[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH][linux 2.6.18] remove pointless error handling in scsiback



>>> On 02.07.14 at 09:26, <"jgross@xxxxxxxx".non-mime.internet> wrote:
> Changeset 610:3bcc901cbd7a added a retry logic for reporting LUNs to the
> frontend. The idea was to retry the operation if a LUN was added between
> counting the LUNs (needed for calculating the needed buffer size) and
> processing them (filling the buffer). The number of retries was limited to
> three. Reasoning was "one error is very unlikely, 4 of them are way off".
> 
> Unfortunately above changeset was wrong: The retry didn't include recounting
> the LUNs, so the buffer size would not be changed. The number of found was not
> resetted, so the first found LUN would again trigger the retry logic until the
> number of retries would be exhausted.
> 
> While one error is really very unlikely (there was no report of failure and
> as shown above, the first error would have led to failure at once), even in
> case of a correct retry logic following errors are not completely impossible:
> If a target with many LUNs is being added to the frontend, the single LUNs are
> added one after another. A parallel running "report LUNs" could see a fast
> increasing number of LUNs, so even a few retries might fail.
> 
> So just remove the retry logic again.

I'm still not convinced of this being the right action: I can see a point
in the original argumentation, so why don't we just fix it instead of
removing it (see below/attached)?

Jan

scsiback: fix retry handling in __report_luns()

Neither did lun_cnt get reset when invoking a retry here, nor did the
intermediate buffer get resized to fit the larger amount. Fix this, at
once
- adding spare extra buffer space growing with the number of retries,
- correct a couple of signed/unsigned issues here and its helper
  function __nr_luns_under_host(),
- drop pointless NULL checks before kfree() invocations (not only does
  kfree() deal fine with NULL, in the first of the two cases it was
  plain impossible for the pointer to be NULL).

Reported-by: Juergen Gross <jgross@xxxxxxxx>
Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>

--- a/drivers/xen/scsiback/emulate.c
+++ b/drivers/xen/scsiback/emulate.c
@@ -193,12 +193,12 @@ static int __maybe_unused __copy_from_sg
        return 0;
 }
 
-static int __nr_luns_under_host(struct vscsibk_info *info)
+static unsigned int __nr_luns_under_host(struct vscsibk_info *info)
 {
        struct v2p_entry *entry;
        struct list_head *head = &(info->v2p_entry_lists);
        unsigned long flags;
-       int lun_cnt = 0;
+       unsigned int lun_cnt = 0;
 
        spin_lock_irqsave(&info->v2p_lock, flags);
        list_for_each_entry(entry, head, l) {
@@ -213,6 +213,7 @@ static int __nr_luns_under_host(struct v
 /* REPORT LUNS Define*/
 #define VSCSI_REPORT_LUNS_HEADER       8
 #define VSCSI_REPORT_LUNS_RETRY                3
+#define VSCSI_REPORT_LUNS_SPARE                3
 
 /* quoted scsi_debug.c/resp_report_luns() */
 static void __report_luns(pending_req_t *pending_req, void *data)
@@ -229,8 +230,8 @@ static void __report_luns(pending_req_t 
        unsigned int req_bufflen = 0;
        unsigned int actual_len = 0;
        unsigned int retry_cnt = 0;
-       int select_report = (int)cmd[2];
-       int i, lun_cnt = 0, lun, upper, err = 0;
+       int err, select_report = (int)cmd[2];
+       unsigned int i, lun_cnt, lun, upper;
        
        struct v2p_entry *entry;
        struct list_head *head = &(info->v2p_entry_lists);
@@ -242,16 +243,18 @@ static void __report_luns(pending_req_t 
        if ((req_bufflen < 4) || (select_report != 0))
                goto fail;
 
-       alloc_luns = __nr_luns_under_host(info);
+retry:
+       alloc_luns = __nr_luns_under_host(info)
+                    + retry_cnt * VSCSI_REPORT_LUNS_SPARE;
        alloc_len  = sizeof(struct scsi_lun) * alloc_luns
                                + VSCSI_REPORT_LUNS_HEADER;
-retry:
        if ((buff = kzalloc(alloc_len, GFP_KERNEL)) == NULL) {
                printk(KERN_ERR "scsiback:%s kmalloc err\n", __FUNCTION__);
                goto fail;
        }
 
-       one_lun = (struct scsi_lun *) &buff[8];
+       one_lun = (struct scsi_lun *)&buff[VSCSI_REPORT_LUNS_HEADER];
+       lun_cnt = 0;
        spin_lock_irqsave(&info->v2p_lock, flags);
        list_for_each_entry(entry, head, l) {
                if ((entry->v.chn == channel) &&
@@ -264,8 +267,7 @@ retry:
 
                                if (retry_cnt < VSCSI_REPORT_LUNS_RETRY) {
                                        retry_cnt++;
-                                       if (buff)
-                                               kfree(buff);
+                                       kfree(buff);
                                        goto retry;
                                }
 
@@ -309,8 +311,7 @@ fail:
                INVALID_FIELD_IN_CDB, 0);
        pending_req->rslt  = check_condition_result;
        pending_req->resid = 0;
-       if (buff)
-               kfree(buff);
+       kfree(buff);
        return;
 }
 


Attachment: xen-scsiback-report-LUNs-retry.patch
Description: Text document

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.