[PATCH] block: Prevent deadlocks when switching elevators

[PATCH] block: Prevent deadlocks when switching elevators

[Damien Le Moal]

Commit af2814149883 ("block: freeze the queue in queue_attr_store")
changed queue_attr_store() to always freeze a sysfs attribute queue
before calling the attribute store() method, to ensure that no IOs are
in-flight when an attribute value is being updated.

However, this change created a potential deadlock situation for the
scheduler queue attribute as changing the queue elevator with
elv_iosched_store() can result in a call to request_module() if the user
requested module is not already registered. If the file of the requested
module is stored on the block device of the frozen queue, a deadlock
will happen as the read operations triggered by request_module() will
wait for the queue freeze to end.

Solve this issue by introducing the load_module method in struct
queue_sysfs_entry, and to calling this method function in
queue_attr_store() before freezing the attribute queue.
The macro definition QUEUE_RW_LOAD_MODULE_ENTRY() is added to define a
queue sysfs attribute that needs loading a module.

The definition of the scheduler atrribute is changed to using
QUEUE_RW_LOAD_MODULE_ENTRY(), with the function
elv_iosched_load_module() defined as the load_module method.
elv_iosched_store() can then be simplified to remove the call to
request_module().

Reported-by: Richard W.M. Jones <rjones@redhat.com>
Reported-by: Jiri Jaburek <jjaburek@redhat.com>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=219166
Fixes: af2814149883 ("block: freeze the queue in queue_attr_store")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
---
 block/blk-sysfs.c | 22 +++++++++++++++++++++-
 block/elevator.c  | 21 +++++++++++++++------
 block/elevator.h  |  2 ++
 3 files changed, 38 insertions(+), 7 deletions(-)

diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 60116d13cb80..e85941bec857 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -23,6 +23,7 @@
 struct queue_sysfs_entry {
 	struct attribute attr;
 	ssize_t (*show)(struct gendisk *disk, char *page);
+	int (*load_module)(struct gendisk *disk, const char *page, size_t count);
 	ssize_t (*store)(struct gendisk *disk, const char *page, size_t count);
 };
 
@@ -413,6 +414,14 @@ static struct queue_sysfs_entry _prefix##_entry = {	\
 	.store	= _prefix##_store,			\
 };
 
+#define QUEUE_RW_LOAD_MODULE_ENTRY(_prefix, _name)		\
+static struct queue_sysfs_entry _prefix##_entry = {		\
+	.attr		= { .name = _name, .mode = 0644 },	\
+	.show		= _prefix##_show,			\
+	.load_module	= _prefix##_load_module,		\
+	.store		= _prefix##_store,			\
+}
+
 QUEUE_RW_ENTRY(queue_requests, "nr_requests");
 QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
 QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
@@ -420,7 +429,7 @@ QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
 QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
 QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
 QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
-QUEUE_RW_ENTRY(elv_iosched, "scheduler");
+QUEUE_RW_LOAD_MODULE_ENTRY(elv_iosched, "scheduler");
 
 QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
 QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
@@ -670,6 +679,17 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
 	if (!entry->store)
 		return -EIO;
 
+	/*
+	 * If the attribute needs to load a module, do it before freezing the
+	 * queue to ensure that the module file can be read when the request
+	 * queue is the one for the device storing the module file.
+	 */
+	if (entry->load_module) {
+		res = entry->load_module(disk, page, length);
+		if (res)
+			return res;
+	}
+
 	blk_mq_freeze_queue(q);
 	mutex_lock(&q->sysfs_lock);
 	res = entry->store(disk, page, length);
diff --git a/block/elevator.c b/block/elevator.c
index f13d552a32c8..c355b55d0107 100644
--- a/block/elevator.c
+++ b/block/elevator.c
@@ -698,17 +698,26 @@ static int elevator_change(struct request_queue *q, const char *elevator_name)
 		return 0;
 
 	e = elevator_find_get(q, elevator_name);
-	if (!e) {
-		request_module("%s-iosched", elevator_name);
-		e = elevator_find_get(q, elevator_name);
-		if (!e)
-			return -EINVAL;
-	}
+	if (!e)
+		return -EINVAL;
 	ret = elevator_switch(q, e);
 	elevator_put(e);
 	return ret;
 }
 
+int elv_iosched_load_module(struct gendisk *disk, const char *buf,
+			    size_t count)
+{
+	char elevator_name[ELV_NAME_MAX];
+
+	if (!elv_support_iosched(disk->queue))
+		return -EOPNOTSUPP;
+
+	strscpy(elevator_name, buf, sizeof(elevator_name));
+
+	return request_module("%s-iosched", strstrip(elevator_name));
+}
+
 ssize_t elv_iosched_store(struct gendisk *disk, const char *buf,
 			  size_t count)
 {
diff --git a/block/elevator.h b/block/elevator.h
index 3fe18e1a8692..2a78544bf201 100644
--- a/block/elevator.h
+++ b/block/elevator.h
@@ -148,6 +148,8 @@ extern void elv_unregister(struct elevator_type *);
  * io scheduler sysfs switching
  */
 ssize_t elv_iosched_show(struct gendisk *disk, char *page);
+int elv_iosched_load_module(struct gendisk *disk, const char *page,
+			    size_t count);
 ssize_t elv_iosched_store(struct gendisk *disk, const char *page, size_t count);
 
 extern bool elv_bio_merge_ok(struct request *, struct bio *);
-- 
2.46.0

Re: [PATCH] block: Prevent deadlocks when switching elevators

[Richard W.M. Jones]

On Sun, Sep 08, 2024 at 09:07:04AM +0900, Damien Le Moal wrote:
> Commit af2814149883 ("block: freeze the queue in queue_attr_store")
> changed queue_attr_store() to always freeze a sysfs attribute queue
> before calling the attribute store() method, to ensure that no IOs are
> in-flight when an attribute value is being updated.
> 
> However, this change created a potential deadlock situation for the
> scheduler queue attribute as changing the queue elevator with
> elv_iosched_store() can result in a call to request_module() if the user
> requested module is not already registered. If the file of the requested
> module is stored on the block device of the frozen queue, a deadlock
> will happen as the read operations triggered by request_module() will
> wait for the queue freeze to end.
> 
> Solve this issue by introducing the load_module method in struct
> queue_sysfs_entry, and to calling this method function in
> queue_attr_store() before freezing the attribute queue.
> The macro definition QUEUE_RW_LOAD_MODULE_ENTRY() is added to define a
> queue sysfs attribute that needs loading a module.
> 
> The definition of the scheduler atrribute is changed to using
> QUEUE_RW_LOAD_MODULE_ENTRY(), with the function
> elv_iosched_load_module() defined as the load_module method.
> elv_iosched_store() can then be simplified to remove the call to
> request_module().
> 
> Reported-by: Richard W.M. Jones <rjones@redhat.com>
> Reported-by: Jiri Jaburek <jjaburek@redhat.com>
> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=219166
> Fixes: af2814149883 ("block: freeze the queue in queue_attr_store")
> Cc: stable@vger.kernel.org
> Signed-off-by: Damien Le Moal <dlemoal@kernel.org>

I tested this (using my "unreliable" test!), and it passed 5,000
iterations, so:

Tested-by: Richard W.M. Jones <rjones@redhat.com>

Thanks,

Rich.

> ---
>  block/blk-sysfs.c | 22 +++++++++++++++++++++-
>  block/elevator.c  | 21 +++++++++++++++------
>  block/elevator.h  |  2 ++
>  3 files changed, 38 insertions(+), 7 deletions(-)
> 
> diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
> index 60116d13cb80..e85941bec857 100644
> --- a/block/blk-sysfs.c
> +++ b/block/blk-sysfs.c
> @@ -23,6 +23,7 @@
>  struct queue_sysfs_entry {
>  	struct attribute attr;
>  	ssize_t (*show)(struct gendisk *disk, char *page);
> +	int (*load_module)(struct gendisk *disk, const char *page, size_t count);
>  	ssize_t (*store)(struct gendisk *disk, const char *page, size_t count);
>  };
>  
> @@ -413,6 +414,14 @@ static struct queue_sysfs_entry _prefix##_entry = {	\
>  	.store	= _prefix##_store,			\
>  };
>  
> +#define QUEUE_RW_LOAD_MODULE_ENTRY(_prefix, _name)		\
> +static struct queue_sysfs_entry _prefix##_entry = {		\
> +	.attr		= { .name = _name, .mode = 0644 },	\
> +	.show		= _prefix##_show,			\
> +	.load_module	= _prefix##_load_module,		\
> +	.store		= _prefix##_store,			\
> +}
> +
>  QUEUE_RW_ENTRY(queue_requests, "nr_requests");
>  QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
>  QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
> @@ -420,7 +429,7 @@ QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
>  QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
>  QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
>  QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
> -QUEUE_RW_ENTRY(elv_iosched, "scheduler");
> +QUEUE_RW_LOAD_MODULE_ENTRY(elv_iosched, "scheduler");
>  
>  QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
>  QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
> @@ -670,6 +679,17 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
>  	if (!entry->store)
>  		return -EIO;
>  
> +	/*
> +	 * If the attribute needs to load a module, do it before freezing the
> +	 * queue to ensure that the module file can be read when the request
> +	 * queue is the one for the device storing the module file.
> +	 */
> +	if (entry->load_module) {
> +		res = entry->load_module(disk, page, length);
> +		if (res)
> +			return res;
> +	}
> +
>  	blk_mq_freeze_queue(q);
>  	mutex_lock(&q->sysfs_lock);
>  	res = entry->store(disk, page, length);
> diff --git a/block/elevator.c b/block/elevator.c
> index f13d552a32c8..c355b55d0107 100644
> --- a/block/elevator.c
> +++ b/block/elevator.c
> @@ -698,17 +698,26 @@ static int elevator_change(struct request_queue *q, const char *elevator_name)
>  		return 0;
>  
>  	e = elevator_find_get(q, elevator_name);
> -	if (!e) {
> -		request_module("%s-iosched", elevator_name);
> -		e = elevator_find_get(q, elevator_name);
> -		if (!e)
> -			return -EINVAL;
> -	}
> +	if (!e)
> +		return -EINVAL;
>  	ret = elevator_switch(q, e);
>  	elevator_put(e);
>  	return ret;
>  }
>  
> +int elv_iosched_load_module(struct gendisk *disk, const char *buf,
> +			    size_t count)
> +{
> +	char elevator_name[ELV_NAME_MAX];
> +
> +	if (!elv_support_iosched(disk->queue))
> +		return -EOPNOTSUPP;
> +
> +	strscpy(elevator_name, buf, sizeof(elevator_name));
> +
> +	return request_module("%s-iosched", strstrip(elevator_name));
> +}
> +
>  ssize_t elv_iosched_store(struct gendisk *disk, const char *buf,
>  			  size_t count)
>  {
> diff --git a/block/elevator.h b/block/elevator.h
> index 3fe18e1a8692..2a78544bf201 100644
> --- a/block/elevator.h
> +++ b/block/elevator.h
> @@ -148,6 +148,8 @@ extern void elv_unregister(struct elevator_type *);
>   * io scheduler sysfs switching
>   */
>  ssize_t elv_iosched_show(struct gendisk *disk, char *page);
> +int elv_iosched_load_module(struct gendisk *disk, const char *page,
> +			    size_t count);
>  ssize_t elv_iosched_store(struct gendisk *disk, const char *page, size_t count);
>  
>  extern bool elv_bio_merge_ok(struct request *, struct bio *);
> -- 
> 2.46.0

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
nbdkit - Flexible, fast NBD server with plugins
https://gitlab.com/nbdkit/nbdkit

Re: [PATCH] block: Prevent deadlocks when switching elevators

[Jens Axboe]

On Sun, 08 Sep 2024 09:07:04 +0900, Damien Le Moal wrote:
> Commit af2814149883 ("block: freeze the queue in queue_attr_store")
> changed queue_attr_store() to always freeze a sysfs attribute queue
> before calling the attribute store() method, to ensure that no IOs are
> in-flight when an attribute value is being updated.
> 
> However, this change created a potential deadlock situation for the
> scheduler queue attribute as changing the queue elevator with
> elv_iosched_store() can result in a call to request_module() if the user
> requested module is not already registered. If the file of the requested
> module is stored on the block device of the frozen queue, a deadlock
> will happen as the read operations triggered by request_module() will
> wait for the queue freeze to end.
> 
> [...]

Applied, thanks!

[1/1] block: Prevent deadlocks when switching elevators
      commit: 3c031b721c0ee1d6237719a6a9d7487ef757487b

Best regards,
-- 
Jens Axboe

Re: [PATCH] block: Prevent deadlocks when switching elevators

[Jiri Slaby]

On 08. 09. 24, 2:07, Damien Le Moal wrote:
> Commit af2814149883 ("block: freeze the queue in queue_attr_store")
> changed queue_attr_store() to always freeze a sysfs attribute queue
> before calling the attribute store() method, to ensure that no IOs are
> in-flight when an attribute value is being updated.
> 
> However, this change created a potential deadlock situation for the
> scheduler queue attribute as changing the queue elevator with
> elv_iosched_store() can result in a call to request_module() if the user
> requested module is not already registered. If the file of the requested
> module is stored on the block device of the frozen queue, a deadlock
> will happen as the read operations triggered by request_module() will
> wait for the queue freeze to end.
> 
> Solve this issue by introducing the load_module method in struct
> queue_sysfs_entry, and to calling this method function in
> queue_attr_store() before freezing the attribute queue.
> The macro definition QUEUE_RW_LOAD_MODULE_ENTRY() is added to define a
> queue sysfs attribute that needs loading a module.
> 
> The definition of the scheduler atrribute is changed to using
> QUEUE_RW_LOAD_MODULE_ENTRY(), with the function
> elv_iosched_load_module() defined as the load_module method.
> elv_iosched_store() can then be simplified to remove the call to
> request_module().

Hi,

this broke udev rules for loop in 6.11:
 > loop1: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write 
ATTR{/sys/devices/virtual/block/loop1/queue/scheduler}="none", ignoring: 
No such file or directory
 > loop0: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write 
ATTR{/sys/devices/virtual/block/loop0/queue/scheduler}="none", ignoring: 
No such file or directory
 > loop5: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write 
ATTR{/sys/devices/virtual/block/loop5/queue/scheduler}="none", ignoring: 
No such file or directory
 > loop3: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write 
ATTR{/sys/devices/virtual/block/loop3/queue/scheduler}="none", ignoring: 
No such file or directory
 > loop2: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write 
ATTR{/sys/devices/virtual/block/loop2/queue/scheduler}="none", ignoring: 
No such file or directory
 > loop7: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write 
ATTR{/sys/devices/virtual/block/loop7/queue/scheduler}="none", ignoring: 
No such file or directory
 > loop4: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write 
ATTR{/sys/devices/virtual/block/loop4/queue/scheduler}="none", ignoring: 
No such file or directory
 > loop6: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write 
ATTR{/sys/devices/virtual/block/loop6/queue/scheduler}="none", ignoring: 
No such file or directory


60-io-scheduler.rules:
>      1  # Set optimal IO schedulers for HDD and SSD
>      2  # Copyright (c) 2021 SUSE LLC
>      3
>      4  # ## DO NOT EDIT. ##
>      5  # To modify the rules, copy this file to /etc/udev/rules.d/60-io-scheduler.rules
>      6  # and edit the copy.
>      7  # Please read the section "Tuning I/O performance" in the System Analysis and Tuning Guide
>      8  # from the SUSE Documentation.
>      9
>     10  # --- DO NOT EDIT THIS PART ----
>     11  SUBSYSTEM!="block", GOTO="scheduler_end"
>     12  ACTION!="add|change", GOTO="scheduler_end"
>     13  ENV{DEVTYPE}!="disk", GOTO="scheduler_end"
>     14  TEST!="%S%p/queue/scheduler", GOTO="scheduler_end"

It apparently exists here ^^^.

>     15
>     16  # For dm devices, the relevant events are "change" events, see 10-dm.rules
>     17  ACTION!="change", KERNEL=="dm-*", GOTO="scheduler_end"
>     18  # "none" with no brackets means scheduler isn't configurable
>     19  ATTR{queue/scheduler}=="none", GOTO="scheduler_end"
>     20  # keep our hands off zoned devices, the kernel auto-configures them
>     21  ATTR{queue/zoned}!="none", GOTO="scheduler_end"
>     22  # Enforce "none" for multipath components.
>     23  ENV{DM_MULTIPATH_DEVICE_PATH}=="1", ATTR{queue/scheduler}="none", GOTO="scheduler_end"
>     24  # Enforce "none" for loop devices
>     25  KERNEL=="loop[0-9]*", ATTR{queue/scheduler}="none", GOTO="scheduler_end"

But cannot be written to ^^^ because it does not exist. What the heck?

>     26
>     27  # --- EDIT BELOW HERE after copying to /etc/udev/rules.d ---
>     28
>     29  # Uncomment these if you want to force virtual devices to use no scheduler
>     30  # KERNEL=="vd[a-z]*", ATTR{queue/scheduler}="none", GOTO="scheduler_end"
>     31  # KERNEL=="xvd[a-z]*", ATTR{queue/scheduler}="none", GOTO="scheduler_end"
>     32
>     33  # Leave virtual devices untouched
>     34  KERNEL=="vd[a-z]*", GOTO="scheduler_end"
>     35  KERNEL=="xvd[a-z]*", GOTO="scheduler_end"
>     36
>     37  # 1. BFQ scheduler for single-queue HDD
>     38  ATTR{queue/rotational}!="0", TEST!="%S%p/mq/1", ATTR{queue/scheduler}="bfq", GOTO="scheduler_end"
>     39
>     40  # 2. BFQ scheduler for every HDD, including "real" multiqueue
>     41  # ATTR{queue/rotational}!="0", ATTR{queue/scheduler}="bfq", GOTO="scheduler_end"
>     42
>     43  # 3. For "real" multiqueue devices, the kernel defaults to no IO scheduling
>     44  # Uncomment this (and select your scheduler) if you need an IO scheduler for them
>     45  # TEST=="%S%p/mq/1", ATTR{queue/scheduler}="kyber", GOTO="scheduler_end"
>     46
>     47  # 4. BFQ scheduler for every device (uncomment if you need ionice or blk-cgroup features)
>     48  # ATTR{queue/scheduler}="bfq", GOTO="scheduler_end"
>     49
>     50  # 5. mq-deadline is the kernel default for devices with just one hardware queue
>     51  # ATTR{queue/scheduler}="mq-deadline"
>     52
>     53  # --- EDIT ABOVE HERE after copying to /etc/udev/rules.d ---
>     54  LABEL="scheduler_end"


Any ideas?





> Reported-by: Richard W.M. Jones <rjones@redhat.com>
> Reported-by: Jiri Jaburek <jjaburek@redhat.com>
> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=219166
> Fixes: af2814149883 ("block: freeze the queue in queue_attr_store")
> Cc: stable@vger.kernel.org
> Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
> ---
>   block/blk-sysfs.c | 22 +++++++++++++++++++++-
>   block/elevator.c  | 21 +++++++++++++++------
>   block/elevator.h  |  2 ++
>   3 files changed, 38 insertions(+), 7 deletions(-)
> 
> diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
> index 60116d13cb80..e85941bec857 100644
> --- a/block/blk-sysfs.c
> +++ b/block/blk-sysfs.c
> @@ -23,6 +23,7 @@
>   struct queue_sysfs_entry {
>   	struct attribute attr;
>   	ssize_t (*show)(struct gendisk *disk, char *page);
> +	int (*load_module)(struct gendisk *disk, const char *page, size_t count);
>   	ssize_t (*store)(struct gendisk *disk, const char *page, size_t count);
>   };
>   
> @@ -413,6 +414,14 @@ static struct queue_sysfs_entry _prefix##_entry = {	\
>   	.store	= _prefix##_store,			\
>   };
>   
> +#define QUEUE_RW_LOAD_MODULE_ENTRY(_prefix, _name)		\
> +static struct queue_sysfs_entry _prefix##_entry = {		\
> +	.attr		= { .name = _name, .mode = 0644 },	\
> +	.show		= _prefix##_show,			\
> +	.load_module	= _prefix##_load_module,		\
> +	.store		= _prefix##_store,			\
> +}
> +
>   QUEUE_RW_ENTRY(queue_requests, "nr_requests");
>   QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
>   QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
> @@ -420,7 +429,7 @@ QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
>   QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
>   QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
>   QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
> -QUEUE_RW_ENTRY(elv_iosched, "scheduler");
> +QUEUE_RW_LOAD_MODULE_ENTRY(elv_iosched, "scheduler");
>   
>   QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
>   QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
> @@ -670,6 +679,17 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
>   	if (!entry->store)
>   		return -EIO;
>   
> +	/*
> +	 * If the attribute needs to load a module, do it before freezing the
> +	 * queue to ensure that the module file can be read when the request
> +	 * queue is the one for the device storing the module file.
> +	 */
> +	if (entry->load_module) {
> +		res = entry->load_module(disk, page, length);
> +		if (res)
> +			return res;
> +	}
> +
>   	blk_mq_freeze_queue(q);
>   	mutex_lock(&q->sysfs_lock);
>   	res = entry->store(disk, page, length);
> diff --git a/block/elevator.c b/block/elevator.c
> index f13d552a32c8..c355b55d0107 100644
> --- a/block/elevator.c
> +++ b/block/elevator.c
> @@ -698,17 +698,26 @@ static int elevator_change(struct request_queue *q, const char *elevator_name)
>   		return 0;
>   
>   	e = elevator_find_get(q, elevator_name);
> -	if (!e) {
> -		request_module("%s-iosched", elevator_name);
> -		e = elevator_find_get(q, elevator_name);
> -		if (!e)
> -			return -EINVAL;
> -	}
> +	if (!e)
> +		return -EINVAL;
>   	ret = elevator_switch(q, e);
>   	elevator_put(e);
>   	return ret;
>   }
>   
> +int elv_iosched_load_module(struct gendisk *disk, const char *buf,
> +			    size_t count)
> +{
> +	char elevator_name[ELV_NAME_MAX];
> +
> +	if (!elv_support_iosched(disk->queue))
> +		return -EOPNOTSUPP;
> +
> +	strscpy(elevator_name, buf, sizeof(elevator_name));
> +
> +	return request_module("%s-iosched", strstrip(elevator_name));
> +}
> +
>   ssize_t elv_iosched_store(struct gendisk *disk, const char *buf,
>   			  size_t count)
>   {
> diff --git a/block/elevator.h b/block/elevator.h
> index 3fe18e1a8692..2a78544bf201 100644
> --- a/block/elevator.h
> +++ b/block/elevator.h
> @@ -148,6 +148,8 @@ extern void elv_unregister(struct elevator_type *);
>    * io scheduler sysfs switching
>    */
>   ssize_t elv_iosched_show(struct gendisk *disk, char *page);
> +int elv_iosched_load_module(struct gendisk *disk, const char *page,
> +			    size_t count);
>   ssize_t elv_iosched_store(struct gendisk *disk, const char *page, size_t count);
>   
>   extern bool elv_bio_merge_ok(struct request *, struct bio *);

thanks,
-- 
js
suse labs

Re: [PATCH] block: Prevent deadlocks when switching elevators

[Holger Hoffstätte]

On 2024-09-24 12:46, Jiri Slaby wrote:
> this broke udev rules for loop in 6.11:
>  > loop1: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write ATTR{/sys/devices/virtual/block/loop1/queue/scheduler}="none", ignoring: No such file or directory

(etc.)

Patch here but it's not in mainline yet:

https://lore.kernel.org/linux-block/20240917133231.134806-1-dlemoal@kernel.org/

cheers
Holger

Re: [PATCH] block: Prevent deadlocks when switching elevators

[Jens Axboe]

On 9/24/24 7:08 AM, Holger Hoffstätte wrote:
> On 2024-09-24 12:46, Jiri Slaby wrote:
>> this broke udev rules for loop in 6.11:
>>  > loop1: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write ATTR{/sys/devices/virtual/block/loop1/queue/scheduler}="none", ignoring: No such file or directory
> 
> (etc.)
> 
> Patch here but it's not in mainline yet:
> 
> https://lore.kernel.org/linux-block/20240917133231.134806-1-dlemoal@kernel.org/

It'll go upstream later this week.

-- 
Jens Axboe

Re: [PATCH] block: Prevent deadlocks when switching elevators

[Jiri Slaby]

On 24. 09. 24, 16:34, Jens Axboe wrote:
> On 9/24/24 7:08 AM, Holger Hoffstätte wrote:
>> On 2024-09-24 12:46, Jiri Slaby wrote:
>>> this broke udev rules for loop in 6.11:
>>>   > loop1: /usr/lib/udev/rules.d/60-io-scheduler.rules:25 Failed to write ATTR{/sys/devices/virtual/block/loop1/queue/scheduler}="none", ignoring: No such file or directory
>>
>> (etc.)
>>
>> Patch here but it's not in mainline yet:
>>
>> https://lore.kernel.org/linux-block/20240917133231.134806-1-dlemoal@kernel.org/
> 
> It'll go upstream later this week.

FWIW works for me.

thanks,
-- 
js