scsi: smartpqi: grab scsi device ref in slave_configure()
Subsystems (2)
PATCH 1/2 scsi: smartpqi: grab scsi device ref in slave_configure()
From: Martin Wilck <[email protected]>
We have observed kernel crashes caused by the smartpqi driver holding a pointer to a struct scsi_device that had been removed already. Fix this by getting and holding a ref for the device as long as the driver uses it.
Use a lock to avoid races between device probing and removal.
--- drivers/scsi/smartpqi/smartpqi.h | 1 + drivers/scsi/smartpqi/smartpqi_init.c | 122 +++++++++++++++++++++----- 2 files changed, 103 insertions(+), 20 deletions(-)
drivers/scsi/smartpqi/smartpqi.h +1 -0
@@ -954,6 +954,7 @@ struct pqi_scsi_dev {struct raid_map *raid_map;/* RAID bypass map */ struct pqi_sas_port *sas_port; +spinlock_t sdev_lock;/* protect access to sdev */ struct scsi_device *sdev;struct list_head scsi_device_list_entry;
drivers/scsi/smartpqi/smartpqi_init.c +102 -20
176 lines not on this page · raw message
Signature · 1 line
2.26.2
PATCH 2/2 scsi: smartpqi: check sdev in pqi_scsi_find_entry
From: Martin Wilck <[email protected]>
If a scsi device has been destroyed e.g. using the sysfs "delete" attribute, subsequent host rescans won't re-discover it. This patch makes it work at least via the smartqpi-specific "rescan" sysfs attribute.
--- drivers/scsi/smartpqi/smartpqi_init.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/scsi/smartpqi/smartpqi_init.c +2 -1
@@ -1612,7 +1612,8 @@ static enum pqi_find_result pqi_scsi_find_entry(struct pqi_ctrl_info *ctrl_info, device->scsi3addr)) { *matching_device = device;if (pqi_device_equal (device_to_find, device)) { -if (device_to_find->volume_offline) +if (device_to_find->volume_offline || + !pqi_get_scsi_device (device))return DEVICE_CHANGED ;return DEVICE_SAME ; }
Signature · 1 line
2.26.2
RE: [PATCH 2/2] scsi: smartpqi: check sdev in pqi_scsi_find_entry
Quoted 40 lines
-----Original Message-----
From: [email protected] [mailto:linux-scsi- [email protected]] On Behalf Of [email protected]
Sent: Wednesday, 17 June 2020 1:32 AM
To: Don Brace <[email protected]>; Martin K. Petersen <[email protected]>
Cc: [email protected]; [email protected]; Martin Wilck <[email protected]>
Subject: [PATCH 2/2] scsi: smartpqi: check sdev inpqi_scsi_find_entryFrom: Martin Wilck <[email protected]>
If a scsi device has been destroyed e.g. using the sysfs "delete" attribute, subsequent host rescans won't re-discover it. This patch makes it work at least via the smartqpi-specific "rescan" sysfs attribute.
Signed-off-by Martin Wilck <[email protected]>--- drivers/scsi/smartpqi/smartpqi_init.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)drivers/scsi/smartpqi/smartpqi_init.c +2
-1@@ -1612,7 +1612,8 @@ static enum pqi_find_result pqi_scsi_find_entry(struct pqi_ctrl_info *ctrl_info, device->scsi3addr)) { *matching_device = device;if (pqi_device_equal (device_to_find, device)) { -if (device_to_find->volume_offline) +if (device_to_find->volume_offline || + !pqi_get_scsi_device (device))return DEVICE_CHANGED ;return DEVICE_SAME ;}
--
2.26.2
Quoted 240 lines
-----Original Message-----
From: [email protected] [mailto:linux-scsi- [email protected]] On Behalf Of [email protected]
Sent: Wednesday, 17 June 2020 1:32 AM
To: Don Brace <[email protected]>; Martin K. Petersen <[email protected]>
Cc: [email protected]; [email protected]; Martin Wilck <[email protected]>
Subject: [PATCH 1/2] scsi: smartpqi: grab scsi device ref inslave_configure()From: Martin Wilck <[email protected]>
We have observed kernel crashes caused by the smartpqi driver holding a pointer to a struct
scsi_devicethat had been removed already. Fix this by getting and holding a ref for the device as long as the driver uses it.Use a lock to avoid races between device probing and removal.
Signed-off-by Martin Wilck <[email protected]>--- drivers/scsi/smartpqi/smartpqi.h | 1 + drivers/scsi/smartpqi/smartpqi_init.c | 122 +++++++++++++++++++++----- 2 files changed, 103 insertions(+), 20 deletions(-)drivers/scsi/smartpqi/smartpqi.h +1
-0@@ -954,6 +954,7 @@ struct pqi_scsi_dev {struct raid_map *raid_map;/* RAID bypass map */ struct pqi_sas_port *sas_port; +spinlock_t sdev_lock;/* protect access to sdev */ struct scsi_device *sdev;struct list_head scsi_device_list_entry;drivers/scsi/smartpqi/smartpqi_init.c +12
-0@@ -1514,6 +1514,18 @@ static int pqi_add_device(struct pqi_ctrl_info *ctrl_info,#define PQI_PENDING_IO_TIMEOUT_SECS 20 +static inline struct scsi_device * +pqi_get_scsi_device (struct pqi_scsi_dev *device) +{ +unsigned long flags; +struct scsi_device *sdev; + +spin_lock_irqsave (&device->sdev_lock, flags); + sdev = device->sdev; +spin_unlock_irqrestore (&device->sdev_lock, flags); +return sdev; +} +static inline void pqi_remove_device (struct pqi_ctrl_info *ctrl_info,struct pqi_scsi_dev *device){
drivers/scsi/smartpqi/smartpqi_init.c +21
-4@@ -1530,9 +1542,26 @@ static inline void pqi_remove_device(struct pqi_ctrl_info *ctrl_info, device->target, device->lun, atomic_read(&device->scsi_cmds_outstanding)); -if (pqi_is_logical_device (device)) -scsi_remove_device (device->sdev); -else +if (pqi_is_logical_device (device)) { +struct scsi_device *sdev; +unsigned long flags; + +spin_lock_irqsave (&device->sdev_lock, flags); + sdev = device->sdev; +if (sdev) +get_device (&sdev->sdev_gendev); +spin_unlock_irqrestore (&device->sdev_lock, flags); + +/* +* As scsi_remove_device() will call pqi_slave_destroy(), +* we can't keep the sdev_lock held. But we've taken a ref, +* so sdev will not go away under us. +*/ +if (sdev) { +scsi_remove_device (sdev); +put_device (&sdev->sdev_gendev); + } + }else pqi_remove_sas_device (device); }@@ -1749,7 +1778,7 @@ static inline bool pqi_is_device_added(struct pqi_scsi_dev *device)if (device->is_expander_smp_device)return device->sas_port !=NULL ; -return device->sdev !=NULL ; +return pqi_get_scsi_device (device) !=NULL ; }static void pqi_update_device_list(struct
pqi_ctrl_info*ctrl_info,drivers/scsi/smartpqi/smartpqi_init.c +18
-5@@ -1861,11 +1890,24 @@ static void pqi_update_device_list(structpqi_ctrl_info *ctrl_info, */ list_for_each_entry (device, &ctrl_info->scsi_device_list, scsi_device_list_entry) { -if (device->sdev && device->queue_depth != - device->advertised_queue_depth) { - device->advertised_queue_depth = device- >queue_depth; -scsi_change_queue_depth (device->sdev, - device->advertised_queue_depth); +struct scsi_device *sdev; +unsigned long flags; + +spin_lock_irqsave (&device->sdev_lock, flags); + sdev = device->sdev; +if (sdev) +get_device (&sdev->sdev_gendev); +spin_unlock_irqrestore (&device->sdev_lock, flags); + +if (sdev) { +if (device->queue_depth != + device->advertised_queue_depth) { + device->advertised_queue_depth = + device->queue_depth; +scsi_change_queue_depth (sdev, + device->advertised_queue_depth); + } +put_device (&sdev->sdev_gendev); }drivers/scsi/smartpqi/smartpqi_init.c +1
-0}@@ -2082,6 +2124,7 @@ static int pqi_update_scsi_devices(struct pqi_ctrl_info *ctrl_info) device =list_first_entry (&new_device_list_head, struct pqi_scsi_dev, new_device_list_entry); +spin_lock_init (&device->sdev_lock);memcpy (device->scsi3addr, scsi3addr,sizeof (device- >scsi3addr));drivers/scsi/smartpqi/smartpqi_init.c +16
-11device->is_physical_device = is_physical_device;if (is_physical_device) {@@ -5785,6 +5828,7 @@ static int pqi_slave_alloc(struct scsi_device *sdev)struct pqi_ctrl_info *ctrl_info;struct scsi_target *starget;struct sas_rphy *rphy; +int ret; ctrl_info =shost_to_hba (sdev->host);@@ -5806,23 +5850,59 @@ static int pqi_slave_alloc(struct scsi_device *sdev)if (device) { sdev->hostdata = device; - device->sdev = sdev; -if (device->queue_depth) { - device->advertised_queue_depth = device- >queue_depth; -scsi_change_queue_depth (sdev, - device->advertised_queue_depth); - } -if (pqi_is_logical_device (device)) -pqi_disable_write_same (sdev); -else - sdev->allow_restart =1 ; - } + ret =0 ; + }else + ret = -ENXIO ;spin_unlock_irqrestore (&ctrl_info->scsi_device_list_lock, flags); +return ret; +} + +static int pqi_slave_configure (struct scsi_device *sdev) +{ +struct pqi_scsi_dev *device = sdev->hostdata; +unsigned long flags; + +/* +* Grab a ref to the SCSI device, lest it be removed under us. It will +* be dropped in pqi_slave_destroy(). +* Don't use scsi_device_get() here, we'd increment our own use count
drivers/scsi/smartpqi/smartpqi_init.c +34
-0+*/ +if (!get_device (&sdev->sdev_gendev)) +return -ENXIO ; + +spin_lock_irqsave (&device->sdev_lock, flags); + device->sdev = sdev; +spin_unlock_irqrestore (&device->sdev_lock, flags); + +if (device->queue_depth) { + device->advertised_queue_depth = device->queue_depth; +scsi_change_queue_depth (sdev, + device->advertised_queue_depth); + } +if (pqi_is_logical_device (device)) +pqi_disable_write_same (sdev); +else + sdev->allow_restart =1 ;return 0 ; } +static void pqi_slave_destroy (struct scsi_device *sdev) +{ +struct pqi_scsi_dev *device = sdev->hostdata; +unsigned long flags; + +if (!device) +return ; + +spin_lock_irqsave (&device->sdev_lock, flags); + sdev = device->sdev; + device->sdev =NULL ; +spin_unlock_irqrestore (&device->sdev_lock, flags); +put_device (&sdev->sdev_gendev); +} +static int pqi_map_queues (struct Scsi_Host *shost) {struct pqi_ctrl_info *ctrl_info =shost_to_hba (shost);@@ -6501,6 +6581,8 @@ static struct scsi_host_template pqi_driver_template = { .eh_device_reset_handler = pqi_eh_device_reset_handler, .ioctl = pqi_ioctl, .slave_alloc = pqi_slave_alloc, + .slave_configure = pqi_slave_configure, + .slave_destroy = pqi_slave_destroy, .map_queues = pqi_map_queues, .sdev_attrs = pqi_sdev_attrs,.shost_attrs =
pqi_shost_attrs,
--
2.26.2