Damien Le Moal [Wed, 11 Aug 2021 03:36:58 +0000 (12:36 +0900)]
block: improve ioprio class description comment
In include/usapi/linux/ioprio.h, change the ioprio class enum comment
to remove the outdated reference to CFQ and mention BFQ and mq-deadline
instead. Also document the high priority NCQ command use for RT class
IOs directed at ATA drives that support NCQ priority.
Damien Le Moal [Wed, 11 Aug 2021 03:36:57 +0000 (12:36 +0900)]
block: bfq: fix bfq_set_next_ioprio_data()
For a request that has a priority level equal to or larger than
IOPRIO_BE_NR, bfq_set_next_ioprio_data() prints a critical warning but
defaults to setting the request new_ioprio field to IOPRIO_BE_NR. This
is not consistent with the warning and the allowed values for priority
levels. Fix this by setting the request new_ioprio field to
IOPRIO_BE_NR - 1, the lowest priority level allowed.
Cc: <stable@vger.kernel.org> Fixes: aee69d78dec0 ("block, bfq: introduce the BFQ-v0 I/O scheduler as an extra scheduler") Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Link: https://lore.kernel.org/r/20210811033702.368488-2-damien.lemoal@wdc.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
Christoph Hellwig [Wed, 4 Aug 2021 09:56:28 +0000 (11:56 +0200)]
bcache: use bvec_virt
Use bvec_virt instead of open coding it. Note that the existing code is
fine despite ignoring bv_offset as the bio is known to contain exactly
one page from the page allocator per bio_vec.
Christoph Hellwig [Mon, 16 Aug 2021 12:26:14 +0000 (14:26 +0200)]
block: ensure the bdi is freed after inode_detach_wb
inode_detach_wb references the "main" bdi of the inode. With the
recent change to move the bdi from the request_queue to the gendisk
this causes a guaranteed use after free when using certain cgroup
configurations. The big itself is older through as any non-default
inode reference (e.g. an open file descriptor) could have injected
this use after free even before that.
Fixes: 52ebea749aae ("writeback: make backing_dev_info host cgroup-specific bdi_writebacks") Reported-by: Qian Cai <quic_qiancai@quicinc.com> Reported-by: syzbot <syzbot+1fb38bb7d3ce0fa3e1c4@syzkaller.appspotmail.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Link: https://lore.kernel.org/r/20210816122614.601358-3-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Christoph Hellwig [Mon, 16 Aug 2021 12:26:13 +0000 (14:26 +0200)]
block: free the extended dev_t minor later
The dev_t is used as the inode hash, so we should only released it
once then block device inode is gone from the inode cache. Move it
to bdev_free_inode to ensure that.
Chunguang Xu [Mon, 2 Aug 2021 03:51:56 +0000 (11:51 +0800)]
blk-throtl: optimize IOPS throttle for large IO scenarios
After patch 54efd50 (block: make generic_make_request handle
arbitrarily sized bios), the IO through io-throttle may be larger,
and these IOs may be further split into more small IOs. However,
IOPS throttle does not seem to be aware of this change, which
makes the calculation of IOPS of large IOs incomplete, resulting
in disk-side IOPS that does not meet expectations. Maybe we should
fix this problem.
We can reproduce it by set max_sectors_kb of disk to 128, set
blkio.write_iops_throttle to 100, run a dd instance inside blkio
and use iostat to watch IOPS:
Christoph Hellwig [Tue, 10 Aug 2021 15:45:09 +0000 (17:45 +0200)]
block: store a gendisk in struct parsed_partitions
Partition scanning only happens on the whole device, so pass a
struct gendisk instead of the whole device block_device to the scanners.
This allows to simplify printing the device name in various places as the
disk name is available in disk->name.
Christoph Hellwig [Mon, 9 Aug 2021 06:40:27 +0000 (08:40 +0200)]
bcache: move the del_gendisk call out of bcache_device_free
Let the callers call del_gendisk so that we can check if add_disk
has been called properly for the cached device case instead of relying
on the block layer internal GENHD_FL_UP flag.
Ming Lei [Thu, 29 Jul 2021 03:42:26 +0000 (11:42 +0800)]
block: return ELEVATOR_DISCARD_MERGE if possible
When merging one bio to request, if they are discard IO and the queue
supports multi-range discard, we need to return ELEVATOR_DISCARD_MERGE
because both block core and related drivers(nvme, virtio-blk) doesn't
handle mixed discard io merge(traditional IO merge together with
discard merge) well.
Fix the issue by returning ELEVATOR_DISCARD_MERGE in this situation,
so both blk-mq and drivers just need to handle multi-range discard.
Christoph Hellwig [Mon, 9 Aug 2021 14:17:44 +0000 (16:17 +0200)]
block: remove the bd_bdi in struct block_device
Just retrieve the bdi from the disk.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Link: https://lore.kernel.org/r/20210809141744.1203023-6-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Christoph Hellwig [Mon, 9 Aug 2021 14:17:43 +0000 (16:17 +0200)]
block: move the bdi from the request_queue to the gendisk
The backing device information only makes sense for file system I/O,
and thus belongs into the gendisk and not the lower level request_queue
structure. Move it there.
Christoph Hellwig [Mon, 9 Aug 2021 14:17:42 +0000 (16:17 +0200)]
block: add a queue_has_disk helper
Add a helper to check if a gendisk is associated with a request_queue.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Link: https://lore.kernel.org/r/20210809141744.1203023-4-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Christoph Hellwig [Mon, 9 Aug 2021 14:17:41 +0000 (16:17 +0200)]
block: pass a gendisk to blk_queue_update_readahead
.. and rename the function to disk_update_readahead. This is in
preparation for moving the BDI from the request_queue to the gendisk.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Link: https://lore.kernel.org/r/20210809141744.1203023-3-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Christoph Hellwig [Mon, 9 Aug 2021 14:17:40 +0000 (16:17 +0200)]
mm: hide laptop_mode_wb_timer entirely behind the BDI API
Don't leak the detaÑ–ls of the timer into the block layer, instead
initialize the timer in bdi_alloc and delete it in bdi_unregister.
Note that this means the timer is initialized (but not armed) for
non-block queues as well now.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Link: https://lore.kernel.org/r/20210809141744.1203023-2-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Christoph Hellwig [Wed, 4 Aug 2021 09:41:46 +0000 (11:41 +0200)]
dm: delay registering the gendisk
device mapper is currently the only outlier that tries to call
register_disk after add_disk, leading to fairly inconsistent state
of these block layer data structures. Instead change device-mapper
to just register the gendisk later now that the holder mechanism
can cope with that.
Note that this introduces a user visible change: the dm kobject is
now only visible after the initial table has been loaded.
Christoph Hellwig [Wed, 4 Aug 2021 09:41:45 +0000 (11:41 +0200)]
dm: move setting md->type into dm_setup_md_queue
Move setting md->type from both callers into dm_setup_md_queue.
This ensures that md->type is only set to a valid value after the queue
has been fully setup, something we'll rely on future changes.
Christoph Hellwig [Wed, 4 Aug 2021 09:41:43 +0000 (11:41 +0200)]
block: support delayed holder registration
device mapper needs to register holders before it is ready to do I/O.
Currently it does so by registering the disk early, which can leave
the disk and queue in a weird half state where the queue is registered
with the disk, except for sysfs and the elevator. And this state has
been a bit promlematic before, and will get more so when sorting out
the responsibilities between the queue and the disk.
Support registering holders on an initialized but not registered disk
instead by delaying the sysfs registration until the disk is registered.
Christoph Hellwig [Wed, 4 Aug 2021 09:41:41 +0000 (11:41 +0200)]
block: remove the extra kobject reference in bd_link_disk_holder
Since commit 0d02129e76ed ("block: merge struct block_device and struct
hd_struct") there is no way for the bdev to go away as long as there is
a holder, so remove the extra references.
Christoph Hellwig [Wed, 4 Aug 2021 09:41:40 +0000 (11:41 +0200)]
block: make the block holder code optional
Move the block holder code into a separate file as it is not in any way
related to the other block_dev.c code, and add a new selectable config
option for it so that we don't have to build it without any remapped
drivers selected.
The Kconfig symbol contains a _DEPRECATED suffix to match the comments
added in commit 49731baa41df
("block: restore multiple bd_link_disk_holder() support").
Bart Van Assche [Thu, 5 Aug 2021 17:42:00 +0000 (10:42 -0700)]
loop: Select I/O scheduler 'none' from inside add_disk()
We noticed that the user interface of Android devices becomes very slow
under memory pressure. This is because Android uses the zram driver on top
of the loop driver for swapping, because under memory pressure the swap
code alternates reads and writes quickly, because mq-deadline is the
default scheduler for loop devices and because mq-deadline delays writes by
five seconds for such a workload with default settings. Fix this by making
the kernel select I/O scheduler 'none' from inside add_disk() for loop
devices. This default can be overridden at any time from user space,
e.g. via a udev rule. This approach has an advantage compared to changing
the I/O scheduler from userspace from 'mq-deadline' into 'none', namely
that synchronize_rcu() does not get called.
This patch changes the default I/O scheduler for loop devices from
'mq-deadline' into 'none'.
Additionally, this patch reduces the Android boot time on my test setup
with 0.5 seconds compared to configuring the loop I/O scheduler from user
space.
Cc: Christoph Hellwig <hch@lst.de> Cc: Ming Lei <ming.lei@redhat.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Martijn Coenen <maco@android.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Signed-off-by: Bart Van Assche <bvanassche@acm.org> Link: https://lore.kernel.org/r/20210805174200.3250718-3-bvanassche@acm.org Signed-off-by: Jens Axboe <axboe@kernel.dk>
Bart Van Assche [Thu, 5 Aug 2021 17:41:59 +0000 (10:41 -0700)]
blk-mq: Introduce the BLK_MQ_F_NO_SCHED_BY_DEFAULT flag
elevator_get_default() uses the following algorithm to select an I/O
scheduler from inside add_disk():
- In case of a single hardware queue or if sharing hardware queues across
multiple request queues (BLK_MQ_F_TAG_HCTX_SHARED), use mq-deadline.
- Otherwise, use 'none'.
This is a good choice for most but not for all block drivers. Make it
possible to override the selection of mq-deadline with a new flag,
namely BLK_MQ_F_NO_SCHED_BY_DEFAULT.
Cc: Christoph Hellwig <hch@lst.de> Cc: Ming Lei <ming.lei@redhat.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Martijn Coenen <maco@android.com> Cc: Jaegeuk Kim <jaegeuk@kernel.org> Signed-off-by: Bart Van Assche <bvanassche@acm.org> Link: https://lore.kernel.org/r/20210805174200.3250718-2-bvanassche@acm.org Signed-off-by: Jens Axboe <axboe@kernel.dk>
Damien Le Moal [Tue, 13 Jul 2021 08:18:37 +0000 (17:18 +0900)]
block: remove blk-mq-sysfs dead code
In block/blk-mq-sysfs.c, struct blk_mq_ctx_sysfs_entry is not used to
define any attribute since the "mq" sysfs directory contains only
sub-directories (no attribute files). As a result, blk_mq_sysfs_show(),
blk_mq_sysfs_store(), and struct sysfs_ops blk_mq_sysfs_ops are all
unused and unnecessary. Remove all this unused code.
block: add a helper to raise a media changed event
Refactor disk_check_events() and move some code into disk_event_uevent().
Then add disk_force_media_change(), a helper which will be used by
devices to force issuing a DISK_EVENT_MEDIA_CHANGE event.
Associating uevents with block devices in userspace is difficult and racy:
the uevent netlink socket is lossy, and on slow and overloaded systems
has a very high latency.
Block devices do not have exclusive owners in userspace, any process can
set one up (e.g. loop devices). Moreover, device names can be reused
(e.g. loop0 can be reused again and again). A userspace process setting
up a block device and watching for its events cannot thus reliably tell
whether an event relates to the device it just set up or another earlier
instance with the same name.
Being able to set a UUID on a loop device would solve the race conditions.
But it does not allow to derive orderings from uevents: if you see a
uevent with a UUID that does not match the device you are waiting for,
you cannot tell whether it's because the right uevent has not arrived yet,
or it was already sent and you missed it. So you cannot tell whether you
should wait for it or not.
Associating a unique, monotonically increasing sequential number to the
lifetime of each block device, which can be retrieved with an ioctl
immediately upon setting it up, allows to solve the race conditions with
uevents, and also allows userspace processes to know whether they should
wait for the uevent they need or if it was dropped and thus they should
move on.
Additionally, increment the disk sequence number when the media change,
i.e. on DISK_EVENT_MEDIA_CHANGE event.
I have compiled the kernel with a cross compiler "hppa-linux-gnu-" v9.3.0
on x86-64 host machine. I got the following warning:
block/genhd.c: In function ‘diskstats_show’:
block/genhd.c:1227:1: warning: the frame size of 1688 bytes is larger
than 1280 bytes [-Wframe-larger-than=]
1227 | }
By Reduced the stack footprint by using the %pg printk specifier instead
of disk_name to remove the need for the on-stack buffer.
Signed-off-by: Abd-Alrhman Masalkhi <abd.masalkhi@gmail.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Link: https://lore.kernel.org/r/20210727062518.122108-2-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Christoph Hellwig [Thu, 22 Jul 2021 07:54:02 +0000 (09:54 +0200)]
block: remove bdput
Now that we've stopped using inode references for anything meaninful
in the block layer get rid of the helper to put it and just open code
the call to iput on the block_device inode.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Josef Bacik <josef@toxicpanda.com> Reviewed-by: Chaitanya Kulkarni <ckulkarnilinux@gmail.com> Link: https://lore.kernel.org/r/20210722075402.983367-10-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Christoph Hellwig [Thu, 22 Jul 2021 07:53:58 +0000 (09:53 +0200)]
block: change the refcounting for partitions
Instead of acquiring an inode reference on open make sure partitions
always hold device model references to the disk while alive, and switch
open to grab only a device model reference to the opened block device.
If that is a partition the disk reference is transitively held by the
partition already.
Christoph Hellwig [Thu, 22 Jul 2021 07:53:56 +0000 (09:53 +0200)]
block: unhash the whole device inode earlier
Unhash the whole device inode early in del_gendisk. This allows to
remove the first GENHD_FL_UP check in the open path as we simply
won't find a just removed inode. The second non-racy check after
taking open_mutex is still kept.
Christoph Hellwig [Tue, 27 Jul 2021 05:56:38 +0000 (07:56 +0200)]
dm-writecache: use bvec_kmap_local instead of bvec_kmap_irq
There is no need to disable interrupts in bio_copy_block, and the local
only mappings helps to avoid any sort of problems with stray writes
into the bio data.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Ira Weiny <ira.weiny@intel.com> Link: https://lore.kernel.org/r/20210727055646.118787-8-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Oliver Hartkopp [Wed, 14 Jul 2021 19:56:55 +0000 (21:56 +0200)]
ioprio: move user space relevant ioprio bits to UAPI includes
systemd added a modified copy of include/linux/ioprio.h into its
code to get the relevant content definitions for the exposed
ioprio_[get|set] system calls.
Move the user space relevant ioprio bits to the UAPI includes to be
able to use the ioprio_[get|set] syscalls as intended.
Linus Torvalds [Sun, 1 Aug 2021 19:25:30 +0000 (12:25 -0700)]
Merge tag 'perf-tools-fixes-for-v5.14-2021-08-01' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux
Pull perf tools fixes from Arnaldo Carvalho de Melo:
- Revert "perf map: Fix dso->nsinfo refcounting", this makes 'perf top'
abort, uncovering a design flaw on how namespace information is kept.
The fix for that is more than we can do right now, leave it for the
next merge window.
- Split --dump-raw-trace by AUX records for ARM's CoreSight, fixing up
the decoding of some records.
- Fix PMU alias matching.
Thanks to James Clark and John Garry for these fixes.
* tag 'perf-tools-fixes-for-v5.14-2021-08-01' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux:
Revert "perf map: Fix dso->nsinfo refcounting"
perf pmu: Fix alias matching
perf cs-etm: Split --dump-raw-trace by AUX records
Linus Torvalds [Sun, 1 Aug 2021 19:18:44 +0000 (12:18 -0700)]
Merge tag 'powerpc-5.14-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux
Pull powerpc fixes from Michael Ellerman:
- Don't use r30 in VDSO code, to avoid breaking existing Go lang
programs.
- Change an export symbol to allow non-GPL modules to use spinlocks
again.
Thanks to Paul Menzel, and Srikar Dronamraju.
* tag 'powerpc-5.14-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
powerpc/vdso: Don't use r30 to avoid breaking Go lang
powerpc/pseries: Fix regression while building external modules
Linus Torvalds [Sun, 1 Aug 2021 19:07:23 +0000 (12:07 -0700)]
Merge tag 'xfs-5.14-fixes-2' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
Pull xfs fixes from Darrick Wong:
"This contains a bunch of bug fixes in XFS.
Dave and I have been busy the last couple of weeks to find and fix as
many log recovery bugs as we can find; here are the results so far. Go
fstests -g recoveryloop! ;)
- Fix a number of coordination bugs relating to cache flushes for
metadata writeback, cache flushes for multi-buffer log writes, and
FUA writes for single-buffer log writes
- Fix a bug with incorrect replay of attr3 blocks
- Fix unnecessary stalls when flushing logs to disk
- Fix spoofing problems when recovering realtime bitmap blocks"
* tag 'xfs-5.14-fixes-2' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
xfs: prevent spoofing of rtbitmap blocks when recovering buffers
xfs: limit iclog tail updates
xfs: need to see iclog flags in tracing
xfs: Enforce attr3 buffer recovery order
xfs: logging the on disk inode LSN can make it go backwards
xfs: avoid unnecessary waits in xfs_log_force_lsn()
xfs: log forces imply data device cache flushes
xfs: factor out forced iclog flushes
xfs: fix ordering violation between cache flushes and tail updates
xfs: fold __xlog_state_release_iclog into xlog_state_release_iclog
xfs: external logs need to flush data device
xfs: flush data dev on external log write
Merge tag '5.14-rc3-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6
Pull cifs fixes from Steve French:
"Three cifs/smb3 fixes, including two for stable, and a fix for an
fallocate problem noticed by Clang"
* tag '5.14-rc3-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6:
cifs: add missing parsing of backupuid
smb3: rc uninitialized in one fallocate path
SMB3: fix readpage for large swap cache
- mlx5e: RX, avoid possible data corruption w/ relaxed ordering and
LRO
- phy: re-add check for PHY_BRCM_DIS_TXCRXC_NOENRGY on the BCM54811
PHY
- sctp: fix return value check in __sctp_rcv_asconf_lookup
Previous releases - always broken:
- bpf:
- more spectre corner case fixes, introduce a BPF nospec
instruction for mitigating Spectre v4
- fix OOB read when printing XDP link fdinfo
- sockmap: fix cleanup related races
- mac80211: fix enabling 4-address mode on a sta vif after assoc
- can:
- raw: raw_setsockopt(): fix raw_rcv panic for sock UAF
- j1939: j1939_session_deactivate(): clarify lifetime of session
object, avoid UAF
- fix number of identical memory leaks in USB drivers
- tipc:
- do not blindly write skb_shinfo frags when doing decryption
- fix sleeping in tipc accept routine"
* tag 'net-5.14-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (91 commits)
gve: Update MAINTAINERS list
can: esd_usb2: fix memory leak
can: ems_usb: fix memory leak
can: usb_8dev: fix memory leak
can: mcba_usb_start(): add missing urb->transfer_dma initialization
can: hi311x: fix a signedness bug in hi3110_cmd()
MAINTAINERS: add Yasushi SHOJI as reviewer for the Microchip CAN BUS Analyzer Tool driver
bpf: Fix leakage due to insufficient speculative store bypass mitigation
bpf: Introduce BPF nospec instruction for mitigating Spectre v4
sis900: Fix missing pci_disable_device() in probe and remove
net: let flow have same hash in two directions
nfc: nfcsim: fix use after free during module unload
tulip: windbond-840: Fix missing pci_disable_device() in probe and remove
sctp: fix return value check in __sctp_rcv_asconf_lookup
nfc: s3fwrn5: fix undefined parameter values in dev_err()
net/mlx5: Fix mlx5_vport_tbl_attr chain from u16 to u32
net/mlx5e: Fix nullptr in mlx5e_hairpin_get_mdev()
net/mlx5: Unload device upon firmware fatal error
net/mlx5e: Fix page allocation failure for ptp-RQ over SF
net/mlx5e: Fix page allocation failure for trap-RQ over SF
...
Merge tag 'acpi-5.14-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Pull ACPI fixes from Rafael Wysocki:
"These revert a recent IRQ resources handling modification that turned
out to be problematic, fix suspend-to-idle handling on AMD platforms
to take upcoming systems into account properly and fix the retrieval
of the DPTF attributes of the PCH FIVR.
Specifics:
- Revert recent change of the ACPI IRQ resources handling that
attempted to improve the ACPI IRQ override selection logic, but
introduced serious regressions on some systems (Hui Wang).
- Fix up quirks for AMD platforms in the suspend-to-idle support code
so as to take upcoming systems using uPEP HID AMDI007 into account
as appropriate (Mario Limonciello).
- Fix the code retrieving DPTF attributes of the PCH FIVR so that it
agrees on the return data type with the ACPI control method
evaluated for this purpose (Srinivas Pandruvada)"
* tag 'acpi-5.14-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
ACPI: DPTF: Fix reading of attributes
Revert "ACPI: resources: Add checks for ACPI IRQ override"
ACPI: PM: Add support for upcoming AMD uPEP HID AMDI007
Since commit 1b6b26ae7053 ("pipe: fix and clarify pipe write wakeup
logic") we have sanitized the pipe write logic, and would only try to
wake up readers if they needed it.
In particular, if the pipe already had data in it before the write,
there was no point in trying to wake up a reader, since any existing
readers must have been aware of the pre-existing data already. Doing
extraneous wakeups will only cause potential thundering herd problems.
However, it turns out that some Android libraries have misused the EPOLL
interface, and expected "edge triggered" be to "any new write will
trigger it". Even if there was no edge in sight.
Quoting Sandeep Patil:
"The commit 1b6b26ae7053 ('pipe: fix and clarify pipe write wakeup
logic') changed pipe write logic to wakeup readers only if the pipe
was empty at the time of write. However, there are libraries that
relied upon the older behavior for notification scheme similar to
what's described in [1]
One such library 'realm-core'[2] is used by numerous Android
applications. The library uses a similar notification mechanism as GNU
Make but it never drains the pipe until it is full. When Android moved
to v5.10 kernel, all applications using this library stopped working.
The library has since been fixed[3] but it will be a while before all
applications incorporate the updated library"
Our regression rule for the kernel is that if applications break from
new behavior, it's a regression, even if it was because the application
did something patently wrong. Also note the original report [4] by
Michal Kerrisk about a test for this epoll behavior - but at that point
we didn't know of any actual broken use case.
So add the extraneous wakeup, to approximate the old behavior.
[ I say "approximate", because the exact old behavior was to do a wakeup
not for each write(), but for each pipe buffer chunk that was filled
in. The behavior introduced by this change is not that - this is just
"every write will cause a wakeup, whether necessary or not", which
seems to be sufficient for the broken library use. ]
It's worth noting that this adds the extraneous wakeup only for the
write side, while the read side still considers the "edge" to be purely
about reading enough from the pipe to allow further writes.
See commit f467a6a66419 ("pipe: fix and clarify pipe read wakeup logic")
for the pipe read case, which remains that "only wake up if the pipe was
full, and we read something from it".
This makes 'perf top' abort in some cases, and the right fix will
involve surgery that is too much to do at this stage, so revert for now
and fix it in the next merge window.
projects / linux.git / log