From: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx>
This patch converts the incoming struct scatterlist array pointer logic
for transport_map_sg_to_mem() to use the include/linux/scatterlist.h
sg_next() macro in order to properly support incoming linked scatterlists.
An issue was observed recently with TCM_Loop + FILEIO backstores using v2.6.35
struct scsi_host_template->sg_tablesize=256. 128+128 linked scatterlists
coming in via struct scsi_cmnd->>sdb.table.sgl where generating bogus struct se_mem
because of the assumption of a single contigious scatterlist array for the incoming
memory to be mapped zero-copy to struct se_mem.
The single array walking scatterlist code in transport_map_sg_to_mem() can
historically be attributed to the matching RAMDISK_DR algortihms existing before
Jens added proper array linked scatterlist support into v2.6.24-rc series way
back in October 2007. For modern TCM purposes, this patch does the "right thing"
and uses Jen's sg_next() macros when walking physical memory for incoming
SHT->queuecommand() I/O.
Thanks Jens!
Signed-off-by: Nicholas A. Bellinger <nab@xxxxxxxxxxxxxxx>
---
drivers/target/target_core_transport.c | 49 +++++++++++++++++++++----------
1 files changed, 33 insertions(+), 16 deletions(-)
diff --git a/drivers/target/target_core_transport.c b/drivers/target/target_core_transport.c
index efb351a..a6b5706 100644
--- a/drivers/target/target_core_transport.c
+++ b/drivers/target/target_core_transport.c
@@ -6686,14 +6686,14 @@ int transport_map_sg_to_mem(
u32 *task_offset)
{
struct se_mem *se_mem;
- struct scatterlist *sg_s;
- u32 j = 0, saved_task_offset = 0, task_size = cmd->data_length;
+ struct scatterlist *sg;
+ u32 sg_count = 0, saved_task_offset = 0, task_size = cmd->data_length;
if (!in_mem) {
printk(KERN_ERR "No source scatterlist\n");
return -1;
}
- sg_s = (struct scatterlist *)in_mem;
+ sg = (struct scatterlist *)in_mem;
while (task_size) {
se_mem = kmem_cache_zalloc(se_mem_cache, GFP_KERNEL);
@@ -6702,14 +6702,19 @@ int transport_map_sg_to_mem(
return -1;
}
INIT_LIST_HEAD(&se_mem->se_list);
+ DEBUG_MEM("sg_to_mem: Starting loop with task_size: %u"
+ " sg_page: %p offset: %d length: %d\n", task_size,
+ sg_page(sg), sg->offset, sg->length);
if (*task_offset == 0) {
- se_mem->se_page = sg_page(&sg_s[j]);
- se_mem->se_off = sg_s[j].offset;
+ se_mem->se_page = sg_page(sg);
+ se_mem->se_off = sg->offset;
- if (task_size >= sg_s[j].length)
- se_mem->se_len = sg_s[j++].length;
- else {
+ if (task_size >= sg->length) {
+ se_mem->se_len = sg->length;
+ sg = sg_next(sg);
+ sg_count++;
+ } else {
se_mem->se_len = task_size;
task_size -= se_mem->se_len;
@@ -6723,10 +6728,10 @@ int transport_map_sg_to_mem(
if (saved_task_offset)
*task_offset = saved_task_offset;
} else {
- se_mem->se_page = sg_page(&sg_s[j]);
- se_mem->se_off = (*task_offset + sg_s[j].offset);
+ se_mem->se_page = sg_page(sg);
+ se_mem->se_off = (*task_offset + sg->offset);
- if ((sg_s[j].length - *task_offset) > task_size) {
+ if ((sg->length - *task_offset) > task_size) {
se_mem->se_len = task_size;
task_size -= se_mem->se_len;
@@ -6734,21 +6739,33 @@ int transport_map_sg_to_mem(
*task_offset += se_mem->se_len;
goto next;
}
- } else
- se_mem->se_len = (sg_s[j++].length -
- *task_offset);
+ } else {
+ se_mem->se_len = (sg->length - *task_offset);
+ sg = sg_next(sg);
+ sg_count++;
+ }
saved_task_offset = *task_offset;
*task_offset = 0;
}
task_size -= se_mem->se_len;
next:
+ DEBUG_MEM("sg_to_mem: *se_mem_cnt: %u task_size: %u, *task_offset: %u "
+ "saved_task_offset: %d\n", *se_mem_cnt, task_size, *task_offset,
+ saved_task_offset);
+
+ DEBUG_MEM("sg_to_mem: Final se_page: %p se_off: %d se_len: %d\n",
+ se_mem->se_page, se_mem->se_off, se_mem->se_len);
+
list_add_tail(&se_mem->se_list, se_mem_list);
(*se_mem_cnt)++;
}
DEBUG_MEM("task[0] - Mapped(%u) struct scatterlist segments to(%u)"
- " struct se_mem\n", j, *se_mem_cnt);
+ " struct se_mem\n", sg_count, *se_mem_cnt);
+
+ if (sg_count != *se_mem_cnt)
+ BUG();
return 0;
}
--
1.5.6.5
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
