[PATCH v2 for-rc] RDMA/rxe: Fix panic when calling kmem_cache_create()

[PATCH v2 for-rc] RDMA/rxe: Fix panic when calling kmem_cache_create()

[Kamal Heib]

To avoid the following kernel panic when calling kmem_cache_create()
with a NULL pointer from pool_cache(), move the rxe_cache_init() to the
context of device initialization.

 BUG: unable to handle kernel NULL pointer dereference at 000000000000000b
 PGD 0 P4D 0
 Oops: 0000 [#1] SMP NOPTI
 CPU: 4 PID: 8512 Comm: modprobe Kdump: loaded Not tainted 4.18.0-231.el8.x86_64 #1
 Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 10/02/2018
 RIP: 0010:kmem_cache_alloc+0xd1/0x1b0
 Code: 8b 57 18 45 8b 77 1c 48 8b 5c 24 30 0f 1f 44 00 00 5b 48 89 e8 5d 41 5c 41 5d 41 5e 41 5f c3 81 e3 00 00 10 00 75 0e 4d 89 fe <41> f6 47 0b 04 0f 84 6c ff ff ff 4c 89 ff e8 cc da 01 00 49 89 c6
 RSP: 0018:ffffa2b8c773f9d0 EFLAGS: 00010246
 RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000005
 RDX: 0000000000000004 RSI: 00000000006080c0 RDI: 0000000000000000
 RBP: ffff8ea0a8634fd0 R08: ffffa2b8c773f988 R09: 00000000006000c0
 R10: 0000000000000000 R11: 0000000000000230 R12: 00000000006080c0
 R13: ffffffffc0a97fc8 R14: 0000000000000000 R15: 0000000000000000
 FS:  00007f9138ed9740(0000) GS:ffff8ea4ae800000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 000000000000000b CR3: 000000046d59a000 CR4: 00000000003406e0
 Call Trace:
  rxe_alloc+0xc8/0x160 [rdma_rxe]
  rxe_get_dma_mr+0x25/0xb0 [rdma_rxe]
  __ib_alloc_pd+0xcb/0x160 [ib_core]
  ib_mad_init_device+0x296/0x8b0 [ib_core]
  add_client_context+0x11a/0x160 [ib_core]
  enable_device_and_get+0xdc/0x1d0 [ib_core]
  ib_register_device+0x572/0x6b0 [ib_core]
  ? crypto_create_tfm+0x32/0xe0
  ? crypto_create_tfm+0x7a/0xe0
  ? crypto_alloc_tfm+0x58/0xf0
  rxe_register_device+0x19d/0x1c0 [rdma_rxe]
  rxe_net_add+0x3d/0x70 [rdma_rxe]
  ? dev_get_by_name_rcu+0x73/0x90
  rxe_param_set_add+0xaf/0xc0 [rdma_rxe]
  parse_args+0x179/0x370
  ? ref_module+0x1b0/0x1b0
  load_module+0x135e/0x17e0
  ? ref_module+0x1b0/0x1b0
  ? __do_sys_init_module+0x13b/0x180
  __do_sys_init_module+0x13b/0x180
  do_syscall_64+0x5b/0x1a0
  entry_SYSCALL_64_after_hwframe+0x65/0xca
 RIP: 0033:0x7f9137ed296e

Fixes: 8700e3e7c485 ("Soft RoCE driver")
Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
---
v2: Call rxe_cache_init() from rxe_init().
---
 drivers/infiniband/sw/rxe/rxe.c      | 14 +++++++-------
 drivers/infiniband/sw/rxe/rxe_pool.c |  3 +++
 2 files changed, 10 insertions(+), 7 deletions(-)

diff --git a/drivers/infiniband/sw/rxe/rxe.c b/drivers/infiniband/sw/rxe/rxe.c
index 5642eefb4ba1..c80f7c183957 100644
--- a/drivers/infiniband/sw/rxe/rxe.c
+++ b/drivers/infiniband/sw/rxe/rxe.c
@@ -252,6 +252,13 @@ static int rxe_init(struct rxe_dev *rxe)
 	/* init default device parameters */
 	rxe_init_device_param(rxe);
 
+	/* initialize slab caches for managed objects */
+	err = rxe_cache_init();
+	if (err) {
+		pr_err("unable to init object pools\n");
+		return err;
+	}
+
 	err = rxe_init_ports(rxe);
 	if (err)
 		goto err1;
@@ -336,13 +343,6 @@ static int __init rxe_module_init(void)
 {
 	int err;
 
-	/* initialize slab caches for managed objects */
-	err = rxe_cache_init();
-	if (err) {
-		pr_err("unable to init object pools\n");
-		return err;
-	}
-
 	err = rxe_net_init();
 	if (err)
 		return err;
diff --git a/drivers/infiniband/sw/rxe/rxe_pool.c b/drivers/infiniband/sw/rxe/rxe_pool.c
index fbcbac52290b..06c6d1f835b7 100644
--- a/drivers/infiniband/sw/rxe/rxe_pool.c
+++ b/drivers/infiniband/sw/rxe/rxe_pool.c
@@ -139,6 +139,9 @@ int rxe_cache_init(void)
 	for (i = 0; i < RXE_NUM_TYPES; i++) {
 		type = &rxe_type_info[i];
 		size = ALIGN(type->size, RXE_POOL_ALIGN);
+		if (type->cache)
+			continue;
+
 		if (!(type->flags & RXE_POOL_NO_ALLOC)) {
 			type->cache =
 				kmem_cache_create(type->name, size,
-- 
2.26.2

Re: [PATCH v2 for-rc] RDMA/rxe: Fix panic when calling kmem_cache_create()

[Jason Gunthorpe]

On Tue, Aug 18, 2020 at 05:25:04PM +0300, Kamal Heib wrote:
> To avoid the following kernel panic when calling kmem_cache_create()
> with a NULL pointer from pool_cache(), move the rxe_cache_init() to the
> context of device initialization.

I think you've hit on a bigger bug than just this oops.

rxe_net_add() should never be called before rxe_module_init(), that
surely subtly breaks all kinds of things.

Maybe it is time to remove these module parameters?

Jason

Re: [PATCH v2 for-rc] RDMA/rxe: Fix panic when calling kmem_cache_create()

[Kamal Heib]

On Tue, Aug 18, 2020 at 01:31:57PM -0300, Jason Gunthorpe wrote:
> On Tue, Aug 18, 2020 at 05:25:04PM +0300, Kamal Heib wrote:
> > To avoid the following kernel panic when calling kmem_cache_create()
> > with a NULL pointer from pool_cache(), move the rxe_cache_init() to the
> > context of device initialization.
> 
> I think you've hit on a bigger bug than just this oops.
> 
> rxe_net_add() should never be called before rxe_module_init(), that
> surely subtly breaks all kinds of things.
> 
> Maybe it is time to remove these module parameters?
>
Yes, I agree, this can be done in for-next.

But at least can we take this patch to for-rc (stable) to fix this issue
in stable releases?

Thanks,
Kamal

> Jason

Re: [PATCH v2 for-rc] RDMA/rxe: Fix panic when calling kmem_cache_create()

[Jason Gunthorpe]

On Wed, Aug 19, 2020 at 12:15:45AM +0300, Kamal Heib wrote:
> On Tue, Aug 18, 2020 at 01:31:57PM -0300, Jason Gunthorpe wrote:
> > On Tue, Aug 18, 2020 at 05:25:04PM +0300, Kamal Heib wrote:
> > > To avoid the following kernel panic when calling kmem_cache_create()
> > > with a NULL pointer from pool_cache(), move the rxe_cache_init() to the
> > > context of device initialization.
> > 
> > I think you've hit on a bigger bug than just this oops.
> > 
> > rxe_net_add() should never be called before rxe_module_init(), that
> > surely subtly breaks all kinds of things.
> > 
> > Maybe it is time to remove these module parameters?
> >
> Yes, I agree, this can be done in for-next.
> 
> But at least can we take this patch to for-rc (stable) to fix this issue
> in stable releases?

If you want to fix something in stable then block the module options
from working as actual module options - eg before rxe_module_init()
runs.

Jason

Re: [PATCH v2 for-rc] RDMA/rxe: Fix panic when calling kmem_cache_create()

[Kamal Heib]

On Thu, Aug 20, 2020 at 08:37:17AM -0300, Jason Gunthorpe wrote:
> On Wed, Aug 19, 2020 at 12:15:45AM +0300, Kamal Heib wrote:
> > On Tue, Aug 18, 2020 at 01:31:57PM -0300, Jason Gunthorpe wrote:
> > > On Tue, Aug 18, 2020 at 05:25:04PM +0300, Kamal Heib wrote:
> > > > To avoid the following kernel panic when calling kmem_cache_create()
> > > > with a NULL pointer from pool_cache(), move the rxe_cache_init() to the
> > > > context of device initialization.
> > > 
> > > I think you've hit on a bigger bug than just this oops.
> > > 
> > > rxe_net_add() should never be called before rxe_module_init(), that
> > > surely subtly breaks all kinds of things.
> > > 
> > > Maybe it is time to remove these module parameters?
> > >
> > Yes, I agree, this can be done in for-next.
> > 
> > But at least can we take this patch to for-rc (stable) to fix this issue
> > in stable releases?
> 
> If you want to fix something in stable then block the module options
> from working as actual module options - eg before rxe_module_init()
> runs.
> 
> Jason

Something like the following patch?

diff --git a/drivers/infiniband/sw/rxe/rxe.c b/drivers/infiniband/sw/rxe/rxe.c
index 907203afbd99..872ebc57ac06 100644
--- a/drivers/infiniband/sw/rxe/rxe.c
+++ b/drivers/infiniband/sw/rxe/rxe.c
@@ -40,6 +40,8 @@ MODULE_AUTHOR("Bob Pearson, Frank Zago, John Groves, Kamal Heib");
 MODULE_DESCRIPTION("Soft RDMA transport");
 MODULE_LICENSE("Dual BSD/GPL");

+bool rxe_is_loaded = false;
+
 /* free resources for a rxe device all objects created for this device must
  * have been destroyed
  */
@@ -315,6 +317,7 @@ static int __init rxe_module_init(void)
                return err;

        rdma_link_register(&rxe_link_ops);
+       rxe_is_loaded = true;
        pr_info("loaded\n");
        return 0;
 }
@@ -326,6 +329,7 @@ static void __exit rxe_module_exit(void)
        rxe_net_exit();
        rxe_cache_exit();

+       rxe_is_loaded = false;
        pr_info("unloaded\n");
 }

diff --git a/drivers/infiniband/sw/rxe/rxe.h b/drivers/infiniband/sw/rxe/rxe.h
index fb07eed9e402..d9b71b5e2fba 100644
--- a/drivers/infiniband/sw/rxe/rxe.h
+++ b/drivers/infiniband/sw/rxe/rxe.h
@@ -67,6 +67,8 @@

 #define RXE_ROCE_V2_SPORT              (0xc000)

+extern bool rxe_is_loaded;
+
 static inline u32 rxe_crc32(struct rxe_dev *rxe,
                            u32 crc, void *next, size_t len)
 {
diff --git a/drivers/infiniband/sw/rxe/rxe_sysfs.c b/drivers/infiniband/sw/rxe/rxe_sysfs.c
index ccda5f5a3bc0..12c7ca0764d5 100644
--- a/drivers/infiniband/sw/rxe/rxe_sysfs.c
+++ b/drivers/infiniband/sw/rxe/rxe_sysfs.c
@@ -61,6 +61,11 @@ static int rxe_param_set_add(const char *val, const struct kernel_param *kp)
        struct net_device *ndev;
        struct rxe_dev *exists;

+       if (!rxe_is_loaded) {
+               pr_err("Please make sure to load the rdma_rxe module first\n");
+               return -EINVAL;
+       }
+
        len = sanitize_arg(val, intf, sizeof(intf));
        if (!len) {
                pr_err("add: invalid interface name\n");

Thanks,
Kamal

Re: [PATCH v2 for-rc] RDMA/rxe: Fix panic when calling kmem_cache_create()

[Leon Romanovsky]

On Sun, Aug 23, 2020 at 10:45:58PM +0300, Kamal Heib wrote:
> On Thu, Aug 20, 2020 at 08:37:17AM -0300, Jason Gunthorpe wrote:
> > On Wed, Aug 19, 2020 at 12:15:45AM +0300, Kamal Heib wrote:
> > > On Tue, Aug 18, 2020 at 01:31:57PM -0300, Jason Gunthorpe wrote:
> > > > On Tue, Aug 18, 2020 at 05:25:04PM +0300, Kamal Heib wrote:
> > > > > To avoid the following kernel panic when calling kmem_cache_create()
> > > > > with a NULL pointer from pool_cache(), move the rxe_cache_init() to the
> > > > > context of device initialization.
> > > >
> > > > I think you've hit on a bigger bug than just this oops.
> > > >
> > > > rxe_net_add() should never be called before rxe_module_init(), that
> > > > surely subtly breaks all kinds of things.
> > > >
> > > > Maybe it is time to remove these module parameters?
> > > >
> > > Yes, I agree, this can be done in for-next.
> > >
> > > But at least can we take this patch to for-rc (stable) to fix this issue
> > > in stable releases?
> >
> > If you want to fix something in stable then block the module options
> > from working as actual module options - eg before rxe_module_init()
> > runs.
> >
> > Jason
>
> Something like the following patch?

If you want to got this path, it will be more like this: request_module("rdma_rxe")

Thanks

>
> diff --git a/drivers/infiniband/sw/rxe/rxe.c b/drivers/infiniband/sw/rxe/rxe.c
> index 907203afbd99..872ebc57ac06 100644
> --- a/drivers/infiniband/sw/rxe/rxe.c
> +++ b/drivers/infiniband/sw/rxe/rxe.c
> @@ -40,6 +40,8 @@ MODULE_AUTHOR("Bob Pearson, Frank Zago, John Groves, Kamal Heib");
>  MODULE_DESCRIPTION("Soft RDMA transport");
>  MODULE_LICENSE("Dual BSD/GPL");
>
> +bool rxe_is_loaded = false;
> +
>  /* free resources for a rxe device all objects created for this device must
>   * have been destroyed
>   */
> @@ -315,6 +317,7 @@ static int __init rxe_module_init(void)
>                 return err;
>
>         rdma_link_register(&rxe_link_ops);
> +       rxe_is_loaded = true;
>         pr_info("loaded\n");
>         return 0;
>  }
> @@ -326,6 +329,7 @@ static void __exit rxe_module_exit(void)
>         rxe_net_exit();
>         rxe_cache_exit();
>
> +       rxe_is_loaded = false;
>         pr_info("unloaded\n");
>  }
>
> diff --git a/drivers/infiniband/sw/rxe/rxe.h b/drivers/infiniband/sw/rxe/rxe.h
> index fb07eed9e402..d9b71b5e2fba 100644
> --- a/drivers/infiniband/sw/rxe/rxe.h
> +++ b/drivers/infiniband/sw/rxe/rxe.h
> @@ -67,6 +67,8 @@
>
>  #define RXE_ROCE_V2_SPORT              (0xc000)
>
> +extern bool rxe_is_loaded;
> +
>  static inline u32 rxe_crc32(struct rxe_dev *rxe,
>                             u32 crc, void *next, size_t len)
>  {
> diff --git a/drivers/infiniband/sw/rxe/rxe_sysfs.c b/drivers/infiniband/sw/rxe/rxe_sysfs.c
> index ccda5f5a3bc0..12c7ca0764d5 100644
> --- a/drivers/infiniband/sw/rxe/rxe_sysfs.c
> +++ b/drivers/infiniband/sw/rxe/rxe_sysfs.c
> @@ -61,6 +61,11 @@ static int rxe_param_set_add(const char *val, const struct kernel_param *kp)
>         struct net_device *ndev;
>         struct rxe_dev *exists;
>
> +       if (!rxe_is_loaded) {
> +               pr_err("Please make sure to load the rdma_rxe module first\n");
> +               return -EINVAL;
> +       }
> +
>         len = sanitize_arg(val, intf, sizeof(intf));
>         if (!len) {
>                 pr_err("add: invalid interface name\n");
>
> Thanks,
> Kamal

Re: [PATCH v2 for-rc] RDMA/rxe: Fix panic when calling kmem_cache_create()

[Jason Gunthorpe]

On Sun, Aug 23, 2020 at 10:45:58PM +0300, Kamal Heib wrote:
> On Thu, Aug 20, 2020 at 08:37:17AM -0300, Jason Gunthorpe wrote:
> > On Wed, Aug 19, 2020 at 12:15:45AM +0300, Kamal Heib wrote:
> > > On Tue, Aug 18, 2020 at 01:31:57PM -0300, Jason Gunthorpe wrote:
> > > > On Tue, Aug 18, 2020 at 05:25:04PM +0300, Kamal Heib wrote:
> > > > > To avoid the following kernel panic when calling kmem_cache_create()
> > > > > with a NULL pointer from pool_cache(), move the rxe_cache_init() to the
> > > > > context of device initialization.
> > > > 
> > > > I think you've hit on a bigger bug than just this oops.
> > > > 
> > > > rxe_net_add() should never be called before rxe_module_init(), that
> > > > surely subtly breaks all kinds of things.
> > > > 
> > > > Maybe it is time to remove these module parameters?
> > > >
> > > Yes, I agree, this can be done in for-next.
> > > 
> > > But at least can we take this patch to for-rc (stable) to fix this issue
> > > in stable releases?
> > 
> > If you want to fix something in stable then block the module options
> > from working as actual module options - eg before rxe_module_init()
> > runs.
> > 
> > Jason
> 
> Something like the following patch?

Yes, something more like that

Jason

Re: [PATCH v2 for-rc] RDMA/rxe: Fix panic when calling kmem_cache_create()

[Kamal Heib]

On Mon, Aug 24, 2020 at 10:47:23AM -0300, Jason Gunthorpe wrote:
> On Sun, Aug 23, 2020 at 10:45:58PM +0300, Kamal Heib wrote:
> > On Thu, Aug 20, 2020 at 08:37:17AM -0300, Jason Gunthorpe wrote:
> > > On Wed, Aug 19, 2020 at 12:15:45AM +0300, Kamal Heib wrote:
> > > > On Tue, Aug 18, 2020 at 01:31:57PM -0300, Jason Gunthorpe wrote:
> > > > > On Tue, Aug 18, 2020 at 05:25:04PM +0300, Kamal Heib wrote:
> > > > > > To avoid the following kernel panic when calling kmem_cache_create()
> > > > > > with a NULL pointer from pool_cache(), move the rxe_cache_init() to the
> > > > > > context of device initialization.
> > > > > 
> > > > > I think you've hit on a bigger bug than just this oops.
> > > > > 
> > > > > rxe_net_add() should never be called before rxe_module_init(), that
> > > > > surely subtly breaks all kinds of things.
> > > > > 
> > > > > Maybe it is time to remove these module parameters?
> > > > >
> > > > Yes, I agree, this can be done in for-next.
> > > > 
> > > > But at least can we take this patch to for-rc (stable) to fix this issue
> > > > in stable releases?
> > > 
> > > If you want to fix something in stable then block the module options
> > > from working as actual module options - eg before rxe_module_init()
> > > runs.
> > > 
> > > Jason
> > 
> > Something like the following patch?
> 
> Yes, something more like that
> 
> Jason

OK, Thanks!

I'll send v3 soon.

Thanks,
Kamal