I also added KCMP_FILEOBJ which compares the objects files referring to, instead of just files itself as required by KCMP_FILE semantic.
Details
- Reviewers
brooks markj - Commits
- rGf006524d6d69: kcmp(2): implement for procdesc
rG5c41d888de1a: kcmp(2): implement for devfs files
rG41fb6dc3d4df: kcmp(2): implement for linuxkpi cdevs
rGf04220c1b064: kcmp(2): implement for vnode files
rGf28526e9466c: kcmp(2): implement for generic file types
rG211bdd601ee5: Add kcmp(2) userspace bits
rGd8decc9ae31a: Add kcmp(2) kernel bits
rG168c7580c632: file: add fo_cmp method
rG58d317169834: Add fget_remote()
rG24fee9771e0b: sys/file.h: style
Diff Detail
- Repository
- rG FreeBSD src repository
- Lint
Lint Skipped - Unit
Tests Skipped
Event Timeline
include/unistd.h | ||
---|---|---|
523 | looks like there's stray (); at the end of the prototype |
I have a few compile errors :
/usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2193:36: error: declaration of 'struct kcmp_args' will not be visible outside of this function [-Werror,-Wvisibility] 2193 | sys_kcmp(struct thread *td, struct kcmp_args *uap) | ^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2195:27: error: incomplete definition of type 'struct kcmp_args' 2195 | return (kern_kcmp(td, uap->pid1, uap->pid2, uap->type, | ~~~^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2193:36: note: forward declaration of 'struct kcmp_args' 2193 | sys_kcmp(struct thread *td, struct kcmp_args *uap) | ^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2195:38: error: incomplete definition of type 'struct kcmp_args' 2195 | return (kern_kcmp(td, uap->pid1, uap->pid2, uap->type, | ~~~^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2193:36: note: forward declaration of 'struct kcmp_args' 2193 | sys_kcmp(struct thread *td, struct kcmp_args *uap) | ^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2195:49: error: incomplete definition of type 'struct kcmp_args' 2195 | return (kern_kcmp(td, uap->pid1, uap->pid2, uap->type, | ~~~^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2193:36: note: forward declaration of 'struct kcmp_args' 2193 | sys_kcmp(struct thread *td, struct kcmp_args *uap) | ^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2196:9: error: incomplete definition of type 'struct kcmp_args' 2196 | uap->idx1, uap->idx2)); | ~~~^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2193:36: note: forward declaration of 'struct kcmp_args' 2193 | sys_kcmp(struct thread *td, struct kcmp_args *uap) | ^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2196:20: error: incomplete definition of type 'struct kcmp_args' 2196 | uap->idx1, uap->idx2)); | ~~~^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2193:36: note: forward declaration of 'struct kcmp_args' 2193 | sys_kcmp(struct thread *td, struct kcmp_args *uap) | ^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2193:1: error: no previous prototype for function 'sys_kcmp' [-Werror,-Wmissing-prototypes] 2193 | sys_kcmp(struct thread *td, struct kcmp_args *uap) | ^ /usr/home/manu/Work/freebsd/src/upstream_main/sys/kern/sys_generic.c:2192:1: note: declare 'static' if the function is not intended to be used outside of this translation unit 2192 | int | ^ | static 7 errors generated. --- sys_generic.o --- *** [sys_generic.o] Error code 1
You need to do make sysent at src/ top level before starting buildworld or buildkernel.
Ah yes sorry, always forget that for new syscalls.
Will try again today and will cook a patch for mesa to test it.
Seems to work fine with https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27184 applied.
This part of FreeBSD is a bit to hard for me to review though :)
FWIW, Chromium also complains about the lack of kcmp.
sys/kern/sys_generic.c | ||
---|---|---|
2089 | style(9) wants a blank line here |
This is missing a manual page.
sys/compat/linuxkpi/common/src/linux_compat.c | ||
---|---|---|
1729 | Shouldn't we compare f_ops as well? DTYPE_DEV is "device-specific fd type", not "has a linux cdev". | |
1734 | fd_install() in the LinuxKPI installs a DTYPE_DEV file without setting f_cdev. I'm not sure if the comparison is right. | |
sys/kern/sys_generic.c | ||
2093 | Why CANDEBUG and not just CANSEE? I'd expect the check to be the same as what sysctl_kern_file() uses. | |
2124 | Rather than exporting fdhold() and fddrop(), I'd make this a function in kern_descrip.c, maybe fget_proc(). There is nothing specific to kcmp() here. | |
2130 | Why is it "unsigned long" instead of "int"? I understand that Linux defines it that way, but it's weird. Ah, for some uses, one of the argument is a pointer. Can we use uintptr_t instead then? I believe it's compatible for the cases we care about, and it'll let you remove some casts below. | |
sys/sys/file.h | ||
502 | Why EBADF instead of ENODEV? |
sys/kern/sys_generic.c | ||
---|---|---|
2130 | Agree on uintptr_t for idx1 and idx2. It's a shame that Linux defines them to be ignored rather than 0 when unused, but so it goes. |
FWIW, this Debian code search query suggests that we could enforce not passing nonsense values to idx[12] when they are ignored. That's probably not a good idea though. https://codesearch.debian.net/search?q=SYS_kcmp+-path%3A.*%2Fmusl%2F.*+-path%3A.*%2Flibc%2F.*+-path%3A.*%2Flinux_like%2F.*+-file%3Apo+-file%3Apot&literal=0
Indeed. Problem is that I did read the man page to write the implementation, which raises the question of 'clean room' text content. It is similar to membarrier(2).
sys/compat/linuxkpi/common/src/linux_compat.c | ||
---|---|---|
1729 | Nothing else sets type to DTYPE_DEV. I think we should rename the file type to e.g. DTYPE_LINUXKPI_DEV. | |
1734 | I do not follow. finit() in Linuxkpi always set f_data to filp, which is populated with f_cdev pointing to lcdev. | |
sys/kern/sys_generic.c | ||
2093 | Linux kcmp(2) manpage states: Permission to employ kcmp() is governed by ptrace access mode PTRACE_MODE_READ_REALCREDS checks against both pid1 and pid2; see ptrace(2). | |
sys/sys/file.h | ||
502 | fo_cmp() should be NULL only for badfileops, but I do not see an importance there. |
fget_remote()
unsigned long->uintptr_t idx
More minor tweaks like error codes returned
I didn't really look at the Linux man page, so took the liberty of writing one: https://reviews.freebsd.org/D43562
sys/compat/linuxkpi/common/src/linux_compat.c | ||
---|---|---|
1729 | That's fine with me. | |
1734 | I think I misread. It appears possible for f_cdev to be NULL, but that's ok. | |
sys/sys/file.h | ||
502 | Well, it distinguishes the "fd is not valid" case from "fd is not a supported type" case. If we think the latter case is impossible, then there should be an assertion instead. |