/*
- * Copyright (C) 2004 Jeff Dike (jdike@addtoit.com)
+ * Copyright (C) 2004 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
* Licensed under the GPL
*/
-#include <stdlib.h>
#include <unistd.h>
+#include <sched.h>
#include <signal.h>
-#include <string.h>
#include <errno.h>
-#include <sched.h>
-#include <sys/syscall.h>
-#include "os.h"
-#include "helper.h"
+#include <sys/time.h>
+#include <asm/unistd.h>
#include "aio.h"
#include "init.h"
+#include "kern_constants.h"
+#include "kern_util.h"
+#include "os.h"
#include "user.h"
-#include "mode.h"
-
-static int aio_req_fd_r = -1;
-static int aio_req_fd_w = -1;
-static int update_aio(struct aio_context *aio, int res)
-{
- if(res < 0)
- aio->len = res;
- else if((res == 0) && (aio->type == AIO_READ)){
- /* This is the EOF case - we have hit the end of the file
- * and it ends in a partial block, so we fill the end of
- * the block with zeros and claim success.
- */
- memset(aio->data, 0, aio->len);
- aio->len = 0;
- }
- else if(res > 0){
- aio->len -= res;
- aio->data += res;
- aio->offset += res;
- return aio->len;
- }
-
- return 0;
-}
+struct aio_thread_req {
+ enum aio_type type;
+ int io_fd;
+ unsigned long long offset;
+ char *buf;
+ int len;
+ struct aio_context *aio;
+};
#if defined(HAVE_AIO_ABI)
#include <linux/aio_abi.h>
-/* If we have the headers, we are going to build with AIO enabled.
+/*
+ * If we have the headers, we are going to build with AIO enabled.
* If we don't have aio in libc, we define the necessary stubs here.
*/
static long io_setup(int n, aio_context_t *ctxp)
{
- return syscall(__NR_io_setup, n, ctxp);
+ return syscall(__NR_io_setup, n, ctxp);
}
static long io_submit(aio_context_t ctx, long nr, struct iocb **iocbpp)
{
- return syscall(__NR_io_submit, ctx, nr, iocbpp);
+ return syscall(__NR_io_submit, ctx, nr, iocbpp);
}
static long io_getevents(aio_context_t ctx_id, long min_nr, long nr,
- struct io_event *events, struct timespec *timeout)
+ struct io_event *events, struct timespec *timeout)
{
- return syscall(__NR_io_getevents, ctx_id, min_nr, nr, events, timeout);
+ return syscall(__NR_io_getevents, ctx_id, min_nr, nr, events, timeout);
}
#endif
-/* The AIO_MMAP cases force the mmapped page into memory here
+/*
+ * The AIO_MMAP cases force the mmapped page into memory here
* rather than in whatever place first touches the data. I used
* to do this by touching the page, but that's delicate because
* gcc is prone to optimizing that away. So, what's done here
* that it now backs the mmapped area.
*/
-static int do_aio(aio_context_t ctx, struct aio_context *aio)
+static int do_aio(aio_context_t ctx, enum aio_type type, int fd, char *buf,
+ int len, unsigned long long offset, struct aio_context *aio)
{
- struct iocb iocb, *iocbp = &iocb;
- char c;
- int err;
-
- iocb = ((struct iocb) { .aio_data = (unsigned long) aio,
- .aio_reqprio = 0,
- .aio_fildes = aio->fd,
- .aio_buf = (unsigned long) aio->data,
- .aio_nbytes = aio->len,
- .aio_offset = aio->offset,
- .aio_reserved1 = 0,
- .aio_reserved2 = 0,
- .aio_reserved3 = 0 });
-
- switch(aio->type){
- case AIO_READ:
- iocb.aio_lio_opcode = IOCB_CMD_PREAD;
- break;
- case AIO_WRITE:
- iocb.aio_lio_opcode = IOCB_CMD_PWRITE;
- break;
- case AIO_MMAP:
- iocb.aio_lio_opcode = IOCB_CMD_PREAD;
- iocb.aio_buf = (unsigned long) &c;
- iocb.aio_nbytes = sizeof(c);
- break;
- default:
- printk("Bogus op in do_aio - %d\n", aio->type);
- err = -EINVAL;
- goto out;
- }
-
- err = io_submit(ctx, 1, &iocbp);
- if(err > 0)
- err = 0;
- else
- err = -errno;
+ struct iocb *iocbp = & ((struct iocb) {
+ .aio_data = (unsigned long) aio,
+ .aio_fildes = fd,
+ .aio_buf = (unsigned long) buf,
+ .aio_nbytes = len,
+ .aio_offset = offset
+ });
+ char c;
+
+ switch (type) {
+ case AIO_READ:
+ iocbp->aio_lio_opcode = IOCB_CMD_PREAD;
+ break;
+ case AIO_WRITE:
+ iocbp->aio_lio_opcode = IOCB_CMD_PWRITE;
+ break;
+ case AIO_MMAP:
+ iocbp->aio_lio_opcode = IOCB_CMD_PREAD;
+ iocbp->aio_buf = (unsigned long) &c;
+ iocbp->aio_nbytes = sizeof(c);
+ break;
+ default:
+ printk(UM_KERN_ERR "Bogus op in do_aio - %d\n", type);
+ return -EINVAL;
+ }
- out:
- return err;
+ return (io_submit(ctx, 1, &iocbp) > 0) ? 0 : -errno;
}
+/* Initialized in an initcall and unchanged thereafter */
static aio_context_t ctx = 0;
static int aio_thread(void *arg)
{
- struct aio_thread_reply reply;
- struct aio_context *aio;
- struct io_event event;
- int err, n;
-
- signal(SIGWINCH, SIG_IGN);
-
- while(1){
- n = io_getevents(ctx, 1, 1, &event, NULL);
- if(n < 0){
- if(errno == EINTR)
- continue;
- printk("aio_thread - io_getevents failed, "
- "errno = %d\n", errno);
- }
- else {
- /* This is safe as we've just a pointer here. */
- aio = (struct aio_context *) (long) event.data;
- if(update_aio(aio, event.res)){
- do_aio(ctx, aio);
- continue;
- }
+ struct aio_thread_reply reply;
+ struct io_event event;
+ int err, n, reply_fd;
+
+ signal(SIGWINCH, SIG_IGN);
- reply = ((struct aio_thread_reply)
- { .data = aio,
- .err = aio->len });
- err = os_write_file(aio->reply_fd, &reply,
- sizeof(reply));
- if(err != sizeof(reply))
- printk("aio_thread - write failed, "
- "fd = %d, err = %d\n", aio->reply_fd,
- -err);
- }
- }
- return 0;
+ while (1) {
+ n = io_getevents(ctx, 1, 1, &event, NULL);
+ if (n < 0) {
+ if (errno == EINTR)
+ continue;
+ printk(UM_KERN_ERR "aio_thread - io_getevents failed, "
+ "errno = %d\n", errno);
+ }
+ else {
+ reply = ((struct aio_thread_reply)
+ { .data = (void *) (long) event.data,
+ .err = event.res });
+ reply_fd = ((struct aio_context *) reply.data)->reply_fd;
+ err = write(reply_fd, &reply, sizeof(reply));
+ if (err != sizeof(reply))
+ printk(UM_KERN_ERR "aio_thread - write failed, "
+ "fd = %d, err = %d\n", reply_fd, errno);
+ }
+ }
+ return 0;
}
#endif
-static int do_not_aio(struct aio_context *aio)
+static int do_not_aio(struct aio_thread_req *req)
{
- char c;
- int err;
-
- switch(aio->type){
- case AIO_READ:
- err = os_seek_file(aio->fd, aio->offset);
- if(err)
- goto out;
-
- err = os_read_file(aio->fd, aio->data, aio->len);
- break;
- case AIO_WRITE:
- err = os_seek_file(aio->fd, aio->offset);
- if(err)
- goto out;
-
- err = os_write_file(aio->fd, aio->data, aio->len);
- break;
- case AIO_MMAP:
- err = os_seek_file(aio->fd, aio->offset);
- if(err)
- goto out;
-
- err = os_read_file(aio->fd, &c, sizeof(c));
- break;
- default:
- printk("do_not_aio - bad request type : %d\n", aio->type);
- err = -EINVAL;
- break;
- }
-
- out:
- return err;
+ char c;
+ unsigned long long actual;
+ int n;
+
+ actual = lseek64(req->io_fd, req->offset, SEEK_SET);
+ if (actual != req->offset)
+ return -errno;
+
+ switch (req->type) {
+ case AIO_READ:
+ n = read(req->io_fd, req->buf, req->len);
+ break;
+ case AIO_WRITE:
+ n = write(req->io_fd, req->buf, req->len);
+ break;
+ case AIO_MMAP:
+ n = read(req->io_fd, &c, sizeof(c));
+ break;
+ default:
+ printk(UM_KERN_ERR "do_not_aio - bad request type : %d\n",
+ req->type);
+ return -EINVAL;
+ }
+
+ if (n < 0)
+ return -errno;
+ return 0;
}
+/* These are initialized in initcalls and not changed */
+static int aio_req_fd_r = -1;
+static int aio_req_fd_w = -1;
+static int aio_pid = -1;
+static unsigned long aio_stack;
+
static int not_aio_thread(void *arg)
{
- struct aio_context *aio;
- struct aio_thread_reply reply;
- int err;
-
- signal(SIGWINCH, SIG_IGN);
- while(1){
- err = os_read_file(aio_req_fd_r, &aio, sizeof(aio));
- if(err != sizeof(aio)){
- if(err < 0)
- printk("not_aio_thread - read failed, "
- "fd = %d, err = %d\n", aio_req_fd_r,
- -err);
- else {
- printk("not_aio_thread - short read, fd = %d, "
- "length = %d\n", aio_req_fd_r, err);
- }
- continue;
- }
- again:
- err = do_not_aio(aio);
-
- if(update_aio(aio, err))
- goto again;
-
- reply = ((struct aio_thread_reply) { .data = aio,
- .err = aio->len });
- err = os_write_file(aio->reply_fd, &reply, sizeof(reply));
- if(err != sizeof(reply))
- printk("not_aio_thread - write failed, fd = %d, "
- "err = %d\n", aio_req_fd_r, -err);
- }
+ struct aio_thread_req req;
+ struct aio_thread_reply reply;
+ int err;
+
+ signal(SIGWINCH, SIG_IGN);
+ while (1) {
+ err = read(aio_req_fd_r, &req, sizeof(req));
+ if (err != sizeof(req)) {
+ if (err < 0)
+ printk(UM_KERN_ERR "not_aio_thread - "
+ "read failed, fd = %d, err = %d\n",
+ aio_req_fd_r,
+ errno);
+ else {
+ printk(UM_KERN_ERR "not_aio_thread - short "
+ "read, fd = %d, length = %d\n",
+ aio_req_fd_r, err);
+ }
+ continue;
+ }
+ err = do_not_aio(&req);
+ reply = ((struct aio_thread_reply) { .data = req.aio,
+ .err = err });
+ err = write(req.aio->reply_fd, &reply, sizeof(reply));
+ if (err != sizeof(reply))
+ printk(UM_KERN_ERR "not_aio_thread - write failed, "
+ "fd = %d, err = %d\n", req.aio->reply_fd, errno);
+ }
+
+ return 0;
}
-static int submit_aio_24(struct aio_context *aio)
+static int init_aio_24(void)
{
- int err;
+ int fds[2], err;
- err = os_write_file(aio_req_fd_w, &aio, sizeof(aio));
- if(err == sizeof(aio))
- err = 0;
+ err = os_pipe(fds, 1, 1);
+ if (err)
+ goto out;
- return err;
-}
+ aio_req_fd_w = fds[0];
+ aio_req_fd_r = fds[1];
-static int aio_pid = -1;
-static int (*submit_proc)(struct aio_context *aio);
+ err = os_set_fd_block(aio_req_fd_w, 0);
+ if (err)
+ goto out_close_pipe;
-static int init_aio_24(void)
-{
- unsigned long stack;
- int fds[2], err;
-
- err = os_pipe(fds, 1, 1);
- if(err)
- goto out;
-
- aio_req_fd_w = fds[0];
- aio_req_fd_r = fds[1];
- err = run_helper_thread(not_aio_thread, NULL,
- CLONE_FILES | CLONE_VM | SIGCHLD, &stack, 0);
- if(err < 0)
- goto out_close_pipe;
-
- aio_pid = err;
- goto out;
-
- out_close_pipe:
- os_close_file(fds[0]);
- os_close_file(fds[1]);
- aio_req_fd_w = -1;
- aio_req_fd_r = -1;
- out:
-#ifndef HAVE_AIO_ABI
- printk("/usr/include/linux/aio_abi.h not present during build\n");
-#endif
- printk("2.6 host AIO support not used - falling back to I/O "
- "thread\n");
+ err = run_helper_thread(not_aio_thread, NULL,
+ CLONE_FILES | CLONE_VM, &aio_stack);
+ if (err < 0)
+ goto out_close_pipe;
- submit_proc = submit_aio_24;
+ aio_pid = err;
+ goto out;
- return 0;
+out_close_pipe:
+ close(fds[0]);
+ close(fds[1]);
+ aio_req_fd_w = -1;
+ aio_req_fd_r = -1;
+out:
+#ifndef HAVE_AIO_ABI
+ printk(UM_KERN_INFO "/usr/include/linux/aio_abi.h not present during "
+ "build\n");
+#endif
+ printk(UM_KERN_INFO "2.6 host AIO support not used - falling back to "
+ "I/O thread\n");
+ return 0;
}
#ifdef HAVE_AIO_ABI
#define DEFAULT_24_AIO 0
-static int submit_aio_26(struct aio_context *aio)
+static int init_aio_26(void)
{
- struct aio_thread_reply reply;
int err;
- err = do_aio(ctx, aio);
- if(err){
- reply = ((struct aio_thread_reply) { .data = aio,
- .err = err });
- err = os_write_file(aio->reply_fd, &reply, sizeof(reply));
- if(err != sizeof(reply))
- printk("submit_aio_26 - write failed, "
- "fd = %d, err = %d\n", aio->reply_fd, -err);
- else err = 0;
+ if (io_setup(256, &ctx)) {
+ err = -errno;
+ printk(UM_KERN_ERR "aio_thread failed to initialize context, "
+ "err = %d\n", errno);
+ return err;
}
- return err;
-}
-
-static int init_aio_26(void)
-{
- unsigned long stack;
- int err;
+ err = run_helper_thread(aio_thread, NULL,
+ CLONE_FILES | CLONE_VM, &aio_stack);
+ if (err < 0)
+ return err;
- if(io_setup(256, &ctx)){
- err = -errno;
- printk("aio_thread failed to initialize context, err = %d\n",
- errno);
- return err;
- }
+ aio_pid = err;
- err = run_helper_thread(aio_thread, NULL,
- CLONE_FILES | CLONE_VM | SIGCHLD, &stack, 0);
- if(err < 0)
- return err;
-
- aio_pid = err;
+ printk(UM_KERN_INFO "Using 2.6 host AIO\n");
+ return 0;
+}
- printk("Using 2.6 host AIO\n");
+static int submit_aio_26(enum aio_type type, int io_fd, char *buf, int len,
+ unsigned long long offset, struct aio_context *aio)
+{
+ struct aio_thread_reply reply;
+ int err;
- submit_proc = submit_aio_26;
+ err = do_aio(ctx, type, io_fd, buf, len, offset, aio);
+ if (err) {
+ reply = ((struct aio_thread_reply) { .data = aio,
+ .err = err });
+ err = write(aio->reply_fd, &reply, sizeof(reply));
+ if (err != sizeof(reply)) {
+ err = -errno;
+ printk(UM_KERN_ERR "submit_aio_26 - write failed, "
+ "fd = %d, err = %d\n", aio->reply_fd, -err);
+ }
+ else err = 0;
+ }
- return 0;
+ return err;
}
#else
#define DEFAULT_24_AIO 1
-static int submit_aio_26(struct aio_context *aio)
+static int init_aio_26(void)
{
- return -ENOSYS;
+ return -ENOSYS;
}
-static int init_aio_26(void)
+static int submit_aio_26(enum aio_type type, int io_fd, char *buf, int len,
+ unsigned long long offset, struct aio_context *aio)
{
- submit_proc = submit_aio_26;
- return -ENOSYS;
+ return -ENOSYS;
}
#endif
+/* Initialized in an initcall and unchanged thereafter */
static int aio_24 = DEFAULT_24_AIO;
static int __init set_aio_24(char *name, int *add)
{
- aio_24 = 1;
- return 0;
+ aio_24 = 1;
+ return 0;
}
__uml_setup("aio=2.4", set_aio_24,
static int init_aio(void)
{
- int err;
-
- CHOOSE_MODE(({
- if(!aio_24){
- printk("Disabling 2.6 AIO in tt mode\n");
- aio_24 = 1;
- } }), (void) 0);
-
- if(!aio_24){
- err = init_aio_26();
- if(err && (errno == ENOSYS)){
- printk("2.6 AIO not supported on the host - "
- "reverting to 2.4 AIO\n");
- aio_24 = 1;
- }
- else return err;
- }
-
- if(aio_24)
- return init_aio_24();
-
- return 0;
+ int err;
+
+ if (!aio_24) {
+ err = init_aio_26();
+ if (err && (errno == ENOSYS)) {
+ printk(UM_KERN_INFO "2.6 AIO not supported on the "
+ "host - reverting to 2.4 AIO\n");
+ aio_24 = 1;
+ }
+ else return err;
+ }
+
+ if (aio_24)
+ return init_aio_24();
+
+ return 0;
}
-/* The reason for the __initcall/__uml_exitcall asymmetry is that init_aio
+/*
+ * The reason for the __initcall/__uml_exitcall asymmetry is that init_aio
* needs to be called when the kernel is running because it calls run_helper,
* which needs get_free_page. exit_aio is a __uml_exitcall because the generic
* kernel does not run __exitcalls on shutdown, and can't because many of them
static void exit_aio(void)
{
- if(aio_pid != -1)
- os_kill_process(aio_pid, 1);
+ if (aio_pid != -1) {
+ os_kill_process(aio_pid, 1);
+ free_stack(aio_stack, 0);
+ }
}
__uml_exitcall(exit_aio);
-int submit_aio(struct aio_context *aio)
+static int submit_aio_24(enum aio_type type, int io_fd, char *buf, int len,
+ unsigned long long offset, struct aio_context *aio)
+{
+ struct aio_thread_req req = { .type = type,
+ .io_fd = io_fd,
+ .offset = offset,
+ .buf = buf,
+ .len = len,
+ .aio = aio,
+ };
+ int err;
+
+ err = write(aio_req_fd_w, &req, sizeof(req));
+ if (err == sizeof(req))
+ err = 0;
+ else err = -errno;
+
+ return err;
+}
+
+int submit_aio(enum aio_type type, int io_fd, char *buf, int len,
+ unsigned long long offset, int reply_fd,
+ struct aio_context *aio)
{
- return (*submit_proc)(aio);
+ aio->reply_fd = reply_fd;
+ if (aio_24)
+ return submit_aio_24(type, io_fd, buf, len, offset, aio);
+ else
+ return submit_aio_26(type, io_fd, buf, len, offset, aio);
}
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff