feat(gov): initial process resource governor (P1-1 + P1-2 core)#53
feat(gov): initial process resource governor (P1-1 + P1-2 core)#53heidi-dang merged 2 commits intomainfrom
Conversation
Summary of ChangesHello @heidi-dang, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request lays the groundwork for process resource governance within the Highlights
Changelog
Activity
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a process resource governor with a custom JSON parser and platform-specific process control logic. A critical vulnerability exists where throwing string-to-number conversion functions (std::stoi, std::stoll) are used on untrusted input without proper exception handling, potentially leading to a process-wide Denial of Service. Additionally, the implementation has critical issues related to correctness and portability, extensively using Linux-specific APIs despite claiming cross-platform compatibility. There's also a significant logic error in how resource limits are retrieved and applied, a potential memory exhaustion issue due to unbounded growth of the rules map, and opportunities to improve performance by avoiding busy-waiting and reducing code duplication.
| int start = std::stoi(left); | ||
| int end = std::stoi(right); | ||
| ranges.push_back({start, end}); | ||
| } else { | ||
| int cpu = std::stoi(part); |
There was a problem hiding this comment.
The use of std::stoi on untrusted input from the affinity string can lead to a Denial of Service. std::stoi throws exceptions (std::invalid_argument, std::out_of_range) when parsing fails or the value is too large. Since these exceptions are not caught in the apply_loop thread, any malformed input will cause the entire process to terminate. Consider using std::from_chars or wrapping these calls in a try-catch block.
| } else if (key == "cpu") { | ||
| if (s.empty() || s.front() != '{') { | ||
| result.ack = AckCode::NACK_PARSE_ERROR; | ||
| result.error_detail = "cpu must be an object"; | ||
| return result; | ||
| } | ||
| size_t brace_end = 1; | ||
| int depth = 1; | ||
| while (brace_end < s.size() && depth > 0) { | ||
| if (s[brace_end] == '{') { | ||
| depth++; | ||
| } else if (s[brace_end] == '}') { | ||
| depth--; | ||
| } | ||
| brace_end++; | ||
| } | ||
| std::string_view cpu_obj = s.substr(0, brace_end); | ||
| s.remove_prefix(brace_end); | ||
|
|
||
| cpu_obj = trim(cpu_obj); | ||
| if (cpu_obj.front() != '{' || cpu_obj.back() != '}') { | ||
| result.ack = AckCode::NACK_PARSE_ERROR; | ||
| result.error_detail = "malformed cpu object"; | ||
| return result; | ||
| } | ||
| cpu_obj.remove_prefix(1); | ||
| cpu_obj.remove_suffix(1); | ||
| cpu_obj = trim(cpu_obj); | ||
|
|
||
| CpuPolicy cpu_policy; | ||
| while (!cpu_obj.empty()) { | ||
| std::string cpu_key; | ||
| if (!parse_key(cpu_obj, cpu_key)) { | ||
| result.ack = AckCode::NACK_PARSE_ERROR; | ||
| result.error_detail = "failed to parse cpu sub-key"; | ||
| return result; | ||
| } | ||
| if (!consume_colon(cpu_obj)) { | ||
| result.ack = AckCode::NACK_PARSE_ERROR; | ||
| result.error_detail = "missing colon in cpu object"; | ||
| return result; | ||
| } | ||
| if (cpu_key == "affinity") { | ||
| std::string val; | ||
| if (!parse_string_value(cpu_obj, val)) { | ||
| result.ack = AckCode::NACK_PARSE_ERROR; | ||
| result.error_detail = "failed to parse affinity value"; | ||
| return result; | ||
| } | ||
| cpu_policy.affinity = val; | ||
| } else if (cpu_key == "nice") { | ||
| int8_t val; | ||
| if (!parse_int8_value(cpu_obj, val)) { | ||
| result.ack = AckCode::NACK_INVALID_RANGE; | ||
| result.error_detail = "nice value out of range (-128 to 127)"; | ||
| return result; | ||
| } | ||
| cpu_policy.nice = val; | ||
| } else if (cpu_key == "max_pct") { | ||
| uint8_t val; | ||
| if (!parse_uint8_value(cpu_obj, val)) { | ||
| result.ack = AckCode::NACK_INVALID_RANGE; | ||
| result.error_detail = "max_pct value out of range (0-255)"; | ||
| return result; | ||
| } | ||
| cpu_policy.max_pct = val; | ||
| } else { | ||
| result.ack = AckCode::NACK_UNKNOWN_FIELD; | ||
| result.error_detail = "unknown cpu field: " + cpu_key; | ||
| return result; | ||
| } | ||
|
|
||
| skip_whitespace(cpu_obj); | ||
| if (!cpu_obj.empty() && cpu_obj[0] == ',') { | ||
| cpu_obj.remove_prefix(1); | ||
| cpu_obj = trim(cpu_obj); | ||
| } | ||
| } | ||
| result.msg.cpu = cpu_policy; | ||
| has_cpu = true; |
There was a problem hiding this comment.
The logic for parsing nested JSON objects is duplicated for cpu, mem, pids, and rlim. This makes the code harder to maintain and more prone to errors. For example, the block for parsing the cpu object is nearly identical to the blocks for other policy types. Consider extracting this logic into a helper function. This function could take the string_view of the object's content and a lambda to handle the specific key-value pairs for each policy type. This would significantly reduce code duplication and improve maintainability.
| int8_t val; | ||
| if (!parse_int8_value(cpu_obj, val)) { | ||
| result.ack = AckCode::NACK_INVALID_RANGE; | ||
| result.error_detail = "nice value out of range (-128 to 127)"; | ||
| return result; | ||
| } | ||
| cpu_policy.nice = val; |
There was a problem hiding this comment.
The validation for the nice value is incorrect. It uses parse_int8_value, which only checks if the value fits in an int8_t (-128 to 127). The valid range for a process's nice value on Linux is -20 to 19. The parser should enforce this specific range to prevent errors during application and to provide a more accurate error message.
int64_t val;
if (!parse_int_value(cpu_obj, val)) {
result.ack = AckCode::NACK_PARSE_ERROR;
result.error_detail = "failed to parse nice value";
return result;
}
if (val < -20 || val > 19) {
result.ack = AckCode::NACK_INVALID_RANGE;
result.error_detail = "nice value out of range (-20 to 19)";
return result;
}
cpu_policy.nice = static_cast<int8_t>(val);| if (getrlimit(RLIMIT_NOFILE, &rl) != 0) { | ||
| result.err = errno; | ||
| result.error_detail = std::strerror(errno); | ||
| return result; | ||
| } | ||
| } | ||
| if (rlim.nofile_hard) { | ||
| rl.rlim_max = *rlim.nofile_hard; | ||
| } else { | ||
| if (getrlimit(RLIMIT_NOFILE, &rl) != 0) { | ||
| result.err = errno; | ||
| result.error_detail = std::strerror(errno); | ||
| return result; | ||
| } | ||
| } | ||
| if (prlimit(pid, RLIMIT_NOFILE, &rl, nullptr) != 0) { | ||
| result.err = errno; | ||
| result.error_detail = "prlimit RLIMIT_NOFILE: " + std::string(std::strerror(errno)); | ||
| return result; | ||
| } | ||
| } | ||
|
|
||
| if (rlim.core_soft || rlim.core_hard) { | ||
| if (rlim.core_soft) { | ||
| rl.rlim_cur = *rlim.core_soft; | ||
| } else { | ||
| if (getrlimit(RLIMIT_CORE, &rl) != 0) { | ||
| result.err = errno; | ||
| result.error_detail = std::strerror(errno); | ||
| return result; | ||
| } | ||
| } | ||
| if (rlim.core_hard) { | ||
| rl.rlim_max = *rlim.core_hard; | ||
| } else { | ||
| if (getrlimit(RLIMIT_CORE, &rl) != 0) { |
There was a problem hiding this comment.
In apply_rlimit, getrlimit is incorrectly used to retrieve current limits. This results in the governor's limits being applied to the target process instead of the target pid's actual limits, leading to a significant logic error in how resource limits are retrieved and applied. Specifically, the logic at lines 276 and 302 overwrites any previously set rl.rlim_cur with the governor's soft limit, ignoring the user-supplied value. A correct and safer implementation would be to first read the existing limits for the target pid using prlimit(pid, resource, nullptr, &rl), modify them with the provided values, and then set them back in a single prlimit call for each resource type.
if (rlim.nofile_soft || rlim.nofile_hard) {
// First, get the current limits for the target process
if (prlimit(pid, RLIMIT_NOFILE, nullptr, &rl) != 0) {
result.err = errno;
result.error_detail = "prlimit get RLIMIT_NOFILE: " + std::string(std::strerror(errno));
return result;
}
// Now, update with new values if they are present
if (rlim.nofile_soft) {
rl.rlim_cur = *rlim.nofile_soft;
}
if (rlim.nofile_hard) {
rl.rlim_max = *rlim.nofile_hard;
}
// Finally, set the new limits
if (prlimit(pid, RLIMIT_NOFILE, &rl, nullptr) != 0) {
result.err = errno;
result.error_detail = "prlimit set RLIMIT_NOFILE: " + std::string(std::strerror(errno));
return result;
}
}
if (rlim.core_soft || rlim.core_hard) {
// First, get the current limits for the target process
if (prlimit(pid, RLIMIT_CORE, nullptr, &rl) != 0) {
result.err = errno;
result.error_detail = "prlimit get RLIMIT_CORE: " + std::string(std::strerror(errno));
return result;
}
// Now, update with new values if they are present
if (rlim.core_soft) {
rl.rlim_cur = *rlim.core_soft;
}
if (rlim.core_hard) {
rl.rlim_max = *rlim.core_hard;
}
// Finally, set the new limits
if (prlimit(pid, RLIMIT_CORE, &rl, nullptr) != 0) {
result.err = errno;
result.error_detail = "prlimit set RLIMIT_CORE: " + std::string(std::strerror(errno));
return result;
}
}| std::lock_guard<std::mutex> lock(rules_mutex_); | ||
| if (result.success) { | ||
| stats_.messages_processed++; | ||
| rules_[msg.pid] = msg; |
There was a problem hiding this comment.
The rules_ map grows indefinitely as new PIDs are governed, as there is no check against the kMaxRules limit (1024) and no mechanism to remove entries for terminated processes. This can lead to memory exhaustion over time. Enforce the kMaxRules limit and implement a cleanup strategy for dead processes. Additionally, the apply_loop function, where this map is managed, currently uses a busy-wait pattern, which is inefficient and consumes unnecessary CPU cycles. Consider using a std::condition_variable for more efficient queue handling.
| out = std::stoll(std::string(s.substr(0, end))); | ||
| s.remove_prefix(end); | ||
| return true; | ||
| } | ||
|
|
||
| bool parse_uint_value(std::string_view& s, uint64_t& out) { | ||
| skip_whitespace(s); | ||
| if (s.empty() || !is_digit(s[0])) { | ||
| return false; | ||
| } | ||
| size_t end = 0; | ||
| while (end < s.size() && is_digit(s[end])) { | ||
| end++; | ||
| } | ||
| out = std::stoull(std::string(s.substr(0, end))); |
There was a problem hiding this comment.
The parse_int_value and parse_uint_value functions use std::stoll and std::stoull, which can throw std::out_of_range exceptions if the numeric string in the payload exceeds the range of a 64-bit integer. Although the payload size is limited, it can still contain numbers large enough to trigger these exceptions. If the caller of parse_gov_apply does not catch them, the process will crash. Using std::from_chars is recommended for exception-safe parsing.
- Add GOV_APPLY schema with strict linear JSON parser (<=512B payload) - Add ACK/NACK codes for validation errors - Implement RuleSet POD structures (cpu/mem/pids/rlim/oom_score_adj) - Add bounded ingress queue (256 capacity) with backpressure - Implement ProcessGovernor with apply loop - Implement affinity (sched_setaffinity), nice (setpriority), rlimits (prlimit RLIMIT_NOFILE/RLIMIT_CORE), oom_score_adj - Add 20 unit tests for parser (passes 20/20)
heidi-dang
force-pushed
the
feat/w2p1-governor
branch
from
33a8683 to
0e1f625
Compare
1 participant
Summary
Pass criteria