| DUPB(9F) | Kernel Functions for Drivers | DUPB(9F) |
dupb - duplicate a message block descriptor
#include <sys/stream.h> mblk_t *dupb(mblk_t *bp);
Architecture independent level 1 (DDI/DKI).
dupb() creates a new mblk_t structure (see msgb(9S)) to reference the message block pointed to by bp.
Unlike copyb(9F), dupb() does not copy the information in the dblk_t structure (see datab(9S)), but creates a new mblk_t structure to point to it. The reference count in the dblk_t structure (db_ref) is incremented. The new mblk_t structure contains the same information as the original. Note that b_rptr and b_wptr are copied from the bp.
Printed copy or docs.sun.com shows a figure that shows a new mblk_t structure created, with the original and new bp both pointing to the dblk_t structure, and db_ref incremented by one
bp
If successful, dupb() returns a pointer to the new message block. A NULL pointer is returned if dupb() cannot allocate a new message block descriptor or if the db_ref field of the data block structure (see datab(9S)) has reached a maximum value (255).
dupb() can be called from user, kernel, or interrupt context.
Example 1 Using dupb()
This srv(9E) (service) routine adds a header to all M_DATA messages before passing them along. dupb is used instead of copyb(9F) because the contents of the header block are not changed.
For each message on the queue, if it is a priority message, pass it along immediately (lines 10-11). Otherwise, if it is anything other than an M_DATA message (line 12), and if it can be sent along (line 13), then do so (line 14). Otherwise, put the message back on the queue and return (lines 16-17). For all M_DATA messages, first check to see if the stream is flow-controlled (line 20). If it is, put the message back on the queue and return (lines 37-38). If it is not, the header block is duplicated (line 21).
dupb() can fail either due to lack of resources or because the message block has already been duplicated 255 times. In order to handle the latter case, the example calls copyb(9F) (line 22). If copyb(9F) fails, it is due to buffer allocation failure. In this case, qbufcall(9F) is used to initiate a callback (lines 30-31) if one is not already pending (lines 26-27).
The callback function, xxxcallback(), clears the recorded qbufcall(9F) callback id and schedules the service procedure (lines 49-50). Note that the close routine, xxxclose(), must cancel any outstanding qbufcall(9F) callback requests (lines 58-59).
If dupb() or copyb(9F) succeed, link the M_DATA message to the new message block (line 34) and pass it along (line 35).
1 xxxsrv(q)
2 queue_t *q;
3 {
4 struct xx *xx = (struct xx *)q->q_ptr;
5 mblk_t *mp;
6 mblk_t *bp;
7 extern mblk_t *hdr;
8
9 while ((mp = getq(q)) != NULL) {
10 if (mp->b_datap->db_type >= QPCTL) {
11 putnext(q, mp);
12 } else if (mp->b_datap->db_type != M_DATA) {
13 if (canputnext(q))
14 putnext(q, mp);
15 else {
16 putbq(q, mp);
17 return;
18 }
19 } else { /* M_DATA */
20 if (canputnext(q)) {
21 if ((bp = dupb(hdr)) == NULL)
22 bp = copyb(hdr);
23 if (bp == NULL) {
24 size_t size = msgdsize(mp);
25 putbq(q, mp);
26 if (xx->xx_qbufcall_id) {
27 /* qbufcall pending */
28 return;
29 }
30 xx->xx_qbufcall_id = qbufcall(q, size,
31 BPRI_MED, xxxcallback, (intptr_t)q);
32 return;
33 }
34 linkb(bp, mp);
35 putnext(q, bp);
36 } else {
37 putbq(q, mp);
38 return;
39 }
40 }
41 }
42 }
43 void
44 xxxcallback(q)
45 queue_t *q;
46 {
47 struct xx *xx = (struct xx *)q->q_ptr;
48
49 xx->xx_qbufcall_id = 0;
50 qenable(q);
51 }
52 xxxclose(q, cflag, crp)
53 queue_t *q;
54 int cflag;
55 cred_t *crp;
56 {
57 struct xx *xx = (struct xx *)q->q_ptr;
...
58 if (xx->xx_qbufcall_id)
59 qunbufcall(q, xx->xx_qbufcall_id);
...
60 }
srv(9E), copyb(9F), qbufcall(9F), datab(9S), msgb(9S)
Writing Device Drivers STREAMS Programming Guide
| June 20, 2021 | OmniOS |