2022-12-14 23:00:43 +01:00
|
|
|
import base64
|
|
|
|
import binascii
|
2022-11-30 19:11:35 +01:00
|
|
|
import os
|
2022-12-14 23:00:43 +01:00
|
|
|
from datetime import timedelta
|
2023-12-05 21:25:00 +01:00
|
|
|
from urllib.parse import quote, urlsplit
|
2020-06-11 00:54:34 +02:00
|
|
|
|
|
|
|
from django.conf import settings
|
2021-11-02 15:42:58 +01:00
|
|
|
from django.contrib.auth.models import AnonymousUser
|
2022-06-21 22:23:34 +02:00
|
|
|
from django.core.files.uploadedfile import UploadedFile
|
2022-12-14 23:00:43 +01:00
|
|
|
from django.core.signing import BadSignature, TimestampSigner
|
2024-06-21 20:58:42 +02:00
|
|
|
from django.db import transaction
|
2022-12-06 22:26:39 +01:00
|
|
|
from django.http import (
|
|
|
|
FileResponse,
|
|
|
|
HttpRequest,
|
|
|
|
HttpResponse,
|
|
|
|
HttpResponseBase,
|
|
|
|
HttpResponseForbidden,
|
|
|
|
HttpResponseNotFound,
|
|
|
|
)
|
2024-07-15 23:23:36 +02:00
|
|
|
from django.http.request import MediaType
|
2016-06-07 01:09:05 +02:00
|
|
|
from django.shortcuts import redirect
|
2022-12-14 23:00:43 +01:00
|
|
|
from django.urls import reverse
|
2023-11-21 22:03:02 +01:00
|
|
|
from django.utils.cache import patch_cache_control, patch_vary_headers
|
2023-04-19 05:16:13 +02:00
|
|
|
from django.utils.http import content_disposition_header
|
2021-04-16 00:57:30 +02:00
|
|
|
from django.utils.translation import gettext as _
|
2016-06-07 01:09:05 +02:00
|
|
|
|
2021-11-02 15:42:58 +01:00
|
|
|
from zerver.context_processors import get_valid_realm_from_request
|
2023-11-21 22:23:22 +01:00
|
|
|
from zerver.decorator import zulip_redirect_to_login
|
2023-12-15 20:03:19 +01:00
|
|
|
from zerver.lib.attachments import validate_attachment_request
|
2021-06-30 18:35:50 +02:00
|
|
|
from zerver.lib.exceptions import JsonableError
|
2024-06-18 23:41:37 +02:00
|
|
|
from zerver.lib.mime_types import guess_type
|
2021-06-30 18:35:50 +02:00
|
|
|
from zerver.lib.response import json_success
|
2023-11-21 22:03:02 +01:00
|
|
|
from zerver.lib.storage import static_path
|
2024-06-21 20:58:42 +02:00
|
|
|
from zerver.lib.thumbnail import (
|
|
|
|
THUMBNAIL_OUTPUT_FORMATS,
|
|
|
|
BaseThumbnailFormat,
|
|
|
|
StoredThumbnailFormat,
|
2024-07-17 03:11:57 +02:00
|
|
|
get_image_thumbnail_path,
|
2024-06-21 20:58:42 +02:00
|
|
|
)
|
2022-12-06 22:24:43 +01:00
|
|
|
from zerver.lib.upload import (
|
|
|
|
check_upload_within_quota,
|
|
|
|
get_public_upload_root_url,
|
2023-02-28 04:02:25 +01:00
|
|
|
upload_message_attachment_from_request,
|
2022-12-06 22:24:43 +01:00
|
|
|
)
|
2022-12-14 21:51:37 +01:00
|
|
|
from zerver.lib.upload.base import INLINE_MIME_TYPES
|
2022-12-14 23:00:43 +01:00
|
|
|
from zerver.lib.upload.local import assert_is_local_storage_path
|
2022-12-14 21:51:37 +01:00
|
|
|
from zerver.lib.upload.s3 import get_signed_upload_url
|
2024-08-29 23:39:07 +02:00
|
|
|
from zerver.models import Attachment, ImageAttachment, UserProfile
|
2024-06-21 20:58:42 +02:00
|
|
|
from zerver.worker.thumbnail import ensure_thumbnails
|
2020-06-11 00:54:34 +02:00
|
|
|
|
2016-06-07 01:09:05 +02:00
|
|
|
|
2024-08-29 23:39:07 +02:00
|
|
|
def patch_disposition_header(response: HttpResponse, filename: str, is_attachment: bool) -> None:
|
2023-04-19 05:16:13 +02:00
|
|
|
content_disposition = content_disposition_header(is_attachment, filename)
|
2022-11-30 19:11:35 +01:00
|
|
|
|
2023-04-19 05:16:13 +02:00
|
|
|
if content_disposition is not None:
|
|
|
|
response.headers["Content-Disposition"] = content_disposition
|
2022-11-30 19:11:35 +01:00
|
|
|
|
|
|
|
|
2024-07-12 02:30:23 +02:00
|
|
|
def internal_nginx_redirect(internal_path: str, content_type: str | None = None) -> HttpResponse:
|
2022-12-06 22:26:39 +01:00
|
|
|
# The following headers from this initial response are
|
|
|
|
# _preserved_, if present, and sent unmodified to the client;
|
|
|
|
# all other headers are overridden by the redirected URL:
|
|
|
|
# - Content-Type
|
|
|
|
# - Content-Disposition
|
|
|
|
# - Accept-Ranges
|
|
|
|
# - Set-Cookie
|
|
|
|
# - Cache-Control
|
|
|
|
# - Expires
|
2023-01-11 17:43:14 +01:00
|
|
|
# As such, we default to unsetting the Content-type header to
|
|
|
|
# allow nginx to set it from the static file; the caller can set
|
|
|
|
# Content-Disposition and Cache-Control on this response as they
|
|
|
|
# desire, and the client will see those values. In some cases
|
|
|
|
# (local files) we do wish to control the Content-Type, so also
|
|
|
|
# support setting it explicitly.
|
|
|
|
response = HttpResponse(content_type=content_type)
|
2022-12-06 22:26:39 +01:00
|
|
|
response["X-Accel-Redirect"] = internal_path
|
2023-01-11 17:43:14 +01:00
|
|
|
if content_type is None:
|
|
|
|
del response["Content-Type"]
|
2022-12-06 22:26:39 +01:00
|
|
|
return response
|
|
|
|
|
|
|
|
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
def serve_s3(request: HttpRequest, path_id: str, force_download: bool = False) -> HttpResponse:
|
|
|
|
url = get_signed_upload_url(path_id, force_download=force_download)
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
assert url.startswith("https://")
|
|
|
|
|
|
|
|
if settings.DEVELOPMENT:
|
|
|
|
# In development, we do not have the nginx server to offload
|
|
|
|
# the response to; serve a redirect to the short-lived S3 URL.
|
|
|
|
# This means the content cannot be cached by the browser, but
|
|
|
|
# this is acceptable in development.
|
|
|
|
return redirect(url)
|
2016-06-08 11:22:06 +02:00
|
|
|
|
uploads: Extra-escape internal S3 paths.
In nginx, `location` blocks operate on the _decoded_ URI[^1]:
> The matching is performed against a normalized URI, after decoding
> the text encoded in the “%XX” form
This means that if a user-uploaded file contains characters that are
not URI-safe, the browser encodes them in UTF-8 and then URI-encodes
them -- and nginx decodes them and reassembles the original character
before running the `location ~ ^/...` match. This means that the `$2`
_is not URI-encoded_ and _may contain non-ASCII characters.
When `proxy_pass` is passed a value containing one or more variables,
it does no encoding on that expanded value, assuming that the bytes
are exactly as they should be passed to the upstream. This means that
directly calling `proxy_pass https://$1/$2` would result in sending
high-bit characters to the S3 upstream, which would rightly balk.
However, a longstanding bug in nginx's `set` directive[^2] means that
the following line:
```nginx
set $download_url https://$1/$2;
```
...results in nginx accidentally URI-encoding $1 and $2 when they are
inserted, resulting in a `$download_url` which is suitable to pass to
`proxy_pass`. This bug is only present with numeric capture
variables, not named captures; this is particularly relevant because
numeric captures are easily overridden by additional regexes
elsewhere, as subsequent commits will add.
Fixing this is complicated; nginx does not supply any way to escape
values[^3], besides a third-party module[^4] which is an undue
complication to begin using. The only variable which nginx exposes
which is _not_ un-escaped already is `$request_uri`, which contains
the very original URL sent by the browser -- and thus can't respect
any work done in Django to generate the `X-Accel-Redirect` (e.g., for
`/user_uploads/temporary/` URLs). We also cannot pass these URLs to
nginx via query-parameters, since `$arg_foo` values are not
URI-decoded by nginx, there is no function to do so[^3], and the
values must be URI-encoded because they themselves are URLs with query
parameters.
Extra-URI-encode the path that we pass to the `X-Accel-Redirect`
location, for S3 redirects. We rely on the `location` block
un-escaping that layer, leaving `$s3_hostname` and `$s3_path` as they
were intended in Django.
This works around the nginx bug, with no behaviour change.
[^1]: http://nginx.org/en/docs/http/ngx_http_core_module.html#location
[^2]: https://trac.nginx.org/nginx/ticket/348
[^3]: https://trac.nginx.org/nginx/ticket/52
[^4]: https://github.com/openresty/set-misc-nginx-module#set_escape_uri
2023-01-31 21:05:12 +01:00
|
|
|
# We over-escape the path, to work around it being impossible to
|
2024-06-09 06:01:03 +02:00
|
|
|
# get the _unescaped_ new internal request URL in nginx.
|
2023-12-05 21:25:00 +01:00
|
|
|
parsed_url = urlsplit(url)
|
uploads: Extra-escape internal S3 paths.
In nginx, `location` blocks operate on the _decoded_ URI[^1]:
> The matching is performed against a normalized URI, after decoding
> the text encoded in the “%XX” form
This means that if a user-uploaded file contains characters that are
not URI-safe, the browser encodes them in UTF-8 and then URI-encodes
them -- and nginx decodes them and reassembles the original character
before running the `location ~ ^/...` match. This means that the `$2`
_is not URI-encoded_ and _may contain non-ASCII characters.
When `proxy_pass` is passed a value containing one or more variables,
it does no encoding on that expanded value, assuming that the bytes
are exactly as they should be passed to the upstream. This means that
directly calling `proxy_pass https://$1/$2` would result in sending
high-bit characters to the S3 upstream, which would rightly balk.
However, a longstanding bug in nginx's `set` directive[^2] means that
the following line:
```nginx
set $download_url https://$1/$2;
```
...results in nginx accidentally URI-encoding $1 and $2 when they are
inserted, resulting in a `$download_url` which is suitable to pass to
`proxy_pass`. This bug is only present with numeric capture
variables, not named captures; this is particularly relevant because
numeric captures are easily overridden by additional regexes
elsewhere, as subsequent commits will add.
Fixing this is complicated; nginx does not supply any way to escape
values[^3], besides a third-party module[^4] which is an undue
complication to begin using. The only variable which nginx exposes
which is _not_ un-escaped already is `$request_uri`, which contains
the very original URL sent by the browser -- and thus can't respect
any work done in Django to generate the `X-Accel-Redirect` (e.g., for
`/user_uploads/temporary/` URLs). We also cannot pass these URLs to
nginx via query-parameters, since `$arg_foo` values are not
URI-decoded by nginx, there is no function to do so[^3], and the
values must be URI-encoded because they themselves are URLs with query
parameters.
Extra-URI-encode the path that we pass to the `X-Accel-Redirect`
location, for S3 redirects. We rely on the `location` block
un-escaping that layer, leaving `$s3_hostname` and `$s3_path` as they
were intended in Django.
This works around the nginx bug, with no behaviour change.
[^1]: http://nginx.org/en/docs/http/ngx_http_core_module.html#location
[^2]: https://trac.nginx.org/nginx/ticket/348
[^3]: https://trac.nginx.org/nginx/ticket/52
[^4]: https://github.com/openresty/set-misc-nginx-module#set_escape_uri
2023-01-31 21:05:12 +01:00
|
|
|
assert parsed_url.hostname is not None
|
|
|
|
assert parsed_url.path is not None
|
|
|
|
assert parsed_url.query is not None
|
|
|
|
escaped_path_parts = parsed_url.hostname + quote(parsed_url.path) + "?" + parsed_url.query
|
|
|
|
response = internal_nginx_redirect("/internal/s3/" + escaped_path_parts)
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
|
|
|
|
# It is important that S3 generate both the Content-Type and
|
|
|
|
# Content-Disposition headers; when the file was uploaded, we
|
|
|
|
# stored the browser-provided value for the former, and set
|
|
|
|
# Content-Disposition according to if that was safe. As such,
|
|
|
|
# only S3 knows if a given attachment is safe to inline; we only
|
|
|
|
# override Content-Disposition to "attachment", and do so by
|
|
|
|
# telling S3 that is what we want in the signed URL.
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
patch_cache_control(response, private=True, immutable=True)
|
|
|
|
return response
|
2020-04-08 00:27:24 +02:00
|
|
|
|
2021-02-12 08:19:30 +01:00
|
|
|
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
def serve_local(
|
2024-08-29 23:39:07 +02:00
|
|
|
request: HttpRequest,
|
|
|
|
path_id: str,
|
|
|
|
filename: str,
|
|
|
|
force_download: bool = False,
|
2024-09-04 20:19:25 +02:00
|
|
|
mimetype: str | None = None,
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
) -> HttpResponseBase:
|
2023-01-06 01:18:13 +01:00
|
|
|
assert settings.LOCAL_FILES_DIR is not None
|
|
|
|
local_path = os.path.join(settings.LOCAL_FILES_DIR, path_id)
|
|
|
|
assert_is_local_storage_path("files", local_path)
|
|
|
|
if not os.path.isfile(local_path):
|
2021-02-12 08:20:45 +01:00
|
|
|
return HttpResponseNotFound("<p>File not found</p>")
|
2018-03-13 07:08:27 +01:00
|
|
|
|
2024-09-04 20:19:25 +02:00
|
|
|
if mimetype is None:
|
|
|
|
mimetype = guess_type(filename)[0]
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
download = force_download or mimetype not in INLINE_MIME_TYPES
|
|
|
|
|
2022-12-06 22:26:39 +01:00
|
|
|
if settings.DEVELOPMENT:
|
|
|
|
# In development, we do not have the nginx server to offload
|
2023-01-11 17:43:14 +01:00
|
|
|
# the response to; serve it directly ourselves. FileResponse
|
|
|
|
# handles setting Content-Type, Content-Disposition, etc.
|
2023-02-02 03:40:54 +01:00
|
|
|
response: HttpResponseBase = FileResponse(
|
2023-12-05 18:45:07 +01:00
|
|
|
open(local_path, "rb"), # noqa: SIM115
|
|
|
|
as_attachment=download,
|
2024-08-29 23:39:07 +02:00
|
|
|
filename=filename,
|
2024-09-04 20:19:25 +02:00
|
|
|
content_type=mimetype,
|
2023-02-02 03:40:54 +01:00
|
|
|
)
|
2022-12-06 22:26:39 +01:00
|
|
|
patch_cache_control(response, private=True, immutable=True)
|
|
|
|
return response
|
|
|
|
|
2023-01-11 17:43:14 +01:00
|
|
|
# For local responses, we are in charge of generating both
|
|
|
|
# Content-Type and Content-Disposition headers; unlike with S3
|
|
|
|
# storage, the Content-Type is not stored with the file in any
|
|
|
|
# way, so Django makes the determination of it, and thus as well
|
|
|
|
# if that type is safe to have a Content-Disposition of "inline".
|
|
|
|
# nginx respects the values we send.
|
|
|
|
response = internal_nginx_redirect(
|
|
|
|
quote(f"/internal/local/uploads/{path_id}"), content_type=mimetype
|
|
|
|
)
|
2024-08-29 23:39:07 +02:00
|
|
|
patch_disposition_header(response, filename, download)
|
2022-12-06 22:26:39 +01:00
|
|
|
patch_cache_control(response, private=True, immutable=True)
|
2019-10-02 00:10:30 +02:00
|
|
|
return response
|
2016-06-09 12:19:56 +02:00
|
|
|
|
2021-02-12 08:19:30 +01:00
|
|
|
|
2022-03-22 04:38:18 +01:00
|
|
|
def serve_file_download_backend(
|
2023-06-07 22:26:04 +02:00
|
|
|
request: HttpRequest,
|
2024-07-12 02:30:23 +02:00
|
|
|
maybe_user_profile: UserProfile | AnonymousUser,
|
2023-06-07 22:26:04 +02:00
|
|
|
realm_id_str: str,
|
|
|
|
filename: str,
|
2022-12-06 22:26:39 +01:00
|
|
|
) -> HttpResponseBase:
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
return serve_file(
|
2023-06-07 22:26:04 +02:00
|
|
|
request, maybe_user_profile, realm_id_str, filename, url_only=False, force_download=True
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
)
|
2022-03-22 04:38:18 +01:00
|
|
|
|
|
|
|
|
2021-02-12 08:19:30 +01:00
|
|
|
def serve_file_backend(
|
2021-11-02 15:42:58 +01:00
|
|
|
request: HttpRequest,
|
2024-07-12 02:30:23 +02:00
|
|
|
maybe_user_profile: UserProfile | AnonymousUser,
|
2021-11-02 15:42:58 +01:00
|
|
|
realm_id_str: str,
|
|
|
|
filename: str,
|
2024-06-21 20:58:42 +02:00
|
|
|
thumbnail_format: str | None = None,
|
2022-12-06 22:26:39 +01:00
|
|
|
) -> HttpResponseBase:
|
2024-06-21 20:58:42 +02:00
|
|
|
return serve_file(
|
|
|
|
request,
|
|
|
|
maybe_user_profile,
|
|
|
|
realm_id_str,
|
|
|
|
filename,
|
|
|
|
url_only=False,
|
|
|
|
thumbnail_format=thumbnail_format,
|
|
|
|
)
|
2020-04-08 00:27:24 +02:00
|
|
|
|
2021-02-12 08:19:30 +01:00
|
|
|
|
|
|
|
def serve_file_url_backend(
|
|
|
|
request: HttpRequest, user_profile: UserProfile, realm_id_str: str, filename: str
|
2022-12-06 22:26:39 +01:00
|
|
|
) -> HttpResponseBase:
|
2020-04-08 00:27:24 +02:00
|
|
|
"""
|
|
|
|
We should return a signed, short-lived URL
|
|
|
|
that the client can use for native mobile download, rather than serving a redirect.
|
|
|
|
"""
|
|
|
|
|
|
|
|
return serve_file(request, user_profile, realm_id_str, filename, url_only=True)
|
|
|
|
|
2021-02-12 08:19:30 +01:00
|
|
|
|
2024-07-12 02:30:23 +02:00
|
|
|
def preferred_accept(request: HttpRequest, served_types: list[str]) -> str | None:
|
2023-11-21 22:03:02 +01:00
|
|
|
# Returns the first of the served_types which the browser will
|
|
|
|
# accept, based on the browser's stated quality preferences.
|
|
|
|
# Returns None if none of the served_types are accepted by the
|
|
|
|
# browser.
|
|
|
|
accepted_types = sorted(
|
|
|
|
request.accepted_types,
|
|
|
|
key=lambda e: float(e.params.get("q", "1.0")),
|
|
|
|
reverse=True,
|
|
|
|
)
|
|
|
|
for potential_type in accepted_types:
|
|
|
|
for served_type in served_types:
|
|
|
|
if potential_type.match(served_type):
|
|
|
|
return served_type
|
|
|
|
return None
|
|
|
|
|
|
|
|
|
2024-07-15 23:23:36 +02:00
|
|
|
def closest_thumbnail_format(
|
|
|
|
requested_format: BaseThumbnailFormat,
|
|
|
|
accepts: list[MediaType],
|
|
|
|
rendered_formats: list[StoredThumbnailFormat],
|
|
|
|
) -> StoredThumbnailFormat:
|
|
|
|
accepted_types = sorted(
|
|
|
|
accepts,
|
|
|
|
key=lambda e: float(e.params.get("q", "1.0")),
|
|
|
|
reverse=True,
|
|
|
|
)
|
|
|
|
|
|
|
|
def q_for(content_type: str) -> float:
|
|
|
|
for potential_type in accepted_types:
|
|
|
|
if potential_type.match(content_type):
|
|
|
|
return float(potential_type.params.get("q", "1.0"))
|
|
|
|
return 0.0
|
|
|
|
|
|
|
|
# Serve a "close" format -- preferring animated which
|
|
|
|
# matches, followed by the format they requested, or one
|
|
|
|
# their browser supports, in the size closest to what they
|
|
|
|
# requested, with the minimum bytes.
|
|
|
|
def grade_format(
|
|
|
|
possible_format: StoredThumbnailFormat,
|
|
|
|
) -> tuple[bool, bool, float, int, int]:
|
|
|
|
return (
|
|
|
|
possible_format.animated != requested_format.animated,
|
|
|
|
possible_format.extension != requested_format.extension,
|
|
|
|
1.0 - q_for(possible_format.content_type),
|
|
|
|
abs(requested_format.max_width - possible_format.max_width),
|
|
|
|
possible_format.byte_size,
|
|
|
|
)
|
|
|
|
|
|
|
|
return sorted(rendered_formats, key=grade_format)[0]
|
|
|
|
|
|
|
|
|
2021-02-12 08:19:30 +01:00
|
|
|
def serve_file(
|
|
|
|
request: HttpRequest,
|
2024-07-12 02:30:23 +02:00
|
|
|
maybe_user_profile: UserProfile | AnonymousUser,
|
2021-02-12 08:19:30 +01:00
|
|
|
realm_id_str: str,
|
|
|
|
filename: str,
|
2024-06-21 20:58:42 +02:00
|
|
|
thumbnail_format: str | None = None,
|
2021-02-12 08:19:30 +01:00
|
|
|
url_only: bool = False,
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
force_download: bool = False,
|
2022-12-06 22:26:39 +01:00
|
|
|
) -> HttpResponseBase:
|
2020-06-10 06:41:04 +02:00
|
|
|
path_id = f"{realm_id_str}/{filename}"
|
2021-11-02 15:42:58 +01:00
|
|
|
realm = get_valid_realm_from_request(request)
|
2024-09-04 20:04:03 +02:00
|
|
|
is_authorized, attachment = validate_attachment_request(maybe_user_profile, path_id, realm)
|
2016-06-17 19:48:17 +02:00
|
|
|
|
2023-11-21 22:03:02 +01:00
|
|
|
def serve_image_error(status: int, image_path: str) -> HttpResponseBase:
|
|
|
|
# We cannot use X-Accel-Redirect to offload the serving of
|
|
|
|
# this image to nginx, because it does not preserve the status
|
|
|
|
# code of this response, nor the Vary: header.
|
|
|
|
return FileResponse(open(static_path(image_path), "rb"), status=status) # noqa: SIM115
|
|
|
|
|
2024-09-04 20:04:03 +02:00
|
|
|
if attachment is None:
|
2023-11-21 22:03:02 +01:00
|
|
|
if preferred_accept(request, ["text/html", "image/png"]) == "image/png":
|
|
|
|
response = serve_image_error(404, "images/errors/image-not-exist.png")
|
|
|
|
else:
|
|
|
|
response = HttpResponseNotFound(
|
|
|
|
_("<p>This file does not exist or has been deleted.</p>")
|
|
|
|
)
|
|
|
|
patch_vary_headers(response, ("Accept",))
|
|
|
|
return response
|
2016-06-17 19:48:17 +02:00
|
|
|
if not is_authorized:
|
2023-11-21 22:03:02 +01:00
|
|
|
if preferred_accept(request, ["text/html", "image/png"]) == "image/png":
|
|
|
|
response = serve_image_error(403, "images/errors/image-no-auth.png")
|
2023-11-21 22:23:22 +01:00
|
|
|
elif isinstance(maybe_user_profile, AnonymousUser):
|
|
|
|
response = zulip_redirect_to_login(request)
|
2023-11-21 22:03:02 +01:00
|
|
|
else:
|
|
|
|
response = HttpResponseForbidden(_("<p>You are not authorized to view this file.</p>"))
|
|
|
|
patch_vary_headers(response, ("Accept",))
|
|
|
|
return response
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
if url_only:
|
|
|
|
url = generate_unauthed_file_access_url(path_id)
|
|
|
|
return json_success(request, data=dict(url=url))
|
|
|
|
|
2024-06-21 20:58:42 +02:00
|
|
|
if thumbnail_format is not None:
|
|
|
|
# Check if this is something that we thumbnail at all
|
|
|
|
try:
|
|
|
|
image_attachment = ImageAttachment.objects.get(path_id=path_id)
|
|
|
|
except ImageAttachment.DoesNotExist:
|
|
|
|
return serve_image_error(404, "images/errors/image-not-exist.png")
|
|
|
|
|
|
|
|
# Validate that this is a potential thumbnail format
|
|
|
|
requested_format = BaseThumbnailFormat.from_string(thumbnail_format)
|
|
|
|
if requested_format is None:
|
|
|
|
return serve_image_error(404, "images/errors/image-not-exist.png")
|
|
|
|
|
|
|
|
rendered_formats = [StoredThumbnailFormat(**f) for f in image_attachment.thumbnail_metadata]
|
|
|
|
|
|
|
|
# We never generate animated versions if the input was still,
|
|
|
|
# so filter those out
|
|
|
|
if image_attachment.frames == 1:
|
|
|
|
potential_output_formats = [
|
|
|
|
thumbnail_format
|
|
|
|
for thumbnail_format in THUMBNAIL_OUTPUT_FORMATS
|
|
|
|
if not thumbnail_format.animated
|
|
|
|
]
|
|
|
|
else:
|
|
|
|
potential_output_formats = THUMBNAIL_OUTPUT_FORMATS
|
|
|
|
if requested_format not in potential_output_formats:
|
2024-07-15 23:23:36 +02:00
|
|
|
if rendered_formats == []:
|
|
|
|
# We haven't rendered anything, and they requested
|
|
|
|
# something we don't support.
|
|
|
|
return serve_image_error(404, "images/errors/image-not-exist.png")
|
|
|
|
elif requested_format in rendered_formats:
|
2024-06-21 20:58:42 +02:00
|
|
|
# Not a _current_ format, but we did render it at the time, so fine to serve
|
|
|
|
pass
|
|
|
|
else:
|
2024-07-15 23:23:36 +02:00
|
|
|
# Find something "close enough". This will not be a
|
|
|
|
# common occurrence -- the client has out of date
|
|
|
|
# information about which formats are supported, and
|
|
|
|
# the thumbnails were generated with an even earlier
|
|
|
|
# set, or the client is just guessing a format and
|
|
|
|
# hoping.
|
|
|
|
requested_format = closest_thumbnail_format(
|
|
|
|
requested_format, request.accepted_types, rendered_formats
|
|
|
|
)
|
2024-06-21 20:58:42 +02:00
|
|
|
elif requested_format not in rendered_formats:
|
|
|
|
# They requested a valid format, but one we've not
|
|
|
|
# rendered yet. Take a lock on the row, then render every
|
|
|
|
# missing format, synchronously. The lock prevents us
|
|
|
|
# from doing double-work if the background worker is
|
|
|
|
# currently processing the row.
|
|
|
|
with transaction.atomic(savepoint=False):
|
|
|
|
ensure_thumbnails(
|
|
|
|
ImageAttachment.objects.select_for_update().get(id=image_attachment.id)
|
|
|
|
)
|
|
|
|
|
|
|
|
# Update the path that we are fetching to be the thumbnail
|
|
|
|
path_id = get_image_thumbnail_path(image_attachment, requested_format)
|
2024-08-29 23:39:07 +02:00
|
|
|
served_filename = str(requested_format)
|
2024-09-04 20:19:25 +02:00
|
|
|
mimetype: str | None = None # Guess from filename
|
2024-08-29 23:39:07 +02:00
|
|
|
else:
|
|
|
|
served_filename = attachment.file_name
|
2024-09-04 20:19:25 +02:00
|
|
|
mimetype = attachment.content_type
|
2024-06-21 20:58:42 +02:00
|
|
|
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
if settings.LOCAL_UPLOADS_DIR is not None:
|
2024-08-29 23:39:07 +02:00
|
|
|
return serve_local(
|
2024-09-04 20:19:25 +02:00
|
|
|
request,
|
|
|
|
path_id,
|
|
|
|
filename=served_filename,
|
|
|
|
force_download=force_download,
|
|
|
|
mimetype=mimetype,
|
2024-08-29 23:39:07 +02:00
|
|
|
)
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
else:
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
return serve_s3(request, path_id, force_download=force_download)
|
2020-04-08 00:27:24 +02:00
|
|
|
|
2021-02-12 08:19:30 +01:00
|
|
|
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
USER_UPLOADS_ACCESS_TOKEN_SALT = "user_uploads_"
|
2022-12-14 23:00:43 +01:00
|
|
|
|
|
|
|
|
|
|
|
def generate_unauthed_file_access_url(path_id: str) -> str:
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
signed_data = TimestampSigner(salt=USER_UPLOADS_ACCESS_TOKEN_SALT).sign(path_id)
|
2022-12-14 23:00:43 +01:00
|
|
|
token = base64.b16encode(signed_data.encode()).decode()
|
|
|
|
|
|
|
|
filename = path_id.split("/")[-1]
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
return reverse("file_unauthed_from_token", args=[token, filename])
|
2022-12-14 23:00:43 +01:00
|
|
|
|
|
|
|
|
2024-07-12 02:30:23 +02:00
|
|
|
def get_file_path_id_from_token(token: str) -> str | None:
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
signer = TimestampSigner(salt=USER_UPLOADS_ACCESS_TOKEN_SALT)
|
2022-12-14 23:00:43 +01:00
|
|
|
try:
|
|
|
|
signed_data = base64.b16decode(token).decode()
|
2023-11-21 15:34:07 +01:00
|
|
|
path_id = signer.unsign(
|
|
|
|
signed_data, max_age=timedelta(seconds=settings.SIGNED_ACCESS_TOKEN_VALIDITY_IN_SECONDS)
|
|
|
|
)
|
2022-12-14 23:00:43 +01:00
|
|
|
except (BadSignature, binascii.Error):
|
|
|
|
return None
|
|
|
|
|
|
|
|
return path_id
|
|
|
|
|
|
|
|
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
def serve_file_unauthed_from_token(
|
|
|
|
request: HttpRequest, token: str, filename: str
|
|
|
|
) -> HttpResponseBase:
|
|
|
|
path_id = get_file_path_id_from_token(token)
|
2020-04-08 00:27:24 +02:00
|
|
|
if path_id is None:
|
2021-06-30 18:35:50 +02:00
|
|
|
raise JsonableError(_("Invalid token"))
|
2021-02-12 08:20:45 +01:00
|
|
|
if path_id.split("/")[-1] != filename:
|
2021-06-30 18:35:50 +02:00
|
|
|
raise JsonableError(_("Invalid filename"))
|
2024-08-29 23:39:07 +02:00
|
|
|
try:
|
|
|
|
attachment = Attachment.objects.get(path_id=path_id)
|
|
|
|
except Attachment.DoesNotExist:
|
|
|
|
raise JsonableError(_("Invalid token"))
|
2016-06-08 11:22:06 +02:00
|
|
|
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
if settings.LOCAL_UPLOADS_DIR is not None:
|
2024-09-04 20:19:25 +02:00
|
|
|
return serve_local(
|
|
|
|
request,
|
|
|
|
path_id,
|
|
|
|
filename=attachment.file_name,
|
|
|
|
mimetype=attachment.content_type,
|
|
|
|
)
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
else:
|
CVE-2023-22735: Provide the Content-Disposition header from S3.
The Content-Type of user-provided uploads was provided by the browser
at initial upload time, and stored in S3; however, 04cf68b45ebb
switched to determining the Content-Disposition merely from the
filename. This makes uploads vulnerable to a stored XSS, wherein a
file uploaded with a content-type of `text/html` and an extension of
`.png` would be served to browsers as `Content-Disposition: inline`,
which is unsafe.
The `Content-Security-Policy` headers in the previous commit mitigate
this, but only for browsers which support them.
Revert parts of 04cf68b45ebb, specifically by allowing S3 to provide
the Content-Disposition header, and using the
`ResponseContentDisposition` argument when necessary to override it to
`attachment`. Because we expect S3 responses to vary based on this
argument, we include it in the cache key; since the query parameter
has dashes in it, we can't use use the helper `$arg_` variables, and
must parse it from the query parameters manually.
Adding the disposition may decrease the cache hit rate somewhat, but
downloads are infrequent enough that it is unlikely to have a
noticeable effect. We take care to not adjust the cache key for
requests which do not specify the disposition.
2023-01-11 17:36:41 +01:00
|
|
|
return serve_s3(request, path_id)
|
2016-06-08 11:22:06 +02:00
|
|
|
|
2021-02-12 08:19:30 +01:00
|
|
|
|
2022-12-06 22:24:43 +01:00
|
|
|
def serve_local_avatar_unauthed(request: HttpRequest, path: str) -> HttpResponseBase:
|
|
|
|
"""Serves avatar images off disk, via nginx (or directly in dev), with no auth.
|
|
|
|
|
|
|
|
This is done unauthed because these need to be accessed from HTML
|
|
|
|
emails, where the client does not have any auth. We rely on the
|
|
|
|
URL being generated using the AVATAR_SALT secret.
|
|
|
|
|
|
|
|
"""
|
|
|
|
if settings.LOCAL_AVATARS_DIR is None:
|
|
|
|
# We do not expect clients to hit this URL when using the S3
|
|
|
|
# backend; however, there is no reason to not serve the
|
|
|
|
# redirect to S3 where the content lives.
|
2024-06-13 21:23:51 +02:00
|
|
|
url = get_public_upload_root_url() + path
|
|
|
|
return redirect(url, permanent=True)
|
2022-12-06 22:24:43 +01:00
|
|
|
|
|
|
|
local_path = os.path.join(settings.LOCAL_AVATARS_DIR, path)
|
|
|
|
assert_is_local_storage_path("avatars", local_path)
|
|
|
|
if not os.path.isfile(local_path):
|
|
|
|
return HttpResponseNotFound("<p>File not found</p>")
|
|
|
|
|
|
|
|
if settings.DEVELOPMENT:
|
2023-02-02 03:40:54 +01:00
|
|
|
response: HttpResponseBase = FileResponse(open(local_path, "rb")) # noqa: SIM115
|
2022-12-06 22:24:43 +01:00
|
|
|
else:
|
uploads: Serve S3 uploads directly from nginx.
When file uploads are stored in S3, this means that Zulip serves as a
302 to S3. Because browsers do not cache redirects, this means that
no image contents can be cached -- and upon every page load or reload,
every recently-posted image must be re-fetched. This incurs extra
load on the Zulip server, as well as potentially excessive bandwidth
usage from S3, and on the client's connection.
Switch to fetching the content from S3 in nginx, and serving the
content from nginx. These have `Cache-control: private, immutable`
headers set on the response, allowing browsers to cache them locally.
Because nginx fetching from S3 can be slow, and requests for uploads
will generally be bunched around when a message containing them are
first posted, we instruct nginx to cache the contents locally. This
is safe because uploaded file contents are immutable; access control
is still mediated by Django. The nginx cache key is the URL without
query parameters, as those parameters include a time-limited signed
authentication parameter which lets nginx fetch the non-public file.
This adds a number of nginx-level configuration parameters to control
the caching which nginx performs, including the amount of in-memory
index for he cache, the maximum storage of the cache on disk, and how
long data is retained in the cache. The currently-chosen figures are
reasonable for small to medium deployments.
The most notable effect of this change is in allowing browsers to
cache uploaded image content; however, while there will be many fewer
requests, it also has an improvement on request latency. The
following tests were done with a non-AWS client in SFO, a server and
S3 storage in us-east-1, and with 100 requests after 10 requests of
warm-up (to fill the nginx cache). The mean and standard deviation
are shown.
| | Redirect to S3 | Caching proxy, hot | Caching proxy, cold |
| ----------------- | ------------------- | ------------------- | ------------------- |
| Time in Django | 263.0 ms ± 28.3 ms | 258.0 ms ± 12.3 ms | 258.0 ms ± 12.3 ms |
| Small file (842b) | 586.1 ms ± 21.1 ms | 266.1 ms ± 67.4 ms | 288.6 ms ± 17.7 ms |
| Large file (660k) | 959.6 ms ± 137.9 ms | 609.5 ms ± 13.0 ms | 648.1 ms ± 43.2 ms |
The hot-cache performance is faster for both large and small files,
since it saves the client the time having to make a second request to
a separate host. This performance improvement remains at least 100ms
even if the client is on the same coast as the server.
Cold nginx caches are only slightly slower than hot caches, because
VPC access to S3 endpoints is extremely fast (assuming it is in the
same region as the host), and nginx can pool connections to S3 and
reuse them.
However, all of the 648ms taken to serve a cold-cache large file is
occupied in nginx, as opposed to the only 263ms which was spent in
nginx when using redirects to S3. This means that to overall spend
less time responding to uploaded-file requests in nginx, clients will
need to find files in their local cache, and skip making an
uploaded-file request, at least 60% of the time. Modeling shows a
reduction in the number of client requests by about 70% - 80%.
The `Content-Disposition` header logic can now also be entirely shared
with the local-file codepath, as can the `url_only` path used by
mobile clients. While we could provide the direct-to-S3 temporary
signed URL to mobile clients, we choose to provide the
served-from-Zulip signed URL, to better control caching headers on it,
and greater consistency. In doing so, we adjust the salt used for the
URL; since these URLs are only valid for 60s, the effect of this salt
change is minimal.
2022-11-22 20:41:35 +01:00
|
|
|
response = internal_nginx_redirect(quote(f"/internal/local/user_avatars/{path}"))
|
2022-12-06 22:24:43 +01:00
|
|
|
|
2024-06-13 14:57:18 +02:00
|
|
|
patch_cache_control(response, max_age=31536000, public=True, immutable=True)
|
2022-12-06 22:24:43 +01:00
|
|
|
return response
|
|
|
|
|
|
|
|
|
2017-11-27 09:28:57 +01:00
|
|
|
def upload_file_backend(request: HttpRequest, user_profile: UserProfile) -> HttpResponse:
|
2016-06-27 19:28:09 +02:00
|
|
|
if len(request.FILES) == 0:
|
2021-06-30 18:35:50 +02:00
|
|
|
raise JsonableError(_("You must specify a file to upload"))
|
2016-06-27 19:28:09 +02:00
|
|
|
if len(request.FILES) != 1:
|
2021-06-30 18:35:50 +02:00
|
|
|
raise JsonableError(_("You may only upload one file at a time"))
|
2016-06-27 19:28:09 +02:00
|
|
|
|
2023-07-22 00:34:11 +02:00
|
|
|
[user_file] = request.FILES.values()
|
2022-06-21 22:23:34 +02:00
|
|
|
assert isinstance(user_file, UploadedFile)
|
2018-02-02 05:43:18 +01:00
|
|
|
file_size = user_file.size
|
2022-06-21 22:23:34 +02:00
|
|
|
assert file_size is not None
|
2024-01-29 00:52:43 +01:00
|
|
|
if file_size > settings.MAX_FILE_UPLOAD_SIZE * 1024 * 1024:
|
2021-06-30 18:35:50 +02:00
|
|
|
raise JsonableError(
|
2023-07-17 22:40:33 +02:00
|
|
|
_("Uploaded file is larger than the allowed limit of {max_size} MiB").format(
|
|
|
|
max_size=settings.MAX_FILE_UPLOAD_SIZE,
|
2021-02-12 08:19:30 +01:00
|
|
|
)
|
|
|
|
)
|
2018-01-26 16:13:33 +01:00
|
|
|
check_upload_within_quota(user_profile.realm, file_size)
|
2016-06-27 19:28:09 +02:00
|
|
|
|
2024-08-30 04:13:01 +02:00
|
|
|
url, filename = upload_message_attachment_from_request(user_file, user_profile)
|
2024-07-15 07:06:38 +02:00
|
|
|
|
|
|
|
# TODO/compatibility: uri is a deprecated alias for url that can
|
|
|
|
# be removed once there are no longer clients relying on it.
|
2024-08-30 04:13:01 +02:00
|
|
|
return json_success(request, data={"uri": url, "url": url, "filename": filename})
|