do_ioctl(): run lsof on EBUSYSometimes despite all the retries ioctl() on ploop device still returns
EBUSY, like this:
> Error in ploop_stop (ploop.c:1011): PLOOP_IOC_STOP: Device or resource busy
In such a case, let's provide lsof /dev/ploopXXX output to the user
so they might have an idea of who's using the device.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
do_ioctl(): change delaysThis is an addition to the previous commit.
Let's start with a delay of 0.01s, increase 2x until up to 0.5s.
Total wait time is still the same -- 60s.
Also, provide a comment telling why we need to do this.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
do_ioctl(): implement exponential wait for retryCommit cc4a397 was way too optimistic -- sometimes IOC_STOP_DEVICE
returns EBUSY for more than 2 seconds, so we need to increase it
back to 60 seconds.
Maxim Patlasov suggested to increase sleep intervals in an exponential
fashion. With this patch, sleep intervals between retries are doubled each
time. It starts with 0.015s and then increases to max of 15s per
interval. Here's a simulation:
s...
read_diskdescriptor(): removeNo one is using it, probably a leftover after conversion to new
reading/locking API.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_umount_image(): simplify unlock on returnThis is just a minor refactoring, instead of calling
ploop_unlock_dd() from multiple places, do "goto out"
that does it for us.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop init: add --nolazyAdd -n, --nolazy option to ploop init command, and document it
in ploop(8) man page.
This is a followup to commit 90ff5a7.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
do_ioctl(): don't hang for a minuteInstead of doing 60 retries with 1 second interval,
let's do 10 retries with 0.2 seconds interval so
the function will return in no longer than 2 seconds.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_get_top_delta_fname(): use ploop_lock(), return SYSEXIT_*A problem with ploop_get_top_delta_fname() is it can only be used
when a DiskDescriptor.xml is read into memory. For example, the
following code won't work:
struct ploop_disk_images_data *di;
char name[PATH_MAX];
ret = ploop_open_dd(&di, argv[0]);
if (ret)
return ret;
ret = ploop_get_top_delta_fname(di, name, sizeof(name));
if (ret)
return ret;
printf("top delta: %s", name);
Ther...
ploop_create_param: add NOLAZY flagAdd flags field to struct ploop_create_param.
Define PLOOP_CREATE_NOLAZY flag. If set, ext4 lazy inode table
initialization feature is disabled from mkfs.
From mke2fs(8) man page:
lazy_itable_init[= <0 to disable, 1 to enable>]
If enabled and the uninit_bg feature is enabled, the inode table
will not be fully initialized by mke2fs. This speeds up filesystem
initialization noticeably, b...
tools/Makefile: fix LDLIBSThe only library needed to directly link tools to is ploop.
This is to avoid overlinking:
https://wiki.openmandriva.org/en/Overlinking_issues_in_packaging
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Makefiles: move -lpthread to the proper placeLDFLAGS should not be populated with libs, as otherwise the order
of options in ld command line will be wrong.
Also, we don't need to build tools with -lpthread.
Fix both issues by moving -lpthread to LDLIBS in lib/
Add pkgconfig fileThis makes it possible to use pkg-config when building stuff that
requires ploop library. For example:
# pkg-config --libs --static ploop
-lploop -lrt -lpthread -lxml2 -lz -llzma -lm
Note it shows libs that are not explicitly listed, but are
needed to statically link with libxml2 (they are of course
taken from libxml-2.0.pc).
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop.spec: no %attr for symlinkRPM from Fedora 22 rightfully complains that %attr should not be used
with symlinks. Fix by removing the %attr, as it is redundant anyway.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop.spec: fix date in %changelogRPM from Fedora 22 rightfully complains the date is bogus.
Indeed it was Tuesday not Thursday.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop_create_image(): do not return -1Found another case when ploop library returns -1 instead of proper
SYSEXIT_* code. Fix is easy.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
is_xml_fname(): fix a potential SIGSEGVIn case argument is NULL, it's definitely not an XML file
so return 0 instead of crashing:
Program received signal SIGSEGV, Segmentation fault.
0x00007ffff757ac5a in __strrchr_sse2 () from /lib64/libc.so.6
Currently it appears to be no calls to is_xml_fname(NULL)
but there was one -- see previous commit for details.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
ploop snapshot-switch: fix SIGSEGV parsing cmdlineThe check for arguments validity is obviously incorrect, leading to
calling is_xml_fname() with NULL argument, leading to SIGSEGV:
# ploop snapshot-switch
Segmentation fault
Another bug is, in case there is a proper number of arguments function
is_xml_fname() is not called at all.
Interestingly, the bug is here from the very beginning,
introduced by commit r733091 (Fri Feb 3 10:51:26 2012)...
include/libploop.h: doc mount and snapshot switch paramsDocument the fields and flags of
* struct ploop_mount_param
* struct ploop_snapshot_switch_param
Also, move defines for ploop_snapshot_switch_param.flags from the
very beginning of the file closer to struct ploop_snapshot_switch_param.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
print_sys_block_ploop(): more compact outputSimplify the /sys/block/ploop printing, also making the output
more compact (and, I hope, more readable as well).
This is the way it was:
==> /sys/block/ploop14553/pdelta/0/image <==
/vz/private/206/root.hdd/root.hdd
==> /sys/block/ploop14553/pdelta/0/io <==
direct
==> /sys/block/ploop14553/pdelta/0/ro <==
0
This is how it is now, after this patch:
/sys/block/ploop14553/pdelta/0/imag...
ploop_replace_image: use mkstemp() not mktemp()This is more secure, and in Fedora 22 linker prints an annoying warning
about using mktemp():
libploop.a(ploop.o): In function `ploop_replace_image':
warning: the use of `mktemp' is dangerous, better use `mkstemp'
Not tested much.
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
mount.ploop: check if mtab is not symlinkApparently, test -f $FILE returns true even if $FILE is a symlink.
That, combined with the fact that when /etc/mtab is a symlink to
/proc/mounts, gives us the following error:
# mount DiskDescriptor.xml mnt -t ploop
...
/sbin/mount.ploop: line 201: echo: write error: Invalid argument
To fix, add a check that /etc/mtab is not a symlink
Signed-off-by: Kir Kolyshkin <kir@openvz.org>
Fix POWER compilation wrt __u64 printf/scanfOn POWER, __u64 is defined in the Linux kernel headers as
"long unsinged", while on x86_64 it is defined as "long long unsigned".
As a result, we have no way to uniformly printf/scanf __u64 values
without type casting.
The closest userspace type equivalent of __u64 is uint64_t
(as defined by ISO C99: 7.18 Integer types <stdint.h>),
and this is what we should cast it to. Unfortunately, uint64_t...
ploop_set_max_delta_size(): fix definitionOn POWER, __u64 is defined as "unsigned long", while on x86_64
it is "unsigned long long", thus the following error:
> ploop.c:3431:5: error: conflicting types for ‘ploop_set_max_delta_size’
> int ploop_set_max_delta_size(struct ploop_disk_images_data *di, __u64 size)
> ^
> In file included from ploop.h:13:0,
> from ploop.c:39:
> ../include/libploop.h:260:5: note: previo...
get_max_ploop_size(): change type of last argUse "unsigned long long" as a portable type, to fix the following
compilation error on POWER (caused by the fact that __u64 is defined
as "unsigned long" on POWER, but "unsigned long long" on x86_64).
> ploop.c: In function ‘ploop_get_max_size’:
> ploop.c:305:2: error: passing argument 3 of ‘get_max_ploop_size’ from incompatible pointer type [-Werror]
> if (get_max_ploop_size(default_fmt_ver...
Kir Kolyshkin