Cpu pressure by daandemeyer · Pull Request #9 · daandemeyer/systemd

daandemeyer · 2026-03-18T11:46:21Z

No description provided.

daandemeyer · 2026-03-18T11:47:12Z

github-actions · 2026-03-18T11:47:29Z

Claude review of PR #9 (`dcdf8a8`)

This PR extends the existing memory pressure monitoring infrastructure in sd-event and PID 1 to also support CPU and IO pressure, using Linux's PSI (Pressure Stall Information) interface. The refactoring approach is clean — the memory-pressure-specific code is generalized to work with a PressureResource enum, reducing duplication across all three resource types.

The PR is well structured across 7 commits:

Test macro migration (prep work)
Unprivileged test support (prep work)
sd-event CPU pressure API
PID 1 / core CPU pressure support
IO pressure support (sd-event + core)
Documentation update (MEMORY_PRESSURE.md → PRESSURE.md)
Minor docs fix

Overall the code quality is high and the generalization is done consistently.

Suggestions

~~CPU auto mode missing cpu_accounting check~~ (dismissed) — src/core/cgroup.h — cgroup_context_want_pressure(PRESSURE_CPU) doesn't check c->cpu_accounting, unlike the memory and IO variants. The docs say auto enables "if CPU accounting is anyway enabled", but CPUAccounting=true alone won't trigger it.
Missing sanitize_environment entries for CPU/IO pressure — src/core/manager.c — sanitize_environment() strips MEMORY_PRESSURE_WATCH/MEMORY_PRESSURE_WRITE but does not strip the corresponding CPU_PRESSURE_WATCH/CPU_PRESSURE_WRITE/IO_PRESSURE_WATCH/IO_PRESSURE_WRITE env vars. If PID 1 inherits these from its environment, they could leak into services.

Nits

Parser function naming — src/core/load-fragment-gperf.gperf.in — config_parse_memory_pressure_watch is reused for CPU/IO but the name still says "memory"
Unused callback parameter in test wrappers — src/test/test-pressure.c — The wrapper functions ignore the callback parameter passed to them
Stale comment in cgroup.c — src/core/cgroup.c:2123 — Comment still says "try to open the memory pressure interface anew" but the code now sets up all pressure resource types
IO test file not cleaned up — src/test/test-pressure.c:496 — /var/tmp/.io-pressure-test is created during the IO pressure test but never cleaned up afterward

Workflow run

daandemeyer · 2026-03-18T11:56:47Z

@claude review

daandemeyer · 2026-03-18T12:16:07Z

@claude review

daandemeyer · 2026-03-18T13:40:27Z

@claude review

github-actions · 2026-03-18T14:00:05Z

src/core/cgroup.c

@@ -2110,7 +2116,8 @@ static int unit_update_cgroup(
         * that PID 1 runs before init.scope is even set up. Hence, whenever init.scope is realized, let's
         * try to open the memory pressure interface anew. */


Claude: nit: Stale comment: "try to open the memory pressure interface anew" should be updated to reflect that this now sets up all pressure resource types (memory, CPU, and IO), e.g. "try to open the pressure interfaces anew".

github-actions · 2026-03-18T14:00:06Z

src/test/test-pressure.c

+        for (;;) {
+                _cleanup_close_ int fd = -EBADF;
+
+                fd = open("/var/tmp/.io-pressure-test", O_WRONLY|O_CREAT|O_TRUNC|O_CLOEXEC, 0600);


Claude: nit: /var/tmp/.io-pressure-test is created by the IO pressure child process but never cleaned up after the test (e.g., in outro() or via _cleanup_). Consider adding (void) unlink("/var/tmp/.io-pressure-test"); to the test cleanup.

daandemeyer · 2026-03-18T14:06:25Z

@claude review

github-actions · 2026-03-18T14:22:54Z

src/core/load-fragment-gperf.gperf.in

+{{type}}.MemoryPressureWatch,                 config_parse_memory_pressure_watch,                 0,                                  offsetof({{type}}, cgroup_context.pressure_watch[PRESSURE_MEMORY])
+{{type}}.CPUPressureThresholdSec,             config_parse_sec,                                   0,                                  offsetof({{type}}, cgroup_context.pressure_threshold_usec[PRESSURE_CPU])
+{{type}}.CPUPressureWatch,                    config_parse_memory_pressure_watch,                 0,                                  offsetof({{type}}, cgroup_context.pressure_watch[PRESSURE_CPU])
 {{type}}.NFTSet,                              config_parse_cgroup_nft_set,                        NFT_SET_PARSE_CGROUP,               offsetof({{type}}, cgroup_context)


Claude: nit: config_parse_memory_pressure_watch is reused for CPUPressureWatch and IOPressureWatch, but the function name still says "memory". Consider renaming it to config_parse_pressure_watch for consistency, since it now handles all pressure resource types.

Move the bit-rate computation out of property_get_bit_rates() in networkd-link-bus.c into a standalone link_get_bit_rates() helper in networkd-speed-meter.c, which already owns the speed meter state. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

Expose the speed meter transmit/receive rates as a BitRates struct in the per-link JSON output and the io.systemd.Network Varlink IDL. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

Add an sd_varlink* parameter to both functions so Varlink callers can receive a deferred reply once all async work completes, symmetrically with the existing sd_bus_message* path. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

The handler calls dhcp4_renew() and logs a warning on failure. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

The handler calls sd_dhcp_server_forcerenew() if the server is running and logs a warning on failure. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

The handler calls link_reconfigure_full() with UNCONDITIONALLY|CLEANLY flags and defers the Varlink reply until reconfiguration completes. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

Bind networkd's reload handler to the generic io.systemd.service.Reload method. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

Co-authored-by: Yu Watanabe <watanabe.yu+github@gmail.com> Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

The handler returns the JSON produced by link_build_json(). Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

This makes networkctl fetch bit-rate statistics and offered DHCP leases via Link.Describe() method instead of D-Bus. Co-authored-by: Yu Watanabe <watanabe.yu+github@gmail.com> Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

…enew Varlink methods The D-Bus counterparts (bus_link_method_renew, bus_link_method_force_renew) reject calls on unmanaged interfaces with BUS_ERROR_UNMANAGED_INTERFACE, but the Varlink methods silently succeed. Add the same guard to both Varlink methods, returning io.systemd.Network.Link.InterfaceUnmanaged, and declare the error in the IDL. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Right now, if a user tries to pass VMADDR_CID_ANY to systemd-ssh-proxy, an assert is triggered: $ ssh vsock%4294967295 Assertion 'cid != VMADDR_CID_ANY' failed at src/ssh-generator/ssh-proxy.c:21, function process_vsock_cid(). Aborting. mm_receive_fd: recvmsg: expected received 1 got 0 proxy dialer did not pass back a connection This is becauase the value returned from vsock_parse_cid is not checked before being passed to process_vsock_string. Add a check to prevent that.

It has been observed on some systems[1] that ssh-issue may print out: Try contacting this VM's SSH server via 'ssh vsock%4294967295' from host. i.e. it suggests connecting with VMADDR_CID_ANY, which is not valid. It seems that IOCTL_VM_SOCKETS_GET_LOCAL_CID may return VMADDR_CID_ANY in some cases, e.g. when vsock is not full initialized or so. Treat VMADDR_CID_ANY as special in vsock_get_local_cid(), the same as VMADDR_CID_LOCAL and VMADDR_CID_HOST, and return an error. [1] https://launchpad.net/bugs/2145027

FORCERENEW message support has been disabled so long time for security concern. Most other implementations of DHCP server/client neither support FORCERENEW. Let's completely drop relevant code.

It never fails.

Call client_stop() on error and return 0 only on callback. Normal non-callback functions should propagate errors. This also makes client_initialize_time_events() use event_reset_time_relative().

dns_packet_unref() unconditionally unrefs p->more on every call, even when n_ref > 1. But dns_packet_ref() doesn't ref p->more. This means if a packet with a ->more chain gets ref'd and unref'd multiple times, the chain gets freed too early while the parent still holds a dangling pointer. Move the p->more unref into the n_ref == 1 block so the chain only gets cleaned up when the packet is actually being freed.

This adds the low-level io.systemd.Manager shutdown support. This is (much) simpler than the logind one. It mimics dbus but uses a shared helper for the simple cases. Note that this is more restrictive than the dbus version. The dbus version uses SD_BUS_VTABLE_CAPABILITY(CAP_SYS_BOOT) but the varlink version uses varlink_check_privileged_peer(link). This is mostly because I'm not sure how to do the equivalent in a race-free way. Thanks to Daan for suggesting this.

Extract a common helper varlink_log_caller() and use in the varlink code when logging the caller of a method. It also logs the method now that was tried (but failed) to be logged with log_notice just like manager_log_caller() would do. I was looking into modifying `manager_log_caller` instead and accept a NULL pidref but could not log more than the method without pidref and would make the manager_log_caller slightly less nice. Thanks to keszybz for suggesting this.

This commit allows unset pidref when calling manager_log_caller(). With that we can log manager calls even if we cannot resolve the caller. Currently when we cannot resolve the caller we are just not logging anything. With this commit we at least log the call (even though we don't know what caller it was). Thanks to keszybz for the suggestion.

Follow-up for 7993922.

The commit 6d7cb9a fixes the assignment of the these options when specified through SendOption=. However, it breaks when specified through UserClass= or SendVendorOption=. When UserClass= or SendVendorOption= is specified, the option length is calculated from the sd_dhcp_client.user_class or .vendor_options. Hence, we can use 0 for the length in that case. Follow-up for 6d7cb9a.

systemd now has a system call wrapper that does a long series of #ifdef's to differentiate between architectures and ABIs. This wrapper has two problems. 1. On mips, it needs to differentiate between O32, N32, N64 ABI. It does that via a code block in src/include/override/sys/generate-syscall.py (and derived files): 76 # elif defined(_MIPS_SIM) 77 # if _MIPS_SIM == _MIPS_SIM_ABI32 78 # define systemd_NR_{syscall} {nr_mipso32} 79 # elif _MIPS_SIM == _MIPS_SIM_NABI32 80 # define systemd_NR_{syscall} {nr_mips64n32} 81 # elif _MIPS_SIM == _MIPS_SIM_ABI64 82 # define systemd_NR_{syscall} {nr_mips64} 83 # else 84 # error "Unknown MIPS ABI" 85 # endif 86 # elif defined(__hppa__) Now the _MIPS_SIM* constants stem from a vendor-specific header file sgidefs.h, which is included with glibc, but not with musl. It is however always present in the Linux kernel headers as asm/sgidefs.h ... 2. To work around this, the syscall wrapper already has a block 47 #ifdef ARCH_MIPS 48 #include <asm/sgidefs.h> 49 #endif Turns out, ARCH_MIPS is defined nowhere in Gentoo, neither on glibc nor on musl. As a result the code (by accident, probably sgidefs.h is included transitively somehow) works on glibc, but not on musl. The simplest fix is to replace line 47 in the generator and the derived file with 47 #ifdef __mips__ Two other source code files require a similar fix since they rely on the constants. Bug: systemd#41239 Bug: https://bugs.gentoo.org/971376 Signed-off-by: Andreas K. Hüttel <dilfridge@gentoo.org>

…systemd#41236)

Originally reported in Ubuntu as https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/2145027.

This is not actually used so it doesn't really matter in practice and the fields are used anyway, but for cleanliness fix it Reported on yeswehack.com as YWH-PGM9780-165 Follow-up for 5699a16

mountfsd applies R/O when the varlink readOnly flag is set to false Reported on yeswehack.com as YWH-PGM9780-164 Follow-up for 702a52f

networkctl previously called networkd over D-Bus for several operations. This replaces all of those calls with Varlink, making it the sole IPC mechanism between networkctl and networkd. New Varlink methods added to networkd: - io.systemd.Network.Link (new sub-interface for link-specific operations): - Link.Describe — replaces Link.BitRates + DHCPServer.Leases D-Bus properties - Link.Up — replaces Link.SetUp D-Bus method (moved from **io.systemd.Network.LinkUp**) - Link.Down — replaces Link.SetDown D-Bus method (moved from **io.systemd.Network.LinkDown**) - Link.Renew — replaces Link.Renew D-Bus method - Link.ForceRenew — replaces Link.ForceRenew D-Bus method - Link.Reconfigure — replaces Link.Reconfigure D-Bus method - io.systemd.service - Reload — replaces Manager.Reload D-Bus method Supporting changes: - link_get_bit_rates() extracted from networkd-link.c into networkd-speed-meter.c - BitRates added to link_build_json() so Link.Describe returns them inline alongside the existing interface description, replacing a separate D-Bus read - link_reconfigure_full() and manager_reload() extended to accept sd_varlink* for deferred async replies (consistent with existing sd_bus_message* path) - DHCP lease display (networkctl status) uses Link.Describe instead of DHCPServer.Leases; falls back to ClientId when hostname is not present

From yeswehack.com reports

Currently translated at 100.0% (266 of 266 strings) Co-authored-by: Arif Budiman <arifpedia@gmail.com> Translate-URL: https://translate.fedoraproject.org/projects/systemd/main/id/ Translation: systemd/main

libtss2-tcti-device0 is not installed by default in the openSUSE image, but is now required when building the test image. Without it, the build fails with ``` Shared library 'libtss2-tcti-device.so.0' is not available: libtss2-tcti-device.so.0: cannot open shared object file: No such file or directory ``` Follow-up for 5f85409

…systemd#41246)

Works the same way as memory pressure notifications. Code is refactored to work on enum arrays to reduce duplication.

Make the doc more generic and mention all pressure types, not just memory.

daandemeyer force-pushed the main branch from 6684bdb to 223ca0e Compare March 18, 2026 13:40

github-actions bot reviewed Mar 18, 2026

View reviewed changes

daandemeyer force-pushed the main branch from 223ca0e to ede018b Compare March 18, 2026 14:05

github-actions bot reviewed Mar 18, 2026

View reviewed changes

daandemeyer force-pushed the cpu-pressure branch from dcdf8a8 to 791ae6d Compare March 18, 2026 19:04

noxiouz and others added 12 commits March 18, 2026 22:55

network: add BitRates field to link_build_json() and the Varlink IDL

872feaa

Expose the speed meter transmit/receive rates as a BitRates struct in the per-link JSON output and the io.systemd.Network Varlink IDL. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

network: add io.systemd.Network.Link.Renew() Varlink method

5e7287c

The handler calls dhcp4_renew() and logs a warning on failure. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

network: add io.systemd.Network.Link.ForceRenew() Varlink method

982a3fa

The handler calls sd_dhcp_server_forcerenew() if the server is running and logs a warning on failure. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

network: add io.systemd.Network.Link.Reconfigure() Varlink method

506890f

The handler calls link_reconfigure_full() with UNCONDITIONALLY|CLEANLY flags and defers the Varlink reply until reconfiguration completes. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

networkctl: use new varlink methods

dde311b

Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

network: implement io.systemd.service.Reload() in networkd

6627dfe

Bind networkd's reload handler to the generic io.systemd.service.Reload method. Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

networkctl: use io.systemd.service.Reload() method

54e8ea6

Co-authored-by: Yu Watanabe <watanabe.yu+github@gmail.com> Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

network: add io.systemd.Network.Link.Describe() Varlink method

e90ea94

The handler returns the JSON produced by link_build_json(). Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

networkctl: drop unused functions

1416d3b

Co-developed-by: Claude Opus 4.6 <noreply@anthropic.com>

daandemeyer force-pushed the cpu-pressure branch from 791ae6d to a01d709 Compare March 19, 2026 08:49

noxiouz and others added 3 commits March 19, 2026 11:50

daandemeyer force-pushed the main branch from ede018b to bb7486d Compare March 20, 2026 19:17

daandemeyer closed this Mar 20, 2026

yuwata added 4 commits March 22, 2026 03:34

sd-dhcp-client: add missing error checks

ddded65

sd-dhcp-client: drop disabled FORCERENEW message support

2fb192d

FORCERENEW message support has been disabled so long time for security concern. Most other implementations of DHCP server/client neither support FORCERENEW. Let's completely drop relevant code.

sd-dhcp-client: voidify client_initialize()

ad4d2ae

It never fails.

sd-dhcp-client: propagate errors in client_initialize_{io,time}_events()

26f769b

Call client_stop() on error and return 0 only on callback. Normal non-callback functions should propagate errors. This also makes client_initialize_time_events() use event_reset_time_relative().

daandemeyer force-pushed the cpu-pressure branch from acaf3f0 to 9f7043b Compare March 21, 2026 19:28

Oblivionsage and others added 24 commits March 21, 2026 20:34

sd-varlink: gracefully reject arrays/maps with a null element

924f5a3

Follow-up for 7993922.

sd-dhcp-client: several trivial cleanups (systemd#41256)

9d84401

[RFC] core: add io.systemd.Manager.{PowerOff,SoftReboot,Halt,Kexec} (…

2bc0b82

…systemd#41236)

ssh: handle VMADDR_CID_ANY in a couple places (systemd#41230)

8878df4

Originally reported in Ubuntu as https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/2145027.

core: also set iov_len when deserializing LogExtraFields=

aaac2d3

This is not actually used so it doesn't really matter in practice and the fields are used anyway, but for cleanliness fix it Reported on yeswehack.com as YWH-PGM9780-165 Follow-up for 5699a16

mountfsd: fix readOnly flag inversion

7b148d7

mountfsd applies R/O when the varlink readOnly flag is set to false Reported on yeswehack.com as YWH-PGM9780-164 Follow-up for 702a52f

Two fixlets (systemd#41273)

4d8e090

From yeswehack.com reports

po: Translated using Weblate (Indonesian)

8070689

Currently translated at 100.0% (266 of 266 strings) Co-authored-by: Arif Budiman <arifpedia@gmail.com> Translate-URL: https://translate.fedoraproject.org/projects/systemd/main/id/ Translation: systemd/main

ci: Replace codeql PotentiallyDangerousFunction query with clang-tidy (…

6fb5ec3

…systemd#41246)

test-mempress: Migrate to new assertion macros

2671c46

test-mempress: Support unprivileged operation

83c857b

sd-event: Add support for cpu pressure notifications

3faad3d

core: Add support for CPU pressure notifications

84d7150

Works the same way as memory pressure notifications. Code is refactored to work on enum arrays to reduce duplication.

core: Add I/O pressure support

73b1284

docs: Update MEMORY_PRESSURE.md => PRESSURE.md

7ec45e5

Make the doc more generic and mention all pressure types, not just memory.

docs: Fix window in PRESSURE.md

acb2b96

daandemeyer force-pushed the cpu-pressure branch from 9f7043b to acb2b96 Compare March 23, 2026 12:08

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Cpu pressure#9

Cpu pressure#9
daandemeyer wants to merge 56 commits intomainfrom
cpu-pressure

daandemeyer commented Mar 18, 2026

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

github-actions bot commented Mar 18, 2026 •

edited

Loading

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

github-actions bot Mar 18, 2026

Uh oh!

github-actions bot Mar 18, 2026

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

github-actions bot Mar 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

11 participants

		@@ -2110,7 +2116,8 @@ static int unit_update_cgroup(
		* that PID 1 runs before init.scope is even set up. Hence, whenever init.scope is realized, let's
		* try to open the memory pressure interface anew. */

Conversation

daandemeyer commented Mar 18, 2026

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

github-actions bot commented Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Claude review of PR #9 (dcdf8a8)

Suggestions

Nits

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

github-actions bot Mar 18, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions bot Mar 18, 2026

Choose a reason for hiding this comment

Uh oh!

daandemeyer commented Mar 18, 2026

Uh oh!

github-actions bot Mar 18, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

11 participants

github-actions bot commented Mar 18, 2026 •

edited

Loading

Claude review of PR #9 (`dcdf8a8`)