summaryrefslogtreecommitdiffstats
path: root/third_party/rust/redox_syscall/src/call.rs
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-28 14:29:10 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-28 14:29:10 +0000
commit2aa4a82499d4becd2284cdb482213d541b8804dd (patch)
treeb80bf8bf13c3766139fbacc530efd0dd9d54394c /third_party/rust/redox_syscall/src/call.rs
parentInitial commit. (diff)
downloadfirefox-upstream.tar.xz
firefox-upstream.zip
Adding upstream version 86.0.1.upstream/86.0.1upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'third_party/rust/redox_syscall/src/call.rs')
-rw-r--r--third_party/rust/redox_syscall/src/call.rs393
1 files changed, 393 insertions, 0 deletions
diff --git a/third_party/rust/redox_syscall/src/call.rs b/third_party/rust/redox_syscall/src/call.rs
new file mode 100644
index 0000000000..8b56541419
--- /dev/null
+++ b/third_party/rust/redox_syscall/src/call.rs
@@ -0,0 +1,393 @@
+use super::arch::*;
+use super::data::{Map, SigAction, Stat, StatVfs, TimeSpec};
+use super::error::Result;
+use super::number::*;
+
+use core::{mem, ptr};
+
+// Signal restorer
+extern "C" fn restorer() -> ! {
+ sigreturn().unwrap();
+ unreachable!();
+}
+
+/// Set the end of the process's heap
+///
+/// When `addr` is `0`, this function will return the current break.
+///
+/// When `addr` is nonzero, this function will attempt to set the end of the process's
+/// heap to `addr` and return the new program break. The new program break should be
+/// checked by the allocator, it may not be exactly `addr`, as it may be aligned to a page
+/// boundary.
+///
+/// On error, `Err(ENOMEM)` will be returned indicating that no memory is available
+pub unsafe fn brk(addr: usize) -> Result<usize> {
+ syscall1(SYS_BRK, addr)
+}
+
+/// Change the process's working directory
+///
+/// This function will attempt to set the process's working directory to `path`, which can be
+/// either a relative, scheme relative, or absolute path.
+///
+/// On success, `Ok(0)` will be returned. On error, one of the following errors will be returned.
+///
+/// # Errors
+///
+/// * `EACCES` - permission is denied for one of the components of `path`, or `path`
+/// * `EFAULT` - `path` does not point to the process's addressible memory
+/// * `EIO` - an I/O error occurred
+/// * `ENOENT` - `path` does not exit
+/// * `ENOTDIR` - `path` is not a directory
+pub fn chdir<T: AsRef<[u8]>>(path: T) -> Result<usize> {
+ unsafe { syscall2(SYS_CHDIR, path.as_ref().as_ptr() as usize, path.as_ref().len()) }
+}
+
+#[deprecated(
+ since = "0.1.55",
+ note = "use fchmod instead"
+)]
+pub fn chmod<T: AsRef<[u8]>>(path: T, mode: usize) -> Result<usize> {
+ unsafe { syscall3(SYS_CHMOD, path.as_ref().as_ptr() as usize, path.as_ref().len(), mode) }
+}
+
+/// Produce a fork of the current process, or a new process thread
+pub unsafe fn clone(flags: usize) -> Result<usize> {
+ syscall1_clobber(SYS_CLONE, flags)
+}
+
+/// Close a file
+pub fn close(fd: usize) -> Result<usize> {
+ unsafe { syscall1(SYS_CLOSE, fd) }
+}
+
+/// Get the current system time
+pub fn clock_gettime(clock: usize, tp: &mut TimeSpec) -> Result<usize> {
+ unsafe { syscall2(SYS_CLOCK_GETTIME, clock, tp as *mut TimeSpec as usize) }
+}
+
+/// Copy and transform a file descriptor
+pub fn dup(fd: usize, buf: &[u8]) -> Result<usize> {
+ unsafe { syscall3(SYS_DUP, fd, buf.as_ptr() as usize, buf.len()) }
+}
+
+/// Copy and transform a file descriptor
+pub fn dup2(fd: usize, newfd: usize, buf: &[u8]) -> Result<usize> {
+ unsafe { syscall4(SYS_DUP2, fd, newfd, buf.as_ptr() as usize, buf.len()) }
+}
+
+/// Exit the current process
+pub fn exit(status: usize) -> Result<usize> {
+ unsafe { syscall1(SYS_EXIT, status) }
+}
+
+/// Change file permissions
+pub fn fchmod(fd: usize, mode: u16) -> Result<usize> {
+ unsafe { syscall2(SYS_FCHMOD, fd, mode as usize) }
+
+}
+
+/// Change file ownership
+pub fn fchown(fd: usize, uid: u32, gid: u32) -> Result<usize> {
+ unsafe { syscall3(SYS_FCHOWN, fd, uid as usize, gid as usize) }
+
+}
+
+/// Change file descriptor flags
+pub fn fcntl(fd: usize, cmd: usize, arg: usize) -> Result<usize> {
+ unsafe { syscall3(SYS_FCNTL, fd, cmd, arg) }
+}
+
+/// Replace the current process with a new executable
+pub fn fexec(fd: usize, args: &[[usize; 2]], vars: &[[usize; 2]]) -> Result<usize> {
+ unsafe { syscall5(SYS_FEXEC, fd, args.as_ptr() as usize, args.len(), vars.as_ptr() as usize, vars.len()) }
+}
+
+/// Map a file into memory
+pub unsafe fn fmap(fd: usize, map: &Map) -> Result<usize> {
+ syscall3(SYS_FMAP, fd, map as *const Map as usize, mem::size_of::<Map>())
+}
+
+/// Unmap a memory-mapped file
+pub unsafe fn funmap(addr: usize) -> Result<usize> {
+ syscall1(SYS_FUNMAP, addr)
+}
+
+/// Retrieve the canonical path of a file
+pub fn fpath(fd: usize, buf: &mut [u8]) -> Result<usize> {
+ unsafe { syscall3(SYS_FPATH, fd, buf.as_mut_ptr() as usize, buf.len()) }
+}
+
+/// Rename a file
+pub fn frename<T: AsRef<[u8]>>(fd: usize, path: T) -> Result<usize> {
+ unsafe { syscall3(SYS_FRENAME, fd, path.as_ref().as_ptr() as usize, path.as_ref().len()) }
+}
+
+/// Get metadata about a file
+pub fn fstat(fd: usize, stat: &mut Stat) -> Result<usize> {
+ unsafe { syscall3(SYS_FSTAT, fd, stat as *mut Stat as usize, mem::size_of::<Stat>()) }
+}
+
+/// Get metadata about a filesystem
+pub fn fstatvfs(fd: usize, stat: &mut StatVfs) -> Result<usize> {
+ unsafe { syscall3(SYS_FSTATVFS, fd, stat as *mut StatVfs as usize, mem::size_of::<StatVfs>()) }
+}
+
+/// Sync a file descriptor to its underlying medium
+pub fn fsync(fd: usize) -> Result<usize> {
+ unsafe { syscall1(SYS_FSYNC, fd) }
+}
+
+/// Truncate or extend a file to a specified length
+pub fn ftruncate(fd: usize, len: usize) -> Result<usize> {
+ unsafe { syscall2(SYS_FTRUNCATE, fd, len) }
+}
+
+// Change modify and/or access times
+pub fn futimens(fd: usize, times: &[TimeSpec]) -> Result<usize> {
+ unsafe { syscall3(SYS_FUTIMENS, fd, times.as_ptr() as usize, times.len() * mem::size_of::<TimeSpec>()) }
+}
+
+/// Fast userspace mutex
+pub unsafe fn futex(addr: *mut i32, op: usize, val: i32, val2: usize, addr2: *mut i32)
+ -> Result<usize> {
+ syscall5(SYS_FUTEX, addr as usize, op, (val as isize) as usize, val2, addr2 as usize)
+}
+
+/// Get the current working directory
+pub fn getcwd(buf: &mut [u8]) -> Result<usize> {
+ unsafe { syscall2(SYS_GETCWD, buf.as_mut_ptr() as usize, buf.len()) }
+}
+
+/// Get the effective group ID
+pub fn getegid() -> Result<usize> {
+ unsafe { syscall0(SYS_GETEGID) }
+}
+
+/// Get the effective namespace
+pub fn getens() -> Result<usize> {
+ unsafe { syscall0(SYS_GETENS) }
+}
+
+/// Get the effective user ID
+pub fn geteuid() -> Result<usize> {
+ unsafe { syscall0(SYS_GETEUID) }
+}
+
+/// Get the current group ID
+pub fn getgid() -> Result<usize> {
+ unsafe { syscall0(SYS_GETGID) }
+}
+
+/// Get the current namespace
+pub fn getns() -> Result<usize> {
+ unsafe { syscall0(SYS_GETNS) }
+}
+
+/// Get the current process ID
+pub fn getpid() -> Result<usize> {
+ unsafe { syscall0(SYS_GETPID) }
+}
+
+/// Get the process group ID
+pub fn getpgid(pid: usize) -> Result<usize> {
+ unsafe { syscall1(SYS_GETPGID, pid) }
+}
+
+/// Get the parent process ID
+pub fn getppid() -> Result<usize> {
+ unsafe { syscall0(SYS_GETPPID) }
+}
+
+/// Get the current user ID
+pub fn getuid() -> Result<usize> {
+ unsafe { syscall0(SYS_GETUID) }
+}
+
+/// Set the I/O privilege level
+///
+/// # Errors
+///
+/// * `EPERM` - `uid != 0`
+/// * `EINVAL` - `level > 3`
+pub unsafe fn iopl(level: usize) -> Result<usize> {
+ syscall1(SYS_IOPL, level)
+}
+
+/// Send a signal `sig` to the process identified by `pid`
+pub fn kill(pid: usize, sig: usize) -> Result<usize> {
+ unsafe { syscall2(SYS_KILL, pid, sig) }
+}
+
+/// Create a link to a file
+pub unsafe fn link(old: *const u8, new: *const u8) -> Result<usize> {
+ syscall2(SYS_LINK, old as usize, new as usize)
+}
+
+/// Seek to `offset` bytes in a file descriptor
+pub fn lseek(fd: usize, offset: isize, whence: usize) -> Result<usize> {
+ unsafe { syscall3(SYS_LSEEK, fd, offset as usize, whence) }
+}
+
+/// Make a new scheme namespace
+pub fn mkns(schemes: &[[usize; 2]]) -> Result<usize> {
+ unsafe { syscall2(SYS_MKNS, schemes.as_ptr() as usize, schemes.len()) }
+}
+
+/// Change mapping flags
+pub unsafe fn mprotect(addr: usize, size: usize, flags: usize) -> Result<usize> {
+ syscall3(SYS_MPROTECT, addr, size, flags)
+}
+
+/// Sleep for the time specified in `req`
+pub fn nanosleep(req: &TimeSpec, rem: &mut TimeSpec) -> Result<usize> {
+ unsafe { syscall2(SYS_NANOSLEEP, req as *const TimeSpec as usize,
+ rem as *mut TimeSpec as usize) }
+}
+
+/// Open a file
+pub fn open<T: AsRef<[u8]>>(path: T, flags: usize) -> Result<usize> {
+ unsafe { syscall3(SYS_OPEN, path.as_ref().as_ptr() as usize, path.as_ref().len(), flags) }
+}
+
+/// Allocate pages, linearly in physical memory
+///
+/// # Errors
+///
+/// * `EPERM` - `uid != 0`
+/// * `ENOMEM` - the system has run out of available memory
+pub unsafe fn physalloc(size: usize) -> Result<usize> {
+ syscall1(SYS_PHYSALLOC, size)
+}
+
+/// Free physically allocated pages
+///
+/// # Errors
+///
+/// * `EPERM` - `uid != 0`
+pub unsafe fn physfree(physical_address: usize, size: usize) -> Result<usize> {
+ syscall2(SYS_PHYSFREE, physical_address, size)
+}
+
+/// Map physical memory to virtual memory
+///
+/// # Errors
+///
+/// * `EPERM` - `uid != 0`
+pub unsafe fn physmap(physical_address: usize, size: usize, flags: usize) -> Result<usize> {
+ syscall3(SYS_PHYSMAP, physical_address, size, flags)
+}
+
+/// Unmap previously mapped physical memory
+///
+/// # Errors
+///
+/// * `EPERM` - `uid != 0`
+/// * `EFAULT` - `virtual_address` has not been mapped
+pub unsafe fn physunmap(virtual_address: usize) -> Result<usize> {
+ syscall1(SYS_PHYSUNMAP, virtual_address)
+}
+
+/// Create a pair of file descriptors referencing the read and write ends of a pipe
+pub fn pipe2(fds: &mut [usize; 2], flags: usize) -> Result<usize> {
+ unsafe { syscall2(SYS_PIPE2, fds.as_ptr() as usize, flags) }
+}
+
+/// Read from a file descriptor into a buffer
+pub fn read(fd: usize, buf: &mut [u8]) -> Result<usize> {
+ unsafe { syscall3(SYS_READ, fd, buf.as_mut_ptr() as usize, buf.len()) }
+}
+
+/// Remove a directory
+pub fn rmdir<T: AsRef<[u8]>>(path: T) -> Result<usize> {
+ unsafe { syscall2(SYS_RMDIR, path.as_ref().as_ptr() as usize, path.as_ref().len()) }
+}
+
+/// Set the process group ID
+pub fn setpgid(pid: usize, pgid: usize) -> Result<usize> {
+ unsafe { syscall2(SYS_SETPGID, pid, pgid) }
+}
+
+/// Set the current process group IDs
+pub fn setregid(rgid: usize, egid: usize) -> Result<usize> {
+ unsafe { syscall2(SYS_SETREGID, rgid, egid) }
+}
+
+/// Make a new scheme namespace
+pub fn setrens(rns: usize, ens: usize) -> Result<usize> {
+ unsafe { syscall2(SYS_SETRENS, rns, ens) }
+}
+
+/// Set the current process user IDs
+pub fn setreuid(ruid: usize, euid: usize) -> Result<usize> {
+ unsafe { syscall2(SYS_SETREUID, ruid, euid) }
+}
+
+/// Set up a signal handler
+pub fn sigaction(sig: usize, act: Option<&SigAction>, oldact: Option<&mut SigAction>) -> Result<usize> {
+ unsafe { syscall4(SYS_SIGACTION, sig,
+ act.map(|x| x as *const _).unwrap_or_else(ptr::null) as usize,
+ oldact.map(|x| x as *mut _).unwrap_or_else(ptr::null_mut) as usize,
+ restorer as usize) }
+}
+
+/// Get and/or set signal masks
+pub fn sigprocmask(how: usize, set: Option<&[u64; 2]>, oldset: Option<&mut [u64; 2]>) -> Result<usize> {
+ unsafe { syscall3(SYS_SIGPROCMASK, how,
+ set.map(|x| x as *const _).unwrap_or_else(ptr::null) as usize,
+ oldset.map(|x| x as *mut _).unwrap_or_else(ptr::null_mut) as usize) }
+}
+
+// Return from signal handler
+pub fn sigreturn() -> Result<usize> {
+ unsafe { syscall0(SYS_SIGRETURN) }
+}
+
+/// Set the file mode creation mask
+pub fn umask(mask: usize) -> Result<usize> {
+ unsafe { syscall1(SYS_UMASK, mask) }
+}
+
+/// Remove a file
+pub fn unlink<T: AsRef<[u8]>>(path: T) -> Result<usize> {
+ unsafe { syscall2(SYS_UNLINK, path.as_ref().as_ptr() as usize, path.as_ref().len()) }
+}
+
+/// Convert a virtual address to a physical one
+///
+/// # Errors
+///
+/// * `EPERM` - `uid != 0`
+pub unsafe fn virttophys(virtual_address: usize) -> Result<usize> {
+ syscall1(SYS_VIRTTOPHYS, virtual_address)
+}
+
+/// Check if a child process has exited or received a signal
+pub fn waitpid(pid: usize, status: &mut usize, options: usize) -> Result<usize> {
+ unsafe { syscall3(SYS_WAITPID, pid, status as *mut usize as usize, options) }
+}
+
+/// Write a buffer to a file descriptor
+///
+/// The kernel will attempt to write the bytes in `buf` to the file descriptor `fd`, returning
+/// either an `Err`, explained below, or `Ok(count)` where `count` is the number of bytes which
+/// were written.
+///
+/// # Errors
+///
+/// * `EAGAIN` - the file descriptor was opened with `O_NONBLOCK` and writing would block
+/// * `EBADF` - the file descriptor is not valid or is not open for writing
+/// * `EFAULT` - `buf` does not point to the process's addressible memory
+/// * `EIO` - an I/O error occurred
+/// * `ENOSPC` - the device containing the file descriptor has no room for data
+/// * `EPIPE` - the file descriptor refers to a pipe or socket whose reading end is closed
+pub fn write(fd: usize, buf: &[u8]) -> Result<usize> {
+ unsafe { syscall3(SYS_WRITE, fd, buf.as_ptr() as usize, buf.len()) }
+}
+
+/// Yield the process's time slice to the kernel
+///
+/// This function will return Ok(0) on success
+pub fn sched_yield() -> Result<usize> {
+ unsafe { syscall0(SYS_YIELD) }
+}