What would be the best way of capturing stderr from GCC? Not sure if I am understanding something incorrectly; the dev I am interacting with suggests capturing stderr from GCC by using dtrace. Has a better way been formulated since the 90s?

ok, I must be blind or something, but why this buffer needs to be 32 bytes?! src.illumos.org/source/xref/illumos…/fs/zfs/zfs_vfsops.c?r=4763305e#789

I can count 17:)

uh, that looks nasty

why does it not use asprintf or expect the actual buffer size as an argument?

asprintf wont work as they are skipping prefix bytes from that buffer ("obj-")

that is, the actual buffer is expected to be + 4 bytes larger

then the callers should just pass the actual size, and use snprintf in that function :)

sprintf is so 1985

At first I took that 32 literally and replaced char *buf by char buf[32], and gained compiler check on buffer size, then I started to look where this size is coming from:)

17? I count 20, or are you taking into account limits on domainid and rid?

well, 21 with the null byte

Oh, hex.. I'll get some coffee.

eee? we are putting those bytes into that buf: sprintf(buf, "%llx", (longlong_t)fuid); nothing more

:D

yes, its hex

so its 16+1

and yes, going to replace sprintf by snprintf there for safety

That function takes a 'len' parameter in openzfs, as well as using snprintf (commit c9e319faae)

tsoome: we should use ilstr for that, tbh

src.illumos.org/source/xref/illumos…s/zfs_vfsops.c?r=4763305e#1576-1587 -- this is not great, for example

yes, that one was blowing on my face with char buf[32] like this: ../../common/fs/zfs/zfs_vfsops.c:849:15: error: 'id_to_fuidstr' accessing 32 bytes in a region of size 24 [-Werror=stringop-overflow=]

oof

You can still use a stack buffer with ilstr_init_prealloc() at least, it just won't be as egregiously unsafe haha

and then I started to check 1) why 32, and 2) what happens with the content there

I guess ilstr may be good answer there, as zfs_vfsops.c is os specific anyhow [in OpenZFS] and therefore we will not suffer the portability.

ofc that 32 must be leftover from some older version;)

ah, ok, quota bits are moved to zfs_quota.c ok....

[illumos-gate] 16997 zfs.h correct comments in zfs_ioc -- Toomas Soome <tsoome⊙mc>

I might have misread above, but I don't see reasons to avoid ilstr for porting purposes

we have an interface that makes our strings safer, that's good, and we should use it

richlowe: Yes, agreed.

If folks want to take patches we've written against ZFS and apply them to other projects, they're welcome to them under the same licence etc

(and thus also welcome to make a copy of ilstr etc)