This change moves all global binder state into
the context struct, thereby completely separating
the state and the locks between two different contexts.
The debugfs entries remain global, printing entries
from all contexts.
Change-Id: If8e3e2bece7bc6f974b66fbcf1d91d529ffa62f0
Signed-off-by: Martijn Coenen <maco@google.com>
binder_fd_array_object starts with a 4-byte header,
followed by a few fields that are 8 bytes when
ANDROID_BINDER_IPC_32BIT=N.
This can cause alignment issues in a 64-bit kernel
with a 32-bit userspace, as on x86_32 an 8-byte primitive
may be aligned to a 4-byte address. Pad with a __u32
to fix this.
Change-Id: I4374ed2cc3ccd3c6a1474cb7209b53ebfd91077b
Signed-off-by: Martijn Coenen <maco@android.com>
The binder allocator assumes that the thread that
called binder_open will never die for the lifetime of
that proc. That thread is normally the group_leader,
however it may not be. Use the group_leader instead
of current.
Bug: 35707103
Test: Created test case to open with temporary thread
Change-Id: Id693f74b3591f3524a8c6e9508e70f3e5a80c588
Signed-off-by: Todd Kjos <tkjos@google.com>
Signed-off-by: Martijn Coenen <maco@android.com>
sock_i_uid() acquires the sk_callback_lock which does not exist for
sockets in TCP_NEW_SYN_RECV state. This results in errors showing up
as spinlock bad magic. Fix this by looking for the full sock as
suggested by Eric.
Callstack for reference -
-003|rwlock_bug
-004|arch_read_lock
-004|do_raw_read_lock
-005|raw_read_lock_bh
-006|sock_i_uid
-007|from_kuid_munged(inline)
-007|reset_timer
-008|idletimer_tg_target
-009|ipt_do_table
-010|iptable_mangle_hook
-011|nf_iterate
-012|nf_hook_slow
-013|NF_HOOK_COND(inline)
-013|ip_output
-014|ip_local_out
-015|ip_build_and_send_pkt
-016|tcp_v4_send_synack
-017|atomic_sub_return(inline)
-017|reqsk_put(inline)
-017|tcp_conn_request
-018|tcp_v4_conn_request
-019|tcp_rcv_state_process
-020|tcp_v4_do_rcv
-021|tcp_v4_rcv
-022|ip_local_deliver_finish
-023|NF_HOOK_THRESH(inline)
-023|NF_HOOK(inline)
-023|ip_local_deliver
-024|ip_rcv_finish
-025|NF_HOOK_THRESH(inline)
-025|NF_HOOK(inline)
-025|ip_rcv
-026|deliver_skb(inline)
-026|deliver_ptype_list_skb(inline)
-026|__netif_receive_skb_core
-027|__netif_receive_skb
-028|netif_receive_skb_internal
-029|netif_receive_skb
Change-Id: Ic8f3a3d2d7af31434d1163b03971994e2125d552
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Cc: Eric Dumazet <edumazet@google.com>
schedule_work(&timer->work) appears to be called after
cancel_work_sync(&info->timer->work) is completed.
Work can be scheduled from the PM_POST_SUSPEND notification event
even after cancel_work_sync is called.
Call stack
-004|notify_netlink_uevent(
| [X19] timer = 0xFFFFFFC0A5DFC780 -> (
| ...
| [NSD:0xFFFFFFC0A5DFC800] kobj = 0x6B6B6B6B6B6B6B6B,
| [NSD:0xFFFFFFC0A5DFC868] timeout = 0x6B6B6B6B,
| [NSD:0xFFFFFFC0A5DFC86C] refcnt = 0x6B6B6B6B,
| [NSD:0xFFFFFFC0A5DFC870] work_pending = 0x6B,
| [NSD:0xFFFFFFC0A5DFC871] send_nl_msg = 0x6B,
| [NSD:0xFFFFFFC0A5DFC872] active = 0x6B,
| [NSD:0xFFFFFFC0A5DFC874] uid = 0x6B6B6B6B,
| [NSD:0xFFFFFFC0A5DFC878] suspend_time_valid = 0x6B))
-005|idletimer_tg_work(
-006|__read_once_size(inline)
-006|static_key_count(inline)
-006|static_key_false(inline)
-006|trace_workqueue_execute_end(inline)
-006|process_one_work(
-007|worker_thread(
-008|kthread(
-009|ret_from_fork(asm)
---|end of frame
Force any pending idletimer_tg_work() to complete before freeing
the associated work struct and after unregistering to the pm_notifier
callback.
Change-Id: I4c5f0a1c142f7d698c092cf7bcafdb0f9fbaa9c1
Signed-off-by: Subash Abhinov Kasiviswanathan <subashab@codeaurora.org>
If table_make_digest() fails, verify_verity_signature() would try to
pass the returned ERR_PTR() to kfree().
This fixes the smatch error:
drivers/md/dm-android-verity.c:601 verify_verity_signature() error: 'pks' dereferencing possible ERR_PTR()
Change-Id: I9b9b7764b538cb4a5f94337660e9b0f149b139be
Signed-off-by: Greg Hackmann <ghackmann@google.com>
Keep time calculation in 64-bit throughout. If we have long times
between idle calculations this can result in deltas > 32 bits
which causes incorrect load percentage calculations and selecting
the wrong frequencies if we truncate here.
Signed-off-by: Chris Redpath <chris.redpath@arm.com>
This comes from the wrapfs patch
2e346c83b26e Wrapfs: support direct-IO (DIO) operations
Signed-off-by: Li Mengyang <li.mengyang@stonybrook.edu>
Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
Signed-off-by: Daniel Rosenberg <drosen@google.com>
Bug: 34133558
Change-Id: I3fd779c510ab70d56b1d918f99c20421b524cdc4
This comes from the wrapfs patch
3dfec0ffe5e2 Wrapfs: implement vm_ops->page_mkwrite
Some file systems (e.g., ext4) require it. Reported by Ted Ts'o.
Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
Signed-off-by: Daniel Rosenberg <drosen@google.com>
Bug: 34133558
Change-Id: I1a389b2422c654a6d3046bb8ec3e20511aebfa8e
There is no point deleting entries from dlist, as
that is a temporary list on the stack from which
contains only entries that are being deleted.
Not all code paths set up dlist, so those that
don't were performing invalid accesses in
hash_del_rcu. As an additional means to prevent
any other issue, we null out the list entries when
we allocate from the cache.
Signed-off-by: Daniel Rosenberg <drosen@google.com>
Bug: 35666680
Change-Id: Ibb1e28c08c3a600c29418d39ba1c0f3db3bf31e5
"ANDROID: sdcardfs: Add GID Derivation to sdcardfs" introduced
an unbalanced pat_get, leading to storage space not being freed
after deleting a file until rebooting. This adds the missing path_put.
Signed-off-by: Daniel Rosenberg <drosen@google.com>
Bug: 34691169
Change-Id: Ia7ef97ec2eca2c555cc06b235715635afc87940e
This adds back the hash calculation removed as part of
the previous patch, as it is in fact necessary.
Signed-off-by: Daniel Rosenberg <drosen@google.com>
Bug: 35307857
Change-Id: Ie607332bcf2c5d2efdf924e4060ef3f576bf25dc
Add a new per-file key derivation method to ext4 encryption defined as:
derived_key[0:127] = AES-256-ENCRYPT(master_key[0:255], nonce)
derived_key[128:255] = AES-256-ENCRYPT(master_key[0:255], nonce ^ 0x01)
derived_key[256:383] = AES-256-ENCRYPT(master_key[256:511], nonce)
derived_key[384:511] = AES-256-ENCRYPT(master_key[256:511], nonce ^ 0x01)
... where the derived key and master key are both 512 bits, the nonce is
128 bits, AES-256-ENCRYPT takes the arguments (key, plaintext), and
'nonce ^ 0x01' denotes flipping the low order bit of the last byte.
The existing key derivation method is
'derived_key = AES-128-ECB-ENCRYPT(key=nonce, plaintext=master_key)'.
We want to make this change because currently, given a derived key you
can easily compute the master key by computing
'AES-128-ECB-DECRYPT(key=nonce, ciphertext=derived_key)'.
This was formerly OK because the previous threat model assumed that the
master key and derived keys are equally hard to obtain by an attacker.
However, we are looking to move the master key into secure hardware in
some cases, so we want to make sure that an attacker with access to a
derived key cannot compute the master key.
We are doing this instead of increasing the nonce to 512 bits because
it's important that the per-file xattr fit in the inode itself. By
default, inodes are 256 bytes, and on Android we're already pretty close
to that limit. If we increase the nonce size, we end up allocating a new
filesystem block for each and every encrypted file, which has a
substantial performance and disk utilization impact.
Another option considered was to use the HMAC-SHA512 of the nonce, keyed
by the master key. However this would be a little less performant,
would be less extensible to other key sizes and MAC algorithms, and
would pull in a dependency (security-wise and code-wise) on SHA-512.
Due to the use of "aes" rather than "ecb(aes)" in the implementation,
the new key derivation method is actually about twice as fast as the old
one, though the old one could be optimized similarly as well.
This patch makes the new key derivation method be used whenever HEH is
used to encrypt filenames. Although these two features are logically
independent, it was decided to bundle them together for now. Note that
neither feature is upstream yet, and it cannot be guaranteed that the
on-disk format won't change if/when these features are upstreamed. For
this reason, and as noted in the previous patch, the features are both
behind a special mode number for now.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: Iee4113f57e59dc8c0b7dc5238d7003c83defb986
Update ext4 encryption to allow filenames to be encrypted using the
Hash-Encrypt-Hash (HEH) block cipher mode of operation, which is
believed to be more secure than CBC, particularly within the constant
initialization vector (IV) constraint of filename encryption. Notably,
HEH avoids the "common prefix" problem of CBC. Both algorithms use
AES-256 as the underlying block cipher and take a 256-bit key.
We assign mode number 126 to HEH, just below 127
(EXT4_ENCRYPTION_MODE_PRIVATE) which in some kernels is reserved for
inline encryption on MSM chipsets. Note that these modes are not yet
upstream, which is why these numbers are being used; it's preferable to
avoid collisions with modes that may be added upstream. Also, although
HEH is not hardware-specific, we aren't currently reserving mode number
5 for HEH upstream, since for now we are tying HEH to the new key
derivation method which might become an independent flag upstream, and
there's also a chance that details of HEH will change after it gets
wider review.
Bug: 32975945
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: I81418709d47da0e0ac607ae3f91088063c2d5dd4
poly_hash is part of the HEH (Hash-Encrypt-Hash) encryption mode,
proposed in Internet Draft
https://tools.ietf.org/html/draft-cope-heh-01. poly_hash is very
similar to GHASH; besides the swapping of the last two coefficients
which we opted to handle in the HEH template, poly_hash just uses a
different finite field representation. As with GHASH, poly_hash becomes
much faster and more secure against timing attacks when implemented
using carryless multiplication instructions instead of tables. This
patch adds an ARMv8-CE optimized version of poly_hash, based roughly on
the existing ARMv8-CE optimized version of GHASH.
Benchmark results are shown below, but note that the resistance to
timing attacks may be even more important than the performance gain.
poly_hash only:
poly_hash-generic:
1,000,000 setkey() takes 1185 ms
hashing is 328 MB/s
poly_hash-ce:
1,000,000 setkey() takes 8 ms
hashing is 1756 MB/s
heh(aes) with 4096-byte inputs (this is the ideal case, as the
improvement is less significant with smaller inputs):
encryption with "heh_base(cmac(aes-ce),poly_hash-generic,ecb-aes-ce)": 118 MB/s
decryption with "heh_base(cmac(aes-ce),poly_hash-generic,ecb-aes-ce)": 120 MB/s
encryption with "heh_base(cmac(aes-ce),poly_hash-ce,ecb-aes-ce)": 291 MB/s
decryption with "heh_base(cmac(aes-ce),poly_hash-ce,ecb-aes-ce)": 293 MB/s
Bug: 32508661
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: I621ec0e1115df7e6f5cbd7e864a4a9d8d2e94cf2
Factor most of poly_hash() out into its own keyed hash algorithm so that
optimized architecture-specific implementations of it will be possible.
For now we call poly_hash through the shash API, since HEH already had
an example of using shash for another algorithm (CMAC), and we will not
be adding any poly_hash implementations that require ahash yet. We can
however switch to ahash later if it becomes useful.
Bug: 32508661
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: I8de54ddcecd1d7fa6e9842a09506a08129bae0b6
Hash-Encrypt-Hash (HEH) is a proposed block cipher mode of operation
which extends the strong pseudo-random permutation property of block
ciphers (e.g. AES) to arbitrary length input strings. This provides a
stronger notion of security than existing block cipher modes of
operation (e.g. CBC, CTR, XTS), though it is usually less performant.
It uses two keyed invertible hash functions with a layer of ECB
encryption applied in-between. The algorithm is currently specified by
the following Internet Draft:
https://tools.ietf.org/html/draft-cope-heh-01
This patch adds HEH as a symmetric cipher only. Support for HEH as an
AEAD is not yet implemented.
HEH will use an existing accelerated ecb(block_cipher) implementation
for the encrypt step if available. Accelerated versions of the hash
step are planned but will be left for later patches.
This patch backports HEH to the 4.4 Android kernel, initially for use by
ext4 filenames encryption. Note that HEH is not yet upstream; however,
patches have been made available on linux-crypto, and as noted there is
also a draft specification available. This backport required updating
the code to conform to the legacy ablkcipher API rather than the
skcipher API, which wasn't complete in 4.4.
Signed-off-by: Alex Cope <alexcope@google.com>
Bug: 32975945
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: I945bcc9c0115916824d701bae91b86e3f059a1a9
Adding ble multiplication to GF128mul, and fixing up comments.
The ble multiplication functions multiply GF(2^128) elements in the
ble format. This format is preferable because the bits within each
byte map to polynomial coefficients in the natural order (lowest order
bit = coefficient of lowest degree polynomial term), and the bytes are
stored in little endian order which matches the endianness of most
modern CPUs.
These new functions will be used by the HEH algorithm.
Signed-off-by: Alex Cope <alexcope@google.com>
Bug: 32975945
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: I39a58e8ee83e6f9b2e6bd51738f816dbfa2f3a47
Rename and clean up the GF(2^128) overflow macros and tables. Their
usage is more general than the name suggested, e.g. what was previously
known as the "bbe" table can actually be used for both "bbe" and "ble"
multiplication.
Bug: 32975945
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: Ie6c47b4075ca40031eb1767e9b468cfd7bf1b2e4
GF(2^128) multiplication tables are typically used for secret
information, so it's a good idea to zero them on free.
Signed-off-by: Alex Cope <alexcope@google.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
(cherry-picked from 75aa0a7caf)
Bug: 32975945
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: I37b1ae9544158007f9ee2caf070120f4a42153ab
Analogous to crypto_grab_skcipher() and crypto_spawn_skcipher_alg(),
these are useful for algorithms that need to use a shash sub-algorithm,
possibly in addition to other sub-algorithms.
Bug: 32975945
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: I44e5a519d73f5f839e3b6ecbf8c66e36ec569557
Add a function blkcipher_ablkcipher_walk_virt() which allows ablkcipher
algorithms to use the blkcipher_walk API to walk over their data. This
will be used by the HEH algorithm, which to support asynchronous ECB
algorithms will be an ablkcipher, but it also needs to make other passes
over the data.
Bug: 32975945
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: I05f9a0e5473ba6115fcc72d5122d6b0b18b2078b
ECB modes don't use an initialization vector. The kernel
/proc/crypto interface doesn't reflect this properly.
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
(cherry picked from bee038a4bd)
Signed-off-by: Eric Biggers <ebiggers@google.com>
Change-Id: Ief9558d2b41be58a2d845d2033a141b5ef7b585f
Refactor the fs readpage/write tracepoints to move the
inode->path lookup outside the tracepoint code, and pass a pointer
to the path into the tracepoint code instead. This is necessary
because the tracepoint code runs non-preemptible. Thanks to
Trilok Soni for catching this in 4.4.
Change-Id: I7486c5947918d155a30c61d6b9cd5027cf8fbe15
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
When dealing with uncompressed data, there is no need to read a whole
block (default 128K) to get the desired page: the pages are
independent from each others.
This patch change the readpages logic so that reading uncompressed
data only read the number of pages advised by the readahead algorithm.
Moreover, if the page actor contains holes (i.e. pages that are already
up-to-date), squashfs skips the buffer_head associated to those pages.
This patch greatly improve the performance of random reads for
uncompressed files because squashfs only read what is needed. It also
reduces the number of unnecessary reads.
Signed-off-by: Adrien Schildknecht <adriens@google.com>
Change-Id: I1850150fbf4b45c9dd138d88409fea1ab44054c0
Squashfs does not implement .readpages(), so the kernel just repeatedly
calls .readpage().
The readpages function tries to pack as much pages as possible in the
same page actor so that only 1 read request is issued.
Now that the read requests are asynchronous, the kernel can truly
prefetch pages using its readahead algorithm.
Signed-off-by: Adrien Schildknecht <adriens@google.com>
Change-Id: Ice70e029dc24526f61e4e5a1a902588be2212498
The 'll_rw_block' has been deprecated and BIO is now the basic container
for block I/O within the kernel.
Switching to BIO offers 2 advantages:
1/ It removes synchronous wait for the up-to-date buffers: SquashFS
now deals with decompressions/copies asynchronously.
Implementing an asynchronous mechanism to read data is needed to
efficiently implement .readpages().
2/ Prior to this patch, merging the read requests entirely depends on
the IO scheduler. SquashFS has more information than the IO
scheduler about what could be merged. Moreover, merging the reads
at the FS level means that we rely less on the IO scheduler.
Signed-off-by: Adrien Schildknecht <adriens@google.com>
Change-Id: I775d2e11f017476e1899518ab52d9d0a8a0bce28
This patch essentially does 3 things:
1/ Always use an array of page to store the data instead of a mix of
buffers and pages.
2/ It is now possible to have 'holes' in a page actor, i.e. NULL
pages in the array.
When reading a block (default 128K), squashfs tries to grab all
the pages covering this block. If a single page is up-to-date or
locked, it falls back to using an intermediate buffer to do the
read and then copy the pages in the actor. Allowing holes in the
page actor remove the need for this intermediate buffer.
3/ Refactor the wrappers to share code that deals with page actors.
Signed-off-by: Adrien Schildknecht <adriens@google.com>
Change-Id: I98128bed5d518cf31b67e788a85b275e9a323bec
FILE_DIRECT is working fine and offers faster results and lower memory
footprint.
Removing FILE_CACHE makes our life easier because we don't have to
maintain 2 differents function that does the same thing.
Signed-off-by: Adrien Schildknecht <adriens@google.com>
Change-Id: I6689ba74d0042c222a806f9edc539995e8e04c6b
(cherry picked from commit 0ddb8e0b78)
Since, arm64 can support all offset within a double word limit. Therefore,
now support other lengths within that range as well.
Signed-off-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Pavel Labath <labath@google.com>
Change-Id: Ibcb263a3903572336ccbf96e0180d3990326545a
Bug: 30919905
(cherry picked from commit fdfeff0f9e)
Arm64 hardware does not always report a watchpoint hit address that
matches one of the watchpoints set. It can also report an address
"near" the watchpoint if a single instruction access both watched and
unwatched addresses. There is no straight-forward way, short of
disassembling the offending instruction, to map that address back to
the watchpoint.
Previously, when the hardware reported a watchpoint hit on an address
that did not match our watchpoint (this happens in case of instructions
which access large chunks of memory such as "stp") the process would
enter a loop where we would be continually resuming it (because we did
not recognise that watchpoint hit) and it would keep hitting the
watchpoint again and again. The tracing process would never get
notified of the watchpoint hit.
This commit fixes the problem by looking at the watchpoints near the
address reported by the hardware. If the address does not exactly match
one of the watchpoints we have set, it attributes the hit to the
nearest watchpoint we have. This heuristic is a bit dodgy, but I don't
think we can do much more, given the hardware limitations.
Signed-off-by: Pavel Labath <labath@google.com>
[panand: reworked to rebase on his patches]
Signed-off-by: Pratyush Anand <panand@redhat.com>
[will: use __ffs instead of ffs - 1]
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Pavel Labath <labath@google.com>
[pavel: trivial fixup in hw_breakpoint.c:watchpoint_handler]
Change-Id: I714dfaa3947d89d89a9e9a1ea84914d44ba0faa3
Bug: 30919905
ARM64 hardware supports watchpoint at any double word aligned address.
However, it can select any consecutive bytes from offset 0 to 7 from that
base address. For example, if base address is programmed as 0x420030 and
byte select is 0x1C, then access of 0x420032,0x420033 and 0x420034 will
generate a watchpoint exception.
Currently, we do not have such modularity. We can only program byte,
halfword, word and double word access exception from any base address.
This patch adds support to overcome above limitations.
Signed-off-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Pavel Labath <labath@google.com>
Change-Id: I28b1ca63f63182c10c3d6b6b3bacf6c56887ddbe
Bug: 30919905
(cherry picked from commit 651be3cb08)
We only support breakpoint/watchpoint of length 1, 2, 4 and 8. If we can
support other length as well, then user may watch more data with less
number of watchpoints (provided hardware supports it). For example: if we
have to watch only 4th, 5th and 6th byte from a 64 bit aligned address, we
will have to use two slots to implement it currently. One slot will watch a
half word at offset 4 and other a byte at offset 6. If we can have a
watchpoint of length 3 then we can watch it with single slot as well.
ARM64 hardware does support such functionality, therefore adding these new
definitions in generic layer.
Signed-off-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Pavel Labath <labath@google.com>
[pavel: tools/include/uapi/linux/hw_breakpoint.h is not present in this branch]
Change-Id: Ie17ed89ca526e4fddf591bb4e556fdfb55fc2eac
Bug: 30919905
commit de5540d088 upstream.
Under extremely heavy uses of padata, crashes occur, and with list
debugging turned on, this happens instead:
[87487.298728] WARNING: CPU: 1 PID: 882 at lib/list_debug.c:33
__list_add+0xae/0x130
[87487.301868] list_add corruption. prev->next should be next
(ffffb17abfc043d0), but was ffff8dba70872c80. (prev=ffff8dba70872b00).
[87487.339011] [<ffffffff9a53d075>] dump_stack+0x68/0xa3
[87487.342198] [<ffffffff99e119a1>] ? console_unlock+0x281/0x6d0
[87487.345364] [<ffffffff99d6b91f>] __warn+0xff/0x140
[87487.348513] [<ffffffff99d6b9aa>] warn_slowpath_fmt+0x4a/0x50
[87487.351659] [<ffffffff9a58b5de>] __list_add+0xae/0x130
[87487.354772] [<ffffffff9add5094>] ? _raw_spin_lock+0x64/0x70
[87487.357915] [<ffffffff99eefd66>] padata_reorder+0x1e6/0x420
[87487.361084] [<ffffffff99ef0055>] padata_do_serial+0xa5/0x120
padata_reorder calls list_add_tail with the list to which its adding
locked, which seems correct:
spin_lock(&squeue->serial.lock);
list_add_tail(&padata->list, &squeue->serial.list);
spin_unlock(&squeue->serial.lock);
This therefore leaves only place where such inconsistency could occur:
if padata->list is added at the same time on two different threads.
This pdata pointer comes from the function call to
padata_get_next(pd), which has in it the following block:
next_queue = per_cpu_ptr(pd->pqueue, cpu);
padata = NULL;
reorder = &next_queue->reorder;
if (!list_empty(&reorder->list)) {
padata = list_entry(reorder->list.next,
struct padata_priv, list);
spin_lock(&reorder->lock);
list_del_init(&padata->list);
atomic_dec(&pd->reorder_objects);
spin_unlock(&reorder->lock);
pd->processed++;
goto out;
}
out:
return padata;
I strongly suspect that the problem here is that two threads can race
on reorder list. Even though the deletion is locked, call to
list_entry is not locked, which means it's feasible that two threads
pick up the same padata object and subsequently call list_add_tail on
them at the same time. The fix is thus be hoist that lock outside of
that block.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Acked-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit f5fe1b5190 upstream.
Commit 79bd99596b ("blk: improve order of bio handling in generic_make_request()")
changed current->bio_list so that it did not contain *all* of the
queued bios, but only those submitted by the currently running
make_request_fn.
There are two places which walk the list and requeue selected bios,
and others that check if the list is empty. These are no longer
correct.
So redefine current->bio_list to point to an array of two lists, which
contain all queued bios, and adjust various code to test or walk both
lists.
Signed-off-by: NeilBrown <neilb@suse.com>
Fixes: 79bd99596b ("blk: improve order of bio handling in generic_make_request()")
Signed-off-by: Jens Axboe <axboe@fb.com>
[jwang: backport to 4.4]
Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[bwh: Restore changes in device-mapper from upstream version]
Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
commit 79bd99596b upstream.
To avoid recursion on the kernel stack when stacked block devices
are in use, generic_make_request() will, when called recursively,
queue new requests for later handling. They will be handled when the
make_request_fn for the current bio completes.
If any bios are submitted by a make_request_fn, these will ultimately
be handled seqeuntially. If the handling of one of those generates
further requests, they will be added to the end of the queue.
This strict first-in-first-out behaviour can lead to deadlocks in
various ways, normally because a request might need to wait for a
previous request to the same device to complete. This can happen when
they share a mempool, and can happen due to interdependencies
particular to the device. Both md and dm have examples where this happens.
These deadlocks can be erradicated by more selective ordering of bios.
Specifically by handling them in depth-first order. That is: when the
handling of one bio generates one or more further bios, they are
handled immediately after the parent, before any siblings of the
parent. That way, when generic_make_request() calls make_request_fn
for some particular device, we can be certain that all previously
submited requests for that device have been completely handled and are
not waiting for anything in the queue of requests maintained in
generic_make_request().
An easy way to achieve this would be to use a last-in-first-out stack
instead of a queue. However this will change the order of consecutive
bios submitted by a make_request_fn, which could have unexpected consequences.
Instead we take a slightly more complex approach.
A fresh queue is created for each call to a make_request_fn. After it completes,
any bios for a different device are placed on the front of the main queue, followed
by any bios for the same device, followed by all bios that were already on
the queue before the make_request_fn was called.
This provides the depth-first approach without reordering bios on the same level.
This, by itself, it not enough to remove all deadlocks. It just makes
it possible for drivers to take the extra step required themselves.
To avoid deadlocks, drivers must never risk waiting for a request
after submitting one to generic_make_request. This includes never
allocing from a mempool twice in the one call to a make_request_fn.
A common pattern in drivers is to call bio_split() in a loop, handling
the first part and then looping around to possibly split the next part.
Instead, a driver that finds it needs to split a bio should queue
(with generic_make_request) the second part, handle the first part,
and then return. The new code in generic_make_request will ensure the
requests to underlying bios are processed first, then the second bio
that was split off. If it splits again, the same process happens. In
each case one bio will be completely handled before the next one is attempted.
With this is place, it should be possible to disable the
punt_bios_to_recover() recovery thread for many block devices, and
eventually it may be possible to remove it completely.
Ref: http://www.spinics.net/lists/raid/msg54680.html
Tested-by: Jinpu Wang <jinpu.wang@profitbricks.com>
Inspired-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
[jwang: backport to 4.4]
Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 0b0408745e upstream.
LPDDR memories can only handle up to 400 uncontrolled power off. Ensure the
proper power off sequence is used before shutting down the platform.
Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: Sebastian Reichel <sre@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 90db10434b upstream.
No caller currently checks the return value of
kvm_io_bus_unregister_dev(). This is evil, as all callers silently go on
freeing their device. A stale reference will remain in the io_bus,
getting at least used again, when the iobus gets teared down on
kvm_destroy_vm() - leading to use after free errors.
There is nothing the callers could do, except retrying over and over
again.
So let's simply remove the bus altogether, print an error and make
sure no one can access this broken bus again (returning -ENOMEM on any
attempt to access it).
Fixes: e93f8a0f82 ("KVM: convert io_bus to SRCU")
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 3bd32722c8 upstream.
On some QNAP NAS devices the rtc can wake the machine. Several people
noticed that once the machine was woken this way it fails to shut down.
That's because the driver fails to acknowledge the interrupt and so it
keeps active and restarts the machine immediatly after shutdown. See
https://bugs.debian.org/794266 for a bug report.
Doing this correctly requires to interpret the INT2 flag of the first read
of the STATUS1 register because this bit is cleared by read.
Note this is not maximally robust though because a pending irq isn't
detected when the STATUS1 register was already read (and so INT2 is not
set) but the irq was not disabled. But that is a hardware imposed problem
that cannot easily be fixed by software.
Signed-off-by: Uwe Kleine-König <uwe@kleine-koenig.org>
Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 8e6583f1b5 upstream.
There were two deviations from the reference manual: you have to wait
half a second when POC is active and you might have to repeat
initialization when POC or BLD are still set after the sequence.
Note however that as POC and BLD are cleared by read the driver might
not be able to detect that a reset is necessary. I don't have a good
idea how to fix this.
Additionally report the value read from STATUS1 to the caller. This
prepares the next patch.
Signed-off-by: Uwe Kleine-König <uwe@kleine-koenig.org>
Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The rtc core calls the .read_alarm with all fields initialized to 0. As
the s35390a driver doesn't touch some fields the returned date is
interpreted as a date in January 1900. So make sure all fields are set
to -1; some of them are then overwritten with the right data depending
on the hardware state.
In mainline this is done by commit d68778b80d ("rtc: initialize output
parameter for read alarm to "uninitialized"") in the core. This is
considered to dangerous for stable as it might have side effects for
other rtc drivers that might for example rely on alarm->time.tm_sec
being initialized to 0.
Signed-off-by: Uwe Kleine-König <uwe@kleine-koenig.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit f87e904ddd upstream.
There are several issues fixed in this patch:
- When alarm isn't enabled, set .enabled to zero instead of returning
-EINVAL.
- Ignore how IRQ1 is configured when determining if IRQ2 is on.
- The three alarm registers have an enable flag which must be
evaluated.
- The chip always triggers when the seconds register gets 0.
Note that the rtc framework however doesn't handle the result correctly
because it doesn't check wday being initialized and so interprets an
alarm being set for 10:00 AM in three days as 10:00 AM tomorrow (or
today if that's not over yet).
Signed-off-by: Uwe Kleine-König <uwe@kleine-koenig.org>
Signed-off-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 6c356eda22 upstream.
With the IRQ stack changes integrated, the XRX200 devices started
emitting a constant stream of kernel messages like this:
[ 565.415310] Spurious IRQ: CAUSE=0x1100c300
This is caused by IP0 getting handled by plat_irq_dispatch() rather than
its vectored interrupt handler, which is fixed by commit de856416e714
("MIPS: IRQ Stack: Fix erroneous jal to plat_irq_dispatch").
Fix plat_irq_dispatch() to handle non-vectored IPI interrupts correctly
by setting up IP2-6 as proper chained IRQ handlers and calling do_IRQ
for all MIPS CPU interrupts.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Acked-by: John Crispin <john@phrozen.org>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/15077/
[james.hogan@imgtec.com: tweaked commit message]
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Signed-off-by: Amit Pundir <amit.pundir@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
