Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
md/r5cache: read data into orig_page for prexor of cached data
With write back cache, we use orig_page to do prexor. This patch
makes sure we read data into orig_page for it.
Flag R5_OrigPageUPTDODATE is added to show whether orig_page
has the latest data from raid disk.
We introduce a helper function uptodate_for_rmw() to simplify
the a couple conditions in handle_stripe_dirtying().
Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Shaohua Li <shli@fb.com>
authored by Song Liu and committed by Shaohua Li 86aa1397 d46d29f0
+2
drivers/md/raid5-cache.c
+2
drivers/md/raid5-cache.c
+35
-9
drivers/md/raid5.c
+35
-9
drivers/md/raid5.c
···
1015
1016
if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1017
WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1018
-
sh->dev[i].vec.bv_page = sh->dev[i].page;
1019
bi->bi_vcnt = 1;
1020
bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1021
bi->bi_io_vec[0].bv_offset = 0;
···
2390
} else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2391
clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2392
2393
if (atomic_read(&rdev->read_errors))
2394
atomic_set(&rdev->read_errors, 0);
2395
} else {
···
3611
break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
3612
}
3613
3614
static int handle_stripe_dirtying(struct r5conf *conf,
3615
struct stripe_head *sh,
3616
struct stripe_head_state *s,
···
3657
if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx ||
3658
test_bit(R5_InJournal, &dev->flags)) &&
3659
!test_bit(R5_LOCKED, &dev->flags) &&
3660
-
!((test_bit(R5_UPTODATE, &dev->flags) &&
3661
-
(!test_bit(R5_InJournal, &dev->flags) ||
3662
-
dev->page != dev->orig_page)) ||
3663
test_bit(R5_Wantcompute, &dev->flags))) {
3664
if (test_bit(R5_Insync, &dev->flags))
3665
rmw++;
···
3669
i != sh->pd_idx && i != sh->qd_idx &&
3670
!test_bit(R5_LOCKED, &dev->flags) &&
3671
!(test_bit(R5_UPTODATE, &dev->flags) ||
3672
-
test_bit(R5_InJournal, &dev->flags) ||
3673
test_bit(R5_Wantcompute, &dev->flags))) {
3674
if (test_bit(R5_Insync, &dev->flags))
3675
rcw++;
···
3722
i == sh->pd_idx || i == sh->qd_idx ||
3723
test_bit(R5_InJournal, &dev->flags)) &&
3724
!test_bit(R5_LOCKED, &dev->flags) &&
3725
-
!((test_bit(R5_UPTODATE, &dev->flags) &&
3726
-
(!test_bit(R5_InJournal, &dev->flags) ||
3727
-
dev->page != dev->orig_page)) ||
3728
test_bit(R5_Wantcompute, &dev->flags)) &&
3729
test_bit(R5_Insync, &dev->flags)) {
3730
if (test_bit(STRIPE_PREREAD_ACTIVE,
···
3749
i != sh->pd_idx && i != sh->qd_idx &&
3750
!test_bit(R5_LOCKED, &dev->flags) &&
3751
!(test_bit(R5_UPTODATE, &dev->flags) ||
3752
-
test_bit(R5_InJournal, &dev->flags) ||
3753
test_bit(R5_Wantcompute, &dev->flags))) {
3754
rcw++;
3755
if (test_bit(R5_Insync, &dev->flags) &&
···
1015
1016
if (test_bit(R5_SkipCopy, &sh->dev[i].flags))
1017
WARN_ON(test_bit(R5_UPTODATE, &sh->dev[i].flags));
1018
+
1019
+
if (!op_is_write(op) &&
1020
+
test_bit(R5_InJournal, &sh->dev[i].flags))
1021
+
/*
1022
+
* issuing read for a page in journal, this
1023
+
* must be preparing for prexor in rmw; read
1024
+
* the data into orig_page
1025
+
*/
1026
+
sh->dev[i].vec.bv_page = sh->dev[i].orig_page;
1027
+
else
1028
+
sh->dev[i].vec.bv_page = sh->dev[i].page;
1029
bi->bi_vcnt = 1;
1030
bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
1031
bi->bi_io_vec[0].bv_offset = 0;
···
2380
} else if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
2381
clear_bit(R5_ReadNoMerge, &sh->dev[i].flags);
2382
2383
+
if (test_bit(R5_InJournal, &sh->dev[i].flags))
2384
+
/*
2385
+
* end read for a page in journal, this
2386
+
* must be preparing for prexor in rmw
2387
+
*/
2388
+
set_bit(R5_OrigPageUPTDODATE, &sh->dev[i].flags);
2389
+
2390
if (atomic_read(&rdev->read_errors))
2391
atomic_set(&rdev->read_errors, 0);
2392
} else {
···
3594
break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
3595
}
3596
3597
+
/*
3598
+
* For RMW in write back cache, we need extra page in prexor to store the
3599
+
* old data. This page is stored in dev->orig_page.
3600
+
*
3601
+
* This function checks whether we have data for prexor. The exact logic
3602
+
* is:
3603
+
* R5_UPTODATE && (!R5_InJournal || R5_OrigPageUPTDODATE)
3604
+
*/
3605
+
static inline bool uptodate_for_rmw(struct r5dev *dev)
3606
+
{
3607
+
return (test_bit(R5_UPTODATE, &dev->flags)) &&
3608
+
(!test_bit(R5_InJournal, &dev->flags) ||
3609
+
test_bit(R5_OrigPageUPTDODATE, &dev->flags));
3610
+
}
3611
+
3612
static int handle_stripe_dirtying(struct r5conf *conf,
3613
struct stripe_head *sh,
3614
struct stripe_head_state *s,
···
3625
if ((dev->towrite || i == sh->pd_idx || i == sh->qd_idx ||
3626
test_bit(R5_InJournal, &dev->flags)) &&
3627
!test_bit(R5_LOCKED, &dev->flags) &&
3628
+
!(uptodate_for_rmw(dev) ||
3629
test_bit(R5_Wantcompute, &dev->flags))) {
3630
if (test_bit(R5_Insync, &dev->flags))
3631
rmw++;
···
3639
i != sh->pd_idx && i != sh->qd_idx &&
3640
!test_bit(R5_LOCKED, &dev->flags) &&
3641
!(test_bit(R5_UPTODATE, &dev->flags) ||
3642
test_bit(R5_Wantcompute, &dev->flags))) {
3643
if (test_bit(R5_Insync, &dev->flags))
3644
rcw++;
···
3693
i == sh->pd_idx || i == sh->qd_idx ||
3694
test_bit(R5_InJournal, &dev->flags)) &&
3695
!test_bit(R5_LOCKED, &dev->flags) &&
3696
+
!(uptodate_for_rmw(dev) ||
3697
test_bit(R5_Wantcompute, &dev->flags)) &&
3698
test_bit(R5_Insync, &dev->flags)) {
3699
if (test_bit(STRIPE_PREREAD_ACTIVE,
···
3722
i != sh->pd_idx && i != sh->qd_idx &&
3723
!test_bit(R5_LOCKED, &dev->flags) &&
3724
!(test_bit(R5_UPTODATE, &dev->flags) ||
3725
test_bit(R5_Wantcompute, &dev->flags))) {
3726
rcw++;
3727
if (test_bit(R5_Insync, &dev->flags) &&
+5
drivers/md/raid5.h
+5
drivers/md/raid5.h
···
322
* data and parity being written are in the journal
323
* device
324
*/
325
+
R5_OrigPageUPTDODATE, /* with write back cache, we read old data into
326
+
* dev->orig_page for prexor. When this flag is
327
+
* set, orig_page contains latest data in the
328
+
* raid disk.
329
+
*/
330
};
331
332
/*