Introduction to the getrusage() system call. Covers the function signature, the three who flags (RUSAGE_SELF, RUSAGE_CHILDREN, RUSAGE_THREAD), what the call returns, and when to use each flag.

→ Open

Deep dive into all 17 fields of struct rusage. Explains which 9 are actually used on Linux, which 8 are BSD legacy zeros, and what each used field means — CPU time, page faults, context switches, RSS.

→ Open

How child resource usage is accumulated through the process tree. The wait() dependency rule, grandchild scenario, ru_maxrss is a max not a sum for children, and the pre-2.6.9 SIGCHLD deviation bug.

→ Open

What resource limits are, why they matter, the ulimit/limit shell commands, and how /proc/PID/limits (since 2.6.24) lets you view any process’s limits without writing code.

→ Open

Function signatures, the struct rlimit with rlim_cur and rlim_max, the rlim_t data type, RLIM_INFINITY, and safe printing with (long long) cast and %lld format specifier.

→ Open

Who can change what: privileged vs unprivileged processes, irreversibility of hard limit lowering, CAP_SYS_RESOURCE, fork/exec inheritance, per-user-ID scoped limits, and the complete Table 36-1 of all 15 constants.

→ Open

Full walkthrough of Listing 36-2 (printRlimit helper) and Listing 36-3 (RLIMIT_NPROC demo). Explains why only 4 children were created (26 pre-existing), EAGAIN error handling, and zombie accumulation.

→ Open

The 32-bit vs 64-bit rlim_t mismatch on x86-32. RLIM_SAVED_CUR and RLIM_SAVED_MAX constants, what happens when _FILE_OFFSET_BITS=64 expands rlim_t to 64 bits, and glibc’s silent RLIM_INFINITY conversion behaviour.

→ Open

The Section 36.3 introduction: how limits get set in practice (ulimit, /proc, setrlimit), complete Table 36-1 with all 15 constants, categorisation by type (memory/CPU/file/IPC), and the three patterns of breach behaviour (signal+error, error only, signal only).

→ Open

Virtual address space size limit. brk/sbrk/mmap/mremap/shmat fail with ENOMEM on breach. How malloc() hits this limit. Interaction with stack growth.

→ Open

Core dump file size limit. Setting to 0 disables core dumps entirely. Security use case (prevent memory contents being dumped to disk). RLIMIT_FSIZE interaction (whichever is smaller wins).

→ Open

CPU time limit in seconds. SIGXCPU at soft limit (once per second thereafter), SIGKILL at hard limit. Portable signal handler pattern. How UNIX implementations differ in post-SIGXCPU behaviour.

→ Open

Data segment size limit — the sum of .data + .bss + heap. sbrk()/brk() fail with ENOMEM when the heap tries to grow past this limit. malloc() returns NULL as a result.

→ Open

Maximum file size limit. SIGXFSZ signal AND EFBIG error are both generated simultaneously when write() or truncate() would exceed the soft limit. Default action is core dump.

→ Open

Locked (non-swappable) memory limit. Affects mlock(), mlockall(), and locking options of mmap() and shmctl(). MCL_FUTURE flag can cause later brk/mmap/mremap to fail if the limit is set too low.

→ Open

POSIX message queue byte limit per real user ID. The byte formula: maxmsg × sizeof(msg_msg*) + maxmsg × msgsize. The sizeof(msg_msg*) addend prevents zero-length message flooding. Linux-specific since 2.6.8.

→ Open

Nice value ceiling for unprivileged processes. The formula: max nice = 20 − rlim_cur. Affects setpriority() and nice(). Linux-specific since 2.6.12. Why the inverted formula is used.

→ Open

Maximum file descriptor number + 1. EMFILE on open/socket/pipe/dup. NR_OPEN hard ceiling (pre-2.6.25) vs /proc/sys/fs/nr_open (2.6.25+). System-wide file-max limit. sysconf(_SC_OPEN_MAX) behaviour.

→ Open

Maximum processes/threads per real user ID. fork/vfork/clone fail with EAGAIN. threads-max system file. sysconf(_SC_CHILD_MAX) bug fixed in 2.6.23. Scanning /proc/PID/status Uid to estimate current count.

→ Open

Resident set size limit — present but not enforced on modern Linux. Historical context: it had a minor effect on MADV_WILLNEED in old 2.4 kernels (up to 2.4.29). Why Linux never fully implemented it.

→ Open

Realtime scheduling priority ceiling. Affects sched_setscheduler() and sched_setparam(). Linux-specific since 2.6.12. Important for embedded real-time Linux systems (PREEMPT_RT patch).

→ Open

Realtime CPU time limit in microseconds — prevents a runaway RT process from locking the system. SIGXCPU at soft limit, SIGKILL at hard limit, same as RLIMIT_CPU. Linux-specific since 2.6.25.

→ Open

Maximum queued signals per real user ID. sigqueue() fails with EAGAIN on breach. kill() still works even at the limit (one instance of each non-queued signal). SigQ in /proc/PID/status. Linux-specific since 2.6.8.

→ Open

Maximum stack size. SIGSEGV on overflow — requires alternate signal stack to catch. Since 2.6.23, this limit also controls space for command-line arguments and environment variables passed to execve().

→ Open

Consolidated summary of the entire chapter. getrusage() and its three who flags, the soft/hard limit model, privilege rules, fork/exec inheritance, and a quick review of all 15 resource limit constants.

→ Open

All three chapter exercises with complete solution code. Exercise 36-1: RUSAGE_CHILDREN wait() dependency. Exercise 36-2: rusage command wrapper (like /usr/bin/time). Exercise 36-3: behaviour when consumption already exceeds the new soft limit.

→ Open