Ah, checked the versions sources.

This was fixed in 4.10, but not fixed in 4.9.

4.9 source code: source3/libsmb/clifile.c

---------------------------------------------
truct tevent_req *cli_notify_send(TALLOC_CTX *mem_ctx,
                                   struct tevent_context *ev,
                                   struct cli_state *cli, uint16_t fnum,
                                   uint32_t buffer_size,
                                   uint32_t completion_filter, bool recursive)
{
        struct tevent_req *req, *subreq;
        struct cli_notify_state *state;
        unsigned old_timeout;

        req = tevent_req_create(mem_ctx, &state, struct cli_notify_state);
        if (req == NULL) {
                return NULL;
        }

        SIVAL(state->setup, 0, completion_filter);
        SSVAL(state->setup, 4, fnum);
        SSVAL(state->setup, 6, recursive);

        /*
         * Notifies should not time out
         */
        old_timeout = cli_set_timeout(cli, 0);

        subreq = cli_trans_send(
                state,                  /* mem ctx. */
                ev,                     /* event ctx. */
                cli,                    /* cli_state. */
                0,                      /* additional_flags2 */
                SMBnttrans,             /* cmd. */
                NULL,                   /* pipe name. */
                -1,                     /* fid. */
                NT_TRANSACT_NOTIFY_CHANGE, /* function. */
                0,                      /* flags. */
                (uint16_t *)state->setup, /* setup. */
                4,                      /* num setup uint16_t words. */
                0,                      /* max returned setup. */
                NULL,                   /* param. */
                0,                      /* num param. */
                buffer_size,            /* max returned param. */
                NULL,                   /* data. */
                0,                      /* num data. */
                0);                     /* max returned data. */
------------------------------------------------------------

4.10 source code:

struct tevent_req *cli_notify_send(TALLOC_CTX *mem_ctx,
                                   struct tevent_context *ev,
                                   struct cli_state *cli, uint16_t fnum,
                                   uint32_t buffer_size,
                                   uint32_t completion_filter, bool recursive)
{
        struct tevent_req *req;
        struct cli_notify_state *state;
        unsigned old_timeout;

        req = tevent_req_create(mem_ctx, &state, struct cli_notify_state);
        if (req == NULL) {
                return NULL;
        }

        if (smbXcli_conn_protocol(cli->conn) >= PROTOCOL_SMB2_02) {
                /*
                 * Notifies should not time out
                 */
                old_timeout = cli_set_timeout(cli, 0);

                state->subreq = cli_smb2_notify_send(
                        state,
                        ev,
                        cli,
                        fnum,
                        buffer_size,
                        completion_filter,
                        recursive);
------------------------------------------------------------

Created attachment 15283 [details]
Patch that went into the 4.10.x code base

This is a git patchstream from 4.10 created by:

git format-patch --stdout 24b1aa9da2e53e0dba89811f6a8d14b810bd48a8..88d82b44c326cfb3ab1cf1f7288eada2566ab4bc

This may or may not apply cleanly to 4.9. Anyway it shows the fix that went into 4.10.

That's a pretty big changeset. Would something of this size usually be backported to the previous version (4.9.x)?

Given that Debian Buster (10) is shipping with 4.9.x, I'm wondering if this is a worthwhile backport? Apparently not many people are trying to use libsmbclient notify.

Unless there's strong demand from the Linux vendors, we'll probably not backport this. If you're OK with that you can close the bug with a "Fixed in 4.10".

It's unfortunate, but not a showstopper. Just another reason to move to 4.10.x I suppose :-)

(In reply to Jonathon Reinhart from comment #5)

Please report this one to Debian also, maybe, just maybe, we can get an approval to move buster up to 4.10. 

I already suggested that, but it was denied because of buster is in frozen state
 now. 
Just, if debian is going to use 4.9.5 in the end for buster, well, then i do expect a wave of problems on the samba list also. 

Debian is really to far again in its patches (again), and samba is moving fast. 

And, as said, this patch set is pretty big, and this is only one patch.

This was fixed in 4.10, no activity for over a year, so closing.