diff --git a/lustre/kernel_patches/patches/raid5-zerocopy.patch b/lustre/kernel_patches/patches/raid5-zerocopy.patch
index 5095906dfbdca8659d00da10eab46fbf93bd9675..f3c1e041c9955655ac8145fd31f58f4340ffa7b4 100644
--- a/lustre/kernel_patches/patches/raid5-zerocopy.patch
+++ b/lustre/kernel_patches/patches/raid5-zerocopy.patch
@@ -1,15 +1,17 @@
-diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
---- linux-2.6.9.orig/drivers/md/raid5.c 2007-07-09 02:43:33.000000000 -0600
-+++ linux-2.6.9/drivers/md/raid5.c 2007-07-13 00:39:15.000000000 -0600
-@@ -412,6 +412,7 @@ static int raid5_end_read_request (struc
+diff -pur linux-2.6.9-67.orig/drivers/md/raid5.c linux-2.6.9-67/drivers/md/raid5.c
+--- linux-2.6.9-67.orig/drivers/md/raid5.c 2009-02-15 10:11:54.000000000 +0800
++++ linux-2.6.9-67/drivers/md/raid5.c 2009-02-15 10:22:51.000000000 +0800
+@@ -412,6 +412,9 @@ static int raid5_end_read_request (struc
clear_buffer_uptodate(bh);
}
#endif
-+ BUG_ON(test_bit(R5_Direct, &sh->dev[i].flags));
++ /* Read on a Directing write is allowable */
++ /* BUG_ON(test_bit(R5_Direct, &sh->dev[i].flags)) */
++ BUG_ON(sh->dev[i].req.bi_io_vec[0].bv_page != sh->dev[i].page);
clear_bit(R5_LOCKED, &sh->dev[i].flags);
set_bit(STRIPE_HANDLE, &sh->state);
release_stripe(sh);
-@@ -450,6 +451,10 @@ static int raid5_end_write_request (stru
+@@ -450,6 +453,10 @@ static int raid5_end_write_request (stru
rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
@@ -20,7 +22,7 @@ diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
clear_bit(R5_LOCKED, &sh->dev[i].flags);
set_bit(STRIPE_HANDLE, &sh->state);
__release_stripe(conf, sh);
-@@ -620,7 +625,27 @@ static sector_t compute_blocknr(struct s
+@@ -620,7 +627,27 @@ static sector_t compute_blocknr(struct s
return r_sector;
}
@@ -48,7 +50,7 @@ diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
/*
* Copy data between a page in the stripe cache, and one or more bion
-@@ -716,8 +741,9 @@ static void compute_parity(struct stripe
+@@ -716,8 +743,9 @@ static void compute_parity(struct stripe
{
raid5_conf_t *conf = sh->raid_conf;
int i, pd_idx = sh->pd_idx, disks = conf->raid_disks, count;
@@ -59,7 +61,7 @@ diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
PRINTK("compute_parity, stripe %llu, method %d\n",
(unsigned long long)sh->sector, method);
-@@ -744,13 +770,14 @@ static void compute_parity(struct stripe
+@@ -744,13 +772,14 @@ static void compute_parity(struct stripe
break;
case RECONSTRUCT_WRITE:
memset(ptr[0], 0, STRIPE_SIZE);
@@ -75,7 +77,7 @@ diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
break;
case CHECK_PARITY:
break;
-@@ -760,34 +787,88 @@ static void compute_parity(struct stripe
+@@ -760,34 +789,90 @@ static void compute_parity(struct stripe
count = 1;
}
@@ -105,6 +107,8 @@ diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
+ page = zero_copy_data(wbi, sector);
+ if (page) {
+ atomic_inc(&conf->writes_zcopy);
++ /* The pointer must be restored whenever the LOCKED
++ * gets cleared. */
+ sh->dev[i].req.bi_io_vec[0].bv_page = page;
+ set_bit(R5_Direct, &sh->dev[i].flags);
+ clear_bit(R5_UPTODATE, &sh->dev[i].flags);
@@ -181,7 +185,7 @@ diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
}
if (count != 1)
xor_block(count, STRIPE_SIZE, ptr);
-@@ -1059,13 +1140,15 @@ static void handle_stripe(struct stripe_
+@@ -1061,13 +1146,15 @@ static void handle_stripe(struct stripe_
if (sh->dev[i].written) {
dev = &sh->dev[i];
if (!test_bit(R5_LOCKED, &dev->flags) &&
@@ -198,7 +202,23 @@ diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
while (wbi && wbi->bi_sector < dev->sector + STRIPE_SECTORS) {
wbi2 = r5_next_bio(wbi, dev->sector);
if (--wbi->bi_phys_segments == 0) {
-@@ -1831,6 +1914,7 @@ memory = conf->max_nr_stripes * (sizeof(
+@@ -1337,6 +1424,15 @@ static void handle_stripe(struct stripe_
+ } else {
+ PRINTK("skip op %ld on disc %d for sector %llu\n",
+ bi->bi_rw, i, (unsigned long long)sh->sector);
++
++ if (test_bit(R5_Direct, &sh->dev[i].flags)) {
++ /* restore the page pointer of req, otherwise,
++ * no any read is permitted on this stripe, this is
++ * not what we want. -jay */
++ BUG_ON(sh->dev[i].req.bi_io_vec[0].bv_page == sh->dev[i].page);
++ sh->dev[i].req.bi_io_vec[0].bv_page = sh->dev[i].page;
++ }
++
+ clear_bit(R5_LOCKED, &sh->dev[i].flags);
+ set_bit(STRIPE_HANDLE, &sh->state);
+ }
+@@ -1835,6 +1931,7 @@ memory = conf->max_nr_stripes * (sizeof(
if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
}
@@ -206,7 +226,7 @@ diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
/* Ok, everything is just fine now */
mddev->array_size = mddev->size * (mddev->raid_disks - 1);
-@@ -1918,9 +2002,11 @@ static void status (struct seq_file *seq
+@@ -1922,9 +2019,11 @@ static void status (struct seq_file *seq
atomic_read(&conf->handled_in_raid5d),
atomic_read(&conf->out_of_stripes),
atomic_read(&conf->handle_called));
@@ -220,9 +240,9 @@ diff -pru linux-2.6.9.orig/drivers/md/raid5.c linux-2.6.9/drivers/md/raid5.c
seq_printf (seq, "\n\t\t%u delayed, %u active, queues: %u in, %u out\n",
atomic_read(&conf->delayed),
atomic_read(&conf->active_stripes),
-diff -pru linux-2.6.9.orig/include/linux/backing-dev.h linux-2.6.9/include/linux/backing-dev.h
---- linux-2.6.9.orig/include/linux/backing-dev.h 2004-10-18 15:53:46.000000000 -0600
-+++ linux-2.6.9/include/linux/backing-dev.h 2007-07-13 00:12:46.000000000 -0600
+diff -pur linux-2.6.9-67.orig/include/linux/backing-dev.h linux-2.6.9-67/include/linux/backing-dev.h
+--- linux-2.6.9-67.orig/include/linux/backing-dev.h 2009-02-15 10:11:54.000000000 +0800
++++ linux-2.6.9-67/include/linux/backing-dev.h 2009-02-15 10:22:40.000000000 +0800
@@ -30,8 +30,11 @@ struct backing_dev_info {
void *congested_data; /* Pointer to aux data for congested func */
void (*unplug_io_fn)(struct backing_dev_info *, struct page *);
@@ -243,9 +263,9 @@ diff -pru linux-2.6.9.orig/include/linux/backing-dev.h linux-2.6.9/include/linux
+ ((mapping)->backing_dev_info->capabilities & BDI_CAP_PAGE_CONST_WRITE)
+
#endif /* _LINUX_BACKING_DEV_H */
-diff -pru linux-2.6.9.orig/include/linux/page-flags.h linux-2.6.9/include/linux/page-flags.h
---- linux-2.6.9.orig/include/linux/page-flags.h 2004-10-18 15:54:39.000000000 -0600
-+++ linux-2.6.9/include/linux/page-flags.h 2007-07-13 00:12:46.000000000 -0600
+diff -pur linux-2.6.9-67.orig/include/linux/page-flags.h linux-2.6.9-67/include/linux/page-flags.h
+--- linux-2.6.9-67.orig/include/linux/page-flags.h 2009-02-15 10:11:54.000000000 +0800
++++ linux-2.6.9-67/include/linux/page-flags.h 2009-02-15 10:22:40.000000000 +0800
@@ -74,6 +74,7 @@
#define PG_swapcache 16 /* Swap page: swp_entry_t in private */
#define PG_mappedtodisk 17 /* Has blocks allocated on-disk */
@@ -266,9 +286,9 @@ diff -pru linux-2.6.9.orig/include/linux/page-flags.h linux-2.6.9/include/linux/
struct page; /* forward declaration */
int test_clear_page_dirty(struct page *page);
-diff -pru linux-2.6.9.orig/include/linux/pagemap.h linux-2.6.9/include/linux/pagemap.h
---- linux-2.6.9.orig/include/linux/pagemap.h 2004-10-18 15:53:06.000000000 -0600
-+++ linux-2.6.9/include/linux/pagemap.h 2007-07-13 00:12:46.000000000 -0600
+diff -pur linux-2.6.9-67.orig/include/linux/pagemap.h linux-2.6.9-67/include/linux/pagemap.h
+--- linux-2.6.9-67.orig/include/linux/pagemap.h 2009-02-15 10:11:54.000000000 +0800
++++ linux-2.6.9-67/include/linux/pagemap.h 2009-02-15 10:22:40.000000000 +0800
@@ -191,6 +191,19 @@ static inline void wait_on_page_writebac
extern void end_page_writeback(struct page *page);
@@ -289,9 +309,9 @@ diff -pru linux-2.6.9.orig/include/linux/pagemap.h linux-2.6.9/include/linux/pag
/*
* Fault a userspace page into pagetables. Return non-zero on a fault.
*
-diff -pru linux-2.6.9.orig/include/linux/raid/raid5.h linux-2.6.9/include/linux/raid/raid5.h
---- linux-2.6.9.orig/include/linux/raid/raid5.h 2007-07-09 02:43:33.000000000 -0600
-+++ linux-2.6.9/include/linux/raid/raid5.h 2007-07-13 00:39:15.000000000 -0600
+diff -pur linux-2.6.9-67.orig/include/linux/raid/raid5.h linux-2.6.9-67/include/linux/raid/raid5.h
+--- linux-2.6.9-67.orig/include/linux/raid/raid5.h 2009-02-15 10:11:54.000000000 +0800
++++ linux-2.6.9-67/include/linux/raid/raid5.h 2009-02-15 10:22:40.000000000 +0800
@@ -153,6 +153,7 @@ struct stripe_head {
#define R5_Wantread 4 /* want to schedule a read */
#define R5_Wantwrite 5
@@ -309,9 +329,9 @@ diff -pru linux-2.6.9.orig/include/linux/raid/raid5.h linux-2.6.9/include/linux/
atomic_t handle_called;
atomic_t delayed;
atomic_t in_reqs_in_queue;
-diff -pru linux-2.6.9.orig/mm/filemap.c linux-2.6.9/mm/filemap.c
---- linux-2.6.9.orig/mm/filemap.c 2007-07-09 02:43:33.000000000 -0600
-+++ linux-2.6.9/mm/filemap.c 2007-07-13 00:12:46.000000000 -0600
+diff -pur linux-2.6.9-67.orig/mm/filemap.c linux-2.6.9-67/mm/filemap.c
+--- linux-2.6.9-67.orig/mm/filemap.c 2009-02-15 10:11:55.000000000 +0800
++++ linux-2.6.9-67/mm/filemap.c 2009-02-15 10:22:40.000000000 +0800
@@ -27,6 +27,8 @@
#include <linux/pagevec.h>
#include <linux/blkdev.h>
@@ -321,7 +341,7 @@ diff -pru linux-2.6.9.orig/mm/filemap.c linux-2.6.9/mm/filemap.c
/*
* This is needed for the following functions:
* - try_to_release_page
-@@ -486,11 +488,52 @@ void end_page_writeback(struct page *pag
+@@ -485,11 +487,52 @@ void end_page_writeback(struct page *pag
BUG();
smp_mb__after_clear_bit();
}
diff --git a/lustre/kernel_patches/patches/raid6-zerocopy.patch b/lustre/kernel_patches/patches/raid6-zerocopy.patch
index 95b713d465fa12a952bbb438e73f67ce514a52e6..27d7f153f90e5fbc80f99c76dcbe724d606f8dbd 100644
--- a/lustre/kernel_patches/patches/raid6-zerocopy.patch
+++ b/lustre/kernel_patches/patches/raid6-zerocopy.patch
@@ -1,15 +1,17 @@
-diff -pur linux-2.6.9.orig/drivers/md/raid6main.c linux-2.6.9/drivers/md/raid6main.c
---- linux-2.6.9.orig/drivers/md/raid6main.c 2008-01-10 14:02:08.000000000 +0800
-+++ linux-2.6.9/drivers/md/raid6main.c 2008-01-10 14:01:56.000000000 +0800
-@@ -430,6 +430,7 @@ static int raid6_end_read_request (struc
+diff -pur linux-2.6.9-67.orig/drivers/md/raid6main.c linux-2.6.9-67/drivers/md/raid6main.c
+--- linux-2.6.9-67.orig/drivers/md/raid6main.c 2009-02-15 10:24:30.000000000 +0800
++++ linux-2.6.9-67/drivers/md/raid6main.c 2009-02-15 10:26:17.000000000 +0800
+@@ -430,6 +430,9 @@ static int raid6_end_read_request (struc
clear_buffer_uptodate(bh);
}
#endif
-+ BUG_ON(test_bit(R5_Direct, &sh->dev[i].flags));
++ /* Read on a Directing write is allowable */
++ /* BUG_ON(test_bit(R5_Direct, &sh->dev[i].flags)) */
++ BUG_ON(sh->dev[i].req.bi_io_vec[0].bv_page != sh->dev[i].page);
clear_bit(R5_LOCKED, &sh->dev[i].flags);
set_bit(STRIPE_HANDLE, &sh->state);
release_stripe(sh);
-@@ -468,6 +469,10 @@ static int raid6_end_write_request (stru
+@@ -468,6 +471,10 @@ static int raid6_end_write_request (stru
rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
@@ -20,7 +22,7 @@ diff -pur linux-2.6.9.orig/drivers/md/raid6main.c linux-2.6.9/drivers/md/raid6ma
clear_bit(R5_LOCKED, &sh->dev[i].flags);
set_bit(STRIPE_HANDLE, &sh->state);
__release_stripe(conf, sh);
-@@ -664,7 +669,27 @@ static sector_t compute_blocknr(struct s
+@@ -664,7 +671,27 @@ static sector_t compute_blocknr(struct s
return r_sector;
}
@@ -48,7 +50,7 @@ diff -pur linux-2.6.9.orig/drivers/md/raid6main.c linux-2.6.9/drivers/md/raid6ma
/*
* Copy data between a page in the stripe cache, and one or more bion
-@@ -731,6 +756,7 @@ static void compute_parity(struct stripe
+@@ -731,6 +758,7 @@ static void compute_parity(struct stripe
raid6_conf_t *conf = sh->raid_conf;
int i, pd_idx = sh->pd_idx, qd_idx, d0_idx, disks = conf->raid_disks, count;
struct bio *chosen;
@@ -56,7 +58,7 @@ diff -pur linux-2.6.9.orig/drivers/md/raid6main.c linux-2.6.9/drivers/md/raid6ma
/**** FIX THIS: This could be very bad if disks is close to 256 ****/
void *ptrs[disks];
-@@ -761,18 +787,46 @@ static void compute_parity(struct stripe
+@@ -761,18 +789,46 @@ static void compute_parity(struct stripe
BUG(); /* Not implemented yet */
}
@@ -112,7 +114,7 @@ diff -pur linux-2.6.9.orig/drivers/md/raid6main.c linux-2.6.9/drivers/md/raid6ma
// switch(method) {
// case RECONSTRUCT_WRITE:
-@@ -783,7 +837,10 @@ static void compute_parity(struct stripe
+@@ -783,7 +839,10 @@ static void compute_parity(struct stripe
count = 0;
i = d0_idx;
do {
@@ -124,7 +126,7 @@ diff -pur linux-2.6.9.orig/drivers/md/raid6main.c linux-2.6.9/drivers/md/raid6ma
i = raid6_next_disk(i, disks);
} while ( i != d0_idx );
-@@ -1185,7 +1242,8 @@ static void handle_stripe(struct stripe_
+@@ -1185,7 +1244,8 @@ static void handle_stripe(struct stripe_
if (sh->dev[i].written) {
dev = &sh->dev[i];
if (!test_bit(R5_LOCKED, &dev->flags) &&
@@ -134,7 +136,7 @@ diff -pur linux-2.6.9.orig/drivers/md/raid6main.c linux-2.6.9/drivers/md/raid6ma
/* We can return any write requests */
struct bio *wbi, *wbi2;
PRINTK("Return write for stripe %llu disc %d\n",
-@@ -1193,6 +1251,7 @@ static void handle_stripe(struct stripe_
+@@ -1193,6 +1253,7 @@ static void handle_stripe(struct stripe_
spin_lock_irq(&conf->device_lock);
wbi = dev->written;
dev->written = NULL;
@@ -142,7 +144,23 @@ diff -pur linux-2.6.9.orig/drivers/md/raid6main.c linux-2.6.9/drivers/md/raid6ma
while (wbi && wbi->bi_sector < dev->sector + STRIPE_SECTORS) {
wbi2 = r5_next_bio(wbi, dev->sector);
if (--wbi->bi_phys_segments == 0) {
-@@ -2008,6 +2067,7 @@ static int run (mddev_t *mddev)
+@@ -1503,6 +1564,15 @@ static void handle_stripe(struct stripe_
+ } else {
+ PRINTK("skip op %ld on disc %d for sector %llu\n",
+ bi->bi_rw, i, (unsigned long long)sh->sector);
++
++ if (test_bit(R5_Direct, &sh->dev[i].flags)) {
++ /* restore the page pointer of req, otherwise,
++ * no any read is permitted on this stripe, this is
++ * not what we want. -jay */
++ BUG_ON(sh->dev[i].req.bi_io_vec[0].bv_page == sh->dev[i].page);
++ sh->dev[i].req.bi_io_vec[0].bv_page = sh->dev[i].page;
++ }
++
+ clear_bit(R5_LOCKED, &sh->dev[i].flags);
+ set_bit(STRIPE_HANDLE, &sh->state);
+ atomic_dec(&conf->delayed);
+@@ -2008,6 +2078,7 @@ static int run (mddev_t *mddev)
if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe)
mddev->queue->backing_dev_info.ra_pages = 2 * stripe;
}
@@ -150,7 +168,7 @@ diff -pur linux-2.6.9.orig/drivers/md/raid6main.c linux-2.6.9/drivers/md/raid6ma
/* Ok, everything is just fine now */
mddev->array_size = mddev->size * (mddev->raid_disks - 2);
-@@ -2095,9 +2155,11 @@ static void status (struct seq_file *seq
+@@ -2095,9 +2166,11 @@ static void status (struct seq_file *seq
atomic_read(&conf->handled_in_raid5d),
atomic_read(&conf->out_of_stripes),
atomic_read(&conf->handle_called));