btrfs: add truncate control struct
[linux-block.git] / fs / btrfs / inode-item.c
index 37f36ffdaf6b7fa9e259f94e246f1fa6015851a9..0946a3912afd61a5a559e85a9f7a46ee1df7a0b6 100644 (file)
@@ -4,6 +4,7 @@
  */
 
 #include "ctree.h"
+#include "inode-item.h"
 #include "disk-io.h"
 #include "transaction.h"
 #include "print-tree.h"
@@ -19,7 +20,7 @@ struct btrfs_inode_ref *btrfs_find_name_in_backref(struct extent_buffer *leaf,
        u32 cur_offset = 0;
        int len;
 
-       item_size = btrfs_item_size_nr(leaf, slot);
+       item_size = btrfs_item_size(leaf, slot);
        ptr = btrfs_item_ptr_offset(leaf, slot);
        while (cur_offset < item_size) {
                ref = (struct btrfs_inode_ref *)(ptr + cur_offset);
@@ -45,7 +46,7 @@ struct btrfs_inode_extref *btrfs_find_name_in_ext_backref(
        u32 cur_offset = 0;
        int ref_name_len;
 
-       item_size = btrfs_item_size_nr(leaf, slot);
+       item_size = btrfs_item_size(leaf, slot);
        ptr = btrfs_item_ptr_offset(leaf, slot);
 
        /*
@@ -139,7 +140,7 @@ static int btrfs_del_inode_extref(struct btrfs_trans_handle *trans,
        }
 
        leaf = path->nodes[0];
-       item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+       item_size = btrfs_item_size(leaf, path->slots[0]);
        if (index)
                *index = btrfs_inode_extref_index(leaf, extref);
 
@@ -208,7 +209,7 @@ int btrfs_del_inode_ref(struct btrfs_trans_handle *trans,
                goto out;
        }
        leaf = path->nodes[0];
-       item_size = btrfs_item_size_nr(leaf, path->slots[0]);
+       item_size = btrfs_item_size(leaf, path->slots[0]);
 
        if (index)
                *index = btrfs_inode_ref_index(leaf, ref);
@@ -256,7 +257,6 @@ static int btrfs_insert_inode_extref(struct btrfs_trans_handle *trans,
        struct btrfs_path *path;
        struct btrfs_key key;
        struct extent_buffer *leaf;
-       struct btrfs_item *item;
 
        key.objectid = inode_objectid;
        key.type = BTRFS_INODE_EXTREF_KEY;
@@ -282,9 +282,8 @@ static int btrfs_insert_inode_extref(struct btrfs_trans_handle *trans,
                goto out;
 
        leaf = path->nodes[0];
-       item = btrfs_item_nr(path->slots[0]);
        ptr = (unsigned long)btrfs_item_ptr(leaf, path->slots[0], char);
-       ptr += btrfs_item_size(leaf, item) - ins_len;
+       ptr += btrfs_item_size(leaf, path->slots[0]) - ins_len;
        extref = (struct btrfs_inode_extref *)ptr;
 
        btrfs_set_inode_extref_name_len(path->nodes[0], extref, name_len);
@@ -332,7 +331,7 @@ int btrfs_insert_inode_ref(struct btrfs_trans_handle *trans,
                if (ref)
                        goto out;
 
-               old_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
+               old_size = btrfs_item_size(path->nodes[0], path->slots[0]);
                btrfs_extend_item(path, ins_len);
                ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
                                     struct btrfs_inode_ref);
@@ -419,3 +418,334 @@ int btrfs_lookup_inode(struct btrfs_trans_handle *trans, struct btrfs_root
        }
        return ret;
 }
+
+/*
+ * Remove inode items from a given root.
+ *
+ * @trans:             A transaction handle.
+ * @root:              The root from which to remove items.
+ * @inode:             The inode whose items we want to remove.
+ * @control:           The btrfs_truncate_control to control how and what we
+ *                     are truncating.
+ *
+ * Remove all keys associated with the inode from the given root that have a key
+ * with a type greater than or equals to @min_type. When @min_type has a value of
+ * BTRFS_EXTENT_DATA_KEY, only remove file extent items that have an offset value
+ * greater than or equals to @new_size. If a file extent item that starts before
+ * @new_size and ends after it is found, its length is adjusted.
+ *
+ * Returns: 0 on success, < 0 on error and NEED_TRUNCATE_BLOCK when @min_type is
+ * BTRFS_EXTENT_DATA_KEY and the caller must truncate the last block.
+ */
+int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
+                              struct btrfs_root *root,
+                              struct btrfs_inode *inode,
+                              struct btrfs_truncate_control *control)
+{
+       struct btrfs_fs_info *fs_info = root->fs_info;
+       struct btrfs_path *path;
+       struct extent_buffer *leaf;
+       struct btrfs_file_extent_item *fi;
+       struct btrfs_key key;
+       struct btrfs_key found_key;
+       u64 new_size = control->new_size;
+       u64 extent_num_bytes = 0;
+       u64 extent_offset = 0;
+       u64 item_end = 0;
+       u64 last_size = new_size;
+       u32 found_type = (u8)-1;
+       int del_item;
+       int pending_del_nr = 0;
+       int pending_del_slot = 0;
+       int extent_type = -1;
+       int ret;
+       u64 ino = btrfs_ino(inode);
+       u64 bytes_deleted = 0;
+       bool be_nice = false;
+       bool should_throttle = false;
+
+       BUG_ON(new_size > 0 && control->min_type != BTRFS_EXTENT_DATA_KEY);
+
+       /*
+        * For shareable roots we want to back off from time to time, this turns
+        * out to be subvolume roots, reloc roots, and data reloc roots.
+        */
+       if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
+               be_nice = true;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+       path->reada = READA_BACK;
+
+       key.objectid = ino;
+       key.offset = (u64)-1;
+       key.type = (u8)-1;
+
+search_again:
+       /*
+        * With a 16K leaf size and 128MiB extents, you can actually queue up a
+        * huge file in a single leaf.  Most of the time that bytes_deleted is
+        * > 0, it will be huge by the time we get here
+        */
+       if (be_nice && bytes_deleted > SZ_32M &&
+           btrfs_should_end_transaction(trans)) {
+               ret = -EAGAIN;
+               goto out;
+       }
+
+       ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+       if (ret < 0)
+               goto out;
+
+       if (ret > 0) {
+               ret = 0;
+               /* There are no items in the tree for us to truncate, we're done */
+               if (path->slots[0] == 0)
+                       goto out;
+               path->slots[0]--;
+       }
+
+       while (1) {
+               u64 clear_start = 0, clear_len = 0, extent_start = 0;
+
+               fi = NULL;
+               leaf = path->nodes[0];
+               btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+               found_type = found_key.type;
+
+               if (found_key.objectid != ino)
+                       break;
+
+               if (found_type < control->min_type)
+                       break;
+
+               item_end = found_key.offset;
+               if (found_type == BTRFS_EXTENT_DATA_KEY) {
+                       fi = btrfs_item_ptr(leaf, path->slots[0],
+                                           struct btrfs_file_extent_item);
+                       extent_type = btrfs_file_extent_type(leaf, fi);
+                       if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
+                               item_end +=
+                                   btrfs_file_extent_num_bytes(leaf, fi);
+
+                               trace_btrfs_truncate_show_fi_regular(
+                                       inode, leaf, fi, found_key.offset);
+                       } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
+                               item_end += btrfs_file_extent_ram_bytes(leaf, fi);
+
+                               trace_btrfs_truncate_show_fi_inline(
+                                       inode, leaf, fi, path->slots[0],
+                                       found_key.offset);
+                       }
+                       item_end--;
+               }
+               if (found_type > control->min_type) {
+                       del_item = 1;
+               } else {
+                       if (item_end < new_size)
+                               break;
+                       if (found_key.offset >= new_size)
+                               del_item = 1;
+                       else
+                               del_item = 0;
+               }
+
+               /* FIXME, shrink the extent if the ref count is only 1 */
+               if (found_type != BTRFS_EXTENT_DATA_KEY)
+                       goto delete;
+
+               control->extents_found++;
+
+               if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
+                       u64 num_dec;
+
+                       clear_start = found_key.offset;
+                       extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
+                       if (!del_item) {
+                               u64 orig_num_bytes =
+                                       btrfs_file_extent_num_bytes(leaf, fi);
+                               extent_num_bytes = ALIGN(new_size -
+                                               found_key.offset,
+                                               fs_info->sectorsize);
+                               clear_start = ALIGN(new_size, fs_info->sectorsize);
+
+                               btrfs_set_file_extent_num_bytes(leaf, fi,
+                                                        extent_num_bytes);
+                               num_dec = (orig_num_bytes - extent_num_bytes);
+                               if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
+                                   extent_start != 0)
+                                       inode_sub_bytes(&inode->vfs_inode,
+                                                       num_dec);
+                               btrfs_mark_buffer_dirty(leaf);
+                       } else {
+                               extent_num_bytes =
+                                       btrfs_file_extent_disk_num_bytes(leaf, fi);
+                               extent_offset = found_key.offset -
+                                       btrfs_file_extent_offset(leaf, fi);
+
+                               /* FIXME blocksize != 4096 */
+                               num_dec = btrfs_file_extent_num_bytes(leaf, fi);
+                               if (extent_start != 0) {
+                                       if (test_bit(BTRFS_ROOT_SHAREABLE,
+                                                    &root->state))
+                                               inode_sub_bytes(&inode->vfs_inode,
+                                                               num_dec);
+                               }
+                       }
+                       clear_len = num_dec;
+               } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
+                       /*
+                        * We can't truncate inline items that have had
+                        * special encodings
+                        */
+                       if (!del_item &&
+                           btrfs_file_extent_encryption(leaf, fi) == 0 &&
+                           btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
+                           btrfs_file_extent_compression(leaf, fi) == 0) {
+                               u32 size = (u32)(new_size - found_key.offset);
+
+                               btrfs_set_file_extent_ram_bytes(leaf, fi, size);
+                               size = btrfs_file_extent_calc_inline_size(size);
+                               btrfs_truncate_item(path, size, 1);
+                       } else if (!del_item) {
+                               /*
+                                * We have to bail so the last_size is set to
+                                * just before this extent.
+                                */
+                               ret = BTRFS_NEED_TRUNCATE_BLOCK;
+                               break;
+                       } else {
+                               /*
+                                * Inline extents are special, we just treat
+                                * them as a full sector worth in the file
+                                * extent tree just for simplicity sake.
+                                */
+                               clear_len = fs_info->sectorsize;
+                       }
+
+                       if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
+                               inode_sub_bytes(&inode->vfs_inode,
+                                               item_end + 1 - new_size);
+               }
+delete:
+               /*
+                * We use btrfs_truncate_inode_items() to clean up log trees for
+                * multiple fsyncs, and in this case we don't want to clear the
+                * file extent range because it's just the log.
+                */
+               if (root == inode->root) {
+                       ret = btrfs_inode_clear_file_extent_range(inode,
+                                                 clear_start, clear_len);
+                       if (ret) {
+                               btrfs_abort_transaction(trans, ret);
+                               break;
+                       }
+               }
+
+               if (del_item)
+                       last_size = found_key.offset;
+               else
+                       last_size = new_size;
+               if (del_item) {
+                       if (!pending_del_nr) {
+                               /* No pending yet, add ourselves */
+                               pending_del_slot = path->slots[0];
+                               pending_del_nr = 1;
+                       } else if (pending_del_nr &&
+                                  path->slots[0] + 1 == pending_del_slot) {
+                               /* Hop on the pending chunk */
+                               pending_del_nr++;
+                               pending_del_slot = path->slots[0];
+                       } else {
+                               BUG();
+                       }
+               } else {
+                       break;
+               }
+               should_throttle = false;
+
+               if (del_item && extent_start != 0 &&
+                   root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
+                       struct btrfs_ref ref = { 0 };
+
+                       bytes_deleted += extent_num_bytes;
+
+                       btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
+                                       extent_start, extent_num_bytes, 0);
+                       btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
+                                       ino, extent_offset,
+                                       root->root_key.objectid, false);
+                       ret = btrfs_free_extent(trans, &ref);
+                       if (ret) {
+                               btrfs_abort_transaction(trans, ret);
+                               break;
+                       }
+                       if (be_nice) {
+                               if (btrfs_should_throttle_delayed_refs(trans))
+                                       should_throttle = true;
+                       }
+               }
+
+               if (found_type == BTRFS_INODE_ITEM_KEY)
+                       break;
+
+               if (path->slots[0] == 0 ||
+                   path->slots[0] != pending_del_slot ||
+                   should_throttle) {
+                       if (pending_del_nr) {
+                               ret = btrfs_del_items(trans, root, path,
+                                               pending_del_slot,
+                                               pending_del_nr);
+                               if (ret) {
+                                       btrfs_abort_transaction(trans, ret);
+                                       break;
+                               }
+                               pending_del_nr = 0;
+                       }
+                       btrfs_release_path(path);
+
+                       /*
+                        * We can generate a lot of delayed refs, so we need to
+                        * throttle every once and a while and make sure we're
+                        * adding enough space to keep up with the work we are
+                        * generating.  Since we hold a transaction here we
+                        * can't flush, and we don't want to FLUSH_LIMIT because
+                        * we could have generated too many delayed refs to
+                        * actually allocate, so just bail if we're short and
+                        * let the normal reservation dance happen higher up.
+                        */
+                       if (should_throttle) {
+                               ret = btrfs_delayed_refs_rsv_refill(fs_info,
+                                                       BTRFS_RESERVE_NO_FLUSH);
+                               if (ret) {
+                                       ret = -EAGAIN;
+                                       break;
+                               }
+                       }
+                       goto search_again;
+               } else {
+                       path->slots[0]--;
+               }
+       }
+out:
+       if (ret >= 0 && pending_del_nr) {
+               int err;
+
+               err = btrfs_del_items(trans, root, path, pending_del_slot,
+                                     pending_del_nr);
+               if (err) {
+                       btrfs_abort_transaction(trans, err);
+                       ret = err;
+               }
+       }
+       if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
+               ASSERT(last_size >= new_size);
+               if (!ret && last_size > new_size)
+                       last_size = new_size;
+               btrfs_inode_safe_disk_i_size_write(inode, last_size);
+       }
+
+       btrfs_free_path(path);
+       return ret;
+}