crypto: revert dangerous uses of std::string_view #57816
Merged
Conversation
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
tniessen
added
crypto
|
Review requested:
|
nodejs-github-bot
added
lib / src
tniessen
force-pushed
the
crypto-revert-stringview
branch
from
a8e27f4 to
6fdef91
Compare
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #57816 +/- ##
==========================================
- Coverage 90.23% 90.23% -0.01%
==========================================
Files 630 630
Lines 185481 185480 -1
Branches 36373 36373
==========================================
- Hits 167375 167362 -13
+ Misses 10994 10991 -3
- Partials 7112 7127 +15
🚀 New features to boost your workflow:
|
An `std::string_view v` is a `const char* v.data()` along with an `std::size_t v.size()` that guarantees that `v.size()` contiguous elements of type `char` can be accessed relative to the pointer `v.data()`. One of the main reasons behind the existence of `std::string_view` is the ability to operate on `char` sequences without requiring null termination, which otherwise often requires expensive copies of strings to be made. As a consequence, it is generally incorrect to assume that `v.data()` points to a null-terminated sequence of `char`, and the only way to obtain a null-terminated string from an `std::string_view` is to make a copy. It is not even possible to check if the sequence pointed to by `v.data()` is null-terminated because the null character would be at position `v.data() + v.size()`, which is outside of the range that `v` guarantees safe access to. (A default-constructed `std::string_view` even sets its own data pointer to a `nullptr`, which is fine because it only needs to guarantee safe access to zero elements, i.e., to no elements). In `deps/ncrypto` and `src/crypto`, there are various APIs that consume `std::string_view v` arguments but then ignore `v.size()` and treat `v.data()` as a C-style string of type `const char*`. However, that is not what call sites would expect from functions that explicitly ask for `std::string_view` arguments, since it makes assumptions beyond the guarantees provided by `std::string_view` and leads to undefined behavior unless the given view either contains an embedded null character or the `char` at address `v.data() + v.size()` is a null character. This is not a reasonable assumption for `std::string_view` in general, and it also defeats the purpose of `std::string_view` for the most part since, when `v.size()` is being ignored, it is essentially just a `const char*`. Constructing an `std::string_view` from a `const char*` is also not "free" but requires computing the length of the C-style string (unless the length can be computed at compile time, e.g., because the value is just a string literal). Repeated conversion between `const char*` as used by OpenSSL and `std::string_view` as used by ncrypto thus incurs the additional overhead of computing the length of the string whenever an `std::string_view` is constructed from a `const char*`. (This seems negligible compared to the safety argument though.) Similarly, returning a `const char*` pointer to a C-style string as an `std::string_view` has two downsides: the function must compute the length of the string in order to construct the view, and the caller can no longer assume that the return value is null-terminated and thus cannot pass the returned view to functions that require their arguments to be null terminated. (And, for the reasons explained above, the caller also cannot check if the value is null-terminated without potentially invoking undefined behavior.) C++20 unfortunately does not have a type similar to Rust's `CStr` or GSL `czstring`. Therefore, this commit changes many occurrences of `std::string_view` back to `const char*`, which is conventional for null-terminated C-style strings and does not require computing the length of strings. There are _a lot_ of occurrences of `std::string_view` in ncrypto and for each one, we need to evaluate if it is safe and a good abstraction. I tried to do so, but I might have changed too few or too many, so please feel free to give feedback on individual occurrences.
tniessen
force-pushed
the
crypto-revert-stringview
branch
from
6fdef91 to
bea5e74
Compare
jasnell
approved these changes
Apr 10, 2025
anonrig
approved these changes
Apr 11, 2025
RaisinTen
approved these changes
Apr 11, 2025
tniessen
added
the
commit-queue
nodejs-github-bot
removed
the
commit-queue
|
Landed in 0a1d5d3 |
RafaelGSS
pushed a commit
that referenced
this pull request
May 1, 2025
An `std::string_view v` is a `const char* v.data()` along with an `std::size_t v.size()` that guarantees that `v.size()` contiguous elements of type `char` can be accessed relative to the pointer `v.data()`. One of the main reasons behind the existence of `std::string_view` is the ability to operate on `char` sequences without requiring null termination, which otherwise often requires expensive copies of strings to be made. As a consequence, it is generally incorrect to assume that `v.data()` points to a null-terminated sequence of `char`, and the only way to obtain a null-terminated string from an `std::string_view` is to make a copy. It is not even possible to check if the sequence pointed to by `v.data()` is null-terminated because the null character would be at position `v.data() + v.size()`, which is outside of the range that `v` guarantees safe access to. (A default-constructed `std::string_view` even sets its own data pointer to a `nullptr`, which is fine because it only needs to guarantee safe access to zero elements, i.e., to no elements). In `deps/ncrypto` and `src/crypto`, there are various APIs that consume `std::string_view v` arguments but then ignore `v.size()` and treat `v.data()` as a C-style string of type `const char*`. However, that is not what call sites would expect from functions that explicitly ask for `std::string_view` arguments, since it makes assumptions beyond the guarantees provided by `std::string_view` and leads to undefined behavior unless the given view either contains an embedded null character or the `char` at address `v.data() + v.size()` is a null character. This is not a reasonable assumption for `std::string_view` in general, and it also defeats the purpose of `std::string_view` for the most part since, when `v.size()` is being ignored, it is essentially just a `const char*`. Constructing an `std::string_view` from a `const char*` is also not "free" but requires computing the length of the C-style string (unless the length can be computed at compile time, e.g., because the value is just a string literal). Repeated conversion between `const char*` as used by OpenSSL and `std::string_view` as used by ncrypto thus incurs the additional overhead of computing the length of the string whenever an `std::string_view` is constructed from a `const char*`. (This seems negligible compared to the safety argument though.) Similarly, returning a `const char*` pointer to a C-style string as an `std::string_view` has two downsides: the function must compute the length of the string in order to construct the view, and the caller can no longer assume that the return value is null-terminated and thus cannot pass the returned view to functions that require their arguments to be null terminated. (And, for the reasons explained above, the caller also cannot check if the value is null-terminated without potentially invoking undefined behavior.) C++20 unfortunately does not have a type similar to Rust's `CStr` or GSL `czstring`. Therefore, this commit changes many occurrences of `std::string_view` back to `const char*`, which is conventional for null-terminated C-style strings and does not require computing the length of strings. There are _a lot_ of occurrences of `std::string_view` in ncrypto and for each one, we need to evaluate if it is safe and a good abstraction. I tried to do so, but I might have changed too few or too many, so please feel free to give feedback on individual occurrences. PR-URL: #57816 Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Yagiz Nizipli <yagiz@nizipli.com> Reviewed-By: Darshan Sen <raisinten@gmail.com>
RafaelGSS
pushed a commit
that referenced
this pull request
May 2, 2025
An `std::string_view v` is a `const char* v.data()` along with an `std::size_t v.size()` that guarantees that `v.size()` contiguous elements of type `char` can be accessed relative to the pointer `v.data()`. One of the main reasons behind the existence of `std::string_view` is the ability to operate on `char` sequences without requiring null termination, which otherwise often requires expensive copies of strings to be made. As a consequence, it is generally incorrect to assume that `v.data()` points to a null-terminated sequence of `char`, and the only way to obtain a null-terminated string from an `std::string_view` is to make a copy. It is not even possible to check if the sequence pointed to by `v.data()` is null-terminated because the null character would be at position `v.data() + v.size()`, which is outside of the range that `v` guarantees safe access to. (A default-constructed `std::string_view` even sets its own data pointer to a `nullptr`, which is fine because it only needs to guarantee safe access to zero elements, i.e., to no elements). In `deps/ncrypto` and `src/crypto`, there are various APIs that consume `std::string_view v` arguments but then ignore `v.size()` and treat `v.data()` as a C-style string of type `const char*`. However, that is not what call sites would expect from functions that explicitly ask for `std::string_view` arguments, since it makes assumptions beyond the guarantees provided by `std::string_view` and leads to undefined behavior unless the given view either contains an embedded null character or the `char` at address `v.data() + v.size()` is a null character. This is not a reasonable assumption for `std::string_view` in general, and it also defeats the purpose of `std::string_view` for the most part since, when `v.size()` is being ignored, it is essentially just a `const char*`. Constructing an `std::string_view` from a `const char*` is also not "free" but requires computing the length of the C-style string (unless the length can be computed at compile time, e.g., because the value is just a string literal). Repeated conversion between `const char*` as used by OpenSSL and `std::string_view` as used by ncrypto thus incurs the additional overhead of computing the length of the string whenever an `std::string_view` is constructed from a `const char*`. (This seems negligible compared to the safety argument though.) Similarly, returning a `const char*` pointer to a C-style string as an `std::string_view` has two downsides: the function must compute the length of the string in order to construct the view, and the caller can no longer assume that the return value is null-terminated and thus cannot pass the returned view to functions that require their arguments to be null terminated. (And, for the reasons explained above, the caller also cannot check if the value is null-terminated without potentially invoking undefined behavior.) C++20 unfortunately does not have a type similar to Rust's `CStr` or GSL `czstring`. Therefore, this commit changes many occurrences of `std::string_view` back to `const char*`, which is conventional for null-terminated C-style strings and does not require computing the length of strings. There are _a lot_ of occurrences of `std::string_view` in ncrypto and for each one, we need to evaluate if it is safe and a good abstraction. I tried to do so, but I might have changed too few or too many, so please feel free to give feedback on individual occurrences. PR-URL: #57816 Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Yagiz Nizipli <yagiz@nizipli.com> Reviewed-By: Darshan Sen <raisinten@gmail.com>
aduh95
added
the
backport-requested-v22.x
|
This needs a manual backport for v22.x |
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.
An
std::string_view vis aconst char* v.data()along with anstd::size_t v.size()that guarantees thatv.size()contiguous elements of typecharcan be accessed relative to the pointerv.data().One of the main reasons behind the existence of
std::string_viewis the ability to operate oncharsequences without requiring null termination, which otherwise often requires expensive copies of strings to be made. As a consequence, it is generally incorrect to assume thatv.data()points to a null-terminated sequence ofchar, and the only way to obtain a null-terminated string from anstd::string_viewis to make a copy. It is not even possible to check if the sequence pointed to byv.data()is null-terminated because the null character would be at positionv.data() + v.size(), which is outside of the range thatvguarantees safe access to. (A default-constructedstd::string_vieweven sets its own data pointer to anullptr, which is fine because it only needs to guarantee safe access to zero elements, i.e., to no elements).In
deps/ncryptoandsrc/crypto, there are various APIs that consumestd::string_view varguments but then ignorev.size()and treatv.data()as a C-style string of typeconst char*. However, that is not what call sites would expect from functions that explicitly ask forstd::string_viewarguments, since it makes assumptions beyond the guarantees provided bystd::string_viewand leads to undefined behavior unless the given view either contains an embedded null character or thecharat addressv.data() + v.size()is a null character. This is not a reasonable assumption forstd::string_viewin general, and it also defeats the purpose ofstd::string_viewfor the most part since, whenv.size()is being ignored, it is essentially just aconst char*.Constructing an
std::string_viewfrom aconst char*is also not "free" but requires computing the length of the C-style string (unless the length can be computed at compile time, e.g., because the value is just a string literal). Repeated conversion betweenconst char*as used by OpenSSL andstd::string_viewas used by ncrypto thus incurs the additional overhead of computing the length of the string whenever anstd::string_viewis constructed from aconst char*. (This seems negligible compared to the safety argument though.)Similarly, returning a
const char*pointer to a C-style string as anstd::string_viewhas two downsides: the function must compute the length of the string in order to construct the view, and the caller can no longer assume that the return value is null-terminated and thus cannot pass the returned view to functions that require their arguments to be null terminated. (And, for the reasons explained above, the caller also cannot check if the value is null-terminated without potentially invoking undefined behavior.)C++20 unfortunately does not have a type similar to Rust's
CStror GSLczstring. Therefore, this commit changes many occurrences ofstd::string_viewback toconst char*, which is conventional for null-terminated C-style strings and does not require computing the length of strings.There are a lot of occurrences of
std::string_viewin ncrypto and for each one, we need to evaluate if it is safe and a good abstraction. I tried to do so, but I might have changed too few or too many, so please feel free to give feedback on individual occurrences.cc @nodejs/cpp-reviewers