Set new allocated bfqq to bic or remove freed bfqq from bic are both
protected by bfqd->lock, however bfq_limit_depth() is deferencing bfqq
from bic without the lock, this can lead to UAF if the io_context is
shared by multiple tasks.
For example, test bfq with io_uring can trigger following UAF in v6.6:
==================================================================
BUG: KASAN: slab-use-after-free in bfqq_group+0x15/0x50
Fix the problem by protecting bic_to_bfqq() with bfqd->lock.
CC: Jan Kara <jack@suse.cz> Fixes: 76f1df88bbc2 ("bfq: Limit number of requests consumed by each cgroup") Signed-off-by: Yu Kuai <yukuai3@huawei.com> Link: https://lore.kernel.org/r/20241129091509.2227136-1-yukuai1@huaweicloud.com Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Sasha Levin <sashal@kernel.org>
Every pNFS SCSI IO wants to do LAYOUTGET, then within the layout find the
device which can drive GETDEVINFO, then finally may need to prep the device
with a reservation. This slow work makes a mess of IO latencies if one of
the later steps is going to fail for awhile.
If we're unable to register a SCSI device, ensure we mark the device as
unavailable so that it will timeout and be re-added via GETDEVINFO. This
avoids repeated doomed attempts to register a device in the IO path.
Add some clarifying comments as well.
Fixes: d869da91cccb ("nfs/blocklayout: Fix premature PR key unregistration") Signed-off-by: Benjamin Coddington <bcodding@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Since commit d869da91cccb ("nfs/blocklayout: Fix premature PR key
unregistration") an unmount of a pNFS SCSI layout-enabled NFS may
dereference a NULL block_device in:
This happens because even though we were able to create the
nfs4_deviceid_node, the lookup for the device was unable to attach the
block device to the pnfs_block_dev.
If we never found a block device to register, we can avoid this case with
the PNFS_BDEV_REGISTERED flag. Move the deref behind the test for the
flag.
Fixes: d869da91cccb ("nfs/blocklayout: Fix premature PR key unregistration") Signed-off-by: Benjamin Coddington <bcodding@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
ip netns add netns_1
ip link add name veth_1_peer type veth peer veth_1
ifconfig veth_1_peer 11.11.0.254 up
ip link set veth_1 netns netns_1
ip netns exec netns_1 ifconfig veth_1 11.11.0.1
ip netns exec netns_1 /root/iptables -A OUTPUT -d 11.11.0.254 -p tcp \
--tcp-flags FIN FIN -j DROP
(note: In my environment, a DESTROY_CLIENTID operation is always sent
immediately, breaking the nfs tcp connection.)
ip netns exec netns_1 timeout -s 9 300 mount -t nfs -o proto=tcp,vers=4.1 \
11.11.0.254:/mnt/nfsshare /mnt/nfs/netns_1
ip netns del netns_1
The reason here is that the tcp socket in netns_1 (nfs side) has been
shutdown and closed (done in xs_destroy), but the FIN message (with ack)
is discarded, and the nfsd side keeps sending retransmission messages.
As a result, when the tcp sock in netns_1 processes the received message,
it sends the message (FIN message) in the sending queue, and the tcp timer
is re-established. When the network namespace is deleted, the net structure
accessed by tcp's timer handler function causes problems.
To fix this problem, let's hold netns refcnt for the tcp kernel socket as
done in other modules. This is an ugly hack which can easily be backported
to earlier kernels. A proper fix which cleans up the interfaces will
follow, but may not be so easy to backport.
Fixes: 26abe14379f8 ("net: Modify sk_alloc to not reference count the netns of kernel sockets.") Signed-off-by: Liu Jian <liujian56@huawei.com> Acked-by: Jeff Layton <jlayton@kernel.org> Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
We've noticed a situation where an unstable TCP connection can cause the
TLS handshake to timeout waiting for userspace to complete it. When this
happens, we don't want to return from xs_tls_handshake_sync() with zero, as
this will cause the upper xprt to be set CONNECTED, and subsequent attempts
to transmit will be returned with -EPIPE. The sunrpc machine does not
recover from this situation and will spin attempting to transmit.
The return value of tls_handshake_cancel() can be used to detect a race
with completion:
* tls_handshake_cancel - cancel a pending handshake
* Return values:
* %true - Uncompleted handshake request was canceled
* %false - Handshake request already completed or not found
If true, we do not want the upper xprt to be connected, so return
-ETIMEDOUT. If false, its possible the handshake request was lost and
that may be the reason for our timeout. Again we do not want the upper
xprt to be connected, so return -ETIMEDOUT.
Ensure that we alway return an error from xs_tls_handshake_sync() if we
call tls_handshake_cancel().
Signed-off-by: Benjamin Coddington <bcodding@redhat.com> Reviewed-by: Chuck Lever <chuck.lever@oracle.com> Fixes: 75eb6af7acdf ("SUNRPC: Add a TCP-with-TLS RPC transport class") Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Since transport->sock has been set to NULL during reset transport,
XPRT_SOCK_UPD_TIMEOUT also needs to be cleared. Otherwise, the
xs_tcp_set_socket_timeouts() may be triggered in xs_tcp_send_request()
to dereference the transport->sock that has been set to NULL.
Fixes: 7196dbb02ea0 ("SUNRPC: Allow changing of the TCP timeout parameters on the fly") Signed-off-by: Li Lingfeng <lilingfeng3@huawei.com> Signed-off-by: Liu Jian <liujian56@huawei.com> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
When exporting only one file system with fsid=0 on the server side, the
client alternately uses the ro/rw mount options to perform the mount
operation, and a new vfsmount is generated each time.
It can be reproduced as follows:
[root@localhost ~]# mount /dev/sda /mnt2
[root@localhost ~]# echo "/mnt2 *(rw,no_root_squash,fsid=0)" >/etc/exports
[root@localhost ~]# systemctl restart nfs-server
[root@localhost ~]# mount -t nfs -o ro,vers=4 127.0.0.1:/ /mnt/sdaa
[root@localhost ~]# mount -t nfs -o rw,vers=4 127.0.0.1:/ /mnt/sdaa
[root@localhost ~]# mount -t nfs -o ro,vers=4 127.0.0.1:/ /mnt/sdaa
[root@localhost ~]# mount -t nfs -o rw,vers=4 127.0.0.1:/ /mnt/sdaa
[root@localhost ~]# mount | grep nfs4
127.0.0.1:/ on /mnt/sdaa type nfs4 (ro,relatime,vers=4.2,rsize=1048576,...
127.0.0.1:/ on /mnt/sdaa type nfs4 (rw,relatime,vers=4.2,rsize=1048576,...
127.0.0.1:/ on /mnt/sdaa type nfs4 (ro,relatime,vers=4.2,rsize=1048576,...
127.0.0.1:/ on /mnt/sdaa type nfs4 (rw,relatime,vers=4.2,rsize=1048576,...
[root@localhost ~]#
We expected that after mounting with the ro option, using the rw option to
mount again would return EBUSY, but the actual situation was not the case.
As shown above, when mounting for the first time, a superblock with the ro
flag will be generated, and at the same time, in do_new_mount_fc -->
do_add_mount, it detects that the superblock corresponding to the current
target directory is inconsistent with the currently generated one
(path->mnt->mnt_sb != newmnt->mnt.mnt_sb), and a new vfsmount will be
generated.
When mounting with the rw option for the second time, since no matching
superblock can be found in the fs_supers list, a new superblock with the
rw flag will be generated again. The superblock in use (ro) is different
from the newly generated superblock (rw), and a new vfsmount will be
generated again.
When mounting with the ro option for the third time, the superblock (ro)
is found in fs_supers, the superblock in use (rw) is different from the
found superblock (ro), and a new vfsmount will be generated again.
We can switch between ro/rw through remount, and only one superblock needs
to be generated, thus avoiding the problem of repeated generation of
vfsmount caused by switching superblocks.
Furthermore, This can also resolve the issue described in the link.
Fixes: 275a5d24bf56 ("NFS: Error when mounting the same filesystem with different options") Link: https://lore.kernel.org/all/20240604112636.236517-3-lilingfeng@huaweicloud.com/ Signed-off-by: Li Lingfeng <lilingfeng3@huawei.com> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Unlock before returning if smb3_sync_session_ctx_passwords() fails.
Fixes: 7e654ab7da03 ("cifs: during remount, make sure passwords are in sync") Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org> Reviewed-by: Bharath SM <bharathsm@microsoft.com> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
This fixes scenarios where remount can overwrite the only currently
working password, breaking reconnect.
We recently introduced a password2 field in both ses and ctx structs.
This was done so as to allow the client to rotate passwords for a mount
without any downtime. However, when the client transparently handles
password rotation, it can swap the values of the two password fields
in the ses struct, but not in smb3_fs_context struct that hangs off
cifs_sb. This can lead to a situation where a remount unintentionally
overwrites a working password in the ses struct.
In order to fix this, we first get the passwords in ctx struct
in-sync with ses struct, before replacing them with what the passwords
that could be passed as a part of remount.
Also, in order to avoid race condition between smb2_reconnect and
smb3_reconfigure, we make sure to lock session_mutex before changing
password and password2 fields of the ses structure.
Fixes: 35f834265e0d ("smb3: fix broken reconnect when password changing on the server by allowing password rotation") Signed-off-by: Shyam Prasad N <sprasad@microsoft.com> Signed-off-by: Meetakshi Setiya <msetiya@microsoft.com> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
This checked if the first character of eisa_device_id::sig was not '\0'.
However, commit ac551828993e changed it as follows:
if (sig[0])
sig[0] is NOT the first character of the eisa_device_id::sig. The
type of 'sig' is 'char (*)[8]', meaning that the type of 'sig[0]' is
'char [8]' instead of 'char'. 'sig[0]' and 'symval' refer to the same
address, which never becomes NULL.
The correct conversion would have been:
if ((*sig)[0])
However, this if-conditional was meaningless because the earlier change
in commit ac551828993e was incorrect.
This commit removes the entire incorrect code, which should never have
been executed.
Fixes: ac551828993e ("modpost: i2c aliases need no trailing wildcard") Fixes: 6543becf26ff ("mod/file2alias: make modalias generation safe for cross compiling") Signed-off-by: Masahiro Yamada <masahiroy@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
A write which goes past the end of the bdev in blkdev_write_iter() will
be truncated. Truncating cannot tolerated for an atomic write, so error
that condition.
Avoid leaking a tcon ref when a lease break races with opening the
cached directory. Processing the leak break might take a reference to
the tcon in cached_dir_lease_break() and then fail to release the ref in
cached_dir_offload_close, since cfid->tcon is still NULL.
Fixes: ebe98f1447bb ("cifs: enable caching of directories for which a lease is held") Signed-off-by: Paul Aurich <paul@darkrain42.org> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Kernels built without CONFIG_MODULES might still want to create -dbg deb
packages but install_linux_image_dbg() assumes modules.order always
exists. This obviously isn't true if no modules were built, so we should
skip reading modules.order in that case.
Fixes: 16c36f8864e3 ("kbuild: deb-pkg: use build ID instead of debug link for dbg package") Signed-off-by: Matt Fleming <mfleming@cloudflare.com> Signed-off-by: Masahiro Yamada <masahiroy@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Commit b1fca27d384e ("kernel debug: support resetting WARN*_ONCE")
added support for clearing the state of once warnings. However,
it is not functional when CONFIG_LD_DEAD_CODE_DATA_ELIMINATION or
CONFIG_LTO_CLANG is enabled, because .data.once matches the
.data.[0-9a-zA-Z_]* pattern in the DATA_MAIN macro.
Commit cb87481ee89d ("kbuild: linker script do not match C names unless
LD_DEAD_CODE_DATA_ELIMINATION is configured") was introduced to suppress
the issue for the default CONFIG_LD_DEAD_CODE_DATA_ELIMINATION=n case,
providing a minimal fix for stable backporting. We were aware this did
not address the issue for CONFIG_LD_DEAD_CODE_DATA_ELIMINATION=y. The
plan was to apply correct fixes and then revert cb87481ee89d. [1]
Seven years have passed since then, yet the #ifdef workaround remains in
place. Meanwhile, commit b1fca27d384e introduced the .data.once section,
and commit dc5723b02e52 ("kbuild: add support for Clang LTO") extended
the #ifdef.
Using a ".." separator in the section name fixes the issue for
CONFIG_LD_DEAD_CODE_DATA_ELIMINATION and CONFIG_LTO_CLANG.
Commit 7ccaba5314ca ("consolidate WARN_...ONCE() static variables")
was intended to collect all .data.unlikely sections into one chunk.
However, this has not worked when CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
or CONFIG_LTO_CLANG is enabled, because .data.unlikely matches the
.data.[0-9a-zA-Z_]* pattern in the DATA_MAIN macro.
Commit cb87481ee89d ("kbuild: linker script do not match C names unless
LD_DEAD_CODE_DATA_ELIMINATION is configured") was introduced to suppress
the issue for the default CONFIG_LD_DEAD_CODE_DATA_ELIMINATION=n case,
providing a minimal fix for stable backporting. We were aware this did
not address the issue for CONFIG_LD_DEAD_CODE_DATA_ELIMINATION=y. The
plan was to apply correct fixes and then revert cb87481ee89d. [1]
Seven years have passed since then, yet the #ifdef workaround remains in
place.
Using a ".." separator in the section name fixes the issue for
CONFIG_LD_DEAD_CODE_DATA_ELIMINATION and CONFIG_LTO_CLANG.
Fixes: cb87481ee89d ("kbuild: linker script do not match C names unless LD_DEAD_CODE_DATA_ELIMINATION is configured") Signed-off-by: Masahiro Yamada <masahiroy@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The undervoltage flags reported by the RTC are useful to know if the
time and date are reliable after a reboot. Although the threshold VLOW1
indicates that the thermometer has been shutdown and time compensation
is off, it doesn't mean that the temperature readout is currently
impossible.
As the system is running, the RTC voltage is now fully established and
we can read the temperature.
SMB1 NT_TRANSACT_IOCTL/FSCTL_GET_REPARSE_POINT even in non-UNICODE mode
returns reparse buffer in UNICODE/UTF-16 format.
This is because FSCTL_GET_REPARSE_POINT is NT-based IOCTL which does not
distinguish between 8-bit non-UNICODE and 16-bit UNICODE modes and its path
buffers are always encoded in UTF-16.
This change fixes reading of native symlinks in SMB1 when UNICODE session
is not active.
Fixes: ed3e0a149b58 ("smb: client: implement ->query_reparse_point() for SMB1") Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
SMB symlink which has SYMLINK_FLAG_RELATIVE set is relative (as opposite of
the absolute) and it can be relative either to the current directory (where
is the symlink stored) or relative to the top level export path. To what it
is relative depends on the first character of the symlink target path.
If the first character is path separator then symlink is relative to the
export, otherwise to the current directory. Linux (and generally POSIX
systems) supports only symlink paths relative to the current directory
where is symlink stored.
Currently if Linux SMB client reads relative SMB symlink with first
character as path separator (slash), it let as is. Which means that Linux
interpret it as absolute symlink pointing from the root (/). But this
location is different than the top level directory of SMB export (unless
SMB export was mounted to the root) and thefore SMB symlinks relative to
the export are interpreted wrongly by Linux SMB client.
Fix this problem. As Linux does not have equivalent of the path relative to
the top of the mount point, convert such symlink target path relative to
the current directory. Do this by prepending "../" pattern N times before
the SMB target path, where N is the number of path separators found in SMB
symlink path.
So for example, if SMB share is mounted to Linux path /mnt/share/, symlink
is stored in file /mnt/share/test/folder1/symlink (so SMB symlink path is
test\folder1\symlink) and SMB symlink target points to \test\folder2\file,
then convert symlink target path to Linux path ../../test/folder2/file.
Deduplicate code for parsing SMB symlinks in native form from functions
smb2_parse_symlink_response() and parse_reparse_native_symlink() into new
function smb2_parse_native_symlink() and pass into this new function a new
full_path parameter from callers, which specify SMB full path where is
symlink stored.
This change fixes resolving of the native Windows symlinks relative to the
top level directory of the SMB share.
Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Steve French <stfrench@microsoft.com>
Stable-dep-of: f4ca4f5a36ea ("cifs: Fix parsing reparse point with native symlink in SMB1 non-UNICODE session") Signed-off-by: Sasha Levin <sashal@kernel.org>
The init_size description of boot protocol has an example of the runtime
start address for the compressed bzImage. For non-relocatable kernel
it relies on the pref_address value (if not 0), but for relocatable case
only pays respect to the load_addres and kernel_alignment, and it is
inaccurate for the latter. Boot loader must consider the pref_address
as the Linux kernel relocates to it before being decompressed as nicely
described in this commit message a year ago:
Due to this documentation inaccuracy some of the bootloaders (*) made a
mistake in the calculations and if kernel image is big enough, this may
lead to unbootable configurations.
*)
In particular, kexec-tools missed that and resently got a couple of
changes which will be part of v2.0.30 release. For the record,
commit 43b1d3e68ee7 only fixed the kernel kexec implementation and
also missed to update the init_size description.
While at it, make an example C-like looking as it's done elsewhere in
the document and fix indentation as presribed by the reStructuredText
specifications, so the syntax highliting will work properly.
Fixes: 43b1d3e68ee7 ("kexec: Allocate kernel above bzImage's pref_address") Fixes: d297366ba692 ("x86: document new bzImage fields") Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Randy Dunlap <rdunlap@infradead.org> Cc: "H. Peter Anvin" <hpa@zytor.com> Link: https://lore.kernel.org/r/20241125105005.1616154-1-andriy.shevchenko@linux.intel.com Signed-off-by: Sasha Levin <sashal@kernel.org>
Setting dir_cache_timeout to zero should disable the caching of
directory contents. Currently, even when dir_cache_timeout is zero,
some caching related functions are still invoked, which is unintended
behavior.
Fix the issue by setting tcon->nohandlecache to true when
dir_cache_timeout is zero, ensuring that directory handle caching
is properly disabled.
Fixes: 238b351d0935 ("smb3: allow controlling length of time directory entries are cached with dir leases") Reviewed-by: Paulo Alcantara (Red Hat) <pc@manguebit.com> Reviewed-by: Enzo Matsumiya <ematsumiya@suse.de> Signed-off-by: Henrique Carvalho <henrique.carvalho@suse.com> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The portid_bits and deviceid_bits were set only for XP type nodes in
the arm_cmn_discover() and it confused other nodes to find XP nodes.
Copy the both bits from the XP nodes directly when it sets up a new
node.
Fixes: e79634b53e39 ("perf/arm-cmn: Refactor node ID handling. Again.") Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Will Deacon <will@kernel.org> Reviewed-by: Robin Murphy <robin.murphy@arm.com> Link: https://lore.kernel.org/r/20241121001334.331334-1-namhyung@kernel.org Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Same as
https://lore.kernel.org/all/20240514180050.182454-1-namhyung@kernel.org/,
we should skip `for_each_sibling_event()` for group leader since it
doesn't have the ctx yet.
Fixes: f3c0eba28704 ("perf: Add a few assertions") Reported-by: Greg Thelen <gthelen@google.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Robin Murphy <robin.murphy@arm.com> Cc: Tuan Phan <tuanphan@os.amperecomputing.com> Signed-off-by: Chun-Tse Shao <ctshao@google.com> Acked-by: Will Deacon <will@kernel.org> Link: https://lore.kernel.org/r/20241108050806.3730811-1-ctshao@google.com Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Large amount of mount hangs observed during hotplugging of 9pfs devices. The
9pfs Xen driver attempts to initialize itself more than once, causing the
frontend and backend to disagree: the backend listens on a channel that the
frontend does not send on, resulting in stalled processing.
The nvme keep-alive operation, which executes at a periodic interval,
could potentially sneak in while shutting down a fabric controller.
This may lead to a race between the fabric controller admin queue
destroy code path (invoked while shutting down controller) and hw/hctx
queue dispatcher called from the nvme keep-alive async request queuing
operation. This race could lead to the kernel crash shown below:
While shutting down fabric controller, if nvme keep-alive request sneaks
in then it would be flushed off. The nvme_keep_alive_end_io function is
then invoked to handle the end of the keep-alive operation which
decrements the admin->q_usage_counter and assuming this is the last/only
request in the admin queue then the admin->q_usage_counter becomes zero.
If that happens then blk-mq destroy queue operation (blk_mq_destroy_
queue()) which could be potentially running simultaneously on another
cpu (as this is the controller shutdown code path) would forward
progress and deletes the admin queue. So, now from this point onward
we are not supposed to access the admin queue resources. However the
issue here's that the nvme keep-alive thread running hw/hctx queue
dispatch operation hasn't yet finished its work and so it could still
potentially access the admin queue resource while the admin queue had
been already deleted and that causes the above crash.
The above kernel crash is regression caused due to changes implemented
in commit a54a93d0e359 ("nvme: move stopping keep-alive into
nvme_uninit_ctrl()"). Ideally we should stop keep-alive before destroyin
g the admin queue and freeing the admin tagset so that it wouldn't sneak
in during the shutdown operation. However we removed the keep alive stop
operation from the beginning of the controller shutdown code path in commit a54a93d0e359 ("nvme: move stopping keep-alive into nvme_uninit_ctrl()")
and added it under nvme_uninit_ctrl() which executes very late in the
shutdown code path after the admin queue is destroyed and its tagset is
removed. So this change created the possibility of keep-alive sneaking in
and interfering with the shutdown operation and causing observed kernel
crash.
To fix the observed crash, we decided to move nvme_stop_keep_alive() from
nvme_uninit_ctrl() to nvme_remove_admin_tag_set(). This change would ensure
that we don't forward progress and delete the admin queue until the keep-
alive operation is finished (if it's in-flight) or cancelled and that would
help contain the race condition explained above and hence avoid the crash.
Moving nvme_stop_keep_alive() to nvme_remove_admin_tag_set() instead of
adding nvme_stop_keep_alive() to the beginning of the controller shutdown
code path in nvme_stop_ctrl(), as was the case earlier before commit a54a93d0e359 ("nvme: move stopping keep-alive into nvme_uninit_ctrl()"),
would help save one callsite of nvme_stop_keep_alive().
Fixes: a54a93d0e359 ("nvme: move stopping keep-alive into nvme_uninit_ctrl()") Link: https://lore.kernel.org/all/1a21f37b-0f2a-4745-8c56-4dc8628d3983@linux.ibm.com/ Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Nilay Shroff <nilay@linux.ibm.com> Signed-off-by: Keith Busch <kbusch@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
blk_mq_clear_flush_rq_mapping() is not called during scsi probe, by
checking blk_queue_init_done(). However, QUEUE_FLAG_INIT_DONE is cleared
in del_gendisk by commit aec89dc5d421 ("block: keep q_usage_counter in
atomic mode after del_gendisk"), hence for disk like scsi, following
blk_mq_destroy_queue() will not clear flush rq from tags->rqs[] as well,
cause following uaf that is found by our syzkaller for v6.6:
==================================================================
BUG: KASAN: slab-use-after-free in blk_mq_find_and_get_req+0x16e/0x1a0 block/blk-mq-tag.c:261
Read of size 4 at addr ffff88811c969c20 by task kworker/1:2H/224909
Other than blk_mq_clear_flush_rq_mapping(), the flag is only used in
blk_register_queue() from initialization path, hence it's safe not to
clear the flag in del_gendisk. And since QUEUE_FLAG_REGISTERED already
make sure that queue should only be registered once, there is no need
to test the flag as well.
Fixes: 6cfeadbff3f8 ("blk-mq: don't clear flush_rq from tags->rqs[]")
Depends-on: commit aec89dc5d421 ("block: keep q_usage_counter in atomic mode after del_gendisk") Signed-off-by: Yu Kuai <yukuai3@huawei.com> Reviewed-by: Ming Lei <ming.lei@redhat.com> Link: https://lore.kernel.org/r/20241104110005.1412161-1-yukuai1@huaweicloud.com Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Sasha Levin <sashal@kernel.org>
Recently we got several deadlock report[1][2][3] caused by
blk_mq_freeze_queue and blk_enter_queue().
Turns out the two are just like acquiring read/write lock, so model them
as read/write lock for supporting lockdep:
1) model q->q_usage_counter as two locks(io and queue lock)
- queue lock covers sync with blk_enter_queue()
- io lock covers sync with bio_enter_queue()
2) make the lockdep class/key as per-queue:
- different subsystem has very different lock use pattern, shared lock
class causes false positive easily
- freeze_queue degrades to no lock in case that disk state becomes DEAD
because bio_enter_queue() won't be blocked any more
- freeze_queue degrades to no lock in case that request queue becomes dying
because blk_enter_queue() won't be blocked any more
3) model blk_mq_freeze_queue() as acquire_exclusive & try_lock
- it is exclusive lock, so dependency with blk_enter_queue() is covered
- it is trylock because blk_mq_freeze_queue() are allowed to run
concurrently
4) model blk_enter_queue() & bio_enter_queue() as acquire_read()
- nested blk_enter_queue() are allowed
- dependency with blk_mq_freeze_queue() is covered
- blk_queue_exit() is often called from other contexts(such as irq), and
it can't be annotated as lock_release(), so simply do it in
blk_enter_queue(), this way still covered cases as many as possible
With lockdep support, such kind of reports may be reported asap and
needn't wait until the real deadlock is triggered.
For example, lockdep report can be triggered in the report[3] with this
patch applied.
[1] occasional block layer hang when setting 'echo noop > /sys/block/sda/queue/scheduler'
https://bugzilla.kernel.org/show_bug.cgi?id=219166
[2] del_gendisk() vs blk_queue_enter() race condition
https://lore.kernel.org/linux-block/20241003085610.GK11458@google.com/
[3] queue_freeze & queue_enter deadlock in scsi
https://lore.kernel.org/linux-block/ZxG38G9BuFdBpBHZ@fedora/T/#u
Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Link: https://lore.kernel.org/r/20241025003722.3630252-4-ming.lei@redhat.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
Stable-dep-of: 3802f73bd807 ("block: fix uaf for flush rq while iterating tags") Signed-off-by: Sasha Levin <sashal@kernel.org>
Add non_owner variant of start_freeze/unfreeze queue APIs, so that the
caller knows that what they are doing, and we can skip lockdep support
for non_owner variant in per-call level.
Prepare for supporting lockdep for freezing/unfreezing queue.
Reviewed-by: Christoph Hellwig <hch@lst.de> Suggested-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Link: https://lore.kernel.org/r/20241025003722.3630252-2-ming.lei@redhat.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
Stable-dep-of: 3802f73bd807 ("block: fix uaf for flush rq while iterating tags") Signed-off-by: Sasha Levin <sashal@kernel.org>
The code currently uses list_for_each_entry_rcu() while holding an SRCU
lock, triggering false positive warnings with CONFIG_PROVE_RCU=y
enabled:
drivers/nvme/host/multipath.c:168 RCU-list traversed in non-reader section!!
drivers/nvme/host/multipath.c:227 RCU-list traversed in non-reader section!!
drivers/nvme/host/multipath.c:260 RCU-list traversed in non-reader section!!
While the list is properly protected by SRCU lock, the code uses the
wrong list traversal primitive. Replace list_for_each_entry_rcu() with
list_for_each_entry_srcu() to correctly indicate SRCU-based protection
and eliminate the false warning.
Signed-off-by: Breno Leitao <leitao@debian.org> Fixes: be647e2c76b2 ("nvme: use srcu for iterating namespace list") Signed-off-by: Keith Busch <kbusch@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
During repetitive namespace remapping operations on the target the
namespace might have changed between the time the initial scan
was performed, and partition scan was invoked by device_add_disk()
in nvme_mpath_set_live(). We then end up with a stuck scanning process:
This happens when we have several paths, some of which are inaccessible,
and the active paths are removed first. Then nvme_find_path() will requeue
I/O in the ns_head (as paths are present), but the requeue list is never
triggered as all remaining paths are inactive.
This patch checks for NVME_NSHEAD_DISK_LIVE in nvme_available_path(),
and requeue I/O after NVME_NSHEAD_DISK_LIVE has been cleared once
the last path has been removed to properly terminate pending I/O.
Signed-off-by: Hannes Reinecke <hare@kernel.org> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Keith Busch <kbusch@kernel.org>
Stable-dep-of: 5dd18f09ce73 ("nvme/multipath: Fix RCU list traversal to use SRCU primitive") Signed-off-by: Sasha Levin <sashal@kernel.org>
This patch appears to assume that if one request is complete, then the
others will complete too before unlocking. That is not a valid
assumption, since other requests could hit a non-fatal error or a short
write that would cause them not to complete.
tm_mon describes months from 0 to 11, but the register contains BCD from
1 to 12. tm_year contains years since 1900, but the BCD contains 20XX.
Apply the offsets when converting these numbers.
When building the kernel with -Wmaybe-uninitialized, the compiler
reports this warning:
In function 'jffs2_mark_erased_block',
inlined from 'jffs2_erase_pending_blocks' at fs/jffs2/erase.c:116:4:
fs/jffs2/erase.c:474:9: warning: 'bad_offset' may be used uninitialized [-Wmaybe-uninitialized]
474 | jffs2_erase_failed(c, jeb, bad_offset);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
fs/jffs2/erase.c: In function 'jffs2_erase_pending_blocks':
fs/jffs2/erase.c:402:18: note: 'bad_offset' was declared here
402 | uint32_t bad_offset;
| ^~~~~~~~~~
When mtd->point() is used, jffs2_erase_pending_blocks can return -EIO
without initializing bad_offset, which is later used at the filebad
label in jffs2_mark_erased_block.
Fix it by initializing this variable.
Fixes: 8a0f572397ca ("[JFFS2] Return values of jffs2_block_check_erase error paths") Signed-off-by: Qingfang Deng <qingfang.deng@siflower.com.cn> Reviewed-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: Richard Weinberger <richard@nod.at> Signed-off-by: Sasha Levin <sashal@kernel.org>
After an insertion in TNC, the tree might split and cause a node to
change its `znode->parent`. A further deletion of other nodes in the
tree (which also could free the nodes), the aforementioned node's
`znode->cparent` could still point to a freed node. This
`znode->cparent` may not be updated when getting nodes to commit in
`ubifs_tnc_start_commit()`. This could then trigger a use-after-free
when accessing the `znode->cparent` in `write_index()` in
`ubifs_tnc_end_commit()`.
The offending `memcpy()` in `ubifs_copy_hash()` has a use-after-free
when a node becomes root in TNC but still has a `cparent` to an already
freed node. More specifically, consider the following TNC:
zroot
/
/
zp1
/
/
zn
Inserting a new node `zn_new` with a key smaller then `zn` will trigger
a split in `tnc_insert()` if `zp1` is full:
zroot
/ \
/ \
zp1 zp2
/ \
/ \
zn_new zn
`zn->parent` has now been moved to `zp2`, *but* `zn->cparent` still
points to `zp1`.
Now, consider a removal of all the nodes _except_ `zn`. Just when
`tnc_delete()` is about to delete `zroot` and `zp2`:
zroot
\
\
zp2
\
\
zn
`zroot` and `zp2` get freed and the tree collapses:
zn
`zn` now becomes the new `zroot`.
`get_znodes_to_commit()` will now only find `zn`, the new `zroot`, and
`write_index()` will check its `znode->cparent` that wrongly points to
the already freed `zp1`. `ubifs_copy_hash()` thus gets wrongly called
with `znode->cparent->zbranch[znode->iip].hash` that triggers the
use-after-free!
Fix this by explicitly setting `znode->cparent` to `NULL` in
`get_znodes_to_commit()` for the root node. The search for the dirty
nodes is bottom-up in the tree. Thus, when `find_next_dirty(znode)`
returns NULL, the current `znode` _is_ the root node. Add an assert for
this.
Since commit 4c39529663b9 ("slab: Warn on duplicate cache names when
DEBUG_VM=y"), the duplicate slab cache names can be detected and a
kernel WARNING is thrown out.
In UBI fast attaching process, alloc_ai() could be invoked twice
with the same slab cache name 'ubi_aeb_slab_cache', which will trigger
following warning messages:
kmem_cache of name 'ubi_aeb_slab_cache' already exists
WARNING: CPU: 0 PID: 7519 at mm/slab_common.c:107
__kmem_cache_create_args+0x100/0x5f0
Modules linked in: ubi(+) nandsim [last unloaded: nandsim]
CPU: 0 UID: 0 PID: 7519 Comm: modprobe Tainted: G 6.12.0-rc2
RIP: 0010:__kmem_cache_create_args+0x100/0x5f0
Call Trace:
__kmem_cache_create_args+0x100/0x5f0
alloc_ai+0x295/0x3f0 [ubi]
ubi_attach+0x3c3/0xcc0 [ubi]
ubi_attach_mtd_dev+0x17cf/0x3fa0 [ubi]
ubi_init+0x3fb/0x800 [ubi]
do_init_module+0x265/0x7d0
__x64_sys_finit_module+0x7a/0xc0
The problem could be easily reproduced by loading UBI device by fastmap
with CONFIG_DEBUG_VM=y.
Fix it by using different slab names for alloc_ai() callers.
Fixes: d2158f69a7d4 ("UBI: Remove alloc_ai() slab name from parameter list") Fixes: fdf10ed710c0 ("ubi: Rework Fastmap attach base code") Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: Richard Weinberger <richard@nod.at> Signed-off-by: Sasha Levin <sashal@kernel.org>
Since commit e874dcde1cbf ("ubifs: Reserve one leb for each journal
head while doing budget"), available space is calulated by deducting
reservation for all journal heads. However, the total block count (
which is only used by statfs) is not updated yet, which will cause
the wrong displaying for used space(total - available).
Fix it by deducting reservation for all journal heads from total
block count.
Fixes: e874dcde1cbf ("ubifs: Reserve one leb for each journal head while doing budget") Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: Richard Weinberger <richard@nod.at> Signed-off-by: Sasha Levin <sashal@kernel.org>
Since commit 14072ee33d5a ("ubi: fastmap: Check wl_pool for free peb
before wear leveling"), wear_leveling_worker() won't schedule fm_work
if wear-leveling pool is empty, which could temporarily disable the
wear-leveling until the fastmap is updated(eg. pool becomes empty).
Fix it by scheduling fm_work if wl_pool is empty during wear-leveing.
Fixes: 14072ee33d5a ("ubi: fastmap: Check wl_pool for free peb before wear leveling") Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: Richard Weinberger <richard@nod.at> Signed-off-by: Sasha Levin <sashal@kernel.org>
If the __rtc_read_time call fails,, the struct rtc_time tm; may contain
uninitialized data, or an illegal date/time read from the RTC hardware.
When calling rtc_tm_to_ktime later, the result may be a very large value
(possibly KTIME_MAX). If there are periodic timers in rtc->timerqueue,
they will continually expire, may causing kernel softlockup.
Fixes: 6610e0893b8b ("RTC: Rework RTC code to use timerqueue for events") Signed-off-by: Yongliang Gao <leonylgao@tencent.com> Acked-by: Jingqun Li <jingqunli@tencent.com> Link: https://lore.kernel.org/r/20241011043153.3788112-1-leonylgao@gmail.com Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
If request_irq() fails in st_rtc_probe(), there is no need to enable
the irq, and if it succeeds, disable_irq() after request_irq() still has
a time gap in which interrupts can come.
request_irq() with IRQF_NO_AUTOEN flag will disable IRQ auto-enable when
request IRQ.
Yang Erkun reports that when two threads are opening files at the same
time, and are forced to abort before a reply is seen, then the call to
nfs_release_seqid() in nfs4_opendata_free() can result in a
use-after-free of the pointer to the defunct rpc task of the other
thread.
The fix is to ensure that if the RPC call is aborted before the call to
nfs_wait_on_sequence() is complete, then we must call nfs_release_seqid()
in nfs4_open_release() before the rpc_task is freed.
Reported-by: Yang Erkun <yangerkun@huawei.com> Fixes: 24ac23ab88df ("NFSv4: Convert open() into an asynchronous RPC call") Reviewed-by: Yang Erkun <yangerkun@huawei.com> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Currently, show_stack() always dumps the trace of the current task.
However, it should dump the trace of the specified task if one is
provided. Otherwise, things like running "echo t > sysrq-trigger"
won't work as expected.
Fixes: 970e51feaddb ("um: Add support for CONFIG_STACKTRACE") Signed-off-by: Tiwei Bie <tiwei.btw@antgroup.com> Link: https://patch.msgid.link/20241106103933.1132365-1-tiwei.btw@antgroup.com Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Currently, the initialization of the disk pointer in the ubd structure
is missing. It should be initialized with the allocated gendisk pointer
in ubd_add().
Fixes: 32621ad7a7ea ("ubd: remove the ubd_gendisk array") Signed-off-by: Tiwei Bie <tiwei.btw@antgroup.com> Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com> Link: https://patch.msgid.link/20241104163203.435515-2-tiwei.btw@antgroup.com Signed-off-by: Johannes Berg <johannes.berg@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Nothing in kfifo.h directly needs dma-mapping.h, only two macros
use DMA_MAPPING_ERROR when actually instantiated. Drop the
dma-mapping.h include to reduce include bloat.
Add an explicity <linux/io.h> include to drivers/mailbox/omap-mailbox.c
as that file uses __raw_readl and __raw_writel through a complicated
include chain involving <linux/dma-mapping.h>
The function `c_show` was called with protection from RCU. This only
ensures that `cp` will not be freed. Therefore, the reference count for
`cp` can drop to zero, which will trigger a refcount use-after-free
warning when `cache_get` is called. To resolve this issue, use
`cache_get_rcu` to ensure that `cp` remains active.
------------[ cut here ]------------
refcount_t: addition on 0; use-after-free.
WARNING: CPU: 7 PID: 822 at lib/refcount.c:25
refcount_warn_saturate+0xb1/0x120
CPU: 7 UID: 0 PID: 822 Comm: cat Not tainted 6.12.0-rc3+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
1.16.1-2.fc37 04/01/2014
RIP: 0010:refcount_warn_saturate+0xb1/0x120
If the tag length is >= U32_MAX - 3 then the "length + 4" addition
can result in an integer overflow. Address this by splitting the
decoding into several steps so that decode_cb_compound4res() does
not have to perform arithmetic on the unsafe length value.
Reported-by: Dan Carpenter <dan.carpenter@linaro.org> Cc: stable@vger.kernel.org Reviewed-by: Jeff Layton <jlayton@kernel.org> Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In cs40l50_add(), the work_data is a local variable and the work_data.work
should initialize with INIT_WORK_ONSTACK() instead of INIT_WORK().
Small error in cs40l50_erase() also fixed in this commit.
Fixes: c38fe1bb5d21 ("Input: cs40l50 - Add support for the CS40L50 haptic driver") Signed-off-by: Yuan Can <yuancan@huawei.com> Reviewed-by: James Ogletree <jogletre@opensource.cirrus.com> Link: https://lore.kernel.org/r/20241106013549.78142-1-yuancan@huawei.com Cc: stable@vger.kernel.org Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Percpu memory allocation may failed during create_ipc_ns however this
fail is not handled properly since ipc sysctls and mq sysctls is not
released properly. Fix this by release these two resource when failure.
The root cause is: on a fuzzed image, blkaddr in nat entry may be
corrupted, then it will cause system panic when using it in
f2fs_invalidate_blocks(), to avoid this, let's add sanity check on
nat blkaddr in truncate_node().
The output of ".%03u" with the unsigned int in range [0, 4294966295] may
get truncated if the target buffer is not 12 bytes. This can't really
happen here as the 'remainder' variable cannot exceed 999 but the
compiler doesn't know it. To make it happy just increase the buffer to
where the warning goes away.
Fixes: 3c9f3681d0b4 ("[SCSI] lib: add generic helper to print sizes rounded to the correct SI range") Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org> Reviewed-by: Andy Shevchenko <andy@kernel.org> Cc: James E.J. Bottomley <James.Bottomley@HansenPartnership.com> Cc: Kees Cook <kees@kernel.org> Cc: stable@vger.kernel.org Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Link: https://lore.kernel.org/r/20241101205453.9353-1-brgl@bgdev.pl Signed-off-by: Kees Cook <kees@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Device nodes accessed via of_get_compatible_child() require
of_node_put() to be called when the node is no longer required to avoid
leaving a reference to the node behind, leaking the resource.
In this case, the usage of 'tnode' is straightforward and there are no
error paths, allowing for a single of_node_put() when 'tnode' is no
longer required.
Cc: stable@vger.kernel.org Fixes: 29646ee33cc3 ("counter: stm32-timer-cnt: add checks on quadrature encoder capability") Signed-off-by: Javier Carrasco <javier.carrasco.cruz@gmail.com> Link: https://lore.kernel.org/r/20241027-stm32-timer-cnt-of_node_put-v1-1-ebd903cdf7ac@gmail.com Signed-off-by: William Breathitt Gray <wbg@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
An error path was introduced without including the required call to
of_node_put() to decrement the node's refcount and avoid leaking memory.
If the call to kzalloc() for 'mgmt' fails, the probe returns without
decrementing the refcount.
Use the automatic cleanup facility to fix the bug and protect the code
against new error paths where the call to of_node_put() might be missing
again.
The dwc3_request->num_queued_sgs is decremented on completion. If a
partially completed request is handled, then the
dwc3_request->num_queued_sgs no longer reflects the total number of
num_queued_sgs (it would be cleared).
Correctly check the number of request SG entries remained to be prepare
and queued. Failure to do this may cause null pointer dereference when
accessing non-existent SG entry.
The check whether the TRB ring is full or empty in dwc3_calc_trbs_left()
is insufficient. It assumes there are active TRBs if there's any request
in the started_list. However, that's not the case for requests with a
large SG list.
That is, if we have a single usb request that requires more TRBs than
the total TRBs in the TRB ring, the queued TRBs will be available when
all the TRBs in the ring are completed. But the request is only
partially completed and remains in the started_list. With the current
logic, the TRB ring is empty, but dwc3_calc_trbs_left() returns 0.
Fix this by additionally checking for the request->num_trbs for active
TRB count.
There is a possibility that a request's callback could be invoked from
usb_ep_queue() (call trace below, supplemented with missing calls):
req->complete from usb_gadget_giveback_request
(drivers/usb/gadget/udc/core.c:999)
usb_gadget_giveback_request from musb_g_giveback
(drivers/usb/musb/musb_gadget.c:147)
musb_g_giveback from rxstate
(drivers/usb/musb/musb_gadget.c:784)
rxstate from musb_ep_restart
(drivers/usb/musb/musb_gadget.c:1169)
musb_ep_restart from musb_ep_restart_resume_work
(drivers/usb/musb/musb_gadget.c:1176)
musb_ep_restart_resume_work from musb_queue_resume_work
(drivers/usb/musb/musb_core.c:2279)
musb_queue_resume_work from musb_gadget_queue
(drivers/usb/musb/musb_gadget.c:1241)
musb_gadget_queue from usb_ep_queue
(drivers/usb/gadget/udc/core.c:300)
According to the docstring of usb_ep_queue(), this should not happen:
"Note that @req's ->complete() callback must never be called from within
usb_ep_queue() as that can create deadlock situations."
In fact, a hardware lockup might occur in the following sequence:
1. The gadget is initialized using musb_gadget_enable().
2. Meanwhile, a packet arrives, and the RXPKTRDY flag is set, raising an
interrupt.
3. If IRQs are enabled, the interrupt is handled, but musb_g_rx() finds an
empty queue (next_request() returns NULL). The interrupt flag has
already been cleared by the glue layer handler, but the RXPKTRDY flag
remains set.
4. The first request is enqueued using usb_ep_queue(), leading to the call
of req->complete(), as shown in the call trace above.
5. If the callback enables IRQs and another packet is waiting, step (3)
repeats. The request queue is empty because usb_g_giveback() removes the
request before invoking the callback.
6. The endpoint remains locked up, as the interrupt triggered by hardware
setting the RXPKTRDY flag has been handled, but the flag itself remains
set.
For this scenario to occur, it is only necessary for IRQs to be enabled at
some point during the complete callback. This happens with the USB Ethernet
gadget, whose rx_complete() callback calls netif_rx(). If called in the
task context, netif_rx() disables the bottom halves (BHs). When the BHs are
re-enabled, IRQs are also enabled to allow soft IRQs to be processed. The
gadget itself is initialized at module load (or at boot if built-in), but
the first request is enqueued when the network interface is brought up,
triggering rx_complete() in the task context via ioctl(). If a packet
arrives while the interface is down, it can prevent the interface from
receiving any further packets from the USB host.
The situation is quite complicated with many parties involved. This
particular issue can be resolved in several possible ways:
1. Ensure that callbacks never enable IRQs. This would be difficult to
enforce, as discovering how netif_rx() interacts with interrupts was
already quite challenging and u_ether is not the only function driver.
Similar "bugs" could be hidden in other drivers as well.
2. Disable MUSB interrupts in musb_g_giveback() before calling the callback
and re-enable them afterwars (by calling musb_{dis,en}able_interrupts(),
for example). This would ensure that MUSB interrupts are not handled
during the callback, even if IRQs are enabled. In fact, it would allow
IRQs to be enabled when releasing the lock. However, this feels like an
inelegant hack.
3. Modify the interrupt handler to clear the RXPKTRDY flag if the request
queue is empty. While this approach also feels like a hack, it wastes
CPU time by attempting to handle incoming packets when the software is
not ready to process them.
4. Flush the Rx FIFO instead of calling rxstate() in musb_ep_restart().
This ensures that the hardware can receive packets when there is at
least one request in the queue. Once IRQs are enabled, the interrupt
handler will be able to correctly process the next incoming packet
(eventually calling rxstate()). This approach may cause one or two
packets to be dropped (two if double buffering is enabled), but this
seems to be a minor issue, as packet loss can occur when the software is
not yet ready to process them. Additionally, this solution makes the
gadget driver compliant with the rule mentioned in the docstring of
usb_ep_queue().
There may be additional solutions, but from these four, the last one has
been chosen as it seems to be the most appropriate, as it addresses the
"bad" behavior of the driver.
Fixes: baebdf48c360 ("net: dev: Makes sure netif_rx() can be invoked in any context.") Cc: stable@vger.kernel.org Signed-off-by: Hubert Wiśniewski <hubert.wisniewski.25632@gmail.com> Link: https://lore.kernel.org/r/4ee1ead4525f78fb5909a8cbf99513ad0082ad21.camel@gmail.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The driver cannot issue the End Transfer command to the SETUP transfer.
Don't clear DWC3_EP_TRANSFER_STARTED flag to make sure that the driver
won't send Start Transfer command again, which can cause no-resource
error. For example this can occur if the host issues a reset to the
device.
On some Lenovo platforms, the patch works around problems with ov2740
sensor initialization, which manifest themself like below:
[ 4.540476] ov2740 i2c-INT3474:01: error -EIO: failed to find sensor
[ 4.542066] ov2740 i2c-INT3474:01: probe with driver ov2740 failed with error -5
or
[ 7.742633] ov2740 i2c-INT3474:01: chip id mismatch: 2740 != 0
[ 7.742638] ov2740 i2c-INT3474:01: error -ENXIO: failed to find sensor
and also by random failures of video stream start.
Issue can be reproduced by this script:
n=0
k=0
while [ $n -lt 50 ] ; do
sudo modprobe -r ov2740
sleep `expr $RANDOM % 5`
sudo modprobe ov2740
if media-ctl -p | grep -q ov2740 ; then
let k++
fi
let n++
done
echo Success rate $k/$n
Without the patch, success rate is approximately 15 or 50 tries.
With the patch it does not fail.
This problem is some hardware or firmware malfunction, that can not be
easy debug and fix. While setting small autosuspend delay is not perfect
workaround as user can configure it to any value, it will prevent
the failures by default.
Additionally setting small autosuspend delay should have positive effect
on power consumption as for most ljca workloads device is used for just
a few milliseconds flowed by long periods of at least 100ms of inactivity
(usually more).
Fixes: acd6199f195d ("usb: Add support for Intel LJCA device") Cc: stable@vger.kernel.org Reviewed-by: Hans de Goede <hdegoede@redhat.com> Tested-by: Hans de Goede <hdegoede@redhat.com> # ThinkPad X1 Yoga Gen 8, ov2740 Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com> Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com> Link: https://lore.kernel.org/r/20241112075514.680712-2-stanislaw.gruszka@linux.intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The unmount process (cifs_kill_sb() calling close_all_cached_dirs()) can
race with various cached directory operations, which ultimately results
in dentries not being dropped and these kernel BUGs:
BUG: Dentry ffff88814f37e358{i=1000000000080,n=/} still in use (2) [unmount of cifs cifs]
VFS: Busy inodes after unmount of cifs (cifs)
------------[ cut here ]------------
kernel BUG at fs/super.c:661!
This happens when a cfid is in the process of being cleaned up when, and
has been removed from the cfids->entries list, including:
- Receiving a lease break from the server
- Server reconnection triggers invalidate_all_cached_dirs(), which
removes all the cfids from the list
- The laundromat thread decides to expire an old cfid.
To solve these problems, dropping the dentry is done in queued work done
in a newly-added cfid_put_wq workqueue, and close_all_cached_dirs()
flushes that workqueue after it drops all the dentries of which it's
aware. This is a global workqueue (rather than scoped to a mount), but
the queued work is minimal.
The final cleanup work for cleaning up a cfid is performed via work
queued in the serverclose_wq workqueue; this is done separate from
dropping the dentries so that close_all_cached_dirs() doesn't block on
any server operations.
Both of these queued works expect to invoked with a cfid reference and
a tcon reference to avoid those objects from being freed while the work
is ongoing.
While we're here, add proper locking to close_all_cached_dirs(), and
locking around the freeing of cfid->dentry.
Fixes: ebe98f1447bb ("cifs: enable caching of directories for which a lease is held") Cc: stable@vger.kernel.org Signed-off-by: Paul Aurich <paul@darkrain42.org> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If open_cached_dir() encounters an error parsing the lease from the
server, the error handling may race with receiving a lease break,
resulting in open_cached_dir() freeing the cfid while the queued work is
pending.
Update open_cached_dir() to drop refs rather than directly freeing the
cfid.
Have cached_dir_lease_break(), cfids_laundromat_worker(), and
invalidate_all_cached_dirs() clear has_lease immediately while still
holding cfids->cfid_list_lock, and then use this to also simplify the
reference counting in cfids_laundromat_worker() and
invalidate_all_cached_dirs().
Fixes this KASAN splat (which manually injects an error and lease break
in open_cached_dir()):
==================================================================
BUG: KASAN: slab-use-after-free in smb2_cached_lease_break+0x27/0xb0
Read of size 8 at addr ffff88811cc24c10 by task kworker/3:1/65
Last potentially related work creation:
kasan_save_stack+0x33/0x60
__kasan_record_aux_stack+0xad/0xc0
insert_work+0x32/0x100
__queue_work+0x5c9/0x870
queue_work_on+0x82/0x90
open_cached_dir+0x1369/0x1fb0
smb2_query_path_info+0x43c/0x6e0
cifs_get_fattr+0x346/0xf10
cifs_get_inode_info+0x157/0x210
cifs_revalidate_dentry_attr+0x2d1/0x460
cifs_getattr+0x173/0x470
vfs_statx_path+0x10f/0x160
vfs_statx+0xe9/0x150
vfs_fstatat+0x5e/0xc0
__do_sys_newfstatat+0x91/0xf0
do_syscall_64+0x95/0x1a0
entry_SYSCALL_64_after_hwframe+0x76/0x7e
The buggy address belongs to the object at ffff88811cc24c00
which belongs to the cache kmalloc-1k of size 1024
The buggy address is located 16 bytes inside of
freed 1024-byte region [ffff88811cc24c00, ffff88811cc25000)
Cc: stable@vger.kernel.org Signed-off-by: Paul Aurich <paul@darkrain42.org> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
open_cached_dir() may either race with the tcon reconnection even before
compound_send_recv() or directly trigger a reconnection via
SMB2_open_init() or SMB_query_info_init().
The reconnection process invokes invalidate_all_cached_dirs() via
cifs_mark_open_files_invalid(), which removes all cfids from the
cfids->entries list but doesn't drop a ref if has_lease isn't true. This
results in the currently-being-constructed cfid not being on the list,
but still having a refcount of 2. It leaks if returned from
open_cached_dir().
Fix this by setting cfid->has_lease when the ref is actually taken; the
cfid will not be used by other threads until it has a valid time.
Addresses these kmemleaks:
unreferenced object 0xffff8881090c4000 (size 1024):
comm "bash", pid 1860, jiffies 4295126592
hex dump (first 32 bytes):
00 01 00 00 00 00 ad de 22 01 00 00 00 00 ad de ........".......
00 ca 45 22 81 88 ff ff f8 dc 4f 04 81 88 ff ff ..E"......O.....
backtrace (crc 6f58c20f):
[<ffffffff8b895a1e>] __kmalloc_cache_noprof+0x2be/0x350
[<ffffffff8bda06e3>] open_cached_dir+0x993/0x1fb0
[<ffffffff8bdaa750>] cifs_readdir+0x15a0/0x1d50
[<ffffffff8b9a853f>] iterate_dir+0x28f/0x4b0
[<ffffffff8b9a9aed>] __x64_sys_getdents64+0xfd/0x200
[<ffffffff8cf6da05>] do_syscall_64+0x95/0x1a0
[<ffffffff8d00012f>] entry_SYSCALL_64_after_hwframe+0x76/0x7e
unreferenced object 0xffff8881044fdcf8 (size 8):
comm "bash", pid 1860, jiffies 4295126592
hex dump (first 8 bytes):
00 cc cc cc cc cc cc cc ........
backtrace (crc 10c106a9):
[<ffffffff8b89a3d3>] __kmalloc_node_track_caller_noprof+0x363/0x480
[<ffffffff8b7d7256>] kstrdup+0x36/0x60
[<ffffffff8bda0700>] open_cached_dir+0x9b0/0x1fb0
[<ffffffff8bdaa750>] cifs_readdir+0x15a0/0x1d50
[<ffffffff8b9a853f>] iterate_dir+0x28f/0x4b0
[<ffffffff8b9a9aed>] __x64_sys_getdents64+0xfd/0x200
[<ffffffff8cf6da05>] do_syscall_64+0x95/0x1a0
[<ffffffff8d00012f>] entry_SYSCALL_64_after_hwframe+0x76/0x7e
And addresses these BUG splats when unmounting the SMB filesystem:
This reproduces eventually with an SMB mount and two shells running
these loops concurrently
- while true; do
cd ~; sleep 1;
for i in {1..3}; do cd /mnt/test/subdir;
echo $PWD; sleep 1; cd ..; echo $PWD; sleep 1;
done;
echo ...;
done
- while true; do
iptables -F OUTPUT; mount -t cifs -a;
for _ in {0..2}; do ls /mnt/test/subdir/ | wc -l; done;
iptables -I OUTPUT -p tcp --dport 445 -j DROP;
sleep 10
echo "unmounting"; umount -l -t cifs -a; echo "done unmounting";
sleep 20
echo "recovering"; iptables -F OUTPUT;
sleep 10;
done
Fixes: ebe98f1447bb ("cifs: enable caching of directories for which a lease is held") Fixes: 5c86919455c1 ("smb: client: fix use-after-free in smb2_query_info_compound()") Cc: stable@vger.kernel.org Signed-off-by: Paul Aurich <paul@darkrain42.org> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(1) Windows Server will fail FSCTL_SET_REPARSE_POINT with
STATUS_IO_REPARSE_DATA_INVALID when input buffer is larger than
16K, as specified in MS-FSA 2.1.5.10.37.
(2) The client won't be able to parse large SMB responses that
includes SMB symlink path within SMB2_CREATE or SMB2_IOCTL
responses.
Fix this by defining a maximum length value (4060) for SMB symlinks
that both client and server can handle.
Cc: David Howells <dhowells@redhat.com> Cc: stable@vger.kernel.org Signed-off-by: Paulo Alcantara (Red Hat) <pc@manguebit.com> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This client was only requesting READ caching, not READ and HANDLE caching
in the LeaseState on the open requests we send for directories. To
delay closing a handle (e.g. for caching directory contents) we should
be requesting HANDLE as well as READ (as we already do for deferred
close of files). See MS-SMB2 3.3.1.4 e.g.
Cc: stable@vger.kernel.org Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Internal Speaker of Infinix Y4 Max remains muted due to incorrect
Pin configuration, and the Internal Mic records high noise. This patch
corrects the Pin configuration for the Internal Speaker and limits
the Internal Mic boost.
HW Probe for device: https://linux-hardware.org/?probe=6d4386c347
Test: Internal Speaker works fine, Mic has low noise.
A driver might allow the mmap access before initializing its
runtime->dma_area properly. Add a proper NULL check before passing to
virt_to_page() for avoiding a panic.
The m1.0 field of UMP Function Block info specifies whether the given
FB is a MIDI 1.0 port or not. When implementing the UMP support on
Linux, I somehow interpreted as if it were bit flags, but the field is
actually an enumeration from 0 to 2, where 2 means MIDI 1.0 *and* low
speed.
This patch corrects the interpretation and sets the right bit flags
depending on the m1.0 field of FB Info. This effectively fixes the
missing detection of MIDI 1.0 FB when m1.0 is 2.
Fixes: 37e0e14128e0 ("ALSA: ump: Support UMP Endpoint and Function Block parsing") Cc: <stable@vger.kernel.org> Link: https://patch.msgid.link/20241127070059.8099-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
At the conversion of locking with guard(), I overlooked that kvfree()
must not be called inside the spinlock unlike kfree(), and this was
caught by syzkaller now.
This patch reverts the conversion partially for restoring the kvfree()
call outside the spinlock. It's not trivial to use guard() in this
context, unfortunately.
Fixes: 84bb065b316e ("ALSA: rawmidi: Use guard() for locking") Reported-by: syzbot+351f8764833934c68836@syzkaller.appspotmail.com Reported-by: Eric Dumazet <eric.dumazet@gmail.com> Closes: https://lore.kernel.org/6744737b.050a0220.1cc393.007e.GAE@google.com Cc: <stable@vger.kernel.org> Link: https://patch.msgid.link/20241125142041.16578-1-tiwai@suse.de Signed-off-by: Takashi Iwai <tiwai@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The v4l2_detect_cvt/gtf functions should check the result against the
timing capabilities: these functions calculate the timings, so if they
are out of bounds, they should be rejected.
To do this, add the struct v4l2_dv_timings_cap as argument to those
functions.
This required updates to the adv7604 and adv7842 drivers since the
prototype of these functions has now changed. The timings struct
that is passed to v4l2_detect_cvt/gtf in those two drivers is filled
with the timings detected by the hardware.
The vivid driver was also updated, but an additional check was added:
the width and height specified by VIDIOC_S_DV_TIMINGS has to match the
calculated result, otherwise something went wrong. Note that vivid
*emulates* hardware, so all the values passed to the v4l2_detect_cvt/gtf
functions came from the timings struct that was filled by userspace
and passed on to the driver via VIDIOC_S_DV_TIMINGS. So these fields
can contain random data. Both the constraints check via
struct v4l2_dv_timings_cap and the additional width/height check
ensure that the resulting timings are sane and not messed up by the
v4l2_detect_cvt/gtf calculations.
Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl> Fixes: 2576415846bc ("[media] v4l2: move dv-timings related code to v4l2-dv-timings.c") Cc: stable@vger.kernel.org Reported-by: syzbot+a828133770f62293563e@syzkaller.appspotmail.com Closes: https://lore.kernel.org/linux-media/000000000000013050062127830a@google.com/ Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
of_find_compatible_node() requires a call to of_node_put() when the
pointer to the node is not required anymore to decrement its refcount
and avoid leaking memory.
Add the missing call to of_node_put() after the node has been used.
The Voltorb device uses a speaker codec different from the original
Corsola device. When the Voltorb device tree was first added, the new
codec was added as a separate node when it should have just replaced the
existing one.
Merge the two nodes. The only differences are the compatible string and
the GPIO line property name. This keeps the device node path for the
speaker codec the same across the MT8186 Chromebook line. Also rename
the related labels and node names from having rt1019p to speaker codec.
Cc: stable@vger.kernel.org # v6.11+ Signed-off-by: Chen-Yu Tsai <wenst@chromium.org> Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com> Link: https://lore.kernel.org/r/20241018082113.1297268-1-wenst@chromium.org Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Some IPU6 devices have shared interrupts. We need to handle properly
case when interrupt is triggered from other device on shared irq line
and IPU6 itself disabled. In such case we get 0xffffffff from
ISR_STATUS register and handle all irq's cases, for what we are not
not prepared and usually hang the whole system.
To avoid the issue use pm_runtime_get_if_active() to check if
the device is enabled and prevent suspending it when we handle irq
until the end of irq. Additionally use synchronize_irq() in suspend
Fixes: ab29a2478e70 ("media: intel/ipu6: add IPU6 buttress interface driver") Cc: stable@vger.kernel.org Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com> Reviewed-by: Hans de Goede <hdegoede@redhat.com> Tested-by: Hans de Goede <hdegoede@redhat.com> # ThinkPad X1 Yoga Gen 8, ov2740 Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com> Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Atomicity violation occurs when the fmc_send_cmd() function is executed
simultaneously with the modification of the fmdev->resp_skb value.
Consider a scenario where, after passing the validity check within the
function, a non-null fmdev->resp_skb variable is assigned a null value.
This results in an invalid fmdev->resp_skb variable passing the validity
check. As seen in the later part of the function, skb = fmdev->resp_skb;
when the invalid fmdev->resp_skb passes the check, a null pointer
dereference error may occur at line 478, evt_hdr = (void *)skb->data;
To address this issue, it is recommended to include the validity check of
fmdev->resp_skb within the locked section of the function. This
modification ensures that the value of fmdev->resp_skb does not change
during the validation process, thereby maintaining its validity.
This possible bug is found by an experimental static analysis tool
developed by our team. This tool analyzes the locking APIs
to extract function pairs that can be concurrently executed, and then
analyzes the instructions in the paired functions to identify possible
concurrency bugs including data races and atomicity violations.
Fixes: e8454ff7b9a4 ("[media] drivers:media:radio: wl128x: FM Driver Common sources") Cc: stable@vger.kernel.org Signed-off-by: Qiu-ji Chen <chenqiuji666@gmail.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
After assembling the new private flags on a PF, the operation to determine
the changed flags uses the wrong bitmaps. Instead of xor-ing orig_flags
with new_flags, it uses the still unchanged pf->flags, thus changed_flags
is always 0.
Fix it by using the correct bitmaps.
The issue was discovered while debugging why disabling source pruning
stopped working with release 6.7. Although the new flags will be copied to
pf->flags later on in that function, disabling source pruning requires
a reset of the PF, which was skipped due to this bug.
Disabling source pruning:
$ sudo ethtool --set-priv-flags eno1 disable-source-pruning on
$ sudo ethtool --show-priv-flags eno1
Private flags for eno1:
MFP : off
total-port-shutdown : off
LinkPolling : off
flow-director-atr : on
veb-stats : off
hw-atr-eviction : off
link-down-on-close : off
legacy-rx : off
disable-source-pruning: on
disable-fw-lldp : off
rs-fec : off
base-r-fec : off
vf-vlan-pruning : off
Regarding reproducing:
I observed the issue with a rather complicated lab setup, where
* two VLAN interfaces are created on eno1
* each with a different MAC address assigned
* each moved into a separate namespace
* both VLANs are bridged externally, so they form a single layer 2 network
The external bridge is done via a channel emulator adding packet loss and
delay and the application in the namespaces tries to send/receive traffic
and measure the performance. Sender and receiver are separated by
namespaces, yet the network card "sees its own traffic" send back to it.
To make that work, source pruning has to be disabled.
Cc: stable@vger.kernel.org Fixes: 70756d0a4727 ("i40e: Use DECLARE_BITMAP for flags and hw_features fields in i40e_pf") Signed-off-by: Peter Große <pegro@friiks.de> Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de> Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com> Tested-by: Pucha Himasekhar Reddy <himasekharx.reddy.pucha@intel.com> (A Contingent worker at Intel) Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com> Link: https://patch.msgid.link/20241113210705.1296408-1-anthony.l.nguyen@intel.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The tilt data contained in the Bluetooth packets of an Intuos Pro are
supposed to be interpreted as signed values. Simply casting the values
to type `char` is not guaranteed to work since it is implementation-
defined whether it is signed or unsigned. At least one user has noticed
the data being reported incorrectly on their system. To ensure that the
data is interpreted properly, we specifically cast to `signed char`
instead.
Link: https://github.com/linuxwacom/input-wacom/issues/445 Fixes: 4922cd26f03c ("HID: wacom: Support 2nd-gen Intuos Pro's Bluetooth classic interface") CC: stable@vger.kernel.org # 4.11+ Signed-off-by: Jason Gerecke <jason.gerecke@wacom.com> Signed-off-by: Jiri Kosina <jkosina@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When configuring flow steering rules, the driver is currently going
through a reset for all errors from the device. Instead, the driver
should only reset when there's a timeout error from the device.
Make sure that the tag_list_lock mutex is not held any longer than
necessary. This change reduces latency if e.g. blk_mq_quiesce_tagset()
is called concurrently from more than one thread. This function is used
by the NVMe core and also by the UFS driver.
Reported-by: Peter Wang <peter.wang@mediatek.com> Cc: Chao Leng <lengchao@huawei.com> Cc: Ming Lei <ming.lei@redhat.com> Cc: stable@vger.kernel.org Fixes: 414dd48e882c ("blk-mq: add tagset quiesce interface") Signed-off-by: Bart Van Assche <bvanassche@acm.org> Reviewed-by: Keith Busch <kbusch@kernel.org> Link: https://lore.kernel.org/r/20241022181617.2716173-1-bvanassche@acm.org Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Supposing first scenario with a virtio_blk driver.
CPU0 CPU1
blk_mq_try_issue_directly()
__blk_mq_issue_directly()
q->mq_ops->queue_rq()
virtio_queue_rq()
blk_mq_stop_hw_queue()
virtblk_done()
blk_mq_request_bypass_insert() 1) store
blk_mq_start_stopped_hw_queue()
clear_bit(BLK_MQ_S_STOPPED) 3) store
blk_mq_run_hw_queue()
if (!blk_mq_hctx_has_pending()) 4) load
return
blk_mq_sched_dispatch_requests()
blk_mq_run_hw_queue()
if (!blk_mq_hctx_has_pending())
return
blk_mq_sched_dispatch_requests()
if (blk_mq_hctx_stopped()) 2) load
return
__blk_mq_sched_dispatch_requests()
Supposing another scenario.
CPU0 CPU1
blk_mq_requeue_work()
blk_mq_insert_request() 1) store
virtblk_done()
blk_mq_start_stopped_hw_queue()
blk_mq_run_hw_queues() clear_bit(BLK_MQ_S_STOPPED) 3) store
blk_mq_run_hw_queue()
if (!blk_mq_hctx_has_pending()) 4) load
return
blk_mq_sched_dispatch_requests()
if (blk_mq_hctx_stopped()) 2) load
continue
blk_mq_run_hw_queue()
Both scenarios are similar, the full memory barrier should be inserted
between 1) and 2), as well as between 3) and 4) to make sure that either
CPU0 sees BLK_MQ_S_STOPPED is cleared or CPU1 sees dispatch list.
Otherwise, either CPU will not rerun the hardware queue causing
starvation of the request.
The easy way to fix it is to add the essential full memory barrier into
helper of blk_mq_hctx_stopped(). In order to not affect the fast path
(hardware queue is not stopped most of the time), we only insert the
barrier into the slow path. Actually, only slow path needs to care about
missing of dispatching the request to the low-level device driver.
Fixes: 320ae51feed5 ("blk-mq: new multi-queue block IO queueing mechanism") Cc: stable@vger.kernel.org Cc: Muchun Song <muchun.song@linux.dev> Signed-off-by: Muchun Song <songmuchun@bytedance.com> Reviewed-by: Ming Lei <ming.lei@redhat.com> Link: https://lore.kernel.org/r/20241014092934.53630-4-songmuchun@bytedance.com Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
blk_mq_insert_request() 1) store
blk_mq_unquiesce_queue()
blk_queue_flag_clear() 3) store
blk_mq_run_hw_queues()
blk_mq_run_hw_queue()
if (!blk_mq_hctx_has_pending()) 4) load
return
blk_mq_run_hw_queue()
if (blk_queue_quiesced()) 2) load
return
blk_mq_sched_dispatch_requests()
The full memory barrier should be inserted between 1) and 2), as well as
between 3) and 4) to make sure that either CPU0 sees QUEUE_FLAG_QUIESCED
is cleared or CPU1 sees dispatch list or setting of bitmap of software
queue. Otherwise, either CPU will not rerun the hardware queue causing
starvation.
So the first solution is to 1) add a pair of memory barrier to fix the
problem, another solution is to 2) use hctx->queue->queue_lock to
synchronize QUEUE_FLAG_QUIESCED. Here, we chose 2) to fix it since
memory barrier is not easy to be maintained.
Fixes: f4560ffe8cec ("blk-mq: use QUEUE_FLAG_QUIESCED to quiesce queue") Cc: stable@vger.kernel.org Cc: Muchun Song <muchun.song@linux.dev> Signed-off-by: Muchun Song <songmuchun@bytedance.com> Reviewed-by: Ming Lei <ming.lei@redhat.com> Link: https://lore.kernel.org/r/20241014092934.53630-3-songmuchun@bytedance.com Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
After CPU0 has marked the queue as stopped, CPU1 will see the queue is
stopped. But before CPU1 puts the request on the dispatch list, CPU2
receives the interrupt of completion of request, so it will run the
hardware queue and marks the queue as non-stopped. Meanwhile, CPU1 also
runs the same hardware queue. After both CPU1 and CPU2 complete
blk_mq_run_hw_queue(), CPU1 just puts the request to the same hardware
queue and returns. It misses dispatching a request. Fix it by running
the hardware queue explicitly. And blk_mq_request_issue_directly()
should handle a similar situation. Fix it as well.
Fixes: d964f04a8fde ("blk-mq: fix direct issue") Cc: stable@vger.kernel.org Cc: Muchun Song <muchun.song@linux.dev> Signed-off-by: Muchun Song <songmuchun@bytedance.com> Reviewed-by: Ming Lei <ming.lei@redhat.com> Link: https://lore.kernel.org/r/20241014092934.53630-2-songmuchun@bytedance.com Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 18011eac28c7 ("arm64: tls: Avoid unconditional zeroing of
tpidrro_el0 for native tasks") tried to optimise the context switching
of tpidrro_el0 by eliding the clearing of the register when switching
to a native task with kpti enabled, on the erroneous assumption that
the kpti trampoline entry code would already have taken care of the
write.
Although the kpti trampoline does zero the register on entry from a
native task, the check in tls_thread_switch() is on the *next* task and
so we can end up leaving a stale, non-zero value in the register if the
previous task was 32-bit.
Drop the broken optimisation and zero tpidrro_el0 unconditionally when
switching to a native 64-bit task.
Cc: Mark Rutland <mark.rutland@arm.com> Cc: stable@vger.kernel.org Fixes: 18011eac28c7 ("arm64: tls: Avoid unconditional zeroing of tpidrro_el0 for native tasks") Signed-off-by: Will Deacon <will@kernel.org> Acked-by: Mark Rutland <mark.rutland@arm.com> Link: https://lore.kernel.org/r/20241114095332.23391-1-will@kernel.org Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
and UBLK_MAX_QUEUE_DEPTH is 4096 and part of UAPI, so 'max_cmd_buf_size'
is always page aligned in 4K page size kernel. However, it isn't true in
64K page size kernel.
Fixes the issue by always rounding up 'max_cmd_buf_size' with PAGE_SIZE.
Cc: stable@vger.kernel.org Fixes: 71f28f3136af ("ublk_drv: add io_uring based userspace block driver") Signed-off-by: Ming Lei <ming.lei@redhat.com> Link: https://lore.kernel.org/r/20241111110718.1394001-1-ming.lei@redhat.com Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Initialize the variable ret at the time of declaration to prevent it from
being returned without a defined value. Fixes smatch warning:
drivers/iio/industrialio-gts-helper.c:256 gain_to_scaletables() error:
uninitialized symbol 'ret'.
When CONFIG_CPUMASK_OFFSTACK and CONFIG_DEBUG_PER_CPU_MAPS are selected,
cpu_max_bits_warn() generates a runtime warning similar as below when
showing /proc/cpuinfo. Fix this by using nr_cpu_ids (the runtime limit)
instead of NR_CPUS to iterate CPUs.
Cc: stable@vger.kernel.org Signed-off-by: Huacai Chen <chenhuacai@loongson.cn> Reviewed-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> Tested-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> Signed-off-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The drvdata is not available in release. Let's just use container_of()
to get the vector_device instance. Otherwise, removing a vector device
will result in a crash:
The function blk_revalidate_disk_zones() calls the function
disk_update_zone_resources() after freezing the device queue. In turn,
disk_update_zone_resources() calls queue_limits_start_update() which
takes a queue limits mutex lock, resulting in the ordering:
q->q_usage_counter check -> q->limits_lock. However, the usual ordering
is to always take a queue limit lock before freezing the queue to commit
the limits updates, e.g., the code pattern:
lim = queue_limits_start_update(q);
...
blk_mq_freeze_queue(q);
ret = queue_limits_commit_update(q, &lim);
blk_mq_unfreeze_queue(q);
Thus, blk_revalidate_disk_zones() introduces a potential circular
locking dependency deadlock that lockdep sometimes catches with the
splat:
[ 51.934109] ======================================================
[ 51.935916] WARNING: possible circular locking dependency detected
[ 51.937561] 6.12.0+ #2107 Not tainted
[ 51.938648] ------------------------------------------------------
[ 51.940351] kworker/u16:4/157 is trying to acquire lock:
[ 51.941805] ffff9fff0aa0bea8 (&q->limits_lock){+.+.}-{4:4}, at: disk_update_zone_resources+0x86/0x170
[ 51.944314]
but task is already holding lock:
[ 51.945688] ffff9fff0aa0b890 (&q->q_usage_counter(queue)#3){++++}-{0:0}, at: blk_revalidate_disk_zones+0x15f/0x340
[ 51.948527]
which lock already depends on the new lock.
Prevent this issue by moving the calls to blk_mq_freeze_queue() and
blk_mq_unfreeze_queue() around the call to queue_limits_commit_update()
in disk_update_zone_resources(). In case of revalidation failure, the
call to disk_free_zone_resources() in blk_revalidate_disk_zones()
is still done with the queue frozen as before.
Fixes: 843283e96e5a ("block: Fake max open zones limit when there is no limit") Cc: stable@vger.kernel.org Signed-off-by: Damien Le Moal <dlemoal@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Link: https://lore.kernel.org/r/20241126104705.183996-1-dlemoal@kernel.org Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The 'fw_vols' fwnode_handle initialized via
device_get_named_child_node() requires explicit calls to
fwnode_handle_put() when the variable is no longer required.
Add the missing calls to fwnode_handle_put() before the function
returns.
Cc: stable@vger.kernel.org Fixes: 51932f9fc487 ("mtd: ubi: populate ubi volume fwnode") Signed-off-by: Javier Carrasco <javier.carrasco.cruz@gmail.com> Reviewed-by: Zhihao Cheng <chengzhihao1@huawei.com> Signed-off-by: Richard Weinberger <richard@nod.at> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When CONFIG_DMA_ENGINE is disabled, the driver now fails to build:
drivers/tty/serial/amba-pl011.c: In function 'pl011_unthrottle_rx':
drivers/tty/serial/amba-pl011.c:1822:16: error: 'struct uart_amba_port' has no member named 'using_rx_dma'
1822 | if (uap->using_rx_dma) {
| ^~
drivers/tty/serial/amba-pl011.c:1823:20: error: 'struct uart_amba_port' has no member named 'dmacr'
1823 | uap->dmacr |= UART011_RXDMAE;
| ^~
drivers/tty/serial/amba-pl011.c:1824:32: error: 'struct uart_amba_port' has no member named 'dmacr'
1824 | pl011_write(uap->dmacr, uap, REG_DMACR);
| ^~
Add the missing #ifdef check around these field accesses, matching
what other parts of this driver do.
Function pl011_throttle_rx() calls pl011_stop_rx() to disable RX, which
also disables the RX DMA by clearing the RXDMAE bit of the DMACR
register. However, to properly unthrottle RX when DMA is used, the
function pl011_unthrottle_rx() is expected to set the RXDMAE bit of
the DMACR register, which it currently lacks. This causes RX to stall
after the throttle API is called.
Set RXDMAE bit in the DMACR register while unthrottling RX if RX DMA is
used.
Currently in omap_8250_shutdown, the dma->rx_running flag is
set to zero in omap_8250_rx_dma_flush. Next pm_runtime_get_sync
is called, which is a runtime resume call stack which can
re-set the flag. When the call omap_8250_shutdown returns, the
flag is expected to be UN-SET, but this is not the case. This
is causing issues the next time UART is re-opened and
omap_8250_rx_dma is called. Fix by moving pm_runtime_get_sync
before the omap_8250_rx_dma_flush.
cc: stable@vger.kernel.org Fixes: 0e31c8d173ab ("tty: serial: 8250_omap: add custom DMA-RX callback") Signed-off-by: Bin Liu <b-liu@ti.com>
[Judith: Add commit message] Signed-off-by: Judith Mendez <jm@ti.com> Reviewed-by: Kevin Hilman <khilman@baylibre.com> Tested-by: Kevin Hilman <khilman@baylibre.com> Link: https://lore.kernel.org/r/20241031172315.453750-1-jm@ti.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The F81216E is a LPC/eSPI to 4 UART Super I/O and is mostly compatible with
the F81216H, but does not support RS-485 auto-direction delays on any port.
Code expects array only with 2 items which should be checked.
But also item checking is not working as it should likely because of
incorrect items description.
The drvdata is not available in release. Let's just use container_of()
to get the uml_net instance. Otherwise, removing a network device will
result in a crash:
projects / linux / log