ref: 4b60d3dd32efd00a1d07cb08662926ba9836719f
parent: 2569cbf9bd35798942a1640bf919162efe92d3b8
author: Noam Preil <noam@pixelhero.dev>
date: Sat Dec 28 23:57:05 EST 2024
server: increase connection stack size from 256K to 512K for vtwrite
--- a/notebook
+++ b/notebook
@@ -2512,4 +2512,100 @@
<stdin>:2: (error) setproc: open /proc/484718/mem: file does not exist: '/proc/484718'
-uh.
\ No newline at end of file
+uh.
+
+venti/read -h tcp!127.1!14011 dc294a9b60bc62b368036181784c7a42688c05c8 | venti/write -h tcp!127.1!14011
+
+write successful, memcpying
+vtsending
+panic: D2B called on non-block 10131d720 (double-free?)
+
+Okay, vtsend is doing a Bad. Hmm.
+
+write successful, memcpying
+vtsending
+sending, marking size...
+settting type
+write syscall... written!
+panic: D2B called on non-block 10131d770 (double-free?)
+
+No, actually, vtsend is working just fine. What's going on??
+
+The request is actually succeeding, we're just crashing _afterwards_?
+
+% venti/read -h tcp!127.1!14011 dc294a9b60bc62b368036181784c7a42688c05c8 | venti/write -h tcp!127.1!14011
+dc294a9b60bc62b368036181784c7a42688c05c8
+
+yep!
+
+Request handled!panic: D2B called on non-block 10131d780 (double-free?)
+
+Mmmm, a bug in the partial packet handling, maybe?
+
+acid: stacks()
+p=(Proc)0x3c588 pid 27808 Running
+ t=(Thread)0x3c6e8 Running /usr/glenda/src/disk/neoventi/server.c:251 serve [threadmain]
+ open()+0xc /sys/src/libc/9syscall/open.s:6
+ listen(dir=0x5c710,newdir=0x5c6e8)+0x34 /sys/src/libc/9sys/announce.c:93
+ buf=0x7063742f74656e2f
+ ctl=0x104c000000007
+ n=0x700000009
+ serve()+0x64 /usr/glenda/src/disk/neoventi/server.c:251
+ adir=0x7063742f74656e2f
+ dir=0x7063742f74656e2f
+ ctl=0x7
+ threadmain(argv=0x1fffeff70)+0x34 /usr/glenda/src/disk/neoventi/neoventi.c:118
+ mainlauncher()+0x10 /sys/src/libthread/main.c:17
+ launcherarm64(arg=0x1fffeff70,f=0x1a98c)+0x8 /sys/src/libthread/arm64.c:8
+ launcherarm64(arg=0x1fffeff70,f=0x1a98c)+0xfffffffffffffff8 /sys/src/libthread/channel.c:583
+ listen+0x34 /sys/src/libc/9sys/announce.c:93
+
+
+p=(Proc)0xde9a8 pid 27838 Sched
+
+hm yeah going to need prints here.
+
+write syscall... written!
+Request handled!
+reading packet...
+packet read
+panic: D2B called on non-block 10135d808 (double-free?)
+
+Oh interesting, packet read, n 0, sz 0
+
+packet read, n 0, sz 0
+hanging up
+panic: D2B called on non-block 10135d838 (double-free?)
+
+So now we're just crashing when hanging up. Lovely.
+
+venti/read on its own works fine and doesn't cause this, though.
+
+hmm
+
+poolfreel(v=0x10131d6f0,p=0x352d0)+0x18 /sys/src/libc/port/pool.c:1152
+ ab=0x30150
+poolfree(p=0x352d0,v=0x10131d6f0)+0x34 /sys/src/libc/port/pool.c:1287
+free()+0x18 /sys/src/libc/port/malloc.c:250
+_schedinit()+0xd0 /sys/src/libthread/sched.c:69
+ p=0xde6d8
+ t=0xde838
+
+ free(t->stk);
+ free(t->cmdname);
+
+That's freeing either the stack or the command name, depending on whether it's accurate.
+
+Stack corruption, maybe? Adding in the vtwrite() buffer for data comparison is probably overflowing.
+
+yep, that was it.
+
+% git/diff server.c
+- proccreate(handleproc, (void*)fd, 256*1024);
++ proccreate(handleproc, (void*)fd, 512*1024);
+
+fixed :P
+
+also, TODO: remove libbio usage, it's probably overkill for what we actually need.
+
+Shorter term TODO: see about bumping buffer size to account for Bungetsz to make reads aligned so devfs patch can be reverted. (Performance improvement, since aligned reads are faster.)
--- a/server.c
+++ b/server.c
@@ -225,7 +225,7 @@
fprint(2, "failed to accept connection\n");
return;
}
- proccreate(handleproc, (void*)fd, 256*1024);
+ proccreate(handleproc, (void*)fd, 512*1024);
}
void