| Age | Commit message (Collapse) | Author |
|---|
|
Returning u64 doesn't make sense: max header->obj_order is 25 and
ceph_file_layout::object_size is u32.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
|
|
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
|
|
As explained in the previous commit, rbd_obj_request machinery (and
rbd_osd_req_create() in particular) shouldn't be used for working with
metadata objects.
Switch to the recently added ceph_osdc_call(). It assumes single pages
for outbound and inbound buffers, but that's OK - none of the callers
need more than that. These pages need to be allocated (messenger is in
dire need of proper iterator interface!), but we are swapping for
pages[] and pagelist allocations in the existing code.
Kill class_name argument - all rbd methods are under "rbd".
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
|
|
To spare checking for "this reply fits into a page, but does it fit
into my buffer?" in some callers, osd_req_op_cls_response_data_pages()
needs to know how big it is.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
|
|
rbd_obj_request machinery is completely unnecessary here; all that's
being done is fetching a metadata object - no striping, cloning, etc.
More importantly, rbd_osd_req_create() grabs pool id from layout and
that is becoming a data pool id.
Kill offset argument - all metadata objects are small and read in full.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
|
|
No reason to delay it until image_id is known. This will be required
by some rbd_obj_method_sync() callers, after rbd_obj_method_sync() is
changed to take oloc.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
|
|
Image format 1 is deprecated and format 2 doesn't have these. Also,
__rbd_dev_create() takes care of zeroing (or otherwise initializing)
format 2 specific fields.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
|
|
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>
|
|
... to accommodate potentially very wide EC pools. This increases the
size of a typical rbd ceph_osd_request by ~12% (from 1040 to 1168 bytes),
but I'd rather go future proof here.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Sage Weil <sage@redhat.com>
|
|
With EC overwrites maturing, the kernel client will be getting exposed
to potentially very wide EC pools. While "min(pi->size, X)" works fine
when the cluster is stable and happy, truncating OSD sets interferes
with resend logic (ceph_is_new_interval(), etc). Abort the mapping if
the pool is too wide, assigning the request to the homeless session.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Sage Weil <sage@redhat.com>
|
|
Much like Arlo Guthrie, I decided that one big pile is better than two
little piles.
Reflects ceph.git commit 95c2df6c7e0b22d2ea9d91db500cf8b9441c73ba.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
Then add it to the working state. It would be very nice if we didn't
have to take a lock to calculate a crush placement. By moving the
permutation array into the working data, we can treat the CRUSH map as
immutable.
Reflects ceph.git commit cbcd039651c0569551cb90d26ce27e1432671f2a.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
Simplify osdmap_decode() and osdmap_apply_incremental() a bit.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
This was causing a build failure for openrisc when using musl and
gcc 5.4.0 since the file is not available in the toolchain.
It doesnt seem this is needed and removing it does not cause any build
warnings for me.
Signed-off-by: Stafford Horne <shorne@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
In commit c3f4688a08f (ceph: don't set req->r_locked_dir in
ceph_d_revalidate), we changed the code to do a GETATTR instead of a
LOOKUP as the parent info isn't strictly necessary to revalidate the
dentry. What we missed there though is that in order to update the lease
on the dentry after revalidating it, we _do_ need parent info.
Change ceph_d_revalidate back to doing a LOOKUP instead of a GETATTR so
that we can get the parent info in order to update the lease from
ceph_fill_trace. Note that we set req->r_parent here, but we cannot set
the CEPH_MDS_R_PARENT_LOCKED flag as we can't guarantee that it is.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
We don't really require that the parent be locked in order to update the
lease on a dentry. Lease info is protected by the d_lock. In the event
that the parent is not locked in ceph_fill_trace, and we have both
parent and target info, go ahead and update the dentry lease.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
In a later patch, we're going to need to allow ceph_fill_trace to
update the dentry's lease when the parent is not locked. This is
potentially racy though -- by the time we get around to processing the
trace, the parent may have already changed.
Change update_dentry_lease to take a ceph_vino pointer and use that to
ensure that the dentry's parent still matches it before updating the
lease.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
This if block updates the dentry lease even in the case where
the MDS didn't grant one.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
struct ceph_mds_request has an r_locked_dir pointer, which is set to
indicate the parent inode and that its i_rwsem is locked. In some
critical places, we need to be able to indicate the parent inode to the
request handling code, even when its i_rwsem may not be locked.
Most of the code that operates on r_locked_dir doesn't require that the
i_rwsem be locked. We only really need it to handle manipulation of the
dcache. The rest (filling of the inode, updating dentry leases, etc.)
already has its own locking.
Add a new r_req_flags bit that indicates whether the parent is locked
when doing the request, and rename the pointer to "r_parent". For now,
all the places that set r_parent also set this flag, but that will
change in a later patch.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
Currently, we have a bunch of bool flags in struct ceph_mds_request. We
need more flags though, but each bool takes (at least) a byte. Those
add up over time.
Merge all of the existing bools in this struct into a single unsigned
long, and use the set/test/clear_bit macros to manipulate them. These
are atomic operations, but that is required here to prevent
load/modify/store races. The existing flags are protected by different
locks, so we can't rely on them for that purpose.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
Just get it from r_session since that's what's always passed in.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
Keeping around commented out code is just asking for it to bitrot and
makes viewing the code under cscope more confusing. If
we really need this, then we can revert this patch and put it under a
Kconfig option.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
__ceph_caps_mds_wanted() ignores caps from stale session. So the
return value of __ceph_caps_mds_wanted() can keep the same across
ceph_renew_caps(). This causes try_get_cap_refs() to keep calling
ceph_renew_caps(). The fix is ignore the session valid check for
the try_get_cap_refs() case. If session is stale, just let the
caps requester sleep.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
|
|
when flushing inode's auth cap changes, we need to move it into the
new auth cap session's cap_flushing list
Signed-off-by: Yan, Zheng <zyan@redhat.com>
|
|
add_to_page_cache_lru() can fails, so the actual pages to read
can be smaller than the initial size of osd request. We need to
update osd request size in that case.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@redhat.com>
|
|
sparse says:
fs/ceph/ioctl.c:100:28: warning: cast to restricted __le64
preferred_osd is a __s64 so we don't need to do any conversion. Also,
just remove the cast in ceph_ioctl_get_layout as it's not needed.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Sage Weil <sage@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
Currently crypto.c gets linux/sched.h indirectly through linux/slab.h
from linux/kasan.h. Include it directly for memalloc_noio_*() inlines.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
I ran into this compile warning, which is the result of BUG_ON(1)
not always leading to the compiler treating the code path as
unreachable:
include/linux/ceph/osdmap.h: In function 'ceph_can_shift_osds':
include/linux/ceph/osdmap.h:62:1: error: control reaches end of non-void function [-Werror=return-type]
Using BUG() here avoids the warning.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
user space may open/close single file frequently. It's not good
to send a clientcaps message to mds for each open/close syscall.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
|
|
This patch sets the io_pages bdi hint based on the rsize mount option.
Without this patch large buffered reads (request size > max readahead)
are processed sequentially in chunks of the readahead size (i.e. read
requests are sent out up to the readahead size, then the
do_generic_file_read() function waits until the first page is received).
With this patch read requests are sent out at once up to the size
specified in the rsize mount option (default: 64 MB).
Signed-off-by: Andreas Gerstmayr <andreas.gerstmayr@catalysts.cc>
Acked-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Yan, Zheng <zyan@redhat.com>
|
|
trivial fix to spelling mistake in debug message
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Yan, Zheng <zyan@redhat.com>
|
|
This removes the uses of ACCESS_ONCE in favor of READ_ONCE
Signed-off-by: Seraphime Kirkovski <kirkseraph@gmail.com>
Signed-off-by: Yan, Zheng <zyan@redhat.com>
|
|
If we have a parent inode reference already, then we don't need to
go back up the directory tree to find one.
Link: http://tracker.ceph.com/issues/18148
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
Accessing d_parent requires some sort of locking or it could vanish
out from under us. Since we take the d_lock anyway, use that to fetch
d_parent and take a reference to it, and then use that reference to
call ceph_encode_inode_release.
Link: http://tracker.ceph.com/issues/18148
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
In the event that we have a parent inode reference in the request, we
can use that instead of mucking about in the dcache. Pass any parent
inode info we have down to build_dentry_path so it can make use of it.
Link: http://tracker.ceph.com/issues/18148
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
While we hold a reference to the dentry when build_dentry_path is
called, we could end up racing with a rename that changes d_parent.
Handle that situation correctly, by using the rcu_read_lock to
ensure that the parent dentry and inode stick around long enough
to safely check ceph_snap and ceph_ino.
Link: http://tracker.ceph.com/issues/18148
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
__choose_mds exists to pick an MDS to use when issuing a call. Doing
that typically involves picking an inode and using the authoritative
MDS for it. In most cases, that's pretty straightforward, as we are
using an inode to which we hold a reference (usually represented by
r_dentry or r_inode in the request).
In the case of a snapshotted directory however, we need to fetch
the non-snapped parent, which involves walking back up the parents
in the tree. The dentries in the snapshot dir are effectively frozen
but the overall parent is _not_, and could vanish if a concurrent
rename were to occur.
Clean this code up and take special care to ensure the validity of
the entries we're working with. First, try to use the inode in
r_locked_dir if one exists. If not and all we have is r_dentry,
then we have to walk back up the tree. Use the rcu_read_lock for
this so we can ensure that any d_parent we find won't go away, and
take extra care to deal with the possibility that the dentries could
go negative.
Change get_nonsnap_parent to return an inode, and take a reference to
that inode before returning (if any). Change all of the other places
where we set "inode" in __choose_mds to also take a reference, and then
call iput on that inode before exiting the function.
Link: http://tracker.ceph.com/issues/18148
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc into HEAD
Paul Mackerras writes:
"Please do a pull from my kvm-ppc-next branch to get some fixes which I
would like to have in 4.11. There are four small commits there; two
are fixes for potential host crashes in the new HPT resizing code, and
the other two are changes to printks to make KVM on PPC a little less
noisy."
|
|
Stupid bug that wrecked the alignment of task_struct and causes WARN()s
in the x86 FPU code on some platforms.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Tested-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: e274795ea7b7 ("locking/mutex: Fix mutex handoff")
Link: http://lkml.kernel.org/r/20170218142645.GH6500@twins.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
Add mixer quirk for Tascam US-16x08 usb interface.
Even that this is an usb compliant device,
the input channels and DSP functions (EQ/Compressor) aren't accessible
by default.
Signed-off-by: Detlef Urban <onkel@paraair.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
|
|
When building a CONFIG_STACK_VALIDATION enabled kernel without the
libelf devel package installed, the Makefile prints a warning:
"Cannot use CONFIG_STACK_VALIDATION, please install libelf-dev, libelf-devel or elfutils-libelf-devel"
But when building an out-of-tree module, the warning doesn't show.
Instead it tries to use objtool, and the build fails with:
/bin/sh: ./tools/objtool/objtool: No such file or directory
Make sure the warning and the disabling of objtool occur in all cases,
by moving the CONFIG_STACK_VALIDATION checks outside the 'ifeq
($(KBUILD_EXTMOD),)' block in the Makefile.
Tested-By: Marc MERLIN <marc@merlins.org>
Suggested-by: Jessica Yu <jeyu@redhat.com>
Reported-by: Marc MERLIN <marc@merlins.org>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Reviewed-by: Jessica Yu <jeyu@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Michal Marek <mmarek@suse.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 3b27a0c85d70 ("objtool: Detect and warn if libelf is missing and don't break the build")
Link: http://lkml.kernel.org/r/b3088ae4a8698143d4851965793c61fec2135b1f.1487182864.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
|
|
Commit fd76863 (RAID1: a new I/O barrier implementation to remove resync
window) introduces a user-after-free bug.
Signed-off-by: Shaohua Li <shli@fb.com>
|
|
When I run a parallel reading performan testing on a md raid1 device with
two NVMe SSDs, I observe very bad throughput in supprise: by fio with 64KB
block size, 40 seq read I/O jobs, 128 iodepth, overall throughput is
only 2.7GB/s, this is around 50% of the idea performance number.
The perf reports locking contention happens at allow_barrier() and
wait_barrier() code,
- 41.41% fio [kernel.kallsyms] [k] _raw_spin_lock_irqsave
- _raw_spin_lock_irqsave
+ 89.92% allow_barrier
+ 9.34% __wake_up
- 37.30% fio [kernel.kallsyms] [k] _raw_spin_lock_irq
- _raw_spin_lock_irq
- 100.00% wait_barrier
The reason is, in these I/O barrier related functions,
- raise_barrier()
- lower_barrier()
- wait_barrier()
- allow_barrier()
They always hold conf->resync_lock firstly, even there are only regular
reading I/Os and no resync I/O at all. This is a huge performance penalty.
The solution is a lockless-like algorithm in I/O barrier code, and only
holding conf->resync_lock when it has to.
The original idea is from Hannes Reinecke, and Neil Brown provides
comments to improve it. I continue to work on it, and make the patch into
current form.
In the new simpler raid1 I/O barrier implementation, there are two
wait barrier functions,
- wait_barrier()
Which calls _wait_barrier(), is used for regular write I/O. If there is
resync I/O happening on the same I/O barrier bucket, or the whole
array is frozen, task will wait until no barrier on same barrier bucket,
or the whold array is unfreezed.
- wait_read_barrier()
Since regular read I/O won't interfere with resync I/O (read_balance()
will make sure only uptodate data will be read out), it is unnecessary
to wait for barrier in regular read I/Os, waiting in only necessary
when the whole array is frozen.
The operations on conf->nr_pending[idx], conf->nr_waiting[idx], conf->
barrier[idx] are very carefully designed in raise_barrier(),
lower_barrier(), _wait_barrier() and wait_read_barrier(), in order to
avoid unnecessary spin locks in these functions. Once conf->
nr_pengding[idx] is increased, a resync I/O with same barrier bucket index
has to wait in raise_barrier(). Then in _wait_barrier() if no barrier
raised in same barrier bucket index and array is not frozen, the regular
I/O doesn't need to hold conf->resync_lock, it can just increase
conf->nr_pending[idx], and return to its caller. wait_read_barrier() is
very similar to _wait_barrier(), the only difference is it only waits when
array is frozen. For heavy parallel reading I/Os, the lockless I/O barrier
code almostly gets rid of all spin lock cost.
This patch significantly improves raid1 reading peroformance. From my
testing, a raid1 device built by two NVMe SSD, runs fio with 64KB
blocksize, 40 seq read I/O jobs, 128 iodepth, overall throughput
increases from 2.7GB/s to 4.6GB/s (+70%).
Changelog
V4:
- Change conf->nr_queued[] to atomic_t.
- Define BARRIER_BUCKETS_NR_BITS by (PAGE_SHIFT - ilog2(sizeof(atomic_t)))
V3:
- Add smp_mb__after_atomic() as Shaohua and Neil suggested.
- Change conf->nr_queued[] from atomic_t to int.
- Change conf->array_frozen from atomic_t back to int, and use
READ_ONCE(conf->array_frozen) to check value of conf->array_frozen
in _wait_barrier() and wait_read_barrier().
- In _wait_barrier() and wait_read_barrier(), add a call to
wake_up(&conf->wait_barrier) after atomic_dec(&conf->nr_pending[idx]),
to fix a deadlock between _wait_barrier()/wait_read_barrier and
freeze_array().
V2:
- Remove a spin_lock/unlock pair in raid1d().
- Add more code comments to explain why there is no racy when checking two
atomic_t variables at same time.
V1:
- Original RFC patch for comments.
Signed-off-by: Coly Li <colyli@suse.de>
Cc: Shaohua Li <shli@fb.com>
Cc: Hannes Reinecke <hare@suse.com>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Cc: Guoqing Jiang <gqjiang@suse.com>
Reviewed-by: Neil Brown <neilb@suse.de>
Signed-off-by: Shaohua Li <shli@fb.com>
|
|
'Commit 79ef3a8aa1cb ("raid1: Rewrite the implementation of iobarrier.")'
introduces a sliding resync window for raid1 I/O barrier, this idea limits
I/O barriers to happen only inside a slidingresync window, for regular
I/Os out of this resync window they don't need to wait for barrier any
more. On large raid1 device, it helps a lot to improve parallel writing
I/O throughput when there are background resync I/Os performing at
same time.
The idea of sliding resync widow is awesome, but code complexity is a
challenge. Sliding resync window requires several variables to work
collectively, this is complexed and very hard to make it work correctly.
Just grep "Fixes: 79ef3a8aa1" in kernel git log, there are 8 more patches
to fix the original resync window patch. This is not the end, any further
related modification may easily introduce more regreassion.
Therefore I decide to implement a much simpler raid1 I/O barrier, by
removing resync window code, I believe life will be much easier.
The brief idea of the simpler barrier is,
- Do not maintain a global unique resync window
- Use multiple hash buckets to reduce I/O barrier conflicts, regular
I/O only has to wait for a resync I/O when both them have same barrier
bucket index, vice versa.
- I/O barrier can be reduced to an acceptable number if there are enough
barrier buckets
Here I explain how the barrier buckets are designed,
- BARRIER_UNIT_SECTOR_SIZE
The whole LBA address space of a raid1 device is divided into multiple
barrier units, by the size of BARRIER_UNIT_SECTOR_SIZE.
Bio requests won't go across border of barrier unit size, that means
maximum bio size is BARRIER_UNIT_SECTOR_SIZE<<9 (64MB) in bytes.
For random I/O 64MB is large enough for both read and write requests,
for sequential I/O considering underlying block layer may merge them
into larger requests, 64MB is still good enough.
Neil also points out that for resync operation, "we want the resync to
move from region to region fairly quickly so that the slowness caused
by having to synchronize with the resync is averaged out over a fairly
small time frame". For full speed resync, 64MB should take less then 1
second. When resync is competing with other I/O, it could take up a few
minutes. Therefore 64MB size is fairly good range for resync.
- BARRIER_BUCKETS_NR
There are BARRIER_BUCKETS_NR buckets in total, which is defined by,
#define BARRIER_BUCKETS_NR_BITS (PAGE_SHIFT - 2)
#define BARRIER_BUCKETS_NR (1<<BARRIER_BUCKETS_NR_BITS)
this patch makes the bellowed members of struct r1conf from integer
to array of integers,
- int nr_pending;
- int nr_waiting;
- int nr_queued;
- int barrier;
+ int *nr_pending;
+ int *nr_waiting;
+ int *nr_queued;
+ int *barrier;
number of the array elements is defined as BARRIER_BUCKETS_NR. For 4KB
kernel space page size, (PAGE_SHIFT - 2) indecates there are 1024 I/O
barrier buckets, and each array of integers occupies single memory page.
1024 means for a request which is smaller than the I/O barrier unit size
has ~0.1% chance to wait for resync to pause, which is quite a small
enough fraction. Also requesting single memory page is more friendly to
kernel page allocator than larger memory size.
- I/O barrier bucket is indexed by bio start sector
If multiple I/O requests hit different I/O barrier units, they only need
to compete I/O barrier with other I/Os which hit the same I/O barrier
bucket index with each other. The index of a barrier bucket which a
bio should look for is calculated by sector_to_idx() which is defined
in raid1.h as an inline function,
static inline int sector_to_idx(sector_t sector)
{
return hash_long(sector >> BARRIER_UNIT_SECTOR_BITS,
BARRIER_BUCKETS_NR_BITS);
}
Here sector_nr is the start sector number of a bio.
- Single bio won't go across boundary of a I/O barrier unit
If a request goes across boundary of barrier unit, it will be split. A
bio may be split in raid1_make_request() or raid1_sync_request(), if
sectors returned by align_to_barrier_unit_end() is smaller than
original bio size.
Comparing to single sliding resync window,
- Currently resync I/O grows linearly, therefore regular and resync I/O
will conflict within a single barrier units. So the I/O behavior is
similar to single sliding resync window.
- But a barrier unit bucket is shared by all barrier units with identical
barrier uinit index, the probability of conflict might be higher
than single sliding resync window, in condition that writing I/Os
always hit barrier units which have identical barrier bucket indexs with
the resync I/Os. This is a very rare condition in real I/O work loads,
I cannot imagine how it could happen in practice.
- Therefore we can achieve a good enough low conflict rate with much
simpler barrier algorithm and implementation.
There are two changes should be noticed,
- In raid1d(), I change the code to decrease conf->nr_pending[idx] into
single loop, it looks like this,
spin_lock_irqsave(&conf->device_lock, flags);
conf->nr_queued[idx]--;
spin_unlock_irqrestore(&conf->device_lock, flags);
This change generates more spin lock operations, but in next patch of
this patch set, it will be replaced by a single line code,
atomic_dec(&conf->nr_queueud[idx]);
So we don't need to worry about spin lock cost here.
- Mainline raid1 code split original raid1_make_request() into
raid1_read_request() and raid1_write_request(). If the original bio
goes across an I/O barrier unit size, this bio will be split before
calling raid1_read_request() or raid1_write_request(), this change
the code logic more simple and clear.
- In this patch wait_barrier() is moved from raid1_make_request() to
raid1_write_request(). In raid_read_request(), original wait_barrier()
is replaced by raid1_read_request().
The differnece is wait_read_barrier() only waits if array is frozen,
using different barrier function in different code path makes the code
more clean and easy to read.
Changelog
V4:
- Add alloc_r1bio() to remove redundant r1bio memory allocation code.
- Fix many typos in patch comments.
- Use (PAGE_SHIFT - ilog2(sizeof(int))) to define BARRIER_BUCKETS_NR_BITS.
V3:
- Rebase the patch against latest upstream kernel code.
- Many fixes by review comments from Neil,
- Back to use pointers to replace arraries in struct r1conf
- Remove total_barriers from struct r1conf
- Add more patch comments to explain how/why the values of
BARRIER_UNIT_SECTOR_SIZE and BARRIER_BUCKETS_NR are decided.
- Use get_unqueued_pending() to replace get_all_pendings() and
get_all_queued()
- Increase bucket number from 512 to 1024
- Change code comments format by review from Shaohua.
V2:
- Use bio_split() to split the orignal bio if it goes across barrier unit
bounday, to make the code more simple, by suggestion from Shaohua and
Neil.
- Use hash_long() to replace original linear hash, to avoid a possible
confilict between resync I/O and sequential write I/O, by suggestion from
Shaohua.
- Add conf->total_barriers to record barrier depth, which is used to
control number of parallel sync I/O barriers, by suggestion from Shaohua.
- In V1 patch the bellowed barrier buckets related members in r1conf are
allocated in memory page. To make the code more simple, V2 patch moves
the memory space into struct r1conf, like this,
- int nr_pending;
- int nr_waiting;
- int nr_queued;
- int barrier;
+ int nr_pending[BARRIER_BUCKETS_NR];
+ int nr_waiting[BARRIER_BUCKETS_NR];
+ int nr_queued[BARRIER_BUCKETS_NR];
+ int barrier[BARRIER_BUCKETS_NR];
This change is by the suggestion from Shaohua.
- Remove some inrelavent code comments, by suggestion from Guoqing.
- Add a missing wait_barrier() before jumping to retry_write, in
raid1_make_write_request().
V1:
- Original RFC patch for comments
Signed-off-by: Coly Li <colyli@suse.de>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Cc: Guoqing Jiang <gqjiang@suse.com>
Reviewed-by: Neil Brown <neilb@suse.de>
Signed-off-by: Shaohua Li <shli@fb.com>
|
|
sctp has changed to use rhlist for transport rhashtable since commit
7fda702f9315 ("sctp: use new rhlist interface on sctp transport
rhashtable").
But rhltable_insert_key doesn't check the duplicate node when inserting
a node, unlike rhashtable_lookup_insert_key. It may cause duplicate
assoc/transport in rhashtable. like:
client (addr A, B) server (addr X, Y)
connect to X INIT (1)
------------>
connect to Y INIT (2)
------------>
INIT_ACK (1)
<------------
INIT_ACK (2)
<------------
After sending INIT (2), one transport will be created and hashed into
rhashtable. But when receiving INIT_ACK (1) and processing the address
params, another transport will be created and hashed into rhashtable
with the same addr Y and EP as the last transport. This will confuse
the assoc/transport's lookup.
This patch is to fix it by returning err if any duplicate node exists
before inserting it.
Fixes: 7fda702f9315 ("sctp: use new rhlist interface on sctp transport rhashtable")
Reported-by: Fabio M. Di Nitto <fdinitto@redhat.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Some Cavium dev boards have firmware which doesn't supply a proper
ethernet-phy-ieee802.3-c22" compatible property. Restore these boards
to working order by whitelisting this compatible value.
Signed-off-by: David Daney <david.daney@cavium.com>
Acked-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add constants and callback functions for the dwmac on rk3328 socs.
As can be seen, the base structure is the same, only registers and the
bits in them moved slightly.
Signed-off-by: david.wu <david.wu@rock-chips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Now when sending a packet, sctp_transport_dst_check will check if dst
is obsolete by calling ipv4/ip6_dst_check. But they return obsolete
only when adding a new cache, after that when the cache's pmtu is
updated again, it will not trigger transport->dst/pmtu's update.
It can be reproduced by reducing route's pmtu twice. At the 1st time
client will add a new cache, and transport->pathmtu gets updated as
sctp_transport_dst_check finds it's obsolete. But at the 2nd time,
cache's mtu is updated, sctp client will never send out any packet,
because transport->pmtu has no chance to update.
This patch is to fix this by also checking if transport pmtu is dst
mtu in sctp_transport_dst_check, so that transport->pmtu can be
updated on time.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Xin Long says:
====================
sctp: add receiver-side procedures for stream reconf ssn reset request chunk
Patch 3/7 and 4/7 are to implement receiver-side procedures for the
Outgoing and Incoming SSN Reset Request Parameter described in rfc6525
section 5.2.2 and 5.2.3
Patch 1/7 and 2/7 are ahead of them to define some apis.
Patch 5/7-7/7 are to add the process of reconf chunk event in rx path.
Note that with this patchset, asoc->reconf_enable has no chance yet to
be set, until the patch "sctp: add get and set sockopt for reconf_enable"
is applied in the future. As we can not just enable it when sctp is not
capable of processing reconf chunk yet.
v1->v2:
- re-split the patchset and make sure it has no dead codes for review.
- rename the titles of the commits and improve some changelogs.
- drop __packed from some structures in patch 1/7.
- fix some kbuild warnings in patch 3/7 by initializing str_p = NULL.
- sctp_chunk_lookup_strreset_param changes to return sctp_paramhdr_t *
and uses sctp_strreset_tsnreq to access request_seq in patch 3/7.
- use __u<size> in uapi sctp.h in patch 1/7.
- do str_list endian conversion when generating stream_reset_event in patch
2/7.
- remove str_list endian conversion, pass resp_seq param with network endian
to lookup_strreset_param in 3/7.
- move str_list endian conversion out of sctp_make_strreset_req, so that
sctp_make_strreset_req can be used more conveniently to process inreq in
patch 4/7.
- remove sctp_merge_reconf_chunk and not support response with multiparam
in patch 6/7.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
