♟ trasz

Remove debugging printf.

sysutils/debootstrap: install mremap(2) apt(8) workaround

linux: mute "unsupported socket(AF_NETLINK, 3, NETLINK_AUDIT)" warnings

It might be useful to also install virtio man pages on all architectures (in share/man/man4/Makefile) then?

linux: mute the "unsupported prctl option 23" warnings

Yup. We're still missing statx(2), but the ones above will do for now.

linux: bump the default version from 3.10.0 to 3.17.0

sysutils/debootstrap: add workarounds necessary for bootstrapping

The current diff seems a bit incomplete - previous one seemed ok, something between Phab and Git, perhaps?

Looks good, thank you!

... And reverted in f53120073b3 - mostly because, as Jessica pointed out, it would break existing applications which properly share the context between threads, eg by using explicit locking.

Revert "bsnmpclient(3): make it thread-safe"

Drop unneeded changes.

I'm not a huge fan of this, to be honest. Not the implementation, but rather the idea.

I don't remember much about how directory entries work, but I don't see anything wrong with the patch. I have a question, though: how did it work before? Is that field actually used?

bsnmpclient(3): make it thread-safe

iscsid(8): fix memory leak by freeing the 'addr'

iscsid(8): free data allocated by getaddrinfo(3)

geom(4): make g_newprovider_event() return if G_P_WITHER is set

libcam(3): make cam_getccb(3) zero the whole ccb, not just the header

devfs(4): defer freeing until we drop devmtx ("cdev")

Which unlock? The gotos at 1048, 1057, 1062, 1076, 1080, 1085, and 1089 all get here without the lock held.

Also remove now useless variable.

Move myself from additional contributors to commiters.

I'm a src commiter now. Wow. I mean... wow. ;-)

Document removal of VSVTX, VSGID and VSUID flags.

Bump __FreeBSD_version after introduction of accmode_t.

Make sure to request IPv4 address.

Ah, right. What do you think about the new patch? I've tried avoiding the relocking, but the code flow there is somewhat complicated and I'd probably introduce more bugs.

Make sure the ip is valid.

Ah, never mind, there's https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=241385 which is much better.

Make "diskinfo -i" also test 1MB reads.

In D27337#613495, @kib wrote:

In D27337#610846, @trasz wrote:

In D27337#610604, @kib wrote:

This is not a proper fix. ip is invalid the very moment fvp is unlocked.

"ip" as the pointer (later used to compare it with "xp"), or vnode data it points to?

Well, is a pointer to freed memory valid ? This is somewhat scholastic question, and I believe that our kernel depends on pointer value itself not becoming a poisoned value after free.
But the data it points to is invalid, and your patch still accesses the potentially freed or reused memory.

In D27338#610605, @kib wrote:

Do you have repro ? Did you sent it to pho ?

In D27132#610070, @kib wrote:

Until you set ast flag for other threads, they can run arbitrary amount of code, including syscalls.

In D27337#610604, @kib wrote:

This is not a proper fix. ip is invalid the very moment fvp is unlocked.

Replace paths in ctladm(8) examples with something sensible.

Assorted tweaks to hier(7): punctuation, some .Xr's, clarify wording.

trasz (Edward Tomasz Napierała)
User

Projects

User Details

Recent Activity
View All

Yesterday

Sat, Jan 16

Fri, Jan 15

Thu, Jan 14

Wed, Jan 13

Mon, Jan 11

Mon, Jan 4

Tue, Dec 29

Mon, Dec 28

Sun, Dec 27

Thu, Dec 24

Wed, Dec 23

Dec 13 2020

Dec 12 2020

Dec 11 2020

Dec 10 2020

Dec 7 2020

Nov 30 2020

Nov 26 2020

Nov 25 2020

Nov 24 2020

Nov 23 2020

trasz (Edward Tomasz Napierała)User