Various fixes in RDMA/core
cover Various fixes in RDMA/core
From: Leon Romanovsky <[email protected]>
Hi,
This is compilation of unrelated fixes, one is an outcome of syzkaller report and other two were found in our regression.
Thanks
Leon Romanovsky (1):
RDMA/ucma: Protect mc during concurrent multicast leaves
Maor Gottlieb (2):
RDMA/cma: Use correct address when leaving multicast group RDMA/core: Set MR type in ib_reg_user_mr
drivers/infiniband/core/cma.c | 61 ++++++++++++++++----------------- drivers/infiniband/core/ucma.c | 34 ++++++++++++------ drivers/infiniband/core/verbs.c | 1 + 3 files changed, 54 insertions(+), 42 deletions(-)
Signature · 1 line
2.34.1
PATCH 1/3 RDMA/cma: Use correct address when leaving multicast group
From: Maor Gottlieb <[email protected]>
In RoCE we should use cma_iboe_set_mgid and not cma_set_mgid to generate the mgid, otherwise we will try to remove incorrect address.
--- drivers/infiniband/core/cma.c | 61 +++++++++++++++++------------------ 1 file changed, 30 insertions(+), 31 deletions(-)
drivers/infiniband/core/cma.c +30 -31
@@ -1830,6 +1830,31 @@ static void cma_release_port(struct rdma_id_private *id_priv)mutex_unlock (&lock); } +static void cma_iboe_set_mgid (struct sockaddr *addr,union ib_gid *mgid, +enum ib_gid_type gid_type) +{ +struct sockaddr_in *sin = (struct sockaddr_in *)addr; +struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr; + +if (!cma_any_addr (addr) && addr->sa_family ==AF_INET6 ) { +memcpy (mgid, &sin6->sin6_addr,sizeof (*mgid)); +return ; + } + +memset (mgid,0 ,sizeof (*mgid)); +if (cma_any_addr (addr)) +return ; + +/* AF_INET4 */ +if (gid_type !=IB_GID_TYPE_ROCE_UDP_ENCAP ) { + mgid->raw[0 ] =0xff ; + mgid->raw[1 ] =0x0e ; + } + mgid->raw[10 ] =0xff ; + mgid->raw[11 ] =0xff ; + *(__be32 *)(&mgid->raw[12 ]) = sin->sin_addr.s_addr; +} +static void destroy_mc (struct rdma_id_private *id_priv,struct cma_multicast *mc) {
@@ -1847,10 +1872,13 @@ static void destroy_mc(struct rdma_id_private *id_priv, ndev =dev_get_by_index (dev_addr->net, dev_addr->bound_dev_if);if (ndev) { +enum ib_gid_type gid_type;union ib_gid mgid; -cma_set_mgid (id_priv, (struct sockaddr *)&mc->addr, - &mgid); + gid_type =cma_get_default_gid_type ( + id_priv->cma_dev, id_priv->id.port_num); +cma_iboe_set_mgid ((struct sockaddr *)&mc->addr, &mgid, + gid_type);if (!send_only)cma_igmp_send (ndev, &mgid, false);
@@ -4702,35 +4730,6 @@ static int cma_join_ib_multicast(struct rdma_id_private *id_priv,return PTR_ERR_OR_ZERO (mc->sa_mc); } -static void cma_iboe_set_mgid (struct sockaddr *addr,union ib_gid *mgid, -enum ib_gid_type gid_type) -{ -struct sockaddr_in *sin = (struct sockaddr_in *)addr; -struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr; - -if (cma_any_addr (addr)) { -memset (mgid,0 ,sizeof *mgid); - }else if (addr->sa_family ==AF_INET6 ) { -memcpy (mgid, &sin6->sin6_addr,sizeof *mgid); - }else { - mgid->raw[0 ] = - (gid_type ==IB_GID_TYPE_ROCE_UDP_ENCAP ) ?0 :0xff ; - mgid->raw[1 ] = - (gid_type ==IB_GID_TYPE_ROCE_UDP_ENCAP ) ?0 :0x0e ; - mgid->raw[2 ] =0 ; - mgid->raw[3 ] =0 ; - mgid->raw[4 ] =0 ; - mgid->raw[5 ] =0 ; - mgid->raw[6 ] =0 ; - mgid->raw[7 ] =0 ; - mgid->raw[8 ] =0 ; - mgid->raw[9 ] =0 ; - mgid->raw[10 ] =0xff ; - mgid->raw[11 ] =0xff ; - *(__be32 *)(&mgid->raw[12 ]) = sin->sin_addr.s_addr; - } -} -static int cma_iboe_join_multicast (struct rdma_id_private *id_priv,struct cma_multicast *mc) {
Signature · 1 line
2.34.1
PATCH 3/3 RDMA/core: Set MR type in ib_reg_user_mr
In reply to Leon RomanovskyFrom: Maor Gottlieb <[email protected]>
Add missing assignment of MR type to IB_MR_TYPE_USER.
--- drivers/infiniband/core/verbs.c | 1 + 1 file changed, 1 insertion(+)
drivers/infiniband/core/verbs.c +1 -0
@@ -2153,6 +2153,7 @@ struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,return mr; mr->device = pd->device; + mr->type =IB_MR_TYPE_USER ; mr->pd = pd; mr->dm =NULL ;atomic_inc (&pd->usecnt);
Signature · 1 line
2.34.1
PATCH 2/3 RDMA/ucma: Protect mc during concurrent multicast leaves
In reply to Leon RomanovskyFrom: Leon Romanovsky <[email protected]>
Partially revert the commit mentioned in the Fixes line to make sure that allocation and erasing multicast struct are locked.
==================================================================
BUG: KASAN: use-after-free in ucma_cleanup_multicast drivers/infiniband/core/ucma.c:491 [inline] BUG: KASAN: use-after-free in ucma_destroy_private_ctx+0x914/0xb70 drivers/infiniband/core/ucma.c:579 Read of size 8 at addr ffff88801bb74b00 by task syz-executor.1/25529 CPU: 0 PID: 25529 Comm: syz-executor.1 Not tainted 5.16.0-rc7-syzkaller #0 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: <TASK> __dump_stack lib/dump_stack.c:88 [inline] dump_stack_lvl+0xcd/0x134 lib/dump_stack.c:106 print_address_description.constprop.0.cold+0x8d/0x320 mm/kasan/report.c:247 __kasan_report mm/kasan/report.c:433 [inline] kasan_report.cold+0x83/0xdf mm/kasan/report.c:450 ucma_cleanup_multicast drivers/infiniband/core/ucma.c:491 [inline] ucma_destroy_private_ctx+0x914/0xb70 drivers/infiniband/core/ucma.c:579 ucma_destroy_id+0x1e6/0x280 drivers/infiniband/core/ucma.c:614 ucma_write+0x25c/0x350 drivers/infiniband/core/ucma.c:1732 vfs_write+0x28e/0xae0 fs/read_write.c:588 ksys_write+0x1ee/0x250 fs/read_write.c:643 do_syscall_x64 arch/x86/entry/common.c:50 [inline] do_syscall_64+0x35/0xb0 arch/x86/entry/common.c:80 entry_SYSCALL_64_after_hwframe+0x44/0xae RIP: 0033:0x7f2fcd207e99 Code: ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b
4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 bc ff ff ff f7 d8 64 89 01 48
RSP: 002b:00007f2fcbb7d168 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 00007f2fcd31af60 RCX: 00007f2fcd207e99
RDX: 0000000000000018 RSI: 00000000200000c0 RDI: 0000000000000004
RBP: 00007f2fcd261ff1 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
R13: 00007fff66a135bf R14: 00007f2fcbb7d300 R15: 0000000000022000
</TASK>
--- drivers/infiniband/core/ucma.c | 34 +++++++++++++++++++++++----------- 1 file changed, 23 insertions(+), 11 deletions(-)
drivers/infiniband/core/ucma.c +23 -11
@@ -95,6 +95,7 @@ struct ucma_context {u64 uid;struct list_head list; +struct list_head mc_list;struct work_struct close_work; };
@@ -105,6 +106,7 @@ struct ucma_multicast {u64 uid;u8 join_state; +struct list_head list;struct sockaddr_storage addr; };
@@ -198,6 +200,7 @@ static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file)INIT_WORK (&ctx->close_work, ucma_close_id);init_completion (&ctx->comp); +INIT_LIST_HEAD (&ctx->mc_list);/* So list_del() will work if we don't do ucma_finish_ctx() */ INIT_LIST_HEAD (&ctx->list); ctx->file = file;
@@ -484,19 +487,19 @@ static ssize_t ucma_create_id(struct ucma_file *file, const char __user *inbuf,static void ucma_cleanup_multicast (struct ucma_context *ctx) { -struct ucma_multicast *mc; -unsigned long index; +struct ucma_multicast *mc, *tmp; -xa_for_each (&multicast_table, index, mc) { -if (mc->ctx != ctx) -continue ; +xa_lock (&multicast_table); +list_for_each_entry_safe (mc, tmp, &ctx->mc_list, list) { +list_del (&mc->list);/* * At this point mc->ctx->ref is 0 so the mc cannot leave the * lock on the reader and this is enough serialization */ -xa_erase (&multicast_table, index); +__xa_erase (&multicast_table, mc->id);kfree (mc); } +xa_unlock (&multicast_table); }static void ucma_cleanup_mc_events (struct ucma_multicast *mc)
@@ -1469,12 +1472,16 @@ static ssize_t ucma_process_join(struct ucma_file *file, mc->uid = cmd->uid;memcpy (&mc->addr, addr, cmd->addr_size); -if (xa_alloc (&multicast_table, &mc->id,NULL , xa_limit_32b, +xa_lock (&multicast_table); +if (__xa_alloc (&multicast_table, &mc->id,NULL , xa_limit_32b,GFP_KERNEL )) { ret = -ENOMEM ; goto err_free_mc; } +list_add_tail (&mc->list, &ctx->mc_list); +xa_unlock (&multicast_table); +mutex_lock (&ctx->mutex); ret =rdma_join_multicast (ctx->cm_id, (struct sockaddr *)&mc->addr, join_state, mc);
@@ -1500,8 +1507,11 @@ static ssize_t ucma_process_join(struct ucma_file *file,mutex_unlock (&ctx->mutex);ucma_cleanup_mc_events (mc); err_xa_erase: -xa_erase (&multicast_table, mc->id); +xa_lock (&multicast_table); +list_del (&mc->list); +__xa_erase (&multicast_table, mc->id); err_free_mc: +xa_unlock (&multicast_table);kfree (mc); err_put_ctx:ucma_put_ctx (ctx);
@@ -1569,15 +1579,17 @@ static ssize_t ucma_leave_multicast(struct ucma_file *file, mc =ERR_PTR (-EINVAL ); else if (!refcount_inc_not_zero(&mc->ctx->ref)) mc =ERR_PTR (-ENXIO ); - else - __xa_erase(&multicast_table, mc->id); -xa_unlock (&multicast_table);if (IS_ERR (mc)) { +xa_unlock (&multicast_table); ret =PTR_ERR (mc); goto out; } +list_del (&mc->list); +__xa_erase (&multicast_table, mc->id); +xa_unlock (&multicast_table); +mutex_lock (&mc->ctx->mutex);rdma_leave_multicast (mc->ctx->cm_id, (struct sockaddr *) &mc->addr);mutex_unlock (&mc->ctx->mutex);
Signature · 1 line
2.34.1
Jason Gunthorpe
maintainerRe: [PATCH rdma-next 1/3] RDMA/cma: Use correct address when leaving multicast group
In reply to Leon RomanovskyQuoted 11 lines · Leon Romanovsky
From: Maor Gottlieb <[email protected]>
In RoCE we should use
cma_iboe_set_mgidand notcma_set_mgidto generate the mgid, otherwise we will try to remove incorrect address.Fixes b5de0c60cc30 ("RDMA/cma: Fix use after free race in roce multicast join")Signed-off-by Maor Gottlieb <[email protected]>Signed-off-by Leon Romanovsky <[email protected]>--- drivers/infiniband/core/cma.c | 61 +++++++++++++++++------------------ 1 file changed, 30 insertions(+), 31 deletions(-)
I didn't like this one so much and changed it into this:
@@ -67,8 +67,8 @@ static const char * const cma_events[] = {
[RDMA_CM_EVENT_TIMEWAIT_EXIT] = "timewait exit",
};
-static void cma_set_mgid(struct rdma_id_private *id_priv, struct sockaddr *addr,
- union ib_gid *mgid);
+static void cma_iboe_set_mgid(struct sockaddr *addr, union ib_gid *mgid,
+ enum ib_gid_type gid_type);
const char *__attribute_const__ rdma_event_msg(enum rdma_cm_event_type event) {
@@ -1846,17 +1846,19 @@ static void destroy_mc(struct rdma_id_private *id_priv,
if (dev_addr->bound_dev_if)
ndev = dev_get_by_index(dev_addr->net,
dev_addr->bound_dev_if);
- if (ndev) {
+ if (ndev && !send_only) {
+ enum ib_gid_type gid_type;
union ib_gid mgid;- cma_set_mgid(id_priv, (struct sockaddr *)&mc->addr,
- &mgid);
-
- if (!send_only)
- cma_igmp_send(ndev, &mgid, false);
-
- dev_put(ndev);
+ gid_type = id_priv->cma_dev->default_gid_type
+ [id_priv->id.port_num -
+ rdma_start_port(
+ id_priv->cma_dev->device)];
+ cma_iboe_set_mgid((struct sockaddr *)&mc->addr, &mgid,
+ gid_type);
+ cma_igmp_send(ndev, &mgid, false);
}+ dev_put(ndev);
cancel_work_sync(&mc->iboe_join.work);
}Thanks,
Jason
Quoted 7 lines · Leon Romanovsky
From: Leon Romanovsky <[email protected]>
Hi,
This is compilation of unrelated fixes, one is an outcome of syzkaller report and other two were found in our regression.
Quoted 7 lines
Thanks
Leon Romanovsky (1):
RDMA/ucma: Protect mc during concurrent multicast leavesMaor Gottlieb (2):
RDMA/cma: Use correct address when leaving multicast group
Applied to for-rc
Quoted 1 line
RDMA/core: Set MR type in
ib_reg_user_mr
Applied to for-next
Thanks,
Jason