Fix CWE-190 - DoS could happen with a specially crafted boundary parameter in a multipart request#445
Merged
Merged
Conversation
There was a problem hiding this comment.
Pull request overview
This PR addresses a security issue (CWE-190) in multipart request handling by constraining and hardening boundary parsing to prevent integer overflow / DoS scenarios during multipart body parsing.
Changes:
- Add validation to reject multipart boundaries that are empty or exceed RFC 2046’s 70-character limit.
- Change multipart boundary parsing state tracking (
_boundaryPosition) tosize_tand update a related loop to avoid overflow issues.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
| src/WebRequest.cpp | Adds boundary length validation and updates multipart parsing loop indexing to use size_t. |
| src/ESPAsyncWebServer.h | Updates _boundaryPosition type to size_t to safely track boundary parsing progress. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
me-no-dev
approved these changes
Jun 7, 2026
mathieucarbou
force-pushed
the
CWE-190
branch
2 times, most recently
from
658afa6 to
ec13b20
Compare
Member
Author
|
@me-no-dev : updated the PR and rebased. |
mathieucarbou
force-pushed
the
CWE-190
branch
2 times, most recently
from
300d91c to
c7114c0
Compare
mathieucarbou
force-pushed
the
CWE-190
branch
2 times, most recently
from
3c7ad90 to
4753be6
Compare
_boundaryPosition was declared as uint8_t. A remote attacker could send a
Content-Type header with a boundary string of exactly 256 bytes, causing the
increment in _parseMultipartPostByte() to overflow back to 0. The parser
would then loop indefinitely in the BOUNDARY_OR_DATA state, consuming 100%
CPU and triggering the FreeRTOS watchdog on ESP32/ESP8266.
Fixes:
- ESPAsyncWebServer.h: change _boundaryPosition from uint8_t to size_t,
eliminating the overflow entirely.
- WebRequest.cpp (_parseMultipartPostByte): change all rewind loop variables
from uint8_t to size_t (BOUNDARY_OR_DATA, DASH3_OR_RETURN2, EXPECT_FEED2).
- WebRequest.cpp (_parseMultipartPostByte / EXPECT_BOUNDARY): add a comment
explaining the intentional unsigned underflow when _parsedLength < 2.
- WebRequest.cpp (_parseReqHeader): validate the boundary value extracted
from the Content-Type header before any body parsing begins:
* Media type comparison changed from case-sensitive startsWith() to a
case-insensitive check on a pre-lowercased copy of the value, so that
e.g. 'Multipart/form-data' is correctly recognised (RFC 2045 §5.1).
* Require the 'boundary=' token to be a proper RFC 2045 parameter
(immediately preceded by ';'), preventing false matches on tokens
whose name merely ends in 'boundary=' (e.g. 'x-boundary=').
* Parse the boundary value as either a token or a quoted-string per
RFC 2046 §5.1. Quoted-string form correctly handles backslash-escaped
quotes ("), pairs of backslashes (\), and rejects unterminated
quoted strings. The closing-quote detector counts consecutive
backslashes before the quote so that an even count (backslashes
cancelling each other) correctly identifies the real closing quote.
* Reject boundaries that are empty or longer than 70 characters
(RFC 2046 §5.1 hard limit); drop the connection immediately with
PARSE_REQ_FAIL + abort() instead of continuing.
* Trim _contentType defensively after extraction.
- AsyncWebHeader.cpp (parse): fix OWS handling per RFC 7230 §3.2.3.
The previous code only skipped a single space after the colon; the
correct rule is OWS = *( SP / HTAB ). This affected every parsed
header: Content-Type, Upgrade, Expect, Transfer-Encoding, Host, etc.
The fix is applied at the source so all header branches benefit.
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
me-no-dev
approved these changes
Jun 8, 2026
Member
|
All good with me @mathieucarbou . Feel free to merge and move this forward |
willmmiles
reviewed
Jun 8, 2026
willmmiles
left a comment
willmmiles
left a comment
There was a problem hiding this comment.
Sorry I'm late to review, been a bit busy of late. The implementation is technically workable but inefficient in a few ways. A tokenizer might be more practical. std::string_view<> could go a long way here too - it'd substantially reduce the number of allocations.
| // would mishandle the \\" case. | ||
| int endQuote = 1; | ||
| while (true) { | ||
| endQuote = _boundary.indexOf('"', endQuote); |
There was a problem hiding this comment.
This would be better implemented by re-using the quote handling logic above -- here we declare a second parser to do exactly the same thing with a different, less efficient algorithm.
Worse, we could be building the final boundary value and handling escapes as we iterate through the string, instead of doing it over again.
| // BOUNDARY_OR_DATA loop from ever reaching the end of the boundary and | ||
| // causing unbounded CPU consumption (watchdog reset on ESP32/ESP8266). | ||
| // Reject boundaries outside the valid range before any parsing begins. | ||
| if (_boundary.length() == 0 || _boundary.length() > 70) { |
There was a problem hiding this comment.
This would be better implemented in the search loops above, without having to make a copy of the string first if it's out of spec.
Member
Author
There was a problem hiding this comment.
Yes the check can be done while looping but still we need to accumulate the characters into _boundary. But at least we will not accumulate more than 70 doing that.
| // trailing whitespace. The value may be either a token or a | ||
| // quoted-string (RFC 2045 §5.1 / RFC 2046 §5.1). | ||
| _boundary = value.substring(bpos + (int)T_BOUNDARY_LEN); | ||
| _boundary.trim(); |
There was a problem hiding this comment.
The trim here is redundant and wasteful. We're guaranteed there's no leading whitespace because it's already been processed by the search loop above; and any trailing whitespace will be trimmed in the value analysis below. (If it's even trailing to the boundary value -- it could belong to some other parameter entirely!)
| // optional OWS) so that e.g. 'x-boundary=' is not matched. | ||
| int bpos = -1; | ||
| bool inQuotes = false; | ||
| for (int i = 0; i < (int)lowcase.length(); i++) { |
There was a problem hiding this comment.
A little tweak: the loop end condition can be lowcase.length() - (T_BOUNDARY_LEN+2), eg. enough space for the '=' and a data byte. If there aren't enough bytes left for a successful match, we can stop up here, and then we also don't have to length-check on line 513 (in fact we can even use strcmp for greater speed!).
Member
Author
I agree I completely forgot about it. This PR was an iterative improvement with copilot and it grew to the point where indeed perf improvements could be done. I will open a new PR with them and maybe some others. |
mathieucarbou
added a commit
that referenced
this pull request
Jun 9, 2026
Tighten multipart boundary parsing in _parseReqHeader(): - Replace String::charAt() inner loop with a raw C-string pointer and pre-computed length for faster, allocation-free scanning. - Add an early-exit upper bound on the scan loop so that positions where 'boundary=' cannot possibly fit are skipped entirely. - Replace the three-pass quoted-string extraction (find close-quote, substring, then unescape) with a single-pass approach that writes directly into _boundary using a std::string_view window into the original header value — no intermediate heap allocations. - Enforce the RFC 2046 §5.1 70-character limit on boundary length in both token and quoted-string paths; abort with PARSE_REQ_FAIL on violation to prevent CWE-190 integer-overflow / DoS. - Replace strncmp length guard with a position-based early break that is clearer and avoids the redundant cast. Ref: #445
Member
Author
|
@willmmiles : opened a PR hereL #447 |
mathieucarbou
added a commit
that referenced
this pull request
Jun 9, 2026
Tighten multipart boundary parsing in _parseReqHeader(): - Replace String::charAt() inner loop with a raw C-string pointer and pre-computed length for faster, allocation-free scanning. - Add an early-exit upper bound on the scan loop so that positions where 'boundary=' cannot possibly fit are skipped entirely. - Replace the three-pass quoted-string extraction (find close-quote, substring, then unescape) with a single-pass approach that writes directly into _boundary using a raw pointer+length pair — no intermediate heap allocations and no C++17 dependency. - Enforce the RFC 2046 §5.1 70-character limit on boundary length in both token and quoted-string paths; abort with PARSE_REQ_FAIL on violation to prevent CWE-190 integer-overflow / DoS. - Replace strncmp length guard with a position-based early break that is clearer and avoids the redundant cast. - Fix ESP8266 build: String(const char*, size_t) is unavailable; use a 71-byte stack buffer + String(const char*) instead (#ifdef ESP8266). - Fix LibreTiny/C++14 build: remove std::string_view (C++17 only); replaced with plain const char* + size_t throughout. Ref: #445
mathieucarbou
added a commit
that referenced
this pull request
Jun 9, 2026
Tighten multipart boundary parsing in _parseReqHeader(): - Replace String::charAt() inner loop with a raw C-string pointer and pre-computed length for faster, allocation-free scanning. - Add an early-exit upper bound on the scan loop so that positions where 'boundary=' cannot possibly fit are skipped entirely. - Replace the three-pass quoted-string extraction (find close-quote, substring, then unescape) with a single-pass approach that writes directly into _boundary using a raw pointer+length pair — no intermediate heap allocations and no C++17 dependency. - Enforce the RFC 2046 §5.1 70-character limit on boundary length in both token and quoted-string paths; abort with PARSE_REQ_FAIL on violation to prevent CWE-190 integer-overflow / DoS. - Replace strncmp length guard with a position-based early break that is clearer and avoids the redundant cast. - Fix ESP8266 build: String(const char*, size_t) is unavailable; use a 71-byte stack buffer + String(const char*) instead (#ifdef ESP8266). - Fix LibreTiny/C++14 build: remove std::string_view (C++17 only); replaced with plain const char* + size_t throughout. Ref: #445
mathieucarbou
added a commit
that referenced
this pull request
Jun 9, 2026
Tighten multipart boundary parsing in _parseReqHeader(): - Replace String::charAt() inner loop with a raw C-string pointer and pre-computed length for faster, allocation-free scanning. - Add an early-exit upper bound on the scan loop so that positions where 'boundary=' cannot possibly fit are skipped entirely. - Replace the three-pass quoted-string extraction (find close-quote, substring, then unescape) with a single-pass approach that writes directly into _boundary using a raw pointer+length pair — no intermediate heap allocations and no C++17 dependency. - Enforce the RFC 2046 §5.1 70-character limit on boundary length in both token and quoted-string paths; abort with PARSE_REQ_FAIL on violation to prevent CWE-190 integer-overflow / DoS. - Replace strncmp length guard with a position-based early break that is clearer and avoids the redundant cast. - Fix ESP8266 build: String(const char*, size_t) is unavailable; use a 71-byte stack buffer + String(const char*) instead (#ifdef ESP8266). - Fix LibreTiny/C++14 build: remove std::string_view (C++17 only); replaced with plain const char* + size_t throughout. Ref: #445
mathieucarbou
added a commit
that referenced
this pull request
Jun 9, 2026
Tighten multipart boundary parsing in _parseReqHeader(): - Replace String::charAt() inner loop with a raw C-string pointer and pre-computed length for faster, allocation-free scanning. - Add an early-exit upper bound on the scan loop so that positions where 'boundary=' cannot possibly fit are skipped entirely. - Replace the three-pass quoted-string extraction (find close-quote, substring, then unescape) with a single-pass approach that writes directly into _boundary using a raw pointer+length pair — no intermediate heap allocations and no C++17 dependency. - Enforce the RFC 2046 §5.1 70-character limit on boundary length in both token and quoted-string paths; abort with PARSE_REQ_FAIL on violation to prevent CWE-190 integer-overflow / DoS. - Replace strncmp length guard with a position-based early break that is clearer and avoids the redundant cast. - Fix ESP8266 build: String(const char*, size_t) is unavailable; use a 71-byte stack buffer + String(const char*) instead (#ifdef ESP8266). - Fix LibreTiny/C++14 build: remove std::string_view (C++17 only); replaced with plain const char* + size_t throughout. Ref: #445
1 task
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
See advisory: GHSA-4phx-fcj6-46r4
Tested with our UploadFlash example.
Fix CWE-190: integer overflow in multipart boundary parser (DoS)
_boundaryPosition was declared as uint8_t. A remote attacker could send a
Content-Type header with a boundary string of exactly 256 bytes, causing the
increment in _parseMultipartPostByte() to overflow back to 0. The parser
would then loop indefinitely in the BOUNDARY_OR_DATA state, consuming 100%
CPU and triggering the FreeRTOS watchdog on ESP32/ESP8266.
Fixes:
eliminating the overflow entirely.
from uint8_t to size_t (BOUNDARY_OR_DATA, DASH3_OR_RETURN2, EXPECT_FEED2).
explaining the intentional unsigned underflow when _parsedLength < 2.
from the Content-Type header before any body parsing begins:
case-insensitive check on a pre-lowercased copy of the value, so that
e.g. 'Multipart/form-data' is correctly recognised (RFC 2045 §5.1).
(immediately preceded by ';'), preventing false matches on tokens
whose name merely ends in 'boundary=' (e.g. 'x-boundary=').
RFC 2046 §5.1. Quoted-string form correctly handles backslash-escaped
quotes ("), pairs of backslashes (), and rejects unterminated
quoted strings. The closing-quote detector counts consecutive
backslashes before the quote so that an even count (backslashes
cancelling each other) correctly identifies the real closing quote.
(RFC 2046 §5.1 hard limit); drop the connection immediately with
PARSE_REQ_FAIL + abort() instead of continuing.
The previous code only skipped a single space after the colon; the
correct rule is OWS = *( SP / HTAB ). This affected every parsed
header: Content-Type, Upgrade, Expect, Transfer-Encoding, Host, etc.
The fix is applied at the source so all header branches benefit.