We only had two callers to nbd_export_new; qemu-nbd.c always
passed a valid offset/length pair (because it already checked
the file length, to ensure that offset was in bounds), while
blockdev-nbd.c always passed 0/-1. Then nbd_export_new reduces
the size to a multiple of BDRV_SECTOR_SIZE (can only happen
when offset is not sector-aligned, since bdrv_getlength()
currently rounds up) (someday, it would be nice to have
byte-accurate lengths - but not today).
However, I'm finding it easier to work with the code if we are
consistent on having both callers pass in a valid length, and
just assert that things are sane in nbd_export_new, meaning
that no negative values were passed, and that offset+size does
not exceed 63 bits (as that really is a fundamental limit to
later operations, whether we use off_t or uint64_t).
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20190117193658.16413-6-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
We only have one caller that wants to export a bitmap name,
which it does right after creation of the export. But there is
still a brief window of time where an NBD client could see the
export but not the dirty bitmap, which a robust client would
have to interpret as meaning the entire image should be treated
as dirty. Better is to eliminate the window entirely, by
inlining nbd_export_bitmap() into nbd_export_new(), and refusing
to create the bitmap in the first place if the requested bitmap
can't be located.
We also no longer need logic for setting a different bitmap
name compared to the bitmap being exported.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20190111194720.15671-8-eblake@redhat.com>
The existing NBD code had a weird split where nbd_export_new()
created an export but did not add it to the list of exported
names until a later nbd_export_set_name() came along and grabbed
a second reference on the object; later, the first call to
nbd_export_close() drops the second reference while removing
the export from the list. This is in part because the QAPI
NbdServerRemoveNode enum documents the possibility of adding a
mode where we could do a soft disconnect: preventing new clients,
but waiting for existing clients to gracefully quit, based on
the mode used when calling nbd_export_close().
But in spite of all that, note that we never change the name of
an NBD export while it is exposed, which means it is easier to
just inline the process of setting the name as part of creating
the export.
Inline the contents of nbd_export_set_name() and
nbd_export_set_description() into the two points in an export
lifecycle where they matter, then adjust both callers to pass
the name up front. Note that for creation, all callers pass a
non-NULL name, (passing NULL at creation was for old style
servers, but we removed support for that in commit 7f7dfe2a),
so we can add an assert and do things unconditionally; but for
cleanup, because of the dual nature of nbd_export_close(), we
still have to be careful to avoid use-after-free. Along the
way, add a comment reminding ourselves of the potential of
adding a middle mode disconnect.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20190111194720.15671-5-eblake@redhat.com>
Our initial implementation of x-nbd-server-add-bitmap put
in a restriction because of incremental backups: in that
usage, we are exporting one qcow2 file (the temporary overlay
target of a blockdev-backup sync:none job) and a dirty bitmap
owned by a second qcow2 file (the source of the
blockdev-backup, which is the backing file of the temporary).
While both qcow2 files are still writable (the target in
order to capture copy-on-write of old contents, and the
source in order to track live guest writes in the meantime),
the NBD client expects to see constant data, including the
dirty bitmap. An enabled bitmap in the source would be
modified by guest writes, which is at odds with the NBD
export being a read-only constant view, hence the initial
code choice of enforcing a disabled bitmap (the intent is
that the exposed bitmap was disabled in the same transaction
that started the blockdev-backup job, although we don't want
to track enough state to actually enforce that).
However, consider the case of a bitmap contained in a read-only
node (including when the bitmap is found in a backing layer of
the active image). Because the node can't be modified, the
bitmap won't change due to writes, regardless of whether it is
still enabled. Forbidding the export unless the bitmap is
disabled is awkward, paritcularly since we can't change the
bitmap to be disabled (because the node is read-only).
Alternatively, consider the case of live storage migration,
where management directs the destination to create a writable
NBD server, then performs a drive-mirror from the source to
the target, prior to doing the rest of the live migration.
Since storage migration can be time-consuming, it may be wise
to let the destination include a dirty bitmap to track which
portions it has already received, where even if the migration
is interrupted and restarted, the source can query the
destination block status in order to potentially minimize
re-sending data that has not changed in the meantime on a
second attempt. Such code has not been written, and might not
be trivial (after all, a cluster being marked dirty in the
bitmap does not necessarily guarantee it has the desired
contents), but it makes sense that letting an active dirty
bitmap be exposed and changing alongside writes may prove
useful in the future.
Solve both issues by gating the restriction against a
disabled bitmap to only happen when the caller has requested
a read-only export, and where the BDS that owns the bitmap
(whether or not it is the BDS handed to nbd_export_new() or
from its backing chain) is still writable. We could drop
the check altogether (if management apps are prepared to
deal with a changing bitmap even on a read-only image), but
for now keeping a check for the read-only case still stands
a chance of preventing management errors.
Update iotest 223 to show the looser behavior by leaving
a bitmap enabled the whole run; note that we have to tear
down and re-export a node when handling an error.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20190111194720.15671-4-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
The NBD spec, and even our code comment, says that if the client
asks for NBD_OPT_LIST_META_CONTEXT with 0 queries, then we should
reply with (a possibly-compressed representation of) ALL contexts
that we are willing to let them try. But commit 3d068aff forgot
to advertise qemu:dirty-bitmap:FOO.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20181130023232.3079982-2-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Commit 37ec36f6 intentionally ignores errors when trying to reply
to an NBD_OPT_ABORT request for plaintext clients, but did not make
the same change for a TLS server. Since NBD_OPT_ABORT is
documented as being a potential for an EPIPE when the client hangs
up without waiting for our reply, we don't need to pollute the
server's output with that failure.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20181117223221.2198751-1-eblake@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
A space was missing after the option number was printed:
Option 0x8not permitted before TLS
becomes
Option 0x8 not permitted before TLS
This fixes
commit 3668328303
Author: Eric Blake <eblake@redhat.com>
Date: Fri Oct 14 13:33:09 2016 -0500
nbd: Send message along with server NBD_REP_ERR errors
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20181116155325.22428-2-berrange@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: move lone space to next line]
Signed-off-by: Eric Blake <eblake@redhat.com>
Whether it's "locked" or "frozen", it's in use and should
not be allowed for the purposes of this operation.
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20181002230218.13949-7-jsnow@redhat.com
Signed-off-by: John Snow <jsnow@redhat.com>
After the previous commit, nbd_client_new's first parameter is always
NULL. Let's drop it with all corresponding old-style negotiation code
path which is unreachable now.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20181003170228.95973-3-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: re-wrap short line]
Signed-off-by: Eric Blake <eblake@redhat.com>
We should not go to structured-read branch on CACHE command, fix that.
Bug introduced in bc37b06a5c "nbd/server: introduce NBD_CMD_CACHE"
with the whole feature and affects 3.0.0 release.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
CC: qemu-stable@nongnu.org
Message-Id: <20181003144738.70670-1-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: commit message typo fix]
Signed-off-by: Eric Blake <eblake@redhat.com>
Taking the address of a field in a packed struct is a bad idea, because
it might not be actually aligned enough for that pointer type (and
thus cause a crash on dereference on some host architectures). Newer
versions of clang warn about this. Avoid the bug by not using the
"modify in place" byte swapping functions.
This patch was produced with the following spatch script:
@@
expression E;
@@
-be16_to_cpus(&E);
+E = be16_to_cpu(E);
@@
expression E;
@@
-be32_to_cpus(&E);
+E = be32_to_cpu(E);
@@
expression E;
@@
-be64_to_cpus(&E);
+E = be64_to_cpu(E);
@@
expression E;
@@
-cpu_to_be16s(&E);
+E = cpu_to_be16(E);
@@
expression E;
@@
-cpu_to_be32s(&E);
+E = cpu_to_be32(E);
@@
expression E;
@@
-cpu_to_be64s(&E);
+E = cpu_to_be64(E);
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20180927164200.15097-1-peter.maydell@linaro.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: rebase, and squash in missed changes]
Signed-off-by: Eric Blake <eblake@redhat.com>
This is necessary for efficient block-status export, for clients which
support it. (qemu is not yet such a client, but could become one.)
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180704112302.471456-3-vsementsov@virtuozzo.com>
[eblake: grammar tweaks]
Signed-off-by: Eric Blake <eblake@redhat.com>
bitmap_to_extents function is broken: it switches dirty variable after
every iteration, however it can process only part of dirty (or zero)
area during one iteration in case when this area is too large for one
extent.
Fortunately, the bug doesn't produce wrong extent flags: it just inserts
a zero-length extent between sequential extents representing large dirty
(or zero) area. However, zero-length extents are forbidden by the NBD
protocol. So, a careful client should consider such a reply as a server
fault, while a less-careful will likely ignore zero-length extents.
The bug can only be triggered by a client that requests block status
for nearly 4G at once (a request of 4G and larger is impossible per
the protocol, and requests smaller than 4G less the bitmap granularity
cause the loop to quit iterating rather than revisit the tail of the
large area); it also cannot trigger if the client used the
NBD_CMD_FLAG_REQ_ONE flag. Since qemu 3.0 as client (using the
x-dirty-bitmap extension) always passes the flag, it is immune; and
we are not aware of other open-source clients that know how to request
qemu:dirty-bitmap:FOO contexts. Clients that want to avoid the bug
could cap block status requests to a smaller length, such as 2G or 3G.
Fix this by more careful handling of dirty variable.
Bug was introduced in 3d068aff16
"nbd/server: implement dirty bitmap export", with the whole function.
and is present in v3.0.0 release.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180914165116.23182-1-vsementsov@virtuozzo.com>
CC: qemu-stable@nongnu.org
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: improved commit message]
Signed-off-by: Eric Blake <eblake@redhat.com>
Call nbd_co_send_extents() with correct length parameter
(extent.length may be smaller than original length).
Also, switch length parameter type to uint32_t, to correspond with
request->len and similar nbd_co_send_bitmap().
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180704112302.471456-2-vsementsov@virtuozzo.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
In my hurry to fix a build failure, I introduced a logic bug.
The assertion conditional is backwards, meaning that qemu will
now abort instead of reporting dirty bitmap status.
The bug can only be tickled by an NBD client using an exported
dirty bitmap (which is still an experimental QMP command), so
it's not the end of the world for supported usage (and neither
'make check' nor qemu-iotests fails); but it also shows that we
really want qemu-io support for reading dirty bitmaps if only
so that I can add iotests coverage to prevent future
brown-bag-of-shame events like this one.
Fixes: 45eb6fb6
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180622153509.375130-1-eblake@redhat.com>
The code has a while() loop that always initialized 'end', and
the loop always executes at least once (as evidenced by the assert()
just prior to the loop). But some versions of gcc still complain
that 'end' is used uninitialized, so silence them.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20180622125814.345274-1-eblake@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Handle nbd CACHE command. Just do read, without sending read data back.
Cache mechanism should be done by exported node driver chain.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180413143156.11409-1-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: fix two missing case labels in switch statements]
Signed-off-by: Eric Blake <eblake@redhat.com>
Handle a new NBD meta namespace: "qemu", and corresponding queries:
"qemu:dirty-bitmap:<export bitmap name>".
With the new metadata context negotiated, BLOCK_STATUS query will reply
with dirty-bitmap data, converted to extents. The new public function
nbd_export_bitmap selects which bitmap to export. For now, only one bitmap
may be exported.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180609151758.17343-5-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: wording tweaks, minor cleanups, additional tracing]
Signed-off-by: Eric Blake <eblake@redhat.com>
Add nbd_meta_pattern() and nbd_meta_empty_or_pattern() helpers for
metadata query parsing. nbd_meta_pattern() will be reused for the
"qemu" namespace in following patches.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180609151758.17343-4-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: comment tweaks]
Signed-off-by: Eric Blake <eblake@redhat.com>
Use NBDExport pointer instead of just export name: there is no need to
store a duplicated name in the struct; moreover, NBDExport will be used
further.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180609151758.17343-3-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: commit message grammar tweak]
Signed-off-by: Eric Blake <eblake@redhat.com>
Return code = 1 doesn't mean that we parsed base:allocation. Use
correct traces in both -parsed and -skipped cases.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180609151758.17343-2-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: comment tweaks]
Signed-off-by: Eric Blake <eblake@redhat.com>
The NBD spec says that behavior is unspecified if the client
requests 0 length for block status; but since the structured
reply is documenting as returning a non-zero length, it's
easier to just diagnose this with an EINVAL error than to
figure out what to return.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180621124937.166549-1-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Having a more detailed log of the interaction between client and
server is invaluable in debugging how meta context negotiation
actually works.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180330130950.1931229-1-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Minimal realization: only one extent in server answer is supported.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180312152126.286890-4-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: tweak whitespace, move constant from .h to .c, improve
logic of check_meta_export_name, simplify nbd_negotiate_options
by doing more in nbd_negotiate_meta_queries]
Signed-off-by: Eric Blake <eblake@redhat.com>
Add helper to read name in format:
uint32 len (<= NBD_MAX_NAME_SIZE)
len bytes string (not 0-terminated)
The helper will be reused in following patch.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180312152126.286890-3-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: grammar fixes, actually check error]
Signed-off-by: Eric Blake <eblake@redhat.com>
NBD_REP_ERR_INVALID is often parameter to nbd_opt_drop and it would
be used more in following patches. So, let's add a helper.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180312152126.286890-2-vsementsov@virtuozzo.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
The NBD spec states that since trim requests can affect disk contents,
then they should allow for FUA semantics just like writes for ensuring
the disk has settled before returning. As bdrv_[co_]pdiscard() does
not support a flags argument, we can't pass FUA down the block layer
stack, and must therefore emulate it with a flush at the NBD layer.
Note that in all reality, generic well-behaved clients will never
send TRIM+FUA (in fact, qemu as a client never does, and we have no
intention to plumb flags into bdrv_pdiscard). This is because the
NBD protocol states that it is unspecified to READ a trimmed area
(you might read stale data, all zeroes, or even random unrelated
data) without first rewriting it, and even the experimental
BLOCK_STATUS extension states that TRIM need not affect reported
status. Thus, in the general case, a client cannot tell the
difference between an arbitrary server that ignores TRIM, a server
that had a power outage without flushing to disk, and a server that
actually affected the disk before returning; so waiting for the
trim actions to flush to disk makes little sense. However, for a
specific client and server pair, where the client knows the server
treats TRIM'd areas as guaranteed reads-zero, waiting for a flush
makes sense, hence why the protocol documents that FUA is valid on
trim. So, even though the NBD protocol doesn't have a way for the
server to advertise what effects (if any) TRIM will actually have,
and thus any client that relies on specific effects is probably
in error, we can at least support a client that requests TRIM+FUA.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180307225732.155835-1-eblake@redhat.com>
Split out request handling logic.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180308184636.178534-6-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: touch up blank line placement]
Signed-off-by: Eric Blake <eblake@redhat.com>
nbd_trip has difficult logic when sending replies: it tries to use one
code path for all replies. It is ok for simple replies, but is not
comfortable for structured replies. Also, two types of error (and
corresponding messages in local_err) - fatal (leading to disconnect)
and not-fatal (just to be sent to the client) are difficult to follow.
To make things a bit clearer, the following is done:
- split CMD_READ logic to separate function. It is the most difficult
command for now, and it is definitely cramped inside nbd_trip. Also,
it is difficult to follow CMD_READ logic, shared between
"case NBD_CMD_READ" and "if"s under "reply:" label.
- create separate helper function nbd_send_generic_reply() and use it
both in new nbd_do_cmd_read and for other commands in nbd_trip instead
of common code-path under "reply:" label in nbd_trip. The helper
supports an error message, so logic with local_err in nbd_trip is
simplified.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180308184636.178534-5-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: grammar tweaks and blank line placement]
Signed-off-by: Eric Blake <eblake@redhat.com>
Since the unchanged code has just set client->recv_coroutine to
NULL before calling nbd_client_receive_next_request(), we are
spawning a new coroutine unconditionally, but the first thing
that coroutine will do is check for client->closing, making it
a no-op if we have already detected that the client is going
away. Furthermore, for any error other than EIO (where we
disconnect, which itself sets client->closing), if the client
has already gone away, we'll probably encounter EIO later
in the function and attempt disconnect at that point. Logically,
as soon as we know the connection is closing, there is no need
to try a likely-to-fail a response or spawn a no-op coroutine.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180308184636.178534-4-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: squash in further reordering: hoist check before spawning
next coroutine, and document rationale in commit message]
Signed-off-by: Eric Blake <eblake@redhat.com>
In case of io error in nbd_co_send_sparse_read we should not
"goto reply:", as it was a fatal error and the common behavior
is to disconnect in this case. We should not try to send the
client an additional error reply, since we already hit a
channel-io error on our previous attempt to send one.
Fix this by handling block-status error in nbd_co_send_sparse_read,
so nbd_co_send_sparse_read fails only on io error. Then just skip
common "reply:" code path in nbd_trip.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180308184636.178534-3-vsementsov@virtuozzo.com>
[eblake: grammar tweaks]
Signed-off-by: Eric Blake <eblake@redhat.com>
To be reused in nbd_co_send_sparse_read() in the following patch.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180308184636.178534-2-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
A new parameter "context" is added to qio_channel_tls_handshake() is to
allow the TLS to be run on a non-default context. Still, no functional
change.
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
1. NBD_REP_ERR_INVALID is not only about length, so, make message more
general
2. hex format is not very good: it's hard to read something like
"option a (set meta context)", so switch to dec.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <1518702707-7077-6-git-send-email-vsementsov@virtuozzo.com>
[eblake: expand scope of patch: ALL uses of nbd_opt_lookup and
nbd_rep_lookup are now decimal]
Signed-off-by: Eric Blake <eblake@redhat.com>
Add command for removing an export. It is needed for cases when we
don't want to keep the export after the operation on it was completed.
The other example is a temporary node, created with blockdev-add.
If we want to delete it we should firstly remove any corresponding
NBD export.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180119135719.24745-3-vsementsov@virtuozzo.com>
[eblake: drop dead nb_clients code]
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20171122101958.17065-6-vsementsov@virtuozzo.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Rather than making every callsite perform length sanity checks
and error reporting, add the helper functions nbd_opt_read()
and nbd_opt_drop() that use the length stored in the client
struct; also add an assertion that optlen is 0 before any
option (ie. any previous option was fully handled), complementing
the assertion added in an earlier patch that optlen is 0 after
all negotiation completes.
Note that the call in nbd_negotiate_handle_export_name() does
not use the new helper (in part because the server cannot
reply to NBD_OPT_EXPORT_NAME - it either succeeds or the
connection drops).
Based on patches by Vladimir Sementsov-Ogievskiy.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180110230825.18321-6-eblake@redhat.com>
This will be useful for the next patch.
Based on a patch by Vladimir Sementsov-Ogievskiy
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180110230825.18321-5-eblake@redhat.com>
When a client abruptly disconnects before we've finished reading
the name sent with NBD_OPT_EXPORT_NAME, we are better off logging
the failure as EIO (we can't communicate with the client), rather
than EINVAL (the client sent bogus data).
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180110230825.18321-4-eblake@redhat.com>
Instead of passing currently negotiating option and its length to
many of negotiation functions let's just store them on NBDClient
struct to be state-variables of negotiation phase.
This unifies semantics of negotiation functions and allows
tracking changes of remaining option length in future patches.
Asssert that optlen is back to 0 after negotiation (including
old-style connections which don't negotiate), although we need
more patches before we can assert optlen is 0 between options
during negotiation.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20171122101958.17065-2-vsementsov@virtuozzo.com>
[eblake: rebase, commit message tweak, assert !optlen after
negotiation completes]
Signed-off-by: Eric Blake <eblake@redhat.com>
No semantic change, but will make it easier for an upcoming patch
to refactor code without having to add forward declarations. Fix
a poor comment while at it.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20180110230825.18321-2-eblake@redhat.com>
This place is not obvious, nbd_export_close may theoretically reduce
refcount to 0. It may happen if someone calls nbd_export_put on named
export not through nbd_export_set_name when refcount is 1.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20171207155102.66622-2-vsementsov@virtuozzo.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
If we are careful to handle 0-length read requests correctly,
we can optimize our sparse read to send the NBD_REPLY_FLAG_DONE
bit on our last OFFSET_DATA or OFFSET_HOLE chunk rather than
needing a separate chunk.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20171107030912.23930-3-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
The reason that NBD added structured reply in the first place was
to allow for efficient reads of sparse files, by allowing the
reply to include chunks to quickly communicate holes to the client
without sending lots of zeroes over the wire. Time to implement
this in the server; our client can already read such data.
We can only skip holes insofar as the block layer can query them;
and only if the client is okay with a fragmented request (if a
client requests NBD_CMD_FLAG_DF and the entire read is a hole, we
could technically return a single NBD_REPLY_TYPE_OFFSET_HOLE, but
that's a fringe case not worth catering to here). Sadly, the
control flow is a bit wonkier than I would have preferred, but
it was minimally invasive to have a split in the action between
a fragmented read (handled directly where we recognize
NBD_CMD_READ with the right conditions, and sending multiple
chunks) vs. a single read (handled at the end of nbd_trip, for
both simple and structured replies, when we know there is only
one thing being read). Likewise, I didn't make any effort to
optimize the final chunk of a fragmented read to set the
NBD_REPLY_FLAG_DONE, but unconditionally send that as a separate
NBD_REPLY_TYPE_NONE.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20171107030912.23930-2-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Introduced in commit f37708f6b8 (2.10). The NBD spec says a client
can request export names up to 4096 bytes in length, even though
they should not expect success on names longer than 256. However,
qemu hard-codes the limit of 256, and fails to filter out a client
that probes for a longer name; the result is a stack smash that can
potentially give an attacker arbitrary control over the qemu
process.
The smash can be easily demonstrated with this client:
$ qemu-io f raw nbd://localhost:10809/$(printf %3000d 1 | tr ' ' a)
If the qemu NBD server binary (whether the standalone qemu-nbd, or
the builtin server of QMP nbd-server-start) was compiled with
-fstack-protector-strong, the ability to exploit the stack smash
into arbitrary execution is a lot more difficult (but still
theoretically possible to a determined attacker, perhaps in
combination with other CVEs). Still, crashing a running qemu (and
losing the VM) is bad enough, even if the attacker did not obtain
full execution control.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
The NBD spec gives us permission to abruptly disconnect on clients
that send outrageously large option requests, rather than having
to spend the time reading to the end of the option. No real
option request requires that much data anyways; and meanwhile, we
already have the practice of abruptly dropping the connection on
any client that sends NBD_CMD_WRITE with a payload larger than 32M.
For comparison, nbdkit drops the connection on any request with
more than 4096 bytes; however, that limit is probably too low
(as the NBD spec states an export name can theoretically be up
to 4096 bytes, which means a valid NBD_OPT_INFO could be even
longer) - even if qemu doesn't permit exports longer than 256
bytes.
It could be argued that a malicious client trying to get us to
read nearly 4G of data on a bad request is a form of denial of
service. In particular, if the server requires TLS, but a client
that does not know the TLS credentials sends any option (other
than NBD_OPT_STARTTLS or NBD_OPT_EXPORT_NAME) with a stated
payload of nearly 4G, then the server was keeping the connection
alive trying to read all the payload, tying up resources that it
would rather be spending on a client that can get past the TLS
handshake. Hence, this warranted a CVE.
Present since at least 2.5 when handling known options, and made
worse in 2.6 when fixing support for NBD_FLAG_C_FIXED_NEWSTYLE
to handle unknown options.
CC: qemu-stable@nongnu.org
Signed-off-by: Eric Blake <eblake@redhat.com>
The NBD spec says an attempt to NBD_CMD_TRIM on a read-only
export should fail with EPERM, as a trim has the potential
to change disk contents, but we were relying on the block
layer to catch that for us, which might not always give the
right error (and even if it does, it does not let us pass
back a sane message for structured replies).
The NBD spec says an attempt to NBD_CMD_WRITE_ZEROES out of
bounds should fail with ENOSPC, not EINVAL.
Our check for u64 offset + u32 length wraparound up front is
pointless; nothing uses offset until after the second round
of sanity checks, and we can just as easily ensure there is
no wraparound by checking whether offset is in bounds (since
a disk size cannot exceed off_t which is 63 bits, adding a
32-bit number for a valid offset can't overflow). Bonus:
dropping the up-front check lets us keep the connection alive
after NBD_CMD_WRITE, whereas before we would drop the
connection (of course, any client sending a packet that would
trigger the failure is already buggy, so it's also okay to
drop the connection, but better quality-of-implementation
never hurts).
Solve all of these issues by some code motion and improved
request validation.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20171115213557.3548-1-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
The NBD spec was recently clarified to state that a read of length 0
should not be attempted by a compliant client; but that a server must
still handle it correctly in an unspecified manner (that is, either
a successful no-op or an error reply, but not a crash) [1]. However,
it also implies that NBD_REPLY_TYPE_OFFSET_DATA must have a non-zero
payload length, but our existing code was replying with a chunk
that a picky client could reject as invalid because it was missing
a payload (our own client implementation was recently patched to be
that picky, after first fixing it to not send 0-length requests).
We are already doing successful no-ops for 0-length writes and for
non-structured reads; so for consistency, we want structured reply
reads to also be a no-op. The easiest way to do this is to return
a NBD_REPLY_TYPE_NONE chunk; this is best done via a new helper
function (especially since future patches for other structured
replies may benefit from using the same helper).
[1] https://github.com/NetworkBlockDevice/nbd/commit/ee926037
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20171108215703.9295-8-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
A closer read of the NBD spec shows that a structured reply chunk
for a hole is not quite identical to the prefix of a data chunk,
because the hole has to also send a 32-bit size field. Although
we do not yet send holes, we should fix the misleading information
in our header and make it easier for a future patch to support
sparse reads. Messed up in commit bae245d1.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20171108215703.9295-5-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Eric Blake