Aren
/
NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

getdfwithoffset - do not issue dummy read to seek to offset zero on a 

newly opened fid as it's already there.  This is redundant and also
seems to confuse inferno when issued against the root directory.  Now
you can ls the mount point of

    styxlisten -A 'tcp!*!styx' export /

While here fix the seek logic to actually repeatedly seek forward
instead of always doing it from zero.
This commit is contained in:
uwe 2020-05-26 19:38:14 +00:00
parent b43a201fc6
commit 795a40f9c2
1 changed files with 37 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

73
usr.sbin/puffs/mount_9p/subr.c
View File

@ -1,4 +1,4 @@
/* $NetBSD: subr.c,v 1.7 2019/06/07 05:34:34 ozaki-r Exp $ */
/* $NetBSD: subr.c,v 1.8 2020/05/26 19:38:14 uwe Exp $ */
/*
* Copyright (c) 2007 Antti Kantee. All Rights Reserved.
@ -27,7 +27,7 @@
#include <sys/cdefs.h>
#ifndef lint
__RCSID("$NetBSD: subr.c,v 1.7 2019/06/07 05:34:34 ozaki-r Exp $");
__RCSID("$NetBSD: subr.c,v 1.8 2020/05/26 19:38:14 uwe Exp $");
#endif /* !lint */
#include <sys/types.h>
@ -104,11 +104,7 @@ getdfwithoffset(struct puffs_usermount *pu, struct p9pnode *p9n, off_t wantoff,
{
struct puffs_cc *pcc = puffs_cc_getcc(pu);
struct puffs9p *p9p = puffs_getspecific(pu);
struct puffs_framebuf *pb;
struct dirfid *dfp = NULL;
p9ptag_t tag = NEXTTAG(p9p);
off_t curoff, advance;
uint32_t count;
int rv;
LIST_FOREACH(dfp, &p9n->dir_openlist, entries) {
@ -120,52 +116,57 @@ getdfwithoffset(struct puffs_usermount *pu, struct p9pnode *p9n, off_t wantoff,
}
/* didn't get off easy? damn, do manual labour */
pb = p9pbuf_makeout();
dfp = ecalloc(1, sizeof(struct dirfid));
dfp->fid = NEXTFID(p9p);
rv = proto_cc_open(pu, p9n->fid_base, dfp->fid, P9PROTO_OMODE_READ);
if (rv)
goto out;
for (curoff = 0;;) {
advance = wantoff - curoff;
off_t curoff = 0;
if (wantoff != 0) {
struct puffs_framebuf *pb = p9pbuf_makeout();
for (;;) {
off_t advance = wantoff - curoff;
tag = NEXTTAG(p9p);
p9pbuf_put_1(pb, P9PROTO_T_READ);
p9pbuf_put_2(pb, tag);
p9pbuf_put_4(pb, dfp->fid);
p9pbuf_put_8(pb, 0);
p9pbuf_put_4(pb, advance);
GETRESPONSE(pb);
p9ptag_t tag = NEXTTAG(p9p);
p9pbuf_put_1(pb, P9PROTO_T_READ);
p9pbuf_put_2(pb, tag);
p9pbuf_put_4(pb, dfp->fid);
p9pbuf_put_8(pb, curoff);
p9pbuf_put_4(pb, advance);
GETRESPONSE(pb);
if (p9pbuf_get_type(pb) != P9PROTO_R_READ) {
rv = proto_handle_rerror(pu, pb);
goto out;
if (p9pbuf_get_type(pb) != P9PROTO_R_READ) {
rv = proto_handle_rerror(pu, pb);
puffs_framebuf_destroy(pb);
goto out;
}
/*
* Check how many bytes we got. If we got the
* amount we wanted, we are at the correct position.
* If we got zero bytes, either the directory
* doesn't "support" the seek offset we want
* (someone has probably inserted an entry
* meantime) or we at the end of directory.
* Either way, let the upper layer deal with it.
*/
uint32_t count;
p9pbuf_get_4(pb, &count);
curoff += count;
if (count == advance || count == 0)
break;
p9pbuf_recycleout(pb);
}
/*
* Check how many bytes we got. If we got the amount we
* wanted, we are at the correct position. If we got
* zero bytes, either the directory doesn't "support" the
* seek offset we want (someone has probably inserted an
* entry meantime) or we at the end of directory. Either
* way, let the upper layer deal with it.
*/
p9pbuf_get_4(pb, &count);
curoff += count;
if (count == advance || count == 0)
break;
p9pbuf_recycleout(pb);
puffs_framebuf_destroy(pb);
}
puffs_framebuf_destroy(pb);
dfp->seekoff = curoff;
*rfid = dfp;
return 0;
out:
puffs_framebuf_destroy(pb);
free(dfp);
return rv;
}