IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
I have recently been getting the above message on fc11 and
I have traced it down to a bug in util_lookup_group.
As of 145 util_lookup_group reads:
gid_t util_lookup_group(struct udev *udev, const char *group)
{
char *endptr;
int buflen = sysconf(_SC_GETGR_R_SIZE_MAX);
char buf[buflen];
struct group grbuf;
struct group *gr;
gid_t gid = 0;
if (strcmp(group, "root") == 0)
return 0;
gid = strtoul(group, &endptr, 10);
if (endptr[0] == '\0')
return gid;
errno = 0;
getgrnam_r(group, &grbuf, buf, buflen, &gr);
if (gr != NULL)
return gr->gr_gid;
if (errno == 0 || errno == ENOENT || errno == ESRCH)
err(udev, "specified group '%s' unknown\n", group);
else
err(udev, "error resolving group '%s': %m\n", group);
return 0;
}
The errno value from getgrnam_r here is ERANGE which is documented as
"Insufficient buffer space supplied".
When I call get getgrnam_r with a large enough buffer everything
works. Indicating that the problem is that sysconf is returning
a value too small.
A quick google search tells me that sysconf(_S_GETGR_R_SIZE_MAX)
is documented as:
> sysconf(_S_GETGR_R_SIZE_MAX) returns either -1 or a good
> suggested starting value for buflen. It does not return the
> worst case possible for buflen.
In my case I have a group with about 50 users in /etc/group
and that is what triggered the problem in udev and caused
all of the udevs group lookups to fail.
The following patch which dynamically allocates the group member buffer
should fix this problem.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Generally ALSA control devices should be the last ones to be processed
for ACL changes and similar operations because they can then be used as
indicators that ACL management finished for all device nodes of a
specific card.
This patch simple moves each controlC device behind all the pcmC devices
(and similar).
We must never access random devices in /dev which do not belong to
the event we are handling. Hard-coding /dev/hidrawX, and looping over all
devices is absolutely not acceptable --> hook into hidraw events.
We can not relay on (rather random) properties merged into the parent
device by earlier rules --> use libudev to find the sibling device
with a matching interface.
Libusb does not fit into udev's use case. We never want want to scan
and open() all usb devices in the system, just to find the device
we are already handling the event for --> put all the stupid scanning
into a single function and prepare for a fixed libusb or drop it later.
The keymap table has some holes in it, which caused the interactive mode to
crash for unknown keys. In these cases, print the numeric key code instead.
What's odd is that this is a huawei modem, not an option modem, so one would
expect it to work better with usb_modeswitch and it's -H (huawei) mode - but
that's not the case, I've tested that as well.
https://launchpad.net/bugs/401655
Remove key map files which have only one override. Instead, use keymap tools'
new feature of specifying scancode/keyname pairs directly at the command line.
Also add a comment to 95-keymap.rules about how to specify key mappings in the
rules.
We may need to handle SIGCHLD before the queued worker message. The last
reference, from the SIGCHLD or the worker message will clean up the worker
context. In case we receive an unexpected SIGCHLD with an error, we let
the event fail and clean up the worker context.
Persistent network rules write out new rules files. When rules change,
we need to kill all workers to update the in-memory copy of the rules.
We need to make sure, that a worker finshes its work for all device
messages it has accepted, before it exits after a SIGTERM from the main
process.
On machines with many thousands of devices:
$ time find /sys -name uevent | wc -l
74876
real 0m33.171s
user 0m3.329s
sys 0m29.719s
the current udevtrigger spends minutes sorting the device list:
$ time /sbin/udevadm trigger --dry-run
real 4m56.739s
user 4m45.743s
sys 0m7.862s
with qsort() it looks better:
$ time udev/udevadm trigger --dry-run
real 0m6.495s
user 0m0.473s
sys 0m5.923s
The callers of prepend_vendor_model both expect < 0 to be returned on
error and the index to be returned otherwise. However
prepend_vendor_model actually returns 1 on error. Fix this by correctly
returning -1.
Older kernels (before e5b3cd42: "SCSI: sanitize INQUIRY strings")
truncated the model field in sysfs (or propagated bad results from the
target) to less than the expected/required 16 characters which meant
that the SCSI id was mangled into:
# /sbin/scsi_id -g -s /block/sdg
S146cee20VIRTUAL-DISK
when it should have been:
# /sbin/scsi_id -g -s /block/sdg
SIET VIRTUAL-DISK 146cee20
Notice how the serial number has been pasted over the vendor+model at
index 1 instead of being added at the end.
In the former case:
# cat /sys/devices/platform/host5/session1/target5:0:0/5:0:0:1/model | od -t c -t x1
0000000 V I R T U A L - D I S K \n
56 49 52 54 55 41 4c 2d 44 49 53 4b 0a
But it should have been:
# cat /sys/devices/platform/host5/session1/target5:0:0/5:0:0:1/model | od -t c -t x1
0000000 V I R T U A L - D I S K
56 49 52 54 55 41 4c 2d 44 49 53 4b 20 20 20 20
0000020 \n
0a
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
