Hi all
proud to present support for raid6 in ORE.
changes since RFC:
Nothing changed. one checkpatch fix. And this time all patches actually
compile (thanks to Tigran)
I have by now tested better. RAID6 code was not actually tested heavily
but RAID5 was, and there are no known regressions to RAID5 so I'm pushing
this for next merge.
.I.E: Yes I did run all tests on RAID6 but user-mode utils do not support
scrubbing and checking of RAID6 yet. So I'm not confident that it is
100% cosher. But I do have a XOR checker and RAID5 is just as good as
current code. So I'm pushing this now and any RAID6 specific problems
can be fixed when they are found.
These simple patches will enable raid6 using the Kernel's raid6_pq engine
for support under exofs and pnfs-objects. There is nothing needed to do
at exofs and pnfs-obj. Just fire your mkfs.exofs with --raid=6 (that was
already supported before) and off you go as usual. The ORE will pick up
the new map and will start writing two devices of redundancy bits.
Including the pNFS server code and pnfs-export bits at exofs, they all
work supporting raid6 without any code modifications.
I want to deeply thank Daniel Gryniewicz who found first all the
bugs in the old raid code, and inspired these patches.
list of patches:
[PATCH 1/4] ore: (trivial) reformat some code
[PATCH 2/4] ore: Remove redundant dev_order(), more cleanups
[PATCH 3/4] ore: Support for raid 6
Cheers
Boaz
rearrange some source lines. Nothing changed.
Signed-off-by: Boaz Harrosh <[email protected]>
---
fs/exofs/ore.c | 19 ++++++++-----------
fs/exofs/ore_raid.c | 4 +---
2 files changed, 9 insertions(+), 14 deletions(-)
diff --git a/fs/exofs/ore.c b/fs/exofs/ore.c
index dae8846..92157b6 100644
--- a/fs/exofs/ore.c
+++ b/fs/exofs/ore.c
@@ -675,8 +675,8 @@ static int _prepare_for_striping(struct ore_io_state *ios)
si->cur_pg = si->unit_off / PAGE_SIZE;
while (length) {
- unsigned comp = dev - first_dev;
- struct ore_per_dev_state *per_dev = &ios->per_dev[comp];
+ struct ore_per_dev_state *per_dev =
+ &ios->per_dev[dev - first_dev];
unsigned cur_len, page_off = 0;
if (!per_dev->length) {
@@ -708,11 +708,9 @@ static int _prepare_for_striping(struct ore_io_state *ios)
if (unlikely(ret))
goto out;
- dev += mirrors_p1;
- dev = (dev % devs_in_group) + first_dev;
-
length -= cur_len;
+ dev = ((dev + mirrors_p1) % devs_in_group) + first_dev;
si->cur_comp = (si->cur_comp + 1) % group_width;
if (unlikely((dev == si->par_dev) || (!length && ios->sp2d))) {
if (!length && ios->sp2d) {
@@ -721,11 +719,6 @@ static int _prepare_for_striping(struct ore_io_state *ios)
*/
dev = si->par_dev;
}
- if (ios->sp2d)
- /* In writes cur_len just means if it's the
- * last one. See _ore_add_parity_unit.
- */
- cur_len = length;
per_dev = &ios->per_dev[dev - first_dev];
if (!per_dev->length) {
/* Only/always the parity unit of the first
@@ -736,7 +729,11 @@ static int _prepare_for_striping(struct ore_io_state *ios)
per_dev->offset = si->obj_offset - si->unit_off;
}
- ret = _ore_add_parity_unit(ios, si, per_dev, cur_len);
+ /* In writes cur_len just means if it's the
+ * last one. See _ore_add_parity_unit.
+ */
+ ret = _ore_add_parity_unit(ios, si, per_dev,
+ ios->sp2d ? length : cur_len);
if (unlikely(ret))
goto out;
diff --git a/fs/exofs/ore_raid.c b/fs/exofs/ore_raid.c
index 4e2c032..af417d3 100644
--- a/fs/exofs/ore_raid.c
+++ b/fs/exofs/ore_raid.c
@@ -226,9 +226,7 @@ static void _gen_xor_unit(struct __stripe_pages_2d *sp2d)
init_async_submit(&_1ps->submit,
ASYNC_TX_XOR_ZERO_DST | ASYNC_TX_ACK,
- NULL,
- NULL, NULL,
- (addr_conv_t *)_1ps->scribble);
+ NULL, NULL, NULL, (addr_conv_t *)_1ps->scribble);
/* TODO: raid6 */
_1ps->tx = async_xor(_1ps->pages[sp2d->data_devs], _1ps->pages,
--
1.9.0
This simple patch adds support for raid6 to the ORE.
Most operations and calculations where already for the general
case. Only things left:
* call async_gen_syndrome() in the case of raid6
(NOTE that the raid6 math is the one supported by the Linux Kernel
see: crypto/async_tx/async_pq.c)
* call _ore_add_parity_unit() twice with only last call generating
the redundancy pages.
* Fix couple BUGS in old code
a. In reads when parity==2 it can happen that per_dev->length=0
but per_dev->offset was set and adjusted by _ore_add_sg_seg().
Don't let it be overwritten.
b. The all 'cur_comp > starting_dev' thing to determine if:
"per_dev->offset is in the current stripe number or the
next one."
Was a complete raid5/4 accident. When parity==2 this is not
at all true usually. All we need to do is increment si->ob_offset
once we pass by the first parity device.
(This also greatly simplifies the code, amen)
c. Calculation of si->dev rotation can overflow when parity==2.
* Then last enable raid6 in ore_verify_layout()
I want to deeply thank Daniel Gryniewicz who found first all the
bugs in the old raid code, and inspired these patches:
Inspired-by Daniel Gryniewicz <[email protected]>
Signed-off-by: Boaz Harrosh <[email protected]>
---
fs/exofs/Kconfig.ore | 2 ++
fs/exofs/ore.c | 75 ++++++++++++++++++++++++++++++++++++----------------
fs/exofs/ore_raid.c | 37 +++++++++++++++++---------
fs/exofs/ore_raid.h | 3 ++-
4 files changed, 80 insertions(+), 37 deletions(-)
diff --git a/fs/exofs/Kconfig.ore b/fs/exofs/Kconfig.ore
index 1ca7fb7..2daf232 100644
--- a/fs/exofs/Kconfig.ore
+++ b/fs/exofs/Kconfig.ore
@@ -9,4 +9,6 @@ config ORE
tristate
depends on EXOFS_FS || PNFS_OBJLAYOUT
select ASYNC_XOR
+ select RAID6_PQ
+ select ASYNC_PQ
default SCSI_OSD_ULD
diff --git a/fs/exofs/ore.c b/fs/exofs/ore.c
index 0e2a835..cfc0205 100644
--- a/fs/exofs/ore.c
+++ b/fs/exofs/ore.c
@@ -58,9 +58,12 @@ int ore_verify_layout(unsigned total_comps, struct ore_layout *layout)
layout->parity = 1;
break;
case PNFS_OSD_RAID_PQ:
+ layout->parity = 2;
+ break;
case PNFS_OSD_RAID_4:
default:
- ORE_ERR("Only RAID_0/5 for now\n");
+ ORE_ERR("Only RAID_0/5/6 for now received-enum=%d\n",
+ layout->raid_algorithm);
return -EINVAL;
}
if (0 != (layout->stripe_unit & ~PAGE_MASK)) {
@@ -112,6 +115,8 @@ int ore_verify_layout(unsigned total_comps, struct ore_layout *layout)
layout->max_io_length /= stripe_length;
layout->max_io_length *= stripe_length;
}
+ ORE_DBGMSG("max_io_length=0x%lx\n", layout->max_io_length);
+
return 0;
}
EXPORT_SYMBOL(ore_verify_layout);
@@ -561,7 +566,8 @@ void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset,
si->par_dev = (group_width + group_width - parity - RxP) %
group_width + first_dev;
- si->dev = (group_width + C - RxP) % group_width + first_dev;
+ si->dev = (group_width + group_width + C - RxP) %
+ group_width + first_dev;
si->bytes_in_stripe = U;
si->first_stripe_start = M * S + G * T + N * U;
} else {
@@ -651,6 +657,43 @@ out: /* we fail the complete unit on an error eg don't advance
return ret;
}
+static int _add_parity_units(struct ore_io_state *ios,
+ struct ore_striping_info *si,
+ unsigned dev, unsigned first_dev,
+ unsigned mirrors_p1, unsigned devs_in_group,
+ unsigned cur_len)
+{
+ unsigned do_parity;
+ int ret = 0;
+
+ for (do_parity = ios->layout->parity; do_parity; --do_parity) {
+ struct ore_per_dev_state *per_dev;
+
+ per_dev = &ios->per_dev[dev - first_dev];
+ if (!per_dev->length && !per_dev->offset) {
+ /* Only/always the parity unit of the first
+ * stripe will be empty. So this is a chance to
+ * initialize the per_dev info.
+ */
+ per_dev->dev = dev;
+ per_dev->offset = si->obj_offset - si->unit_off;
+ }
+
+ ret = _ore_add_parity_unit(ios, si, per_dev, cur_len,
+ do_parity == 1);
+ if (unlikely(ret))
+ break;
+
+ if (do_parity != 1) {
+ dev = ((dev + mirrors_p1) % devs_in_group) + first_dev;
+ si->cur_comp = (si->cur_comp + 1) %
+ ios->layout->group_width;
+ }
+ }
+
+ return ret;
+}
+
static int _prepare_for_striping(struct ore_io_state *ios)
{
struct ore_striping_info *si = &ios->si;
@@ -660,7 +703,6 @@ static int _prepare_for_striping(struct ore_io_state *ios)
unsigned devs_in_group = group_width * mirrors_p1;
unsigned dev = si->dev;
unsigned first_dev = dev - (dev % devs_in_group);
- unsigned dev_order;
unsigned cur_pg = ios->pages_consumed;
u64 length = ios->length;
int ret = 0;
@@ -672,14 +714,13 @@ static int _prepare_for_striping(struct ore_io_state *ios)
BUG_ON(length > si->length);
- dev_order = si->cur_comp;
-
while (length) {
struct ore_per_dev_state *per_dev =
&ios->per_dev[dev - first_dev];
unsigned cur_len, page_off = 0;
- if (!per_dev->length) {
+ if (!per_dev->length && !per_dev->offset) {
+ /* First time initialize the per_dev info. */
per_dev->dev = dev;
if (dev == si->dev) {
WARN_ON(dev == si->par_dev);
@@ -688,13 +729,7 @@ static int _prepare_for_striping(struct ore_io_state *ios)
page_off = si->unit_off & ~PAGE_MASK;
BUG_ON(page_off && (page_off != ios->pgbase));
} else {
- if (si->cur_comp > dev_order)
- per_dev->offset =
- si->obj_offset - si->unit_off;
- else /* si->cur_comp < dev_order */
- per_dev->offset =
- si->obj_offset + stripe_unit -
- si->unit_off;
+ per_dev->offset = si->obj_offset - si->unit_off;
cur_len = stripe_unit;
}
} else {
@@ -721,20 +756,12 @@ static int _prepare_for_striping(struct ore_io_state *ios)
/* If last stripe operate on parity comp */
si->cur_comp = group_width - ios->layout->parity;
}
- per_dev = &ios->per_dev[dev - first_dev];
- if (!per_dev->length) {
- /* Only/always the parity unit of the first
- * stripe will be empty. So this is a chance to
- * initialize the per_dev info.
- */
- per_dev->dev = dev;
- per_dev->offset = si->obj_offset - si->unit_off;
- }
/* In writes cur_len just means if it's the
* last one. See _ore_add_parity_unit.
*/
- ret = _ore_add_parity_unit(ios, si, per_dev,
+ ret = _add_parity_units(ios, si, dev, first_dev,
+ mirrors_p1, devs_in_group,
ios->sp2d ? length : cur_len);
if (unlikely(ret))
goto out;
@@ -746,6 +773,8 @@ static int _prepare_for_striping(struct ore_io_state *ios)
/* Next stripe, start fresh */
si->cur_comp = 0;
si->cur_pg = 0;
+ si->obj_offset += cur_len;
+ si->unit_off = 0;
}
}
out:
diff --git a/fs/exofs/ore_raid.c b/fs/exofs/ore_raid.c
index d58a952..7f20f25 100644
--- a/fs/exofs/ore_raid.c
+++ b/fs/exofs/ore_raid.c
@@ -218,20 +218,28 @@ static unsigned _sp2d_max_pg(struct __stripe_pages_2d *sp2d)
static void _gen_xor_unit(struct __stripe_pages_2d *sp2d)
{
unsigned p;
+ unsigned tx_flags = ASYNC_TX_ACK;
+
+ if (sp2d->parity == 1)
+ tx_flags |= ASYNC_TX_XOR_ZERO_DST;
+
for (p = 0; p < sp2d->pages_in_unit; p++) {
struct __1_page_stripe *_1ps = &sp2d->_1p_stripes[p];
if (!_1ps->write_count)
continue;
- init_async_submit(&_1ps->submit,
- ASYNC_TX_XOR_ZERO_DST | ASYNC_TX_ACK,
+ init_async_submit(&_1ps->submit, tx_flags,
NULL, NULL, NULL, (addr_conv_t *)_1ps->scribble);
- /* TODO: raid6 */
- _1ps->tx = async_xor(_1ps->pages[sp2d->data_devs], _1ps->pages,
- 0, sp2d->data_devs, PAGE_SIZE,
- &_1ps->submit);
+ if (sp2d->parity == 1)
+ _1ps->tx = async_xor(_1ps->pages[sp2d->data_devs],
+ _1ps->pages, 0, sp2d->data_devs,
+ PAGE_SIZE, &_1ps->submit);
+ else /* parity == 2 */
+ _1ps->tx = async_gen_syndrome(_1ps->pages, 0,
+ sp2d->data_devs + sp2d->parity,
+ PAGE_SIZE, &_1ps->submit);
}
for (p = 0; p < sp2d->pages_in_unit; p++) {
@@ -616,7 +624,7 @@ static int _read_4_write_execute(struct ore_io_state *ios)
int _ore_add_parity_unit(struct ore_io_state *ios,
struct ore_striping_info *si,
struct ore_per_dev_state *per_dev,
- unsigned cur_len)
+ unsigned cur_len, bool do_xor)
{
if (ios->reading) {
if (per_dev->cur_sg >= ios->sgs_per_dev) {
@@ -641,9 +649,11 @@ int _ore_add_parity_unit(struct ore_io_state *ios,
/* If first stripe, Read in all read4write pages
* (if needed) before we calculate the first parity.
*/
- _read_4_write_first_stripe(ios);
+ if (do_xor)
+ _read_4_write_first_stripe(ios);
}
- if (!cur_len) /* If last stripe r4w pages of last stripe */
+ if (!cur_len && do_xor)
+ /* If last stripe r4w pages of last stripe */
_read_4_write_last_stripe(ios);
_read_4_write_execute(ios);
@@ -655,7 +665,7 @@ int _ore_add_parity_unit(struct ore_io_state *ios,
++(ios->cur_par_page);
}
- BUG_ON(si->cur_comp != sp2d->data_devs);
+ BUG_ON(si->cur_comp < sp2d->data_devs);
BUG_ON(si->cur_pg + num_pages > sp2d->pages_in_unit);
ret = _ore_add_stripe_unit(ios, &array_start, 0, pages,
@@ -663,9 +673,10 @@ int _ore_add_parity_unit(struct ore_io_state *ios,
if (unlikely(ret))
return ret;
- /* TODO: raid6 if (last_parity_dev) */
- _gen_xor_unit(sp2d);
- _sp2d_reset(sp2d, ios->r4w, ios->private);
+ if (do_xor) {
+ _gen_xor_unit(sp2d);
+ _sp2d_reset(sp2d, ios->r4w, ios->private);
+ }
}
return 0;
}
diff --git a/fs/exofs/ore_raid.h b/fs/exofs/ore_raid.h
index d365bda..cf6375d 100644
--- a/fs/exofs/ore_raid.h
+++ b/fs/exofs/ore_raid.h
@@ -38,7 +38,8 @@ void _ore_free_raid_stuff(struct ore_io_state *ios);
void _ore_add_sg_seg(struct ore_per_dev_state *per_dev, unsigned cur_len,
bool not_last);
int _ore_add_parity_unit(struct ore_io_state *ios, struct ore_striping_info *si,
- struct ore_per_dev_state *per_dev, unsigned cur_len);
+ struct ore_per_dev_state *per_dev, unsigned cur_len,
+ bool do_xor);
void _ore_add_stripe_page(struct __stripe_pages_2d *sp2d,
struct ore_striping_info *si, struct page *page);
static inline void _add_stripe_page(struct __stripe_pages_2d *sp2d,
--
1.9.0
Two cleanups:
* si->cur_comp, si->cur_pg where always calculated after
the call to ore_calc_stripe_info() with the help of
_dev_order(...). But these are already calculated by
ore_calc_stripe_info() and can be just set there.
(This is left over from the time that si->cur_comp, si->cur_pg
were only used by raid code, but now the main loop manages
them anyway even though they are ultimately not used in
none raid code)
* si->cur_comp - For the very last stripe case, was set inside
_ore_add_parity_unit(). This is not clear and will be wrong
for coming raid6 so move this to only caller. Now si->cur_comp
is only manipulated within _prepare_for_striping(), always next
to the manipulation of cur_dev.
Which is much easier to understand and follow.
Signed-off-by: Boaz Harrosh <[email protected]>
---
fs/exofs/ore.c | 10 ++++++----
fs/exofs/ore_raid.c | 13 ++++---------
fs/exofs/ore_raid.h | 18 ------------------
3 files changed, 10 insertions(+), 31 deletions(-)
diff --git a/fs/exofs/ore.c b/fs/exofs/ore.c
index 92157b6..0e2a835 100644
--- a/fs/exofs/ore.c
+++ b/fs/exofs/ore.c
@@ -545,17 +545,19 @@ void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset,
/* "H - (N * U)" is just "H % U" so it's bound to u32 */
u32 C = (u32)(H - (N * U)) / stripe_unit + G * group_width;
+ u32 first_dev = C - C % group_width;
div_u64_rem(file_offset, stripe_unit, &si->unit_off);
si->obj_offset = si->unit_off + (N * stripe_unit) +
(M * group_depth * stripe_unit);
+ si->cur_comp = C - first_dev;
+ si->cur_pg = si->unit_off / PAGE_SIZE;
if (parity) {
u32 LCMdP = lcm(group_width, parity) / parity;
/* R = N % LCMdP; */
u32 RxP = (N % LCMdP) * parity;
- u32 first_dev = C - C % group_width;
si->par_dev = (group_width + group_width - parity - RxP) %
group_width + first_dev;
@@ -670,9 +672,7 @@ static int _prepare_for_striping(struct ore_io_state *ios)
BUG_ON(length > si->length);
- dev_order = _dev_order(devs_in_group, mirrors_p1, si->par_dev, dev);
- si->cur_comp = dev_order;
- si->cur_pg = si->unit_off / PAGE_SIZE;
+ dev_order = si->cur_comp;
while (length) {
struct ore_per_dev_state *per_dev =
@@ -718,6 +718,8 @@ static int _prepare_for_striping(struct ore_io_state *ios)
* stripe. then operate on parity dev.
*/
dev = si->par_dev;
+ /* If last stripe operate on parity comp */
+ si->cur_comp = group_width - ios->layout->parity;
}
per_dev = &ios->per_dev[dev - first_dev];
if (!per_dev->length) {
diff --git a/fs/exofs/ore_raid.c b/fs/exofs/ore_raid.c
index af417d3..d58a952 100644
--- a/fs/exofs/ore_raid.c
+++ b/fs/exofs/ore_raid.c
@@ -402,9 +402,8 @@ static int _add_to_r4w_last_page(struct ore_io_state *ios, u64 *offset)
ore_calc_stripe_info(ios->layout, *offset, 0, &si);
- p = si.unit_off / PAGE_SIZE;
- c = _dev_order(ios->layout->group_width * ios->layout->mirrors_p1,
- ios->layout->mirrors_p1, si.par_dev, si.dev);
+ p = si.cur_pg;
+ c = si.cur_comp;
page = ios->sp2d->_1p_stripes[p].pages[c];
pg_len = PAGE_SIZE - (si.unit_off % PAGE_SIZE);
@@ -532,9 +531,8 @@ static int _read_4_write_last_stripe(struct ore_io_state *ios)
goto read_it;
ore_calc_stripe_info(ios->layout, offset, 0, &read_si);
- p = read_si.unit_off / PAGE_SIZE;
- c = _dev_order(ios->layout->group_width * ios->layout->mirrors_p1,
- ios->layout->mirrors_p1, read_si.par_dev, read_si.dev);
+ p = read_si.cur_pg;
+ c = read_si.cur_comp;
if (min_p == sp2d->pages_in_unit) {
/* Didn't do it yet */
@@ -638,9 +636,6 @@ int _ore_add_parity_unit(struct ore_io_state *ios,
si->cur_pg = _sp2d_min_pg(sp2d);
num_pages = _sp2d_max_pg(sp2d) + 1 - si->cur_pg;
- if (!cur_len) /* If last stripe operate on parity comp */
- si->cur_comp = sp2d->data_devs;
-
if (!per_dev->length) {
per_dev->offset += si->cur_pg * PAGE_SIZE;
/* If first stripe, Read in all read4write pages
diff --git a/fs/exofs/ore_raid.h b/fs/exofs/ore_raid.h
index 2ffd2c3..d365bda 100644
--- a/fs/exofs/ore_raid.h
+++ b/fs/exofs/ore_raid.h
@@ -31,24 +31,6 @@
#define ORE_DBGMSG2(M...) do {} while (0)
/* #define ORE_DBGMSG2 ORE_DBGMSG */
-/* Calculate the component order in a stripe. eg the logical data unit
- * address within the stripe of @dev given the @par_dev of this stripe.
- */
-static inline unsigned _dev_order(unsigned devs_in_group, unsigned mirrors_p1,
- unsigned par_dev, unsigned dev)
-{
- unsigned first_dev = dev - dev % devs_in_group;
-
- dev -= first_dev;
- par_dev -= first_dev;
-
- if (devs_in_group == par_dev) /* The raid 0 case */
- return dev / mirrors_p1;
- /* raid4/5/6 case */
- return ((devs_in_group + dev - par_dev - mirrors_p1) % devs_in_group) /
- mirrors_p1;
-}
-
/* ios_raid.c stuff needed by ios.c */
int _ore_post_alloc_raid_stuff(struct ore_io_state *ios);
void _ore_free_raid_stuff(struct ore_io_state *ios);
--
1.9.0