         num_bytes = min(offset + len, extent_end) - offset;
5425         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5426                                 goto out;
5427         /*
5428          * all of the above have passed, it is safe to overwrite this extent
5429          * without cow
5430          */
5431         ret = 1;
5432 out:
5433         btrfs_free_path(path);
5434         return ret;
5435 }
5436
5437 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5438                                    struct buffer_head *bh_result, int create)
5439 {
5440         struct extent_map *em;
5441         struct btrfs_root *root = BTRFS_I(inode)->root;
5442         u64 start = iblock << inode->i_blkbits;
5443         u64 len = bh_result->b_size;
5444         struct btrfs_trans_handle *trans;
5445
5446         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5447         if (IS_ERR(em))
5448                 return PTR_ERR(em);
5449
5450         /*
5451          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5452          * io.  INLINE is special, and we could probably kludge it in here, but
5453          * it's still buffered so for safety lets just fall back to the generic
5454          * buffered path.
5455          *
5456          * For COMPRESSED we _have_ to read the entire extent in so we can
5457          * decompress it, so there will be buffering required no matter what we
5458          * do, so go ahead and fallback to buffered.
5459          *
5460          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5461          * to buffered IO.  Don't blame me, this is the price we pay for using
5462          * the generic code.
5463          */
5464         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5465             em->block_start == EXTENT_MAP_INLINE) {
5466                 free_extent_map(em);
5467                 return -ENOTBLK;
5468         }
5469
5470         /* Just a good old fashioned hole, return */
5471         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5472                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5473                 free_extent_map(em);
5474                 /* DIO will do one hole at a time, so just unlock a sector */
5475                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5476                               start + root->sectorsize - 1, GFP_NOFS);
5477                 return 0;
5478         }
5479
5480         /*
5481          * We don't allocate a new extent in the following cases
5482          *
5483          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5484          * existing extent.
5485          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5486          * just use the extent.
5487          *
5488          */
5489         if (!create) {
5490                 len = em->len - (start - em->start);
5491                 goto map;
5492         }
5493
5494         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5495             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5496              em->block_start != EXTENT_MAP_HOLE)) {
5497                 int type;
5498                 int ret;
5499                 u64 block_start;
5500
5501                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5502                         type = BTRFS_ORDERED_PREALLOC;
5503                 else
5504                         type = BTRFS_ORDERED_NOCOW;
5505                 len = min(len, em->len - (start - em->start));
5506                 block_start = em->block_start + (start - em->start);
5507
5508                 /*
5509                  * we're not going to log anything, but we do need
5510                  * to make sure the current transaction stays open
5511                  * while we look for nocow cross refs
5512                  */
5513                 trans = btrfs_join_transaction(root);
5514                 if (IS_ERR(trans))
5515                         goto must_cow;
5516
5517                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5518                         ret = btrfs_add_ordered_extent_dio(inode, start,
5519                                            block_start, len, len, type);
5520                         btrfs_end_transaction(trans, root);
5521                         if (ret) {
5522                                 free_extent_map(em);
5523                                 return ret;
5524                         }
5525                         goto unlock;
5526                 }
5527                 btrfs_end_transaction(trans, root);
5528         }
5529 must_cow:
5530         /*
5531          * this will cow the extent, reset the len in case we changed
5532          * it above
5533          */
5534         len = bh_result->b_size;
5535         em = btrfs_new_extent_direct(inode, em, start, len);
5536         if (IS_ERR(em))
5537                 return PTR_ERR(em);
5538         len = min(len, em->len - (start - em->start));
5539 unlock:
5540         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5541                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5542                           0, NULL, GFP_NOFS);
5543 map:
5544         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5545                 inode->i_blkbits;
5546         bh_result->b_size = len;
5547         bh_result->b_bdev = em->bdev;
5548         set_buffer_mapped(bh_result);
5549         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5550                 set_buffer_new(bh_result);
5551
5552         free_extent_map(em);
5553
5554         return 0;
5555 }
5556
5557 struct btrfs_dio_private {
5558         struct inode *inode;
5559         u64 logical_offset;
5560         u64 disk_bytenr;
5561         u64 bytes;
5562         u32 *csums;
5563         void *private;
5564
5565         /* number of bios pending for this dio */
5566         atomic_t pending_bios;
5567
5568         /* IO errors */
5569         int errors;
5570
5571         struct bio *orig_bio;
5572 };
5573
5574 static void btrfs_endio_direct_read(struct bio *bio, int err)
5575 {
5576         struct btrfs_dio_private *dip = bio->bi_private;
5577         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5578         struct bio_vec *bvec = bio->bi_io_vec;
5579         struct inode *inode = dip->inode;
5580         struct btrfs_root *root = BTRFS_I(inode)->root;
5581         u64 start;
5582         u32 *private = dip->csums;
5583
5584         start = dip->logical_offset;
5585         do {
5586                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5587                         struct page *page = bvec->bv_page;
5588                         char *kaddr;
5589                         u32 csum = ~(u32)0;
5590                         unsigned long flags;
5591
5592                         local_irq_save(flags);
5593                         kaddr = kmap_atomic(page, KM_IRQ0);
5594                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5595                                                csum, bvec->bv_len);
5596                         btrfs_csum_final(csum, (char *)&csum);
5597                         kunmap_atomic(kaddr, KM_IRQ0);
5598                         local_irq_restore(flags);
5599
5600                         flush_dcache_page(bvec->bv_page);
5601                         if (csum != *private) {
5602                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5603                                       " %llu csum %u private %u\n",
5604                                       (unsigned long long)btrfs_ino(inode),
5605                                       (unsigned long long)start,
5606                                       csum, *private);
5607                                 err = -EIO;
5608                         }
5609                 }
5610
5611                 start += bvec->bv_len;
5612                 private++;
5613                 bvec++;
5614         } while (bvec <= bvec_end);
5615
5616         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5617                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5618         bio->bi_private = dip->private;
5619
5620         kfree(dip->csums);
5621         kfree(dip);
5622
5623         /* If we had a csum failure make sure to clear the uptodate flag */
5624         if (err)
5625                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5626         dio_end_io(bio, err);
5627 }
5628
5629 static void btrfs_endio_direct_write(struct bio *bio, int err)
5630 {
5631         struct btrfs_dio_private *dip = bio->bi_private;
5632         struct inode *inode = dip->inode;
5633         struct btrfs_root *root = BTRFS_I(inode)->root;
5634         struct btrfs_trans_handle *trans;
5635         struct btrfs_ordered_extent *ordered = NULL;
5636         struct extent_state *cached_state = NULL;
5637         u64 ordered_offset = dip->logical_offset;
5638         u64 ordered_bytes = dip->bytes;
5639         int ret;
5640
5641         if (err)
5642                 goto out_done;
5643 again:
5644         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5645                                                    &ordered_offset,
5646                                                    ordered_bytes);
5647         if (!ret)
5648                 goto out_test;
5649
5650         BUG_ON(!ordered);
5651
5652         trans = btrfs_join_transaction(root);
5653         if (IS_ERR(trans)) {
5654                 err = -ENOMEM;
5655                 goto out;
5656         }
5657         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5658
5659         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5660                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5661                 if (!ret)
5662                         err = btrfs_update_inode_fallback(trans, root, inode);
5663                 goto out;
5664         }
5665
5666         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5667                          ordered->file_offset + ordered->len - 1, 0,
5668                          &cached_state, GFP_NOFS);
5669
5670         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5671                 ret = btrfs_mark_extent_written(trans, inode,
5672                                                 ordered->file_offset,
5673                                                 ordered->file_offset +
5674                                                 ordered->len);
5675                 if (ret) {
5676                         err = ret;
5677                         goto out_unlock;
5678                 }
5679         } else {
5680                 ret = insert_reserved_file_extent(trans, inode,
5681                                                   ordered->file_offset,
5682                                                   ordered->start,
5683                                                   ordered->disk_len,
5684                                                   ordered->len,
5685                                                   ordered->len,
5686                                                   0, 0, 0,
5687                                                   BTRFS_FILE_EXTENT_REG);
5688                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5689                                    ordered->file_offset, ordered->len);
5690                 if (ret) {
5691                         err = ret;
5692                         WARN_ON(1);
5693                         goto out_unlock;
5694                 }
5695         }
5696
5697         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5698         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5699         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5700                 btrfs_update_inode_fallback(trans, root, inode);
5701         ret = 0;
5702 out_unlock:
5703         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5704                              ordered->file_offset + ordered->len - 1,
5705                              &cached_state, GFP_NOFS);
5706 out:
5707         btrfs_delalloc_release_metadata(inode, ordered->len);
5708         btrfs_end_transaction(trans, root);
5709         ordered_offset = ordered->file_offset + ordered->len;
5710         btrfs_put_ordered_extent(ordered);
5711         btrfs_put_ordered_extent(ordered);
5712
5713 out_test:
5714         /*
5715          * our bio might span multiple ordered extents.  If we haven't
5716          * completed the accounting for the whole dio, go back and try again
5717          */
5718         if (ordered_offset < dip->logical_offset + dip->bytes) {
5719                 ordered_bytes = dip->logical_offset + dip->bytes -
5720                         ordered_offset;
5721                 goto again;
5722         }
5723 out_done:
5724         bio->bi_private = dip->private;
5725
5726         kfree(dip->csums);
5727         kfree(dip);
5728
5729         /* If we had an error make sure to clear the uptodate flag */
5730         if (err)
5731                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5732         dio_end_io(bio, err);
5733 }
5734
5735 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5736                                     struct bio *bio, int mirror_num,
5737                                     unsigned long bio_flags, u64 offset)
5738 {
5739         int ret;
5740         struct btrfs_root *root = BTRFS_I(inode)->root;
5741         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5742         BUG_ON(ret);
5743         return 0;
5744 }
5745
5746 static void btrfs_end_dio_bio(struct bio *bio, int err)
5747 {
5748         struct btrfs_dio_private *dip = bio->bi_private;
5749
5750         if (err) {
5751                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5752                       "sector %#Lx len %u err no %d\n",
5753                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5754                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5755                 dip->errors = 1;
5756
5757                 /*
5758                  * before atomic variable goto zero, we must make sure
5759                  * dip->errors is perceived to be set.
5760                  */
5761                 smp_mb__before_atomic_dec();
5762         }
5763
5764         /* if there are more bios still pending for this dio, just exit */
5765         if (!atomic_dec_and_test(&dip->pending_bios))
5766                 goto out;
5767
5768         if (dip->errors)
5769                 bio_io_error(dip->orig_bio);
5770         else {
5771                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5772                 bio_endio(dip->orig_bio, 0);
5773         }
5774 out:
5775         bio_put(bio);
5776 }
5777
5778 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5779                                        u64 first_sector, gfp_t gfp_flags)
5780 {
5781         int nr_vecs = bio_get_nr_vecs(bdev);
5782         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5783 }
5784
5785 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5786                                          int rw, u64 file_offset, int skip_sum,
5787                                          u32 *csums, int async_submit)
5788 {
5789         int write = rw & REQ_WRITE;
5790         struct btrfs_root *root = BTRFS_I(inode)->root;
5791         int ret;
5792
5793         bio_get(bio);
5794         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5795         if (ret)
5796                 goto err;
5797
5798         if (skip_sum)
5799                 goto map;
5800
5801         if (write && async_submit) {
5802                 ret = btrfs_wq_submit_bio(root->fs_info,
5803                                    inode, rw, bio, 0, 0,
5804                                    file_offset,
5805                                    __btrfs_submit_bio_start_direct_io,
5806                                    __btrfs_submit_bio_done);
5807                 goto err;
5808         } else if (write) {
5809                 /*
5810                  * If we aren't doing async submit, calculate the csum of the
5811                  * bio now.
5812                  */
5813                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5814                 if (ret)
5815                         goto err;
5816         } else if (!skip_sum) {
5817                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5818                                           file_offset, csums);
5819                 if (ret)
5820                         goto err;
5821         }
5822
5823 map:
5824         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5825 err:
5826         bio_put(bio);
5827         return ret;
5828 }
5829
5830 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5831                                     int skip_sum)
5832 {
5833         struct inode *inode = dip->inode;
5834         struct btrfs_root *root = BTRFS_I(inode)->root;
5835         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5836         struct bio *bio;
5837         struct bio *orig_bio = dip->orig_bio;
5838         struct bio_vec *bvec = orig_bio->bi_io_vec;
5839         u64 start_sector = orig_bio->bi_sector;
5840         u64 file_offset = dip->logical_offset;
5841         u64 submit_len = 0;
5842         u64 map_length;
5843         int nr_pages = 0;
5844         u32 *csums = dip->csums;
5845         int ret = 0;
5846         int async_submit = 0;
5847         int write = rw & REQ_WRITE;
5848
5849         map_length = orig_bio->bi_size;
5850         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5851                               &map_length, NULL, 0);
5852         if (ret) {
5853                 bio_put(orig_bio);
5854                 return -EIO;
5855         }
5856
5857         if (map_length >= orig_bio->bi_size) {
5858                 bio = orig_bio;
5859                 goto submit;
5860         }
5861
5862         async_submit = 1;
5863         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5864         if (!bio)
5865                 return -ENOMEM;
5866         bio->bi_private = dip;
5867         bio->bi_end_io = btrfs_end_dio_bio;
5868         atomic_inc(&dip->pending_bios);
5869
5870         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5871                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5872                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5873                                  bvec->bv_offset) < bvec->bv_len)) {
5874                         /*
5875                          * inc the count before we submit the bio so
5876                          * we know the end IO handler won't happen before
5877                          * we inc the count. Otherwise, the dip might get freed
5878                          * before we're done setting it up
5879                          */
5880                         atomic_inc(&dip->pending_bios);
5881                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5882                                                      file_offset, skip_sum,
5883                                                      csums, async_submit);
5884                         if (ret) {
5885                                 bio_put(bio);
5886                                 atomic_dec(&dip->pending_bios);
5887                                 goto out_err;
5888                         }
5889
5890                         /* Write's use the ordered csums */
5891                         if (!write && !skip_sum)
5892                                 csums = csums + nr_pages;
5893                         start_sector += submit_len >> 9;
5894                         file_offset += submit_len;
5895
5896                         submit_len = 0;
5897                         nr_pages = 0;
5898
5899                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5900                                                   start_sector, GFP_NOFS);
5901                         if (!bio)
5902                                 goto out_err;
5903                         bio->bi_private = dip;
5904                         bio->bi_end_io = btrfs_end_dio_bio;
5905
5906                         map_length = orig_bio->bi_size;
5907                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5908                                               &map_length, NULL, 0);
5909                         if (ret) {
5910                                 bio_put(bio);
5911                                 goto out_err;
5912                         }
5913                 } else {
5914                         submit_len += bvec->bv_len;
5915                         nr_pages ++;
5916                         bvec++;
5917                 }
5918         }
5919
5920 submit:
5921         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
5922                                      csums, async_submit);
5923         if (!ret)
5924                 return 0;
5925
5926         bio_put(bio);
5927 out_err:
5928         dip->errors = 1;
5929         /*
5930          * before atomic variable goto zero, we must
5931          * make sure dip->errors is perceived to be set.
5932          */
5933         smp_mb__before_atomic_dec();
5934         if (atomic_dec_and_test(&dip->pending_bios))
5935                 bio_io_error(dip->orig_bio);
5936
5937         /* bio_end_io() will handle error, so we needn't return it */
5938         return 0;
5939 }
5940
5941 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5942                                 loff_t file_offset)
5943 {
5944         struct btrfs_root *root = BTRFS_I(inode)->root;
5945         struct btrfs_dio_private *dip;
5946         struct bio_vec *bvec = bio->bi_io_vec;
5947         int skip_sum;
5948         int write = rw & REQ_WRITE;
5949         int ret = 0;
5950
5951         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5952
5953         dip = kmalloc(sizeof(*dip), GFP_NOFS);
5954         if (!dip) {
5955                 ret = -ENOMEM;
5956                 goto free_ordered;
5957         }
5958         dip->csums = NULL;
5959
5960         /* Write's use the ordered csum stuff, so we don't need dip->csums */
5961         if (!write && !skip_sum) {
5962                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5963                 if (!dip->csums) {
5964                         kfree(dip);
5965                         ret = -ENOMEM;
5966                         goto free_ordered;
5967                 }
5968         }
5969
5970         dip->private = bio->bi_private;
5971         dip->inode = inode;
5972         dip->logical_offset = file_offset;
5973
5974         dip->bytes = 0;
5975         do {
5976                 dip->bytes += bvec->bv_len;
5977                 bvec++;
5978         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5979
5980         dip->disk_bytenr = (u64)bio->bi_sector << 9;
5981         bio->bi_private = dip;
5982         dip->errors = 0;
5983         dip->orig_bio = bio;
5984         atomic_set(&dip->pending_bios, 0);
5985
5986         if (write)
5987                 bio->bi_end_io = btrfs_endio_direct_write;
5988         else
5989                 bio->bi_end_io = btrfs_endio_direct_read;
5990
5991         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
5992         if (!ret)
5993                 return;
5994 free_ordered:
5995         /*
5996          * If this is a write, we need to clean up the reserved space and kill
5997          * the ordered extent.
5998          */
5999         if (write) {
6000                 struct btrfs_ordered_extent *ordered;
6001                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6002                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6003                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6004                         btrfs_free_reserved_extent(root, ordered->start,
6005                                                    ordered->disk_len);
6006                 btrfs_put_ordered_extent(ordered);
6007                 btrfs_put_ordered_extent(ordered);
6008         }
6009         bio_endio(bio, ret);
6010 }
6011
6012 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6013                         const struct iovec *iov, loff_t offset,
6014                         unsigned long nr_segs)
6015 {
6016         int seg;
6017         int i;
6018         size_t size;
6019         unsigned long addr;
6020         unsigned blocksize_mask = root->sectorsize - 1;
6021         ssize_t retval = -EINVAL;
6022         loff_t end = offset;
6023
6024         if (offset & blocksize_mask)
6025                 goto out;
6026
6027         /* Check the memory alignment.  Blocks cannot straddle pages */
6028         for (seg = 0; seg < nr_segs; seg++) {
6029                 addr = (unsigned long)iov[seg].iov_base;
6030                 size = iov[seg].iov_len;
6031                 end += size;
6032                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6033                         goto out;
6034
6035                 /* If this is a write we don't need to check anymore */
6036                 if (rw & WRITE)
6037                         continue;
6038
6039                 /*
6040                  * Check to make sure we don't have duplicate iov_base's in this
6041                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6042                  * when reading back.
6043                  */
6044                 for (i = seg + 1; i < nr_segs; i++) {
6045                         if (iov[seg].iov_base == iov[i].iov_base)
6046                                 goto out;
6047                 }
6048         }
6049         retval = 0;
6050 out:
6051         return retval;
6052 }
6053 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6054                         const struct iovec *iov, loff_t offset,
6055                         unsigned long nr_segs)
6056 {
6057         struct file *file = iocb->ki_filp;
6058         struct inode *inode = file->f_mapping->host;
6059         struct btrfs_ordered_extent *ordered;
6060         struct extent_state *cached_state = NULL;
6061         u64 lockstart, lockend;
6062         ssize_t ret;
6063         int writing = rw & WRITE;
6064         int write_bits = 0;
6065         size_t count = iov_length(iov, nr_segs);
6066
6067         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6068                             offset, nr_segs)) {
6069                 return 0;
6070         }
6071
6072         lockstart = offset;
6073         lockend = offset + count - 1;
6074
6075         if (writing) {
6076                 ret = btrfs_delalloc_reserve_space(inode, count);
6077                 if (ret)
6078                         goto out;
6079         }
6080
6081         while (1) {
6082                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6083                                  0, &cached_state, GFP_NOFS);
6084                 /*
6085                  * We're concerned with the entire range that we're going to be
6086                  * doing DIO to, so we need to make sure theres no ordered
6087                  * extents in this range.
6088                  */
6089                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6090                                                      lockend - lockstart + 1);
6091                 if (!ordered)
6092                         break;
6093                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6094                                      &cached_state, GFP_NOFS);
6095                 btrfs_start_ordered_extent(inode, ordered, 1);
6096                 btrfs_put_ordered_extent(ordered);
6097                 cond_resched();
6098         }
6099
6100         /*
6101          * we don't use btrfs_set_extent_delalloc because we don't want
6102          * the dirty or uptodate bits
6103          */
6104         if (writing) {
6105                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6106                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6107                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6108                                      GFP_NOFS);
6109                 if (ret) {
6110                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6111                                          lockend, EXTENT_LOCKED | write_bits,
6112                                          1, 0, &cached_state, GFP_NOFS);
6113                         goto out;
6114                 }
6115         }
6116
6117         free_extent_state(cached_state);
6118         cached_state = NULL;
6119
6120         ret = __blockdev_direct_IO(rw, iocb, inode,
6121                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6122                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6123                    btrfs_submit_direct, 0);
6124
6125         if (ret < 0 && ret != -EIOCBQUEUED) {
6126                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6127                               offset + iov_length(iov, nr_segs) - 1,
6128                               EXTENT_LOCKED | write_bits, 1, 0,
6129                               &cached_state, GFP_NOFS);
6130         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6131                 /*
6132                  * We're falling back to buffered, unlock the section we didn't
6133                  * do IO on.
6134                  */
6135                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6136                               offset + iov_length(iov, nr_segs) - 1,
6137                               EXTENT_LOCKED | write_bits, 1, 0,
6138                               &cached_state, GFP_NOFS);
6139         }
6140 out:
6141         free_extent_state(cached_state);
6142         return ret;
6143 }
6144
6145 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6146                 __u64 start, __u64 len)
6147 {
6148         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6149 }
6150
6151 int btrfs_readpage(struct file *file, struct page *page)
6152 {
6153         struct extent_io_tree *tree;
6154         tree = &BTRFS_I(page->mapping->host)->io_tree;
6155         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6156 }
6157
6158 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6159 {
6160         struct extent_io_tree *tree;
6161
6162
6163         if (current->flags & PF_MEMALLOC) {
6164                 redirty_page_for_writepage(wbc, page);
6165                 unlock_page(page);
6166                 return 0;
6167         }
6168         tree = &BTRFS_I(page->mapping->host)->io_tree;
6169         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6170 }
6171
6172 int btrfs_writepages(struct address_space *mapping,
6173                      struct writeback_control *wbc)
6174 {
6175         struct extent_io_tree *tree;
6176
6177         tree = &BTRFS_I(mapping->host)->io_tree;
6178         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6179 }
6180
6181 static int
6182 btrfs_readpages(struct file *file, struct address_space *mapping,
6183                 struct list_head *pages, unsigned nr_pages)
6184 {
6185         struct extent_io_tree *tree;
6186         tree = &BTRFS_I(mapping->host)->io_tree;
6187         return extent_readpages(tree, mapping, pages, nr_pages,
6188                                 btrfs_get_extent);
6189 }
6190 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6191 {
6192         struct extent_io_tree *tree;
6193         struct extent_map_tree *map;
6194         int ret;
6195
6196         tree = &BTRFS_I(page->mapping->host)->io_tree;
6197         map = &BTRFS_I(page->mapping->host)->extent_tree;
6198         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6199         if (ret == 1) {
6200                 ClearPagePrivate(page);
6201                 set_page_private(page, 0);
6202                 page_cache_release(page);
6203         }
6204         return ret;
6205 }
6206
6207 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6208 {
6209         if (PageWriteback(page) || PageDirty(page))
6210                 return 0;
6211         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6212 }
6213
6214 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6215 {
6216         struct extent_io_tree *tree;
6217         struct btrfs_ordered_extent *ordered;
6218         struct extent_state *cached_state = NULL;
6219         u64 page_start = page_offset(page);
6220         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6221
6222
6223         /*
6224          * we have the page locked, so new writeback can't start,
6225          * and the dirty bit won't be cleared while we are here.
6226          *
6227          * Wait for IO on this page so that we can safely clear
6228          * the PagePrivate2 bit and do ordered accounting
6229          */
6230         wait_on_page_writeback(page);
6231
6232         tree = &BTRFS_I(page->mapping->host)->io_tree;
6233         if (offset) {
6234                 btrfs_releasepage(page, GFP_NOFS);
6235                 return;
6236         }
6237         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6238                          GFP_NOFS);
6239         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6240                                            page_offset(page));
6241         if (ordered) {
6242                 /*
6243                  * IO on this page will never be started, so we need
6244                  * to account for any ordered extents now
6245                  */
6246                 clear_extent_bit(tree, page_start, page_end,
6247                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6248                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6249                                  &cached_state, GFP_NOFS);
6250                 /*
6251                  * whoever cleared the private bit is responsible
6252                  * for the finish_ordered_io
6253                  */
6254                 if (TestClearPagePrivate2(page)) {
6255                         btrfs_finish_ordered_io(page->mapping->host,
6256                                                 page_start, page_end);
6257                 }
6258                 btrfs_put_ordered_extent(ordered);
6259                 cached_state = NULL;
6260                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6261                                  GFP_NOFS);
6262         }
6263         clear_extent_bit(tree, page_start, page_end,
6264                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6265                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6266         __btrfs_releasepage(page, GFP_NOFS);
6267
6268         ClearPageChecked(page);
6269         if (PagePrivate(page)) {
6270                 ClearPagePrivate(page);
6271                 set_page_private(page, 0);
6272                 page_cache_release(page);
6273         }
6274 }
6275
6276 /*
6277  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6278  * called from a page fault handler when a page is first dirtied. Hence we must
6279  * be careful to check for EOF conditions here. We set the page up correctly
6280  * for a written page which means we get ENOSPC checking when writing into
6281  * holes and correct delalloc and unwritten extent mapping on filesystems that
6282  * support these features.
6283  *
6284  * We are not allowed to take the i_mutex here so we have to play games to
6285  * protect against truncate races as the page could now be beyond EOF.  Because
6286  * vmtruncate() writes the inode size before removing pages, once we have the
6287  * page lock we can determine safely if the page is beyond EOF. If it is not
6288  * beyond EOF, then the page is guaranteed safe against truncation until we
6289  * unlock the page.
6290  */
6291 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6292 {
6293         struct page *page = vmf->page;
6294         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6295         struct btrfs_root *root = BTRFS_I(inode)->root;
6296         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6297         struct btrfs_ordered_extent *ordered;
6298         struct extent_state *cached_state = NULL;
6299         char *kaddr;
6300         unsigned long zero_start;
6301         loff_t size;
6302         int ret;
6303         u64 page_start;
6304         u64 page_end;
6305
6306         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6307         if (ret) {
6308                 if (ret == -ENOMEM)
6309                         ret = VM_FAULT_OOM;
6310                 else /* -ENOSPC, -EIO, etc */
6311                         ret = VM_FAULT_SIGBUS;
6312                 goto out;
6313         }
6314
6315         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6316 again:
6317         lock_page(page);
6318         size = i_size_read(inode);
6319         page_start = page_offset(page);
6320         page_end = page_start + PAGE_CACHE_SIZE - 1;
6321
6322         if ((page->mapping != inode->i_mapping) ||
6323             (page_start >= size)) {
6324                 /* page got truncated out from underneath us */
6325                 goto out_unlock;
6326         }
6327         wait_on_page_writeback(page);
6328
6329         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6330                          GFP_NOFS);
6331         set_page_extent_mapped(page);
6332
6333         /*
6334          * we can't set the delalloc bits if there are pending ordered
6335          * extents.  Drop our locks and wait for them to finish
6336          */
6337         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6338         if (ordered) {
6339                 unlock_extent_cached(io_tree, page_start, page_end,
6340                                      &cached_state, GFP_NOFS);
6341                 unlock_page(page);
6342                 btrfs_start_ordered_extent(inode, ordered, 1);
6343                 btrfs_put_ordered_extent(ordered);
6344                 goto again;
6345         }
6346
6347         /*
6348          * XXX - page_mkwrite gets called every time the page is dirtied, even
6349          * if it was already dirty, so for space accounting reasons we need to
6350          * clear any delalloc bits for the range we are fixing to save.  There
6351          * is probably a better way to do this, but for now keep consistent with
6352          * prepare_pages in the normal write path.
6353          */
6354         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6355                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6356                           0, 0, &cached_state, GFP_NOFS);
6357
6358         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6359                                         &cached_state);
6360         if (ret) {
6361                 unlock_extent_cached(io_tree, page_start, page_end,
6362                                      &cached_state, GFP_NOFS);
6363                 ret = VM_FAULT_SIGBUS;
6364                 goto out_unlock;
6365         }
6366         ret = 0;
6367
6368         /* page is wholly or partially inside EOF */
6369         if (page_start + PAGE_CACHE_SIZE > size)
6370                 zero_start = size & ~PAGE_CACHE_MASK;
6371         else
6372                 zero_start = PAGE_CACHE_SIZE;
6373
6374         if (zero_start != PAGE_CACHE_SIZE) {
6375                 kaddr = kmap(page);
6376                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6377                 flush_dcache_page(page);
6378                 kunmap(page);
6379         }
6380         ClearPageChecked(page);
6381         set_page_dirty(page);
6382         SetPageUptodate(page);
6383
6384         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6385         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6386
6387         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6388
6389 out_unlock:
6390         if (!ret)
6391                 return VM_FAULT_LOCKED;
6392         unlock_page(page);
6393         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6394 out:
6395         return ret;
6396 }
6397
6398 static int btrfs_truncate(struct inode *inode)
6399 {
6400         struct btrfs_root *root = BTRFS_I(inode)->root;
6401         struct btrfs_block_rsv *rsv;
6402         int ret;
6403         int err = 0;
6404         struct btrfs_trans_handle *trans;
6405         unsigned long nr;
6406         u64 mask = root->sectorsize - 1;
6407         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6408
6409         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6410         if (ret)
6411                 return ret;
6412
6413         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6414         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6415
6416         /*
6417          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6418          * 3 things going on here
6419          *
6420          * 1) We need to reserve space for our orphan item and the space to
6421          * delete our orphan item.  Lord knows we don't want to have a dangling
6422          * orphan item because we didn't reserve space to remove it.
6423          *
6424          * 2) We need to reserve space to update our inode.
6425          *
6426          * 3) We need to have something to cache all the space that is going to
6427          * be free'd up by the truncate operation, but also have some slack
6428          * space reserved in case it uses space during the truncate (thank you
6429          * very much snapshotting).
6430          *
6431          * And we need these to all be seperate.  The fact is we can use alot of
6432          * space doing the truncate, and we have no earthly idea how much space
6433          * we will use, so we need the truncate reservation to be seperate so it
6434          * doesn't end up using space reserved for updating the inode or
6435          * removing the orphan item.  We also need to be able to stop the
6436          * transaction and start a new one, which means we need to be able to
6437          * update the inode several times, and we have no idea of knowing how
6438          * many times that will be, so we can't just reserve 1 item for the
6439          * entirety of the opration, so that has to be done seperately as well.
6440          * Then there is the orphan item, which does indeed need to be held on
6441          * to for the whole operation, and we need nobody to touch this reserved
6442          * space except the orphan code.
6443          *
6444          * So that leaves us with
6445          *
6446          * 1) root->orphan_block_rsv - for the orphan deletion.
6447          * 2) rsv - for the truncate reservation, which we will steal from the
6448          * transaction reservation.
6449          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6450          * updating the inode.
6451          */
6452         rsv = btrfs_alloc_block_rsv(root);
6453         if (!rsv)
6454                 return -ENOMEM;
6455         rsv->size = min_size;
6456
6457         /*
6458          * 1 for the truncate slack space
6459          * 1 for the orphan item we're going to add
6460          * 1 for the orphan item deletion
6461          * 1 for updating the inode.
6462          */
6463         trans = btrfs_start_transaction(root, 4);
6464         if (IS_ERR(trans)) {
6465                 err = PTR_ERR(trans);
6466                 goto out;
6467         }
6468
6469         /* Migrate the slack space for the truncate to our reserve */
6470         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6471                                       min_size);
6472         BUG_ON(ret);
6473
6474         ret = btrfs_orphan_add(trans, inode);
6475         if (ret) {
6476                 btrfs_end_transaction(trans, root);
6477                 goto out;
6478         }
6479
6480         /*
6481          * setattr is responsible for setting the ordered_data_close flag,
6482          * but that is only tested during the last file release.  That
6483          * could happen well after the next commit, leaving a great big
6484          * window where new writes may get lost if someone chooses to write
6485          * to this file after truncating to zero
6486          *
6487          * The inode doesn't have any dirty data here, and so if we commit
6488          * this is a noop.  If someone immediately starts writing to the inode
6489          * it is very likely we'll catch some of their writes in this
6490          * transaction, and the commit will find this file on the ordered
6491          * data list with good things to send down.
6492          *
6493          * This is a best effort solution, there is still a window where
6494          * using truncate to replace the contents of the file will
6495          * end up with a zero length file after a crash.
6496          */
6497         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6498                 btrfs_add_ordered_operation(trans, root, inode);
6499
6500         while (1) {
6501                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6502                 if (ret) {
6503                         /*
6504                          * This can only happen with the original transaction we
6505                          * started above, every other time we shouldn't have a
6506                          * transaction started yet.
6507                          */
6508                         if (ret == -EAGAIN)
6509                                 goto end_trans;
6510                         err = ret;
6511                         break;
6512                 }
6513
6514                 if (!trans) {
6515                         /* Just need the 1 for updating the inode */
6516                         trans = btrfs_start_transaction(root, 1);
6517                         if (IS_ERR(trans)) {
6518                                 err = PTR_ERR(trans);
6519                                 goto out;
6520                         }
6521                 }
6522
6523                 trans->block_rsv = rsv;
6524
6525                 ret = btrfs_truncate_inode_items(trans, root, inode,
6526                                                  inode->i_size,
6527                                                  BTRFS_EXTENT_DATA_KEY);
6528                 if (ret != -EAGAIN) {
6529                         err = ret;
6530                         break;
6531                 }
6532
6533                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6534                 ret = btrfs_update_inode(trans, root, inode);
6535                 if (ret) {
6536                         err = ret;
6537                         break;
6538                 }
6539 end_trans:
6540                 nr = trans->blocks_used;
6541                 btrfs_end_transaction(trans, root);
6542                 trans = NULL;
6543                 btrfs_btree_balance_dirty(root, nr);
6544         }
6545
6546         if (ret == 0 && inode->i_nlink > 0) {
6547                 trans->block_rsv = root->orphan_block_rsv;
6548                 ret = btrfs_orphan_del(trans, inode);
6549                 if (ret)
6550                         err = ret;
6551         } else if (ret && inode->i_nlink > 0) {
6552                 /*
6553                  * Failed to do the truncate, remove us from the in memory
6554                  * orphan list.
6555                  */
6556                 ret = btrfs_orphan_del(NULL, inode);
6557         }
6558
6559         if (trans) {
6560                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6561                 ret = btrfs_update_inode(trans, root, inode);
6562                 if (ret && !err)
6563                         err = ret;
6564
6565                 nr = trans->blocks_used;
6566                 ret = btrfs_end_transaction_throttle(trans, root);
6567                 btrfs_btree_balance_dirty(root, nr);
6568         }
6569
6570 out:
6571         btrfs_free_block_rsv(root, rsv);
6572
6573         if (ret && !err)
6574                 err = ret;
6575
6576         return err;
6577 }
6578
6579 /*
6580  * create a new subvolume directory/inode (helper for the ioctl).
6581  */
6582 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6583                              struct btrfs_root *new_root, u64 new_dirid)
6584 {
6585         struct inode *inode;
6586         int err;
6587         u64 index = 0;
6588
6589         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6590                                 new_dirid, S_IFDIR | 0700, &index);
6591         if (IS_ERR(inode))
6592                 return PTR_ERR(inode);
6593         inode->i_op = &btrfs_dir_inode_operations;
6594         inode->i_fop = &btrfs_dir_file_operations;
6595
6596         set_nlink(inode, 1);
6597         btrfs_i_size_write(inode, 0);
6598
6599         err = btrfs_update_inode(trans, new_root, inode);
6600         BUG_ON(err);
6601
6602         iput(inode);
6603         return 0;
6604 }
6605
6606 struct inode *btrfs_alloc_inode(struct super_block *sb)
6607 {
6608         struct btrfs_inode *ei;
6609         struct inode *inode;
6610
6611         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6612         if (!ei)
6613                 return NULL;
6614
6615         ei->root = NULL;
6616         ei->space_info = NULL;
6617         ei->generation = 0;
6618         ei->sequence = 0;
6619         ei->last_trans = 0;
6620         ei->last_sub_trans = 0;
6621         ei->logged_trans = 0;
6622         ei->delalloc_bytes = 0;
6623         ei->disk_i_size = 0;
6624         ei->flags = 0;
6625         ei->csum_bytes = 0;
6626         ei->index_cnt = (u64)-1;
6627         ei->last_unlink_trans = 0;
6628
6629         spin_lock_init(&ei->lock);
6630         ei->outstanding_extents = 0;
6631         ei->reserved_extents = 0;
6632
6633         ei->ordered_data_close = 0;
6634         ei->orphan_meta_reserved = 0;
6635         ei->dummy_inode = 0;
6636         ei->in_defrag = 0;
6637         ei->delalloc_meta_reserved = 0;
6638         ei->force_compress = BTRFS_COMPRESS_NONE;
6639
6640         ei->delayed_node = NULL;
6641
6642         inode = &ei->vfs_inode;
6643         extent_map_tree_init(&ei->extent_tree);
6644         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6645         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6646         mutex_init(&ei->log_mutex);
6647         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6648         INIT_LIST_HEAD(&ei->i_orphan);
6649         INIT_LIST_HEAD(&ei->delalloc_inodes);
6650         INIT_LIST_HEAD(&ei->ordered_operations);
6651         RB_CLEAR_NODE(&ei->rb_node);
6652
6653         return inode;
6654 }
6655
6656 static void btrfs_i_callback(struct rcu_head *head)
6657 {
6658         struct inode *inode = container_of(head, struct inode, i_rcu);
6659         INIT_LIST_HEAD(&inode->i_dentry);
6660         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6661 }
6662
6663 void btrfs_destroy_inode(struct inode *inode)
6664 {
6665         struct btrfs_ordered_extent *ordered;
6666         struct btrfs_root *root = BTRFS_I(inode)->root;
6667
6668         WARN_ON(!list_empty(&inode->i_dentry));
6669         WARN_ON(inode->i_data.nrpages);
6670         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6671         WARN_ON(BTRFS_I(inode)->reserved_extents);
6672         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6673         WARN_ON(BTRFS_I(inode)->csum_bytes);
6674
6675         /*
6676          * This can happen where we create an inode, but somebody else also
6677          * created the same inode and we need to destroy the one we already
6678          * created.
6679          */
6680         if (!root)
6681                 goto free;
6682
6683         /*
6684          * Make sure we're properly removed from the ordered operation
6685          * lists.
6686          */
6687         smp_mb();
6688         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6689                 spin_lock(&root->fs_info->ordered_extent_lock);
6690                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6691                 spin_unlock(&root->fs_info->ordered_extent_lock);
6692         }
6693
6694         spin_lock(&root->orphan_lock);
6695         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6696                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6697                        (unsigned long long)btrfs_ino(inode));
6698                 list_del_init(&BTRFS_I(inode)->i_orphan);
6699         }
6700         spin_unlock(&root->orphan_lock);
6701
6702         while (1) {
6703                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6704                 if (!ordered)
6705                         break;
6706                 else {
6707                         printk(KERN_ERR "btrfs found ordered "
6708                                "extent %llu %llu on inode cleanup\n",
6709                                (unsigned long long)ordered->file_offset,
6710                                (unsigned long long)ordered->len);
6711                         btrfs_remove_ordered_extent(inode, ordered);
6712                         btrfs_put_ordered_extent(ordered);
6713                         btrfs_put_ordered_extent(ordered);
6714                 }
6715         }
6716         inode_tree_del(inode);
6717         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6718 free:
6719         btrfs_remove_delayed_node(inode);
6720         call_rcu(&inode->i_rcu, btrfs_i_callback);
6721 }
6722
6723 int btrfs_drop_inode(struct inode *inode)
6724 {
6725         struct btrfs_root *root = BTRFS_I(inode)->root;
6726
6727         if (btrfs_root_refs(&root->root_item) == 0 &&
6728             !btrfs_is_free_space_inode(root, inode))
6729                 return 1;
6730         else
6731                 return generic_drop_inode(inode);
6732 }
6733
6734 static void init_once(void *foo)
6735 {
6736         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6737
6738         inode_init_once(&ei->vfs_inode);
6739 }
6740
6741 void btrfs_destroy_cachep(void)
6742 {
6743         if (btrfs_inode_cachep)
6744                 kmem_cache_destroy(btrfs_inode_cachep);
6745         if (btrfs_trans_handle_cachep)
6746                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6747         if (btrfs_transaction_cachep)
6748                 kmem_cache_destroy(btrfs_transaction_cachep);
6749         if (btrfs_path_cachep)
6750                 kmem_cache_destroy(btrfs_path_cachep);
6751         if (btrfs_free_space_cachep)
6752                 kmem_cache_destroy(btrfs_free_space_cachep);
6753 }
6754
6755 int btrfs_init_cachep(void)
6756 {
6757         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6758                         sizeof(struct btrfs_inode), 0,
6759                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6760         if (!btrfs_inode_cachep)
6761                 goto fail;
6762
6763         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6764                         sizeof(struct btrfs_trans_handle), 0,
6765                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6766         if (!btrfs_trans_handle_cachep)
6767                 goto fail;
6768
6769         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6770                         sizeof(struct btrfs_transaction), 0,
6771                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6772         if (!btrfs_transaction_cachep)
6773                 goto fail;
6774
6775         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6776                         sizeof(struct btrfs_path), 0,
6777                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6778         if (!btrfs_path_cachep)
6779                 goto fail;
6780
6781         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6782                         sizeof(struct btrfs_free_space), 0,
6783                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6784         if (!btrfs_free_space_cachep)
6785                 goto fail;
6786
6787         return 0;
6788 fail:
6789         btrfs_destroy_cachep();
6790         return -ENOMEM;
6791 }
6792
6793 static int btrfs_getattr(struct vfsmount *mnt,
6794                          struct dentry *dentry, struct kstat *stat)
6795 {
6796         struct inode *inode = dentry->d_inode;
6797         generic_fillattr(inode, stat);
6798         stat->dev = BTRFS_I(inode)->root->anon_dev;
6799         stat->blksize = PAGE_CACHE_SIZE;
6800         stat->blocks = (inode_get_bytes(inode) +
6801                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6802         return 0;
6803 }
6804
6805 /*
6806  * If a file is moved, it will inherit the cow and compression flags of the new
6807  * directory.
6808  */
6809 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6810 {
6811         struct btrfs_inode *b_dir = BTRFS_I(dir);
6812         struct btrfs_inode *b_inode = BTRFS_I(inode);
6813
6814         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6815                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6816         else
6817                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6818
6819         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6820                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6821         else
6822                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6823 }
6824
6825 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6826                            struct inode *new_dir, struct dentry *new_dentry)
6827 {
6828         struct btrfs_trans_handle *trans;
6829         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6830         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6831         struct inode *new_inode = new_dentry->d_inode;
6832         struct inode *old_inode = old_dentry->d_inode;
6833         struct timespec ctime = CURRENT_TIME;
6834         u64 index = 0;
6835         u64 root_objectid;
6836         int ret;
6837         u64 old_ino = btrfs_ino(old_inode);
6838
6839         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6840                 return -EPERM;
6841
6842         /* we only allow rename subvolume link between subvolumes */
6843         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6844                 return -EXDEV;
6845
6846         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6847             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6848                 return -ENOTEMPTY;
6849
6850         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6851             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6852                 return -ENOTEMPTY;
6853         /*
6854          * we're using rename to replace one file with another.
6855          * and the replacement file is large.  Start IO on it now so
6856          * we don't add too much work to the end of the transaction
6857          */
6858         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6859             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6860                 filemap_flush(old_inode->i_mapping);
6861
6862         /* close the racy window with snapshot create/destroy ioctl */
6863         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6864                 down_read(&root->fs_info->subvol_sem);
6865         /*
6866          * We want to reserve the absolute worst case amount of items.  So if
6867          * both inodes are subvols and we need to unlink them then that would
6868          * require 4 item modifications, but if they are both normal inodes it
6869          * would require 5 item modifications, so we'll assume their normal
6870          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6871          * should cover the worst case number of items we'll modify.
6872          */
6873         trans = btrfs_start_transaction(root, 20);
6874         if (IS_ERR(trans)) {
6875                 ret = PTR_ERR(trans);
6876                 goto out_notrans;
6877         }
6878
6879         if (dest != root)
6880                 btrfs_record_root_in_trans(trans, dest);
6881
6882         ret = btrfs_set_inode_index(new_dir, &index);
6883         if (ret)
6884                 goto out_fail;
6885
6886         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6887                 /* force full log commit if subvolume involved. */
6888                 root->fs_info->last_trans_log_full_commit = trans->transid;
6889         } else {
6890                 ret = btrfs_insert_inode_ref(trans, dest,
6891                                              new_dentry->d_name.name,
6892                                              new_dentry->d_name.len,
6893                                              old_ino,
6894                                              btrfs_ino(new_dir), index);
6895                 if (ret)
6896                         goto out_fail;
6897                 /*
6898                  * this is an ugly little race, but the rename is required
6899                  * to make sure that if we crash, the inode is either at the
6900                  * old name or the new one.  pinning the log transaction lets
6901                  * us make sure we don't allow a log commit to come in after
6902                  * we unlink the name but before we add the new name back in.
6903                  */
6904                 btrfs_pin_log_trans(root);
6905         }
6906         /*
6907          * make sure the inode gets flushed if it is replacing
6908          * something.
6909          */
6910         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
6911                 btrfs_add_ordered_operation(trans, root, old_inode);
6912
6913         old_dir->i_ctime = old_dir->i_mtime = ctime;
6914         new_dir->i_ctime = new_dir->i_mtime = ctime;
6915         old_inode->i_ctime = ctime;
6916
6917         if (old_dentry->d_parent != new_dentry->d_parent)
6918                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6919
6920         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6921                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6922                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6923                                         old_dentry->d_name.name,
6924                                         old_dentry->d_name.len);
6925         } else {
6926                 ret = __btrfs_unlink_inode(trans, root, old_dir,
6927                                         old_dentry->d_inode,
6928                                         old_dentry->d_name.name,
6929                                         old_dentry->d_name.len);
6930                 if (!ret)
6931                         ret = btrfs_update_inode(trans, root, old_inode);
6932         }
6933         BUG_ON(ret);
6934
6935         if (new_inode) {
6936                 new_inode->i_ctime = CURRENT_TIME;
6937                 if (unlikely(btrfs_ino(new_inode) ==
6938                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6939                         root_objectid = BTRFS_I(new_inode)->location.objectid;
6940                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
6941                                                 root_objectid,
6942                                                 new_dentry->d_name.name,
6943                                                 new_dentry->d_name.len);
6944                         BUG_ON(new_inode->i_nlink == 0);
6945                 } else {
6946                         ret = btrfs_unlink_inode(trans, dest, new_dir,
6947                                                  new_dentry->d_inode,
6948                                                  new_dentry->d_name.name,
6949                                                  new_dentry->d_name.len);
6950                 }
6951                 BUG_ON(ret);
6952                 if (new_inode->i_nlink == 0) {
6953                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6954                         BUG_ON(ret);
6955                 }
6956         }
6957
6958         fixup_inode_flags(new_dir, old_inode);
6959
6960         ret = btrfs_add_link(trans, new_dir, old_inode,
6961                              new_dentry->d_name.name,
6962                              new_dentry->d_name.len, 0, index);
6963         BUG_ON(ret);
6964
6965         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
6966                 struct dentry *parent = new_dentry->d_parent;
6967                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6968                 btrfs_end_log_trans(root);
6969         }
6970 out_fail:
6971         btrfs_end_transaction_throttle(trans, root);
6972 out_notrans:
6973         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6974                 up_read(&root->fs_info->subvol_sem);
6975
6976         return ret;
6977 }
6978
6979 /*
6980  * some fairly slow code that needs optimization. This walks the list
6981  * of all the inodes with pending delalloc and forces them to disk.
6982  */
6983 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6984 {
6985         struct list_head *head = &root->fs_info->delalloc_inodes;
6986         struct btrfs_inode *binode;
6987         struct inode *inode;
6988
6989         if (root->fs_info->sb->s_flags & MS_RDONLY)
6990                 return -EROFS;
6991
6992         spin_lock(&root->fs_info->delalloc_lock);
6993         while (!list_empty(head)) {
6994                 binode = list_entry(head->next, struct btrfs_inode,
6995                                     delalloc_inodes);
6996                 inode = igrab(&binode->vfs_inode);
6997                 if (!inode)
6998                         list_del_init(&binode->delalloc_inodes);
6999                 spin_unlock(&root->fs_info->delalloc_lock);
7000                 if (inode) {
7001                         filemap_flush(inode->i_mapping);
7002                         if (delay_iput)
7003                                 btrfs_add_delayed_iput(inode);
7004                         else
7005                                 iput(inode);
7006                 }
7007                 cond_resched();
7008                 spin_lock(&root->fs_info->delalloc_lock);
7009         }
7010         spin_unlock(&root->fs_info->delalloc_lock);
7011
7012         /* the filemap_flush will queue IO into the worker threads, but
7013          * we have to make sure the IO is actually started and that
7014          * ordered extents get created before we return
7015          */
7016         atomic_inc(&root->fs_info->async_submit_draining);
7017         while (atomic_read(&root->fs_info->nr_async_submits) ||
7018               atomic_read(&root->fs_info->async_delalloc_pages)) {
7019                 wait_event(root->fs_info->async_submit_wait,
7020                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7021                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7022         }
7023         atomic_dec(&root->fs_info->async_submit_draining);
7024         return 0;
7025 }
7026
7027 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7028                          const char *symname)
7029 {
7030         struct btrfs_trans_handle *trans;
7031         struct btrfs_root *root = BTRFS_I(dir)->root;
7032         struct btrfs_path *path;
7033         struct btrfs_key key;
7034         struct inode *inode = NULL;
7035         int err;
7036         int drop_inode = 0;
7037         u64 objectid;
7038         u64 index = 0 ;
7039         int name_len;
7040         int datasize;
7041         unsigned long ptr;
7042         struct btrfs_file_extent_item *ei;
7043         struct extent_buffer *leaf;
7044         unsigned long nr = 0;
7045
7046         name_len = strlen(symname) + 1;
7047         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7048                 return -ENAMETOOLONG;
7049
7050         /*
7051          * 2 items for inode item and ref
7052          * 2 items for dir items
7053          * 1 item for xattr if selinux is on
7054          */
7055         trans = btrfs_start_transaction(root, 5);
7056         if (IS_ERR(trans))
7057                 return PTR_ERR(trans);
7058
7059         err = btrfs_find_free_ino(root, &objectid);
7060         if (err)
7061                 goto out_unlock;
7062
7063         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7064                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7065                                 S_IFLNK|S_IRWXUGO, &index);
7066         if (IS_ERR(inode)) {
7067                 err = PTR_ERR(inode);
7068                 goto out_unlock;
7069         }
7070
7071         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7072         if (err) {
7073                 drop_inode = 1;
7074                 goto out_unlock;
7075         }
7076
7077         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7078         if (err)
7079                 drop_inode = 1;
7080         else {
7081                 inode->i_mapping->a_ops = &btrfs_aops;
7082                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7083                 inode->i_fop = &btrfs_file_operations;
7084                 inode->i_op = &btrfs_file_inode_operations;
7085                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7086         }
7087         if (drop_inode)
7088                 goto out_unlock;
7089
7090         path = btrfs_alloc_path();
7091         if (!path) {
7092                 err = -ENOMEM;
7093                 drop_inode = 1;
7094                 goto out_unlock;
7095         }
7096         key.objectid = btrfs_ino(inode);
7097         key.offset = 0;
7098         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7099         datasize = btrfs_file_extent_calc_inline_size(name_len);
7100         err = btrfs_insert_empty_item(trans, root, path, &key,
7101                                       datasize);
7102         if (err) {
7103                 drop_inode = 1;
7104                 btrfs_free_path(path);
7105                 goto out_unlock;
7106         }
7107         leaf = path->nodes[0];
7108         ei = btrfs_item_ptr(leaf, path->slots[0],
7109                             struct btrfs_file_extent_item);
7110         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7111         btrfs_set_file_extent_type(leaf, ei,
7112                                    BTRFS_FILE_EXTENT_INLINE);
7113         btrfs_set_file_extent_encryption(leaf, ei, 0);
7114         btrfs_set_file_extent_compression(leaf, ei, 0);
7115         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7116         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7117
7118         ptr = btrfs_file_extent_inline_start(ei);
7119         write_extent_buffer(leaf, symname, ptr, name_len);
7120         btrfs_mark_buffer_dirty(leaf);
7121         btrfs_free_path(path);
7122
7123         inode->i_op = &btrfs_symlink_inode_operations;
7124         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7125         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7126         inode_set_bytes(inode, name_len);
7127         btrfs_i_size_write(inode, name_len - 1);
7128         err = btrfs_update_inode(trans, root, inode);
7129         if (err)
7130                 drop_inode = 1;
7131
7132 out_unlock:
7133         nr = trans->blocks_used;
7134         btrfs_end_transaction_throttle(trans, root);
7135         if (drop_inode) {
7136                 inode_dec_link_count(inode);
7137                 iput(inode);
7138         }
7139         btrfs_btree_balance_dirty(root, nr);
7140         return err;
7141 }
7142
7143 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7144                                        u64 start, u64 num_bytes, u64 min_size,
7145                                        loff_t actual_len, u64 *alloc_hint,
7146                                        struct btrfs_trans_handle *trans)
7147 {
7148         struct btrfs_root *root = BTRFS_I(inode)->root;
7149         struct btrfs_key ins;
7150         u64 cur_offset = start;
7151         u64 i_size;
7152         int ret = 0;
7153         bool own_trans = true;
7154
7155         if (trans)
7156                 own_trans = false;
7157         while (num_bytes > 0) {
7158                 if (own_trans) {
7159                         trans = btrfs_start_transaction(root, 3);
7160                         if (IS_ERR(trans)) {
7161                                 ret = PTR_ERR(trans);
7162                                 break;
7163                         }
7164                 }
7165
7166                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7167                                            0, *alloc_hint, (u64)-1, &ins, 1);
7168                 if (ret) {
7169                         if (own_trans)
7170                                 btrfs_end_transaction(trans, root);
7171                         break;
7172                 }
7173
7174                 ret = insert_reserved_file_extent(trans, inode,
7175                                                   cur_offset, ins.objectid,
7176                                                   ins.offset, ins.offset,
7177                                                   ins.offset, 0, 0, 0,
7178                                                   BTRFS_FILE_EXTENT_PREALLOC);
7179                 BUG_ON(ret);
7180                 btrfs_drop_extent_cache(inode, cur_offset,
7181                                         cur_offset + ins.offset -1, 0);
7182
7183                 num_bytes -= ins.offset;
7184                 cur_offset += ins.offset;
7185                 *alloc_hint = ins.objectid + ins.offset;
7186
7187                 inode->i_ctime = CURRENT_TIME;
7188                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7189                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7190                     (actual_len > inode->i_size) &&
7191                     (cur_offset > inode->i_size)) {
7192                         if (cur_offset > actual_len)
7193                                 i_size = actual_len;
7194                         else
7195                                 i_size = cur_offset;
7196                         i_size_write(inode, i_size);
7197                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7198                 }
7199
7200                 ret = btrfs_update_inode(trans, root, inode);
7201                 BUG_ON(ret);
7202
7203                 if (own_trans)
7204                         btrfs_end_transaction(trans, root);
7205         }
7206         return ret;
7207 }
7208
7209 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7210                               u64 start, u64 num_bytes, u64 min_size,
7211                               loff_t actual_len, u64 *alloc_hint)
7212 {
7213         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7214                                            min_size, actual_len, alloc_hint,
7215                                            NULL);
7216 }
7217
7218 int btrfs_prealloc_file_range_trans(struct inode *inode,
7219                                     struct btrfs_trans_handle *trans, int mode,
7220                                     u64 start, u64 num_bytes, u64 min_size,
7221                                     loff_t actual_len, u64 *alloc_hint)
7222 {
7223         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7224                                            min_size, actual_len, alloc_hint, trans);
7225 }
7226
7227 static int btrfs_set_page_dirty(struct page *page)
7228 {
7229         return __set_page_dirty_nobuffers(page);
7230 }
7231
7232 static int btrfs_permission(struct inode *inode, int mask)
7233 {
7234         struct btrfs_root *root = BTRFS_I(inode)->root;
7235         umode_t mode = inode->i_mode;
7236
7237         if (mask & MAY_WRITE &&
7238             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7239                 if (btrfs_root_readonly(root))
7240                         return -EROFS;
7241                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7242                         return -EACCES;
7243         }
7244         return generic_permission(inode, mask);
7245 }
7246
7247 static const struct inode_operations btrfs_dir_inode_operations = {
7248         .getattr        = btrfs_getattr,
7249         .lookup         = btrfs_lookup,
7250         .create         = btrfs_create,
7251         .unlink         = btrfs_unlink,
7252         .link           = btrfs_link,
7253         .mkdir          = btrfs_mkdir,
7254         .rmdir          = btrfs_rmdir,
7255         .rename         = btrfs_rename,
7256         .symlink        = btrfs_symlink,
7257         .setattr        = btrfs_setattr,
7258         .mknod          = btrfs_mknod,
7259         .setxattr       = btrfs_setxattr,
7260         .getxattr       = btrfs_getxattr,
7261         .listxattr      = btrfs_listxattr,
7262         .removexattr    = btrfs_removexattr,
7263         .permission     = btrfs_permission,
7264         .get_acl        = btrfs_get_acl,
7265 };
7266 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7267         .lookup         = btrfs_lookup,
7268         .permission     = btrfs_permission,
7269         .get_acl        = btrfs_get_acl,
7270 };
7271
7272 static const struct file_operations btrfs_dir_file_operations = {
7273         .llseek         = generic_file_llseek,
7274         .read           = generic_read_dir,
7275         .readdir        = btrfs_real_readdir,
7276         .unlocked_ioctl = btrfs_ioctl,
7277 #ifdef CONFIG_COMPAT
7278         .compat_ioctl   = btrfs_ioctl,
7279 #endif
7280         .release        = btrfs_release_file,
7281         .fsync          = btrfs_sync_file,
7282 };
7283
7284 static struct extent_io_ops btrfs_extent_io_ops = {
7285         .fill_delalloc = run_delalloc_range,
7286         .submit_bio_hook = btrfs_submit_bio_hook,
7287         .merge_bio_hook = btrfs_merge_bio_hook,
7288         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7289         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7290         .writepage_start_hook = btrfs_writepage_start_hook,
7291         .set_bit_hook = btrfs_set_bit_hook,
7292         .clear_bit_hook = btrfs_clear_bit_hook,
7293         .merge_extent_hook = btrfs_merge_extent_hook,
7294         .split_extent_hook = btrfs_split_extent_hook,
7295 };
7296
7297 /*
7298  * btrfs doesn't support the bmap operation because swapfiles
7299  * use bmap to make a mapping of extents in the file.  They assume
7300  * these extents won't change over the life of the file and they
7301  * use the bmap result to do IO directly to the drive.
7302  *
7303  * the btrfs bmap call would return logical addresses that aren't
7304  * suitable for IO and they also will change frequently as COW
7305  * operations happen.  So, swapfile + btrfs == corruption.
7306  *
7307  * For now we're avoiding this by dropping bmap.
7308  */
7309 static const struct address_space_operations btrfs_aops = {
7310         .readpage       = btrfs_readpage,
7311         .writepage      = btrfs_writepage,
7312         .writepages     = btrfs_writepages,
7313         .readpages      = btrfs_readpages,
7314         .direct_IO      = btrfs_direct_IO,
7315         .invalidatepage = btrfs_invalidatepage,
7316         .releasepage    = btrfs_releasepage,
7317         .set_page_dirty = btrfs_set_page_dirty,
7318         .error_remove_page = generic_error_remove_page,
7319 };
7320
7321 static const struct address_space_operations btrfs_symlink_aops = {
7322         .readpage       = btrfs_readpage,
7323         .writepage      = btrfs_writepage,
7324         .invalidatepage = btrfs_invalidatepage,
7325         .releasepage    = btrfs_releasepage,
7326 };
7327
7328 static const struct inode_operations btrfs_file_inode_operations = {
7329         .getattr        = btrfs_getattr,
7330         .setattr        = btrfs_setattr,
7331         .setxattr       = btrfs_setxattr,
7332         .getxattr       = btrfs_getxattr,
7333         .listxattr      = btrfs_listxattr,
7334         .removexattr    = btrfs_removexattr,
7335         .permission     = btrfs_permission,
7336         .fiemap         = btrfs_fiemap,
7337         .get_acl        = btrfs_get_acl,
7338 };
7339 static const struct inode_operations btrfs_special_inode_operations = {
7340         .getattr        = btrfs_getattr,
7341         .setattr        = btrfs_setattr,
7342         .permission     = btrfs_permission,
7343         .setxattr       = btrfs_setxattr,
7344         .getxattr       = btrfs_getxattr,
7345         .listxattr      = btrfs_listxattr,
7346         .removexattr    = btrfs_removexattr,
7347         .get_acl        = btrfs_get_acl,
7348 };
7349 static const struct inode_operations btrfs_symlink_inode_operations = {
7350         .readlink       = generic_readlink,
7351         .follow_link    = page_follow_link_light,
7352         .put_link       = page_put_link,
7353         .getattr        = btrfs_getattr,
7354         .permission     = btrfs_permission,
7355         .setxattr       = btrfs_setxattr,
7356         .getxattr       = btrfs_getxattr,
7357         .listxattr      = btrfs_listxattr,
7358         .removexattr    = btrfs_removexattr,
7359         .get_acl        = btrfs_get_acl,
7360 };
7361
7362 const struct dentry_operations btrfs_dentry_operations = {
7363         .d_delete       = btrfs_dentry_delete,
7364         .d_release      = btrfs_dentry_release,
7365 };