load package_info when startup

[Admirepowered]

Only need to read and write once when starting up
开机只需读写一次，如果有大量排除可能会出问题

[Copilot]

Pull request overview

This PR introduces a new supercall to load APatch package_config into kernel state and triggers that load during Android init (second stage), aiming to avoid repeated reads/writes at runtime.

Changes:

• Added SUPERCALL_AP_LOAD_PACKAGE_CONFIG UAPI definition and a userspace wrapper (sc_ap_load_package_config).
• Wired the new command into the kernel supercall dispatcher (supercall.c).
• Implemented CSV parsing + application of config entries in kernel/patch/android/userd.c, and invoked it during init second stage.
• Updated .gitignore to ignore .claude.

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
user/supercall.h	Adds userspace inline wrapper for the new supercall command.
kernel/patch/include/uapi/scdefs.h	Defines new supercall command ID SUPERCALL_AP_LOAD_PACKAGE_CONFIG.
kernel/patch/common/supercall.c	Routes the new command to load_ap_package_config().
kernel/patch/android/userd.c	Adds file loading + CSV parsing logic and calls it during init second stage.
.gitignore	Ignores .claude.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

PR title/description mentions loading package_info, but the implementation loads /data/adb/ap/package_config via load_ap_package_config(). Please align the PR title/description (or the code) so it’s clear which file/format is being loaded at startup.

[Copilot]

*line_end = '\0' can write one byte past the allocated buffer when the last line does not end with a newline (i.e., line_end == content + len). Since kernel_read_file() allocates exactly len bytes, this is an out-of-bounds write. Consider ensuring the read buffer is NUL-terminated (e.g., copy into a len+1 buffer and set the final byte to 0) or guarding against line_end == content + len before writing.

[Copilot]

parse_csv_field() modifies the delimiter character before checking *end == ','. For empty fields (start points at ,), end == start, line 151 overwrites the comma with \0, and then the pointer-advance logic won’t see the comma and won’t advance *line_ptr, which can break parsing for lines with empty fields. Preserve the delimiter (or handle end == start / empty fields explicitly) before writing a terminator.

Suggested change

	// Remove trailing whitespace
	char *trim_end = end - 1;
	while (trim_end > start && (*trim_end == ' ' || *trim_end == '\t')) {
	trim_end--;
	}
	*(trim_end + 1) = '\0';
	// Update pointer position
	if (*end == ',') {
	*line_ptr = end + 1;
	} else {
	*line_ptr = end;
	// Preserve delimiter before modifying buffer
	{
	char delim = *end;
	// Remove trailing whitespace only if field is non-empty
	if (end > start) {
	char *trim_end = end - 1;
	while (trim_end > start && (*trim_end == ' ' || *trim_end == '\t')) {
	trim_end--;
	}
	*(trim_end + 1) = '\0';
	} else {
	// Empty field: terminate at start so caller sees an empty string
	*start = '\0';
	}
	// Update pointer position based on original delimiter
	if (delim == ',') {
	*line_ptr = end + 1;
	} else {
	*line_ptr = end;
	}

[Copilot]

Return values from kstrtoull() are ignored. If uid_str/to_uid_str fails to parse, the corresponding *_tmp stays 0, which can incorrectly apply config to UID 0 (and potentially grant privileges) instead of rejecting the line. Check kstrtoull() results (and range-check casts to uid_t/int) and skip/log invalid lines.

[Copilot]

sctx comes from the config file and can be arbitrarily long; passing it directly to su_add_allow_uid() is risky because su_add_allow_uid() copies a fixed SUPERCALL_SCONTEXT_LEN without guaranteeing NUL-termination. Consider truncating into a local buffer of SUPERCALL_SCONTEXT_LEN and forcing a trailing \0 before calling su_add_allow_uid() to avoid unterminated strings and log overreads.