Skip to contentSkip to conversation

Get drbd-9.0 branch building with Linux 5.9 and 5.10

Drbd-dev New Michael D Labriola 2 participants 11 messages Dec 13, 2020 – Dec 14, 2020 DRBD driver

lore
Share & tools
lore
#0 lore raw

Hey, everyone! I had the chance to try building the latest drbd-9.0 code against the 5.9 and 5.10 kernels this afternoon. Things almost built, but not quite. I made a series of commits to get things compiling. Most of it is trivial compat test changes and following along w/ what's already been done in the mainline drbd8 module to get it ready for 5.10.

I did not create a compat test for the switch from sched_setscheduler() to sched_set_fifo_low() yet. Mostly, because I've never done that and have to wrap my head around how to add another test. ;-)

I tried submitting a pull request via github, but got the auto-close message. So, here's take two.

#1 lore raw

PATCH 1/8 compat: fix include for submit_bio test

Looks like this test is including the wrong header to get at struct block_device_operations.

---
 drbd/drbd-kernel-compat/tests/have_submit_bio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drbd/drbd-kernel-compat/tests/have_submit_bio.c +1 -1
index 26bc7e8c..ca94cd27 100644
@@ -1,6 +1,6 @@
 /* { "version": "v5.8", "commit": "c62b37d96b6eb3ec5ae4cbe00db107bf15aebc93", "comment": "Since 5.8 make_request_fn has been replaced by a block_device_operations method called submit_bio", "author": "Christoph Hellwig <[email protected]>", "date": "Wed Jul 1 10:59:43 2020 +0200" } */
 
-#include <linux/blk_types.h>
+#include <linux/blkdev.h>
 
 void foo(struct block_device_operations *ops)
 {
Signature · 1 line
2.17.1
#2 lore raw

PATCH 2/8 compat: fix include for blkdev_get_by_path test

In reply to Michael D Labriola

This function lives in linux/blkdev.h

---
 drbd/drbd-kernel-compat/tests/have_blkdev_get_by_path.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drbd/drbd-kernel-compat/tests/have_blkdev_get_by_path.c +1 -1
index d0bae496..f795c1d1 100644
@@ -1,4 +1,4 @@
-#include <linux/fs.h>
+#include <linux/blkdev.h>
 
 /*
  * In kernel version 2.6.38-rc1, open_bdev_exclusive() was replaced by
Signature · 1 line
2.17.1
#3 lore raw

PATCH 3/8 drbd: don't include vermagic.h

In reply to Michael D Labriola

Compilation fails because "This header can be included from kernel/module.c or *.mod.c only". Leaving it out seems to work just fine.

---
 drbd/drbd_main.c | 1 -
 1 file changed, 1 deletion(-)
drbd/drbd_main.c +0 -1
index 98a94a16..fe1a0b31 100644
@@ -16,7 +16,6 @@
 #define pr_fmt(fmt)	KBUILD_MODNAME ": " fmt
 
 #include <linux/module.h>
-#include <linux/vermagic.h>
 #include <linux/jiffies.h>
 #include <linux/drbd.h>
 #include <linux/uaccess.h>
Signature · 1 line
2.17.1
#4 lore raw

PATCH 4/8 drbd: fix drbd_submit_bio argument handling

In reply to Michael D Labriola

Looks like we changed arguments from q to bio, but left the line that gets device from q untouched. Oopsie.

---
 drbd/drbd_req.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drbd/drbd_req.c +1 -1
index 8db2a0e0..92b3b863 100644
@@ -2196,7 +2196,7 @@ static bool drbd_fail_request_early(struct drbd_device *device, struct bio *bio)
 
 blk_qc_t drbd_submit_bio(struct bio *bio)
 {
-	struct drbd_device *device = (struct drbd_device *) q->queuedata;
+	struct drbd_device *device = bio->bi_disk->private_data;
 #ifdef CONFIG_DRBD_TIMING_STATS
 	ktime_t start_kt;
 #endif
Signature · 1 line
2.17.1
#5 lore raw

PATCH 5/8 drbd: use sock_set_keepalive() instead of kernel_setsockopt()

In reply to Michael D Labriola

Looks like the rest of the code has been updated to use 5.8's sock_set_* and tcp_set_* functions, but we've missed this one case.

---
 drbd/drbd_transport_tcp.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drbd/drbd_transport_tcp.c +1 -4
index 203dc4e0..3b1d4273 100644
@@ -883,7 +883,6 @@ static int dtt_connect(struct drbd_transport *transport)
 	struct socket *dsocket, *csocket;
 	struct net_conf *nc;
 	int timeout, err;
-	int one = 1;
 	bool ok;
 
 	dsocket = NULL;
@@ -1065,9 +1064,7 @@ randomize:
 	dsocket->sk->sk_sndtimeo = timeout;
 	csocket->sk->sk_sndtimeo = timeout;
 
-	err = kernel_setsockopt(dsocket, SOL_SOCKET, SO_KEEPALIVE, (char *)&one, sizeof(one));
-	if (err)
-		tr_warn(transport, "Failed to enable SO_KEEPALIVE %d\n", err);
+	sock_set_keepalive(dsocket->sk);
 
 	return 0;
 
Signature · 1 line
2.17.1
#6 lore raw

PATCH 6/8 drbd: get rid of set_fs() usage

In reply to Michael D Labriola

5.10 is getting rid of set_fs(). When I went looking through git commits to figure out how to properly protect sendpage, what I found was this commit from 4.13:

    10428957 drbd: ->sendpage() never needed set_fs()
---
 drbd/drbd_transport_tcp.c | 3 ---
 1 file changed, 3 deletions(-)
drbd/drbd_transport_tcp.c +0 -3
index 3b1d4273..4e076460 100644
@@ -1138,7 +1138,6 @@ static int dtt_send_page(struct drbd_transport *transport, enum drbd_stream stre
 	struct drbd_tcp_transport *tcp_transport =
 		container_of(transport, struct drbd_tcp_transport, transport);
 	struct socket *socket = tcp_transport->stream[stream];
-	mm_segment_t oldfs = get_fs();
 	int len = size;
 	int err = -EIO;
 
@@ -1147,7 +1146,6 @@ static int dtt_send_page(struct drbd_transport *transport, enum drbd_stream stre
 
 	msg_flags |= MSG_NOSIGNAL;
 	dtt_update_congested(tcp_transport);
-	set_fs(KERNEL_DS);
 	do {
 		int sent;
 
@@ -1173,7 +1171,6 @@ static int dtt_send_page(struct drbd_transport *transport, enum drbd_stream stre
 		 * and add that to the while() condition below.
 		 */
 	} while (len > 0 /* THINK && peer_device->repl_state[NOW] >= L_ESTABLISHED */);
-	set_fs(oldfs);
 	clear_bit(NET_CONGESTED, &tcp_transport->transport.flags);
 
 	if (len == 0)
Signature · 1 line
2.17.1
#7 lore raw

PATCH 7/8 drbd: don't use sched_setscheduler()

In reply to Michael D Labriola

The sched_setscheduler() function was removed in 5.9. Use sched_set_fifo_low() like the mainline drbd8 module did here:

c9ec0524 sched,drbd: Convert to sched_set_fifo*()

---
 drbd/drbd_receiver.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drbd/drbd_receiver.c +1 -4
index 46368949..6fb2ec39 100644
@@ -9390,13 +9390,10 @@ int drbd_ack_receiver(struct drbd_thread *thi)
 	unsigned int header_size = drbd_header_size(connection);
 	int expect   = header_size;
 	bool ping_timeout_active = false;
-	struct sched_param param = { .sched_priority = 2 };
 	struct drbd_transport *transport = &connection->transport;
 	struct drbd_transport_ops *tr_ops = transport->ops;
 
-	rv = sched_setscheduler(current, SCHED_RR, &param);
-	if (rv < 0)
-		drbd_err(connection, "drbd_ack_receiver: ERROR set priority, ret=%d\n", rv);
+	sched_set_fifo_low(current);
 
 	while (get_t_state(thi) == RUNNING) {
 		drbd_thread_current_set_cpu(thi);
Signature · 1 line
2.17.1
#8 lore raw

PATCH 8/8 compat: remove set_bdi_cap_stable_writes

In reply to Michael D Labriola

BDI_CAP_STABLE_WRITES goes away in 5.10, resulting in tons of warnings at compile time. It doesn't look like anything in either drbd or drbd-utils is using the resulting set_bdi_cap_stable_writes macro anyways, so this commit removes it.

---
 drbd/drbd-kernel-compat/drbd_wrappers.h | 7 -------
 1 file changed, 7 deletions(-)
drbd/drbd-kernel-compat/drbd_wrappers.h +0 -7
index c0d8981c..01fdbf22 100644
@@ -127,13 +127,6 @@ static inline void blk_queue_write_cache(struct request_queue *q, bool enabled,
 #define KREF_INIT(N) { ATOMIC_INIT(N) }
 #endif
 
-#ifdef BDI_CAP_STABLE_WRITES /* >= v3.9 */
-#define set_bdi_cap_stable_writes(cap)	do { (cap) |= BDI_CAP_STABLE_WRITES; } while (0)
-#else /* < v3.9 */
-#warning "BDI_CAP_STABLE_WRITES not available"
-#define set_bdi_cap_stable_writes(cap)	do { } while (0)
-#endif
-
 #ifdef COMPAT_HAVE_POINTER_BACKING_DEV_INFO /* >= v4.11 */
 #define bdi_from_device(device) (device->ldev->backing_bdev->bd_disk->queue->backing_dev_info)
 #else /* < v4.11 */
Signature · 1 line
2.17.1
#9 lore raw
In reply to Michael D Labriola

Hi Michael,

Thanks for this great contribution!

Unfortunately, we duplicated some work here, but neither of us could have known that :)
Over the last few days I fabricated some patches that fix some of the same problems, ultimately leading to the same result.

I'll review your patches – on first glance it seems like some of your approaches may actually be better than mine. If the patches are exactly the same, I'll attribute you since you published your patches first. I'll mix and match with my patches, since they are more thorough in some places (for example, I did go to the trouble of creating a compat test for sched_setscheduler()).

Some of these patches are still under internal review, so this may take some time, but I'll let you know which of yours I chose.

Thanks again!

--
Christoph Böhmwalder
LINBIT | Keeping the Digital World Running DRBD HA — Disaster Recovery — Software defined Storage

Quoted 21 lines · Michael D Labriola

Hey, everyone! I had the chance to try building the latest drbd-9.0 code against the 5.9 and 5.10 kernels this afternoon. Things almost built, but not quite. I made a series of commits to get things compiling. Most of it is trivial compat test changes and following along w/ what's already been done in the mainline drbd8 module to get it ready for 5.10.

I did not create a compat test for the switch from sched_setscheduler() to sched_set_fifo_low() yet. Mostly, because I've never done that and have to wrap my head around how to add another test. ;-)

I tried submitting a pull request via github, but got the auto-close message. So, here's take two.

_______________________________________________
drbd-dev mailing list
[email protected]
https://lists.linbit.com/mailman/listinfo/drbd-dev

#10 lore raw

Thanks, Christoph. I've been using DRBD for 15+ years, so it's about time I officially contributed something other than bug reports. ;-)

Quoted 49 lines · Christoph Böhmwalder

Hi Michael,

Thanks for this great contribution!

Unfortunately, we duplicated some work here, but neither of us could have known that :)
Over the last few days I fabricated some patches that fix some of the same problems, ultimately leading to the same result.

I'll review your patches – on first glance it seems like some of your approaches may actually be better than mine. If the patches are exactly the same, I'll attribute you since you published your patches first. I'll mix and match with my patches, since they are more thorough in some places (for example, I did go to the trouble of creating a compat test for sched_setscheduler()).

Some of these patches are still under internal review, so this may take some time, but I'll let you know which of yours I chose.

Thanks again!

--
Christoph Böhmwalder
LINBIT | Keeping the Digital World Running DRBD HA — Disaster Recovery — Software defined Storage

Quoted 21 lines · Michael D Labriola

Hey, everyone! I had the chance to try building the latest drbd-9.0 code against the 5.9 and 5.10 kernels this afternoon. Things almost built, but not quite. I made a series of commits to get things compiling. Most of it is trivial compat test changes and following along w/ what's already been done in the mainline drbd8 module to get it ready for 5.10.

I did not create a compat test for the switch from sched_setscheduler() to sched_set_fifo_low() yet. Mostly, because I've never done that and have to wrap my head around how to add another test. ;-)

I tried submitting a pull request via github, but got the auto-close message. So, here's take two.

_______________________________________________
drbd-dev mailing list
[email protected]
https://lists.linbit.com/mailman/listinfo/drbd-dev

Signature · 4 lines
Michael D Labriola
21 Rip Van Winkle Cir
Warwick, RI 02886
401-316-9844 (cell)