summaryrefslogtreecommitdiff
path: root/drivers/block/rbd.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/block/rbd.c')
-rw-r--r--drivers/block/rbd.c129
1 files changed, 88 insertions, 41 deletions
diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
index 4ad2ad9a5bb0..cb1db2979d3d 100644
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -397,15 +397,19 @@ static ssize_t rbd_remove(struct bus_type *bus, const char *buf,
static int rbd_dev_image_probe(struct rbd_device *rbd_dev, bool mapping);
static void rbd_spec_put(struct rbd_spec *spec);
-static struct bus_attribute rbd_bus_attrs[] = {
- __ATTR(add, S_IWUSR, NULL, rbd_add),
- __ATTR(remove, S_IWUSR, NULL, rbd_remove),
- __ATTR_NULL
+static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
+static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
+
+static struct attribute *rbd_bus_attrs[] = {
+ &bus_attr_add.attr,
+ &bus_attr_remove.attr,
+ NULL,
};
+ATTRIBUTE_GROUPS(rbd_bus);
static struct bus_type rbd_bus_type = {
.name = "rbd",
- .bus_attrs = rbd_bus_attrs,
+ .bus_groups = rbd_bus_groups,
};
static void rbd_root_dev_release(struct device *dev)
@@ -927,12 +931,14 @@ static const char *rbd_dev_v1_snap_name(struct rbd_device *rbd_dev,
u64 snap_id)
{
u32 which;
+ const char *snap_name;
which = rbd_dev_snap_index(rbd_dev, snap_id);
if (which == BAD_SNAP_INDEX)
- return NULL;
+ return ERR_PTR(-ENOENT);
- return _rbd_dev_v1_snap_name(rbd_dev, which);
+ snap_name = _rbd_dev_v1_snap_name(rbd_dev, which);
+ return snap_name ? snap_name : ERR_PTR(-ENOMEM);
}
static const char *rbd_snap_name(struct rbd_device *rbd_dev, u64 snap_id)
@@ -1557,11 +1563,12 @@ rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
obj_request, obj_request->img_request, obj_request->result,
xferred, length);
/*
- * ENOENT means a hole in the image. We zero-fill the
- * entire length of the request. A short read also implies
- * zero-fill to the end of the request. Either way we
- * update the xferred count to indicate the whole request
- * was satisfied.
+ * ENOENT means a hole in the image. We zero-fill the entire
+ * length of the request. A short read also implies zero-fill
+ * to the end of the request. An error requires the whole
+ * length of the request to be reported finished with an error
+ * to the block layer. In each case we update the xferred
+ * count to indicate the whole request was satisfied.
*/
rbd_assert(obj_request->type != OBJ_REQUEST_NODATA);
if (obj_request->result == -ENOENT) {
@@ -1570,14 +1577,13 @@ rbd_img_obj_request_read_callback(struct rbd_obj_request *obj_request)
else
zero_pages(obj_request->pages, 0, length);
obj_request->result = 0;
- obj_request->xferred = length;
} else if (xferred < length && !obj_request->result) {
if (obj_request->type == OBJ_REQUEST_BIO)
zero_bio_chain(obj_request->bio_list, xferred);
else
zero_pages(obj_request->pages, xferred, length);
- obj_request->xferred = length;
}
+ obj_request->xferred = length;
obj_request_done_set(obj_request);
}
@@ -2163,9 +2169,9 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,
struct rbd_obj_request *obj_request = NULL;
struct rbd_obj_request *next_obj_request;
bool write_request = img_request_write_test(img_request);
- struct bio *bio_list = 0;
+ struct bio *bio_list = NULL;
unsigned int bio_offset = 0;
- struct page **pages = 0;
+ struct page **pages = NULL;
u64 img_offset;
u64 resid;
u16 opcode;
@@ -2203,6 +2209,11 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,
rbd_segment_name_free(object_name);
if (!obj_request)
goto out_unwind;
+ /*
+ * set obj_request->img_request before creating the
+ * osd_request so that it gets the right snapc
+ */
+ rbd_img_obj_request_add(img_request, obj_request);
if (type == OBJ_REQUEST_BIO) {
unsigned int clone_size;
@@ -2244,11 +2255,6 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,
obj_request->pages, length,
offset & ~PAGE_MASK, false, false);
- /*
- * set obj_request->img_request before formatting
- * the osd_request so that it gets the right snapc
- */
- rbd_img_obj_request_add(img_request, obj_request);
if (write_request)
rbd_osd_req_format_write(obj_request);
else
@@ -2808,7 +2814,7 @@ out_err:
obj_request_done_set(obj_request);
}
-static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id)
+static int rbd_obj_notify_ack_sync(struct rbd_device *rbd_dev, u64 notify_id)
{
struct rbd_obj_request *obj_request;
struct ceph_osd_client *osdc = &rbd_dev->rbd_client->client->osdc;
@@ -2823,16 +2829,17 @@ static int rbd_obj_notify_ack(struct rbd_device *rbd_dev, u64 notify_id)
obj_request->osd_req = rbd_osd_req_create(rbd_dev, false, obj_request);
if (!obj_request->osd_req)
goto out;
- obj_request->callback = rbd_obj_request_put;
osd_req_op_watch_init(obj_request->osd_req, 0, CEPH_OSD_OP_NOTIFY_ACK,
notify_id, 0, 0);
rbd_osd_req_format_read(obj_request);
ret = rbd_obj_request_submit(osdc, obj_request);
-out:
if (ret)
- rbd_obj_request_put(obj_request);
+ goto out;
+ ret = rbd_obj_request_wait(obj_request);
+out:
+ rbd_obj_request_put(obj_request);
return ret;
}
@@ -2852,7 +2859,7 @@ static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
if (ret)
rbd_warn(rbd_dev, "header refresh error (%d)\n", ret);
- rbd_obj_notify_ack(rbd_dev, notify_id);
+ rbd_obj_notify_ack_sync(rbd_dev, notify_id);
}
/*
@@ -3324,6 +3331,31 @@ static void rbd_exists_validate(struct rbd_device *rbd_dev)
clear_bit(RBD_DEV_FLAG_EXISTS, &rbd_dev->flags);
}
+static void rbd_dev_update_size(struct rbd_device *rbd_dev)
+{
+ sector_t size;
+ bool removing;
+
+ /*
+ * Don't hold the lock while doing disk operations,
+ * or lock ordering will conflict with the bdev mutex via:
+ * rbd_add() -> blkdev_get() -> rbd_open()
+ */
+ spin_lock_irq(&rbd_dev->lock);
+ removing = test_bit(RBD_DEV_FLAG_REMOVING, &rbd_dev->flags);
+ spin_unlock_irq(&rbd_dev->lock);
+ /*
+ * If the device is being removed, rbd_dev->disk has
+ * been destroyed, so don't try to update its size
+ */
+ if (!removing) {
+ size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
+ dout("setting size to %llu sectors", (unsigned long long)size);
+ set_capacity(rbd_dev->disk, size);
+ revalidate_disk(rbd_dev->disk);
+ }
+}
+
static int rbd_dev_refresh(struct rbd_device *rbd_dev)
{
u64 mapping_size;
@@ -3343,12 +3375,7 @@ static int rbd_dev_refresh(struct rbd_device *rbd_dev)
up_write(&rbd_dev->header_rwsem);
if (mapping_size != rbd_dev->mapping.size) {
- sector_t size;
-
- size = (sector_t)rbd_dev->mapping.size / SECTOR_SIZE;
- dout("setting size to %llu sectors", (unsigned long long)size);
- set_capacity(rbd_dev->disk, size);
- revalidate_disk(rbd_dev->disk);
+ rbd_dev_update_size(rbd_dev);
}
return ret;
@@ -3702,12 +3729,14 @@ static int _rbd_dev_v2_snap_size(struct rbd_device *rbd_dev, u64 snap_id,
if (ret < sizeof (size_buf))
return -ERANGE;
- if (order)
+ if (order) {
*order = size_buf.order;
+ dout(" order %u", (unsigned int)*order);
+ }
*snap_size = le64_to_cpu(size_buf.size);
- dout(" snap_id 0x%016llx order = %u, snap_size = %llu\n",
- (unsigned long long)snap_id, (unsigned int)*order,
+ dout(" snap_id 0x%016llx snap_size = %llu\n",
+ (unsigned long long)snap_id,
(unsigned long long)*snap_size);
return 0;
@@ -4055,8 +4084,13 @@ static u64 rbd_v2_snap_id_by_name(struct rbd_device *rbd_dev, const char *name)
snap_id = snapc->snaps[which];
snap_name = rbd_dev_v2_snap_name(rbd_dev, snap_id);
- if (IS_ERR(snap_name))
- break;
+ if (IS_ERR(snap_name)) {
+ /* ignore no-longer existing snapshots */
+ if (PTR_ERR(snap_name) == -ENOENT)
+ continue;
+ else
+ break;
+ }
found = !strcmp(name, snap_name);
kfree(snap_name);
}
@@ -4135,8 +4169,8 @@ static int rbd_dev_spec_update(struct rbd_device *rbd_dev)
/* Look up the snapshot name, and make a copy */
snap_name = rbd_snap_name(rbd_dev, spec->snap_id);
- if (!snap_name) {
- ret = -ENOMEM;
+ if (IS_ERR(snap_name)) {
+ ret = PTR_ERR(snap_name);
goto out_err;
}
@@ -5126,7 +5160,7 @@ static ssize_t rbd_remove(struct bus_type *bus,
bool already = false;
int ret;
- ret = strict_strtoul(buf, 10, &ul);
+ ret = kstrtoul(buf, 10, &ul);
if (ret)
return ret;
@@ -5157,10 +5191,23 @@ static ssize_t rbd_remove(struct bus_type *bus,
if (ret < 0 || already)
return ret;
- rbd_bus_del_dev(rbd_dev);
ret = rbd_dev_header_watch_sync(rbd_dev, false);
if (ret)
rbd_warn(rbd_dev, "failed to cancel watch event (%d)\n", ret);
+
+ /*
+ * flush remaining watch callbacks - these must be complete
+ * before the osd_client is shutdown
+ */
+ dout("%s: flushing notifies", __func__);
+ ceph_osdc_flush_notifies(&rbd_dev->rbd_client->client->osdc);
+ /*
+ * Don't free anything from rbd_dev->disk until after all
+ * notifies are completely processed. Otherwise
+ * rbd_bus_del_dev() will race with rbd_watch_cb(), resulting
+ * in a potential use after free of rbd_dev->disk or rbd_dev.
+ */
+ rbd_bus_del_dev(rbd_dev);
rbd_dev_image_release(rbd_dev);
module_put(THIS_MODULE);
generated by cgit v1.2.3 (git 2.0.4) at 2024-05-23 12:38:22 +0000