rds: prevent BUG_ON triggering on congestion map updates
Neil Horman [Wed, 2 Mar 2011 06:28:22 +0000 (06:28 +0000)]
Recently had this bug halt reported to me:

kernel BUG at net/rds/send.c:329!
Oops: Exception in kernel mode, sig: 5 [#1]
SMP NR_CPUS=1024 NUMA pSeries
Modules linked in: rds sunrpc ipv6 dm_mirror dm_region_hash dm_log ibmveth sg
ext4 jbd2 mbcache sd_mod crc_t10dif ibmvscsic scsi_transport_srp scsi_tgt
dm_mod [last unloaded: scsi_wait_scan]
NIP: d000000003ca68f4 LR: d000000003ca67fc CTR: d000000003ca8770
REGS: c000000175cab980 TRAP: 0700   Not tainted  (2.6.32-118.el6.ppc64)
MSR: 8000000000029032 <EE,ME,CE,IR,DR>  CR: 44000022  XER: 00000000
TASK = c00000017586ec90[1896] 'krdsd' THREAD: c000000175ca8000 CPU: 0
GPR00: 0000000000000150 c000000175cabc00 d000000003cb7340 0000000000002030
GPR04: ffffffffffffffff 0000000000000030 0000000000000000 0000000000000030
GPR08: 0000000000000001 0000000000000001 c0000001756b1e30 0000000000010000
GPR12: d000000003caac90 c000000000fa2500 c0000001742b2858 c0000001742b2a00
GPR16: c0000001742b2a08 c0000001742b2820 0000000000000001 0000000000000001
GPR20: 0000000000000040 c0000001742b2814 c000000175cabc70 0800000000000000
GPR24: 0000000000000004 0200000000000000 0000000000000000 c0000001742b2860
GPR28: 0000000000000000 c0000001756b1c80 d000000003cb68e8 c0000001742b27b8
NIP [d000000003ca68f4] .rds_send_xmit+0x4c4/0x8a0 [rds]
LR [d000000003ca67fc] .rds_send_xmit+0x3cc/0x8a0 [rds]
Call Trace:
[c000000175cabc00] [d000000003ca67fc] .rds_send_xmit+0x3cc/0x8a0 [rds]
(unreliable)
[c000000175cabd30] [d000000003ca7e64] .rds_send_worker+0x54/0x100 [rds]
[c000000175cabdb0] [c0000000000b475c] .worker_thread+0x1dc/0x3c0
[c000000175cabed0] [c0000000000baa9c] .kthread+0xbc/0xd0
[c000000175cabf90] [c000000000032114] .kernel_thread+0x54/0x70
Instruction dump:
4bfffd50 60000000 60000000 39080001 935f004c f91f0040 41820024 813d017c
7d094a78 7d290074 7929d182 394a0020 <0b090000> 40e2ff68 4bffffa4 39200000
Kernel panic - not syncing: Fatal exception
Call Trace:
[c000000175cab560] [c000000000012e04] .show_stack+0x74/0x1c0 (unreliable)
[c000000175cab610] [c0000000005a365c] .panic+0x80/0x1b4
[c000000175cab6a0] [c00000000002fbcc] .die+0x21c/0x2a0
[c000000175cab750] [c000000000030000] ._exception+0x110/0x220
[c000000175cab910] [c000000000004b9c] program_check_common+0x11c/0x180

Signed-off-by: David S. Miller <davem@davemloft.net>

net/rds/ib_send.c
net/rds/loop.c

index 71f373c..c47a511 100644 (file)
@@ -551,7 +551,10 @@ int rds_ib_xmit(struct rds_connection *conn, struct rds_message *rm,
        if (conn->c_loopback
            && rm->m_inc.i_hdr.h_flags & RDS_FLAG_CONG_BITMAP) {
                rds_cong_map_updated(conn->c_fcong, ~(u64) 0);
-               return sizeof(struct rds_header) + RDS_CONG_MAP_BYTES;
+               scat = &rm->data.op_sg[sg];
+               ret = sizeof(struct rds_header) + RDS_CONG_MAP_BYTES;
+               ret = min_t(int, ret, scat->length - conn->c_xmit_data_off);
+               return ret;
        }
 
        /* FIXME we may overallocate here */
index aeec1d4..bca6761 100644 (file)
@@ -61,10 +61,15 @@ static int rds_loop_xmit(struct rds_connection *conn, struct rds_message *rm,
                         unsigned int hdr_off, unsigned int sg,
                         unsigned int off)
 {
+       struct scatterlist *sgp = &rm->data.op_sg[sg];
+       int ret = sizeof(struct rds_header) +
+                       be32_to_cpu(rm->m_inc.i_hdr.h_len);
+
        /* Do not send cong updates to loopback */
        if (rm->m_inc.i_hdr.h_flags & RDS_FLAG_CONG_BITMAP) {
                rds_cong_map_updated(conn->c_fcong, ~(u64) 0);
-               return sizeof(struct rds_header) + RDS_CONG_MAP_BYTES;
+               ret = min_t(int, ret, sgp->length - conn->c_xmit_data_off);
+               goto out;
        }
 
        BUG_ON(hdr_off || sg || off);
@@ -80,8 +85,8 @@ static int rds_loop_xmit(struct rds_connection *conn, struct rds_message *rm,
                            NULL);
 
        rds_inc_put(&rm->m_inc);
-
-       return sizeof(struct rds_header) + be32_to_cpu(rm->m_inc.i_hdr.h_len);
+out:
+       return ret;
 }
 
 /*