At 09/30/2016 01:37 AM, David Sterba wrote:
On Wed, Sep 28, 2016 at 04:30:03PM +0800, Qu Wenruo wrote:
Add a new function raid5_gen_result() to calculate raid5 parity or
recover data stripe.
Since now that raid6.c handles both raid5 and raid6, rename it to
raid56.c.
Please split changes in this patch to the following:
- rename the file
- error handling of memory allocation failures
- the actual fix
A test would be very velcome, for all the cases the code handles, 2
devices, and more. But I'm not sure if we have support for that in the
testing suite.
Makes sense.
I'll split first and try if I can create some test cases for RAID5/6 codes.
But the later work may be delayed since I'm working on user-space scrub.
(Which will lead to kernel scrub fix).
Thanks,
Qu
@@ -107,3 +108,47 @@ void raid6_gen_syndrome(int disks, size_t bytes, void
**ptrs)
}
}
+static void xor_range(void *src, void *dst, size_t size)
+{
+ while (size) {
+ *(unsigned long *) dst ^= *(unsigned long *) src;
This could lead to unaligned access, please update the types and
possibly add some sanity checks (alignemnt, length).
+ src += sizeof(unsigned long);
+ dst += sizeof(unsigned long);
+ size -= sizeof(unsigned long);
+ }
+}
+
+/*
+ * Generate desired data/parity for RAID5
+ *
+ * @nr_devs: Total number of devices, including parity
+ * @stripe_len: Stripe length
+ * @data: Data, with special layout:
+ * data[0]: Data stripe 0
+ * data[nr_devs-2]: Last data stripe
+ * data[nr_devs-1]: RAID5 parity
+ * @dest: To generate which data. should follow above data layout
+ */
+int raid5_gen_result(int nr_devs, size_t stripe_len, int dest, void **data)
+{
+ int i;
+ char *buf = data[dest];
+
+ if (dest >= nr_devs || nr_devs < 2) {
+ error("invalid parameter for %s", __func__);
+ return -EINVAL;
+ }
+ /* Quich hack, 2 devs RAID5 is just RAID1, no need to calculate */
This is not a hack IMO, it could be a shortcut, a special case, an
optimization.
+ if (nr_devs == 2) {
+ memcpy(data[dest], data[1 - dest], stripe_len);
+ return 0;
+ }
+ /* Just in case */
Such comment is not very helpful.
+ memset(buf, 0, stripe_len);
+ for (i = 0; i < nr_devs; i++) {
+ if (i == dest)
+ continue;
+ xor_range(data[i], buf, stripe_len);
+ }
+ return 0;
+}
diff --git a/volumes.c b/volumes.c
index da79751..718e67c 100644
--- a/volumes.c
+++ b/volumes.c
@@ -2108,12 +2108,14 @@ int write_raid56_with_parity(struct btrfs_fs_info *info,
{
struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
int i;
- int j;
int ret;
int alloc_size = eb->len;
+ void **pointers;
I see you're moving existing code, so if you're going to fix the types,
please do that in a separate patch as well.
- ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
- BUG_ON(!ebs);
+ ebs = malloc(sizeof(*ebs) * multi->num_stripes);
+ pointers = malloc(sizeof(void *) * multi->num_stripes);
+ if (!ebs || !pointers)
+ return -ENOMEM;
if (stripe_len > alloc_size)
alloc_size = stripe_len;
@@ -2143,12 +2145,6 @@ int write_raid56_with_parity(struct btrfs_fs_info *info,
q_eb = new_eb;
}
if (q_eb) {
- void **pointers;
-
- pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
- GFP_NOFS);
- BUG_ON(!pointers);
-
ebs[multi->num_stripes - 2] = p_eb;
ebs[multi->num_stripes - 1] = q_eb;
@@ -2159,17 +2155,14 @@ int write_raid56_with_parity(struct btrfs_fs_info *info,
kfree(pointers);
} else {
ebs[multi->num_stripes - 1] = p_eb;
- memcpy(p_eb->data, ebs[0]->data, stripe_len);
- for (j = 1; j < multi->num_stripes - 1; j++) {
- for (i = 0; i < stripe_len; i += sizeof(u64)) {
- u64 p_eb_data;
- u64 ebs_data;
-
- p_eb_data = get_unaligned_64(p_eb->data + i);
- ebs_data = get_unaligned_64(ebs[j]->data + i);
- p_eb_data ^= ebs_data;
- put_unaligned_64(p_eb_data, p_eb->data + i);
- }
+ for (i = 0; i < multi->num_stripes; i++)
+ pointers[i] = ebs[i]->data;
+ ret = raid5_gen_result(multi->num_stripes, stripe_len,
+ multi->num_stripes - 1, pointers);
+ if (ret < 0) {
+ free(ebs);
+ free(pointers);
+ return ret;
}
}
@@ -2180,7 +2173,8 @@ int write_raid56_with_parity(struct btrfs_fs_info *info,
kfree(ebs[i]);
}
- kfree(ebs);
+ free(ebs);
+ free(pointers);
return 0;
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
