SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/disk-io.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/version.h>
  20 #include <linux/fs.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/scatterlist.h>
  23 #include <linux/swap.h>
  24 #include <linux/radix-tree.h>
  25 #include <linux/writeback.h>
  26 #include <linux/buffer_head.h> // for block_sync_page
  27 #include <linux/workqueue.h>
  28 #include <linux/kthread.h>
  29 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
  30 # include <linux/freezer.h>
  31 #else
  32 # include <linux/sched.h>
  33 #endif
  34 #include "crc32c.h"
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "async-thread.h"
  42 #include "locking.h"
  43 #include "ref-cache.h"
  44
  45 #if 0
  46 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  47 {
  48         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  49                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  50                        (unsigned long long)extent_buffer_blocknr(buf),
  51                        (unsigned long long)btrfs_header_blocknr(buf));
  52                 return 1;
  53         }
  54         return 0;
  55 }
  56 #endif
  57
  58 static struct extent_io_ops btree_extent_io_ops;
  59 static void end_workqueue_fn(struct btrfs_work *work);
  60
  61 struct end_io_wq {
  62         struct bio *bio;
  63         bio_end_io_t *end_io;
  64         void *private;
  65         struct btrfs_fs_info *info;
  66         int error;
  67         int metadata;
  68         struct list_head list;
  69         struct btrfs_work work;
  70 };
  71
  72 struct async_submit_bio {
  73         struct inode *inode;
  74         struct bio *bio;
  75         struct list_head list;
  76         extent_submit_bio_hook_t *submit_bio_hook;
  77         int rw;
  78         int mirror_num;
  79         struct btrfs_work work;
  80 };
  81
  82 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  83                                     size_t page_offset, u64 start, u64 len,
  84                                     int create)
  85 {
  86         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  87         struct extent_map *em;
  88         int ret;
  89
  90         spin_lock(&em_tree->lock);
  91         em = lookup_extent_mapping(em_tree, start, len);
  92         if (em) {
  93                 em->bdev =
  94                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
  95                 spin_unlock(&em_tree->lock);
  96                 goto out;
  97         }
  98         spin_unlock(&em_tree->lock);
  99
 100         em = alloc_extent_map(GFP_NOFS);
 101         if (!em) {
 102                 em = ERR_PTR(-ENOMEM);
 103                 goto out;
 104         }
 105         em->start = 0;
 106         em->len = (u64)-1;
 107         em->block_start = 0;
 108         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 109
 110         spin_lock(&em_tree->lock);
 111         ret = add_extent_mapping(em_tree, em);
 112         if (ret == -EEXIST) {
 113                 u64 failed_start = em->start;
 114                 u64 failed_len = em->len;
 115
 116                 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
 117                        em->start, em->len, em->block_start);
 118                 free_extent_map(em);
 119                 em = lookup_extent_mapping(em_tree, start, len);
 120                 if (em) {
 121                         printk("after failing, found %Lu %Lu %Lu\n",
 122                                em->start, em->len, em->block_start);
 123                         ret = 0;
 124                 } else {
 125                         em = lookup_extent_mapping(em_tree, failed_start,
 126                                                    failed_len);
 127                         if (em) {
 128                                 printk("double failure lookup gives us "
 129                                        "%Lu %Lu -> %Lu\n", em->start,
 130                                        em->len, em->block_start);
 131                                 free_extent_map(em);
 132                         }
 133                         ret = -EIO;
 134                 }
 135         } else if (ret) {
 136                 free_extent_map(em);
 137                 em = NULL;
 138         }
 139         spin_unlock(&em_tree->lock);
 140
 141         if (ret)
 142                 em = ERR_PTR(ret);
 143 out:
 144         return em;
 145 }
 146
 147 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 148 {
 149         return btrfs_crc32c(seed, data, len);
 150 }
 151
 152 void btrfs_csum_final(u32 crc, char *result)
 153 {
 154         *(__le32 *)result = ~cpu_to_le32(crc);
 155 }
 156
 157 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 158                            int verify)
 159 {
 160         char result[BTRFS_CRC32_SIZE];
 161         unsigned long len;
 162         unsigned long cur_len;
 163         unsigned long offset = BTRFS_CSUM_SIZE;
 164         char *map_token = NULL;
 165         char *kaddr;
 166         unsigned long map_start;
 167         unsigned long map_len;
 168         int err;
 169         u32 crc = ~(u32)0;
 170
 171         len = buf->len - offset;
 172         while(len > 0) {
 173                 err = map_private_extent_buffer(buf, offset, 32,
 174                                         &map_token, &kaddr,
 175                                         &map_start, &map_len, KM_USER0);
 176                 if (err) {
 177                         printk("failed to map extent buffer! %lu\n",
 178                                offset);
 179                         return 1;
 180                 }
 181                 cur_len = min(len, map_len - (offset - map_start));
 182                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 183                                       crc, cur_len);
 184                 len -= cur_len;
 185                 offset += cur_len;
 186                 unmap_extent_buffer(buf, map_token, KM_USER0);
 187         }
 188         btrfs_csum_final(crc, result);
 189
 190         if (verify) {
 191                 /* FIXME, this is not good */
 192                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 193                         u32 val;
 194                         u32 found = 0;
 195                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 196
 197                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 198                         printk("btrfs: %s checksum verify failed on %llu "
 199                                "wanted %X found %X level %d\n",
 200                                root->fs_info->sb->s_id,
 201                                buf->start, val, found, btrfs_header_level(buf));
 202                         return 1;
 203                 }
 204         } else {
 205                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 206         }
 207         return 0;
 208 }
 209
 210 static int verify_parent_transid(struct extent_io_tree *io_tree,
 211                                  struct extent_buffer *eb, u64 parent_transid)
 212 {
 213         int ret;
 214
 215         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 216                 return 0;
 217
 218         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 219         if (extent_buffer_uptodate(io_tree, eb) &&
 220             btrfs_header_generation(eb) == parent_transid) {
 221                 ret = 0;
 222                 goto out;
 223         }
 224         printk("parent transid verify failed on %llu wanted %llu found %llu\n",
 225                (unsigned long long)eb->start,
 226                (unsigned long long)parent_transid,
 227                (unsigned long long)btrfs_header_generation(eb));
 228         ret = 1;
 229         clear_extent_buffer_uptodate(io_tree, eb);
 230 out:
 231         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 232                       GFP_NOFS);
 233         return ret;
 234
 235 }
 236
 237 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 238                                           struct extent_buffer *eb,
 239                                           u64 start, u64 parent_transid)
 240 {
 241         struct extent_io_tree *io_tree;
 242         int ret;
 243         int num_copies = 0;
 244         int mirror_num = 0;
 245
 246         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 247         while (1) {
 248                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 249                                                btree_get_extent, mirror_num);
 250                 if (!ret &&
 251                     !verify_parent_transid(io_tree, eb, parent_transid))
 252                         return ret;
 253
 254                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 255                                               eb->start, eb->len);
 256                 if (num_copies == 1)
 257                         return ret;
 258
 259                 mirror_num++;
 260                 if (mirror_num > num_copies)
 261                         return ret;
 262         }
 263         return -EIO;
 264 }
 265
 266 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 267 {
 268         struct extent_io_tree *tree;
 269         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 270         u64 found_start;
 271         int found_level;
 272         unsigned long len;
 273         struct extent_buffer *eb;
 274         int ret;
 275
 276         tree = &BTRFS_I(page->mapping->host)->io_tree;
 277
 278         if (page->private == EXTENT_PAGE_PRIVATE)
 279                 goto out;
 280         if (!page->private)
 281                 goto out;
 282         len = page->private >> 2;
 283         if (len == 0) {
 284                 WARN_ON(1);
 285         }
 286         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 287         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 288                                              btrfs_header_generation(eb));
 289         BUG_ON(ret);
 290         found_start = btrfs_header_bytenr(eb);
 291         if (found_start != start) {
 292                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 293                        start, found_start, len);
 294                 WARN_ON(1);
 295                 goto err;
 296         }
 297         if (eb->first_page != page) {
 298                 printk("bad first page %lu %lu\n", eb->first_page->index,
 299                        page->index);
 300                 WARN_ON(1);
 301                 goto err;
 302         }
 303         if (!PageUptodate(page)) {
 304                 printk("csum not up to date page %lu\n", page->index);
 305                 WARN_ON(1);
 306                 goto err;
 307         }
 308         found_level = btrfs_header_level(eb);
 309         spin_lock(&root->fs_info->hash_lock);
 310         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
 311         spin_unlock(&root->fs_info->hash_lock);
 312         csum_tree_block(root, eb, 0);
 313 err:
 314         free_extent_buffer(eb);
 315 out:
 316         return 0;
 317 }
 318
 319 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 320 {
 321         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 322
 323         csum_dirty_buffer(root, page);
 324         return 0;
 325 }
 326
 327 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 328                                struct extent_state *state)
 329 {
 330         struct extent_io_tree *tree;
 331         u64 found_start;
 332         int found_level;
 333         unsigned long len;
 334         struct extent_buffer *eb;
 335         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 336         int ret = 0;
 337
 338         tree = &BTRFS_I(page->mapping->host)->io_tree;
 339         if (page->private == EXTENT_PAGE_PRIVATE)
 340                 goto out;
 341         if (!page->private)
 342                 goto out;
 343         len = page->private >> 2;
 344         if (len == 0) {
 345                 WARN_ON(1);
 346         }
 347         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 348
 349         found_start = btrfs_header_bytenr(eb);
 350         if (found_start != start) {
 351                 ret = -EIO;
 352                 goto err;
 353         }
 354         if (eb->first_page != page) {
 355                 printk("bad first page %lu %lu\n", eb->first_page->index,
 356                        page->index);
 357                 WARN_ON(1);
 358                 ret = -EIO;
 359                 goto err;
 360         }
 361         if (memcmp_extent_buffer(eb, root->fs_info->fsid,
 362                                  (unsigned long)btrfs_header_fsid(eb),
 363                                  BTRFS_FSID_SIZE)) {
 364                 printk("bad fsid on block %Lu\n", eb->start);
 365                 ret = -EIO;
 366                 goto err;
 367         }
 368         found_level = btrfs_header_level(eb);
 369
 370         ret = csum_tree_block(root, eb, 1);
 371         if (ret)
 372                 ret = -EIO;
 373
 374         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 375         end = eb->start + end - 1;
 376 err:
 377         free_extent_buffer(eb);
 378 out:
 379         return ret;
 380 }
 381
 382 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 383 static void end_workqueue_bio(struct bio *bio, int err)
 384 #else
 385 static int end_workqueue_bio(struct bio *bio,
 386                                    unsigned int bytes_done, int err)
 387 #endif
 388 {
 389         struct end_io_wq *end_io_wq = bio->bi_private;
 390         struct btrfs_fs_info *fs_info;
 391
 392 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 393         if (bio->bi_size)
 394                 return 1;
 395 #endif
 396
 397         fs_info = end_io_wq->info;
 398         end_io_wq->error = err;
 399         end_io_wq->work.func = end_workqueue_fn;
 400         end_io_wq->work.flags = 0;
 401         if (bio->bi_rw & (1 << BIO_RW))
 402                 btrfs_queue_worker(&fs_info->endio_write_workers,
 403                                    &end_io_wq->work);
 404         else
 405                 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
 406
 407 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
 408         return 0;
 409 #endif
 410 }
 411
 412 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 413                         int metadata)
 414 {
 415         struct end_io_wq *end_io_wq;
 416         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 417         if (!end_io_wq)
 418                 return -ENOMEM;
 419
 420         end_io_wq->private = bio->bi_private;
 421         end_io_wq->end_io = bio->bi_end_io;
 422         end_io_wq->info = info;
 423         end_io_wq->error = 0;
 424         end_io_wq->bio = bio;
 425         end_io_wq->metadata = metadata;
 426
 427         bio->bi_private = end_io_wq;
 428         bio->bi_end_io = end_workqueue_bio;
 429         return 0;
 430 }
 431
 432 static void run_one_async_submit(struct btrfs_work *work)
 433 {
 434         struct btrfs_fs_info *fs_info;
 435         struct async_submit_bio *async;
 436
 437         async = container_of(work, struct  async_submit_bio, work);
 438         fs_info = BTRFS_I(async->inode)->root->fs_info;
 439         atomic_dec(&fs_info->nr_async_submits);
 440         async->submit_bio_hook(async->inode, async->rw, async->bio,
 441                                async->mirror_num);
 442         kfree(async);
 443 }
 444
 445 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 446                         int rw, struct bio *bio, int mirror_num,
 447                         extent_submit_bio_hook_t *submit_bio_hook)
 448 {
 449         struct async_submit_bio *async;
 450
 451         async = kmalloc(sizeof(*async), GFP_NOFS);
 452         if (!async)
 453                 return -ENOMEM;
 454
 455         async->inode = inode;
 456         async->rw = rw;
 457         async->bio = bio;
 458         async->mirror_num = mirror_num;
 459         async->submit_bio_hook = submit_bio_hook;
 460         async->work.func = run_one_async_submit;
 461         async->work.flags = 0;
 462         atomic_inc(&fs_info->nr_async_submits);
 463         btrfs_queue_worker(&fs_info->workers, &async->work);
 464         return 0;
 465 }
 466
 467 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 468                                  int mirror_num)
 469 {
 470         struct btrfs_root *root = BTRFS_I(inode)->root;
 471         u64 offset;
 472         int ret;
 473
 474         offset = bio->bi_sector << 9;
 475
 476         /*
 477          * when we're called for a write, we're already in the async
 478          * submission context.  Just jump ingo btrfs_map_bio
 479          */
 480         if (rw & (1 << BIO_RW)) {
 481                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 482                                      mirror_num, 0);
 483         }
 484
 485         /*
 486          * called for a read, do the setup so that checksum validation
 487          * can happen in the async kernel threads
 488          */
 489         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
 490         BUG_ON(ret);
 491
 492         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 493 }
 494
 495 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 496                                  int mirror_num)
 497 {
 498         /*
 499          * kthread helpers are used to submit writes so that checksumming
 500          * can happen in parallel across all CPUs
 501          */
 502         if (!(rw & (1 << BIO_RW))) {
 503                 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
 504         }
 505         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 506                                    inode, rw, bio, mirror_num,
 507                                    __btree_submit_bio_hook);
 508 }
 509
 510 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 511 {
 512         struct extent_io_tree *tree;
 513         tree = &BTRFS_I(page->mapping->host)->io_tree;
 514         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 515 }
 516
 517 static int btree_writepages(struct address_space *mapping,
 518                             struct writeback_control *wbc)
 519 {
 520         struct extent_io_tree *tree;
 521         tree = &BTRFS_I(mapping->host)->io_tree;
 522         if (wbc->sync_mode == WB_SYNC_NONE) {
 523                 u64 num_dirty;
 524                 u64 start = 0;
 525                 unsigned long thresh = 96 * 1024 * 1024;
 526
 527                 if (wbc->for_kupdate)
 528                         return 0;
 529
 530                 if (current_is_pdflush()) {
 531                         thresh = 96 * 1024 * 1024;
 532                 } else {
 533                         thresh = 8 * 1024 * 1024;
 534                 }
 535                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 536                                              thresh, EXTENT_DIRTY);
 537                 if (num_dirty < thresh) {
 538                         return 0;
 539                 }
 540         }
 541         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 542 }
 543
 544 int btree_readpage(struct file *file, struct page *page)
 545 {
 546         struct extent_io_tree *tree;
 547         tree = &BTRFS_I(page->mapping->host)->io_tree;
 548         return extent_read_full_page(tree, page, btree_get_extent);
 549 }
 550
 551 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 552 {
 553         struct extent_io_tree *tree;
 554         struct extent_map_tree *map;
 555         int ret;
 556
 557         tree = &BTRFS_I(page->mapping->host)->io_tree;
 558         map = &BTRFS_I(page->mapping->host)->extent_tree;
 559
 560         ret = try_release_extent_state(map, tree, page, gfp_flags);
 561         if (!ret) {
 562                 return 0;
 563         }
 564
 565         ret = try_release_extent_buffer(tree, page);
 566         if (ret == 1) {
 567                 ClearPagePrivate(page);
 568                 set_page_private(page, 0);
 569                 page_cache_release(page);
 570         }
 571
 572         return ret;
 573 }
 574
 575 static void btree_invalidatepage(struct page *page, unsigned long offset)
 576 {
 577         struct extent_io_tree *tree;
 578         tree = &BTRFS_I(page->mapping->host)->io_tree;
 579         extent_invalidatepage(tree, page, offset);
 580         btree_releasepage(page, GFP_NOFS);
 581         if (PagePrivate(page)) {
 582                 printk("warning page private not zero on page %Lu\n",
 583                        page_offset(page));
 584                 ClearPagePrivate(page);
 585                 set_page_private(page, 0);
 586                 page_cache_release(page);
 587         }
 588 }
 589
 590 #if 0
 591 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 592 {
 593         struct buffer_head *bh;
 594         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 595         struct buffer_head *head;
 596         if (!page_has_buffers(page)) {
 597                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 598                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 599         }
 600         head = page_buffers(page);
 601         bh = head;
 602         do {
 603                 if (buffer_dirty(bh))
 604                         csum_tree_block(root, bh, 0);
 605                 bh = bh->b_this_page;
 606         } while (bh != head);
 607         return block_write_full_page(page, btree_get_block, wbc);
 608 }
 609 #endif
 610
 611 static struct address_space_operations btree_aops = {
 612         .readpage       = btree_readpage,
 613         .writepage      = btree_writepage,
 614         .writepages     = btree_writepages,
 615         .releasepage    = btree_releasepage,
 616         .invalidatepage = btree_invalidatepage,
 617         .sync_page      = block_sync_page,
 618 };
 619
 620 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 621                          u64 parent_transid)
 622 {
 623         struct extent_buffer *buf = NULL;
 624         struct inode *btree_inode = root->fs_info->btree_inode;
 625         int ret = 0;
 626
 627         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 628         if (!buf)
 629                 return 0;
 630         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 631                                  buf, 0, 0, btree_get_extent, 0);
 632         free_extent_buffer(buf);
 633         return ret;
 634 }
 635
 636 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 637                                             u64 bytenr, u32 blocksize)
 638 {
 639         struct inode *btree_inode = root->fs_info->btree_inode;
 640         struct extent_buffer *eb;
 641         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 642                                 bytenr, blocksize, GFP_NOFS);
 643         return eb;
 644 }
 645
 646 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 647                                                  u64 bytenr, u32 blocksize)
 648 {
 649         struct inode *btree_inode = root->fs_info->btree_inode;
 650         struct extent_buffer *eb;
 651
 652         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 653                                  bytenr, blocksize, NULL, GFP_NOFS);
 654         return eb;
 655 }
 656
 657
 658 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 659                                       u32 blocksize, u64 parent_transid)
 660 {
 661         struct extent_buffer *buf = NULL;
 662         struct inode *btree_inode = root->fs_info->btree_inode;
 663         struct extent_io_tree *io_tree;
 664         int ret;
 665
 666         io_tree = &BTRFS_I(btree_inode)->io_tree;
 667
 668         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 669         if (!buf)
 670                 return NULL;
 671
 672         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 673
 674         if (ret == 0) {
 675                 buf->flags |= EXTENT_UPTODATE;
 676         }
 677         return buf;
 678
 679 }
 680
 681 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 682                      struct extent_buffer *buf)
 683 {
 684         struct inode *btree_inode = root->fs_info->btree_inode;
 685         if (btrfs_header_generation(buf) ==
 686             root->fs_info->running_transaction->transid) {
 687                 WARN_ON(!btrfs_tree_locked(buf));
 688                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 689                                           buf);
 690         }
 691         return 0;
 692 }
 693
 694 int wait_on_tree_block_writeback(struct btrfs_root *root,
 695                                  struct extent_buffer *buf)
 696 {
 697         struct inode *btree_inode = root->fs_info->btree_inode;
 698         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
 699                                         buf);
 700         return 0;
 701 }
 702
 703 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 704                         u32 stripesize, struct btrfs_root *root,
 705                         struct btrfs_fs_info *fs_info,
 706                         u64 objectid)
 707 {
 708         root->node = NULL;
 709         root->inode = NULL;
 710         root->commit_root = NULL;
 711         root->ref_tree = NULL;
 712         root->sectorsize = sectorsize;
 713         root->nodesize = nodesize;
 714         root->leafsize = leafsize;
 715         root->stripesize = stripesize;
 716         root->ref_cows = 0;
 717         root->track_dirty = 0;
 718
 719         root->fs_info = fs_info;
 720         root->objectid = objectid;
 721         root->last_trans = 0;
 722         root->highest_inode = 0;
 723         root->last_inode_alloc = 0;
 724         root->name = NULL;
 725         root->in_sysfs = 0;
 726
 727         INIT_LIST_HEAD(&root->dirty_list);
 728         INIT_LIST_HEAD(&root->orphan_list);
 729         INIT_LIST_HEAD(&root->dead_list);
 730         spin_lock_init(&root->node_lock);
 731         spin_lock_init(&root->list_lock);
 732         mutex_init(&root->objectid_mutex);
 733
 734         btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
 735         root->ref_tree = &root->ref_tree_struct;
 736
 737         memset(&root->root_key, 0, sizeof(root->root_key));
 738         memset(&root->root_item, 0, sizeof(root->root_item));
 739         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 740         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 741         root->defrag_trans_start = fs_info->generation;
 742         init_completion(&root->kobj_unregister);
 743         root->defrag_running = 0;
 744         root->defrag_level = 0;
 745         root->root_key.objectid = objectid;
 746         return 0;
 747 }
 748
 749 static int find_and_setup_root(struct btrfs_root *tree_root,
 750                                struct btrfs_fs_info *fs_info,
 751                                u64 objectid,
 752                                struct btrfs_root *root)
 753 {
 754         int ret;
 755         u32 blocksize;
 756
 757         __setup_root(tree_root->nodesize, tree_root->leafsize,
 758                      tree_root->sectorsize, tree_root->stripesize,
 759                      root, fs_info, objectid);
 760         ret = btrfs_find_last_root(tree_root, objectid,
 761                                    &root->root_item, &root->root_key);
 762         BUG_ON(ret);
 763
 764         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 765         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 766                                      blocksize, 0);
 767         BUG_ON(!root->node);
 768         return 0;
 769 }
 770
 771 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 772                                                struct btrfs_key *location)
 773 {
 774         struct btrfs_root *root;
 775         struct btrfs_root *tree_root = fs_info->tree_root;
 776         struct btrfs_path *path;
 777         struct extent_buffer *l;
 778         u64 highest_inode;
 779         u32 blocksize;
 780         int ret = 0;
 781
 782         root = kzalloc(sizeof(*root), GFP_NOFS);
 783         if (!root)
 784                 return ERR_PTR(-ENOMEM);
 785         if (location->offset == (u64)-1) {
 786                 ret = find_and_setup_root(tree_root, fs_info,
 787                                           location->objectid, root);
 788                 if (ret) {
 789                         kfree(root);
 790                         return ERR_PTR(ret);
 791                 }
 792                 goto insert;
 793         }
 794
 795         __setup_root(tree_root->nodesize, tree_root->leafsize,
 796                      tree_root->sectorsize, tree_root->stripesize,
 797                      root, fs_info, location->objectid);
 798
 799         path = btrfs_alloc_path();
 800         BUG_ON(!path);
 801         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 802         if (ret != 0) {
 803                 if (ret > 0)
 804                         ret = -ENOENT;
 805                 goto out;
 806         }
 807         l = path->nodes[0];
 808         read_extent_buffer(l, &root->root_item,
 809                btrfs_item_ptr_offset(l, path->slots[0]),
 810                sizeof(root->root_item));
 811         memcpy(&root->root_key, location, sizeof(*location));
 812         ret = 0;
 813 out:
 814         btrfs_release_path(root, path);
 815         btrfs_free_path(path);
 816         if (ret) {
 817                 kfree(root);
 818                 return ERR_PTR(ret);
 819         }
 820         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 821         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 822                                      blocksize, 0);
 823         BUG_ON(!root->node);
 824 insert:
 825         root->ref_cows = 1;
 826         ret = btrfs_find_highest_inode(root, &highest_inode);
 827         if (ret == 0) {
 828                 root->highest_inode = highest_inode;
 829                 root->last_inode_alloc = highest_inode;
 830         }
 831         return root;
 832 }
 833
 834 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 835                                         u64 root_objectid)
 836 {
 837         struct btrfs_root *root;
 838
 839         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 840                 return fs_info->tree_root;
 841         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 842                 return fs_info->extent_root;
 843
 844         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 845                                  (unsigned long)root_objectid);
 846         return root;
 847 }
 848
 849 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 850                                               struct btrfs_key *location)
 851 {
 852         struct btrfs_root *root;
 853         int ret;
 854
 855         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 856                 return fs_info->tree_root;
 857         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 858                 return fs_info->extent_root;
 859         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
 860                 return fs_info->chunk_root;
 861         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
 862                 return fs_info->dev_root;
 863
 864         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 865                                  (unsigned long)location->objectid);
 866         if (root)
 867                 return root;
 868
 869         root = btrfs_read_fs_root_no_radix(fs_info, location);
 870         if (IS_ERR(root))
 871                 return root;
 872         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 873                                 (unsigned long)root->root_key.objectid,
 874                                 root);
 875         if (ret) {
 876                 free_extent_buffer(root->node);
 877                 kfree(root);
 878                 return ERR_PTR(ret);
 879         }
 880         ret = btrfs_find_dead_roots(fs_info->tree_root,
 881                                     root->root_key.objectid, root);
 882         BUG_ON(ret);
 883
 884         return root;
 885 }
 886
 887 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 888                                       struct btrfs_key *location,
 889                                       const char *name, int namelen)
 890 {
 891         struct btrfs_root *root;
 892         int ret;
 893
 894         root = btrfs_read_fs_root_no_name(fs_info, location);
 895         if (!root)
 896                 return NULL;
 897
 898         if (root->in_sysfs)
 899                 return root;
 900
 901         ret = btrfs_set_root_name(root, name, namelen);
 902         if (ret) {
 903                 free_extent_buffer(root->node);
 904                 kfree(root);
 905                 return ERR_PTR(ret);
 906         }
 907
 908         ret = btrfs_sysfs_add_root(root);
 909         if (ret) {
 910                 free_extent_buffer(root->node);
 911                 kfree(root->name);
 912                 kfree(root);
 913                 return ERR_PTR(ret);
 914         }
 915         root->in_sysfs = 1;
 916         return root;
 917 }
 918 #if 0
 919 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 920         struct btrfs_hasher *hasher;
 921
 922         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 923         if (!hasher)
 924                 return -ENOMEM;
 925         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 926         if (!hasher->hash_tfm) {
 927                 kfree(hasher);
 928                 return -EINVAL;
 929         }
 930         spin_lock(&info->hash_lock);
 931         list_add(&hasher->list, &info->hashers);
 932         spin_unlock(&info->hash_lock);
 933         return 0;
 934 }
 935 #endif
 936
 937 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 938 {
 939         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
 940         int ret = 0;
 941         int limit = 256 * info->fs_devices->open_devices;
 942         struct list_head *cur;
 943         struct btrfs_device *device;
 944         struct backing_dev_info *bdi;
 945
 946         if ((bdi_bits & (1 << BDI_write_congested)) &&
 947             atomic_read(&info->nr_async_submits) > limit) {
 948                 return 1;
 949         }
 950
 951         list_for_each(cur, &info->fs_devices->devices) {
 952                 device = list_entry(cur, struct btrfs_device, dev_list);
 953                 if (!device->bdev)
 954                         continue;
 955                 bdi = blk_get_backing_dev_info(device->bdev);
 956                 if (bdi && bdi_congested(bdi, bdi_bits)) {
 957                         ret = 1;
 958                         break;
 959                 }
 960         }
 961         return ret;
 962 }
 963
 964 /*
 965  * this unplugs every device on the box, and it is only used when page
 966  * is null
 967  */
 968 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 969 {
 970         struct list_head *cur;
 971         struct btrfs_device *device;
 972         struct btrfs_fs_info *info;
 973
 974         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
 975         list_for_each(cur, &info->fs_devices->devices) {
 976                 device = list_entry(cur, struct btrfs_device, dev_list);
 977                 bdi = blk_get_backing_dev_info(device->bdev);
 978                 if (bdi->unplug_io_fn) {
 979                         bdi->unplug_io_fn(bdi, page);
 980                 }
 981         }
 982 }
 983
 984 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 985 {
 986         struct inode *inode;
 987         struct extent_map_tree *em_tree;
 988         struct extent_map *em;
 989         struct address_space *mapping;
 990         u64 offset;
 991
 992         /* the generic O_DIRECT read code does this */
 993         if (!page) {
 994                 __unplug_io_fn(bdi, page);
 995                 return;
 996         }
 997
 998         /*
 999          * page->mapping may change at any time.  Get a consistent copy
1000          * and use that for everything below
1001          */
1002         smp_mb();
1003         mapping = page->mapping;
1004         if (!mapping)
1005                 return;
1006
1007         inode = mapping->host;
1008         offset = page_offset(page);
1009
1010         em_tree = &BTRFS_I(inode)->extent_tree;
1011         spin_lock(&em_tree->lock);
1012         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1013         spin_unlock(&em_tree->lock);
1014         if (!em) {
1015                 __unplug_io_fn(bdi, page);
1016                 return;
1017         }
1018
1019         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1020                 free_extent_map(em);
1021                 __unplug_io_fn(bdi, page);
1022                 return;
1023         }
1024         offset = offset - em->start;
1025         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1026                           em->block_start + offset, page);
1027         free_extent_map(em);
1028 }
1029
1030 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1031 {
1032 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1033         bdi_init(bdi);
1034 #endif
1035         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1036         bdi->state              = 0;
1037         bdi->capabilities       = default_backing_dev_info.capabilities;
1038         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1039         bdi->unplug_io_data     = info;
1040         bdi->congested_fn       = btrfs_congested_fn;
1041         bdi->congested_data     = info;
1042         return 0;
1043 }
1044
1045 static int bio_ready_for_csum(struct bio *bio)
1046 {
1047         u64 length = 0;
1048         u64 buf_len = 0;
1049         u64 start = 0;
1050         struct page *page;
1051         struct extent_io_tree *io_tree = NULL;
1052         struct btrfs_fs_info *info = NULL;
1053         struct bio_vec *bvec;
1054         int i;
1055         int ret;
1056
1057         bio_for_each_segment(bvec, bio, i) {
1058                 page = bvec->bv_page;
1059                 if (page->private == EXTENT_PAGE_PRIVATE) {
1060                         length += bvec->bv_len;
1061                         continue;
1062                 }
1063                 if (!page->private) {
1064                         length += bvec->bv_len;
1065                         continue;
1066                 }
1067                 length = bvec->bv_len;
1068                 buf_len = page->private >> 2;
1069                 start = page_offset(page) + bvec->bv_offset;
1070                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1071                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1072         }
1073         /* are we fully contained in this bio? */
1074         if (buf_len <= length)
1075                 return 1;
1076
1077         ret = extent_range_uptodate(io_tree, start + length,
1078                                     start + buf_len - 1);
1079         if (ret == 1)
1080                 return ret;
1081         return ret;
1082 }
1083
1084 /*
1085  * called by the kthread helper functions to finally call the bio end_io
1086  * functions.  This is where read checksum verification actually happens
1087  */
1088 static void end_workqueue_fn(struct btrfs_work *work)
1089 {
1090         struct bio *bio;
1091         struct end_io_wq *end_io_wq;
1092         struct btrfs_fs_info *fs_info;
1093         int error;
1094
1095         end_io_wq = container_of(work, struct end_io_wq, work);
1096         bio = end_io_wq->bio;
1097         fs_info = end_io_wq->info;
1098
1099         /* metadata bios are special because the whole tree block must
1100          * be checksummed at once.  This makes sure the entire block is in
1101          * ram and up to date before trying to verify things.  For
1102          * blocksize <= pagesize, it is basically a noop
1103          */
1104         if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1105                 btrfs_queue_worker(&fs_info->endio_workers,
1106                                    &end_io_wq->work);
1107                 return;
1108         }
1109         error = end_io_wq->error;
1110         bio->bi_private = end_io_wq->private;
1111         bio->bi_end_io = end_io_wq->end_io;
1112         kfree(end_io_wq);
1113 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1114         bio_endio(bio, bio->bi_size, error);
1115 #else
1116         bio_endio(bio, error);
1117 #endif
1118 }
1119
1120 static int cleaner_kthread(void *arg)
1121 {
1122         struct btrfs_root *root = arg;
1123
1124         do {
1125                 smp_mb();
1126                 if (root->fs_info->closing)
1127                         break;
1128
1129                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1130                 mutex_lock(&root->fs_info->cleaner_mutex);
1131                 btrfs_clean_old_snapshots(root);
1132                 mutex_unlock(&root->fs_info->cleaner_mutex);
1133
1134                 if (freezing(current)) {
1135                         refrigerator();
1136                 } else {
1137                         smp_mb();
1138                         if (root->fs_info->closing)
1139                                 break;
1140                         set_current_state(TASK_INTERRUPTIBLE);
1141                         schedule();
1142                         __set_current_state(TASK_RUNNING);
1143                 }
1144         } while (!kthread_should_stop());
1145         return 0;
1146 }
1147
1148 static int transaction_kthread(void *arg)
1149 {
1150         struct btrfs_root *root = arg;
1151         struct btrfs_trans_handle *trans;
1152         struct btrfs_transaction *cur;
1153         unsigned long now;
1154         unsigned long delay;
1155         int ret;
1156
1157         do {
1158                 smp_mb();
1159                 if (root->fs_info->closing)
1160                         break;
1161
1162                 delay = HZ * 30;
1163                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1164                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1165
1166                 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1167                         printk("btrfs: total reference cache size %Lu\n",
1168                                 root->fs_info->total_ref_cache_size);
1169                 }
1170
1171                 mutex_lock(&root->fs_info->trans_mutex);
1172                 cur = root->fs_info->running_transaction;
1173                 if (!cur) {
1174                         mutex_unlock(&root->fs_info->trans_mutex);
1175                         goto sleep;
1176                 }
1177
1178                 now = get_seconds();
1179                 if (now < cur->start_time || now - cur->start_time < 30) {
1180                         mutex_unlock(&root->fs_info->trans_mutex);
1181                         delay = HZ * 5;
1182                         goto sleep;
1183                 }
1184                 mutex_unlock(&root->fs_info->trans_mutex);
1185                 trans = btrfs_start_transaction(root, 1);
1186                 ret = btrfs_commit_transaction(trans, root);
1187 sleep:
1188                 wake_up_process(root->fs_info->cleaner_kthread);
1189                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1190
1191                 if (freezing(current)) {
1192                         refrigerator();
1193                 } else {
1194                         if (root->fs_info->closing)
1195                                 break;
1196                         set_current_state(TASK_INTERRUPTIBLE);
1197                         schedule_timeout(delay);
1198                         __set_current_state(TASK_RUNNING);
1199                 }
1200         } while (!kthread_should_stop());
1201         return 0;
1202 }
1203
1204 struct btrfs_root *open_ctree(struct super_block *sb,
1205                               struct btrfs_fs_devices *fs_devices,
1206                               char *options)
1207 {
1208         u32 sectorsize;
1209         u32 nodesize;
1210         u32 leafsize;
1211         u32 blocksize;
1212         u32 stripesize;
1213         struct buffer_head *bh;
1214         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1215                                                  GFP_NOFS);
1216         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1217                                                GFP_NOFS);
1218         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1219                                                 GFP_NOFS);
1220         struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1221                                                 GFP_NOFS);
1222         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1223                                               GFP_NOFS);
1224         int ret;
1225         int err = -EINVAL;
1226
1227         struct btrfs_super_block *disk_super;
1228
1229         if (!extent_root || !tree_root || !fs_info) {
1230                 err = -ENOMEM;
1231                 goto fail;
1232         }
1233         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1234         INIT_LIST_HEAD(&fs_info->trans_list);
1235         INIT_LIST_HEAD(&fs_info->dead_roots);
1236         INIT_LIST_HEAD(&fs_info->hashers);
1237         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1238         spin_lock_init(&fs_info->hash_lock);
1239         spin_lock_init(&fs_info->delalloc_lock);
1240         spin_lock_init(&fs_info->new_trans_lock);
1241         spin_lock_init(&fs_info->ref_cache_lock);
1242
1243         init_completion(&fs_info->kobj_unregister);
1244         fs_info->tree_root = tree_root;
1245         fs_info->extent_root = extent_root;
1246         fs_info->chunk_root = chunk_root;
1247         fs_info->dev_root = dev_root;
1248         fs_info->fs_devices = fs_devices;
1249         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1250         INIT_LIST_HEAD(&fs_info->space_info);
1251         btrfs_mapping_init(&fs_info->mapping_tree);
1252         atomic_set(&fs_info->nr_async_submits, 0);
1253         atomic_set(&fs_info->throttles, 0);
1254         atomic_set(&fs_info->throttle_gen, 0);
1255         fs_info->sb = sb;
1256         fs_info->max_extent = (u64)-1;
1257         fs_info->max_inline = 8192 * 1024;
1258         setup_bdi(fs_info, &fs_info->bdi);
1259         fs_info->btree_inode = new_inode(sb);
1260         fs_info->btree_inode->i_ino = 1;
1261         fs_info->btree_inode->i_nlink = 1;
1262         fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1263
1264         INIT_LIST_HEAD(&fs_info->ordered_extents);
1265         spin_lock_init(&fs_info->ordered_extent_lock);
1266
1267         sb->s_blocksize = 4096;
1268         sb->s_blocksize_bits = blksize_bits(4096);
1269
1270         /*
1271          * we set the i_size on the btree inode to the max possible int.
1272          * the real end of the address space is determined by all of
1273          * the devices in the system
1274          */
1275         fs_info->btree_inode->i_size = OFFSET_MAX;
1276         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1277         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1278
1279         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1280                              fs_info->btree_inode->i_mapping,
1281                              GFP_NOFS);
1282         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1283                              GFP_NOFS);
1284
1285         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1286
1287         extent_io_tree_init(&fs_info->free_space_cache,
1288                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1289         extent_io_tree_init(&fs_info->block_group_cache,
1290                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1291         extent_io_tree_init(&fs_info->pinned_extents,
1292                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1293         extent_io_tree_init(&fs_info->pending_del,
1294                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1295         extent_io_tree_init(&fs_info->extent_ins,
1296                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1297         fs_info->do_barriers = 1;
1298
1299         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1300         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1301                sizeof(struct btrfs_key));
1302         insert_inode_hash(fs_info->btree_inode);
1303         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1304
1305         mutex_init(&fs_info->trans_mutex);
1306         mutex_init(&fs_info->drop_mutex);
1307         mutex_init(&fs_info->alloc_mutex);
1308         mutex_init(&fs_info->chunk_mutex);
1309         mutex_init(&fs_info->transaction_kthread_mutex);
1310         mutex_init(&fs_info->cleaner_mutex);
1311         mutex_init(&fs_info->volume_mutex);
1312         init_waitqueue_head(&fs_info->transaction_throttle);
1313         init_waitqueue_head(&fs_info->transaction_wait);
1314
1315 #if 0
1316         ret = add_hasher(fs_info, "crc32c");
1317         if (ret) {
1318                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1319                 err = -ENOMEM;
1320                 goto fail_iput;
1321         }
1322 #endif
1323         __setup_root(4096, 4096, 4096, 4096, tree_root,
1324                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1325
1326
1327         bh = __bread(fs_devices->latest_bdev,
1328                      BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1329         if (!bh)
1330                 goto fail_iput;
1331
1332         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1333         brelse(bh);
1334
1335         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1336
1337         disk_super = &fs_info->super_copy;
1338         if (!btrfs_super_root(disk_super))
1339                 goto fail_sb_buffer;
1340
1341         err = btrfs_parse_options(tree_root, options);
1342         if (err)
1343                 goto fail_sb_buffer;
1344
1345         /*
1346          * we need to start all the end_io workers up front because the
1347          * queue work function gets called at interrupt time, and so it
1348          * cannot dynamically grow.
1349          */
1350         btrfs_init_workers(&fs_info->workers, fs_info->thread_pool_size);
1351         btrfs_init_workers(&fs_info->submit_workers, fs_info->thread_pool_size);
1352
1353         /* a higher idle thresh on the submit workers makes it much more
1354          * likely that bios will be send down in a sane order to the
1355          * devices
1356          */
1357         fs_info->submit_workers.idle_thresh = 64;
1358
1359         btrfs_init_workers(&fs_info->fixup_workers, 1);
1360         btrfs_init_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1361         btrfs_init_workers(&fs_info->endio_write_workers,
1362                            fs_info->thread_pool_size);
1363
1364         /*
1365          * endios are largely parallel and should have a very
1366          * low idle thresh
1367          */
1368         fs_info->endio_workers.idle_thresh = 4;
1369         fs_info->endio_write_workers.idle_thresh = 4;
1370
1371         btrfs_start_workers(&fs_info->workers, 1);
1372         btrfs_start_workers(&fs_info->submit_workers, 1);
1373         btrfs_start_workers(&fs_info->fixup_workers, 1);
1374         btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1375         btrfs_start_workers(&fs_info->endio_write_workers,
1376                             fs_info->thread_pool_size);
1377
1378         err = -EINVAL;
1379         if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1380                 printk("Btrfs: wanted %llu devices, but found %llu\n",
1381                        (unsigned long long)btrfs_super_num_devices(disk_super),
1382                        (unsigned long long)fs_devices->open_devices);
1383                 if (btrfs_test_opt(tree_root, DEGRADED))
1384                         printk("continuing in degraded mode\n");
1385                 else {
1386                         goto fail_sb_buffer;
1387                 }
1388         }
1389
1390         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1391
1392         nodesize = btrfs_super_nodesize(disk_super);
1393         leafsize = btrfs_super_leafsize(disk_super);
1394         sectorsize = btrfs_super_sectorsize(disk_super);
1395         stripesize = btrfs_super_stripesize(disk_super);
1396         tree_root->nodesize = nodesize;
1397         tree_root->leafsize = leafsize;
1398         tree_root->sectorsize = sectorsize;
1399         tree_root->stripesize = stripesize;
1400
1401         sb->s_blocksize = sectorsize;
1402         sb->s_blocksize_bits = blksize_bits(sectorsize);
1403
1404         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1405                     sizeof(disk_super->magic))) {
1406                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1407                 goto fail_sb_buffer;
1408         }
1409
1410         mutex_lock(&fs_info->chunk_mutex);
1411         ret = btrfs_read_sys_array(tree_root);
1412         mutex_unlock(&fs_info->chunk_mutex);
1413         if (ret) {
1414                 printk("btrfs: failed to read the system array on %s\n",
1415                        sb->s_id);
1416                 goto fail_sys_array;
1417         }
1418
1419         blocksize = btrfs_level_size(tree_root,
1420                                      btrfs_super_chunk_root_level(disk_super));
1421
1422         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1423                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1424
1425         chunk_root->node = read_tree_block(chunk_root,
1426                                            btrfs_super_chunk_root(disk_super),
1427                                            blocksize, 0);
1428         BUG_ON(!chunk_root->node);
1429
1430         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1431                  (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1432                  BTRFS_UUID_SIZE);
1433
1434         mutex_lock(&fs_info->chunk_mutex);
1435         ret = btrfs_read_chunk_tree(chunk_root);
1436         mutex_unlock(&fs_info->chunk_mutex);
1437         BUG_ON(ret);
1438
1439         btrfs_close_extra_devices(fs_devices);
1440
1441         blocksize = btrfs_level_size(tree_root,
1442                                      btrfs_super_root_level(disk_super));
1443
1444
1445         tree_root->node = read_tree_block(tree_root,
1446                                           btrfs_super_root(disk_super),
1447                                           blocksize, 0);
1448         if (!tree_root->node)
1449                 goto fail_sb_buffer;
1450
1451
1452         ret = find_and_setup_root(tree_root, fs_info,
1453                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1454         if (ret)
1455                 goto fail_tree_root;
1456         extent_root->track_dirty = 1;
1457
1458         ret = find_and_setup_root(tree_root, fs_info,
1459                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1460         dev_root->track_dirty = 1;
1461
1462         if (ret)
1463                 goto fail_extent_root;
1464
1465         btrfs_read_block_groups(extent_root);
1466
1467         fs_info->generation = btrfs_super_generation(disk_super) + 1;
1468         fs_info->data_alloc_profile = (u64)-1;
1469         fs_info->metadata_alloc_profile = (u64)-1;
1470         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1471         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1472                                                "btrfs-cleaner");
1473         if (!fs_info->cleaner_kthread)
1474                 goto fail_extent_root;
1475
1476         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1477                                                    tree_root,
1478                                                    "btrfs-transaction");
1479         if (!fs_info->transaction_kthread)
1480                 goto fail_cleaner;
1481
1482
1483         return tree_root;
1484
1485 fail_cleaner:
1486         kthread_stop(fs_info->cleaner_kthread);
1487 fail_extent_root:
1488         free_extent_buffer(extent_root->node);
1489 fail_tree_root:
1490         free_extent_buffer(tree_root->node);
1491 fail_sys_array:
1492 fail_sb_buffer:
1493         btrfs_stop_workers(&fs_info->fixup_workers);
1494         btrfs_stop_workers(&fs_info->workers);
1495         btrfs_stop_workers(&fs_info->endio_workers);
1496         btrfs_stop_workers(&fs_info->endio_write_workers);
1497         btrfs_stop_workers(&fs_info->submit_workers);
1498 fail_iput:
1499         iput(fs_info->btree_inode);
1500 fail:
1501         btrfs_close_devices(fs_info->fs_devices);
1502         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1503
1504         kfree(extent_root);
1505         kfree(tree_root);
1506 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1507         bdi_destroy(&fs_info->bdi);
1508 #endif
1509         kfree(fs_info);
1510         return ERR_PTR(err);
1511 }
1512
1513 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1514 {
1515         char b[BDEVNAME_SIZE];
1516
1517         if (uptodate) {
1518                 set_buffer_uptodate(bh);
1519         } else {
1520                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1521                         printk(KERN_WARNING "lost page write due to "
1522                                         "I/O error on %s\n",
1523                                        bdevname(bh->b_bdev, b));
1524                 }
1525                 /* note, we dont' set_buffer_write_io_error because we have
1526                  * our own ways of dealing with the IO errors
1527                  */
1528                 clear_buffer_uptodate(bh);
1529         }
1530         unlock_buffer(bh);
1531         put_bh(bh);
1532 }
1533
1534 int write_all_supers(struct btrfs_root *root)
1535 {
1536         struct list_head *cur;
1537         struct list_head *head = &root->fs_info->fs_devices->devices;
1538         struct btrfs_device *dev;
1539         struct btrfs_super_block *sb;
1540         struct btrfs_dev_item *dev_item;
1541         struct buffer_head *bh;
1542         int ret;
1543         int do_barriers;
1544         int max_errors;
1545         int total_errors = 0;
1546         u32 crc;
1547         u64 flags;
1548
1549         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1550         do_barriers = !btrfs_test_opt(root, NOBARRIER);
1551
1552         sb = &root->fs_info->super_for_commit;
1553         dev_item = &sb->dev_item;
1554         list_for_each(cur, head) {
1555                 dev = list_entry(cur, struct btrfs_device, dev_list);
1556                 if (!dev->bdev) {
1557                         total_errors++;
1558                         continue;
1559                 }
1560                 if (!dev->in_fs_metadata)
1561                         continue;
1562
1563                 btrfs_set_stack_device_type(dev_item, dev->type);
1564                 btrfs_set_stack_device_id(dev_item, dev->devid);
1565                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1566                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1567                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1568                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1569                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1570                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1571                 flags = btrfs_super_flags(sb);
1572                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1573
1574
1575                 crc = ~(u32)0;
1576                 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1577                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1578                 btrfs_csum_final(crc, sb->csum);
1579
1580                 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1581                               BTRFS_SUPER_INFO_SIZE);
1582
1583                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1584                 dev->pending_io = bh;
1585
1586                 get_bh(bh);
1587                 set_buffer_uptodate(bh);
1588                 lock_buffer(bh);
1589                 bh->b_end_io = btrfs_end_buffer_write_sync;
1590
1591                 if (do_barriers && dev->barriers) {
1592                         ret = submit_bh(WRITE_BARRIER, bh);
1593                         if (ret == -EOPNOTSUPP) {
1594                                 printk("btrfs: disabling barriers on dev %s\n",
1595                                        dev->name);
1596                                 set_buffer_uptodate(bh);
1597                                 dev->barriers = 0;
1598                                 get_bh(bh);
1599                                 lock_buffer(bh);
1600                                 ret = submit_bh(WRITE, bh);
1601                         }
1602                 } else {
1603                         ret = submit_bh(WRITE, bh);
1604                 }
1605                 if (ret)
1606                         total_errors++;
1607         }
1608         if (total_errors > max_errors) {
1609                 printk("btrfs: %d errors while writing supers\n", total_errors);
1610                 BUG();
1611         }
1612         total_errors = 0;
1613
1614         list_for_each(cur, head) {
1615                 dev = list_entry(cur, struct btrfs_device, dev_list);
1616                 if (!dev->bdev)
1617                         continue;
1618                 if (!dev->in_fs_metadata)
1619                         continue;
1620
1621                 BUG_ON(!dev->pending_io);
1622                 bh = dev->pending_io;
1623                 wait_on_buffer(bh);
1624                 if (!buffer_uptodate(dev->pending_io)) {
1625                         if (do_barriers && dev->barriers) {
1626                                 printk("btrfs: disabling barriers on dev %s\n",
1627                                        dev->name);
1628                                 set_buffer_uptodate(bh);
1629                                 get_bh(bh);
1630                                 lock_buffer(bh);
1631                                 dev->barriers = 0;
1632                                 ret = submit_bh(WRITE, bh);
1633                                 BUG_ON(ret);
1634                                 wait_on_buffer(bh);
1635                                 if (!buffer_uptodate(bh))
1636                                         total_errors++;
1637                         } else {
1638                                 total_errors++;
1639                         }
1640
1641                 }
1642                 dev->pending_io = NULL;
1643                 brelse(bh);
1644         }
1645         if (total_errors > max_errors) {
1646                 printk("btrfs: %d errors while writing supers\n", total_errors);
1647                 BUG();
1648         }
1649         return 0;
1650 }
1651
1652 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1653                       *root)
1654 {
1655         int ret;
1656
1657         ret = write_all_supers(root);
1658         return ret;
1659 }
1660
1661 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1662 {
1663         radix_tree_delete(&fs_info->fs_roots_radix,
1664                           (unsigned long)root->root_key.objectid);
1665         if (root->in_sysfs)
1666                 btrfs_sysfs_del_root(root);
1667         if (root->inode)
1668                 iput(root->inode);
1669         if (root->node)
1670                 free_extent_buffer(root->node);
1671         if (root->commit_root)
1672                 free_extent_buffer(root->commit_root);
1673         if (root->name)
1674                 kfree(root->name);
1675         kfree(root);
1676         return 0;
1677 }
1678
1679 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1680 {
1681         int ret;
1682         struct btrfs_root *gang[8];
1683         int i;
1684
1685         while(1) {
1686                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1687                                              (void **)gang, 0,
1688                                              ARRAY_SIZE(gang));
1689                 if (!ret)
1690                         break;
1691                 for (i = 0; i < ret; i++)
1692                         btrfs_free_fs_root(fs_info, gang[i]);
1693         }
1694         return 0;
1695 }
1696
1697 int close_ctree(struct btrfs_root *root)
1698 {
1699         int ret;
1700         struct btrfs_trans_handle *trans;
1701         struct btrfs_fs_info *fs_info = root->fs_info;
1702
1703         fs_info->closing = 1;
1704         smp_mb();
1705
1706         kthread_stop(root->fs_info->transaction_kthread);
1707         kthread_stop(root->fs_info->cleaner_kthread);
1708
1709         btrfs_clean_old_snapshots(root);
1710         trans = btrfs_start_transaction(root, 1);
1711         ret = btrfs_commit_transaction(trans, root);
1712         /* run commit again to  drop the original snapshot */
1713         trans = btrfs_start_transaction(root, 1);
1714         btrfs_commit_transaction(trans, root);
1715         ret = btrfs_write_and_wait_transaction(NULL, root);
1716         BUG_ON(ret);
1717
1718         write_ctree_super(NULL, root);
1719
1720         if (fs_info->delalloc_bytes) {
1721                 printk("btrfs: at unmount delalloc count %Lu\n",
1722                        fs_info->delalloc_bytes);
1723         }
1724         if (fs_info->total_ref_cache_size) {
1725                 printk("btrfs: at umount reference cache size %Lu\n",
1726                         fs_info->total_ref_cache_size);
1727         }
1728
1729         if (fs_info->extent_root->node)
1730                 free_extent_buffer(fs_info->extent_root->node);
1731
1732         if (fs_info->tree_root->node)
1733                 free_extent_buffer(fs_info->tree_root->node);
1734
1735         if (root->fs_info->chunk_root->node);
1736                 free_extent_buffer(root->fs_info->chunk_root->node);
1737
1738         if (root->fs_info->dev_root->node);
1739                 free_extent_buffer(root->fs_info->dev_root->node);
1740
1741         btrfs_free_block_groups(root->fs_info);
1742         del_fs_roots(fs_info);
1743
1744         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1745
1746         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1747
1748         btrfs_stop_workers(&fs_info->fixup_workers);
1749         btrfs_stop_workers(&fs_info->workers);
1750         btrfs_stop_workers(&fs_info->endio_workers);
1751         btrfs_stop_workers(&fs_info->endio_write_workers);
1752         btrfs_stop_workers(&fs_info->submit_workers);
1753
1754         iput(fs_info->btree_inode);
1755 #if 0
1756         while(!list_empty(&fs_info->hashers)) {
1757                 struct btrfs_hasher *hasher;
1758                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1759                                     hashers);
1760                 list_del(&hasher->hashers);
1761                 crypto_free_hash(&fs_info->hash_tfm);
1762                 kfree(hasher);
1763         }
1764 #endif
1765         btrfs_close_devices(fs_info->fs_devices);
1766         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1767
1768 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1769         bdi_destroy(&fs_info->bdi);
1770 #endif
1771
1772         kfree(fs_info->extent_root);
1773         kfree(fs_info->tree_root);
1774         kfree(fs_info->chunk_root);
1775         kfree(fs_info->dev_root);
1776         return 0;
1777 }
1778
1779 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1780 {
1781         int ret;
1782         struct inode *btree_inode = buf->first_page->mapping->host;
1783
1784         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1785         if (!ret)
1786                 return ret;
1787
1788         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1789                                     parent_transid);
1790         return !ret;
1791 }
1792
1793 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1794 {
1795         struct inode *btree_inode = buf->first_page->mapping->host;
1796         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1797                                           buf);
1798 }
1799
1800 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1801 {
1802         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1803         u64 transid = btrfs_header_generation(buf);
1804         struct inode *btree_inode = root->fs_info->btree_inode;
1805
1806         WARN_ON(!btrfs_tree_locked(buf));
1807         if (transid != root->fs_info->generation) {
1808                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1809                         (unsigned long long)buf->start,
1810                         transid, root->fs_info->generation);
1811                 WARN_ON(1);
1812         }
1813         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1814 }
1815
1816 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1817 {
1818         /*
1819          * looks as though older kernels can get into trouble with
1820          * this code, they end up stuck in balance_dirty_pages forever
1821          */
1822         struct extent_io_tree *tree;
1823         u64 num_dirty;
1824         u64 start = 0;
1825         unsigned long thresh = 16 * 1024 * 1024;
1826         tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1827
1828         if (current_is_pdflush())
1829                 return;
1830
1831         num_dirty = count_range_bits(tree, &start, (u64)-1,
1832                                      thresh, EXTENT_DIRTY);
1833         if (num_dirty > thresh) {
1834                 balance_dirty_pages_ratelimited_nr(
1835                                    root->fs_info->btree_inode->i_mapping, 1);
1836         }
1837         return;
1838 }
1839
1840 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1841 {
1842         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1843         int ret;
1844         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1845         if (ret == 0) {
1846                 buf->flags |= EXTENT_UPTODATE;
1847         }
1848         return ret;
1849 }
1850
1851 static struct extent_io_ops btree_extent_io_ops = {
1852         .writepage_io_hook = btree_writepage_io_hook,
1853         .readpage_end_io_hook = btree_readpage_end_io_hook,
1854         .submit_bio_hook = btree_submit_bio_hook,
1855         /* note we're sharing with inode.c for the merge bio hook */
1856         .merge_bio_hook = btrfs_merge_bio_hook,
1857 };