]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/commitdiff
[SCSI] Handle disk devices which can not process medium access commands
authorMartin K. Petersen <martin.petersen@oracle.com>
Thu, 9 Feb 2012 18:48:53 +0000 (13:48 -0500)
committerJames Bottomley <JBottomley@Parallels.com>
Sun, 19 Feb 2012 16:14:52 +0000 (10:14 -0600)
We have experienced several devices which fail in a fashion we do not
currently handle gracefully in SCSI. After a failure these devices will
respond to the SCSI primary command set (INQUIRY, TEST UNIT READY, etc.)
but any command accessing the storage medium will time out.

The following patch adds an callback that can be used by upper level
drivers to inspect the results of an error handling command. This in
turn has been used to implement additional checking in the SCSI disk
driver.

If a medium access command fails twice but TEST UNIT READY succeeds both
times in the subsequent error handling we will offline the device. The
maximum number of failed commands required to take a device offline can
be tweaked in sysfs.

Also add a new error flag to scsi_debug which allows this scenario to be
easily reproduced.

[jejb: fix up integer parsing to use kstrtouint]
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
drivers/scsi/scsi.c
drivers/scsi/scsi_debug.c
drivers/scsi/scsi_error.c
drivers/scsi/sd.c
drivers/scsi/sd.h
include/scsi/scsi_cmnd.h
include/scsi/scsi_driver.h

index 2aeb2e9c4d3bc708d5c12c7a117b021ffc160409..07322ecff90dc97e3a81d2f098293b8c6eaba98b 100644 (file)
@@ -782,12 +782,6 @@ static void scsi_done(struct scsi_cmnd *cmd)
        blk_complete_request(cmd->request);
 }
 
-/* Move this to a header if it becomes more generally useful */
-static struct scsi_driver *scsi_cmd_to_driver(struct scsi_cmnd *cmd)
-{
-       return *(struct scsi_driver **)cmd->request->rq_disk->private_data;
-}
-
 /**
  * scsi_finish_command - cleanup and pass command back to upper layer
  * @cmd: the command
index d2fd0efca565f0adef4adcca392211a77dc10ce6..8917154d96c7b39f447077f938b358eb075afcf3 100644 (file)
@@ -126,6 +126,7 @@ static const char * scsi_debug_version_date = "20100324";
 #define SCSI_DEBUG_OPT_TRANSPORT_ERR   16
 #define SCSI_DEBUG_OPT_DIF_ERR   32
 #define SCSI_DEBUG_OPT_DIX_ERR   64
+#define SCSI_DEBUG_OPT_MAC_TIMEOUT  128
 /* When "every_nth" > 0 then modulo "every_nth" commands:
  *   - a no response is simulated if SCSI_DEBUG_OPT_TIMEOUT is set
  *   - a RECOVERED_ERROR is simulated on successful read and write
@@ -3615,6 +3616,9 @@ int scsi_debug_queuecommand_lck(struct scsi_cmnd *SCpnt, done_funct_t done)
                        scsi_debug_every_nth = -1;
                if (SCSI_DEBUG_OPT_TIMEOUT & scsi_debug_opts)
                        return 0; /* ignore command causing timeout */
+               else if (SCSI_DEBUG_OPT_MAC_TIMEOUT & scsi_debug_opts &&
+                        scsi_medium_access_command(SCpnt))
+                       return 0; /* time out reads and writes */
                else if (SCSI_DEBUG_OPT_RECOVERED_ERR & scsi_debug_opts)
                        inj_recovered = 1; /* to reads and writes below */
                else if (SCSI_DEBUG_OPT_TRANSPORT_ERR & scsi_debug_opts)
index f66e90db3bee39bca8cfd7870e8d9c74475fba3b..2cfcbffa41fd559b075e7c75eae57e2c7f637488 100644 (file)
@@ -30,6 +30,7 @@
 #include <scsi/scsi_cmnd.h>
 #include <scsi/scsi_dbg.h>
 #include <scsi/scsi_device.h>
+#include <scsi/scsi_driver.h>
 #include <scsi/scsi_eh.h>
 #include <scsi/scsi_transport.h>
 #include <scsi/scsi_host.h>
@@ -141,11 +142,11 @@ enum blk_eh_timer_return scsi_times_out(struct request *req)
        else if (host->hostt->eh_timed_out)
                rtn = host->hostt->eh_timed_out(scmd);
 
+       scmd->result |= DID_TIME_OUT << 16;
+
        if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
-                    !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
-               scmd->result |= DID_TIME_OUT << 16;
+                    !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
                rtn = BLK_EH_HANDLED;
-       }
 
        return rtn;
 }
@@ -778,6 +779,7 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
                             int cmnd_size, int timeout, unsigned sense_bytes)
 {
        struct scsi_device *sdev = scmd->device;
+       struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
        struct Scsi_Host *shost = sdev->host;
        DECLARE_COMPLETION_ONSTACK(done);
        unsigned long timeleft;
@@ -832,6 +834,10 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
        }
 
        scsi_eh_restore_cmnd(scmd, &ses);
+
+       if (sdrv->eh_action)
+               rtn = sdrv->eh_action(scmd, cmnd, cmnd_size, rtn);
+
        return rtn;
 }
 
index 8c525aa1b8582a40c00687238811e743bc527d55..bd17cf8af01322d012d820e0f2b31cc3f0f73610 100644 (file)
@@ -107,6 +107,7 @@ static int sd_suspend(struct device *, pm_message_t state);
 static int sd_resume(struct device *);
 static void sd_rescan(struct device *);
 static int sd_done(struct scsi_cmnd *);
+static int sd_eh_action(struct scsi_cmnd *, unsigned char *, int, int);
 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
 static void scsi_disk_release(struct device *cdev);
 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
@@ -346,6 +347,31 @@ sd_store_provisioning_mode(struct device *dev, struct device_attribute *attr,
        return count;
 }
 
+static ssize_t
+sd_show_max_medium_access_timeouts(struct device *dev,
+                                  struct device_attribute *attr, char *buf)
+{
+       struct scsi_disk *sdkp = to_scsi_disk(dev);
+
+       return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
+}
+
+static ssize_t
+sd_store_max_medium_access_timeouts(struct device *dev,
+                                   struct device_attribute *attr,
+                                   const char *buf, size_t count)
+{
+       struct scsi_disk *sdkp = to_scsi_disk(dev);
+       int err;
+
+       if (!capable(CAP_SYS_ADMIN))
+               return -EACCES;
+
+       err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
+
+       return err ? err : count;
+}
+
 static struct device_attribute sd_disk_attrs[] = {
        __ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
               sd_store_cache_type),
@@ -360,6 +386,9 @@ static struct device_attribute sd_disk_attrs[] = {
        __ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
        __ATTR(provisioning_mode, S_IRUGO|S_IWUSR, sd_show_provisioning_mode,
               sd_store_provisioning_mode),
+       __ATTR(max_medium_access_timeouts, S_IRUGO|S_IWUSR,
+              sd_show_max_medium_access_timeouts,
+              sd_store_max_medium_access_timeouts),
        __ATTR_NULL,
 };
 
@@ -382,6 +411,7 @@ static struct scsi_driver sd_template = {
        },
        .rescan                 = sd_rescan,
        .done                   = sd_done,
+       .eh_action              = sd_eh_action,
 };
 
 /*
@@ -1313,6 +1343,55 @@ static const struct block_device_operations sd_fops = {
        .unlock_native_capacity = sd_unlock_native_capacity,
 };
 
+/**
+ *     sd_eh_action - error handling callback
+ *     @scmd:          sd-issued command that has failed
+ *     @eh_cmnd:       The command that was sent during error handling
+ *     @eh_cmnd_len:   Length of eh_cmnd in bytes
+ *     @eh_disp:       The recovery disposition suggested by the midlayer
+ *
+ *     This function is called by the SCSI midlayer upon completion of
+ *     an error handling command (TEST UNIT READY, START STOP UNIT,
+ *     etc.) The command sent to the device by the error handler is
+ *     stored in eh_cmnd. The result of sending the eh command is
+ *     passed in eh_disp.
+ **/
+static int sd_eh_action(struct scsi_cmnd *scmd, unsigned char *eh_cmnd,
+                       int eh_cmnd_len, int eh_disp)
+{
+       struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
+
+       if (!scsi_device_online(scmd->device) ||
+           !scsi_medium_access_command(scmd))
+               return eh_disp;
+
+       /*
+        * The device has timed out executing a medium access command.
+        * However, the TEST UNIT READY command sent during error
+        * handling completed successfully. Either the device is in the
+        * process of recovering or has it suffered an internal failure
+        * that prevents access to the storage medium.
+        */
+       if (host_byte(scmd->result) == DID_TIME_OUT && eh_disp == SUCCESS &&
+           eh_cmnd_len && eh_cmnd[0] == TEST_UNIT_READY)
+               sdkp->medium_access_timed_out++;
+
+       /*
+        * If the device keeps failing read/write commands but TEST UNIT
+        * READY always completes successfully we assume that medium
+        * access is no longer possible and take the device offline.
+        */
+       if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
+               scmd_printk(KERN_ERR, scmd,
+                           "Medium access timeout failure. Offlining disk!\n");
+               scsi_device_set_state(scmd->device, SDEV_OFFLINE);
+
+               return FAILED;
+       }
+
+       return eh_disp;
+}
+
 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
 {
        u64 start_lba = blk_rq_pos(scmd->request);
@@ -1402,6 +1481,8 @@ static int sd_done(struct scsi_cmnd *SCpnt)
            (!sense_valid || sense_deferred))
                goto out;
 
+       sdkp->medium_access_timed_out = 0;
+
        switch (sshdr.sense_key) {
        case HARDWARE_ERROR:
        case MEDIUM_ERROR:
@@ -2523,6 +2604,7 @@ static void sd_probe_async(void *data, async_cookie_t cookie)
        sdkp->RCD = 0;
        sdkp->ATO = 0;
        sdkp->first_scan = 1;
+       sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
 
        sd_revalidate_disk(gd);
 
index 4163f2910e3de300ca7d2c37a2ef38fc65d59206..f703f4827b6f290c6272c5e4b6474a73a4f66b0a 100644 (file)
@@ -20,6 +20,7 @@
  */
 #define SD_MAX_RETRIES         5
 #define SD_PASSTHROUGH_RETRIES 1
+#define SD_MAX_MEDIUM_TIMEOUTS 2
 
 /*
  * Size of the initial data buffer for mode and read capacity data
@@ -59,6 +60,8 @@ struct scsi_disk {
        u32             unmap_alignment;
        u32             index;
        unsigned int    physical_block_size;
+       unsigned int    max_medium_access_timeouts;
+       unsigned int    medium_access_timed_out;
        u8              media_present;
        u8              write_prot;
        u8              protection_type;/* Data Integrity Field */
@@ -88,6 +91,38 @@ static inline struct scsi_disk *scsi_disk(struct gendisk *disk)
                    (sdsk)->disk->disk_name, ##a) :                     \
        sdev_printk(prefix, (sdsk)->device, fmt, ##a)
 
+static inline int scsi_medium_access_command(struct scsi_cmnd *scmd)
+{
+       switch (scmd->cmnd[0]) {
+       case READ_6:
+       case READ_10:
+       case READ_12:
+       case READ_16:
+       case SYNCHRONIZE_CACHE:
+       case VERIFY:
+       case VERIFY_12:
+       case VERIFY_16:
+       case WRITE_6:
+       case WRITE_10:
+       case WRITE_12:
+       case WRITE_16:
+       case WRITE_SAME:
+       case WRITE_SAME_16:
+       case UNMAP:
+               return 1;
+       case VARIABLE_LENGTH_CMD:
+               switch (scmd->cmnd[9]) {
+               case READ_32:
+               case VERIFY_32:
+               case WRITE_32:
+               case WRITE_SAME_32:
+                       return 1;
+               }
+       }
+
+       return 0;
+}
+
 /*
  * A DIF-capable target device can be formatted with different
  * protection schemes.  Currently 0 through 3 are defined:
index 9be0128bf303a7a798b11c6042c185554259946d..377df4a28512bd7d97aad68b9a72d1f449f2cba8 100644 (file)
@@ -10,6 +10,7 @@
 
 struct Scsi_Host;
 struct scsi_device;
+struct scsi_driver;
 
 /*
  * MAX_COMMAND_SIZE is:
@@ -131,6 +132,11 @@ struct scsi_cmnd {
        unsigned char tag;      /* SCSI-II queued command tag */
 };
 
+static inline struct scsi_driver *scsi_cmd_to_driver(struct scsi_cmnd *cmd)
+{
+       return *(struct scsi_driver **)cmd->request->rq_disk->private_data;
+}
+
 extern struct scsi_cmnd *scsi_get_command(struct scsi_device *, gfp_t);
 extern struct scsi_cmnd *__scsi_get_command(struct Scsi_Host *, gfp_t);
 extern void scsi_put_command(struct scsi_cmnd *);
index 9fd6702f02e2ee540effd667c3af3e37f87e4b42..d443aa06a72287e4a1ca45c70243b38628686a30 100644 (file)
@@ -16,6 +16,7 @@ struct scsi_driver {
 
        void (*rescan)(struct device *);
        int (*done)(struct scsi_cmnd *);
+       int (*eh_action)(struct scsi_cmnd *, unsigned char *, int, int);
 };
 #define to_scsi_driver(drv) \
        container_of((drv), struct scsi_driver, gendrv)