+/*
+ * this does all the hard work for inserting an inline extent into
+ * the btree. The caller should have done a btrfs_drop_extents so that
+ * no overlapping inline items exist in the btree
+ */
+static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root, struct inode *inode,
+ u64 start, size_t size, size_t compressed_size,
+ struct page **compressed_pages)
+{
+ struct btrfs_key key;
+ struct btrfs_path *path;
+ struct extent_buffer *leaf;
+ struct page *page = NULL;
+ char *kaddr;
+ unsigned long ptr;
+ struct btrfs_file_extent_item *ei;
+ int err = 0;
+ int ret;
+ size_t cur_size = size;
+ size_t datasize;
+ unsigned long offset;
+ int use_compress = 0;
+
+ if (compressed_size && compressed_pages) {
+ use_compress = 1;
+ cur_size = compressed_size;
+ }
+
+ path = btrfs_alloc_path();
+ if (!path)
+ return -ENOMEM;
+
+ btrfs_set_trans_block_group(trans, inode);
+
+ key.objectid = inode->i_ino;
+ key.offset = start;
+ btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
+ datasize = btrfs_file_extent_calc_inline_size(cur_size);
+
+ inode_add_bytes(inode, size);
+ ret = btrfs_insert_empty_item(trans, root, path, &key,
+ datasize);
+ BUG_ON(ret);
+ if (ret) {
+ err = ret;
+ goto fail;
+ }
+ leaf = path->nodes[0];
+ ei = btrfs_item_ptr(leaf, path->slots[0],
+ struct btrfs_file_extent_item);
+ btrfs_set_file_extent_generation(leaf, ei, trans->transid);
+ btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
+ btrfs_set_file_extent_encryption(leaf, ei, 0);
+ btrfs_set_file_extent_other_encoding(leaf, ei, 0);
+ btrfs_set_file_extent_ram_bytes(leaf, ei, size);
+ ptr = btrfs_file_extent_inline_start(ei);
+
+ if (use_compress) {
+ struct page *cpage;
+ int i = 0;
+ while (compressed_size > 0) {
+ cpage = compressed_pages[i];
+ cur_size = min_t(unsigned long, compressed_size,
+ PAGE_CACHE_SIZE);
+
+ kaddr = kmap(cpage);
+ write_extent_buffer(leaf, kaddr, ptr, cur_size);
+ kunmap(cpage);
+
+ i++;
+ ptr += cur_size;
+ compressed_size -= cur_size;
+ }
+ btrfs_set_file_extent_compression(leaf, ei,
+ BTRFS_COMPRESS_ZLIB);
+ } else {
+ page = find_get_page(inode->i_mapping,
+ start >> PAGE_CACHE_SHIFT);
+ btrfs_set_file_extent_compression(leaf, ei, 0);
+ kaddr = kmap_atomic(page, KM_USER0);
+ offset = start & (PAGE_CACHE_SIZE - 1);
+ write_extent_buffer(leaf, kaddr + offset, ptr, size);
+ kunmap_atomic(kaddr, KM_USER0);
+ page_cache_release(page);
+ }
+ btrfs_mark_buffer_dirty(leaf);
+ btrfs_free_path(path);
+
+ BTRFS_I(inode)->disk_i_size = inode->i_size;
+ btrfs_update_inode(trans, root, inode);
+ return 0;
+fail:
+ btrfs_free_path(path);
+ return err;
+}
+
+
+/*
+ * conditionally insert an inline extent into the file. This
+ * does the checks required to make sure the data is small enough
+ * to fit as an inline extent.
+ */
+static int cow_file_range_inline(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ struct inode *inode, u64 start, u64 end,
+ size_t compressed_size,
+ struct page **compressed_pages)
+{
+ u64 isize = i_size_read(inode);
+ u64 actual_end = min(end + 1, isize);
+ u64 inline_len = actual_end - start;
+ u64 aligned_end = (end + root->sectorsize - 1) &
+ ~((u64)root->sectorsize - 1);
+ u64 hint_byte;
+ u64 data_len = inline_len;
+ int ret;
+
+ if (compressed_size)
+ data_len = compressed_size;
+
+ if (start > 0 ||
+ actual_end >= PAGE_CACHE_SIZE ||
+ data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
+ (!compressed_size &&
+ (actual_end & (root->sectorsize - 1)) == 0) ||
+ end + 1 < isize ||
+ data_len > root->fs_info->max_inline) {
+ return 1;
+ }
+
+ ret = btrfs_drop_extents(trans, root, inode, start,
+ aligned_end, start, &hint_byte);
+ BUG_ON(ret);
+
+ if (isize > actual_end)
+ inline_len = min_t(u64, isize, actual_end);
+ ret = insert_inline_extent(trans, root, inode, start,
+ inline_len, compressed_size,
+ compressed_pages);
+ BUG_ON(ret);
+ btrfs_drop_extent_cache(inode, start, aligned_end, 0);
+ return 0;
+}
+
+struct async_extent {
+ u64 start;
+ u64 ram_size;
+ u64 compressed_size;
+ struct page **pages;
+ unsigned long nr_pages;
+ struct list_head list;
+};
+
+struct async_cow {
+ struct inode *inode;
+ struct btrfs_root *root;
+ struct page *locked_page;
+ u64 start;
+ u64 end;
+ struct list_head extents;
+ struct btrfs_work work;
+};
+
+static noinline int add_async_extent(struct async_cow *cow,
+ u64 start, u64 ram_size,
+ u64 compressed_size,
+ struct page **pages,
+ unsigned long nr_pages)
+{
+ struct async_extent *async_extent;
+
+ async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
+ async_extent->start = start;
+ async_extent->ram_size = ram_size;
+ async_extent->compressed_size = compressed_size;
+ async_extent->pages = pages;
+ async_extent->nr_pages = nr_pages;
+ list_add_tail(&async_extent->list, &cow->extents);
+ return 0;
+}
+
+/*
+ * we create compressed extents in two phases. The first
+ * phase compresses a range of pages that have already been
+ * locked (both pages and state bits are locked).
+ *
+ * This is done inside an ordered work queue, and the compression
+ * is spread across many cpus. The actual IO submission is step
+ * two, and the ordered work queue takes care of making sure that
+ * happens in the same order things were put onto the queue by
+ * writepages and friends.
+ *
+ * If this code finds it can't get good compression, it puts an
+ * entry onto the work queue to write the uncompressed bytes. This
+ * makes sure that both compressed inodes and uncompressed inodes
+ * are written in the same order that pdflush sent them down.
+ */
+static noinline int compress_file_range(struct inode *inode,
+ struct page *locked_page,
+ u64 start, u64 end,
+ struct async_cow *async_cow,
+ int *num_added)
+{
+ struct btrfs_root *root = BTRFS_I(inode)->root;
+ struct btrfs_trans_handle *trans;
+ u64 num_bytes;
+ u64 orig_start;
+ u64 disk_num_bytes;
+ u64 blocksize = root->sectorsize;
+ u64 actual_end;
+ u64 isize = i_size_read(inode);
+ int ret = 0;
+ struct page **pages = NULL;
+ unsigned long nr_pages;
+ unsigned long nr_pages_ret = 0;
+ unsigned long total_compressed = 0;
+ unsigned long total_in = 0;
+ unsigned long max_compressed = 128 * 1024;
+ unsigned long max_uncompressed = 128 * 1024;
+ int i;
+ int will_compress;
+
+ orig_start = start;
+
+ actual_end = min_t(u64, isize, end + 1);
+again:
+ will_compress = 0;
+ nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
+ nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
+
+ total_compressed = actual_end - start;
+
+ /* we want to make sure that amount of ram required to uncompress
+ * an extent is reasonable, so we limit the total size in ram
+ * of a compressed extent to 128k. This is a crucial number
+ * because it also controls how easily we can spread reads across
+ * cpus for decompression.
+ *
+ * We also want to make sure the amount of IO required to do
+ * a random read is reasonably small, so we limit the size of
+ * a compressed extent to 128k.
+ */
+ total_compressed = min(total_compressed, max_uncompressed);
+ num_bytes = (end - start + blocksize) & ~(blocksize - 1);
+ num_bytes = max(blocksize, num_bytes);
+ disk_num_bytes = num_bytes;
+ total_in = 0;
+ ret = 0;
+
+ /*
+ * we do compression for mount -o compress and when the
+ * inode has not been flagged as nocompress. This flag can
+ * change at any time if we discover bad compression ratios.
+ */
+ if (!btrfs_test_flag(inode, NOCOMPRESS) &&
+ btrfs_test_opt(root, COMPRESS)) {
+ WARN_ON(pages);
+ pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
+
+ ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
+ total_compressed, pages,
+ nr_pages, &nr_pages_ret,
+ &total_in,
+ &total_compressed,
+ max_compressed);
+
+ if (!ret) {
+ unsigned long offset = total_compressed &
+ (PAGE_CACHE_SIZE - 1);
+ struct page *page = pages[nr_pages_ret - 1];
+ char *kaddr;
+
+ /* zero the tail end of the last page, we might be
+ * sending it down to disk
+ */
+ if (offset) {
+ kaddr = kmap_atomic(page, KM_USER0);
+ memset(kaddr + offset, 0,
+ PAGE_CACHE_SIZE - offset);
+ kunmap_atomic(kaddr, KM_USER0);
+ }
+ will_compress = 1;
+ }
+ }
+ if (start == 0) {
+ trans = btrfs_join_transaction(root, 1);
+ BUG_ON(!trans);
+ btrfs_set_trans_block_group(trans, inode);
+
+ /* lets try to make an inline extent */
+ if (ret || total_in < (actual_end - start)) {
+ /* we didn't compress the entire range, try
+ * to make an uncompressed inline extent.
+ */
+ ret = cow_file_range_inline(trans, root, inode,
+ start, end, 0, NULL);
+ } else {
+ /* try making a compressed inline extent */
+ ret = cow_file_range_inline(trans, root, inode,
+ start, end,
+ total_compressed, pages);
+ }
+ btrfs_end_transaction(trans, root);
+ if (ret == 0) {
+ /*
+ * inline extent creation worked, we don't need
+ * to create any more async work items. Unlock
+ * and free up our temp pages.
+ */
+ extent_clear_unlock_delalloc(inode,
+ &BTRFS_I(inode)->io_tree,
+ start, end, NULL, 1, 0,
+ 0, 1, 1, 1);
+ ret = 0;
+ goto free_pages_out;
+ }
+ }
+
+ if (will_compress) {
+ /*
+ * we aren't doing an inline extent round the compressed size
+ * up to a block size boundary so the allocator does sane
+ * things
+ */
+ total_compressed = (total_compressed + blocksize - 1) &
+ ~(blocksize - 1);
+
+ /*
+ * one last check to make sure the compression is really a
+ * win, compare the page count read with the blocks on disk
+ */
+ total_in = (total_in + PAGE_CACHE_SIZE - 1) &
+ ~(PAGE_CACHE_SIZE - 1);
+ if (total_compressed >= total_in) {
+ will_compress = 0;
+ } else {
+ disk_num_bytes = total_compressed;
+ num_bytes = total_in;
+ }
+ }
+ if (!will_compress && pages) {
+ /*
+ * the compression code ran but failed to make things smaller,
+ * free any pages it allocated and our page pointer array
+ */
+ for (i = 0; i < nr_pages_ret; i++) {
+ WARN_ON(pages[i]->mapping);
+ page_cache_release(pages[i]);
+ }
+ kfree(pages);
+ pages = NULL;
+ total_compressed = 0;
+ nr_pages_ret = 0;
+
+ /* flag the file so we don't compress in the future */
+ btrfs_set_flag(inode, NOCOMPRESS);
+ }
+ if (will_compress) {
+ *num_added += 1;
+
+ /* the async work queues will take care of doing actual
+ * allocation on disk for these compressed pages,
+ * and will submit them to the elevator.
+ */
+ add_async_extent(async_cow, start, num_bytes,
+ total_compressed, pages, nr_pages_ret);
+
+ if (start + num_bytes < end && start + num_bytes < actual_end) {
+ start += num_bytes;
+ pages = NULL;
+ cond_resched();
+ goto again;
+ }
+ } else {
+ /*
+ * No compression, but we still need to write the pages in
+ * the file we've been given so far. redirty the locked
+ * page if it corresponds to our extent and set things up
+ * for the async work queue to run cow_file_range to do
+ * the normal delalloc dance
+ */
+ if (page_offset(locked_page) >= start &&
+ page_offset(locked_page) <= end) {
+ __set_page_dirty_nobuffers(locked_page);
+ /* unlocked later on in the async handlers */
+ }
+ add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
+ *num_added += 1;
+ }
+
+out:
+ return 0;
+
+free_pages_out:
+ for (i = 0; i < nr_pages_ret; i++) {
+ WARN_ON(pages[i]->mapping);
+ page_cache_release(pages[i]);
+ }
+ kfree(pages);
+
+ goto out;
+}
+
+/*
+ * phase two of compressed writeback. This is the ordered portion
+ * of the code, which only gets called in the order the work was
+ * queued. We walk all the async extents created by compress_file_range
+ * and send them down to the disk.
+ */
+static noinline int submit_compressed_extents(struct inode *inode,
+ struct async_cow *async_cow)
+{
+ struct async_extent *async_extent;
+ u64 alloc_hint = 0;
+ struct btrfs_trans_handle *trans;
+ struct btrfs_key ins;
+ struct extent_map *em;
+ struct btrfs_root *root = BTRFS_I(inode)->root;
+ struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+ struct extent_io_tree *io_tree;
+ int ret;
+
+ if (list_empty(&async_cow->extents))
+ return 0;
+
+ trans = btrfs_join_transaction(root, 1);
+
+ while (!list_empty(&async_cow->extents)) {
+ async_extent = list_entry(async_cow->extents.next,
+ struct async_extent, list);
+ list_del(&async_extent->list);
+
+ io_tree = &BTRFS_I(inode)->io_tree;
+
+ /* did the compression code fall back to uncompressed IO? */
+ if (!async_extent->pages) {
+ int page_started = 0;
+ unsigned long nr_written = 0;
+
+ lock_extent(io_tree, async_extent->start,
+ async_extent->start +
+ async_extent->ram_size - 1, GFP_NOFS);
+
+ /* allocate blocks */
+ cow_file_range(inode, async_cow->locked_page,
+ async_extent->start,
+ async_extent->start +
+ async_extent->ram_size - 1,
+ &page_started, &nr_written, 0);
+
+ /*
+ * if page_started, cow_file_range inserted an
+ * inline extent and took care of all the unlocking
+ * and IO for us. Otherwise, we need to submit
+ * all those pages down to the drive.
+ */
+ if (!page_started)
+ extent_write_locked_range(io_tree,
+ inode, async_extent->start,
+ async_extent->start +
+ async_extent->ram_size - 1,
+ btrfs_get_extent,
+ WB_SYNC_ALL);
+ kfree(async_extent);
+ cond_resched();
+ continue;
+ }
+
+ lock_extent(io_tree, async_extent->start,
+ async_extent->start + async_extent->ram_size - 1,
+ GFP_NOFS);
+ /*
+ * here we're doing allocation and writeback of the
+ * compressed pages
+ */
+ btrfs_drop_extent_cache(inode, async_extent->start,
+ async_extent->start +
+ async_extent->ram_size - 1, 0);
+
+ ret = btrfs_reserve_extent(trans, root,
+ async_extent->compressed_size,
+ async_extent->compressed_size,
+ 0, alloc_hint,
+ (u64)-1, &ins, 1);
+ BUG_ON(ret);
+ em = alloc_extent_map(GFP_NOFS);
+ em->start = async_extent->start;
+ em->len = async_extent->ram_size;
+ em->orig_start = em->start;
+
+ em->block_start = ins.objectid;
+ em->block_len = ins.offset;
+ em->bdev = root->fs_info->fs_devices->latest_bdev;
+ set_bit(EXTENT_FLAG_PINNED, &em->flags);
+ set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
+
+ while (1) {
+ spin_lock(&em_tree->lock);
+ ret = add_extent_mapping(em_tree, em);
+ spin_unlock(&em_tree->lock);
+ if (ret != -EEXIST) {
+ free_extent_map(em);
+ break;
+ }
+ btrfs_drop_extent_cache(inode, async_extent->start,
+ async_extent->start +
+ async_extent->ram_size - 1, 0);
+ }
+
+ ret = btrfs_add_ordered_extent(inode, async_extent->start,
+ ins.objectid,
+ async_extent->ram_size,
+ ins.offset,
+ BTRFS_ORDERED_COMPRESSED);
+ BUG_ON(ret);
+
+ btrfs_end_transaction(trans, root);
+
+ /*
+ * clear dirty, set writeback and unlock the pages.
+ */
+ extent_clear_unlock_delalloc(inode,
+ &BTRFS_I(inode)->io_tree,
+ async_extent->start,
+ async_extent->start +
+ async_extent->ram_size - 1,
+ NULL, 1, 1, 0, 1, 1, 0);
+
+ ret = btrfs_submit_compressed_write(inode,
+ async_extent->start,
+ async_extent->ram_size,
+ ins.objectid,
+ ins.offset, async_extent->pages,
+ async_extent->nr_pages);
+
+ BUG_ON(ret);
+ trans = btrfs_join_transaction(root, 1);
+ alloc_hint = ins.objectid + ins.offset;
+ kfree(async_extent);
+ cond_resched();
+ }
+
+ btrfs_end_transaction(trans, root);
+ return 0;
+}
+
+/*
+ * when extent_io.c finds a delayed allocation range in the file,
+ * the call backs end up in this code. The basic idea is to
+ * allocate extents on disk for the range, and create ordered data structs
+ * in ram to track those extents.
+ *
+ * locked_page is the page that writepage had locked already. We use
+ * it to make sure we don't do extra locks or unlocks.
+ *
+ * *page_started is set to one if we unlock locked_page and do everything
+ * required to start IO on it. It may be clean and already done with
+ * IO when we return.
+ */
+static noinline int cow_file_range(struct inode *inode,
+ struct page *locked_page,
+ u64 start, u64 end, int *page_started,
+ unsigned long *nr_written,
+ int unlock)