Currently we dump pages directly from parasite into image files. This
is bad for several reasons:
1. We cannot use any more-or-less custom format for pages easily, since
parasite code cannot be linked with any libraries;
2. We will not be able to optimize migration with preliminary memory
migration (a.k.a. iterative migration) with it -- if we send pages
from parasite over network we are not able to let the task we dump
continue running.
That said, what is done is -- pages from target task are put into a
page-pipe in one go, then (not in this patch) parasite can be released
and we can do with pages whatever we want. For now pages are just
spliced from pipe into image file.
Some numbers:
In order to drain 1Gb of memory from task we need 1.5M of shared map
in args (for iovecs) and 4 pipes (8 descriptors) each referencing 128Mb
of pages, which int turn requires 4 x 640K chunks of sequential kernel
memory (for pipe_buffer). Not that big I guess.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Just make use of previous patch. The creds dumping args are tuned to
fit one page (minimal static args size).
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Sometimes we don't know the exact amount of data we would want
to send to parasite via args area (e.g. -- while draining fds).
Fix this, by moving the args area behind the parasite blob and
mmap-ing it with the run-time calculated size.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
No need to walk up the directories if we need
to include protobuf file. This was always a bad
use of ability to walk the filesystem from other
headers.
Same time we don't need -I$(SRC_DIR)/protobuf/
in general makefile anymore.
[xemul: Small fixlet in head Makefile, since patch
it out-of-order]
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Make them look like __CR_<smth>_H__ with
sed -e '1,2s/#\(ifndef\|define\) _\?_\?\(CR_\)\?/#\1 __CR_/' -e '1,2s/_H_\?_\?.*$/_H__/'
on every header file.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
I've moved dump_thread helper a bit lower in file
since I've to call for find_thread_state helper.
After all this groups all thread related functions
in one slab.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Just to be consistent with types we're using.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
We will need to extend the structure to keep not only tid/tid_addr
but blocked signals as well, thus rename it to more generic
parasite_dump_thread.
The command PARASITE_CMD_DUMP_TID_ADDR renamed to
PARASITE_CMD_DUMP_THREAD for the same sake.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
These helpers will be needed to save a blocked signals
mask for dumpee threads.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
The per-thread information requires own space in parasite data.
In particular we will keep the blocked signals bound to thread
pids.
For this sake the caller need to provide the parasite how many
threads will be used to calculate space.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Will need it to fetch tty link parameters. This is
because the kernel provides SID/PGID related to the
caller context, and if we're dumping the process inside
namespace -- we need local SID/PGIDs, not global ones.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Dumping them is performed via parasite, since calling the getgroups
is the only way of getting the complete list. Currently the nr of
groups to dump is limited explicitly with the size of shared memory
between crtools and parasite. This is MUCH more that we have seen
on real apps so far.
Restoring is done early, before restorer blob not to carry the undefined
array of grpous in there. This is OK, since groups do not affect us at
that point and are not affected by subsequent creds restore.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Currently move there the secbits dumping, which is not dumped
via misc-dumping command. This patch is required to support
per-task groups dumping (setgroups/getgroups) -- we'll have to
drain the groups from parasite.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
In commit 97cfb707 we extended scm_fdset structure
up to 7K thus we need to increase the parasite stack
size enough to hold it. It was not noticed simply because
we were poking arguments area of memory.
The memory map of args and stack for parasite look like
+-----------+ min address
| | |
| arguments | v (8K)
| |
+-----------+
| |
| stack | ^ (16K)
| | |
+-----------+ max address
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
It will be used for dumping zombies.
v2: Add a fast path, if /proc belongs to the target pipns. it's usually for CT.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
* Don't dumplicate array when sending to parasite
* Allocate necessary space in cr-dump for local fds/flags
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
v2:
- Use regular uint types in message proto
- Use PB engine for "show"
v3:
- drop usage of temp. variable in prepare_shmem_pid
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Same as with previous patch -- we have to pull image-independent entry out of the victim
task to be able to use protobuf library for encoding.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Pull the array of rt_sigact_t-s to the crtools instead and fill+write sa_entries there.
This is required to switch to protobuf, since we cannot use pb lib code in parasite for
encoding sa_entries.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Before this patch a transport socket was created on-demand for each operation
requiring it. It will be problematic, when we start supporting net namespaces.
That said -- create a transport sockets and interconnect them early.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Acked-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
In olden times, when people didn't know how to write,
they sent a courier with a letter.
Now we can write messages from parasite, so SET_PARASITE_RET may be dropped.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Looks good to me as well. Maybe we will refine messages to be able
to identify where exactly parasite failed (say line). But it better
to do in another patch.
Acked-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
because now it cares about tid_addr and thread id.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
If we try to dump a process from another pid ns,
we have not another way to get its pid.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
The task is not complete - this is just a part of what have to be done. I.e.
looks like a lot of excessive deps can be fixed.
Signed-off-by: Stanislav Kinsbursky <skinsbursky@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Dumped pages are those taken to the image
Skipped pages are those, sitting in RAM, but left there (file shared pages)
Total is the total amount of pages that fit into vmarea
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Inline assembly is very convenient if a couple of
instructions is used but when it grows better to
move it out of wrapper C code and write in plain
assembly, after all we need a very precise control
in bootstrap code.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Instead of generating offsets from early compiled
object files (one day the offsets obtained from
there might be changed during linkage stage) better
to get them from a final stage where all object
files involved are linked into complete binary blob.
That happened that at early stage we indeed were using
only single file per parasite and restorer but at present
there a couple of file involved (and will be more in
future) so we need a safe approach.
Also note the symbols being exported are prefixed as
"__export_". This is easier approach for now. Putting
such symbols into separate section requires a way
more efforts to handle.
The main reason of having two files (Elf object
and binary blob) is to get 1:1 mapping between
symbols definition and their position in binary
target.
The exported symbols name addresses are obtained
from object file and used as offsets in binary
target.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
There is no need for specific parasite or restorer
head sections, a general "head" one is enough.
Also .stack section is dropped since it's never used.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
It has been used when debuggin without PTRACE_SEIZE,
not needed anymore.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Since now we have local copies of a remove FDs we can dump socket queus without entering
a parasite code. This makes the code MUCH simpler.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Just implemented but not yet used in dumping procedure,
this will be addressed in further patches.
Note the space for file descriptors is statically allocated
in 8K arguments area, not on stack.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Now it has only one descritor for dumping pages
v2: remove rudiments
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
It happened some routines in parasite service code
were not following calling convention so I fixed the
callers and added a comment about adding new code here.
At the same time the 3 error code madness is dropped
as being requested by Pavel -- now we return one error
code only.
The PARASITE_ERR_ codes were dropped as well due to
become redundant.
The status of parasite service routine is set via
SET_PARASITE_RET helper. In case if there is no error --
just return 0. The calling code should not expect to
find anything sane in parasite_status_t because parasite
code might not touch it at all.
Same time, due to this convention the parasite's
dump_socket_queue is getting rid of third @err
member, because it's now returned as a regular
error code.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
This patch adds sockets queue dump functionality. Key ideas
1) sockets info is passed as plain array in parasite args.
2) new socket option SO_PEEK_OFF with MSG_PEEK is used to read the get the
queue's packets.
3) Buffer for packet will be allocated for each socket separately and with
size of socket sending buffer. For stream sockets is means, that it's queue
will be dumped in chunks of this size.
Note: loop around sys_msgrcv() is required for DGRAM sockets - sys_msgrcv()
with MSG_PEEK will return only one packet.
Based on xemul@ patches.
Signed-off-by: Stanislav Kinsbursky <skinsbursky@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Stack heap size is probably small for dumping unix sockest skb's (it's hard to
discover, how big it could be; thus let's assume that 10MB is enough,
otherwise give up and throw error).
So let' replace this stack heap wil 10 MB anonimous private mapping.
Signed-off-by: Stanislav Kinsbursky <skinsbursky@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
The ptrace seize doesn't prevent signals from delivery. That said,
we should block the signals in the target task before dumping
anything which is signals-related, i.e. memory and registers.
But once we've blocked signals, we should dump registers before
unblocking them, since any postponed signal will screw things up.
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
