NFSv4.1 supports an optional lock notification feature which notifies
the client when a lock comes available. (Normally NFSv4 clients just
poll for locks if necessary.) To make that work, we need to request a
blocking lock from the filesystem.
We turned that off for NFS in commit f657f8eef3 ("nfs: don't atempt
blocking locks on nfs reexports") [sic] because it actually blocks the
nfsd thread while waiting for the lock.
Thanks to Vasily Averin for pointing out that NFS isn't the only
filesystem with that problem.
Any filesystem that leaves ->lock NULL will use posix_lock_file(), which
does the right thing. Simplest is just to assume that any filesystem
that defines its own ->lock is not safe to request a blocking lock from.
So, this patch mostly reverts commit f657f8eef3 ("nfs: don't atempt
blocking locks on nfs reexports") [sic] and commit b840be2f00 ("lockd:
don't attempt blocking locks on nfs reexports"), and instead uses a
check of ->lock (Vasily's suggestion) to decide whether to support
blocking lock notifications on a given filesystem. Also add a little
documentation.
Perhaps someday we could add back an export flag later to allow
filesystems with "good" ->lock methods to support blocking lock
notifications.
Reported-by: Vasily Averin <vvs@virtuozzo.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
[ cel: Description rewritten to address checkpatch nits ]
[ cel: Fixed warning when SUNRPC debugging is disabled ]
[ cel: Fixed NULL check ]
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Vasily Averin <vvs@virtuozzo.com>
All profile_handoff_task does is notify the task_free_notifier chain.
The helpers task_handoff_register and task_handoff_unregister are used
to add and delete entries from that chain and are never called.
So remove the dead code and make it much easier to read and reason
about __put_task_struct.
Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
Link: https://lkml.kernel.org/r/87fspyw6m0.fsf@email.froward.int.ebiederm.org
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
When I say remove I mean remove. All profile_task_exit and
profile_munmap do is call a blocking notifier chain. The helpers
profile_task_register and profile_task_unregister are not called
anywhere in the tree. Which means this is all dead code.
So remove the dead code and make it easier to read do_exit.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lkml.kernel.org/r/20220103213312.9144-1-ebiederm@xmission.com
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
This helper is misleading. It tests for an ongoing exec as well as
the process having received a fatal signal.
Sometimes it is appropriate to treat an on-going exec differently than
a process that is shutting down due to a fatal signal. In particular
taking the fast path out of exit_signals instead of retargeting
signals is not appropriate during exec, and not changing the the exit
code in do_group_exit during exec.
Removing the helper makes it more obvious what is going on as both
cases must be coded for explicitly.
While removing the helper fix the two cases where I have observed
using signal_group_exit resulted in the wrong result.
In exit_signals only test for SIGNAL_GROUP_EXIT so that signals are
retargetted during an exec.
In do_group_exit use 0 as the exit code during an exec as de_thread
does not set group_exit_code. As best as I can determine
group_exit_code has been is set to 0 most of the time during
de_thread. During a thread group stop group_exit_code is set to the
stop signal and when the thread group receives SIGCONT group_exit_code
is reset to 0.
Link: https://lkml.kernel.org/r/20211213225350.27481-8-ebiederm@xmission.com
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
The only remaining user of group_exit_task is exec. Rename the field
so that it is clear which part of the code uses it.
Update the comment above the definition of group_exec_task to document
how it is currently used.
Link: https://lkml.kernel.org/r/20211213225350.27481-7-ebiederm@xmission.com
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
After the previous cleanups "signal->core_state" is set whenever
SIGNAL_GROUP_COREDUMP is set and "signal->core_state" is tested
whenver the code wants to know if a coredump is in progress. The
remaining tests of SIGNAL_GROUP_COREDUMP also test to see if
SIGNAL_GROUP_EXIT is set. Similarly the only place that sets
SIGNAL_GROUP_COREDUMP also sets SIGNAL_GROUP_EXIT.
Which makes SIGNAL_GROUP_COREDUMP unecessary and redundant. So stop
setting SIGNAL_GROUP_COREDUMP, stop testing SIGNAL_GROUP_COREDUMP, and
remove it's definition.
With the setting of SIGNAL_GROUP_COREDUMP gone, coredump_finish no
longer needs to clear SIGNAL_GROUP_COREDUMP out of signal->flags
by setting SIGNAL_GROUP_EXIT.
Link: https://lkml.kernel.org/r/20211213225350.27481-5-ebiederm@xmission.com
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
When CONFIG_SYSFS is not enabled provide stubs for the helper functions
to not break their callers.
Fixes: 539b9c94ac ("power: supply: add helpers for charge_behaviour sysfs")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
Link: https://lore.kernel.org/r/20220108153158.189489-1-linux@weissschuh.net
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
The point of using set_child_tid to hold the kthread pointer was that
it already did what is necessary. There are now restrictions on when
set_child_tid can be initialized and when set_child_tid can be used in
schedule_tail. Which indicates that continuing to use set_child_tid
to hold the kthread pointer is a bad idea.
Instead of continuing to use the set_child_tid field of task_struct
generalize the pf_io_worker field of task_struct and use it to hold
the kthread pointer.
Rename pf_io_worker (which is a void * pointer) to worker_private so
it can be used to store kthreads struct kthread pointer. Update the
kthread code to store the kthread pointer in the worker_private field.
Remove the places where set_child_tid had to be dealt with carefully
because kthreads also used it.
Link: https://lkml.kernel.org/r/CAHk-=wgtFAA9SbVYg0gR1tqPMC17-NYcs0GQkaYg1bGhh1uJQQ@mail.gmail.com
Link: https://lkml.kernel.org/r/87a6grvqy8.fsf_-_@email.froward.int.ebiederm.org
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
We currently store large folios as 2^N consecutive entries. While this
consumes rather more memory than necessary, it also turns out to be buggy.
A writeback operation which starts within a tail page of a dirty folio will
not write back the folio as the xarray's dirty bit is only set on the
head index. With multi-index entries, the dirty bit will be found no
matter where in the folio the operation starts.
This does end up simplifying the page cache slightly, although not as
much as I had hoped.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Add a new helper function to help iterate over multi-index entries.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
All of its callers now call folio_batch_remove_exceptionals().
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
find_lock_entries() already only returned the head page of folios, so
convert it to return a folio_batch instead of a pagevec. That cascades
through converting truncate_inode_pages_range() to
delete_from_page_cache_batch() and page_cache_delete_batch().
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
The callers have all been converted to work on folios, so convert
find_get_entries() to return a batch of folios instead of pages.
We also now return multiple large folios in a single call.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Convert all callers of truncate_inode_page() to call
truncate_inode_folio() instead, and move the declaration to mm/internal.h.
Move the assertion that the caller is not passing in a tail page to
generic_error_remove_page(). We can't entirely remove the struct page
from the callers yet because the page pointer in the pvec might be a
shadow/dax/swap entry instead of actually a page.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Convert both callers of unmap_mapping_page() to call unmap_mapping_folio()
instead. Also move zap_details from linux/mm.h to mm/memory.c
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
amdgpu:
- suspend/resume fix
- fix runtime PM regression
-----BEGIN PGP SIGNATURE-----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=Y3vO
-----END PGP SIGNATURE-----
Merge tag 'drm-fixes-2022-01-07' of git://anongit.freedesktop.org/drm/drm
Pull drm fixes from Dave Airlie:
"There is only the amdgpu runtime pm regression fix in here:
amdgpu:
- suspend/resume fix
- fix runtime PM regression"
* tag 'drm-fixes-2022-01-07' of git://anongit.freedesktop.org/drm/drm:
drm/amdgpu: disable runpm if we are the primary adapter
fbdev: fbmem: add a helper to determine if an aperture is used by a fw fb
drm/amd/pm: keep the BACO feature enabled for suspend
Commit b42faeee71 ("spi: Add a PTP system timestamp
to the transfer structure") added an include of ptp_clock_kernel.h
to spi.h for struct ptp_system_timestamp but a forward declaration
is enough. Let's use that to limit the number of objects we have
to rebuild every time we touch networking headers.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://lore.kernel.org/r/20210904013140.2377609-1-kuba@kernel.org
Signed-off-by: Mark Brown <broonie@kernel.org>
This adds basic support for delivering 2 level event channels to a guest.
Initially, it only supports delivery via the IRQ routing table, triggered
by an eventfd. In order to do so, it has a kvm_xen_set_evtchn_fast()
function which will use the pre-mapped shared_info page if it already
exists and is still valid, while the slow path through the irqfd_inject
workqueue will remap the shared_info page if necessary.
It sets the bits in the shared_info page but not the vcpu_info; that is
deferred to __kvm_xen_has_interrupt() which raises the vector to the
appropriate vCPU.
Add a 'verbose' mode to xen_shinfo_test while adding test cases for this.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <20211210163625.2886-5-dwmw2@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This can be used in two modes. There is an atomic mode where the cached
mapping is accessed while holding the rwlock, and a mode where the
physical address is used by a vCPU in guest mode.
For the latter case, an invalidation will wake the vCPU with the new
KVM_REQ_GPC_INVALIDATE, and the architecture will need to refresh any
caches it still needs to access before entering guest mode again.
Only one vCPU can be targeted by the wake requests; it's simple enough
to make it wake all vCPUs or even a mask but I don't see a use case for
that additional complexity right now.
Invalidation happens from the invalidate_range_start MMU notifier, which
needs to be able to sleep in order to wake the vCPU and wait for it.
This means that revalidation potentially needs to "wait" for the MMU
operation to complete and the invalidate_range_end notifier to be
invoked. Like the vCPU when it takes a page fault in that period, we
just spin — fixing that in a future patch by implementing an actual
*wait* may be another part of shaving this particularly hirsute yak.
As noted in the comments in the function itself, the only case where
the invalidate_range_start notifier is expected to be called *without*
being able to sleep is when the OOM reaper is killing the process. In
that case, we expect the vCPU threads already to have exited, and thus
there will be nothing to wake, and no reason to wait. So we clear the
KVM_REQUEST_WAIT bit and send the request anyway, then complain loudly
if there actually *was* anything to wake up.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <20211210163625.2886-3-dwmw2@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The various kvm_write_guest() and mark_page_dirty() functions must only
ever be called in the context of an active vCPU, because if dirty ring
tracking is enabled it may simply oops when kvm_get_running_vcpu()
returns NULL for the vcpu and then kvm_dirty_ring_get() dereferences it.
This oops was reported by "butt3rflyh4ck" <butterflyhuangxx@gmail.com> in
https://lore.kernel.org/kvm/CAFcO6XOmoS7EacN_n6v4Txk7xL7iqRa2gABg3F7E3Naf5uG94g@mail.gmail.com/
That actual bug will be fixed under separate cover but this warning
should help to prevent new ones from being added.
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Message-Id: <20211210163625.2886-2-dwmw2@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
- Simplification of the 'vcpu first run' by integrating it into
KVM's 'pid change' flow
- Refactoring of the FP and SVE state tracking, also leading to
a simpler state and less shared data between EL1 and EL2 in
the nVHE case
- Tidy up the header file usage for the nvhe hyp object
- New HYP unsharing mechanism, finally allowing pages to be
unmapped from the Stage-1 EL2 page-tables
- Various pKVM cleanups around refcounting and sharing
- A couple of vgic fixes for bugs that would trigger once
the vcpu xarray rework is merged, but not sooner
- Add minimal support for ARMv8.7's PMU extension
- Rework kvm_pgtable initialisation ahead of the NV work
- New selftest for IRQ injection
- Teach selftests about the lack of default IPA space and
page sizes
- Expand sysreg selftest to deal with Pointer Authentication
- The usual bunch of cleanups and doc update
-----BEGIN PGP SIGNATURE-----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=HkZk
-----END PGP SIGNATURE-----
Merge tag 'kvmarm-5.17' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmarm/kvmarm into HEAD
KVM/arm64 updates for Linux 5.16
- Simplification of the 'vcpu first run' by integrating it into
KVM's 'pid change' flow
- Refactoring of the FP and SVE state tracking, also leading to
a simpler state and less shared data between EL1 and EL2 in
the nVHE case
- Tidy up the header file usage for the nvhe hyp object
- New HYP unsharing mechanism, finally allowing pages to be
unmapped from the Stage-1 EL2 page-tables
- Various pKVM cleanups around refcounting and sharing
- A couple of vgic fixes for bugs that would trigger once
the vcpu xarray rework is merged, but not sooner
- Add minimal support for ARMv8.7's PMU extension
- Rework kvm_pgtable initialisation ahead of the NV work
- New selftest for IRQ injection
- Teach selftests about the lack of default IPA space and
page sizes
- Expand sysreg selftest to deal with Pointer Authentication
- The usual bunch of cleanups and doc update
1) Expose FEC per lane block counters via ethtool
2) Trivial fixes/updates/cleanup to mlx5e netdev driver
3) Fix htmldoc build warning
4) Spread mlx5 SFs (sub-functions) to all available CPU cores: Commits 1..5
Shay Drory Says:
================
Before this patchset, mlx5 subfunction shared the same IRQs (MSI-X) with
their peers subfunctions, causing them to use same CPU cores.
In large scale, this is very undesirable, SFs use small number of cpu
cores and all of them will be packed on the same CPU cores, not
utilizing all CPU cores in the system.
In this patchset we want to achieve two things.
a) Spread IRQs used by SFs to all cpu cores
b) Pack less SFs in the same IRQ, will result in multiple IRQs per core.
In this patchset, we spread SFs over all online cpus available to mlx5
irqs in Round-Robin manner. e.g.: Whenever a SF is created, pick the next
CPU core with least number of SF IRQs bound to it, SFs will share IRQs on
the same core until a certain limit, when such limit is reached, we
request a new IRQ and add it to that CPU core IRQ pool, when out of IRQs,
pick any IRQ with least number of SF users.
This enhancement is done in order to achieve a better distribution of
the SFs over all the available CPUs, which reduces application latency,
as shown bellow.
Machine details:
Intel(R) Xeon(R) CPU E5-2697 v3 @ 2.60GHz with 56 cores.
PCI Express 3 with BW of 126 Gb/s.
ConnectX-5 Ex; EDR IB (100Gb/s) and 100GbE; dual-port QSFP28; PCIe4.0
x16.
Base line test description:
Single SF on the system. One instance of netperf is running on-top the
SF.
Numbers: latency = 15.136 usec, CPU Util = 35%
Test description:
There are 250 SFs on the system. There are 3 instances of netperf
running, on-top three different SFs, in parallel.
Perf numbers:
# netperf SFs latency(usec) latency CPU utilization
affinity affinity (lower is better) increase %
1 cpu=0 cpu={0} ~23 (app 1-3) 35% 75%
2 cpu=0,2,4 cpu={0} app 1: 21.625 30% 68% (CPU 0)
app 2-3: 16.5 9% 15% (CPU 2,4)
3 cpu=0 cpu={0,2,4} app 1: ~16 7% 84% (CPU 0)
app 2-3: ~17.9 14% 22% (CPU 2,4)
4 cpu=0,2,4 cpu={0,2,4} 15.2 (app 1-3) 0% 33% (CPU 0,2,4)
- The first two entries (#1 and #2) show current state. e.g.: SFs are
using the same CPU. The last two entries (#3 and #4) shows the latency
reduction improvement of this patch. e.g.: SFs are on different CPUs.
- Whenever we use several CPUs, in case there is a different CPU
utilization, write the utilization of each CPU separately.
- Whenever the latency result of the netperf instances were different,
write the latency of each netperf instances separately.
Commands:
- for netperf CPU=0:
$ for i in {1..3}; do taskset -c 0 netperf -H 1${i}.1.1.1 -t TCP_RR -- \
-o RT_LATENCY -r8 & done
- for netperf CPU=0,2,4
$ for i in {1..3}; do taskset -c $(( ($i - 1) * 2 )) netperf -H \
1${i}.1.1.1 -t TCP_RR -- -o RT_LATENCY -r8 & done
================
-----BEGIN PGP SIGNATURE-----
iQEzBAABCAAdFiEEGhZs6bAKwk/OTgTpSD+KveBX+j4FAmHXh+AACgkQSD+KveBX
+j68fQgAghUX4TFS2JSwa7+XSCtzz7GIu2Xrz8aWTAnydRLlNXuFuuHYcNed6I0l
7DaVOZwHp1tp3tnx3WMGPUU6ujDPEgasaDDblvG2UXix5LPVEHDXY44ittQX8mpC
SC8Yj9mNo6DSfOMUZklFDMbw57XuLJ+HEGnwnrOEEyLX7ruDXGEViUmVBd4IoC3B
F2fJHBkdTJfHWTJRB4pWbZD1dw7WbKd0RyPla3OkoHugEUCKnbjii8cMwNM64Bbp
Pjz/SiShVy+NTotqPzRNjcx7y4tHOXCYt33zt1VlGtdUxs5eCA5jkjHFz0jb12Lu
rvfHaBaU+elMKTw5G/WMGJxZQx0kEQ==
=VBWY
-----END PGP SIGNATURE-----
Merge tag 'mlx5-updates-2022-01-06' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux
Saeed Mahameed says:
====================
mlx5-updates-2022-01-06
1) Expose FEC per lane block counters via ethtool
2) Trivial fixes/updates/cleanup to mlx5e netdev driver
3) Fix htmldoc build warning
4) Spread mlx5 SFs (sub-functions) to all available CPU cores: Commits 1..5
Shay Drory Says:
================
Before this patchset, mlx5 subfunction shared the same IRQs (MSI-X) with
their peers subfunctions, causing them to use same CPU cores.
In large scale, this is very undesirable, SFs use small number of cpu
cores and all of them will be packed on the same CPU cores, not
utilizing all CPU cores in the system.
In this patchset we want to achieve two things.
a) Spread IRQs used by SFs to all cpu cores
b) Pack less SFs in the same IRQ, will result in multiple IRQs per core.
In this patchset, we spread SFs over all online cpus available to mlx5
irqs in Round-Robin manner. e.g.: Whenever a SF is created, pick the next
CPU core with least number of SF IRQs bound to it, SFs will share IRQs on
the same core until a certain limit, when such limit is reached, we
request a new IRQ and add it to that CPU core IRQ pool, when out of IRQs,
pick any IRQ with least number of SF users.
This enhancement is done in order to achieve a better distribution of
the SFs over all the available CPUs, which reduces application latency,
as shown bellow.
Machine details:
Intel(R) Xeon(R) CPU E5-2697 v3 @ 2.60GHz with 56 cores.
PCI Express 3 with BW of 126 Gb/s.
ConnectX-5 Ex; EDR IB (100Gb/s) and 100GbE; dual-port QSFP28; PCIe4.0
x16.
Base line test description:
Single SF on the system. One instance of netperf is running on-top the
SF.
Numbers: latency = 15.136 usec, CPU Util = 35%
Test description:
There are 250 SFs on the system. There are 3 instances of netperf
running, on-top three different SFs, in parallel.
Perf numbers:
# netperf SFs latency(usec) latency CPU utilization
affinity affinity (lower is better) increase %
1 cpu=0 cpu={0} ~23 (app 1-3) 35% 75%
2 cpu=0,2,4 cpu={0} app 1: 21.625 30% 68% (CPU 0)
app 2-3: 16.5 9% 15% (CPU 2,4)
3 cpu=0 cpu={0,2,4} app 1: ~16 7% 84% (CPU 0)
app 2-3: ~17.9 14% 22% (CPU 2,4)
4 cpu=0,2,4 cpu={0,2,4} 15.2 (app 1-3) 0% 33% (CPU 0,2,4)
- The first two entries (#1 and #2) show current state. e.g.: SFs are
using the same CPU. The last two entries (#3 and #4) shows the latency
reduction improvement of this patch. e.g.: SFs are on different CPUs.
- Whenever we use several CPUs, in case there is a different CPU
utilization, write the utilization of each CPU separately.
- Whenever the latency result of the netperf instances were different,
write the latency of each netperf instances separately.
Commands:
- for netperf CPU=0:
$ for i in {1..3}; do taskset -c 0 netperf -H 1${i}.1.1.1 -t TCP_RR -- \
-o RT_LATENCY -r8 & done
- for netperf CPU=0,2,4
$ for i in {1..3}; do taskset -c $(( ($i - 1) * 2 )) netperf -H \
1${i}.1.1.1 -t TCP_RR -- -o RT_LATENCY -r8 & done
================
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Cachefiles has a problem in that it needs to keep the backing file for a
cookie open whilst there are local modifications pending that need to be
written to it. However, we don't want to keep the file open indefinitely,
as that causes EMFILE/ENFILE/ENOMEM problems.
Reopening the cache file, however, is a problem if this is being done due
to writeback triggered by exit(). Some filesystems will oops if we try to
open a file in that context because they want to access current->fs or
other resources that have already been dismantled.
To get around this, I added the following:
(1) An inode flag, I_PINNING_FSCACHE_WB, to be set on a network filesystem
inode to indicate that we have a usage count on the cookie caching
that inode.
(2) A flag in struct writeback_control, unpinned_fscache_wb, that is set
when __writeback_single_inode() clears the last dirty page from
i_pages - at which point it clears I_PINNING_FSCACHE_WB and sets this
flag.
This has to be done here so that clearing I_PINNING_FSCACHE_WB can be
done atomically with the check of PAGECACHE_TAG_DIRTY that clears
I_DIRTY_PAGES.
(3) A function, fscache_set_page_dirty(), which if it is not set, sets
I_PINNING_FSCACHE_WB and calls fscache_use_cookie() to pin the cache
resources.
(4) A function, fscache_unpin_writeback(), to be called by ->write_inode()
to unuse the cookie.
(5) A function, fscache_clear_inode_writeback(), to be called when the
inode is evicted, before clear_inode() is called. This cleans up any
lingering I_PINNING_FSCACHE_WB.
The network filesystem can then use these tools to make sure that
fscache_write_to_cache() can write locally modified data to the cache as
well as to the server.
For the future, I'm working on write helpers for netfs lib that should
allow this facility to be removed by keeping track of the dirty regions
separately - but that's incomplete at the moment and is also going to be
affected by folios, one way or another, since it deals with pages
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/163819615157.215744.17623791756928043114.stgit@warthog.procyon.org.uk/ # v1
Link: https://lore.kernel.org/r/163906917856.143852.8224898306177154573.stgit@warthog.procyon.org.uk/ # v2
Link: https://lore.kernel.org/r/163967124567.1823006.14188359004568060298.stgit@warthog.procyon.org.uk/ # v3
Link: https://lore.kernel.org/r/164021524705.640689.17824932021727663017.stgit@warthog.procyon.org.uk/ # v4
Provide a higher-level function than fscache_write() to perform a write
from an inode's pagecache to the cache, whilst fending off concurrent
writes by means of the PG_fscache mark on a page:
void fscache_write_to_cache(struct fscache_cookie *cookie,
struct address_space *mapping,
loff_t start,
size_t len,
loff_t i_size,
netfs_io_terminated_t term_func,
void *term_func_priv,
bool caching);
If caching is false, this function does nothing except call (*term_func)()
if given. It assumes that, in such a case, PG_fscache will not have been
set on the pages.
Otherwise, if caching is true, this function requires the source pages to
have had PG_fscache set on them before calling. start and len define the
region of the file to be modified and i_size indicates the new file size.
The source pages are extracted from the mapping.
term_func and term_func_priv work as for fscache_write(). The PG_fscache
marks will be cleared at the end of the operation, before term_func is
called or the function otherwise returns.
There is an additonal helper function to clear the PG_fscache bits from a
range of pages:
void fscache_clear_page_bits(struct fscache_cookie *cookie,
struct address_space *mapping,
loff_t start, size_t len,
bool caching);
If caching is true, the pages to be managed are expected to be located on
mapping in the range defined by start and len. If caching is false, it
does nothing.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/163819614155.215744.5528123235123721230.stgit@warthog.procyon.org.uk/ # v1
Link: https://lore.kernel.org/r/163906916346.143852.15632773570362489926.stgit@warthog.procyon.org.uk/ # v2
Link: https://lore.kernel.org/r/163967123599.1823006.12946816026724657428.stgit@warthog.procyon.org.uk/ # v3
Link: https://lore.kernel.org/r/164021522672.640689.4381958316198807813.stgit@warthog.procyon.org.uk/ # v4
Provide a pair of functions to perform raw I/O on the cache. The first
function allows an arbitrary asynchronous direct-IO read to be made against
a cache object, though the read should be aligned and sized appropriately
for the backing device:
int fscache_read(struct netfs_cache_resources *cres,
loff_t start_pos,
struct iov_iter *iter,
enum netfs_read_from_hole read_hole,
netfs_io_terminated_t term_func,
void *term_func_priv);
The cache resources must have been previously initialised by
fscache_begin_read_operation(). A read operation is sent to the backing
filesystem, starting at start_pos within the file. The size of the read is
specified by the iterator, as is the location of the output buffer.
If there is a hole in the data it can be ignored and left to the backing
filesystem to deal with (NETFS_READ_HOLE_IGNORE), a hole at the beginning
can be skipped over and the buffer padded with zeros
(NETFS_READ_HOLE_CLEAR) or -ENODATA can be given (NETFS_READ_HOLE_FAIL).
If term_func is not NULL, the operation may be performed asynchronously.
Upon completion, successful or otherwise, (*term_func)() will be called and
passed term_func_priv, along with an error or the amount of data
transferred. If the op is run asynchronously, fscache_read() will return
-EIOCBQUEUED.
The second function allows an arbitrary asynchronous direct-IO write to be
made against a cache object, though the write should be aligned and sized
appropriately for the backing device:
int fscache_write(struct netfs_cache_resources *cres,
loff_t start_pos,
struct iov_iter *iter,
netfs_io_terminated_t term_func,
void *term_func_priv);
This works in very similar way to fscache_read(), except that there's no
need to deal with holes (they're just overwritten).
The caller is responsible for preventing concurrent overlapping writes.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/163819613224.215744.7877577215582621254.stgit@warthog.procyon.org.uk/ # v1
Link: https://lore.kernel.org/r/163906915386.143852.16936177636106480724.stgit@warthog.procyon.org.uk/ # v2
Link: https://lore.kernel.org/r/163967122632.1823006.7487049517698562172.stgit@warthog.procyon.org.uk/ # v3
Link: https://lore.kernel.org/r/164021521420.640689.12747258780542678309.stgit@warthog.procyon.org.uk/ # v4
Provide a function to begin a read operation:
int fscache_begin_read_operation(
struct netfs_cache_resources *cres,
struct fscache_cookie *cookie)
This is primarily intended to be called by network filesystems on behalf of
netfslib, but may also be called to use the I/O access functions directly.
It attaches the resources required by the cache to cres struct from the
supplied cookie.
This holds access to the cache behind the cookie for the duration of the
operation and forces cache withdrawal and cookie invalidation to perform
synchronisation on the operation. cres->inval_counter is set from the
cookie at this point so that it can be compared at the end of the
operation.
Note that this does not guarantee that the cache state is fully set up and
able to perform I/O immediately; looking up and creation may be left in
progress in the background. The operations intended to be called by the
network filesystem, such as reading and writing, are expected to wait for
the cookie to move to the correct state.
This will, however, potentially sleep, waiting for a certain minimum state
to be set or for operations such as invalidate to advance far enough that
I/O can resume.
Also provide a function for the cache to call to wait for the cache object
to get to a state where it can be used for certain things:
bool fscache_wait_for_operation(struct netfs_cache_resources *cres,
enum fscache_want_stage stage);
This looks at the cache resources provided by the begin function and waits
for them to get to an appropriate stage. There's a choice of wanting just
some parameters (FSCACHE_WANT_PARAM) or the ability to do I/O
(FSCACHE_WANT_READ or FSCACHE_WANT_WRITE).
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/163819603692.215744.146724961588817028.stgit@warthog.procyon.org.uk/ # v1
Link: https://lore.kernel.org/r/163906910672.143852.13856103384424986357.stgit@warthog.procyon.org.uk/ # v2
Link: https://lore.kernel.org/r/163967110245.1823006.2239170567540431836.stgit@warthog.procyon.org.uk/ # v3
Link: https://lore.kernel.org/r/164021513617.640689.16627329360866150606.stgit@warthog.procyon.org.uk/ # v4
Provide a pair of functions to count the number of users of a cookie (open
files, writeback, invalidation, resizing, reads, writes), to obtain and pin
resources for the cookie and to prevent culling for the whilst there are
users.
The first function marks a cookie as being in use:
void fscache_use_cookie(struct fscache_cookie *cookie,
bool will_modify);
The caller should indicate the cookie to use and whether or not the caller
is in a context that may modify the cookie (e.g. a file open O_RDWR).
If the cookie is not already resourced, fscache will ask the cache backend
in the background to do whatever it needs to look up, create or otherwise
obtain the resources necessary to access data. This is pinned to the
cookie and may not be culled, though it may be withdrawn if the cache as a
whole is withdrawn.
The second function removes the in-use mark from a cookie and, optionally,
updates the coherency data:
void fscache_unuse_cookie(struct fscache_cookie *cookie,
const void *aux_data,
const loff_t *object_size);
If non-NULL, the aux_data buffer and/or the object_size will be saved into
the cookie and will be set on the backing store when the object is
committed.
If this removes the last usage on a cookie, the cookie is placed onto an
LRU list from which it will be removed and closed after a couple of seconds
if it doesn't get reused. This prevents resource overload in the cache -
in particular it prevents it from holding too many files open.
Changes
=======
ver #2:
- Fix fscache_unuse_cookie() to use atomic_dec_and_lock() to avoid a
potential race if the cookie gets reused before it completes the
unusement.
- Added missing transition to LRU_DISCARDING state.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/163819600612.215744.13678350304176542741.stgit@warthog.procyon.org.uk/ # v1
Link: https://lore.kernel.org/r/163906907567.143852.16979631199380722019.stgit@warthog.procyon.org.uk/ # v2
Link: https://lore.kernel.org/r/163967106467.1823006.6790864931048582667.stgit@warthog.procyon.org.uk/ # v3
Link: https://lore.kernel.org/r/164021511674.640689.10084988363699111860.stgit@warthog.procyon.org.uk/ # v4
Add a number of helper functions to manage access to a cookie, pinning the
cache object in place for the duration to prevent cache withdrawal from
removing it:
(1) void fscache_init_access_gate(struct fscache_cookie *cookie);
This function initialises the access count when a cache binds to a
cookie. An extra ref is taken on the access count to prevent wakeups
while the cache is active. We're only interested in the wakeup when a
cookie is being withdrawn and we're waiting for it to quiesce - at
which point the counter will be decremented before the wait.
The FSCACHE_COOKIE_NACC_ELEVATED flag is set on the cookie to keep
track of the extra ref in order to handle a race between
relinquishment and withdrawal both trying to drop the extra ref.
(2) bool fscache_begin_cookie_access(struct fscache_cookie *cookie,
enum fscache_access_trace why);
This function attempts to begin access upon a cookie, pinning it in
place if it's cached. If successful, it returns true and leaves a the
access count incremented.
(3) void fscache_end_cookie_access(struct fscache_cookie *cookie,
enum fscache_access_trace why);
This function drops the access count obtained by (2), permitting
object withdrawal to take place when it reaches zero.
A tracepoint is provided to track changes to the access counter on a
cookie.
Changes
=======
ver #2:
- Don't hold n_accesses elevated whilst cache is bound to a cookie, but
rather add a flag that prevents the state machine from being queued when
n_accesses reaches 0.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/163819595085.215744.1706073049250505427.stgit@warthog.procyon.org.uk/ # v1
Link: https://lore.kernel.org/r/163906895313.143852.10141619544149102193.stgit@warthog.procyon.org.uk/ # v2
Link: https://lore.kernel.org/r/163967095980.1823006.1133648159424418877.stgit@warthog.procyon.org.uk/ # v3
Link: https://lore.kernel.org/r/164021503063.640689.8870918985269528670.stgit@warthog.procyon.org.uk/ # v4
Add a pair of helper functions to manage access to a volume, pinning the
volume in place for the duration to prevent cache withdrawal from removing
it:
bool fscache_begin_volume_access(struct fscache_volume *volume,
enum fscache_access_trace why);
void fscache_end_volume_access(struct fscache_volume *volume,
enum fscache_access_trace why);
The way the access gate on the volume works/will work is:
(1) If the cache tests as not live (state is not FSCACHE_CACHE_IS_ACTIVE),
then we return false to indicate access was not permitted.
(2) If the cache tests as live, then we increment the volume's n_accesses
count and then recheck the cache liveness, ending the access if it
ceased to be live.
(3) When we end the access, we decrement the volume's n_accesses and wake
up the any waiters if it reaches 0.
(4) Whilst the cache is caching, the volume's n_accesses is kept
artificially incremented to prevent wakeups from happening.
(5) When the cache is taken offline, the state is changed to prevent new
accesses, the volume's n_accesses is decremented and we wait for it to
become 0.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/163819594158.215744.8285859817391683254.stgit@warthog.procyon.org.uk/ # v1
Link: https://lore.kernel.org/r/163906894315.143852.5454793807544710479.stgit@warthog.procyon.org.uk/ # v2
Link: https://lore.kernel.org/r/163967095028.1823006.9173132503876627466.stgit@warthog.procyon.org.uk/ # v3
Link: https://lore.kernel.org/r/164021501546.640689.9631510472149608443.stgit@warthog.procyon.org.uk/ # v4
Add functions to the fscache API to allow data file cookies to be acquired
and relinquished by the network filesystem. It is intended that the
filesystem will create such cookies per-inode under a volume.
To request a cookie, the filesystem should call:
struct fscache_cookie *
fscache_acquire_cookie(struct fscache_volume *volume,
u8 advice,
const void *index_key,
size_t index_key_len,
const void *aux_data,
size_t aux_data_len,
loff_t object_size)
The filesystem must first have created a volume cookie, which is passed in
here. If it passes in NULL then the function will just return a NULL
cookie.
A binary key should be passed in index_key and is of size index_key_len.
This is saved in the cookie and is used to locate the associated data in
the cache.
A coherency data buffer of size aux_data_len will be allocated and
initialised from the buffer pointed to by aux_data. This is used to
validate cache objects when they're opened and is stored on disk with them
when they're committed. The data is stored in the cookie and will be
updateable by various functions in later patches.
The object_size must also be given. This is also used to perform a
coherency check and to size the backing storage appropriately.
This function disallows a cookie from being acquired twice in parallel,
though it will cause the second user to wait if the first is busy
relinquishing its cookie.
When a network filesystem has finished with a cookie, it should call:
void
fscache_relinquish_cookie(struct fscache_volume *volume,
bool retire)
If retire is true, any backing data will be discarded immediately.
Changes
=======
ver #3:
- fscache_hash()'s size parameter is now in bytes. Use __le32 as the unit
to round up to.
- When comparing cookies, simply see if the attributes are the same rather
than subtracting them to produce a strcmp-style return[1].
- Add a check to see if the cookie is still hashed at the point of
freeing.
ver #2:
- Don't hold n_accesses elevated whilst cache is bound to a cookie, but
rather add a flag that prevents the state machine from being queued when
n_accesses reaches 0.
- Remove the unused cookie pointer field from the fscache_acquire
tracepoint.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/CAHk-=whtkzB446+hX0zdLsdcUJsJ=8_-0S1mE_R+YurThfUbLA@mail.gmail.com/ [1]
Link: https://lore.kernel.org/r/163819590658.215744.14934902514281054323.stgit@warthog.procyon.org.uk/ # v1
Link: https://lore.kernel.org/r/163906891983.143852.6219772337558577395.stgit@warthog.procyon.org.uk/ # v2
Link: https://lore.kernel.org/r/163967088507.1823006.12659006350221417165.stgit@warthog.procyon.org.uk/ # v3
Link: https://lore.kernel.org/r/164021498432.640689.12743483856927722772.stgit@warthog.procyon.org.uk/ # v4
Add functions to the fscache API to allow volumes to be acquired and
relinquished by the network filesystem. A volume is an index of data
storage cache objects. A volume is represented by a volume cookie in the
API. A filesystem would typically create a volume for a superblock and
then create per-inode cookies within it.
To request a volume, the filesystem calls:
struct fscache_volume *
fscache_acquire_volume(const char *volume_key,
const char *cache_name,
const void *coherency_data,
size_t coherency_len)
The volume_key is a printable string used to match the volume in the cache.
It should not contain any '/' characters. For AFS, for example, this would
be "afs,<cellname>,<volume_id>", e.g. "afs,example.com,523001".
The cache_name can be NULL, but if not it should be a string indicating the
name of the cache to use if there's more than one available.
The coherency data, if given, is an arbitrarily-sized blob that's attached
to the volume and is compared when the volume is looked up. If it doesn't
match, the old volume is judged to be out of date and it and everything
within it is discarded.
Acquiring a volume twice concurrently is disallowed, though the function
will wait if an old volume cookie is being relinquishing.
When a network filesystem has finished with a volume, it should return the
volume cookie by calling:
void
fscache_relinquish_volume(struct fscache_volume *volume,
const void *coherency_data,
bool invalidate)
If invalidate is true, the entire volume will be discarded; if false, the
volume will be synced and the coherency data will be updated.
Changes
=======
ver #4:
- Removed an extraneous param from kdoc on fscache_relinquish_volume()[3].
ver #3:
- fscache_hash()'s size parameter is now in bytes. Use __le32 as the unit
to round up to.
- When comparing cookies, simply see if the attributes are the same rather
than subtracting them to produce a strcmp-style return[2].
- Make the coherency data an arbitrary blob rather than a u64, but don't
store it for the moment.
ver #2:
- Fix error check[1].
- Make a fscache_acquire_volume() return errors, including EBUSY if a
conflicting volume cookie already exists. No error is printed now -
that's left to the netfs.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
cc: linux-cachefs@redhat.com
Link: https://lore.kernel.org/r/20211203095608.GC2480@kili/ [1]
Link: https://lore.kernel.org/r/CAHk-=whtkzB446+hX0zdLsdcUJsJ=8_-0S1mE_R+YurThfUbLA@mail.gmail.com/ [2]
Link: https://lore.kernel.org/r/20211220224646.30e8205c@canb.auug.org.au/ [3]
Link: https://lore.kernel.org/r/163819588944.215744.1629085755564865996.stgit@warthog.procyon.org.uk/ # v1
Link: https://lore.kernel.org/r/163906890630.143852.13972180614535611154.stgit@warthog.procyon.org.uk/ # v2
Link: https://lore.kernel.org/r/163967086836.1823006.8191672796841981763.stgit@warthog.procyon.org.uk/ # v3
Link: https://lore.kernel.org/r/164021495816.640689.4403156093668590217.stgit@warthog.procyon.org.uk/ # v4
