1 files changed, 62 insertions, 18 deletions

diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index 9e1f2cd5e67a..1d1ba083ca6e 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -94,8 +94,10 @@
 #define LOG_WALK_REPLAY_ALL 3
 
 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
-                             struct btrfs_root *root, struct inode *inode,
-                             int inode_only);
+                           struct btrfs_root *root, struct inode *inode,
+                           int inode_only,
+                           const loff_t start,
+                           const loff_t end);
 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
                              struct btrfs_root *root,
                              struct btrfs_path *path, u64 objectid);
@@ -3298,7 +3300,7 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
         struct list_head ordered_sums;
         int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
         bool has_extents = false;
-        bool need_find_last_extent = (*last_extent == 0);
+        bool need_find_last_extent = true;
         bool done = false;
 
         INIT_LIST_HEAD(&ordered_sums);
@@ -3352,8 +3354,7 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
                  */
                 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
                         has_extents = true;
-                        if (need_find_last_extent &&
-                            first_key.objectid == (u64)-1)
+                        if (first_key.objectid == (u64)-1)
                                 first_key = ins_keys[i];
                 } else {
                         need_find_last_extent = false;
@@ -3427,6 +3428,16 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
         if (!has_extents)
                 return ret;
 
+        if (need_find_last_extent && *last_extent == first_key.offset) {
+                /*
+                 * We don't have any leafs between our current one and the one
+                 * we processed before that can have file extent items for our
+                 * inode (and have a generation number smaller than our current
+                 * transaction id).
+                 */
+                need_find_last_extent = false;
+        }
+
         /*
          * Because we use btrfs_search_forward we could skip leaves that were
          * not modified and then assume *last_extent is valid when it really
@@ -3537,7 +3548,7 @@ fill_holes:
                                                0, 0);
                 if (ret)
                         break;
-                *last_extent = offset + len;
+                *last_extent = extent_end;
         }
         /*
          * Need to let the callers know we dropped the path so they should
@@ -3849,8 +3860,10 @@ process:
  * This handles both files and directories.
  */
 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
-                             struct btrfs_root *root, struct inode *inode,
-                             int inode_only)
+                           struct btrfs_root *root, struct inode *inode,
+                           int inode_only,
+                           const loff_t start,
+                           const loff_t end)
 {
         struct btrfs_path *path;
         struct btrfs_path *dst_path;
@@ -3867,6 +3880,7 @@ static int btrfs_log_inode(struct btrfs_trans_handle *trans,
         int ins_nr;
         bool fast_search = false;
         u64 ino = btrfs_ino(inode);
+        struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 
         path = btrfs_alloc_path();
         if (!path)
@@ -4040,13 +4054,35 @@ log_extents:
                         goto out_unlock;
                 }
         } else if (inode_only == LOG_INODE_ALL) {
-                struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
                 struct extent_map *em, *n;
 
-                write_lock(&tree->lock);
-                list_for_each_entry_safe(em, n, &tree->modified_extents, list)
-                        list_del_init(&em->list);
-                write_unlock(&tree->lock);
+                write_lock(&em_tree->lock);
+                /*
+                 * We can't just remove every em if we're called for a ranged
+                 * fsync - that is, one that doesn't cover the whole possible
+                 * file range (0 to LLONG_MAX). This is because we can have
+                 * em's that fall outside the range we're logging and therefore
+                 * their ordered operations haven't completed yet
+                 * (btrfs_finish_ordered_io() not invoked yet). This means we
+                 * didn't get their respective file extent item in the fs/subvol
+                 * tree yet, and need to let the next fast fsync (one which
+                 * consults the list of modified extent maps) find the em so
+                 * that it logs a matching file extent item and waits for the
+                 * respective ordered operation to complete (if it's still
+                 * running).
+                 *
+                 * Removing every em outside the range we're logging would make
+                 * the next fast fsync not log their matching file extent items,
+                 * therefore making us lose data after a log replay.
+                 */
+                list_for_each_entry_safe(em, n, &em_tree->modified_extents,
+                                         list) {
+                        const u64 mod_end = em->mod_start + em->mod_len - 1;
+
+                        if (em->mod_start >= start && mod_end <= end)
+                                list_del_init(&em->list);
+                }
+                write_unlock(&em_tree->lock);
         }
 
         if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
@@ -4056,6 +4092,7 @@ log_extents:
                         goto out_unlock;
                 }
         }
+
         BTRFS_I(inode)->logged_trans = trans->transid;
         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
 out_unlock:
@@ -4152,7 +4189,10 @@ out:
  */
 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
                                   struct btrfs_root *root, struct inode *inode,
-                                  struct dentry *parent, int exists_only,
+                                  struct dentry *parent,
+                                  const loff_t start,
+                                  const loff_t end,
+                                  int exists_only,
                                   struct btrfs_log_ctx *ctx)
 {
         int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
@@ -4198,7 +4238,7 @@ static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
         if (ret)
                 goto end_no_trans;
 
-        ret = btrfs_log_inode(trans, root, inode, inode_only);
+        ret = btrfs_log_inode(trans, root, inode, inode_only, start, end);
         if (ret)
                 goto end_trans;
 
@@ -4226,7 +4266,8 @@ static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
 
                 if (BTRFS_I(inode)->generation >
                     root->fs_info->last_trans_committed) {
-                        ret = btrfs_log_inode(trans, root, inode, inode_only);
+                        ret = btrfs_log_inode(trans, root, inode, inode_only,
+                                              0, LLONG_MAX);
                         if (ret)
                                 goto end_trans;
                 }
@@ -4260,13 +4301,15 @@ end_no_trans:
  */
 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
                           struct btrfs_root *root, struct dentry *dentry,
+                          const loff_t start,
+                          const loff_t end,
                           struct btrfs_log_ctx *ctx)
 {
         struct dentry *parent = dget_parent(dentry);
         int ret;
 
         ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent,
-                                     0, ctx);
+                                     start, end, 0, ctx);
         dput(parent);
 
         return ret;
@@ -4503,6 +4546,7 @@ int btrfs_log_new_name(struct btrfs_trans_handle *trans,
                     root->fs_info->last_trans_committed))
                 return 0;
 
-        return btrfs_log_inode_parent(trans, root, inode, parent, 1, NULL);
+        return btrfs_log_inode_parent(trans, root, inode, parent, 0,
+                                      LLONG_MAX, 1, NULL);
 }