SG_SCSI_RESET ioctl: add no_escalate values
Subsystems (2)
Douglas Gilbert
maintainerPATCH 1/1 SG_SCSI_RESET ioctl: add no_escalate values
Further to the thread titled: "[PATCH] SG_SCSI_RESET ioctl should only perform requested operation" by Jeremy Linton a patch is presented that adds "no_escalate" versions to the existing ioctl. This should not break any existing code.
This patches applies to lk 3.7.7 and lk 3.8.0-rc7 . I will extend sg_reset in the sg3_utils package to use it.
ChangeLog:
- modify SG_SCSI_RESET ioctl so the SG_SCSI_RESET_NO_ESCALATE value may be added to the existing values. If so the existing device->target->bus->host escalation does not occur.
- modify scsi_reset_provider() in the scsi_error.c file in a similar way to support this new functionality.
drivers/scsi/scsi_error.c +10 -0
@@ -2013,8 +2013,18 @@ scsi_reset_provider(struct scsi_device *dev, int flag)break ;/* FALLTHROUGH */ case SCSI_TRY_RESET_HOST : +case SCSI_TRY_RESET_HOST +SCSI_TRY_RESET_NO_ESCALATE : rtn =scsi_try_host_reset (scmd);break ; +case SCSI_TRY_RESET_DEVICE +SCSI_TRY_RESET_NO_ESCALATE : + rtn =scsi_try_bus_device_reset (scmd); +break ; +case SCSI_TRY_RESET_TARGET +SCSI_TRY_RESET_NO_ESCALATE : + rtn =scsi_try_target_reset (scmd); +break ; +case SCSI_TRY_RESET_BUS +SCSI_TRY_RESET_NO_ESCALATE : + rtn =scsi_try_bus_reset (scmd); +break ;default : rtn =FAILED ; }
include/scsi/scsi_eh.h +5 -0
@@ -62,11 +62,16 @@ extern void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq);/* * Reset request from external source +* Note: if SCSI_TRY_RESET_DEVICE fails then it will escalate to +* SCSI_TRY_RESET_TARGET which if it fails will escalate to +* SCSI_TRY_RESET_BUS which if it fails will escalate to SCSI_TRY_RESET_HOST. +* To prevent escalation add SCSI_TRY_RESET_NO_ESCALATE. */ #define SCSI_TRY_RESET_DEVICE 1#define SCSI_TRY_RESET_BUS 2#define SCSI_TRY_RESET_HOST 3#define SCSI_TRY_RESET_TARGET 4 +#define SCSI_TRY_RESET_NO_ESCALATE 0x100/* may add to prior defines */ extern int scsi_reset_provider (struct scsi_device *,int );
drivers/scsi/sg.c +11 -6
@@ -810,7 +810,7 @@ sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) {void __user *p = (void __user *)arg;int __user *ip = p; -int result, val, read_only; +int result, val, val2, read_only;Sg_device *sdp;Sg_fd *sfp;Sg_request *srp;
@@ -1045,27 +1045,32 @@ sg_ioctl(struct file *filp, unsigned int cmd_in, unsigned long arg) result =get_user (val, ip);if (result)return result; +if (val >=SG_SCSI_RESET_NO_ESCALATE ) { + val -=SG_SCSI_RESET_NO_ESCALATE ; + val2 =SCSI_TRY_RESET_NO_ESCALATE ; + }else + val2 =0 ;if (SG_SCSI_RESET_NOTHING == val)return 0 ;switch (val) {case SG_SCSI_RESET_DEVICE : - val =SCSI_TRY_RESET_DEVICE ; + val2 +=SCSI_TRY_RESET_DEVICE ;break ;case SG_SCSI_RESET_TARGET : - val =SCSI_TRY_RESET_TARGET ; + val2 +=SCSI_TRY_RESET_TARGET ;break ;case SG_SCSI_RESET_BUS : - val =SCSI_TRY_RESET_BUS ; + val2 +=SCSI_TRY_RESET_BUS ;break ;case SG_SCSI_RESET_HOST : - val =SCSI_TRY_RESET_HOST ; + val2 +=SCSI_TRY_RESET_HOST ;break ;default :return -EINVAL ; }if (!capable (CAP_SYS_ADMIN ) || !capable (CAP_SYS_RAWIO ))return -EACCES ; -return (scsi_reset_provider (sdp->device, val) == +return (scsi_reset_provider (sdp->device, val2) ==SUCCESS ) ?0 : -EIO ;case SCSI_IOCTL_SEND_COMMAND :if (sdp->detached)
include/scsi/sg.h +4 -1
@@ -204,12 +204,15 @@ typedef struct sg_req_info { /* used by SG_GET_REQUEST_TABLE ioctl() *//* Returns -EBUSY if occupied. 3rd argument pointer to int (see next) */ #define SG_SCSI_RESET 0x2284 -/* Associated values that can be given to SG_SCSI_RESET follow */ +/* Associated values that can be given to SG_SCSI_RESET follow. +* SG_SCSI_RESET_NO_ESCALATE may be added to the _DEVICE, _TARGET, _BUS +* or _HOST reset value so only that action is attempted. */ #define SG_SCSI_RESET_NOTHING 0#define SG_SCSI_RESET_DEVICE 1#define SG_SCSI_RESET_BUS 2#define SG_SCSI_RESET_HOST 3#define SG_SCSI_RESET_TARGET 4 +#define SG_SCSI_RESET_NO_ESCALATE 0x100/* synchronous SCSI command ioctl, (only in version 3 interface) */ #define SG_IO 0x2285/* similar effect as write() followed by read() */
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Quoted 4 lines · Douglas Gilbert
Further to the thread titled: "[PATCH]
SG_SCSI_RESETioctl should only perform requested operation" by Jeremy Linton a patch is presented that adds "no_escalate" versions to the existing ioctl. This should not break any existing code.
Looks good, I will apply it here and try it out.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (MingW32)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
iQEcBAEBAgAGBQJRHpCDAAoJEL5i86xrzcy7lvoIAK5pxe8rdx8WinxG/NqhLkRb +537Ln3asshkWcH385gaa1JzizoWhiS1+H+tGcLdEwPDwvSSGGqtUAn/qtxB64cM F5X0XqjS/XVup7GrkcoQ4ZzXG2rdI6Lb5gwbpb98QeCGzaVKfRMvJCCtXSIFuPXA S7gL0Xl5d5KapPCWVRpucE05XVaAZq2vrxC3/8hDwB8+3HYUW4gUTcwSAM0pvcUb d7w984jlcFHVxT3Yk+ZHuQ3QOKifbqWBE9WlsgqkPnGjlAK8Q19NQsD6/C9Q9wTg OhMjj9HKHOlcBE65e/xKlcncTXjaa1qcLfa94hOukjSQBkf0vqJH3XeWuACESE0= =Oqr2
-----END PGP SIGNATURE-----
Quoted 8 lines · Douglas Gilbert
Further to the thread titled: "[PATCH]
SG_SCSI_RESETioctl should only perform requested operation" by Jeremy Linton a patch is presented that adds "no_escalate" versions to the existing ioctl. This should not break any existing code.This patches applies to lk 3.7.7 and lk 3.8.0-rc7 . I will extend
sg_resetin thesg3_utilspackage to use it.
Quoted 8 lines
ChangeLog:
- modifySG_SCSI_RESETioctl so theSG_SCSI_RESET_NO_ESCALATEvalue may be added to the existing values. If so the existingdevice->target->bus->hostescalation does not occur.
- modifyscsi_reset_provider()in thescsi_error.cfile in a similar way to support this new functionality.Signed-off-by Douglas Gilbert <[email protected]>
Some drivers rely on more invasive eh callbacks to be called if they return FAILED in a eh callbacks. Is there a way for drivers to tell scsi-ml it needs the old behavior?
Douglas Gilbert
maintainerQuoted 16 lines · Mike Christie
Quoted 15 lines · Douglas Gilbert
Further to the thread titled: "[PATCH]
SG_SCSI_RESETioctl should only perform requested operation" by Jeremy Linton a patch is presented that adds "no_escalate" versions to the existing ioctl. This should not break any existing code.This patches applies to lk 3.7.7 and lk 3.8.0-rc7 . I will extend
sg_resetin thesg3_utilspackage to use it.ChangeLog:
- modifySG_SCSI_RESETioctl so theSG_SCSI_RESET_NO_ESCALATEvalue may be added to the existing values. If so the existingdevice->target->bus->hostescalation does not occur.
- modifyscsi_reset_provider()in thescsi_error.cfile in a similar way to support this new functionality.
Quoted 5 lines
Quoted 1 line
Signed-off-by Douglas Gilbert <[email protected]>Some drivers rely on more invasive eh callbacks to be called if they return FAILED in a eh callbacks. Is there a way for drivers to tell scsi-ml it needs the old behavior?
Mike,
The old behaviour hasn't changed both at the ioctl(SG_SCSI_RESET) level and the underlying kernel scsi_reset_provider() function. In both cases extra values have been added: to third argument of the ioctl and the second argument ('flag') of the scsi_reset_provider() function. The existing values will do exactly the same thing (i.e. escalate) with the same return values.
The only way it is different is that values that were previously errors (precisely 0x101 to 0x104) now cause a non-escalating device(LU)/target/bus reset.
Doug Gilbert
Quoted 33 lines · Douglas Gilbert
Quoted 16 lines · Mike Christie
Quoted 15 lines · Douglas Gilbert
Further to the thread titled: "[PATCH]
SG_SCSI_RESETioctl should only perform requested operation" by Jeremy Linton a patch is presented that adds "no_escalate" versions to the existing ioctl. This should not break any existing code.This patches applies to lk 3.7.7 and lk 3.8.0-rc7 . I will extend
sg_resetin thesg3_utilspackage to use it.ChangeLog:
- modifySG_SCSI_RESETioctl so theSG_SCSI_RESET_NO_ESCALATEvalue may be added to the existing values. If so the existingdevice->target->bus->hostescalation does not occur.
- modifyscsi_reset_provider()in thescsi_error.cfile in a similar way to support this new functionality.Quoted 5 lines
Quoted 1 line
Signed-off-by Douglas Gilbert <[email protected]>Some drivers rely on more invasive eh callbacks to be called if they return FAILED in a eh callbacks. Is there a way for drivers to tell scsi-ml it needs the old behavior?
Mike,
The old behaviour hasn't changed both at theioctl(SG_SCSI_RESET)level and the underlying kernelscsi_reset_provider()function. In both cases extra values have been added: to third argument of the ioctl and the second argument ('flag') of thescsi_reset_provider()function. The existing values will do exactly the same thing (i.e. escalate) with the same return values.
Quoted 3 lines
The only way it is different is that values that were previously errors (precisely
0x101to0x104) now cause a non-escalating device(LU)/target/bus reset.
Sorry for the confusion. I should have written is there a way for drivers to tell scsi-ml it only supports the old behavior and does not support the new calls? If not then I am guessing this is mandatory to support and we need to fix the drivers right?
If userspace sends a command that performs the non-escalting *reset to some drivers, it is going to leave them in a state where they may not process IO or will crash.
Quoted 5 lines · Michael Christie
Quoted 2 lines
Quoted 2 lines
Some drivers rely on more invasive eh callbacks to be called if they return FAILED in a eh callbacks.
If userspace sends a command that performs the non-escalting *reset to some drivers, it is going to leave them in a state where they may not process IO or will crash.
Actually, I take those 2 comments back. I think it is just some drivers abort callbacks that might be issues, so we should be safe. I did not look at all drivers though.