🔒️(backend) prevnt file change post checks

Before this commit, post a file check, the policy could be
reused to change the verified file.
Now, files are uplaoded to a temporary location, then inside
a transaction that prevents concurrent calls, the file is
copied to its final destination and the checks are run on that one.
A new file can still be updated with the policy but it will never be read, etc.

As part of this change, all files in the new tmp directory
on s3 should have an expiration policy.
This commit is contained in:
Florent Chehab
2026-06-01 14:51:41 +02:00
parent 5ef6e8b5ea
commit a82d7f885a
7 changed files with 180 additions and 86 deletions
+96 -76
View File
@@ -1196,96 +1196,116 @@ class FileViewSet(
"""
Check the actual uploaded file and mark it as ready.
"""
file = self.get_object()
if not file.is_pending_upload:
raise drf_exceptions.ValidationError(
{"file": "This action is only available for files in PENDING state."},
code="file_upload_state_not_pending",
)
# Ensures we go through authorization checks
self.get_object()
s3_client = default_storage.connection.meta.client
validation_error = None
head_response = s3_client.head_object(
Bucket=default_storage.bucket_name, Key=file.file_key
)
file_size = head_response["ContentLength"]
with transaction.atomic():
file = self.get_queryset().select_for_update().get(pk=kwargs["pk"])
if settings.FILE_UPLOAD_APPLY_RESTRICTIONS:
config_for_file_type = settings.FILE_UPLOAD_RESTRICTIONS[file.type]
if file_size > config_for_file_type["max_size"]:
self._complete_file_deletion(file)
logger.info(
"upload_ended: file size (%s) for file %s higher than the allowed max size",
file_size,
file.file_key,
)
if not file.is_pending_upload:
raise drf_exceptions.ValidationError(
detail="The file size is higher than the allowed max size.",
code="file_size_exceeded",
{
"file": "This action is only available for files in PENDING state."
},
code="file_upload_state_not_pending",
)
# python-magic recommends using at least the first 2048 bytes
# to reduce incorrect identification.
# This is a tradeoff between pulling in the whole file and the most likely relevant bytes
# of the file for mime type identification.
if file_size > 2048:
range_response = s3_client.get_object(
Bucket=default_storage.bucket_name,
Key=file.file_key,
Range="bytes=0-2047",
)
file_head = range_response["Body"].read()
else:
file_head = s3_client.get_object(
Bucket=default_storage.bucket_name, Key=file.file_key
)["Body"].read()
# Use improved MIME type detection combining magic bytes and file extension
logger.info("upload_ended: detecting mimetype for file: %s", file.file_key)
mimetype = utils.detect_mimetype(file_head, filename=file.filename)
if settings.FILE_UPLOAD_APPLY_RESTRICTIONS:
config_for_file_type = settings.FILE_UPLOAD_RESTRICTIONS[file.type]
allowed_file_mimetypes = config_for_file_type["allowed_mimetypes"]
if mimetype not in allowed_file_mimetypes:
self._complete_file_deletion(file)
logger.warning(
"upload_ended: mimetype not allowed %s for file %s",
mimetype,
file.file_key,
)
raise drf_exceptions.ValidationError(
detail="The file type is not allowed.",
code="file_type_not_allowed",
)
file.upload_state = models.FileUploadStateChoices.READY
file.mimetype = mimetype
file.size = file_size
file.save(update_fields=["upload_state", "mimetype", "size"])
if head_response["ContentType"] != mimetype:
logger.info(
"upload_ended: content type mismatch between object storage and file,"
" updating from %s to %s",
head_response["ContentType"],
mimetype,
)
# We copy the file to its final destination, we will run the checks on that
# final file and ignore any updates to the temporary file. (We cannot revoke the policy,
# so the temporary file might still be updated after that.)
# The temporary folders will need to be cleaned periodically
s3_client.copy_object(
Bucket=default_storage.bucket_name,
Key=file.file_key,
CopySource={
"Bucket": default_storage.bucket_name,
"Key": file.file_key,
"Key": file.temporary_file_key,
},
ContentType=mimetype,
Metadata=head_response["Metadata"],
MetadataDirective="REPLACE",
)
head_response = s3_client.head_object(
Bucket=default_storage.bucket_name, Key=file.file_key
)
file_size = head_response["ContentLength"]
# python-magic recommends using at least the first 2048 bytes
# to reduce incorrect identification.
# This is a tradeoff between pulling in the whole file and
# the most likely relevant bytes
# of the file for mime type identification.
if file_size > 2048:
range_response = s3_client.get_object(
Bucket=default_storage.bucket_name,
Key=file.file_key,
Range="bytes=0-2047",
)
file_head = range_response["Body"].read()
else:
file_head = s3_client.get_object(
Bucket=default_storage.bucket_name, Key=file.file_key
)["Body"].read()
logger.info("upload_ended: detecting mimetype for file: %s", file.file_key)
mimetype = utils.detect_mimetype(file_head, filename=file.filename)
if settings.FILE_UPLOAD_APPLY_RESTRICTIONS:
config_for_file_type = settings.FILE_UPLOAD_RESTRICTIONS[file.type]
if file_size > config_for_file_type["max_size"]:
logger.info(
"upload_ended: file size (%s) for file %s higher than the allowed max size",
file_size,
file.file_key,
)
validation_error = drf_exceptions.ValidationError(
detail="The file size is higher than the allowed max size.",
code="file_size_exceeded",
)
else:
# Use improved MIME type detection combining magic bytes and file extension
allowed_file_mimetypes = config_for_file_type["allowed_mimetypes"]
if mimetype not in allowed_file_mimetypes:
logger.warning(
"upload_ended: mimetype not allowed %s for file %s",
mimetype,
file.file_key,
)
validation_error = drf_exceptions.ValidationError(
detail="The file type is not allowed.",
code="file_type_not_allowed",
)
if validation_error is not None:
self._complete_file_deletion(file)
else:
file.upload_state = models.FileUploadStateChoices.READY
file.mimetype = mimetype
file.size = file_size
file.save(update_fields=["upload_state", "mimetype", "size"])
if head_response["ContentType"] != mimetype:
logger.info(
"upload_ended: content type mismatch between object storage and file,"
" updating from %s to %s",
head_response["ContentType"],
mimetype,
)
s3_client.copy_object(
Bucket=default_storage.bucket_name,
Key=file.file_key,
CopySource={
"Bucket": default_storage.bucket_name,
"Key": file.file_key,
},
ContentType=mimetype,
Metadata=head_response["Metadata"],
MetadataDirective="REPLACE",
)
if validation_error:
raise validation_error
# Not yet implemented
# Change the file.upload_state when this will be done
# malware_detection.analyse_file(file.file_key, file_id=file.id)
@@ -1298,7 +1318,7 @@ class FileViewSet(
"""Delete a file completely."""
file.soft_delete()
file.hard_delete()
process_file_deletion.delay(file.id)
transaction.on_commit(lambda: process_file_deletion.delay(file.id))
def _authorize_subrequest(self, request, pattern):
"""