diff --git a/PKG-INFO b/PKG-INFO
index 16f6d6dc..48c28de8 100644
--- a/PKG-INFO
+++ b/PKG-INFO
@@ -1,10 +1,10 @@
 Metadata-Version: 1.0
 Name: swh.storage
-Version: 0.0.11
+Version: 0.0.12
 Summary: Software Heritage storage manager
 Home-page: https://forge.softwareheritage.org/diffusion/DSTO/
 Author: Software Heritage developers
 Author-email: swh-devel@inria.fr
 License: UNKNOWN
 Description: UNKNOWN
 Platform: UNKNOWN
diff --git a/sql/swh-func.sql b/sql/swh-func.sql
index c7b7af8c..d19fed55 100644
--- a/sql/swh-func.sql
+++ b/sql/swh-func.sql
@@ -1,739 +1,741 @@
 
 -- create a temporary table called tmp_TBLNAME, mimicking existing table
 -- TBLNAME
 --
 -- Args:
 --     tblname: name of the table to mimick
 create or replace function swh_mktemp(tblname regclass)
     returns void
     language plpgsql
 as $$
 begin
     execute format('
 	create temporary table tmp_%I
 	    (like %I including defaults)
 	    on commit drop
 	', tblname, tblname);
     return;
 end
 $$;
 
 
 -- create a temporary table for directory entries called tmp_TBLNAME,
 -- mimicking existing table TBLNAME with an extra dir_id (sha1_git)
 -- column, and dropping the id column.
 --
 -- This is used to create the tmp_directory_entry_<foo> tables.
 --
 -- Args:
 --     tblname: name of the table to mimick
 create or replace function swh_mktemp_dir_entry(tblname regclass)
     returns void
     language plpgsql
 as $$
 begin
     execute format('
 	create temporary table tmp_%I
 	    (like %I including defaults, dir_id sha1_git)
 	    on commit drop;
         alter table tmp_%I drop column id;
 	', tblname, tblname, tblname, tblname);
     return;
 end
 $$;
 
 
 -- create a temporary table for revisions called tmp_revisions,
 -- mimicking existing table revision, replacing the foreign keys to
 -- people with an email and name field
 --
 create or replace function swh_mktemp_revision()
     returns void
     language sql
 as $$
     create temporary table tmp_revision (
         like revision including defaults,
         author_name bytea not null default '',
         author_email bytea not null default '',
         committer_name bytea not null default '',
         committer_email bytea not null default ''
     ) on commit drop;
     alter table tmp_revision drop column author;
     alter table tmp_revision drop column committer;
 $$;
 
 
 -- create a temporary table for releases called tmp_release,
 -- mimicking existing table release, replacing the foreign keys to
 -- people with an email and name field
 --
 create or replace function swh_mktemp_release()
     returns void
     language sql
 as $$
     create temporary table tmp_release (
         like release including defaults,
         author_name bytea not null default '',
         author_email bytea not null default ''
     ) on commit drop;
     alter table tmp_release drop column author;
 $$;
 
 
 -- a content signature is a set of cryptographic checksums that we use to
 -- uniquely identify content, for the purpose of verifying if we already have
 -- some content or not during content injection
 create type content_signature as (
     sha1      sha1,
     sha1_git  sha1_git,
     sha256    sha256
 );
 
 
 -- check which entries of tmp_content are missing from content
 --
 -- operates in bulk: 0. swh_mktemp(content), 1. COPY to tmp_content,
 -- 2. call this function
 create or replace function swh_content_missing()
     returns setof content_signature
     language plpgsql
 as $$
 begin
     -- This query is critical for (single-algorithm) hash collision detection,
     -- so we cannot rely only on the fact that a single hash (e.g., sha1) is
     -- missing from the table content to conclude that a given content is
     -- missing. Ideally, we would want to (try to) add to content all entries
     -- in tmp_content that, when considering all columns together, are missing
     -- from content.
     --
     -- But doing that naively would require a *compound* index on all checksum
     -- columns; that index would not be significantly smaller than the content
     -- table itself, and therefore won't be used. Therefore we union together
     -- all contents that differ on at least one column from what is already
     -- available. If there is a collision on some (but not all) columns, the
     -- relevant tmp_content entry will be included in the set of content to be
     -- added, causing a downstream violation of unicity constraint.
     return query
 	(select sha1, sha1_git, sha256 from tmp_content as tmp
 	 where not exists
 	     (select 1 from content as c where c.sha1 = tmp.sha1))
 	union
 	(select sha1, sha1_git, sha256 from tmp_content as tmp
 	 where not exists
 	     (select 1 from content as c where c.sha1_git = tmp.sha1_git))
 	union
 	(select sha1, sha1_git, sha256 from tmp_content as tmp
 	 where not exists
 	     (select 1 from content as c where c.sha256 = tmp.sha256));
     return;
 end
 $$;
 
 
 -- check which entries of tmp_skipped_content are missing from skipped_content
 --
 -- operates in bulk: 0. swh_mktemp(skipped_content), 1. COPY to tmp_skipped_content,
 -- 2. call this function
 create or replace function swh_skipped_content_missing()
     returns setof content_signature
     language plpgsql
 as $$
 begin
     return query
 	select sha1, sha1_git, sha256 from tmp_skipped_content t
 	where not exists
 	(select 1 from skipped_content s where
 	    s.sha1 is not distinct from t.sha1 and
 	    s.sha1_git is not distinct from t.sha1_git and
 	    s.sha256 is not distinct from t.sha256);
     return;
 end
 $$;
 
 
 -- Look up content based on one or several different checksums. Return all
 -- content information if the content is found; a NULL row otherwise.
 --
 -- At least one checksum should be not NULL. If several are not NULL, they will
 -- be AND-ed together in the lookup query.
 --
 -- Note: this function is meant to be used to look up individual contents
 -- (e.g., for the web app), for batch lookup of missing content (e.g., to be
 -- added) see swh_content_missing
 create or replace function swh_content_find(
     sha1      sha1     default NULL,
     sha1_git  sha1_git default NULL,
     sha256    sha256   default NULL
 )
     returns content
     language plpgsql
 as $$
 declare
     con content;
     filters text[] := array[] :: text[];  -- AND-clauses used to filter content
     q text;
 begin
     if sha1 is not null then
         filters := filters || format('sha1 = %L', sha1);
     end if;
     if sha1_git is not null then
         filters := filters || format('sha1_git = %L', sha1_git);
     end if;
     if sha256 is not null then
         filters := filters || format('sha256 = %L', sha256);
     end if;
 
     if cardinality(filters) = 0 then
         return null;
     else
         q = format('select * from content where %s',
 	        array_to_string(filters, ' and '));
         execute q into con;
 	return con;
     end if;
 end
 $$;
 
 
 -- add tmp_content entries to content, skipping duplicates
 --
 -- operates in bulk: 0. swh_mktemp(content), 1. COPY to tmp_content,
 -- 2. call this function
 create or replace function swh_content_add()
     returns void
     language plpgsql
 as $$
 begin
     insert into content (sha1, sha1_git, sha256, length, status)
 	select distinct sha1, sha1_git, sha256, length, status
 	from tmp_content
 	where (sha1, sha1_git, sha256) in
 	    (select * from swh_content_missing());
 	    -- TODO XXX use postgres 9.5 "UPSERT" support here, when available.
 	    -- Specifically, using "INSERT .. ON CONFLICT IGNORE" we can avoid
 	    -- the extra swh_content_missing() query here.
     return;
 end
 $$;
 
 
 -- add tmp_skipped_content entries to skipped_content, skipping duplicates
 --
 -- operates in bulk: 0. swh_mktemp(skipped_content), 1. COPY to tmp_skipped_content,
 -- 2. call this function
 create or replace function swh_skipped_content_add()
     returns void
     language plpgsql
 as $$
 begin
     insert into skipped_content (sha1, sha1_git, sha256, length, status, reason, origin)
 	select distinct sha1, sha1_git, sha256, length, status, reason, origin
 	from tmp_skipped_content
 	where (coalesce(sha1, ''), coalesce(sha1_git, ''), coalesce(sha256, '')) in
 	    (select coalesce(sha1, ''), coalesce(sha1_git, ''), coalesce(sha256, '') from swh_skipped_content_missing());
 	    -- TODO XXX use postgres 9.5 "UPSERT" support here, when available.
 	    -- Specifically, using "INSERT .. ON CONFLICT IGNORE" we can avoid
 	    -- the extra swh_content_missing() query here.
     return;
 end
 $$;
 
 
 -- check which entries of tmp_directory are missing from directory
 --
 -- operates in bulk: 0. swh_mktemp(directory), 1. COPY to tmp_directory,
 -- 2. call this function
 create or replace function swh_directory_missing()
     returns setof sha1_git
     language plpgsql
 as $$
 begin
     return query
 	select id from tmp_directory t
 	where not exists (
 	    select 1 from directory d
 	    where d.id = t.id);
     return;
 end
 $$;
 
 
 create type directory_entry_type as enum('file', 'dir', 'rev');
 
 
 -- Add tmp_directory_entry_* entries to directory_entry_* and directory,
 -- skipping duplicates in directory_entry_*.  This is a generic function that
 -- works on all kind of directory entries.
 --
 -- operates in bulk: 0. swh_mktemp_dir_entry('directory_entry_*'), 1 COPY to
 -- tmp_directory_entry_*, 2. call this function
 --
 -- Assumption: this function is used in the same transaction that inserts the
 -- context directory in table "directory".
 create or replace function swh_directory_entry_add(typ directory_entry_type)
     returns void
     language plpgsql
 as $$
 begin
     execute format('
     insert into directory_entry_%1$s (target, name, perms)
     select distinct t.target, t.name, t.perms
     from tmp_directory_entry_%1$s t
     where not exists (
     select 1
     from directory_entry_%1$s i
     where t.target = i.target and t.name = i.name and t.perms = i.perms)
    ', typ);
 
     execute format('
     with new_entries as (
 	select t.dir_id, array_agg(i.id) as entries
 	from tmp_directory_entry_%1$s t
 	inner join directory_entry_%1$s i
 	using (target, name, perms)
 	group by t.dir_id
     )
     update directory as d
     set %1$s_entries = new_entries.entries
     from new_entries
     where d.id = new_entries.dir_id
     ', typ);
 
     return;
 end
 $$;
 
 
 -- a directory listing entry with all the metadata
 --
 -- can be used to list a directory, and retrieve all the data in one go.
 create type directory_entry as
 (
   dir_id  sha1_git,     -- id of the parent directory
   type    directory_entry_type,  -- type of entry
   target  sha1_git,     -- id of target
   name    unix_path,    -- path name, relative to containing dir
   perms   file_perms    -- unix-like permissions
 );
 
 
 -- List a single level of directory walked_dir_id
 create or replace function swh_directory_walk_one(walked_dir_id sha1_git)
     returns setof directory_entry
     language sql
+    stable
 as $$
     with dir as (
 	select id as dir_id, dir_entries, file_entries, rev_entries
 	from directory
 	where id = walked_dir_id),
     ls_d as (select dir_id, unnest(dir_entries) as entry_id from dir),
     ls_f as (select dir_id, unnest(file_entries) as entry_id from dir),
     ls_r as (select dir_id, unnest(rev_entries) as entry_id from dir)
     (select dir_id, 'dir'::directory_entry_type as type,
 	    target, name, perms
      from ls_d
      left join directory_entry_dir d on ls_d.entry_id = d.id)
     union
     (select dir_id, 'file'::directory_entry_type as type,
 	    target, name, perms
      from ls_f
      left join directory_entry_file d on ls_f.entry_id = d.id)
     union
     (select dir_id, 'rev'::directory_entry_type as type,
 	    target, name, perms
      from ls_r
      left join directory_entry_rev d on ls_r.entry_id = d.id)
     order by name;
 $$;
 
 
 -- List all revision IDs starting from a given revision, going back in time
 --
 -- TODO ordering: should be breadth-first right now (what do we want?)
 -- TODO ordering: ORDER BY parent_rank somewhere?
 create or replace function swh_revision_list(root_revision sha1_git)
     returns setof sha1_git
     language sql
+    stable
 as $$
     with recursive rev_list(id) as (
 	(select id from revision where id = root_revision)
 	union
 	(select parent_id
 	 from revision_history as h
 	 join rev_list on h.id = rev_list.id)
     )
     select * from rev_list;
 $$;
 
+-- List all the children of a given revision
+create or replace function swh_revision_list_children(root_revision sha1_git)
+    returns setof sha1_git
+    language sql
+    stable
+as $$
+    with recursive rev_list(id) as (
+	(select id from revision where id = root_revision)
+	union
+	(select h.id
+	 from revision_history as h
+	 join rev_list on h.parent_id = rev_list.id)
+    )
+    select * from rev_list;
+$$;
+
 
 -- Detailed entry in a revision log
 create type revision_log_entry as
 (
   id                     sha1_git,
   date                   timestamptz,
   date_offset            smallint,
   committer_date         timestamptz,
   committer_date_offset  smallint,
   type                   revision_type,
   directory              sha1_git,
   message                bytea,
   author_name            bytea,
   author_email           bytea,
   committer_name         bytea,
   committer_email        bytea
 );
 
 
 -- "git style" revision log. Similar to swh_revision_list(), but returning all
 -- information associated to each revision, and expanding authors/committers
 create or replace function swh_revision_log(root_revision sha1_git)
     returns setof revision_log_entry
     language sql
+    stable
 as $$
     select revision.id, date, date_offset,
 	committer_date, committer_date_offset,
 	type, directory, message,
 	author.name as author_name, author.email as author_email,
 	committer.name as committer_name, committer.email as committer_email
     from swh_revision_list(root_revision) as rev_list
     join revision on revision.id = rev_list
     join person as author on revision.author = author.id
     join person as committer on revision.committer = committer.id;
 $$;
 
 -- Detailed entry for a revision
 create type revision_entry as
 (
   id                     sha1_git,
   date                   timestamptz,
   date_offset            smallint,
   committer_date         timestamptz,
   committer_date_offset  smallint,
   type                   revision_type,
   directory              sha1_git,
   message                bytea,
   author_name            bytea,
   author_email           bytea,
   committer_name         bytea,
   committer_email        bytea,
   parents                bytea[]
 );
 
 
 -- Retrieve revisions from tmp_revision in bulk
 create or replace function swh_revision_get()
     returns setof revision_entry
     language plpgsql
 as $$
 begin
     return query
         select t.id, r.date, r.date_offset,
                r.committer_date, r.committer_date_offset,
                r.type, r.directory, r.message,
                a.name, a.email, c.name, c.email,
 	       array_agg(rh.parent_id::bytea order by rh.parent_rank)
                    as parents
         from tmp_revision t
         left join revision r on t.id = r.id
         left join person a on a.id = r.author
         left join person c on c.id = r.committer
         left join revision_history rh on rh.id = r.id
         group by t.id, a.name, a.email, r.date, r.date_offset,
                c.name, c.email, r.committer_date, r.committer_date_offset,
                r.type, r.directory, r.message;
     return;
 end
 $$;
 
 -- List missing revisions from tmp_revision
 create or replace function swh_revision_missing()
     returns setof sha1_git
     language plpgsql
 as $$
 begin
     return query
         select id from tmp_revision t
 	where not exists (
 	    select 1 from revision r
 	    where r.id = t.id);
     return;
 end
 $$;
 
 
 -- Create entries in person from tmp_revision
 create or replace function swh_person_add_from_revision()
     returns void
     language plpgsql
 as $$
 begin
     with t as (
         select author_name as name, author_email as email from tmp_revision
     union
         select committer_name as name, committer_email as email from tmp_revision
     ) insert into person (name, email)
     select distinct name, email from t
     where not exists (
         select 1
 	from person p
 	where t.name = p.name and t.email = p.email
     );
     return;
 end
 $$;
 
 
 -- Create entries in revision from tmp_revision
 create or replace function swh_revision_add()
     returns void
     language plpgsql
 as $$
 begin
     perform swh_person_add_from_revision();
 
     insert into revision (id, date, date_offset, committer_date, committer_date_offset, type, directory, message, author, committer)
     select t.id, t.date, t.date_offset, t.committer_date, t.committer_date_offset, t.type, t.directory, t.message, a.id, c.id
     from tmp_revision t
     left join person a on a.name = t.author_name and a.email = t.author_email
     left join person c on c.name = t.committer_name and c.email = t.committer_email;
     return;
 end
 $$;
 
 
 -- List missing releases from tmp_release
 create or replace function swh_release_missing()
     returns setof sha1_git
     language plpgsql
 as $$
 begin
     return query
         select id from tmp_release t
 	where not exists (
 	select 1 from release r
 	where r.id = t.id);
     return;
 end
 $$;
 
 
 -- Create entries in person from tmp_release
 create or replace function swh_person_add_from_release()
     returns void
     language plpgsql
 as $$
 begin
     with t as (
         select distinct author_name as name, author_email as email from tmp_release
     ) insert into person (name, email)
     select name, email from t
     where not exists (
         select 1
 	from person p
 	where t.name = p.name and t.email = p.email
     );
     return;
 end
 $$;
 
 
 -- Create entries in release from tmp_release
 create or replace function swh_release_add()
     returns void
     language plpgsql
 as $$
 begin
     perform swh_person_add_from_release();
 
     insert into release (id, revision, date, date_offset, name, comment, author)
     select t.id, t.revision, t.date, t.date_offset, t.name, t.comment, a.id
     from tmp_release t
     left join person a on a.name = t.author_name and a.email = t.author_email;
     return;
 end
 $$;
 
 -- add tmp_occurrence_history entries to occurrence_history
 --
 -- operates in bulk: 0. swh_mktemp(occurrence_history), 1. COPY to tmp_occurrence_history,
 -- 2. call this function
 create or replace function swh_occurrence_history_add()
     returns void
     language plpgsql
 as $$
 begin
     -- Update intervals we have the data to update
     with new_intervals as (
         select t.origin, t.branch, t.authority, t.validity,
 	       o.validity - t.validity as new_validity
 	from tmp_occurrence_history t
         left join occurrence_history o
         using (origin, branch, authority)
 	where o.origin is not null),
     -- do not update intervals if they would become empty (perfect overlap)
     to_update as (
         select * from new_intervals
 	where not isempty(new_validity))
     update occurrence_history o set validity = t.new_validity
     from to_update t
     where o.origin = t.origin and o.branch = t.branch and o.authority = t.authority;
 
     -- Now only insert intervals that aren't already present
     insert into occurrence_history (origin, branch, revision, authority, validity)
 	select distinct origin, branch, revision, authority, validity
 	from tmp_occurrence_history t
 	where not exists (
 	    select 1 from occurrence_history o
 	    where o.origin = t.origin and o.branch = t.branch and
 	          o.authority = t.authority and o.revision = t.revision and
 		  o.validity = t.validity);
     return;
 end
 $$;
 
 
 -- Absolute path: directory reference + complete path relative to it
 create type content_dir as (
     directory  sha1_git,
     path       unix_path
 );
 
 
 -- Find the containing directory of a given content, specified by sha1
 -- (note: *not* sha1_git).
 --
 -- Return a pair (dir_it, path) where path is a UNIX path that, from the
 -- directory root, reach down to a file with the desired content. Return NULL
 -- if no match is found.
 --
 -- In case of multiple paths (i.e., pretty much always), an arbitrary one is
 -- chosen.
 create or replace function swh_content_find_directory(content_id sha1)
     returns content_dir
     language sql
+    stable
 as $$
     with recursive path as (
 	-- Recursively build a path from the requested content to a root
 	-- directory. Each iteration returns a pair (dir_id, filename) where
 	-- filename is relative to dir_id. Stops when no parent directory can
 	-- be found.
 	(select dir.id as dir_id, dir_entry_f.name as name, 0 as depth
 	 from directory_entry_file as dir_entry_f
 	 join content on content.sha1_git = dir_entry_f.target
 	 join directory as dir on dir.file_entries @> array[dir_entry_f.id]
 	 where content.sha1 = content_id
 	 limit 1)
 	union all
 	(select dir.id as dir_id,
 		(dir_entry_d.name || '/' || path.name)::unix_path as name,
 		path.depth + 1
 	 from path
 	 join directory_entry_dir as dir_entry_d on dir_entry_d.target = path.dir_id
 	 join directory as dir on dir.dir_entries @> array[dir_entry_d.id]
 	 limit 1)
     )
     select dir_id, name from path order by depth desc limit 1;
 $$;
 
 
 -- Walk the revision history starting from a given revision, until a matching
 -- occurrence is found. Return all occurrence information if one is found, NULL
 -- otherwise.
 create or replace function swh_revision_find_occurrence(revision_id sha1_git)
     returns occurrence
     language plpgsql
 as $$
 declare
     occ occurrence%ROWTYPE;
     rev sha1_git;
 begin
     -- first check to see if revision_id is already pointed by an occurrence
     select origin, branch, revision
     from occurrence_history as occ_hist
     where occ_hist.revision = revision_id
     order by upper(occ_hist.validity)  -- TODO filter by authority?
     limit 1
     into occ;
 
     -- no occurrence point to revision_id, walk up the history
     if not found then
-        -- recursively walk the history, stopping immediately before a revision
-        -- pointed to by an occurrence.
-	-- TODO find a nicer way to stop at, but *including*, that revision
-	with recursive revlog as (
-	    (select revision_id as rev_id, 0 as depth)
-	    union all
-	    (select hist.parent_id as rev_id, revlog.depth + 1
-	     from revlog
-	     join revision_history as hist on hist.id = revlog.rev_id
-	     and not exists(select 1 from occurrence_history
-			    where revision = hist.parent_id)
-	     limit 1)
-	)
-	select rev_id from revlog order by depth desc limit 1
-	into rev;
-	if not found then return null; end if;
-
-	-- as we stopped before a pointed by revision, look it up again and
-	-- return its data
 	select origin, branch, revision
-	from revision_history as rev_hist, occurrence_history as occ_hist
-	where rev_hist.id = rev
-	and occ_hist.revision = rev_hist.parent_id
+	from swh_revision_list_children(revision_id) as rev_list(sha1_git)
+	left join occurrence_history occ_hist
+	on rev_list.sha1_git = occ_hist.revision
+	where occ_hist.origin is not null
 	order by upper(occ_hist.validity)  -- TODO filter by authority?
 	limit 1
 	into occ;
     end if;
 
     return occ;  -- might be NULL
 end
 $$;
 
 
 -- Occurrence of some content in a given context
 create type content_occurrence as (
     origin_type	 text,
     origin_url	 text,
     branch	 text,
     revision_id	 sha1_git,
     path	 unix_path
 );
 
 
 -- Given the sha1 of some content, look up an occurrence that points to a
 -- revision, which in turns reference (transitively) a tree containing the
 -- content. Answer the question: "where/when did SWH see a given content"?
 -- Return information about an arbitrary occurrence/revision/tree if one is
 -- found, NULL otherwise.
 create or replace function swh_content_find_occurrence(content_id sha1)
     returns content_occurrence
     language plpgsql
 as $$
 declare
     dir content_dir;
     rev sha1_git;
     occ occurrence%ROWTYPE;
     coc content_occurrence;
 begin
     -- each step could fail if no results are found, and that's OK
     select * from swh_content_find_directory(content_id)     -- look up directory
 	into dir;
     if not found then return null; end if;
 
     select id from revision where directory = dir.directory  -- look up revision
 	limit 1
 	into rev;
     if not found then return null; end if;
 
     select * from swh_revision_find_occurrence(rev)	     -- look up occurrence
 	into occ;
     if not found then return null; end if;
 
     select origin.type, origin.url, occ.branch, rev, dir.path
     from origin
     where origin.id = occ.origin
     into coc;
 
     return coc;  -- might be NULL
 end
 $$;
diff --git a/sql/swh-schema.sql b/sql/swh-schema.sql
index 059ca4a0..1918d020 100644
--- a/sql/swh-schema.sql
+++ b/sql/swh-schema.sql
@@ -1,324 +1,332 @@
 ---
 --- Software Heritage Data Model
 ---
 
 -- drop schema if exists swh cascade;
 -- create schema swh;
 -- set search_path to swh;
 
 create table dbversion
 (
   version     int primary key,
   release     timestamptz,
   description text
 );
 
 insert into dbversion(version, release, description)
-      values(23, now(), 'Work In Progress');
+      values(24, now(), 'Work In Progress');
 
 -- a SHA1 checksum (not necessarily originating from Git)
 create domain sha1 as bytea check (length(value) = 20);
 
 -- a Git object ID, i.e., a SHA1 checksum
 create domain sha1_git as bytea check (length(value) = 20);
 
 -- a SHA256 checksum
 create domain sha256 as bytea check (length(value) = 32);
 
 -- UNIX path (absolute, relative, individual path component, etc.)
 create domain unix_path as bytea;
 
 -- a set of UNIX-like access permissions, as manipulated by, e.g., chmod
 create domain file_perms as int;
 
 create type content_status as enum ('absent', 'visible', 'hidden');
 
 -- An origin is a place, identified by an URL, where software can be found. We
 -- support different kinds of origins, e.g., git and other VCS repositories,
 -- web pages that list tarballs URLs (e.g., http://www.kernel.org), indirect
 -- tarball URLs (e.g., http://www.example.org/latest.tar.gz), etc. The key
 -- feature of an origin is that it can be *fetched* (wget, git clone, svn
 -- checkout, etc.) to retrieve all the contained software.
 create table origin
 (
   id         bigserial primary key,
   type       text, -- TODO use an enum here (?)
   url        text not null
 );
 
 -- Checksums about actual file content. Note that the content itself is not
 -- stored in the DB, but on external (key-value) storage. A single checksum is
 -- used as key there, but the other can be used to verify that we do not inject
 -- content collisions not knowingly.
 create table content
 (
   sha1      sha1 primary key,
   sha1_git  sha1_git not null,
   sha256    sha256 not null,
   length    bigint not null,
   ctime     timestamptz not null default now(),
             -- creation time, i.e. time of (first) injection into the storage
   status    content_status not null default 'visible'
 );
 
 create unique index on content(sha1_git);
 create unique index on content(sha256);
 
 -- Content we have seen but skipped for some reason. This table is
 -- separate from the content table as we might not have the sha1
 -- checksum of that data (for instance when we inject git
 -- repositories, objects that are too big will be skipped here, and we
 -- will only know their sha1_git). 'reason' contains the reason the
 -- content was skipped. origin is a nullable column allowing to find
 -- out which origin contains that skipped content.
 create table skipped_content
 (
   sha1      sha1,
   sha1_git  sha1_git,
   sha256    sha256,
   length    bigint not null,
   ctime     timestamptz not null default now(),
   status    content_status not null default 'absent',
   reason    text not null,
   origin    bigint references origin(id),
   unique (sha1, sha1_git, sha256)
 );
 
 -- those indexes support multiple NULL values.
 create unique index on skipped_content(sha1);
 create unique index on skipped_content(sha1_git);
 create unique index on skipped_content(sha256);
 
 -- An organization (or part thereof) that might be in charge of running
 -- software projects. Examples: Debian, GNU, GitHub, Apache, The Linux
 -- Foundation. The data model is hierarchical (via parent_id) and might store
 -- sub-branches of existing organizations. The key feature of an organization
 -- is that it can be *listed* to retrieve information about its content, i.e:
 -- sub-organizations, projects, origins.
 create table organization
 (
   id           bigserial primary key,
   parent_id    bigint references organization(id),
   name         text not null,
   description  text,
   homepage     text,
   list_engine  text,  -- crawler to be used to org's content
   list_url     text,  -- root URL to start the listing
   list_params  json,  -- org-specific listing parameter
   latest_list  timestamptz  -- last time the org's content has been listed
 );
 
 -- Log of all organization listings (i.e., organization crawling) that have
 -- been done in the past, or are still ongoing. Similar to fetch_history, but
 -- for organizations.
 create table list_history
 (
   id           bigserial primary key,
   organization bigint references organization(id),
   date         timestamptz not null,
   status       boolean,  -- true if and only if the listing has been successful
   result       json,     -- more detailed return value, depending on status
   stdout       text,
   stderr       text,
   duration     interval  -- fetch duration of NULL if still ongoing
 );
 
 -- Log of all origin fetches (i.e., origin crawling) that have been done in the
 -- past, or are still ongoing. Similar to list_history, but for origins.
 create table fetch_history
 (
   id        bigserial primary key,
   origin    bigint references origin(id),
   date      timestamptz not null,
   status    boolean,  -- true if and only if the fetch has been successful
   result    json,     -- more detailed returned values, times, etc...
   stdout    text,
   stderr    text,     -- null when status is true, filled otherwise
   duration  interval  -- fetch duration of NULL if still ongoing
 );
 
 -- A specific software project, e.g., the Linux kernel, Apache httpd. A
 -- software project is version-less at this level, but is associated to several
 -- metadata. Metadata can evolve over time, this table only contains the most
 -- recent version of them; for old versions of project see table
 -- project_history.
 create table project
 (
   id            bigserial primary key,
   organization  bigint references organization(id),  -- the "owning" organization
   origin        bigint references origin(id),  -- where to find project releases
   name          text,
   description   text,
   homepage      text,
   doap          jsonb
   -- other kinds of metadata/software project description ontologies can be
   -- added here, in addition to DOAP
 );
 
 -- History of project metadata. Time-sensitive version of the table project.
 create table project_history
 (
   id            bigserial primary key,
   project       bigint references project(id),
   validity      tstzrange,
   organization  bigint references organization(id),
   origin        bigint references origin(id),
   name          text,
   description   text,
   homepage      text,
   doap          jsonb
 );
 
 -- A file-system directory.  A directory is a list of directory entries (see
 -- tables: directory_entry_{dir,file}).
 --
 -- To list the contents of a directory:
 -- 1. list the contained directory_entry_dir using array dir_entries
 -- 2. list the contained directory_entry_file using array file_entries
 -- 3. list the contained directory_entry_rev using array rev_entries
 -- 4. UNION
 --
 -- Synonyms/mappings:
 -- * git: tree
 create table directory
 (
   id            sha1_git primary key,
   dir_entries   bigint[],  -- sub-directories, reference directory_entry_dir
   file_entries  bigint[],  -- contained files, reference directory_entry_file
   rev_entries   bigint[]   -- mounted revisions, reference directory_entry_rev
 );
 
 create index on directory using gin (dir_entries);
 create index on directory using gin (file_entries);
 create index on directory using gin (rev_entries);
 
 -- A directory entry pointing to a sub-directory.
 create table directory_entry_dir
 (
   id      bigserial primary key,
   target  sha1_git,   -- id of target directory
   name    unix_path,  -- path name, relative to containing dir
   perms   file_perms  -- unix-like permissions
 );
 
 create unique index on directory_entry_dir(target, name, perms);
 
 -- A directory entry pointing to a file.
 create table directory_entry_file
 (
   id      bigserial primary key,
   target  sha1_git,   -- id of target file
   name    unix_path,  -- path name, relative to containing dir
   perms   file_perms  -- unix-like permissions
 );
 
 create unique index on directory_entry_file(target, name, perms);
 
 -- A directory entry pointing to a revision.
 create table directory_entry_rev
 (
   id      bigserial primary key,
   target  sha1_git,   -- id of target revision
   name    unix_path,  -- path name, relative to containing dir
   perms   file_perms  -- unix-like permissions
 );
 
 create unique index on directory_entry_rev(target, name, perms);
 
 create table person
 (
   id     bigserial primary key,
   name   bytea not null default '',
   email  bytea not null default ''
 );
 
 create unique index on person(name, email);
 
 create type revision_type as enum ('git', 'tar', 'dsc');
 
 -- A snapshot of a software project at a specific point in time.
 --
 -- Synonyms/mappings:
 -- * git / subversion / etc: commit
 -- * tarball: a specific tarball
 --
 -- Revisions are organized as DAGs. Each revision points to 0, 1, or more (in
 -- case of merges) parent revisions. Each revision points to a directory, i.e.,
 -- a file-system tree containing files and directories.
 create table revision
 (
   id                    sha1_git primary key,
   date                  timestamptz,
   date_offset           smallint,
   committer_date        timestamptz,
   committer_date_offset smallint,
   type                  revision_type not null,
   directory             sha1_git,  -- file-system tree
   message               bytea,
   author                bigint references person(id),
   committer             bigint references person(id)
 );
 
+create index on revision(directory);
+
 -- either this table or the sha1_git[] column on the revision table
 create table revision_history
 (
   id           sha1_git references revision(id),
   parent_id    sha1_git,
   parent_rank  int not null default 0,
     -- parent position in merge commits, 0-based
   primary key (id, parent_rank)
 );
 
+create index on revision_history(parent_id);
+
 -- The content of software origins is indexed starting from top-level pointers
 -- called "branches". Every time we fetch some origin we store in this table
 -- where the branches pointed to at fetch time.
 --
 -- Synonyms/mappings:
 -- * git: ref (in the "git update-ref" sense)
 create table occurrence_history
 (
   origin     bigint references origin(id),
   branch     text,  -- e.g., "master" (for VCS), or "sid" (for Debian)
   revision   sha1_git,  -- ref target, e.g., commit id
   authority  bigint references organization(id) not null,
                       -- who is claiming to have seen the occurrence.
                       -- Note: SWH is such an authority, and has an entry in
                       -- the organization table.
   validity   tstzrange,  -- The time validity of this table entry. If the upper
                          -- bound is missing, the entry is still valid.
   exclude using gist (origin with =,
                       branch with =,
                       revision with =,
                       authority with =,
                       validity with &&),
   -- unicity exclusion constraint on lines where the same value is found for
   -- `origin`, `reference`, `revision`, `authority` and overlapping values for
   -- `validity`.
   primary key (origin, branch, revision, authority, validity)
 );
 
+create index on occurrence_history(revision);
+
 -- Materialized view of occurrence_history, storing the *current* value of each
 -- branch, as last seen by SWH.
 create table occurrence
 (
   origin    bigint references origin(id),
   branch    text,
   revision  sha1_git,
   primary key(origin, branch, revision)
 );
 
 -- A "memorable" point in the development history of a project.
 --
 -- Synonyms/mappings:
 -- * git: tag (of the annotated kind, otherwise they are just references)
 -- * tarball: the release version number
 create table release
 (
   id          sha1_git primary key,
   revision    sha1_git,
   date        timestamptz,
   date_offset smallint,
   name        text,
   comment     bytea,
   author      bigint references person(id)
 );
+
+create index on release(revision);
diff --git a/swh.storage.egg-info/PKG-INFO b/swh.storage.egg-info/PKG-INFO
index 16f6d6dc..48c28de8 100644
--- a/swh.storage.egg-info/PKG-INFO
+++ b/swh.storage.egg-info/PKG-INFO
@@ -1,10 +1,10 @@
 Metadata-Version: 1.0
 Name: swh.storage
-Version: 0.0.11
+Version: 0.0.12
 Summary: Software Heritage storage manager
 Home-page: https://forge.softwareheritage.org/diffusion/DSTO/
 Author: Software Heritage developers
 Author-email: swh-devel@inria.fr
 License: UNKNOWN
 Description: UNKNOWN
 Platform: UNKNOWN
diff --git a/swh/storage/objstorage.py b/swh/storage/objstorage.py
index bf03ed70..0094a3da 100644
--- a/swh/storage/objstorage.py
+++ b/swh/storage/objstorage.py
@@ -1,391 +1,392 @@
 # Copyright (C) 2015  The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
 import gzip
 import os
 import shutil
 import tempfile
 
 from contextlib import contextmanager
 
 from swh.core import hashutil
 
 
 ID_HASH_ALGO = 'sha1'
 # ID_HASH_ALGO = 'sha1_git'
 
 GZIP_BUFSIZ = 1048576
 
 
 class Error(Exception):
 
     def __str__(self):
         return 'storage error on object: %s' % self.args
 
 
 class ObjNotFoundError(Error):
 
     def __str__(self):
         return 'object not found: %s' % self.args
 
 
 def _obj_dir(hex_obj_id, root_dir, depth):
     """compute the storage directory of an object
 
     Args:
         hex_obj_id: object id as hexlified string
         root_dir: object storage root directory
         depth: slicing depth of object IDs in the storage
 
     see also: `_obj_path`
 
     """
     if len(hex_obj_id) < depth * 2:
         raise ValueError('object id "%s" is too short for slicing at depth %d'
                          % (hex_obj_id, depth))
 
     # compute [depth] substrings of [obj_id], each of length 2, starting from
     # the beginning
     id_steps = [hex_obj_id[i*2:i*2+2] for i in range(0, depth)]
     steps = [root_dir] + id_steps
 
     return os.path.join(*steps)
 
 
 def _obj_path(hex_obj_id, root_dir, depth):
     """similar to `obj_dir`, but also include the actual object file name in the
     returned path
 
     """
     return os.path.join(_obj_dir(hex_obj_id, root_dir, depth), hex_obj_id)
 
 
 @contextmanager
 def _write_obj_file(hex_obj_id, root_dir, depth):
     """context manager for writing object files to the object storage
 
     During writing data are written to a temporary file, which is atomically
     renamed to the right file name after closing. This context manager also
     takes care of (gzip) compressing the data on the fly.
 
     Yields:
         a file-like object open for writing bytes
 
     Sample usage:
 
     with _write_obj_file(hex_obj_id, root_dir, depth) as f:
         f.write(obj_data)
 
     """
     dir = _obj_dir(hex_obj_id, root_dir, depth)
     if not os.path.isdir(dir):
         os.makedirs(dir)
 
     path = os.path.join(dir, hex_obj_id)
     (tmp, tmp_path) = tempfile.mkstemp(suffix='.tmp', prefix='hex_obj_id.',
                                        dir=dir)
     tmp_f = os.fdopen(tmp, 'wb')
     with gzip.GzipFile(filename=tmp_path, fileobj=tmp_f) as f:
         yield f
     tmp_f.close()
+    os.chmod(tmp_path, 0o644)
     os.rename(tmp_path, path)
 
 
 class ObjStorage:
     """high-level API to manipulate the Software Heritage object storage
 
     Conceptually, the object storage offers 4 methods:
 
     - add()           add a new object, returning an object id
     - __contains__()  check if an object is present, by object id
     - get()           retrieve the content of an object, by object id
     - check()         check the integrity of an object, by object id
 
     Variants of the above methods are implemented by this class, depending on
     how the content of an object is specified (bytes, file-like object, etc.).
 
     On disk, an object storage is a directory tree containing files named after
     their object IDs. An object ID is a checksum of its content, depending on
     the value of the ID_HASH_ALGO constant (see hashutil for its meaning).
 
     To avoid directories that contain too many files, the object storage has a
     given depth (default: 3). Each depth level consumes two characters of the
     object id. So for instance a file with (git) SHA1 of
     34973274ccef6ab4dfaaf86599792fa9c3fe4689 will be stored in an object
     storage configured at depth 3 at
     34/97/32/34973274ccef6ab4dfaaf86599792fa9c3fe4689.
 
     The actual files in the storage are stored in gzipped compressed format.
 
     Each file can hence be self-verified (on the shell) with something like:
 
     actual_id=34973274ccef6ab4dfaaf86599792fa9c3fe4689
     expected_id=$(zcat $filename | sha1sum | cut -f 1 -d' ')
     if [ $actual_id != $expected_id ] ; then
        echo "AYEEE, invalid object $actual_id /o\"
     fi
 
     """
 
     def __init__(self, root, depth=3):
         """create a proxy object to the object storage
 
         Args:
             root: object storage root directory
             depth: slicing depth of object IDs in the storage
 
         """
         if not os.path.isdir(root):
             raise ValueError('obj storage root "%s" is not a directory'
                              % root)
 
         self._root_dir = root
         self._depth = depth
 
         self._temp_dir = os.path.join(root, 'tmp')
         if not os.path.isdir(self._temp_dir):
             os.makedirs(self._temp_dir)
 
     def __obj_dir(self, hex_obj_id):
         """_obj_dir wrapper using this storage configuration"""
         return _obj_dir(hex_obj_id, self._root_dir, self._depth)
 
     def __obj_path(self, hex_obj_id):
         """_obj_path wrapper using this storage configuration"""
         return _obj_path(hex_obj_id, self._root_dir, self._depth)
 
     def __contains__(self, obj_id):
         """check whether a given object id is present in the storage or not
 
         Return:
             True iff the object id is present in the storage
 
         """
         hex_obj_id = hashutil.hash_to_hex(obj_id)
 
         return os.path.exists(_obj_path(hex_obj_id, self._root_dir,
                                         self._depth))
 
     def add_bytes(self, bytes, obj_id=None):
         """add a new object to the object storage
 
         Args:
             bytes: content of the object to be added to the storage
             obj_id: checksums of `bytes` as computed by ID_HASH_ALGO. When
                 given, obj_id will be trusted to match bytes. If missing,
                 obj_id will be computed on the fly.
 
         """
         if obj_id is None:
             # missing checksum, compute it in memory and write to file
             h = hashutil._new_hash(ID_HASH_ALGO, len(bytes))
             h.update(bytes)
             obj_id = h.digest()
 
         if obj_id in self:
             return obj_id
 
         hex_obj_id = hashutil.hash_to_hex(obj_id)
 
         # object is either absent, or present but overwrite is requested
         with _write_obj_file(hex_obj_id,
                              root_dir=self._root_dir,
                              depth=self._depth) as f:
             f.write(bytes)
 
         return obj_id
 
     def add_file(self, f, length, obj_id=None):
         """similar to `add_bytes`, but add the content of file-like object f to the
         object storage
 
         add_file will read the file content only once, and avoid storing all of
         it in memory
 
         """
         if obj_id is None:
             # unknkown object id: work on temp file, compute checksum as we go,
             # mv temp file into place
             (tmp, tmp_path) = tempfile.mkstemp(dir=self._temp_dir)
             try:
                 t = os.fdopen(tmp, 'wb')
                 tz = gzip.GzipFile(fileobj=t)
                 sums = hashutil._hash_file_obj(f, length,
                                                algorithms=[ID_HASH_ALGO],
                                                chunk_cb=lambda b: tz.write(b))
                 tz.close()
                 t.close()
 
                 obj_id = sums[ID_HASH_ALGO]
                 if obj_id in self:
                     return obj_id
 
                 hex_obj_id = hashutil.hash_to_hex(obj_id)
 
                 dir = self.__obj_dir(hex_obj_id)
                 if not os.path.isdir(dir):
                     os.makedirs(dir)
                 path = os.path.join(dir, hex_obj_id)
                 os.rename(tmp_path, path)
             finally:
                 if os.path.exists(tmp_path):
                     os.unlink(tmp_path)
         else:
             # known object id: write to .new file, rename
             if obj_id in self:
                 return obj_id
 
             hex_obj_id = hashutil.hash_to_hex(obj_id)
 
             with _write_obj_file(hex_obj_id,
                                  root_dir=self._root_dir,
                                  depth=self._depth) as obj:
                 shutil.copyfileobj(f, obj)
 
         return obj_id
 
     @contextmanager
     def get_file_obj(self, obj_id):
         """context manager to read the content of an object
 
         Args:
             obj_id: object id
 
         Yields:
             a file-like object open for reading (bytes)
 
         Raises:
             ObjNotFoundError: if the requested object is missing
 
         Sample usage:
 
            with objstorage.get_file_obj(obj_id) as f:
                do_something(f.read())
 
         """
         if obj_id not in self:
             raise ObjNotFoundError(obj_id)
 
         hex_obj_id = hashutil.hash_to_hex(obj_id)
 
         path = self.__obj_path(hex_obj_id)
         with gzip.GzipFile(path, 'rb') as f:
             yield f
 
     def get_bytes(self, obj_id):
         """retrieve the content of a given object
 
         Args:
             obj_id: object id
 
         Returns:
             the content of the requested objects as bytes
 
         Raises:
             ObjNotFoundError: if the requested object is missing
 
         """
         with self.get_file_obj(obj_id) as f:
             return f.read()
 
     def _get_file_path(self, obj_id):
         """retrieve the path of a given object in the objects storage
 
         Note that the path point to a gzip-compressed file, so you need
         gzip.open() or equivalent to get the actual object content.
 
         Args:
             obj_id: object id
 
         Returns:
             a file path pointing into the object storage
 
         Raises:
             ObjNotFoundError: if the requested object is missing
 
         """
         if obj_id not in self:
             raise ObjNotFoundError(obj_id)
 
         hex_obj_id = hashutil.hash_to_hex(obj_id)
 
         return self.__obj_path(hex_obj_id)
 
     def check(self, obj_id):
         """integrity check for a given object
 
         verify that the file object is in place, and that the gzipped content
         matches the object id
 
         Args:
             obj_id: object id
 
         Raises:
             ObjNotFoundError: if the requested object is missing
             Error: if the requested object is corrupt
 
         """
         if obj_id not in self:
             raise ObjNotFoundError(obj_id)
 
         hex_obj_id = hashutil.hash_to_hex(obj_id)
 
         try:
             with gzip.open(self.__obj_path(hex_obj_id)) as f:
                 length = None
                 if ID_HASH_ALGO.endswith('_git'):
                     # if the hashing algorithm is git-like, we need to know the
                     # content size to hash on the fly. Do a first pass here to
                     # compute the size
                     length = 0
                     while True:
                         chunk = f.read(GZIP_BUFSIZ)
                         length += len(chunk)
                         if not chunk:
                             break
                     f.rewind()
 
                 checksums = hashutil._hash_file_obj(f, length,
                                                     algorithms=[ID_HASH_ALGO])
                 actual_obj_id = checksums[ID_HASH_ALGO]
                 if obj_id != actual_obj_id:
                     raise Error('corrupt object %s should have id %s' %
                                 (obj_id, actual_obj_id))
         except (OSError, IOError):
             # IOError is for compatibility with older python versions
             raise Error('corrupt object %s is not a gzip file' % obj_id)
 
     def __iter__(self):
         """iterate over the object identifiers currently available in the storage
 
         Warning: with the current implementation of the object storage, this
         method will walk the filesystem to list objects, meaning that listing
         all objects will be very slow for large storages. You almost certainly
         don't want to use this method in production.
 
         Return:
             iterator over object IDs
 
         """
         def obj_iterator():
             # XXX hackish: it does not verify that the depth of found files
             # matches the slicing depth of the storage
             for root, _dirs, files in os.walk(self._root_dir):
                 for f in files:
                     yield bytes.fromhex(f)
 
         return obj_iterator()
 
     def __len__(self):
         """compute the number of objects available in the storage
 
         Warning: this currently uses `__iter__`, its warning about bad
         performances applies
 
         Return:
             number of objects contained in the storage
 
         """
         return sum(1 for i in self)
diff --git a/swh/storage/tests/test_objstorage.py b/swh/storage/tests/test_objstorage.py
index 7c7905c2..67c5152e 100644
--- a/swh/storage/tests/test_objstorage.py
+++ b/swh/storage/tests/test_objstorage.py
@@ -1,139 +1,146 @@
 # Copyright (C) 2015  The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
 import gzip
 import os
 import shutil
+import stat
 import tempfile
 import unittest
 
 from io import BytesIO
 from nose.tools import istest
 
 from swh.core import hashutil
 from swh.storage import objstorage
 
 
 class TestObjStorage(unittest.TestCase):
 
     def setUp(self):
         self.content = b'42\n'
 
         # sha1
         self.hex_obj_id = '34973274ccef6ab4dfaaf86599792fa9c3fe4689'
 
         # sha1_git
         # self.hex_obj_id = 'd81cc0710eb6cf9efd5b920a8453e1e07157b6cd'
 
         self.obj_id = hashutil.hex_to_hash(self.hex_obj_id)
         self.obj_steps = [self.hex_obj_id[0:2], self.hex_obj_id[2:4],
                           self.hex_obj_id[4:6]]
         self.obj_relpath = os.path.join(*(self.obj_steps + [self.hex_obj_id]))
 
         self.tmpdir = tempfile.mkdtemp()
         self.obj_path = os.path.join(self.tmpdir, self.obj_relpath)
 
         self.storage = objstorage.ObjStorage(root=self.tmpdir, depth=3)
 
         self.missing_obj_id = hashutil.hex_to_hash(
             'f1d2d2f924e986ac86fdf7b36c94bcdf32beec15')
 
     def tearDown(self):
         shutil.rmtree(self.tmpdir)
 
     def assertGzipContains(self, gzip_path, content):  # noqa
         self.assertEqual(gzip.open(gzip_path, 'rb').read(), content)
 
     @istest
     def add_bytes_w_id(self):
         r = self.storage.add_bytes(self.content, obj_id=self.obj_id)
         self.assertEqual(r, self.obj_id)
         self.assertGzipContains(self.obj_path, self.content)
 
     @istest
     def add_bytes_wo_id(self):
         r = self.storage.add_bytes(self.content)
         self.assertEqual(r, self.obj_id)
         self.assertGzipContains(self.obj_path, self.content)
 
     @istest
     def add_file_w_id(self):
         r = self.storage.add_file(BytesIO(self.content),
                                   len(self.content),
                                   obj_id=self.obj_id)
         self.assertEqual(r, self.obj_id)
         self.assertGzipContains(self.obj_path, self.content)
 
     @istest
     def add_file_wo_id(self):
         r = self.storage.add_file(BytesIO(self.content), len(self.content))
         self.assertEqual(r, self.obj_id)
         self.assertGzipContains(self.obj_path, self.content)
 
     @istest
     def contains(self):
         self.storage.add_bytes(self.content, obj_id=self.obj_id)
         self.assertIn(self.obj_id, self.storage)
         self.assertNotIn(self.missing_obj_id, self.storage)
 
     @istest
     def check_ok(self):
         self.storage.add_bytes(self.content, obj_id=self.obj_id)
         try:
             self.storage.check(self.obj_id)
         except:
             self.fail('integrity check failed')
 
     @istest
     def check_missing(self):
         with self.assertRaises(objstorage.Error):
             self.storage.check(self.obj_id)
 
+    @istest
+    def check_file_mode(self):
+        self.storage.add_bytes(self.content, obj_id=self.obj_id)
+        stat_res = os.stat(self.obj_path)
+        self.assertEquals(stat.S_IMODE(stat_res.st_mode), 0o644)
+
     @istest
     def check_not_gzip(self):
         self.storage.add_bytes(self.content, obj_id=self.obj_id)
         with open(self.obj_path, 'ab') as f:  # add trailing garbage
             f.write(b'garbage')
         with self.assertRaises(objstorage.Error):
             self.storage.check(self.obj_id)
 
     @istest
     def check_id_mismatch(self):
         self.storage.add_bytes(self.content, obj_id=self.obj_id)
         with gzip.open(self.obj_path, 'wb') as f:  # replace gzipped content
             f.write(b'unexpected content')
         with self.assertRaises(objstorage.Error):
             self.storage.check(self.obj_id)
 
     @istest
     def get_bytes(self):
         self.storage.add_bytes(self.content, obj_id=self.obj_id)
         self.assertEqual(self.storage.get_bytes(self.obj_id),
                          self.content)
 
     @istest
     def get_file_path(self):
         self.storage.add_bytes(self.content, obj_id=self.obj_id)
         path = self.storage._get_file_path(self.obj_id)
         self.assertEqual(os.path.basename(path), self.hex_obj_id)
         self.assertEqual(gzip.open(path, 'rb').read(), self.content)
 
     @istest
     def get_missing(self):
         with self.assertRaises(objstorage.Error):
             with self.storage.get_file_obj(self.missing_obj_id) as f:
                 f.read()
 
     @istest
     def iter(self):
         self.assertEqual(list(iter(self.storage)), [])
         self.storage.add_bytes(self.content, obj_id=self.obj_id)
         self.assertEqual(list(iter(self.storage)), [self.obj_id])
 
     @istest
     def len(self):
         self.assertEqual(len(self.storage), 0)
         self.storage.add_bytes(self.content, obj_id=self.obj_id)
         self.assertEqual(len(self.storage), 1)
diff --git a/version.txt b/version.txt
index bfcc8205..902d4634 100644
--- a/version.txt
+++ b/version.txt
@@ -1 +1 @@
-v0.0.11-0-gcefe4d0
\ No newline at end of file
+v0.0.12-0-g672dd1c
\ No newline at end of file