fs: Add utimens and lutimens

[xiaoxiang781216]

Summary

https://pubs.opengroup.org/onlinepubs/9699919799/functions/futimens.html
https://www.daemon-systems.org/man/utimens.2.html

Impact

New API, follow up the change from #4153

Testing

[gustavonihei]

I am a bit confused here. I am not familiar with these APIs.
According to the latest release of the OpenGroup standard, neither utimens nor lutimens are standardized functions.
I saw that utimes has been moved the XSI and been somewhat deprecated.
But, the standard defines utimensat as an equivalent function for utimes.
Isn't utimensat the implementation you intended to add instead of utimens?

[xiaoxiang781216]

I am a bit confused here. I am not familiar with these APIs.
According to the latest release of the OpenGroup standard, neither utimens nor lutimens are standardized functions.
I saw that utimes has been moved the XSI and been somewhat deprecated.
But, the standard defines utimensat as an equivalent function for utimes.
Isn't utimensat the implementation you intended to add instead of utimens?

Yes, utimensat is the final goal: @Donny9 is preparing a patch series to support utimensat(and other similar function e.g. openat, unlinkat...). This PR could be used before that patch finish. Once utimensat is ready, all these time related functions will simply forward to utimensat.

[xiaoxiang781216]

@gustavonihei I append a new patch to this PR, you can see from it that:

1. The sim implementation of hostfs call utimensat:
   d7631be#diff-4f28509098f7178398166fad417074658b353f7e72d7a4fddb93e917cb91e055R646
2. The rpmsg implmentation of hostfs call utimens:
   d7631be#diff-9cefb67caeab4d76ef979687e08c0faed73fc127bf2c5ac128fe63ee2164dbd3R779

Without utimens, it look ulgy to convert timespec to timeval and invoke utimes instead in the second implementation. On the other hand, when utimensat is ready, rpmsg's version will change to call utimensat too.

[gustavonihei]

@gustavonihei I append a new patch to this PR, you can see from it that:

1. The sim implementation of hostfs call utimensat:
   d7631be#diff-4f28509098f7178398166fad417074658b353f7e72d7a4fddb93e917cb91e055R646
2. The rpmsg implmentation of hostfs call utimens:
   d7631be#diff-9cefb67caeab4d76ef979687e08c0faed73fc127bf2c5ac128fe63ee2164dbd3R779

Without utimens, it look ulgy to convert timespec to timeval and invoke utimes instead in the second implementation. On the other hand, when utimensat is ready, rpmsg's version will change to call utimensat too.

Ok, I got your point, I think it makes sense.
So utimens and lutimens usage will later be restricted as just internal helper functions.
LGTM, but the CI is still failing on some compiler errors.

[xiaoxiang781216]

@gustavonihei I append a new patch to this PR, you can see from it that:

1. The sim implementation of hostfs call utimensat:
   d7631be#diff-4f28509098f7178398166fad417074658b353f7e72d7a4fddb93e917cb91e055R646
2. The rpmsg implmentation of hostfs call utimens:
   d7631be#diff-9cefb67caeab4d76ef979687e08c0faed73fc127bf2c5ac128fe63ee2164dbd3R779

Without utimens, it look ulgy to convert timespec to timeval and invoke utimes instead in the second implementation. On the other hand, when utimensat is ready, rpmsg's version will change to call utimensat too.

Ok, I got your point, I think it makes sense.
So utimens and lutimens usage will later be restricted as just internal helper functions.

Yes, it will be replace with utimensat.

LGTM, but the CI is still failing on some compiler errors.

It should OK, now.

[yamt]

I am a bit confused here. I am not familiar with these APIs.
According to the latest release of the OpenGroup standard, neither utimens nor lutimens are standardized functions.
I saw that utimes has been moved the XSI and been somewhat deprecated.
But, the standard defines utimensat as an equivalent function for utimes.
Isn't utimensat the implementation you intended to add instead of utimens?

Yes, utimensat is the final goal: @Donny9 is preparing a patch series to support utimensat(and other similar function e.g. openat, unlinkat...). This PR could be used before that patch finish. Once utimensat is ready, all these time related functions will simply forward to utimensat.

@Donny9
are you (or someone) still working on openat and friends?

[xiaoxiang781216]

yes, we have a draft huge patchset.

[xiaoxiang781216]

yes, we have a draft huge patchset.

[yamt]

yes, we have a draft huge patchset.

do you have a plan to submit it anytime soon?

[Donny9]

I'm free at the end of this month to organize this patch, there are other missions recently.

[yamt]

ok. thank you for an update.

[yamt]

btw, my interest in the openat stuff is for wasm-micro-runtime libc-wasi.

[xiaoxiang781216]

Yes, we use this runtime too.

[yamt]

Yes, we use this runtime too.

good to hear.
do you have patches for nuttx apps and product-mini/platforms/nuttx/wamr.mk etc as well?
(just asking because my colleagues seem plan to write them by themselves)

[xiaoxiang781216]

@no1wudi do you know?

[no1wudi]

We plan to support libc-wasi in future, but we don't have such patches now.

[yamt]

ok. thank you.