[casaroli]

Summary

<syslog.h> defined openlog() as a macro that is a block of statements and closelog() as an empty macro. Programs that write (void) openlog(...); or (void) closelog(); did not compile (tcp-wrappers' tcpd.c, for example), and a program that declares them itself, takes their address, or a configure link test did not find them.

The first commit makes them functions of libc, as in musl: openlog() keeps a copy of the ident (up to 31 characters), the options and the facility in the task information of the process (task_info_s.ta_openlog, one per task group, in every build mode), and the messages of the process get ident: before them, ident[pid]: with LOG_PID. With LOG_PERROR, syslog() writes the message to stderr too. The other options and the facility have no effect (NuttX has no system logger to connect to); closelog() has no descriptor to close, and the settings stay, as in musl. The OS adds the ident in nx_vsyslog(), in the stream that already writes the header, so syslog() needs no buffer and no lock: vsyslog() of the programs sets the internal flag SYSLOG_OPENLOG in the priority, and only then nx_vsyslog() reads the settings, during a syslog() call of that process. Messages of the OS and from interrupt handlers have no ident. The second commit describes them in Documentation/standards/posix.rst and on the logging page.

Impact

Two new libc functions; the two macros are gone. The same in flat, protected and kernel builds; nothing per architecture. A file that defines its own openlog() or closelog() macro before <syslog.h> does not compile any more, because the macro now expands inside the declaration. nuttx-apps' interpreters/luamodules/luasyslog did that (on the command line); apache/nuttx-apps#3939 removes those macros and works with both the old and the new <syslog.h>, so it must merge first (sim:lua builds luasyslog).

Testing

Test program openlog.c (expr: (void) openlog(...) and (void) closelog() compile; decl: the program declares both itself and takes the address of openlog(); log: syslog() after openlog() and after closelog()), compiled against the export of qemu-armv8a:knsh (kernel build) and run under QEMU (HVF):

	master aeab6fb	this PR
openlog	does not compile (macro "closelog" passed 1 arguments, expected identifier before '{')	expr PASS, decl PASS, log PASS

Test program openlog2.c (stderr is a pipe, the test reads what syslog() wrote), qemu-armv8a:knsh (kernel) and qemu-armv8a:nsh (flat, builtin app), both with CONFIG_PIPES: ident olt: one 1, LOG_PID olt[5]: two, a message with its own newline, an ident of 40 characters cut to 31, the settings after closelog(), no ident: PASS; flat only: a new task group has its own settings (its openlog() does not change the ident of the parent): PASS. The console shows the same lines: the prefix that nx_vsyslog() adds.

ostest exits with status 0 on qemu-armv8a:knsh (kernel) and qemu-armv8a:nsh (flat).

Builds, each without new warnings: stm32f4discovery:nsh (flat, Cortex-M4), lm3s6965-ek:qemu-protected and mps2-an500:knsh (protected, Cortex-M3), rv-virt:knsh64 (kernel, RISC-V), qemu-armv8a:nsh with CONFIG_SYSLOG_RFC5424 (flat, arm64) and sim:lua (macOS host, with luasyslog and the nuttx-apps change). A file with the old luasyslog macros and openlog()/closelog() calls, compiled against the knsh export: master compiles it, this PR fails ("macro "closelog" passed 1 arguments"); without the macros both compile.

tools/checkpatch.sh -c -u -m -g passes.

depends-on: apache/nuttx-apps#3939

[cederom]

Not sure if this is okay to state that functions are "SUPPORTED" but DOES NOTHING as per proposition, over stating that functions are not implemented. Sounds like cheating :D

Either we support them and their functionality is operational or not at all :-)

[xiaoxiang781216]

why not always put into task tls even for kernel mode

[casaroli]

Done.

[xiaoxiang781216]

syslog is called in interrupt context, malloc and mutex need be avoided

[casaroli]

Done.

[xiaoxiang781216]

why not do the format in kernel side, so we can avoid malloc and mutex

[casaroli]

Done.

[casaroli]

Either we support them and their functionality is operational or not at all :-)

Done.

[github-actions]

MemBrowse Memory Report · Explore what changed →

arduino-mega2560

• flash: .data +18 B, .text +514 B (+0.8%, 65,664 B / 262,144 B, total: 25% used)
• sram: .bss +32 B, .data +18 B (+1.9%, 2,740 B / 8,192 B, total: 33% used)

esp32-devkitc

• ROM: .flash.rodata +16 B, .flash.text +364 B (+0.3%, 128,164 B / 4,194,272 B, total: 3% used)
• dram0_0_seg: .dram0.bss +32 B (+0.3%, 9,928 B / 180,736 B, total: 5% used)
• drom0_0_seg: .flash.rodata +16 B (+0.1%, 13,812 B / 4,194,272 B, total: 0% used)
• irom0_0_seg: .flash.text +364 B (+0.4%, 94,124 B / 3,342,304 B, total: 3% used)

hifive1-revb

• flash: .text +344 B (+0.4%, 84,876 B / 4,194,304 B, total: 2% used)
• sram: .bss +32 B (+0.8%, 3,924 B / 16,384 B, total: 24% used)

mirtoo

• kseg0_progmem: .text +596 B (+0.9%, 67,396 B / 131,072 B, total: 51% used)
• kseg1_datamem: .bss +48 B (+1.9%, 2,568 B / 32,256 B, total: 8% used)

qemu-armv8a

• Code: .rodata +16 B, .text.nx_vsyslog +132 B, .text.vsyslog +20 B (+0.2%, 354,256 B)
• Data: .bss.g_kthread_group +40 B (+0.1%, 77,599 B)

qemu-intel64

• Code: .text +640 B (+0.0%, 8,687,623 B)
• Data: .bss +64 B, .rodata +32 B (+0.1%, 125,113 B)

rx65n-rsk2mb

• RAM: .bss +48 B (+0.7%, 6,586 B / 262,144 B, total: 3% used)
• ROM: .rodata +20 B, .text +288 B (+0.3%, 90,220 B / 2,097,152 B, total: 4% used)

s698pm-dkit

• Code: .text +1,136 B (+0.3%, 383,728 B)
• Data: .bss +48 B (+0.4%, 11,867 B)

stm32-nucleo-f103rb

• flash: .text +312 B (+0.9%, 34,432 B / 131,072 B, total: 26% used)
• sram: .bss +40 B (+1.7%, 2,428 B / 20,480 B, total: 12% used)

---

_{Catch memory regressions in CI with MemBrowse. Free for open source.}