huojianchaoyy opened a new pull request, #21787: URL: https://github.com/apache/nuttx/pull/21787
*Note: Please adhere to [Contributing Guidelines](https://github.com/apache/nuttx/blob/master/CONTRIBUTING.md).* ## Summary Fixes in the syslog write path and in the RPMsg syslog server: - **drivers/syslog: fix syslog write busy loop when a channel is full**: `syslog_write_foreach()` treated a zero-length channel write the same as a successful one. When a channel buffer is full its `sc_write()` returns 0, but the CR/LF path still emitted the trailing CR/LF and advanced `nwritten` past data that was never stored, and the tail path counted the blocked channel as having written 0 bytes and let it update `nwritten_max`. With every channel full the function returned 0, which the caller takes as "no progress" and retries, spinning the CPU. A return of 0 is now handled like a write error. - **drivers/syslog: keep the partial rpmsg line in a fixed buffer**: `syslog_rpmsg_ept_cb()` kept the unterminated tail of each transfer in a buffer that it grew with `kmm_realloc()` from the endpoint callback. The tail now goes to a `SYSLOG_RPMSG_MAXLEN` buffer inside `struct syslog_rpmsg_server_s`, so the callback never allocates. The part of a partial line that does not fit is dropped; `syslog_rpmsg_write_internal()` could merge at most `SYSLOG_RPMSG_MAXLEN` bytes with the rest of the line anyway. - **drivers/syslog: drop the NUL from the color reset write**: `sizeof("\e[0m")` counts the string terminator, so with `CONFIG_SYSLOG_COLOR_OUTPUT` every message ended with a 0x00 byte after its newline. Terminals ignore it, but tools that read the log byte by byte see the NUL at the start of the next line. - **drivers/syslog: check the CRLF write result**: the result of the CR/LF write was never used, so a channel filling up between the payload write and the CR/LF write had its failure discarded while `nwritten` still claimed the '\n' was consumed. The '\n' is now accounted only when its CR/LF expansion was stored; otherwise it is left to the tail write. - **drivers/syslog: keep the CR of an empty line**: with the first commit, the write of the empty text in front of the '\n' of an empty line (or of a '\n' at the start of the buffer) returns 0 and counts as a full channel, so that '\n' went out without its CR. The empty text is no longer written. ## Impact - A full syslog channel no longer makes `syslog_write_foreach()` spin, and no longer gets a CR/LF for text it did not store. - The RPMsg syslog server does not allocate in its endpoint callback. `struct syslog_rpmsg_server_s` grows by `SYSLOG_RPMSG_MAXLEN` (256) bytes per remote CPU, and a partial line longer than that is cut. - `CONFIG_SYSLOG_COLOR_OUTPUT`: no NUL byte in the output any more. - No API or Kconfig change. ## Testing Host: Ubuntu 22.04 x86_64, GCC 13.4.0, CMake 3.23 + Ninja. Base: master dc9481383ed. Targets: `sim:nsh` (it has the default `CONFIG_SYSLOG_CRLF=y`), `sim:nsh` with `CONFIG_SYSLOG_COLOR_OUTPUT=y`, and `sim:rpserver` + `sim:rpproxy`. `./tools/checkpatch.sh -c -u -m -g HEAD~5..HEAD` passes. A small test program (not part of this PR) calls `syslog(LOG_INFO, "slogA\n\nslogB\n")`, `syslog(LOG_INFO, "\nslogC\n")` and `syslog(LOG_INFO, "slogD\n")`. The bytes on the console from `slogA` to `slogD`: | Build | Output | |---|---| | master | `slogA\r\n\r\nslogB\r\n\r\nslogC\r\nslogD` | | the first four commits only | `slogA\r\n\nslogB\r\n\nslogC\r\nslogD` | | this PR | `slogA\r\n\r\nslogB\r\n\r\nslogC\r\nslogD` | With `CONFIG_SYSLOG_COLOR_OUTPUT=y`, master writes 3 NUL bytes in that range, this PR none: ``` master: slogA\r\n\r\nslogB\r\n\x1b[0m\x00\x1b[0m\r\nslogC\r\n\x1b[0m\x00\x1b[0mslogD this PR: slogA\r\n\r\nslogB\r\n\x1b[0m\x1b[0m\r\nslogC\r\n\x1b[0m\x1b[0mslogD ``` On `sim:rpserver` + `sim:rpproxy`, the same program run on the proxy (through `cu` on the server) still shows up on the server console through the RPMsg syslog server: ``` proxy> upstest syslog syslog: done proxy> [ 7.010636] [proxy] slogA slogB [ 7.010710] [proxy] slogC [ 7.010777] [proxy] slogD ``` A channel that really fills up (the case of the first and fourth commits) was not reproduced here. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
