using O_DIRECT
A message that is not in the archive (sent off-list or not yet received): [email protected]
I have written a small program to use O_DIRECT (attached), after applying your patch o_direct-14 to kernel 2.4.9. Opening the file with O_DIRECT is successful, but attempts to write to the fd return EINVAL.
Am I doing something wrong? Should I have to recompile glibc too?
(0)<root@draal:/usr/src/mjpeg_play> gcc -o testopen testopen.c (0)<root@draal:/usr/src/mjpeg_play> strace ./testopen > /dev/null execve("./testopen", ["./testopen"], [/* 54 vars */]) = 0 uname({sys="Linux", node="draal.physics.wisc.edu", ...}) = 0
brk(0) = 0x120010d90
open("/etc/ld.so.preload", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/etc/ld.so.cache", O_RDONLY) = 3fstat(3, {st_mode=S_IFREG|0644, st_size=62527, ...}) = 0mmap(NULL, 62527, PROT_READ, MAP_PRIVATE, 3, 0) = 0x20000030000
close(3) = 0open("/lib/libc.so.6.1", O_RDONLY) = 3
read(3, "\177ELF\2\1\1\0\0\0\0\0\0\0\0\0\3\0&\220\1\0\0\0`\201\3"..., 1024) = 1024 fstat(3, {st_mode=S_IFREG|0755, st_size=9653762, ...}) = 0mmap(NULL, 1558648, PROT_READ|PROT_EXEC, MAP_PRIVATE, 3, 0) = 0x20000040000 mprotect(0x2000019a000, 141432, PROT_NONE) = 0
mmap(0x200001a0000, 98304, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_FIXED, 3, 0x150000) = 0x200001a0000 mmap(0x200001b8000, 18552, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0x200001b8000 close(3) = 0 munmap(0x20000030000, 62527) = 0 getxpid() = 2387
open("testopen.txt", O_RDWR|O_CREAT|O_SYNC|0x100000, 0644) = 3fstat(1, {st_mode=S_IFCHR|0666, st_rdev=makedev(1, 3), ...}) = 0 mmap(NULL, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x2000001e000ioctl(1, TCGETS, 0x11fffea50) = -1 ENOTTY (Inappropriate ioctl for device)write(3, "Welcome to the winter of my disc"..., 40) = -1 EINVAL (Invalid argument)
close(3) = 0write(1, "Got fd 3.\nUnable to write to fil"..., 77) = 77munmap(0x2000001e000, 8192) = 0
exit(0) = ?
Cheers,
-- Bob
Bob McElrath ([email protected]) Univ. of Wisconsin at Madison, Department of Physics
/* Test opening of a file with O_DIRECT */
#include <asm/types.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <linux/types.h>
#include <asm/fcntl.h> /* Note: <fcntl.h> DOES NOT HAVE O_DIRECT! */
/* <fcntl.h> is a glibc header... */#include <errno.h>
#include <stdio.h>
int main(void) {
char message[] = "Welcome to the winter of my discontent.\n";
int fd = open("testopen.txt", O_RDWR|O_CREAT|O_SYNC|O_DIRECT, S_IRUSR|S_IRGRP|S_IROTH|S_IWUSR);
int len;
if(fd < 0) {
printf("Unable to open file, errno is %d.\n", errno);
} else {
printf("Got fd %d.\n", fd);
if((len = write(fd, message, strlen(message))) < 0) {
printf("Unable to write to file, len is %d, errno is %d.\n", len, errno);
} else {
printf("%d bytes written to file.\n", len);
}
}
close(fd);
printf("End of program...\n");
return 0;
}Quoted 5 lines · Bob McElrath
I have written a small program to use
O_DIRECT(attached), after applying your patch o_direct-14 to kernel 2.4.9. Opening the file withO_DIRECTis successful, but attempts to write to the fd return EINVAL.Am I doing something wrong? Should I have to recompile glibc too?
eh, the alignment and size of the buffer have basically the same restrictions of running read/write on a raw device, with the only difference that for rawio the granularity is the hardblocksize, while for O_DIRECT the granularity is the softblocksize of the filesystem.
We'll have to relax the granularity of the I/O down to the hardblocksize for O_DIRECT too eventually.
Andrea
Quoted 11 lines · Andrea Arcangeli [[email protected]] wrote:
Quoted 5 lines · Bob McElrath
I have written a small program to use
O_DIRECT(attached), after applying your patch o_direct-14 to kernel 2.4.9. Opening the file withO_DIRECTis successful, but attempts to write to the fd return EINVAL.Am I doing something wrong? Should I have to recompile glibc too?
eh, the alignment and size of the buffer have basically the same restrictions of running read/write on a raw device, with the only difference that for rawio the granularity is the hardblocksize, while for
O_DIRECTthe granularity is the softblocksize of the filesystem.
Ah, thanks, it's working now. Makes sense that it should use blocksize granularity...
Quoted 2 lines
We'll have to relax the granularity of the I/O down to the hardblocksize for
O_DIRECTtoo eventually.
I don't really care about the granularity. The bigger the better, actually, since I'm streaming data to disk quickly. ;)
Thanks,
-- Bob
Bob McElrath ([email protected]) Univ. of Wisconsin at Madison, Department of Physics
Quoted 2 lines · Bob McElrath
I don't really care about the granularity. The bigger the better, actually, since I'm streaming data to disk quickly. ;)
Yes ;).
Andrea