#include <linux/bio.h>
#include <linux/device.h>
#include <linux/completion.h>
-
+#include <linux/wait.h>
#include <sys/kernel/segdev.h>
#include "xsegbd.h"
+#include <xseg/protocol.h>
#define XSEGBD_MINORS 1
/* define max request size to be used in xsegbd */
//FIXME should we make this 4MB instead of 256KB ?
-#define XSEGBD_MAX_REQUEST_SIZE 262144U
+//#define XSEGBD_MAX_REQUEST_SIZE 262144U
+#define XSEGBD_MAX_REQUEST_SIZE 4194304U
MODULE_DESCRIPTION("xsegbd");
MODULE_AUTHOR("XSEG");
static DEFINE_SPINLOCK(xsegbd_devices_lock);
-
-static struct xsegbd_device *__xsegbd_get_dev(unsigned long id)
+struct xsegbd_device *__xsegbd_get_dev(unsigned long id)
{
struct xsegbd_device *xsegbd_dev = NULL;
static void xseg_request_fn(struct request_queue *rq);
static int xsegbd_get_size(struct xsegbd_device *xsegbd_dev);
+static int xsegbd_mapclose(struct xsegbd_device *xsegbd_dev);
static int xsegbd_dev_init(struct xsegbd_device *xsegbd_dev)
{
xsegbd_dev->xsegbd = &xsegbd;
- xsegbd_dev->blk_queue = blk_alloc_queue(GFP_KERNEL);
+ /* allocates and initializes queue */
+ xsegbd_dev->blk_queue = blk_init_queue(xseg_request_fn, &xsegbd_dev->rqlock);
if (!xsegbd_dev->blk_queue)
goto out;
- if (!blk_init_allocated_queue(xsegbd_dev->blk_queue,
- xseg_request_fn, &xsegbd_dev->rqlock))
- goto outqueue;
-
xsegbd_dev->blk_queue->queuedata = xsegbd_dev;
blk_queue_flush(xsegbd_dev->blk_queue, REQ_FLUSH | REQ_FUA);
blk_queue_physical_block_size(xsegbd_dev->blk_queue, blksize);
blk_queue_bounce_limit(xsegbd_dev->blk_queue, BLK_BOUNCE_ANY);
- //blk_queue_max_segments(dev->blk_queue, 512);
max_request_size_bytes = XSEGBD_MAX_REQUEST_SIZE;
blk_queue_max_hw_sectors(xsegbd_dev->blk_queue, max_request_size_bytes >> 9);
+// blk_queue_max_sectors(xsegbd_dev->blk_queue, max_request_size_bytes >> 10);
+ blk_queue_max_segments(xsegbd_dev->blk_queue, 1024);
blk_queue_max_segment_size(xsegbd_dev->blk_queue, max_request_size_bytes);
blk_queue_io_min(xsegbd_dev->blk_queue, max_request_size_bytes);
blk_queue_io_opt(xsegbd_dev->blk_queue, max_request_size_bytes);
/* vkoukis says we don't need partitions */
xsegbd_dev->gd = disk = alloc_disk(1);
if (!disk)
- goto outqueue;
+ goto out;
disk->major = xsegbd_dev->major;
disk->first_minor = 0; // id * XSEGBD_MINORS;
snprintf(disk->disk_name, 32, "xsegbd%u", xsegbd_dev->id);
ret = 0;
-
+
/* allow a non-zero sector_size parameter to override the disk size */
if (sector_size)
xsegbd_dev->sectors = sector_size;
else {
ret = xsegbd_get_size(xsegbd_dev);
if (ret)
- goto outdisk;
+ goto out;
}
set_capacity(disk, xsegbd_dev->sectors);
XSEGLOG("xsegbd active...");
add_disk(disk); /* immediately activates the device */
- return 0;
-
-
-outdisk:
- put_disk(xsegbd_dev->gd);
-outqueue:
- blk_cleanup_queue(xsegbd_dev->blk_queue);
out:
- xsegbd_dev->gd = NULL;
+ /* on error, everything is cleaned up in xsegbd_dev_release */
return ret;
}
static void xsegbd_dev_release(struct device *dev)
{
struct xsegbd_device *xsegbd_dev = dev_to_xsegbd(dev);
-
- xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
+
/* cleanup gendisk and blk_queue the right way */
if (xsegbd_dev->gd) {
if (xsegbd_dev->gd->flags & GENHD_FL_UP)
del_gendisk(xsegbd_dev->gd);
+ xsegbd_mapclose(xsegbd_dev);
+ }
+
+ spin_lock(&xsegbd_devices_lock);
+ BUG_ON(xsegbd_devices[xsegbd_dev->src_portno] != xsegbd_dev);
+ xsegbd_devices[xsegbd_dev->src_portno] = NULL;
+ spin_unlock(&xsegbd_devices_lock);
+
+ XSEGLOG("releasing id: %d", xsegbd_dev->id);
+// xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
+ xseg_quit_local_signal(xsegbd_dev->xseg, xsegbd_dev->src_portno);
+
+ if (xsegbd_dev->blk_queue)
blk_cleanup_queue(xsegbd_dev->blk_queue);
+ if (xsegbd_dev->gd)
put_disk(xsegbd_dev->gd);
- }
// if (xseg_free_requests(xsegbd_dev->xseg,
// xsegbd_dev->src_portno, xsegbd_dev->nr_requests) < 0)
// XSEGLOG("Error trying to free requests!\n");
+ if (xsegbd_dev->xseg){
+ xseg_leave(xsegbd_dev->xseg);
+ xsegbd_dev->xseg = NULL;
+ }
unregister_blkdev(xsegbd_dev->major, XSEGBD_NAME);
- spin_lock(&xsegbd_devices_lock);
- BUG_ON(xsegbd_devices[xsegbd_dev->src_portno] != xsegbd_dev);
- xsegbd_devices[xsegbd_dev->src_portno] = NULL;
- spin_unlock(&xsegbd_devices_lock);
-
- if (xsegbd_dev->blk_req_pending)
+ if (xsegbd_dev->blk_req_pending){
kfree(xsegbd_dev->blk_req_pending);
+ xsegbd_dev->blk_req_pending = NULL;
+ }
xq_free(&xsegbd_dev->blk_queue_pending);
-
kfree(xsegbd_dev);
-
module_put(THIS_MODULE);
}
uint64_t datalen;
xport p;
int r;
+ unsigned long flags;
+ spin_unlock_irq(&xsegbd_dev->rqlock);
for (;;) {
+ if (current_thread_info()->preempt_count || irqs_disabled()){
+ XSEGLOG("Current thread preempt_count: %d, irqs_disabled(): %lu ",
+ current_thread_info()->preempt_count, irqs_disabled());
+ }
+ //XSEGLOG("Priority: %d", current_thread_info()->task->prio);
+ //XSEGLOG("Static priority: %d", current_thread_info()->task->static_prio);
+ //XSEGLOG("Normal priority: %d", current_thread_info()->task->normal_prio);
+ //XSEGLOG("Rt_priority: %u", current_thread_info()->task->rt_priority);
blkreq_idx = Noneidx;
xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
xsegbd_dev->dst_portno, X_ALLOC);
xsegbd_dev->src_portno);
if (blkreq_idx == Noneidx)
break;
-
+
if (blkreq_idx >= xsegbd_dev->nr_requests) {
XSEGLOG("blkreq_idx >= xsegbd_dev->nr_requests");
BUG_ON(1);
break;
}
+
+ spin_lock_irqsave(&xsegbd_dev->rqlock, flags);
blkreq = blk_fetch_request(rq);
- if (!blkreq)
+ if (!blkreq){
+ spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
break;
+ }
if (blkreq->cmd_type != REQ_TYPE_FS) {
- //we lose xreq here
+ //FIXME we lose xreq here
XSEGLOG("non-fs cmd_type: %u. *shrug*", blkreq->cmd_type);
__blk_end_request_all(blkreq, 0);
+ spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
continue;
}
+ spin_unlock_irqrestore(&xsegbd_dev->rqlock, flags);
+ if (current_thread_info()->preempt_count || irqs_disabled()){
+ XSEGLOG("Current thread preempt_count: %d, irqs_disabled(): %lu ",
+ current_thread_info()->preempt_count, irqs_disabled());
+ }
datalen = blk_rq_bytes(blkreq);
r = xseg_prep_request(xsegbd_dev->xseg, xreq,
xsegbd_dev->targetlen, datalen);
if (r < 0) {
XSEGLOG("couldn't prep request");
- __blk_end_request_err(blkreq, r);
+ blk_end_request_err(blkreq, r);
BUG_ON(1);
break;
}
r = -ENOMEM;
if (xreq->bufferlen - xsegbd_dev->targetlen < datalen){
XSEGLOG("malformed req buffers");
- __blk_end_request_err(blkreq, r);
+ blk_end_request_err(blkreq, r);
BUG_ON(1);
break;
}
pending->dev = xsegbd_dev;
pending->request = blkreq;
pending->comp = NULL;
-
+
xreq->size = datalen;
xreq->offset = blk_rq_pos(blkreq) << 9;
xreq->priv = (uint64_t) blkreq_idx;
xreq->flags |= XF_FUA;
if (rq_data_dir(blkreq)) {
- /* unlock for data transfers? */
blk_to_xseg(xsegbd_dev->xseg, xreq, blkreq);
xreq->op = X_WRITE;
} else {
}
+// XSEGLOG("%s : %lu (%lu)", xsegbd_dev->target, xreq->offset, xreq->datalen);
r = -EIO;
p = xseg_submit(xsegbd_dev->xseg, xreq,
xsegbd_dev->src_portno, X_ALLOC);
if (p == NoPort) {
XSEGLOG("coundn't submit req");
- BUG_ON(1);
- __blk_end_request_err(blkreq, r);
+ WARN_ON(1);
+ blk_end_request_err(blkreq, r);
break;
}
WARN_ON(xseg_signal(xsegbd_dev->xsegbd->xseg, p) < 0);
if (blkreq_idx != Noneidx)
BUG_ON(xq_append_head(&xsegbd_dev->blk_queue_pending,
blkreq_idx, xsegbd_dev->src_portno) == Noneidx);
+ spin_lock_irq(&xsegbd_dev->rqlock);
}
int update_dev_sectors_from_request( struct xsegbd_device *xsegbd_dev,
struct xseg_request *xreq )
{
void *data;
+ if (!xreq) {
+ XSEGLOG("Invalid xreq");
+ return -EIO;
+ }
if (xreq->state & XS_FAILED)
return -ENOENT;
return -EIO;
data = xseg_get_data(xsegbd_dev->xseg, xreq);
+ if (!data) {
+ XSEGLOG("Invalid req data");
+ return -EIO;
+ }
+ if (!xsegbd_dev) {
+ XSEGLOG("Invalid xsegbd_dev");
+ return -ENOENT;
+ }
xsegbd_dev->sectors = *((uint64_t *) data) / 512ULL;
return 0;
}
{
struct xseg_request *xreq;
char *target;
- uint64_t datalen;
xqindex blkreq_idx;
struct xsegbd_pending *pending;
struct completion comp;
xport p;
- void *data;
- int ret = -EBUSY, r;
+ int ret = -EBUSY;
+
xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
xsegbd_dev->dst_portno, X_ALLOC);
if (!xreq)
goto out;
- datalen = sizeof(uint64_t);
- BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen, datalen));
- BUG_ON(xreq->bufferlen - xsegbd_dev->targetlen < datalen);
+ BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen,
+ sizeof(struct xseg_reply_info)));
init_completion(&comp);
blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending, 1);
if (blkreq_idx == Noneidx)
- goto out;
-
+ goto out_put;
+
pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
pending->dev = xsegbd_dev;
pending->request = NULL;
pending->comp = ∁
-
+
xreq->priv = (uint64_t) blkreq_idx;
target = xseg_get_target(xsegbd_dev->xseg, xreq);
strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
- xreq->size = datalen;
+ xreq->size = xreq->datalen;
xreq->offset = 0;
xreq->op = X_INFO;
goto out_queue;
}
WARN_ON(xseg_signal(xsegbd_dev->xseg, p) < 0);
-
+ XSEGLOG("Before wait for completion, comp %lx [%llu]", (unsigned long) pending->comp, (unsigned long long) blkreq_idx);
wait_for_completion_interruptible(&comp);
- //XSEGLOG("Woken up after wait_for_completion_interruptible()\n");
+ XSEGLOG("Woken up after wait_for_completion_interruptible(), comp: %lx [%llu]", (unsigned long) pending->comp, (unsigned long long) blkreq_idx);
ret = update_dev_sectors_from_request(xsegbd_dev, xreq);
- //XSEGLOG("get_size: sectors = %ld\n", (long)xsegbd_dev->sectors);
-out:
+ XSEGLOG("get_size: sectors = %ld\n", (long)xsegbd_dev->sectors);
+
+out_queue:
+ pending->dev = NULL;
+ pending->comp = NULL;
+ xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
+out_put:
BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) == -1);
+out:
return ret;
+}
+
+static int xsegbd_mapclose(struct xsegbd_device *xsegbd_dev)
+{
+ struct xseg_request *xreq;
+ char *target;
+ xqindex blkreq_idx;
+ struct xsegbd_pending *pending;
+ struct completion comp;
+ xport p;
+ int ret = -EBUSY;
+
+ xreq = xseg_get_request(xsegbd_dev->xseg, xsegbd_dev->src_portno,
+ xsegbd_dev->dst_portno, X_ALLOC);
+ if (!xreq)
+ goto out;
+
+ BUG_ON(xseg_prep_request(xsegbd_dev->xseg, xreq, xsegbd_dev->targetlen, 0));
+
+ init_completion(&comp);
+ blkreq_idx = xq_pop_head(&xsegbd_dev->blk_queue_pending, 1);
+ if (blkreq_idx == Noneidx)
+ goto out_put;
+
+ pending = &xsegbd_dev->blk_req_pending[blkreq_idx];
+ pending->dev = xsegbd_dev;
+ pending->request = NULL;
+ pending->comp = ∁
+
+
+ xreq->priv = (uint64_t) blkreq_idx;
+
+ target = xseg_get_target(xsegbd_dev->xseg, xreq);
+ strncpy(target, xsegbd_dev->target, xsegbd_dev->targetlen);
+ xreq->size = xreq->datalen;
+ xreq->offset = 0;
+ xreq->op = X_CLOSE;
+
+ xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
+ p = xseg_submit(xsegbd_dev->xseg, xreq,
+ xsegbd_dev->src_portno, X_ALLOC);
+ if ( p == NoPort) {
+ XSEGLOG("couldn't submit request");
+ BUG_ON(1);
+ goto out_queue;
+ }
+ WARN_ON(xseg_signal(xsegbd_dev->xseg, p) < 0);
+ wait_for_completion_interruptible(&comp);
+ ret = 0;
+ if (xreq->state & XS_FAILED)
+ XSEGLOG("Couldn't close disk on mapper");
out_queue:
+ pending->dev = NULL;
+ pending->comp = NULL;
xq_append_head(&xsegbd_dev->blk_queue_pending, blkreq_idx, 1);
-
- goto out;
+out_put:
+ BUG_ON(xseg_put_request(xsegbd_dev->xseg, xreq, xsegbd_dev->src_portno) == -1);
+out:
+ return ret;
}
static void xseg_callback(xport portno)
unsigned long flags;
xqindex blkreq_idx, ridx;
int err;
- void *data;
xsegbd_dev = __xsegbd_get_dev(portno);
if (!xsegbd_dev) {
for (;;) {
xseg_prepare_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
- xreq = xseg_receive(xsegbd_dev->xseg, portno);
+ xreq = xseg_receive(xsegbd_dev->xseg, portno, 0);
if (!xreq)
break;
- xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
+// xseg_cancel_wait(xsegbd_dev->xseg, xsegbd_dev->src_portno);
blkreq_idx = (xqindex) xreq->priv;
if (blkreq_idx >= xsegbd_dev->nr_requests) {
/* someone is blocking on this request
and will handle it when we wake them up. */
complete(pending->comp);
- pending->comp = NULL;
/* the request is blocker's responsibility so
we will not put_request(); */
continue;
if (xsegbd_dev != pending->dev) {
//FIXME maybe put request?
XSEGLOG("xsegbd_dev != pending->dev");
- BUG_ON(1);
+ WARN_ON(1);
continue;
}
pending->dev = NULL;
if (!blkreq){
//FIXME maybe put request?
XSEGLOG("blkreq does not exist");
- BUG_ON(1);
+ WARN_ON(1);
continue;
}
goto blk_end;
err = 0;
- /* unlock for data transfer? */
if (!rq_data_dir(blkreq)){
xseg_to_blk(xsegbd_dev->xseg, xreq, blkreq);
- }
+ }
blk_end:
blk_end_request_all(blkreq, err);
-
+
ridx = xq_append_head(&xsegbd_dev->blk_queue_pending,
blkreq_idx, xsegbd_dev->src_portno);
if (ridx == Noneidx) {
BUG_ON(1);
}
}
-
if (xsegbd_dev) {
spin_lock_irqsave(&xsegbd_dev->rqlock, flags);
xseg_request_fn(xsegbd_dev->blk_queue);
return ret;
}
+//FIXME
+//maybe try callback, first and then do a more invasive cleanup
static ssize_t xsegbd_cleanup(struct device *dev,
struct device_attribute *attr,
const char *buf,
struct completion *comp = NULL;
mutex_lock_nested(&xsegbd_mutex, SINGLE_DEPTH_NESTING);
- for (i = 0; i < xsegbd_dev->nr_requests; i++) {
- xlock_acquire(&xsegbd_dev->blk_queue_pending.lock,
+ xlock_acquire(&xsegbd_dev->blk_queue_pending.lock,
xsegbd_dev->src_portno);
+ for (i = 0; i < xsegbd_dev->nr_requests; i++) {
if (!__xq_check(&xsegbd_dev->blk_queue_pending, i)) {
pending = &xsegbd_dev->blk_req_pending[i];
blkreq = pending->request;
pending->request = NULL;
comp = pending->comp;
pending->comp = NULL;
+ if (blkreq){
+ XSEGLOG("Cleaning up blkreq %lx [%d]", (unsigned long) blkreq, i);
+ blk_end_request_all(blkreq, -EIO);
+ }
+ if (comp){
+ XSEGLOG("Cleaning up comp %lx [%d]", (unsigned long) comp, i);
+ complete(comp);
+ }
+ __xq_append_tail(&xsegbd_dev->blk_queue_pending, i);
}
- if (blkreq)
- blk_end_request_all(blkreq, -EIO);
- if (comp)
- complete(comp);
- __xq_append_tail(&xsegbd_dev->blk_queue_pending, i);
- xlock_release(&xsegbd_dev->blk_queue_pending.lock);
}
+ xlock_release(&xsegbd_dev->blk_queue_pending.lock);
mutex_unlock(&xsegbd_mutex);
return ret;
xsegbd_dev->nr_requests *sizeof(struct xsegbd_pending),
GFP_KERNEL);
if (!xsegbd_dev->blk_req_pending)
- goto out_freeq;
+ goto out_bus;
XSEGLOG("joining segment");
"segdev",
xseg_callback );
if (!xsegbd_dev->xseg)
- goto out_freepending;
+ goto out_bus;
-
- XSEGLOG("binding to source port %u (destination %u)",
+ XSEGLOG("%s binding to source port %u (destination %u)", xsegbd_dev->target,
xsegbd_dev->src_portno, xsegbd_dev->dst_portno);
- port = xseg_bind_port(xsegbd_dev->xseg, xsegbd_dev->src_portno);
+ port = xseg_bind_port(xsegbd_dev->xseg, xsegbd_dev->src_portno, NULL);
if (!port) {
XSEGLOG("cannot bind to port");
ret = -EFAULT;
- goto out_xseg;
+ goto out_bus;
}
if (xsegbd_dev->src_portno != xseg_portno(xsegbd_dev->xseg, port)) {
XSEGLOG("portno != xsegbd_dev->src_portno");
BUG_ON(1);
ret = -EFAULT;
- goto out_xseg;
+ goto out_bus;
}
-
+ xseg_init_local_signal(xsegbd_dev->xseg, xsegbd_dev->src_portno);
+
+
/* make sure we don't get any requests until we're ready to handle them */
xseg_cancel_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
ret = xsegbd_dev_init(xsegbd_dev);
if (ret)
- goto out_xseg;
+ goto out_bus;
xseg_prepare_wait(xsegbd_dev->xseg, xseg_portno(xsegbd_dev->xseg, port));
return count;
-out_xseg:
- xseg_leave(xsegbd_dev->xseg);
-
-out_freepending:
- kfree(xsegbd_dev->blk_req_pending);
-
-out_freeq:
- xq_free(&xsegbd_dev->blk_queue_pending);
-
out_bus:
xsegbd_bus_del_dev(xsegbd_dev);
return ret;
projects / archipelago / blobdiff