A follow-up on earlier discussions:
[PATCH] bsg referencing bus driver module
https://www.spinics.net/lists/linux-scsi/msg119631.html
[PATCH] Waiting for scsi_host_template release
https://www.spinics.net/lists/linux-scsi/msg119432.html
All these discussions are attempts to fix a crash after
SCSI transport driver unload if a user-mode process
holds a handle in BSG layer towards the unloaded driver
via SCSI mid-layer:
[16834.636216,07] Call Trace:
... scsi_proc_hostdir_rm
[16834.641944,07] [<ffffffff8141723f>] scsi_host_dev_release+0x3f/0x130
[16834.647740,07] [<ffffffff813e4f82>] device_release+0x32/0xa0
[16834.653423,07] [<ffffffff812dc6c7>] kobject_cleanup+0x77/0x190
[16834.659002,07] [<ffffffff812dc585>] kobject_put+0x25/0x50
[16834.664430,07] [<ffffffff813e5277>] put_device+0x17/0x20
[16834.669740,07] [<ffffffff812d0334>] bsg_kref_release_function+0x24/0x30
[16834.675007,07] [<ffffffff812d14a6>] bsg_release+0x166/0x1d0
[16834.680148,07] [<ffffffff8119ba2b>] __fput+0xcb/0x1d0
[16834.685156,07] [<ffffffff8119bb6e>] ____fput+0xe/0x10
[16834.690017,07] [<ffffffff81077476>] task_work_run+0x86/0xb0
[16834.694781,07] [<ffffffff81057043>] exit_to_usermode_loop+0x6b/0x9a
[16834.699466,07] [<ffffffff81002875>] syscall_return_slowpath+0x55/0x60
[16834.704110,07] [<ffffffff8172d615>] int_ret_from_sys_call+0x25/0x9f
The latest input from earlier discussions was to cut off access
to the unloaded driver at bsg_unregister_queue time by calling
blk_cleanup_queue. If we do that we still have to release
the reference to the parent device (otherwise we crash with
the same stack). The next logical step is, rather than maintaining
a "part-time" reference to be dropped early, we discard
referencing completely.
Discarding the reference turns out to be the only thing needed
to fix the problem: all transport drivers already do blk_cleanup_queue
before releasing their reference to the parent device.
From 7eaa9b43f0b99766b1d197044eb4d2e549d11a24 Mon Sep 17 00:00:00 2001
From: Anatoliy Glagolev <[email protected]>
Date: Wed, 16 May 2018 17:03:09 -0600
Subject: [PATCH] Bsg referencing parent device
Signed-off-by: Anatoliy Glagolev <[email protected]>
Bsg holding reference to a parent device may result in a crash
if user closes a bsg file handle after the parent device driver
has unloaded.
Holding a reference is not really needed: parent device must exist
between bsg_register_queue and bsg_unregister_queue. Before
the device goes away the caller does blk_cleanup_queue so that
all in-flight requests to the device are gone and all new requests
cannot pass beyond the queue. The queue itself is a refcounted
object and it will stay alive with a bsg file.
---
block/bsg.c | 37 +++++++++++++++++--------------------
1 file changed, 17 insertions(+), 20 deletions(-)
diff --git a/block/bsg.c b/block/bsg.c
index defa06c..f9e4b91 100644
--- a/block/bsg.c
+++ b/block/bsg.c
@@ -650,18 +650,6 @@ static struct bsg_device *bsg_alloc_device(void)
return bd;
}
-static void bsg_kref_release_function(struct kref *kref)
-{
- struct bsg_class_device *bcd =
- container_of(kref, struct bsg_class_device, ref);
- struct device *parent = bcd->parent;
-
- if (bcd->release)
- bcd->release(bcd->parent);
-
- put_device(parent);
-}
-
static int bsg_put_device(struct bsg_device *bd)
{
int ret = 0, do_free;
@@ -694,7 +682,6 @@ static int bsg_put_device(struct bsg_device *bd)
kfree(bd);
out:
- kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
if (do_free)
blk_put_queue(q);
return ret;
@@ -760,8 +747,6 @@ static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
*/
mutex_lock(&bsg_mutex);
bcd = idr_find(&bsg_minor_idr, iminor(inode));
- if (bcd)
- kref_get(&bcd->ref);
mutex_unlock(&bsg_mutex);
if (!bcd)
@@ -772,8 +757,6 @@ static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
return bd;
bd = bsg_add_device(inode, bcd->queue, file);
- if (IS_ERR(bd))
- kref_put(&bcd->ref, bsg_kref_release_function);
return bd;
}
@@ -902,6 +885,8 @@ static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
void bsg_unregister_queue(struct request_queue *q)
{
+ struct device *parent;
+ void (*release)(struct device *dev);
struct bsg_class_device *bcd = &q->bsg_dev;
if (!bcd->class_dev)
@@ -913,8 +898,21 @@ void bsg_unregister_queue(struct request_queue *q)
sysfs_remove_link(&q->kobj, "bsg");
device_unregister(bcd->class_dev);
bcd->class_dev = NULL;
- kref_put(&bcd->ref, bsg_kref_release_function);
+ parent = bcd->parent;
+ release = bcd->release;
+ bcd->parent = NULL;
+ bcd->release = NULL;
mutex_unlock(&bsg_mutex);
+
+ /*
+ * The caller of bsg_[un]register_queue must hold a reference
+ * to the parent device between ..register.. and ..unregister..
+ * so we do not maintain a refcount to parent here.
+ * Note: the caller is expected to blk_cleanup_queue(q)
+ * before releasing the reference to the parent device.
+ */
+ if (release)
+ release(parent);
}
EXPORT_SYMBOL_GPL(bsg_unregister_queue);
@@ -955,10 +953,9 @@ int bsg_register_queue(struct request_queue *q, struct device *parent,
bcd->minor = ret;
bcd->queue = q;
- bcd->parent = get_device(parent);
+ bcd->parent = parent;
bcd->release = release;
bcd->ops = ops;
- kref_init(&bcd->ref);
dev = MKDEV(bsg_major, bcd->minor);
class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
if (IS_ERR(class_dev)) {
--
1.9.1
The idea looks pretty reasonable, but once that is done we can get rid of
the ->release callback entirely and just handle it in the callers.
Something like the untested patch below:
diff --git a/block/bsg-lib.c b/block/bsg-lib.c
index fc2e5ff2c4b9..9419def8c017 100644
--- a/block/bsg-lib.c
+++ b/block/bsg-lib.c
@@ -303,11 +303,9 @@ static void bsg_exit_rq(struct request_queue *q, struct request *req)
* @name: device to give bsg device
* @job_fn: bsg job handler
* @dd_job_size: size of LLD data needed for each job
- * @release: @dev release function
*/
struct request_queue *bsg_setup_queue(struct device *dev, const char *name,
- bsg_job_fn *job_fn, int dd_job_size,
- void (*release)(struct device *))
+ bsg_job_fn *job_fn, int dd_job_size)
{
struct request_queue *q;
int ret;
@@ -331,7 +329,7 @@ struct request_queue *bsg_setup_queue(struct device *dev, const char *name,
blk_queue_softirq_done(q, bsg_softirq_done);
blk_queue_rq_timeout(q, BLK_DEFAULT_SG_TIMEOUT);
- ret = bsg_register_queue(q, dev, name, &bsg_transport_ops, release);
+ ret = bsg_register_queue(q, dev, name, &bsg_transport_ops);
if (ret) {
printk(KERN_ERR "%s: bsg interface failed to "
"initialize - register queue\n", dev->kobj.name);
diff --git a/block/bsg.c b/block/bsg.c
index defa06c11858..fe1e5632e5d1 100644
--- a/block/bsg.c
+++ b/block/bsg.c
@@ -650,18 +650,6 @@ static struct bsg_device *bsg_alloc_device(void)
return bd;
}
-static void bsg_kref_release_function(struct kref *kref)
-{
- struct bsg_class_device *bcd =
- container_of(kref, struct bsg_class_device, ref);
- struct device *parent = bcd->parent;
-
- if (bcd->release)
- bcd->release(bcd->parent);
-
- put_device(parent);
-}
-
static int bsg_put_device(struct bsg_device *bd)
{
int ret = 0, do_free;
@@ -694,7 +682,6 @@ static int bsg_put_device(struct bsg_device *bd)
kfree(bd);
out:
- kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
if (do_free)
blk_put_queue(q);
return ret;
@@ -760,8 +747,6 @@ static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
*/
mutex_lock(&bsg_mutex);
bcd = idr_find(&bsg_minor_idr, iminor(inode));
- if (bcd)
- kref_get(&bcd->ref);
mutex_unlock(&bsg_mutex);
if (!bcd)
@@ -772,8 +757,6 @@ static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
return bd;
bd = bsg_add_device(inode, bcd->queue, file);
- if (IS_ERR(bd))
- kref_put(&bcd->ref, bsg_kref_release_function);
return bd;
}
@@ -913,25 +896,17 @@ void bsg_unregister_queue(struct request_queue *q)
sysfs_remove_link(&q->kobj, "bsg");
device_unregister(bcd->class_dev);
bcd->class_dev = NULL;
- kref_put(&bcd->ref, bsg_kref_release_function);
mutex_unlock(&bsg_mutex);
}
EXPORT_SYMBOL_GPL(bsg_unregister_queue);
int bsg_register_queue(struct request_queue *q, struct device *parent,
- const char *name, const struct bsg_ops *ops,
- void (*release)(struct device *))
+ const char *name, const struct bsg_ops *ops)
{
struct bsg_class_device *bcd;
dev_t dev;
int ret;
struct device *class_dev = NULL;
- const char *devname;
-
- if (name)
- devname = name;
- else
- devname = dev_name(parent);
/*
* we need a proper transport to send commands, not a stacked device
@@ -955,15 +930,12 @@ int bsg_register_queue(struct request_queue *q, struct device *parent,
bcd->minor = ret;
bcd->queue = q;
- bcd->parent = get_device(parent);
- bcd->release = release;
bcd->ops = ops;
- kref_init(&bcd->ref);
dev = MKDEV(bsg_major, bcd->minor);
- class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
+ class_dev = device_create(bsg_class, parent, dev, NULL, "%s", name);
if (IS_ERR(class_dev)) {
ret = PTR_ERR(class_dev);
- goto put_dev;
+ goto idr_remove;
}
bcd->class_dev = class_dev;
@@ -978,8 +950,7 @@ int bsg_register_queue(struct request_queue *q, struct device *parent,
unregister_class_dev:
device_unregister(class_dev);
-put_dev:
- put_device(parent);
+idr_remove:
idr_remove(&bsg_minor_idr, bcd->minor);
unlock:
mutex_unlock(&bsg_mutex);
@@ -993,7 +964,7 @@ int bsg_scsi_register_queue(struct request_queue *q, struct device *parent)
return -EINVAL;
}
- return bsg_register_queue(q, parent, NULL, &bsg_scsi_ops, NULL);
+ return bsg_register_queue(q, parent, dev_name(parent), &bsg_scsi_ops);
}
EXPORT_SYMBOL_GPL(bsg_scsi_register_queue);
diff --git a/drivers/scsi/scsi_transport_fc.c b/drivers/scsi/scsi_transport_fc.c
index be3be0f9cb2d..d97f689b9278 100644
--- a/drivers/scsi/scsi_transport_fc.c
+++ b/drivers/scsi/scsi_transport_fc.c
@@ -3781,8 +3781,7 @@ fc_bsg_hostadd(struct Scsi_Host *shost, struct fc_host_attrs *fc_host)
snprintf(bsg_name, sizeof(bsg_name),
"fc_host%d", shost->host_no);
- q = bsg_setup_queue(dev, bsg_name, fc_bsg_dispatch, i->f->dd_bsg_size,
- NULL);
+ q = bsg_setup_queue(dev, bsg_name, fc_bsg_dispatch, i->f->dd_bsg_size);
if (IS_ERR(q)) {
dev_err(dev,
"fc_host%d: bsg interface failed to initialize - setup queue\n",
@@ -3827,8 +3826,8 @@ fc_bsg_rportadd(struct Scsi_Host *shost, struct fc_rport *rport)
if (!i->f->bsg_request)
return -ENOTSUPP;
- q = bsg_setup_queue(dev, NULL, fc_bsg_dispatch, i->f->dd_bsg_size,
- NULL);
+ q = bsg_setup_queue(dev, dev_name(dev), fc_bsg_dispatch,
+ i->f->dd_bsg_size);
if (IS_ERR(q)) {
dev_err(dev, "failed to setup bsg queue\n");
return PTR_ERR(q);
diff --git a/drivers/scsi/scsi_transport_iscsi.c b/drivers/scsi/scsi_transport_iscsi.c
index 65f6c94f2e9b..6fd2fe210fc3 100644
--- a/drivers/scsi/scsi_transport_iscsi.c
+++ b/drivers/scsi/scsi_transport_iscsi.c
@@ -1542,7 +1542,7 @@ iscsi_bsg_host_add(struct Scsi_Host *shost, struct iscsi_cls_host *ihost)
return -ENOTSUPP;
snprintf(bsg_name, sizeof(bsg_name), "iscsi_host%d", shost->host_no);
- q = bsg_setup_queue(dev, bsg_name, iscsi_bsg_host_dispatch, 0, NULL);
+ q = bsg_setup_queue(dev, bsg_name, iscsi_bsg_host_dispatch, 0);
if (IS_ERR(q)) {
shost_printk(KERN_ERR, shost, "bsg interface failed to "
"initialize - no request queue\n");
diff --git a/drivers/scsi/scsi_transport_sas.c b/drivers/scsi/scsi_transport_sas.c
index 08acbabfae07..e2953b416746 100644
--- a/drivers/scsi/scsi_transport_sas.c
+++ b/drivers/scsi/scsi_transport_sas.c
@@ -187,16 +187,6 @@ static int sas_smp_dispatch(struct bsg_job *job)
return 0;
}
-static void sas_host_release(struct device *dev)
-{
- struct Scsi_Host *shost = dev_to_shost(dev);
- struct sas_host_attrs *sas_host = to_sas_host_attrs(shost);
- struct request_queue *q = sas_host->q;
-
- if (q)
- blk_cleanup_queue(q);
-}
-
static int sas_bsg_initialize(struct Scsi_Host *shost, struct sas_rphy *rphy)
{
struct request_queue *q;
@@ -208,7 +198,7 @@ static int sas_bsg_initialize(struct Scsi_Host *shost, struct sas_rphy *rphy)
if (rphy) {
q = bsg_setup_queue(&rphy->dev, dev_name(&rphy->dev),
- sas_smp_dispatch, 0, NULL);
+ sas_smp_dispatch, 0);
if (IS_ERR(q))
return PTR_ERR(q);
rphy->q = q;
@@ -217,7 +207,7 @@ static int sas_bsg_initialize(struct Scsi_Host *shost, struct sas_rphy *rphy)
snprintf(name, sizeof(name), "sas_host%d", shost->host_no);
q = bsg_setup_queue(&shost->shost_gendev, name,
- sas_smp_dispatch, 0, sas_host_release);
+ sas_smp_dispatch, 0);
if (IS_ERR(q))
return PTR_ERR(q);
to_sas_host_attrs(shost)->q = q;
@@ -260,8 +250,11 @@ static int sas_host_remove(struct transport_container *tc, struct device *dev,
struct Scsi_Host *shost = dev_to_shost(dev);
struct request_queue *q = to_sas_host_attrs(shost)->q;
- if (q)
+ if (q) {
bsg_unregister_queue(q);
+ blk_cleanup_queue(q);
+ }
+
return 0;
}
diff --git a/include/linux/bsg-lib.h b/include/linux/bsg-lib.h
index 28a7ccc55c89..6aeaf6472665 100644
--- a/include/linux/bsg-lib.h
+++ b/include/linux/bsg-lib.h
@@ -72,8 +72,7 @@ struct bsg_job {
void bsg_job_done(struct bsg_job *job, int result,
unsigned int reply_payload_rcv_len);
struct request_queue *bsg_setup_queue(struct device *dev, const char *name,
- bsg_job_fn *job_fn, int dd_job_size,
- void (*release)(struct device *));
+ bsg_job_fn *job_fn, int dd_job_size);
void bsg_job_put(struct bsg_job *job);
int __must_check bsg_job_get(struct bsg_job *job);
diff --git a/include/linux/bsg.h b/include/linux/bsg.h
index 0c7dd9ceb139..dac37b6e00ec 100644
--- a/include/linux/bsg.h
+++ b/include/linux/bsg.h
@@ -17,17 +17,13 @@ struct bsg_ops {
struct bsg_class_device {
struct device *class_dev;
- struct device *parent;
int minor;
struct request_queue *queue;
- struct kref ref;
const struct bsg_ops *ops;
- void (*release)(struct device *);
};
int bsg_register_queue(struct request_queue *q, struct device *parent,
- const char *name, const struct bsg_ops *ops,
- void (*release)(struct device *));
+ const char *name, const struct bsg_ops *ops);
int bsg_scsi_register_queue(struct request_queue *q, struct device *parent);
void bsg_unregister_queue(struct request_queue *q);
#else