erts: Disable unsafe optimization in bs_append #9372
Merged
Conversation
This file contains 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
CT Test Results 1 files 11 suites 2m 9s ⏱️ Results for commit b4c3fe1. ♻️ This comment has been updated with latest results. To speed up review, make sure that you have read Contributing to Erlang/OTP and that all checks pass. See the TESTING and DEVELOPMENT HowTo guides for details about how to run test locally. Artifacts// Erlang/OTP Github Action Bot |
bjorng
previously approved these changes
Feb 3, 2025
Imagine that we execute the following code, where `X` and `Y` are integers: ``` A = <<0, X>>, B = <<A/bits, Y:X>>. ``` The first line sets up `erts_current_bin`, `erts_bin_offset`, etc because it obviously should. However, the second line does not do so if X is zero, as `build_size_in_bits` being zero short-circuits the relevant logic. This leaves a dangling pointer that, when we execute `erts_new_bs_put_integer` on the second line, we can land in a code path that ostensibly does not modify the buffer, but may nevertheless do a dummy read/write that is masked off, which can race with a write to the same address on another scheduler. For example, this can occur if there is a garbage collection between `A` and `B`, where the old heap that `A` lived on is picked up by another process before `B` executes. Note that there does not need to be an actual link between `A` and `B` here: all that is necessary is that an expression like `<<A/bits, Y:X>>` is executed where `X` is zero. This issue was mostly benign before 9256aad that introduced an read-modify-write in certain cases that would be okay if `erts_current_bin` and `erts_bin_offset` had been set correctly. While we could fix the read-modify-write issue separately, the invariant that `erts_current_bin` and `erts_bin_offset` are up to date has been broken, and it's difficult to tell whether other parts may suffer the same problem. Therefore, we will disable this optimization and reintroduce it again safely at a later time.
jhogberg
force-pushed
the
john/erts/disable-unsafe-bs_append-optimization/ERIERL-1177/OTP-19462
branch
from
dc0da04 to
b4c3fe1
Compare
bjorng
approved these changes
Feb 4, 2025
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.
Imagine that we execute the following code, where
XandYare integers:The first line sets up
erts_current_bin,erts_bin_offset, etc because it obviously should.However, the second line does not do so if
Xis zero, asbuild_size_in_bitsbeing zero short-circuits the relevant logic.This leaves a dangling pointer that, when we execute
erts_new_bs_put_integeron the second line, we can land in a code path that ostensibly does not modify the buffer, but may nevertheless do a dummy read/write that is masked off, which can race with a write to the same address on another scheduler. For example, this can occur if there is a garbage collection betweenAandB, where the old heap thatAlived on is picked up by another process on another scheduler beforeBexecutes.Note that there does not need to be an actual link between
AandBhere: all that is necessary is that an expression like<<A/bits, Y:X>>is executed whereXis zero. This issue was mostly benign before 9256aad that introduced an read-modify-write in certain cases that would be okay iferts_current_binanderts_bin_offsethad been set correctly.While we could fix the read-modify-write issue separately, the invariant that
erts_current_binanderts_bin_offsetare up to date has been broken, and it's difficult to tell whether other parts may suffer the same problem. Therefore, we will disable this optimization and reintroduce it again safely at a later time.