initial trim support

This commit is contained in:
comatose 2012-12-01 00:42:29 +09:00
parent a1219f920f
commit 01919ef8a1
3 changed files with 155 additions and 125 deletions

234
buse.c
View File

@ -18,149 +18,157 @@
*/
#ifdef WORDS_BIGENDIAN
u_int64_t ntohll(u_int64_t a) {
return a;
return a;
}
#else
u_int64_t ntohll(u_int64_t a) {
u_int32_t lo = a & 0xffffffff;
u_int32_t hi = a >> 32U;
lo = ntohl(lo);
hi = ntohl(hi);
return ((u_int64_t) lo) << 32U | hi;
u_int32_t lo = a & 0xffffffff;
u_int32_t hi = a >> 32U;
lo = ntohl(lo);
hi = ntohl(hi);
return ((u_int64_t) lo) << 32U | hi;
}
#endif
#define htonll ntohll
static int read_all(int fd, void *buf, size_t count)
{
int bytes_read;
int bytes_read;
while (count > 0) {
bytes_read = read(fd, buf, count);
assert(bytes_read > 0);
buf = (char *)buf + bytes_read;
count -= bytes_read;
}
assert(count == 0);
while (count > 0) {
bytes_read = read(fd, buf, count);
assert(bytes_read > 0);
buf = (char *)buf + bytes_read;
count -= bytes_read;
}
assert(count == 0);
return 0;
return 0;
}
static int write_all(int fd, const void *buf, size_t count)
{
int bytes_written;
int bytes_written;
while (count > 0) {
bytes_written = write(fd, buf, count);
assert(bytes_written > 0);
buf = (char *)buf + bytes_written;
count -= bytes_written;
}
assert(count == 0);
while (count > 0) {
bytes_written = write(fd, buf, count);
assert(bytes_written > 0);
buf = (char *)buf + bytes_written;
count -= bytes_written;
}
assert(count == 0);
return 0;
return 0;
}
int buse_main(int argc, char *argv[], const struct buse_operations *aop, void *userdata)
{
int sp[2];
int nbd, sk, err, tmp_fd;
u_int64_t from;
u_int32_t len, bytes_read;
char *dev_file;
struct nbd_request request;
struct nbd_reply reply;
void *chunk;
int sp[2];
int nbd, sk, err, tmp_fd;
u_int64_t from;
u_int32_t len;
ssize_t bytes_read;
char *dev_file;
struct nbd_request request;
struct nbd_reply reply;
void *chunk;
(void) userdata;
(void) userdata;
assert(argc == 3);
dev_file = argv[2];
assert(argc == 3);
dev_file = argv[2];
assert(!socketpair(AF_UNIX, SOCK_STREAM, 0, sp));
assert(!socketpair(AF_UNIX, SOCK_STREAM, 0, sp));
nbd = open(dev_file, O_RDWR);
assert(nbd != -1);
nbd = open(dev_file, O_RDWR);
assert(nbd != -1);
assert(ioctl(nbd, NBD_SET_SIZE, aop->size) != -1);
assert(ioctl(nbd, NBD_SET_SIZE, aop->size) != -1);
assert(ioctl(nbd, NBD_CLEAR_SOCK) != -1);
if (!fork()) {
/* The child needs to continue setting things up. */
close(sp[0]);
sk = sp[1];
if(ioctl(nbd, NBD_SET_SOCK, sk) == -1){
fprintf(stderr, "ioctl(nbd, NBD_SET_SOCK, sk) failed.[%s]\n", strerror(errno));
}
else if(ioctl(nbd, NBD_SET_FLAGS, NBD_FLAG_SEND_TRIM) == -1){
fprintf(stderr, "ioctl(nbd, NBD_SET_FLAGS, NBD_FLAG_SEND_TRIM) failed.[%s]\n", strerror(errno));
}
else{
err = ioctl(nbd, NBD_DO_IT);
fprintf(stderr, "nbd device terminated with code %d\n", err);
if (err == -1)
fprintf(stderr, "%s\n", strerror(errno));
}
assert(ioctl(nbd, NBD_CLEAR_QUE) != -1);
assert(ioctl(nbd, NBD_CLEAR_SOCK) != -1);
if (!fork()) {
/* The child needs to continue setting things up. */
close(sp[0]);
sk = sp[1];
exit(0);
}
assert(ioctl(nbd, NBD_SET_SOCK, sk) != -1);
err = ioctl(nbd, NBD_DO_IT);
fprintf(stderr, "nbd device terminated with code %d\n", err);
if (err == -1)
fprintf(stderr, "%s\n", strerror(errno));
/* The parent opens the device file at least once, to make sure the
* partition table is updated. Then it closes it and starts serving up
* requests. */
assert(ioctl(nbd, NBD_CLEAR_QUE) != -1);
assert(ioctl(nbd, NBD_CLEAR_SOCK) != -1);
tmp_fd = open(dev_file, O_RDONLY);
assert(tmp_fd != -1);
close(tmp_fd);
exit(0);
close(sp[1]);
sk = sp[0];
reply.magic = htonl(NBD_REPLY_MAGIC);
reply.error = htonl(0);
while ((bytes_read = read(sk, &request, sizeof(request))) > 0) {
assert(bytes_read == sizeof(request));
memcpy(reply.handle, request.handle, sizeof(reply.handle));
len = ntohl(request.len);
from = ntohll(request.from);
assert(request.magic == htonl(NBD_REQUEST_MAGIC));
switch(ntohl(request.type)) {
/* I may at some point need to deal with the the fact that the
* official nbd server has a maximum buffer size, and divides up
* oversized requests into multiple pieces. This applies to reads
* and writes.
*/
case NBD_CMD_READ:
/* fprintf(stderr, "Request for read of size %d\n", len); */
chunk = malloc(len + sizeof(struct nbd_reply));
aop->read((char *)chunk + sizeof(struct nbd_reply), len, from);
memcpy(chunk, &reply, sizeof(struct nbd_reply));
write_all(sk, chunk, len + sizeof(struct nbd_reply));
free(chunk);
break;
case NBD_CMD_WRITE:
/* fprintf(stderr, "Request for write of size %d\n", len); */
chunk = malloc(len);
read_all(sk, chunk, len);
aop->write(chunk, len, from);
free(chunk);
write_all(sk, &reply, sizeof(struct nbd_reply));
break;
case NBD_CMD_DISC:
/* Handle a disconnect request. */
aop->disc();
return 0;
case NBD_CMD_FLUSH:
aop->flush();
break;
case NBD_CMD_TRIM:
aop->trim(from, len);
break;
default:
assert(0);
}
/* The parent opens the device file at least once, to make sure the
* partition table is updated. Then it closes it and starts serving up
* requests. */
tmp_fd = open(dev_file, O_RDONLY);
assert(tmp_fd != -1);
close(tmp_fd);
close(sp[1]);
sk = sp[0];
reply.magic = htonl(NBD_REPLY_MAGIC);
reply.error = htonl(0);
while (1) {
bytes_read = read(sk, &request, sizeof(request));
assert(bytes_read == sizeof(request));
memcpy(reply.handle, request.handle, sizeof(reply.handle));
len = ntohl(request.len);
from = ntohll(request.from);
assert(request.magic == htonl(NBD_REQUEST_MAGIC));
switch(ntohl(request.type)) {
/* I may at some point need to deal with the the fact that the
* official nbd server has a maximum buffer size, and divides up
* oversized requests into multiple pieces. This applies to reads
* and writes.
*/
case NBD_CMD_READ:
/* fprintf(stderr, "Request for read of size %d\n", len); */
chunk = malloc(len + sizeof(struct nbd_reply));
aop->read((char *)chunk + sizeof(struct nbd_reply), len, from);
memcpy(chunk, &reply, sizeof(struct nbd_reply));
write_all(sk, chunk, len + sizeof(struct nbd_reply));
free(chunk);
break;
case NBD_CMD_WRITE:
/* fprintf(stderr, "Request for write of size %d\n", len); */
chunk = malloc(len);
read_all(sk, chunk, len);
aop->write(chunk, len, from);
free(chunk);
write_all(sk, &reply, sizeof(struct nbd_reply));
break;
case NBD_CMD_DISC:
/* Handle a disconnect request. */
aop->disc();
return 0;
case NBD_CMD_FLUSH:
aop->flush();
break;
case NBD_CMD_TRIM:
aop->trim(from, len);
break;
default:
assert(0);
}
}
return 0;
}
if (bytes_read == -1)
fprintf(stderr, "%s\n", strerror(errno));
return 0;
}

6
buse.h
View File

@ -26,6 +26,12 @@ enum {
NBD_CMD_TRIM = 4
};
/* values for flags field */
#define NBD_FLAG_HAS_FLAGS (1 << 0)
#define NBD_FLAG_READ_ONLY (1 << 1)
/* there is a gap here to match userspace */
#define NBD_FLAG_SEND_TRIM (1 << 5) /* send trim/discard */
/* Magic numbers */
#define NBD_REQUEST_MAGIC 0x25609513
#define NBD_REPLY_MAGIC 0x67446698

View File

@ -8,32 +8,48 @@ static void *data;
static int xmp_read(void *buf, u_int32_t len, u_int64_t offset)
{
memcpy(buf, (char *)data + offset, len);
return 0;
memcpy(buf, (char *)data + offset, len);
fprintf(stderr, "R %lu %u\n", offset, len);
return 0;
}
static int xmp_write(const void *buf, u_int32_t len, u_int64_t offset)
{
memcpy((char *)data + offset, buf, len);
return 0;
memcpy((char *)data + offset, buf, len);
fprintf(stderr, "W %lu %u\n", offset, len);
return 0;
}
static int xmp_disc()
{
fprintf(stderr, "Received a disconnect request.\n");
return 0;
fprintf(stderr, "Received a disconnect request.\n");
return 0;
}
static int xmp_flush()
{
fprintf(stderr, "Received a flush request.\n");
return 0;
}
static int xmp_trim(u_int64_t from, u_int32_t len)
{
fprintf(stderr, "T %lu %u\n", from, len);
return 0;
}
static struct buse_operations aop = {
.read = xmp_read,
.write = xmp_write,
.disc = xmp_disc,
.size = 128 * 1024 * 1024,
.read = xmp_read,
.write = xmp_write,
.disc = xmp_disc,
.flush = xmp_flush,
.trim = xmp_trim,
.size = 128 * 1024 * 1024,
};
int main(int argc, char *argv[])
{
data = malloc(aop.size);
data = malloc(aop.size);
return buse_main(argc, argv, &aop, NULL);
return buse_main(argc, argv, &aop, NULL);
}