linux: avoid repeated zram ENOENT probes

[Komzpa]

Summary

This changes Linux zram accounting to enumerate existing /sys/block/zramN devices instead of probing zram0, zram1, and so on until a file open fails.

On systems without zram, htop otherwise repeats failed sysfs lookups for /sys/block/zram0/disksize and /sys/block/zram0/mm_stat on refresh. Enumerating /sys/block avoids that repeated ENOENT path and also naturally handles non-contiguous zram numbering.

The implementation follows the existing Linux compat style: each enumerated zram directory is opened relative to the /sys/block directory using the openat-style path, and disksize / mm_stat are read with Compat_readfileat rather than fopen.

Validation

• git diff --check
• ./autogen.sh && ./configure && make -j2

[BenBE]

Please avoid the fopen API (they cause trouble in high-memory-pressure situations.

Also, given we are already enumerating a directory using its fd, try to use the openat style APIs that help the kernel with resolving paths quicker.

Also take a look at the Compat_readfileat function to read the entire file contents at once.

Regarding building: Try to ./autogen.sh && ./configure && make, which should give a good starting point to check your changes.

[Komzpa]

Thanks, fixed. The zram scan no longer uses fopen for the sysfs files: it opens each enumerated zram directory relative to /sys/block using the openat-style compat path and reads disksize/mm_stat with Compat_readfileat.

I also reran the requested build path successfully: ./autogen.sh && ./configure && make -j2