memory leak using udp

[riweirui]

When using udp protocol to read and write data, nfs-ganesha received KILL signal
View dmsge, the memory usage is too high and KILL

version V3.5

kernel
3.10.0-957.el7.x86_64

The following information can be found through the valgrind
16,568,600 (142,080 direct, 16,426,520 indirect) bytes in 1,110 blocks are definitely lost in loss record 2,128 of 2,129
==12775== at 0x4C2C089: calloc (vg_replace_malloc.c:762)
==12775== by 0x4F7A602: nfs3_readdirplus (nfs3_readdirplus.c:354)
==12775== by 0x4E902F5: nfs_rpc_process_request (nfs_worker_thread.c:1509)
==12775== by 0x4E9064F: nfs_rpc_valid_NFS (nfs_worker_thread.c:1636)
==12775== by 0x4FD6F2D: svc_dg_decode (svc_dg.c:362)
==12775== by 0x4FDAFC9: svc_request (svc_rqst.c:1202)
==12775== by 0x4FD6DED: svc_dg_recv (svc_dg.c:329)
==12775== by 0x4E91218: nfs_rpc_dispatch_udp_NFS (nfs_rpc_dispatcher_thread.c:279)
==12775== by 0x4FD6DC5: svc_dg_rendezvous (svc_dg.c:318)
==12775== by 0x4FDAF4A: svc_rqst_xprt_task_recv (svc_rqst.c:1183)
==12775== by 0x4FDBADE: svc_rqst_epoll_loop (svc_rqst.c:1564)
==12775== by 0x4FE5875: work_pool_thread (work_pool.c:183)

This problem appears to be caused by an update udp readdir bug
nfs-ganesha/nfs-ganesha#718

I think I found the issue,
The issue is at xdr_mem.c
The code
if (!(--uio->uio_references)) {
if (uio->uio_release) {
uio->uio_release(uio, UIO_FLAG_NONE);
} else {
__warnx(TIRPC_DEBUG_FLAG_ERROR,
"%s() memory leak, unexpected or no release flags (%u)\n",
func, uio->uio_flags);
abort();
}
}

my modification suggestion

if (uio->uio_references == 0) {
uio->uio_references = 1;
}
//fixbug memory leak
//if (!(--uio->uio_references)) {
if (uio->uio_release) {
uio->uio_release(uio, UIO_FLAG_NONE);
} else {
__warnx(TIRPC_DEBUG_FLAG_ERROR,
"%s() memory leak, unexpected or no release flags (%u)\n",
func, uio->uio_flags);
abort();
}
//}

After the modification, the bug of memory leak was solved, and no new problems were found in the test for a week

[chengmao2010]

I have the same problem on V3.5，any suggestions is welcome

[ffilz]

There is a pull request: #271

[ffilz]

Could you verify this still exists with the current code? It may have been resolved with some of the other changes that have gone in.