On Saturday 03 November 2007 10:17, Heiko Carstens wrote:
> > +static void zfcp_fsf_req_latency(struct zfcp_fsf_req *fsf_req)
> > +{
> > + struct fsf_qual_latency_info *lat_inf;
> > + struct zfcp_unit *unit;
> > +
> > + lat_inf = &fsf_req->qtcb->prefix.prot_status_qual.latency_info;
> > + unit = fsf_req->unit;
> > +
> > + switch (fsf_req->qtcb->bottom.io.data_direction) {
> > + case FSF_DATADIR_READ:
> > + unit->latencies.read.channel += lat_inf->channel_lat;
> > + unit->latencies.read.fabric += lat_inf->fabric_lat;
> > + unit->latencies.read.counter++;
> > + break;
> > + case FSF_DATADIR_WRITE:
> > + unit->latencies.write.channel += lat_inf->channel_lat;
> > + unit->latencies.write.fabric += lat_inf->fabric_lat;
> > + unit->latencies.write.counter++;
> > + break;
> > + case FSF_DATADIR_CMND:
> > + unit->latencies.cmd.channel += lat_inf->channel_lat;
> > + unit->latencies.cmd.fabric += lat_inf->fabric_lat;
> > + unit->latencies.cmd.counter++;
> > + break;
> > + }
> > +}
>
> These statistics are concurrently updated from several cpus without
> any locking. That looks like a bug.
>
> > +zfcp_sysfs_unit_##_name##_latency_show(struct device *dev, \
> > + struct device_attribute *attr, \
> > + char *buf) { \
> > + struct scsi_device *sdev = to_scsi_device(dev); \
> > + struct zfcp_unit *unit = sdev->hostdata; \
> > + struct zfcp_latencies *lat = &unit->latencies; \
> > + struct zfcp_adapter *adapter = unit->port->adapter; \
> > + \
> > + return sprintf(buf, "%u %u %u\n", \
> > + lat->_name.fabric * adapter->timer_ticks / 1000, \
> > + lat->_name.channel * adapter->timer_ticks / 1000,\
> > + lat->_name.counter); \
>
> In addition they can be read concurrently from userspace without any
> locking... Since you put several values together in the output I assume
> this is supposed to be some sort of snapshot, which it currently isn't.
>
> > +static int
> > +zfcp_sysfs_adapter_ex_config(struct class_device *cdev,
> > + struct fsf_qtcb_bottom_config **qtcb_config)
> > +{
> > + struct Scsi_Host *scsi_host = class_to_shost(cdev);
> > + struct zfcp_adapter *adapter = (struct zfcp_adapter *)
> > + scsi_host->hostdata[0];
> > +
> > + if (!(adapter->adapter_features & FSF_FEATURE_MEASUREMENT_DATA)) {
> > + ZFCP_LOG_NORMAL("error: Enhanced measurement feature not "
> > + "supported");
> > + return -EOPNOTSUPP;
> > + }
> > +
> > + *qtcb_config = kzalloc(sizeof(struct fsf_qtcb_bottom_config),
> > + GFP_KERNEL);
> > + if (!*qtcb_config)
> > + return -ENOMEM;
> > +
> > + return zfcp_fsf_exchange_config_data_sync(adapter, *qtcb_config);
> > +}
> > +
> > +static ssize_t
> > +zfcp_sysfs_adapter_request_show(struct class_device *cdev, char *buf)
> > +{
> > + struct fsf_qtcb_bottom_config *qtcb_config;
> > + int retval;
> > +
> > + retval = zfcp_sysfs_adapter_ex_config(cdev, &qtcb_config);
> > +
> > + if (!retval)
> > + retval = sprintf(buf, "%lu %lu %lu\n",
> > + qtcb_config->stat_info.input_req,
> > + qtcb_config->stat_info.output_req,
> > + qtcb_config->stat_info.control_req);
> > +
> > + kfree(qtcb_config);
> > + return retval;
> > +}
>
> You're going to call kfree with some random value if the adapter doesn't
> support the measurement data feature.
>
Ok, valid points.
I changed the patch to meet the above described issues.
Before I will post the modified version I want to do some testing (and an internal review).
The updated version will follow soon (hopefully this week), thanks for reviewing.
Cheers Swen
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
