| Age | Commit message (Collapse) | Author |
|---|
|
Hold off setting the device id and formatting the device name
in rbd_add() until just before it's needed.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Read the rbd header information and call rbd_dev_set_mapping()
earlier--before registering the block device or setting up the sysfs
entries for the image. The sysfs entries provide users access to
some information that's only available after doing the rbd header
initialization, so this will make sure it's valid right away.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
rbd_header_set_snap() is a simple initialization routine for an rbd
device's mapping. It has to be called after the snapshot context
for the rbd_dev has been updated, but can be done before snapshot
devices have been registered.
Change the name to rbd_dev_set_mapping() to better reflect its
purpose, and call it a little sooner, before registering snapshot
devices.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
When a new snapshot is found in an rbd device's updated snapshot
context, __rbd_add_snap_dev() is called to create and insert an
entry in the rbd devices list of snapshots. In addition, a Linux
device is registered to represent the snapshot.
For version 2 rbd images, it will be undesirable to initialize the
device right away. So in anticipation of that, this patch separates
the insertion of a snapshot entry in the snaps list from the
creation of devices for those snapshots.
To do this, create a new function rbd_dev_snaps_register() which
traverses the list of snapshots and calls rbd_register_snap_dev()
on any that have not yet been registered.
Rename rbd_dev_snap_devs_update() to be rbd_dev_snaps_update()
to better reflect that only the entry in the snaps list and not
the snapshot's device is affected by the function.
For now, call rbd_dev_snaps_register() immediately after each
call to rbd_dev_snaps_update().
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Move the assignment of the header name for an rbd image a bit later,
outside rbd_add_parse_args() and into its caller.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
An rbd_dev structure maintains a list of current snapshots that have
already been fully initialized. The entries on the list have type
struct rbd_snap, and each entry contains a copy of information
that's found in the rbd_dev's snapshot context and header.
The only caller of snap_by_name() is rbd_header_set_snap(). In that
call site any positive return value (the index in the snapshot
array) is ignored, so there's no need to return the index in
the snapshot context's id array when it's found.
rbd_header_set_snap() also has only one caller--rbd_add()--and that
call is made after a call to rbd_dev_snap_devs_update(). Because
the rbd_snap structures are initialized in that function, the
current snapshot list can be used instead of the snapshot context to
look up a snapshot's information by name.
Change snap_by_name() so it uses the snapshot list rather than the
rbd_dev's snapshot context in looking up snapshot information.
Return 0 if it's found rather than the snapshot id.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
When rbd_bus_add_dev() is called (one spot--in rbd_add()), the rbd
image header has not even been read yet. This means that the list
of snapshots will be empty at the time of the call. As a result,
there is no need for the code that calls rbd_register_snap_dev()
for each entry in that list--so get rid of it.
Once the header has been read (just after returning), a call will
be made to rbd_dev_snap_devs_update(), which will then find every
snapshot in the context to be new and will therefore call
rbd_register_snap_dev() via __rbd_add_snap_dev() accomplishing
the same thing.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Move the calls to get the header semaphore out of
rbd_header_set_snap() and into its caller.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
This just simplifies a few things in rbd_init_disk(), now that the
previous patch has moved a bunch of initialization code out if it.
Done separately to facilitate review.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Move some of the code that initializes an rbd header out of
rbd_init_disk() and into its caller.
Move the code at the end of rbd_init_disk() that sets the device
capacity and activates the Linux device out of that function and
into the caller, ensuring we still have the disk size available
where we need it.
Update rbd_free_disk() so it still aligns well as an inverse of
rbd_init_disk(), moving the rbd_header_free() call out to its
caller.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
There is only one caller of snap_by_name(), and it passes two values
to be assigned, both of which are found within an rbd device
structure.
Change the interface so it just passes the address of the rbd_dev,
and make the assignments to its fields directly.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
With the exception of the snapshot name, all of the mapping-specific
fields in an rbd device structure are set in rbd_header_set_snap().
Pass the snapshot name to be assigned into rbd_header_set_snap()
to keep all of the mapping assignments together.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
This is the first of two patches aimed at isolating the code that
sets the mapping information into a single spot.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Add the size of the mapped image to the set of mapping-specific
fields in an rbd_device, and use it when setting the capacity of the
disk.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Several fields in a struct rbd_dev are related to what is mapped, as
opposed to the actual base rbd image. If the base image is mapped
these are almost unneeded, but if a snapshot is mapped they describe
information about that snapshot.
In some contexts this can be a little bit confusing. So group these
mapping-related field into a structure to make it clear what they
are describing.
This also includes a minor change that rearranges the fields in the
in-core image header structure so that invariant fields are at the
top, followed by those that change.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
The "total_snaps" field in an rbd header structure is never any
different from the value of "num_snaps" stored within a snapshot
context. Avoid any confusion by just using the value held within
the snapshot context, and get rid of the "total_snaps" field.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
A copy of rbd_dev->disk->queue is held in rbd_dev->q, but it's
never actually used. So get just get rid of the field.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
The name __rbd_init_snaps_header() doesn't really convey what that
function does very well. Its purpose is to scan a new snapshot
context and either create or destroy snapshot device entries so
that local host's view is consistent with the reality maintained
on the OSDs. This patch just changes the name of this function,
to be rbd_dev_snap_devs_update(). Still not perfect, but I think
better.
Also add some dynamic debug statements to this function.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
This should have been done as part of this commit:
commit de71a2970d57463d3d965025e33ec3adcf391248
Author: Alex Elder <elder@inktank.com>
Date: Tue Jul 3 16:01:19 2012 -0500
rbd: rename rbd_device->id
rbd_id_get() is assigning the rbd_dev->dev_id field. Change the
name of that function as well as rbd_id_put() and rbd_id_max
to reflect what they are affecting.
Add some dynamic debug statements related to rbd device id activity.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Define rbd_assert() and use it in place of various BUG_ON() calls
now present in the code. By default assertion checking is enabled;
we want to do this differently at some point.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
There are two places where rbd_get_segment() is called. One, in
rbd_rq_fn(), only needs to know the length within a segment that an
I/O request should be. The other, in rbd_do_op(), also needs the
name of the object and the offset within it for the I/O request.
Split out rbd_segment_name() into three dedicated functions:
- rbd_segment_name() allocates and formats the name of the
object for a segment containing a given rbd image offset
- rbd_segment_offset() computes the offset within a segment for
a given rbd image offset
- rbd_segment_length() computes the length to use for I/O within
a segment for a request, not to exceed the end of a segment
object.
In the new functions be a bit more careful, checking for possible
error conditions:
- watch for errors or overflows returned by snprintf()
- catch (using BUG_ON()) potential overflow conditions
when computing segment length
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
It is possible in rbd_get_num_segments() for an overflow to occur
when adding the offset and length. This is easily avoided.
Since the function returns an int and the one caller is already
prepared to handle errors, have it return -ERANGE if overflow would
occur.
The overflow check would not work if a zero-length request was
being tested, so short-circuit that case, returning 0 for the
number of segments required. (This condition might be avoided
elsewhere already, I don't know.)
Have the caller end the request if either an error or 0 is returned.
The returned value is passed to __blk_end_request_all(), meaning
a 0 length request is not treated an error.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
There's a test for null rq pointer inside the while loop in
rbd_rq_fn() that's not needed. That same test already occurred
in the immediatly preceding loop condition test.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
In bio_chain_clone(), at the end of the function the bi_next field
of the tail of the new bio chain is nulled. This isn't necessary,
because if "tail" is non-null, its value will be the last bio
structure allocated at the top of the while loop in that function.
And before that structure is added to the end of the new chain, its
bi_next pointer is always made null.
While touching that function, clean a few other things:
- define each local variable on its own line
- move the definition of "tmp" to an inner scope
- move the modification of gfpmask closer to where it's used
- rearrange the logic that sets the chain's tail pointer
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
The "notify_timeout" rbd device option is never used, so get rid of
it.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
Add the ability to map an rbd image read-only, by specifying either
"read_only" or "ro" as an option on the rbd "command line." Also
allow the inverse to be explicitly specified using "read_write" or
"rw".
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
The rbd options don't really apply to the ceph client. So don't
store a pointer to it in the ceph_client structure, and put them
(a struct, not a pointer) into the rbd_dev structure proper.
Pass the rbd device structure to rbd_client_create() so it can
assign rbd_dev->rbdc if successful, and have it return an error code
instead of the rbd client pointer.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
This just rearranges things a bit more in rbd_header_from_disk()
so that the snapshot sizes are initialized right after the buffer
to hold them is allocated and doing a little further consolidation
that follows from that. Also adds a few simple comments.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
The only thing the on-disk snap_names_len field is needed is to
size the buffer allocated to hold a copy of the snapshot names
for an rbd image.
So don't bother saving it in the in-core rbd_image_header structure.
Just use a local variable to hold the required buffer size while
it's needed.
Move the code that actually copies the snapshot names up closer
to where the required length is saved.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
In rbd_header_from_disk() the object prefix buffer is sized based on
the maximum size it's block_name equivalent on disk could be.
Instead, only allocate enough to hold null-terminated string from
the on-disk header--or the maximum size of no NUL is found.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
There is only caller of __rbd_client_find(), and it somewhat
clumsily gets the appropriate lock and gets a reference to the
existing ceph_client structure if it's found.
Instead, have that function handle its own locking, and acquire the
reference if found while it holds the lock. Drop the underscores
from the name because there's no need to signify anything special
about this function.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
|
|
This fixes a bug that went in with this commit:
commit f6e0c99092cca7be00fca4080cfc7081739ca544
Author: Alex Elder <elder@inktank.com>
Date: Thu Aug 2 11:29:46 2012 -0500
rbd: simplify __rbd_init_snaps_header()
The problem is that a new rbd snapshot needs to go either after an
existing snapshot entry, or at the *end* of an rbd device's snapshot
list. As originally coded, it is placed at the beginning. This was
based on the assumption the list would be empty (so it wouldn't
matter), but in fact if multiple new snapshots are added to an empty
list in one shot the list will be non-empty after the first one is
added.
This addresses http://tracker.newdream.net/issues/3063
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
In the on-disk image header structure there is a field "block_name"
which represents what we now call the "object prefix" for an rbd
image. Rename this field "object_prefix" to be consistent with
modern usage.
This appears to be the only remaining vestige of the use of "block"
in symbols that represent objects in the rbd code.
This addresses http://tracker.newdream.net/issues/1761
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
Reviewed-by: Dan Mick <dan.mick@inktank.com>
|
|
Right now rbd_read_header() both reads the header object for an rbd
image and decodes its contents. It does this repeatedly if needed,
in order to ensure a complete and intact header is obtained.
Separate this process into two steps--reading of the raw header
data (in new function, rbd_dev_v1_header_read()) and separately
decoding its contents (in rbd_header_from_disk()). As a result,
the latter function no longer requires its allocated_snaps argument.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Add checks on the validity of the snap_count and snap_names_len
field values in rbd_dev_ondisk_valid(). This eliminates the
need to do them in rbd_header_from_disk().
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
The only caller of rbd_header_from_disk() is rbd_read_header().
It passes as allocated_snaps the number of snapshots it will
have received from the server for the snapshot context that
rbd_header_from_disk() is to interpret. The first time through
it provides 0--mainly to extract the number of snapshots from
the snapshot context header--so that it can allocate an
appropriately-sized buffer to receive the entire snapshot
context from the server in a second request.
rbd_header_from_disk() will not fill in the array of snapshot ids
unless the number in the snapshot matches the number the caller
had allocated.
This patch adjusts that logic a little further to be more efficient.
rbd_read_header() doesn't even examine the snapshot context unless
the snapshot count (stored in header->total_snaps) matches the
number of snapshots allocated. So rbd_header_from_disk() doesn't
need to allocate or fill in the snapshot context field at all in
that case.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
This just moves code around for the most part. It was pulled out as
a separate patch to avoid cluttering up some upcoming patches which
are more substantive. The point is basically to group everything
related to initializing the snapshot context together.
The only functional change is that rbd_header_from_disk() now
ensures the (in-core) header it is passed is zero-filled. This
allows a simpler error handling path in rbd_header_from_disk().
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Fix a few spots in rbd_header_from_disk() to use sizeof (object)
rather than sizeof (type). Use a local variable to record sizes
to shorten some lines and improve readability.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Fix a number of spots where a pointer value that is known to
have become invalid but was not reset to null.
Also, toss in a change so we use sizeof (object) rather than
sizeof (type).
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
The snap_names_len field of an rbd_image_header structure is defined
with type size_t. That field is used as both the source and target
of 64-bit byte-order swapping operations though, so it's best to
define it with type u64 instead.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
The purpose of __rbd_init_snaps_header() is to compare a new
snapshot context with an rbd device's list of existing snapshots.
It updates the list by adding any new snapshots or removing any
that are not present in the new snapshot context.
The code as written is a little confusing, because it traverses both
the existing snapshot list and the set of snapshots in the snapshot
context in reverse. This was done based on an assumption about
snapshots that is not true--namely that a duplicate snapshot name
could cause an error in intepreting things if they were not
processed in ascending order.
These precautions are not necessary, because:
- all snapshots are uniquely identified by their snapshot id
- a new snapshot cannot be created if the rbd device has another
snapshot with the same name
(It is furthermore not currently possible to rename a snapshot.)
This patch re-implements __rbd_init_snaps_header() so it passes
through both the existing snapshot list and the entries in the
snapshot context in forward order. It still does the same thing
as before, but I find the logic considerably easier to understand.
By going forward through the names in the snapshot context, there
is no longer a need for the rbd_prev_snap_name() helper function.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Pull PCI changes from Bjorn Helgaas:
"Host bridge hotplug
- Protect acpi_pci_drivers and acpi_pci_roots (Taku Izumi)
- Clear host bridge resource info to avoid issue when releasing
(Yinghai Lu)
- Notify acpi_pci_drivers when hot-plugging host bridges (Jiang Liu)
- Use standard list ops for acpi_pci_drivers (Jiang Liu)
Device hotplug
- Use pci_get_domain_bus_and_slot() to close hotplug races (Jiang
Liu)
- Remove fakephp driver (Bjorn Helgaas)
- Fix VGA ref count in hotplug remove path (Yinghai Lu)
- Allow acpiphp to handle PCIe ports without native hotplug (Jiang
Liu)
- Implement resume regardless of pciehp_force param (Oliver Neukum)
- Make pci_fixup_irqs() work after init (Thierry Reding)
Miscellaneous
- Add pci_pcie_type(dev) and remove pci_dev.pcie_type (Yijing Wang)
- Factor out PCI Express Capability accessors (Jiang Liu)
- Add pcibios_window_alignment() so powerpc EEH can use generic
resource assignment (Gavin Shan)
- Make pci_error_handlers const (Stephen Hemminger)
- Cleanup drivers/pci/remove.c (Bjorn Helgaas)
- Improve Vendor-Specific Extended Capability support (Bjorn
Helgaas)
- Use standard list ops for bus->devices (Bjorn Helgaas)
- Avoid kmalloc in pci_get_subsys() and pci_get_class() (Feng Tang)
- Reassign invalid bus number ranges (Intel DP43BF workaround)
(Yinghai Lu)"
* tag 'for-3.7' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci: (102 commits)
PCI: acpiphp: Handle PCIe ports without native hotplug capability
PCI/ACPI: Use acpi_driver_data() rather than searching acpi_pci_roots
PCI/ACPI: Protect acpi_pci_roots list with mutex
PCI/ACPI: Use acpi_pci_root info rather than looking it up again
PCI/ACPI: Pass acpi_pci_root to acpi_pci_drivers' add/remove interface
PCI/ACPI: Protect acpi_pci_drivers list with mutex
PCI/ACPI: Notify acpi_pci_drivers when hot-plugging PCI root bridges
PCI/ACPI: Use normal list for struct acpi_pci_driver
PCI/ACPI: Use DEVICE_ACPI_HANDLE rather than searching acpi_pci_roots
PCI: Fix default vga ref_count
ia64/PCI: Clear host bridge aperture struct resource
x86/PCI: Clear host bridge aperture struct resource
PCI: Stop all children first, before removing all children
Revert "PCI: Use hotplug-safe pci_get_domain_bus_and_slot()"
PCI: Provide a default pcibios_update_irq()
PCI: Discard __init annotations for pci_fixup_irqs() and related functions
PCI: Use correct type when freeing bus resource list
PCI: Check P2P bridge for invalid secondary/subordinate range
PCI: Convert "new_id"/"remove_id" into generic pci_bus driver attributes
xen-pcifront: Use hotplug-safe pci_get_domain_bus_and_slot()
...
|
|
Pull NVMe driver fixes from Matthew Wilcox:
"Now that actual hardware has been released (don't have any yet
myself), people are starting to want some of these fixes merged."
Willy doesn't have hardware? Guys...
* git://git.infradead.org/users/willy/linux-nvme:
NVMe: Cancel outstanding IOs on queue deletion
NVMe: Free admin queue memory on initialisation failure
NVMe: Use ida for nvme device instance
NVMe: Fix whitespace damage in nvme_init
NVMe: handle allocation failure in nvme_map_user_pages()
NVMe: Fix uninitialized iod compiler warning
NVMe: Do not set IO queue depth beyond device max
NVMe: Set block queue max sectors
NVMe: use namespace id for nvme_get_features
NVMe: replace nvme_ns with nvme_dev for user admin
NVMe: Fix nvme module init when nvme_major is set
NVMe: Set request queue logical block size
|
|
This reduces unnecessary interrupts that host could send to guest while
guest is in the progress of irq handling.
If one vcpu is handling the irq, while another interrupt comes, in
handle_edge_irq(), the guest will mask the interrupt via mask_msi_irq()
which is a very heavy operation that goes all the way down to host.
Here are some performance numbers on qemu:
Before:
-------------------------------------
seq-read : io=0 B, bw=269730KB/s, iops=67432 , runt= 62200msec
seq-write : io=0 B, bw=339716KB/s, iops=84929 , runt= 49386msec
rand-read : io=0 B, bw=270435KB/s, iops=67608 , runt= 62038msec
rand-write: io=0 B, bw=354436KB/s, iops=88608 , runt= 47335msec
clat (usec): min=101 , max=138052 , avg=14822.09, stdev=11771.01
clat (usec): min=96 , max=81543 , avg=11798.94, stdev=7735.60
clat (usec): min=128 , max=140043 , avg=14835.85, stdev=11765.33
clat (usec): min=109 , max=147207 , avg=11337.09, stdev=5990.35
cpu : usr=15.93%, sys=60.37%, ctx=7764972, majf=0, minf=54
cpu : usr=32.73%, sys=120.49%, ctx=7372945, majf=0, minf=1
cpu : usr=18.84%, sys=58.18%, ctx=7775420, majf=0, minf=1
cpu : usr=24.20%, sys=59.85%, ctx=8307886, majf=0, minf=0
vdb: ios=8389107/8368136, merge=0/0, ticks=19457874/14616506,
in_queue=34206098, util=99.68%
43: interrupt in total: 887320
fio --exec_prerun="echo 3 > /proc/sys/vm/drop_caches" --group_reporting
--ioscheduler=noop --thread --bs=4k --size=512MB --direct=1 --numjobs=16
--ioengine=libaio --iodepth=64 --loops=3 --ramp_time=0
--filename=/dev/vdb --name=seq-read --stonewall --rw=read
--name=seq-write --stonewall --rw=write --name=rnd-read --stonewall
--rw=randread --name=rnd-write --stonewall --rw=randwrite
After:
-------------------------------------
seq-read : io=0 B, bw=309503KB/s, iops=77375 , runt= 54207msec
seq-write : io=0 B, bw=448205KB/s, iops=112051 , runt= 37432msec
rand-read : io=0 B, bw=311254KB/s, iops=77813 , runt= 53902msec
rand-write: io=0 B, bw=377152KB/s, iops=94287 , runt= 44484msec
clat (usec): min=81 , max=90588 , avg=12946.06, stdev=9085.94
clat (usec): min=57 , max=72264 , avg=8967.97, stdev=5951.04
clat (usec): min=29 , max=101046 , avg=12889.95, stdev=9067.91
clat (usec): min=52 , max=106152 , avg=10660.56, stdev=4778.19
cpu : usr=15.05%, sys=57.92%, ctx=7710941, majf=0, minf=54
cpu : usr=26.78%, sys=101.40%, ctx=7387891, majf=0, minf=2
cpu : usr=19.03%, sys=58.17%, ctx=7681976, majf=0, minf=8
cpu : usr=24.65%, sys=58.34%, ctx=8442632, majf=0, minf=4
vdb: ios=8389086/8361888, merge=0/0, ticks=17243780/12742010,
in_queue=30078377, util=99.59%
43: interrupt in total: 1259639
fio --exec_prerun="echo 3 > /proc/sys/vm/drop_caches" --group_reporting
--ioscheduler=noop --thread --bs=4k --size=512MB --direct=1 --numjobs=16
--ioengine=libaio --iodepth=64 --loops=3 --ramp_time=0
--filename=/dev/vdb --name=seq-read --stonewall --rw=read
--name=seq-write --stonewall --rw=write --name=rnd-read --stonewall
--rw=randread --name=rnd-write --stonewall --rw=randwrite
Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
|
|
Smatch complains about the inconsistent NULL checking here. Fix it to
return NULL on failure.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> (fixed accidental deletion)
|
|
We need to support both REQ_FLUSH and REQ_FUA for bio based path since
it does not get the sequencing of REQ_FUA into REQ_FLUSH that request
based drivers can request.
REQ_FLUSH is emulated by:
A) If the bio has no data to write:
1. Send VIRTIO_BLK_T_FLUSH to device,
2. In the flush I/O completion handler, finish the bio
B) If the bio has data to write:
1. Send VIRTIO_BLK_T_FLUSH to device
2. In the flush I/O completion handler, send the actual write data to device
3. In the write I/O completion handler, finish the bio
REQ_FUA is emulated by:
1. Send the actual write data to device
2. In the write I/O completion handler, send VIRTIO_BLK_T_FLUSH to device
3. In the flush I/O completion handler, finish the bio
Changes in v7:
- Using vbr->flags to trace request type
- Dropped unnecessary struct virtio_blk *vblk parameter
- Reuse struct virtblk_req in bio done function
Cahnges in v6:
- Reworked REQ_FLUSH and REQ_FUA emulatation order
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Tejun Heo <tj@kernel.org>
Cc: Shaohua Li <shli@kernel.org>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: kvm@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: virtualization@lists.linux-foundation.org
Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
|
|
This patch introduces bio-based IO path for virtio-blk.
Compared to request-based IO path, bio-based IO path uses driver
provided ->make_request_fn() method to bypasses the IO scheduler. It
handles the bio to device directly without allocating a request in block
layer. This reduces the IO path in guest kernel to achieve high IOPS
and lower latency. The downside is that guest can not use the IO
scheduler to merge and sort requests. However, this is not a big problem
if the backend disk in host side uses faster disk device.
When the bio-based IO path is not enabled, virtio-blk still uses the
original request-based IO path, no performance difference is observed.
Using a slow device e.g. normal SATA disk, the bio-based IO path for
sequential read and write are slower than req-based IO path due to lack
of merge in guest kernel. So we make the bio-based path optional.
Performance evaluation:
-----------------------------
1) Fio test is performed in a 8 vcpu guest with ramdisk based guest using
kvm tool.
Short version:
With bio-based IO path, sequential read/write, random read/write
IOPS boost : 28%, 24%, 21%, 16%
Latency improvement: 32%, 17%, 21%, 16%
Long version:
With bio-based IO path:
seq-read : io=2048.0MB, bw=116996KB/s, iops=233991 , runt= 17925msec
seq-write : io=2048.0MB, bw=100829KB/s, iops=201658 , runt= 20799msec
rand-read : io=3095.7MB, bw=112134KB/s, iops=224268 , runt= 28269msec
rand-write: io=3095.7MB, bw=96198KB/s, iops=192396 , runt= 32952msec
clat (usec): min=0 , max=2631.6K, avg=58716.99, stdev=191377.30
clat (usec): min=0 , max=1753.2K, avg=66423.25, stdev=81774.35
clat (usec): min=0 , max=2915.5K, avg=61685.70, stdev=120598.39
clat (usec): min=0 , max=1933.4K, avg=76935.12, stdev=96603.45
cpu : usr=74.08%, sys=703.84%, ctx=29661403, majf=21354, minf=22460954
cpu : usr=70.92%, sys=702.81%, ctx=77219828, majf=13980, minf=27713137
cpu : usr=72.23%, sys=695.37%, ctx=88081059, majf=18475, minf=28177648
cpu : usr=69.69%, sys=654.13%, ctx=145476035, majf=15867, minf=26176375
With request-based IO path:
seq-read : io=2048.0MB, bw=91074KB/s, iops=182147 , runt= 23027msec
seq-write : io=2048.0MB, bw=80725KB/s, iops=161449 , runt= 25979msec
rand-read : io=3095.7MB, bw=92106KB/s, iops=184211 , runt= 34416msec
rand-write: io=3095.7MB, bw=82815KB/s, iops=165630 , runt= 38277msec
clat (usec): min=0 , max=1932.4K, avg=77824.17, stdev=170339.49
clat (usec): min=0 , max=2510.2K, avg=78023.96, stdev=146949.15
clat (usec): min=0 , max=3037.2K, avg=74746.53, stdev=128498.27
clat (usec): min=0 , max=1363.4K, avg=89830.75, stdev=114279.68
cpu : usr=53.28%, sys=724.19%, ctx=37988895, majf=17531, minf=23577622
cpu : usr=49.03%, sys=633.20%, ctx=205935380, majf=18197, minf=27288959
cpu : usr=55.78%, sys=722.40%, ctx=101525058, majf=19273, minf=28067082
cpu : usr=56.55%, sys=690.83%, ctx=228205022, majf=18039, minf=26551985
2) Fio test is performed in a 8 vcpu guest with Fusion-IO based guest using
kvm tool.
Short version:
With bio-based IO path, sequential read/write, random read/write
IOPS boost : 11%, 11%, 13%, 10%
Latency improvement: 10%, 10%, 12%, 10%
Long Version:
With bio-based IO path:
read : io=2048.0MB, bw=58920KB/s, iops=117840 , runt= 35593msec
write: io=2048.0MB, bw=64308KB/s, iops=128616 , runt= 32611msec
read : io=3095.7MB, bw=59633KB/s, iops=119266 , runt= 53157msec
write: io=3095.7MB, bw=62993KB/s, iops=125985 , runt= 50322msec
clat (usec): min=0 , max=1284.3K, avg=128109.01, stdev=71513.29
clat (usec): min=94 , max=962339 , avg=116832.95, stdev=65836.80
clat (usec): min=0 , max=1846.6K, avg=128509.99, stdev=89575.07
clat (usec): min=0 , max=2256.4K, avg=121361.84, stdev=82747.25
cpu : usr=56.79%, sys=421.70%, ctx=147335118, majf=21080, minf=19852517
cpu : usr=61.81%, sys=455.53%, ctx=143269950, majf=16027, minf=24800604
cpu : usr=63.10%, sys=455.38%, ctx=178373538, majf=16958, minf=24822612
cpu : usr=62.04%, sys=453.58%, ctx=226902362, majf=16089, minf=23278105
With request-based IO path:
read : io=2048.0MB, bw=52896KB/s, iops=105791 , runt= 39647msec
write: io=2048.0MB, bw=57856KB/s, iops=115711 , runt= 36248msec
read : io=3095.7MB, bw=52387KB/s, iops=104773 , runt= 60510msec
write: io=3095.7MB, bw=57310KB/s, iops=114619 , runt= 55312msec
clat (usec): min=0 , max=1532.6K, avg=142085.62, stdev=109196.84
clat (usec): min=0 , max=1487.4K, avg=129110.71, stdev=114973.64
clat (usec): min=0 , max=1388.6K, avg=145049.22, stdev=107232.55
clat (usec): min=0 , max=1465.9K, avg=133585.67, stdev=110322.95
cpu : usr=44.08%, sys=590.71%, ctx=451812322, majf=14841, minf=17648641
cpu : usr=48.73%, sys=610.78%, ctx=418953997, majf=22164, minf=26850689
cpu : usr=45.58%, sys=581.16%, ctx=714079216, majf=21497, minf=22558223
cpu : usr=48.40%, sys=599.65%, ctx=656089423, majf=16393, minf=23824409
3) Fio test is performed in a 8 vcpu guest with normal SATA based guest
using kvm tool.
Short version:
With bio-based IO path, sequential read/write, random read/write
IOPS boost : -10%, -10%, 4.4%, 0.5%
Latency improvement: -12%, -15%, 2.5%, 0.8%
Long Version:
With bio-based IO path:
read : io=124812KB, bw=36537KB/s, iops=9060 , runt= 3416msec
write: io=169180KB, bw=24406KB/s, iops=6065 , runt= 6932msec
read : io=256200KB, bw=2089.3KB/s, iops=520 , runt=122630msec
write: io=257988KB, bw=1545.7KB/s, iops=384 , runt=166910msec
clat (msec): min=1 , max=1527 , avg=28.06, stdev=89.54
clat (msec): min=2 , max=344 , avg=41.12, stdev=38.70
clat (msec): min=8 , max=1984 , avg=490.63, stdev=207.28
clat (msec): min=33 , max=4131 , avg=659.19, stdev=304.71
cpu : usr=4.85%, sys=17.15%, ctx=31593, majf=0, minf=7
cpu : usr=3.04%, sys=11.45%, ctx=39377, majf=0, minf=0
cpu : usr=0.47%, sys=1.59%, ctx=262986, majf=0, minf=16
cpu : usr=0.47%, sys=1.46%, ctx=337410, majf=0, minf=0
With request-based IO path:
read : io=150120KB, bw=40420KB/s, iops=10037 , runt= 3714msec
write: io=194932KB, bw=27029KB/s, iops=6722 , runt= 7212msec
read : io=257136KB, bw=2001.1KB/s, iops=498 , runt=128443msec
write: io=258276KB, bw=1537.2KB/s, iops=382 , runt=168028msec
clat (msec): min=1 , max=1542 , avg=24.84, stdev=32.45
clat (msec): min=3 , max=628 , avg=35.62, stdev=39.71
clat (msec): min=8 , max=2540 , avg=503.28, stdev=236.97
clat (msec): min=41 , max=4398 , avg=653.88, stdev=302.61
cpu : usr=3.91%, sys=15.75%, ctx=26968, majf=0, minf=23
cpu : usr=2.50%, sys=10.56%, ctx=19090, majf=0, minf=0
cpu : usr=0.16%, sys=0.43%, ctx=20159, majf=0, minf=16
cpu : usr=0.18%, sys=0.53%, ctx=81364, majf=0, minf=0
How to use:
-----------------------------
Add 'virtio_blk.use_bio=1' to kernel cmdline or 'modprobe virtio_blk
use_bio=1' to enable ->make_request_fn() based I/O path.
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Tejun Heo <tj@kernel.org>
Cc: Shaohua Li <shli@kernel.org>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: kvm@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: virtualization@lists.linux-foundation.org
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Asias He <asias@redhat.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client
Pull two ceph fixes from Sage Weil:
"The first fixes a leak in the rbd setup error path, and the second
fixes a more serious problem with mismatched kmap/kunmap that surfaced
after the recent refactoring work."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
libceph: only kunmap kmapped pages
rbd: drop dev reference on error in rbd_open()
|
|
If a read-only rbd device is opened for writing in rbd_open(), it
returns without dropping the just-acquired device reference.
Fix this by moving the read-only check before getting the reference.
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Yehuda Sadeh <yehuda@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
|
|
Pull networking updates from David Miller:
"More bug fixes, nothing gets past these guys"
1) More kernel info leaks found by Mathias Krause, this time in the
IPSEC configuration layers.
2) When IPSEC policies change, we do not properly make sure that cached
routes (which could now be stale) throughout the system will be
revalidated. Fix this by generalizing the generation count
invalidation scheme used by ipv4. From Nicolas Dichtel.
3) When repairing TCP sockets, we need to allow to restore not just the
send window scale, but the receive one too. Extend the existing
interface to achieve this in a backwards compatible way. From
Andrey Vagin.
4) A fix for FCOE scatter gather feature validation erroneously caused
scatter gather to be disabled for things like AOE too. From Ed L
Cashin.
5) Several cases of mishandling of error pointers, from Mathias Krause,
Wei Yongjun, and Devendra Naga.
6) Fix gianfar build, from Richard Cochran.
7) CAP_NET_* failures should return -EPERM not -EACCES, from Zhao
Hongjiang.
8) Hardware reset fix in janz-ican3 CAN driver, from Ira W Snyder.
9) Fix oops during rmmod in ti_hecc CAN driver, from Marc Kleine-Budde.
10) The removal of the conditional compilation of the clk support code
in the stmmac driver broke things. This is because the interfaces
used are the ones that don't also perform the enable/disable of the
clk. Fix from Stefan Roese.
11) The QFQ packet scheduler can record out of range virtual start
times, resulting later in misbehavior and even crashes. Fix from
Paolo Valente.
12) If MSG_WAITALL is used with IOAT DMA under TCP, we can wedge the
receiver when the advertised receive window goes to zero. Detect
this case and force the processing of the IOAT DMA queue when it
happens to avoid getting stuck. Fix from Michal Kubecek.
13) batman-adv assumes that test_bit() returns only 0 or 1, but this is
not true for x86 (which returns -1 or 0, via the 'sbb' instruction).
Fix from Linus Lussing.
14) Fix small packet corruption in e1000, from Tushar Dave.
15) make_blackhole() in the IPSEC policy code can do one read unlock too
many, fix from Li RongQing.
16) The new tcp_try_coalesce() code introduced a bug in TCP URG
handling, fix from Eric Dumazet.
17) Fix memory leak in __netif_receive_skb() when doing zerocopy and
when hit an OOM condition. From Michael S Tsirkin.
18) netxen blindly deferences pdev->bus->self, which is not guarenteed
to be non-NULL. Fix from Nikolay Aleksandrov.
19) Fix a performance regression caused by mistakes in ipv6 checksum
validation in the bnx2x driver, fix from Michal Schmidt.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (45 commits)
net/stmmac: Use clk_prepare_enable and clk_disable_unprepare
net: change return values from -EACCES to -EPERM
net/irda: sh_sir: fix return value check in sh_sir_set_baudrate()
stmmac: fix return value check in stmmac_open_ext_timer()
gianfar: fix phc index build failure
ipv6: fix return value check in fib6_add()
bnx2x: remove false warning regarding interrupt number
can: ti_hecc: fix oops during rmmod
can: janz-ican3: fix support for older hardware revisions
net: do not disable sg for packets requiring no checksum
aoe: assert AoE packets marked as requiring no checksum
at91ether: return PTR_ERR if call to clk_get fails
xfrm_user: don't copy esn replay window twice for new states
xfrm_user: ensure user supplied esn replay window is valid
xfrm_user: fix info leak in copy_to_user_tmpl()
xfrm_user: fix info leak in copy_to_user_policy()
xfrm_user: fix info leak in copy_to_user_state()
xfrm_user: fix info leak in copy_to_user_auth()
net: qmi_wwan: adding Huawei E367, ZTE MF683 and Pantech P4200
tcp: restore rcv_wscale in a repair mode (v2)
...
|
