Merge branch '13146-shared-rails' refs #13146
[arvados.git] / services / fuse / arvados_fuse / fusedir.py
index 76530844b5724867f1f7c0895c765de63e503460..2d58012fa805b506215d3ce5f1ee4e6699efb3b1 100644 (file)
@@ -1,3 +1,7 @@
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
 import logging
 import re
 import time
@@ -146,12 +150,12 @@ class Directory(FreshBase):
         # delete any other directory entries that were not in found in 'items'
         for i in oldentries:
             _logger.debug("Forgetting about entry '%s' on inode %i", i, self.inode)
-            self.inodes.invalidate_entry(self.inode, i.encode(self.inodes.encoding))
+            self.inodes.invalidate_entry(self, i)
             self.inodes.del_entry(oldentries[i])
             changed = True
 
         if changed:
-            self.inodes.invalidate_inode(self.inode)
+            self.inodes.invalidate_inode(self)
             self._mtime = time.time()
 
         self.fresh()
@@ -178,16 +182,21 @@ class Directory(FreshBase):
         self._entries = {}
         for n in oldentries:
             oldentries[n].clear()
-            self.inodes.invalidate_entry(self.inode, n.encode(self.inodes.encoding))
             self.inodes.del_entry(oldentries[n])
-        self.inodes.invalidate_inode(self.inode)
         self.invalidate()
 
     def kernel_invalidate(self):
-        for n, e in self._entries.iteritems():
-            self.inodes.invalidate_entry(self.inode, n.encode(self.inodes.encoding))
-            e.kernel_invalidate()
-        self.inodes.invalidate_inode(self.inode)
+        # Invalidating the dentry on the parent implies invalidating all paths
+        # below it as well.
+        parent = self.inodes[self.parent_inode]
+
+        # Find self on the parent in order to invalidate this path.
+        # Calling the public items() method might trigger a refresh,
+        # which we definitely don't want, so read the internal dict directly.
+        for k,v in parent._entries.items():
+            if v is self:
+                self.inodes.invalidate_entry(parent, k)
+                break
 
     def mtime(self):
         return self._mtime
@@ -262,13 +271,13 @@ class CollectionDirectoryBase(Directory):
                 elif event == arvados.collection.DEL:
                     ent = self._entries[name]
                     del self._entries[name]
-                    self.inodes.invalidate_entry(self.inode, name.encode(self.inodes.encoding))
+                    self.inodes.invalidate_entry(self, name)
                     self.inodes.del_entry(ent)
                 elif event == arvados.collection.MOD:
                     if hasattr(item, "fuse_entry") and item.fuse_entry is not None:
-                        self.inodes.invalidate_inode(item.fuse_entry.inode)
+                        self.inodes.invalidate_inode(item.fuse_entry)
                     elif name in self._entries:
-                        self.inodes.invalidate_inode(self._entries[name].inode)
+                        self.inodes.invalidate_inode(self._entries[name])
 
     def populate(self, mtime):
         self._mtime = mtime
@@ -507,6 +516,8 @@ class CollectionDirectory(CollectionDirectoryBase):
             self.collection.stop_threads()
 
     def clear(self):
+        if self.collection is not None:
+            self.collection.stop_threads()
         super(CollectionDirectory, self).clear()
         self._manifest_size = 0
 
@@ -541,7 +552,7 @@ class TmpCollectionDirectory(CollectionDirectoryBase):
         if self.collection_record_file:
             with llfuse.lock:
                 self.collection_record_file.invalidate()
-            self.inodes.invalidate_inode(self.collection_record_file.inode)
+            self.inodes.invalidate_inode(self.collection_record_file)
             _logger.debug("%s invalidated collection record", self)
 
     def collection_record(self):
@@ -598,12 +609,13 @@ class MagicDirectory(Directory):
     README_TEXT = """
 This directory provides access to Arvados collections as subdirectories listed
 by uuid (in the form 'zzzzz-4zz18-1234567890abcde') or portable data hash (in
-the form '1234567890abcdef0123456789abcdef+123').
+the form '1234567890abcdef0123456789abcdef+123'), and Arvados projects by uuid
+(in the form 'zzzzz-j7d0g-1234567890abcde').
 
 Note that this directory will appear empty until you attempt to access a
-specific collection subdirectory (such as trying to 'cd' into it), at which
-point the collection will actually be looked up on the server and the directory
-will appear if it exists.
+specific collection or project subdirectory (such as trying to 'cd' into it),
+at which point the collection or project will actually be looked up on the server
+and the directory will appear if it exists.
 
 """.lstrip()
 
@@ -633,8 +645,18 @@ will appear if it exists.
             return False
 
         try:
-            e = self.inodes.add_entry(CollectionDirectory(
-                    self.inode, self.inodes, self.api, self.num_retries, k))
+            e = None
+
+            if group_uuid_pattern.match(k):
+                project = self.api.groups().list(
+                    filters=[['group_class', '=', 'project'], ["uuid", "=", k]]).execute(num_retries=self.num_retries)
+                if project[u'items_available'] == 0:
+                    return False
+                e = self.inodes.add_entry(ProjectDirectory(
+                    self.inode, self.inodes, self.api, self.num_retries, project[u'items'][0]))
+            else:
+                e = self.inodes.add_entry(CollectionDirectory(
+                        self.inode, self.inodes, self.api, self.num_retries, k))
 
             if e.update():
                 if k not in self._entries:
@@ -643,12 +665,13 @@ will appear if it exists.
                     self.inodes.del_entry(e)
                 return True
             else:
-                self.inodes.invalidate_entry(self.inode, k)
+                self.inodes.invalidate_entry(self, k)
                 self.inodes.del_entry(e)
                 return False
         except Exception as ex:
-            _logger.debug('arv-mount exception keep %s', ex)
-            self.inodes.del_entry(e)
+            _logger.exception("arv-mount lookup '%s':", k)
+            if e is not None:
+                self.inodes.del_entry(e)
             return False
 
     def __getitem__(self, item):
@@ -673,6 +696,7 @@ class TagsDirectory(Directory):
         self.num_retries = num_retries
         self._poll = True
         self._poll_time = poll_time
+        self._extra = set()
 
     def want_event_subscribe(self):
         return True
@@ -681,15 +705,41 @@ class TagsDirectory(Directory):
     def update(self):
         with llfuse.lock_released:
             tags = self.api.links().list(
-                filters=[['link_class', '=', 'tag']],
-                select=['name'], distinct=True
+                filters=[['link_class', '=', 'tag'], ["name", "!=", ""]],
+                select=['name'], distinct=True, limit=1000
                 ).execute(num_retries=self.num_retries)
         if "items" in tags:
-            self.merge(tags['items'],
+            self.merge(tags['items']+[{"name": n} for n in self._extra],
                        lambda i: i['name'],
                        lambda a, i: a.tag == i['name'],
                        lambda i: TagDirectory(self.inode, self.inodes, self.api, self.num_retries, i['name'], poll=self._poll, poll_time=self._poll_time))
 
+    @use_counter
+    @check_update
+    def __getitem__(self, item):
+        if super(TagsDirectory, self).__contains__(item):
+            return super(TagsDirectory, self).__getitem__(item)
+        with llfuse.lock_released:
+            tags = self.api.links().list(
+                filters=[['link_class', '=', 'tag'], ['name', '=', item]], limit=1
+            ).execute(num_retries=self.num_retries)
+        if tags["items"]:
+            self._extra.add(item)
+            self.update()
+        return super(TagsDirectory, self).__getitem__(item)
+
+    @use_counter
+    @check_update
+    def __contains__(self, k):
+        if super(TagsDirectory, self).__contains__(k):
+            return True
+        try:
+            self[k]
+            return True
+        except KeyError:
+            pass
+        return False
+
 
 class TagDirectory(Directory):
     """A special directory that contains as subdirectories all collections visible
@@ -738,6 +788,7 @@ class ProjectDirectory(Directory):
         self._poll_time = poll_time
         self._updating_lock = threading.Lock()
         self._current_user = None
+        self._full_listing = False
 
     def want_event_subscribe(self):
         return True
@@ -760,27 +811,35 @@ class ProjectDirectory(Directory):
     def uuid(self):
         return self.project_uuid
 
+    def items(self):
+        self._full_listing = True
+        return super(ProjectDirectory, self).items()
+
+    def namefn(self, i):
+        if 'name' in i:
+            if i['name'] is None or len(i['name']) == 0:
+                return None
+            elif "uuid" in i and (collection_uuid_pattern.match(i['uuid']) or group_uuid_pattern.match(i['uuid'])):
+                # collection or subproject
+                return i['name']
+            elif link_uuid_pattern.match(i['uuid']) and i['head_kind'] == 'arvados#collection':
+                # name link
+                return i['name']
+            elif 'kind' in i and i['kind'].startswith('arvados#'):
+                # something else
+                return "{}.{}".format(i['name'], i['kind'][8:])
+        else:
+            return None
+
+
     @use_counter
     def update(self):
         if self.project_object_file == None:
             self.project_object_file = ObjectFile(self.inode, self.project_object)
             self.inodes.add_entry(self.project_object_file)
 
-        def namefn(i):
-            if 'name' in i:
-                if i['name'] is None or len(i['name']) == 0:
-                    return None
-                elif collection_uuid_pattern.match(i['uuid']) or group_uuid_pattern.match(i['uuid']):
-                    # collection or subproject
-                    return i['name']
-                elif link_uuid_pattern.match(i['uuid']) and i['head_kind'] == 'arvados#collection':
-                    # name link
-                    return i['name']
-                elif 'kind' in i and i['kind'].startswith('arvados#'):
-                    # something else
-                    return "{}.{}".format(i['name'], i['kind'][8:])
-            else:
-                return None
+        if not self._full_listing:
+            return True
 
         def samefn(a, i):
             if isinstance(a, CollectionDirectory) or isinstance(a, ProjectDirectory):
@@ -802,31 +861,63 @@ class ProjectDirectory(Directory):
                     self.project_object = self.api.users().get(
                         uuid=self.project_uuid).execute(num_retries=self.num_retries)
 
-                contents = arvados.util.list_all(self.api.groups().contents,
-                                                 self.num_retries, uuid=self.project_uuid)
+                contents = arvados.util.list_all(self.api.groups().list,
+                                                 self.num_retries,
+                                                 filters=[["owner_uuid", "=", self.project_uuid],
+                                                          ["group_class", "=", "project"]])
+                contents.extend(arvados.util.list_all(self.api.collections().list,
+                                                      self.num_retries,
+                                                      filters=[["owner_uuid", "=", self.project_uuid]]))
 
             # end with llfuse.lock_released, re-acquire lock
 
             self.merge(contents,
-                       namefn,
+                       self.namefn,
                        samefn,
                        self.createDirectory)
+            return True
         finally:
             self._updating_lock.release()
 
+    def _add_entry(self, i, name):
+        ent = self.createDirectory(i)
+        self._entries[name] = self.inodes.add_entry(ent)
+        return self._entries[name]
+
     @use_counter
     @check_update
-    def __getitem__(self, item):
-        if item == '.arvados#project':
+    def __getitem__(self, k):
+        if k == '.arvados#project':
             return self.project_object_file
-        else:
-            return super(ProjectDirectory, self).__getitem__(item)
+        elif self._full_listing or super(ProjectDirectory, self).__contains__(k):
+            return super(ProjectDirectory, self).__getitem__(k)
+        with llfuse.lock_released:
+            contents = self.api.groups().list(filters=[["owner_uuid", "=", self.project_uuid],
+                                                       ["group_class", "=", "project"],
+                                                       ["name", "=", k]],
+                                              limit=1).execute(num_retries=self.num_retries)["items"]
+            if not contents:
+                contents = self.api.collections().list(filters=[["owner_uuid", "=", self.project_uuid],
+                                                                ["name", "=", k]],
+                                                       limit=1).execute(num_retries=self.num_retries)["items"]
+        if contents:
+            name = sanitize_filename(self.namefn(contents[0]))
+            if name != k:
+                raise KeyError(k)
+            return self._add_entry(contents[0], name)
+
+        # Didn't find item
+        raise KeyError(k)
 
     def __contains__(self, k):
         if k == '.arvados#project':
             return True
-        else:
-            return super(ProjectDirectory, self).__contains__(k)
+        try:
+            self[k]
+            return True
+        except KeyError:
+            pass
+        return False
 
     @use_counter
     @check_update
@@ -834,7 +925,7 @@ class ProjectDirectory(Directory):
         with llfuse.lock_released:
             if not self._current_user:
                 self._current_user = self.api.users().current().execute(num_retries=self.num_retries)
-            return self._current_user["uuid"] in self.project_object["writable_by"]
+            return self._current_user["uuid"] in self.project_object.get("writable_by", [])
 
     def persisted(self):
         return True
@@ -890,7 +981,51 @@ class ProjectDirectory(Directory):
         # Acually move the entry from source directory to this directory.
         del src._entries[name_old]
         self._entries[name_new] = ent
-        self.inodes.invalidate_entry(src.inode, name_old.encode(self.inodes.encoding))
+        self.inodes.invalidate_entry(src, name_old)
+
+    @use_counter
+    def child_event(self, ev):
+        properties = ev.get("properties") or {}
+        old_attrs = properties.get("old_attributes") or {}
+        new_attrs = properties.get("new_attributes") or {}
+        old_attrs["uuid"] = ev["object_uuid"]
+        new_attrs["uuid"] = ev["object_uuid"]
+        old_name = sanitize_filename(self.namefn(old_attrs))
+        new_name = sanitize_filename(self.namefn(new_attrs))
+
+        # create events will have a new name, but not an old name
+        # delete events will have an old name, but not a new name
+        # update events will have an old and new name, and they may be same or different
+        # if they are the same, an unrelated field changed and there is nothing to do.
+
+        if old_attrs.get("owner_uuid") != self.project_uuid:
+            # Was moved from somewhere else, so don't try to remove entry.
+            old_name = None
+        if ev.get("object_owner_uuid") != self.project_uuid:
+            # Was moved to somewhere else, so don't try to add entry
+            new_name = None
+
+        if old_attrs.get("is_trashed"):
+            # Was previously deleted
+            old_name = None
+        if new_attrs.get("is_trashed"):
+            # Has been deleted
+            new_name = None
+
+        if new_name != old_name:
+            ent = None
+            if old_name in self._entries:
+                ent = self._entries[old_name]
+                del self._entries[old_name]
+                self.inodes.invalidate_entry(self, old_name)
+
+            if new_name:
+                if ent is not None:
+                    self._entries[new_name] = ent
+                else:
+                    self._add_entry(new_attrs, new_name)
+            elif ent is not None:
+                self.inodes.del_entry(ent)
 
 
 class SharedDirectory(Directory):
@@ -904,64 +1039,91 @@ class SharedDirectory(Directory):
         self.current_user = api.users().current().execute(num_retries=num_retries)
         self._poll = True
         self._poll_time = poll_time
+        self._updating_lock = threading.Lock()
 
     @use_counter
     def update(self):
-        with llfuse.lock_released:
-            all_projects = arvados.util.list_all(
-                self.api.groups().list, self.num_retries,
-                filters=[['group_class','=','project']])
-            objects = {}
-            for ob in all_projects:
-                objects[ob['uuid']] = ob
-
-            roots = []
-            root_owners = {}
-            for ob in all_projects:
-                if ob['owner_uuid'] != self.current_user['uuid'] and ob['owner_uuid'] not in objects:
-                    roots.append(ob)
-                    root_owners[ob['owner_uuid']] = True
-
-            lusers = arvados.util.list_all(
-                self.api.users().list, self.num_retries,
-                filters=[['uuid','in', list(root_owners)]])
-            lgroups = arvados.util.list_all(
-                self.api.groups().list, self.num_retries,
-                filters=[['uuid','in', list(root_owners)]])
-
-            users = {}
-            groups = {}
-
-            for l in lusers:
-                objects[l["uuid"]] = l
-            for l in lgroups:
-                objects[l["uuid"]] = l
-
-            contents = {}
-            for r in root_owners:
-                if r in objects:
-                    obr = objects[r]
-                    if obr.get("name"):
-                        contents[obr["name"]] = obr
-                    #elif obr.get("username"):
-                    #    contents[obr["username"]] = obr
-                    elif "first_name" in obr:
-                        contents[u"{} {}".format(obr["first_name"], obr["last_name"])] = obr
-
-
-            for r in roots:
-                if r['owner_uuid'] not in objects:
-                    contents[r['name']] = r
-
-        # end with llfuse.lock_released, re-acquire lock
-
         try:
+            with llfuse.lock_released:
+                self._updating_lock.acquire()
+                if not self.stale():
+                    return
+
+                contents = {}
+                roots = []
+                root_owners = set()
+                objects = {}
+
+                methods = self.api._rootDesc.get('resources')["groups"]['methods']
+                if 'httpMethod' in methods.get('shared', {}):
+                    page = []
+                    while True:
+                        resp = self.api.groups().shared(filters=[['group_class', '=', 'project']]+page,
+                                                        order="uuid",
+                                                        limit=10000,
+                                                        count="none",
+                                                        include="owner_uuid").execute()
+                        if not resp["items"]:
+                            break
+                        page = [["uuid", ">", resp["items"][len(resp["items"])-1]["uuid"]]]
+                        for r in resp["items"]:
+                            objects[r["uuid"]] = r
+                            roots.append(r["uuid"])
+                        for r in resp["included"]:
+                            objects[r["uuid"]] = r
+                            root_owners.add(r["uuid"])
+                else:
+                    all_projects = arvados.util.list_all(
+                        self.api.groups().list, self.num_retries,
+                        filters=[['group_class','=','project']],
+                        select=["uuid", "owner_uuid"])
+                    for ob in all_projects:
+                        objects[ob['uuid']] = ob
+
+                    current_uuid = self.current_user['uuid']
+                    for ob in all_projects:
+                        if ob['owner_uuid'] != current_uuid and ob['owner_uuid'] not in objects:
+                            roots.append(ob['uuid'])
+                            root_owners.add(ob['owner_uuid'])
+
+                    lusers = arvados.util.list_all(
+                        self.api.users().list, self.num_retries,
+                        filters=[['uuid','in', list(root_owners)]])
+                    lgroups = arvados.util.list_all(
+                        self.api.groups().list, self.num_retries,
+                        filters=[['uuid','in', list(root_owners)+roots]])
+
+                    for l in lusers:
+                        objects[l["uuid"]] = l
+                    for l in lgroups:
+                        objects[l["uuid"]] = l
+
+                for r in root_owners:
+                    if r in objects:
+                        obr = objects[r]
+                        if obr.get("name"):
+                            contents[obr["name"]] = obr
+                        #elif obr.get("username"):
+                        #    contents[obr["username"]] = obr
+                        elif "first_name" in obr:
+                            contents[u"{} {}".format(obr["first_name"], obr["last_name"])] = obr
+
+                for r in roots:
+                    if r in objects:
+                        obr = objects[r]
+                        if obr['owner_uuid'] not in objects:
+                            contents[obr["name"]] = obr
+
+            # end with llfuse.lock_released, re-acquire lock
+
             self.merge(contents.items(),
                        lambda i: i[0],
                        lambda a, i: a.uuid() == i[1]['uuid'],
                        lambda i: ProjectDirectory(self.inode, self.inodes, self.api, self.num_retries, i[1], poll=self._poll, poll_time=self._poll_time))
         except Exception:
-            _logger.exception()
+            _logger.exception("arv-mount shared dir error")
+        finally:
+            self._updating_lock.release()
 
     def want_event_subscribe(self):
         return True