Commit Graph

80 Commits

Author SHA1 Message Date
paul luse
34146a7a52 lib/reduce: change and move the max IOVEC define for reduce
Changed to 17 from 32 per conversation with Jim and moved to
header file so it can be shared with the vbdev module.

Change-Id: Ic516b6466bf21d21d8073e9180deaa9929c6b742
Signed-off-by: paul luse <paul.e.luse@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463368
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com>
2019-07-29 04:36:59 +00:00
paul luse
c2d441226a lib/reduce: close hole in compress optimization
If a write operation comes in and is not a read-modify-write but
we don't end up compressing the data (because of either error or
we decide the compression ratio isn't good enough) then we not
only need to copy the data to the host buffers but we need to
zero any offset and/or remainder in the decomp scratch buffer.

Change-Id: Ifb2235507826f9ef1110dd9dbaf88045d8979e7c
Signed-off-by: paul luse <paul.e.luse@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463337
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
2019-07-29 03:33:40 +00:00
paul luse
f9cae9d20f lib/reduce: minor cleanup for consistency in recent changes
Had used a few different means to determine if there's space
remaining in a chunk after accounting for offset and host data,
just making them all consistent. Nothing functionally changes.

Change-Id: Iffd3b0b6d8ab1253f5b1bebda88d6f63d5061e19
Signed-off-by: paul luse <paul.e.luse@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463023
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
2019-07-26 17:44:18 +00:00
paul luse
7384ca015b lib/reduce: for compress path, copy data from host buffers if needed
If the data was not compressed for whatever reason, it needs to be
copied from the host buffers much in the same way as the patch before
this did.

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I5c92db16afbce3ca573c556ee78a51ba6511c564
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/462895
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-07-25 14:50:25 +00:00
paul luse
907ec7268d lib/reduce: conditionally memcpy data to host follow decompression
If the decompression code path runs into a chunk that wasn't
compressed, the data needs to be copied to the host as the
compression engine was not engaged.

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: Ibabd6d6bd2bc5db79953a0a55d7f49d556c08af7
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/462892
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-07-25 14:50:25 +00:00
paul luse
0242a80793 lib/reduce: fix problem with size of chunk map area in PMEM
This fixes a segfault found by using a very small logical volume.
The PMEM file chunk map area wasn't including the chunk map struct
itself.  In PMEM, a single entry as returned by _reduce_vol_get_chunk_map()
is made up of a 'struct spdk_reduce_chunk_map' plus one uint64_t for
each backing IO unit for the volume (# depends on vol parms). So when
an app got close to the end of the chunk map indices it would run off
the end of PMEM and segfault.

With this fix that no longer happens however data integrity issues now
show up with fio. I believe this fix is valid though and am pursuing the data
integrity issue as something new.

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I69a223064c7bbd44bcd0a62d155cf409603970a3
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/461316
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-25 14:50:25 +00:00
paul luse
eb1ddcb664 lib/reduce: remove dead code clearing unused backing io units
At the start of time all of values of io_unit_index[] array in
the chunk maps are FF's.  When a partial chunk is written, only
some of the io_unit_index[] vals will be assigned values from the
allocated_backing_io_units bit array.  When that chunk is written
again, it is not done so in place and at that time in when
_write_write_done() is called, the io_unit_index[] values are
set to FF again and the bit array bits are cleared for the old
chunk so that it is fresh again for use by another write.

This change removes code from an early stage of the write process
where a 'fresh' chuck in allocated.  Because of points made earlier,
it is not possible for any of the io_unit_index[] values to be
assigned or have a corresponding bit set in the bit array.

The prev version of this patch cleared the bit array because the
io_unit_index[] array was being cleared however after further
investigation, and for the reasons stated earlier, the io_unit_index[]
did not have a valid value so clearing the bit array was simply wrong.

Removing this dead code clears things up and leaves it to one place
to clear the io_unit_index[], when a chunk is overwritten.  Well, 2
places including init time.

Change-Id: I5c4e5360c9c13817d2fa7d3c7d004b44537af7f4
Signed-off-by: paul luse <paul.e.luse@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/458520
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-24 18:12:05 +00:00
paul luse
a755fb5f14 lib/reduce: use global zero buffer instead of zeroing scratch buffer
Eliminates need for memset on parts of our scratch buffer

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I00b7213d3c15562ceeda4d7a3ac2bb7cfd41bf66
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/460010
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-07-18 03:21:59 +00:00
Darek Stojaczyk
4617707d07 reduce: switch to spdk_*malloc()
spdk_dma_*malloc() is about to be deprecated.

Change-Id: I140e10b2fd07efb48e664cfa00e1d60f604abd21
Signed-off-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449797
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-07-16 10:51:47 +00:00
paul luse
0b3fb2403e lib/reduce: fix bug with adding up req->decomp_iovcnt
In the memcpy elimination patches, the same bug exists in 3
places.  When building req->decomp_iov using the host buffers,
req->decomp_iovcnt was being incremented in the loop and also
being used as part of the index messing everything up.

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I485ac32502801c1e11b8392b2df7eba06b4f5a9b
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/461053
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
2019-07-12 04:05:52 +00:00
paul luse
b9bc6254a8 lib/reduce: fix critical issue with reduce optimization
The first optimization to eliminate memcpy was too aggressive and
did so for the read-modify-write operation as well. This didn't
affect the fio tests used that the time but bdevio catches it
right away.  When over writing a chunk with data, we first need
to read the old data before applying the new. This patch uses
the scratch buffer for old data as sending it to the user buffer
results in it not being written at the end of the read-modify-write.

There is at least one more bug fix coming after this also found
with bdevio but passed with fio

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I8fe074056434bb4757c68077e2df446861edfd94
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/461032
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-07-12 04:05:52 +00:00
paul luse
63d9d2e2b0 lib/reduce: eliminate RMW on writes with chunk_size length
Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I6545a91e2ae4805f7bd1d92baa6dcbce0f1f8fba
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/459864
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2019-07-05 11:56:03 +00:00
paul luse
d23c36a169 lib/reduce: eliminate two more memcpy operations
For callers of _reduce_vol_compress_chunk()

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I9ac1da8f9bcfd902fe58e4c5ffc20ce16e9bafcd
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/459863
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
2019-07-05 11:56:03 +00:00
paul luse
89a9a50497 lib/reduce: eliminate memcpy in read decompression path
This is the first in a series of patches to eliminate memcpy ops
in the comp/decomp paths. Currently the lib uses 2 scratch buffers
and copies all data in and out of them to the user buffers following
a comp/decomp. This patch replaces the memcpy in one of the paths by
constructing an iovec array that points to a combination of the
scratch buffer and user buffer so that user data decompresses
directly into the user buffer and any data in the chunk that isn't
needed by the user will be sent to the scratch buffer.

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: Ib1956875729a82d218527bc81795f750d1df2b89
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/459662
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-07-05 11:56:03 +00:00
Chunyang Hui
863f17d609 reduce: check pmem buf before unmap
Fixed issue #831

Change-Id: Id589290f3aa729572fa81daf735cecdc8e2adb84
Signed-off-by: Chunyang Hui <chunyang.hui@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/458563
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-06-25 13:39:09 +00:00
paul luse
11bbb26e4f lib/reduce: use helper function to get correct length of logical map
When setting the offset into pmem for the chunk mamps, the wrong
value was used for the length of the logical map resulting in
map corruption after extended fio run.

Fixes issue #813

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: Icb51cf8f38a04f05f69595ebd510ced7f1b608f3
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/457818
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-06-18 04:06:41 +00:00
paul luse
d571336b4b lib/reduce: add debug helper function
Prints key info about a volume and its PMEM. As the metadata is
mapped, there's no real reason to dump all this to a file, it
can be accessed via gdb using the addresses printed out.

Signed-off-by: paul luse <paul.e.luse@intel.com>
Change-Id: I2c7b8c32b6142ba0fded623a660b4d7d0c5b19bc
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/458117
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-06-18 04:06:41 +00:00
paul luse
12f7aeb4ce lib/reduce: avoid extra operation in write path
Change-Id: I92355d3deb25f818e6d4c7a72d2f4fd45e6879ac
Signed-off-by: paul luse <paul.e.luse@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/457523
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-06-11 18:13:10 +00:00
paul luse
6dc095c430 lib/reduce: add new API to retrieve volume parameters
Change-Id: Ic10812e7ababbec2964ee26ce7ffad28bd0daad5
Signed-off-by: paul luse <paul.e.luse@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/452905
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-05-13 15:41:35 +00:00
Jim Harris
d25f2db94a reduce: account for compressed buffers
When we decompress, pass the size of the compressed
data - not the size of the buffer itself.

We also need to remove an assert that's no longer
true since we can now handle compressed chunks.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ifb7896db12ccbefdd49815b5037af11ddd82e625

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449513
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2019-04-02 21:40:21 +00:00
Jim Harris
fa0d54d431 reduce: write uncompressed data when compress returns error
Most typically, the error will be -ENOSPC, meaning the
destination buffer wasn't big enough to hold the compressed
data.  But even for other types of errors, just write the
uncompressed data rather than treating it as an I/O
failure.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Id6526cab95488a1a587b183237f9980c83287761

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449511
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2019-04-02 21:40:21 +00:00
Jim Harris
df082c931e reduce: fix chunk_is_compressed calculation
Stupid bug - a chunk is compressed when the number
of io_units for the compressed data does *not* equal
the number of io_units per chunk.

Paul found this during some of his initial DPDK
framework integration testing.  I have some better
unit tests coming up to test this further, but want
to get this obvious fix in for now.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Iaf9c78a1c1d6045070e7625c6a54ab3d227c8ea7

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449498
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
2019-03-29 02:34:21 +00:00
Jim Harris
2edc652913 reduce: save num_io_units and chunk_is_compressed to req object
When a chunk cannot be compressed enough to save one I/O
unit, we will just write the uncompressed data to disk.  In
this case, we want the helper routine to use the decomp_buf
instead of the comp_buf.

To facilitate that, save some information in the request
object on the number of io_units to be read from or written
to disk, and whether the data is compressed or not.  We
could easily deduce if the data is compressed based on the
number of io_units but saving it in the structure will
save a few instructions and simplify the code a bit.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I1537130577943e19fe4dfcee58eac453eca3c29b

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449259
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-03-28 03:45:29 +00:00
Jim Harris
309f7791b9 reduce: plumb basic compress/decompress callbacks
The unit tests don't really try to compress anything
yet, but this at least gets the pipeline in place.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ic413850b30e4d9631f3ece2bab40d9026225e5b2

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449097
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-03-28 03:45:29 +00:00
Jim Harris
72da9b7562 reduce: use consistent callback names
There are a sequence of operations needed for both
read and write I/O.  For example, on a write I/O,
we may need to do a read, modify the chunk, then
write the data back to disk.

For read I/O, the name of the callback is
_read_read_done - the first "read" indicates the
I/O type, the second "read" indicates the disk read
part of the sequence.

The write I/O steps aren't using this same naming
convention, so change that here.  This will be
important in upcoming patches where we'll be adding
compress/decompress steps into the I/O pipeline -
having a consistent naming strategy will make the code
easier to follow.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Icd525874918378118d807acdbe5dd42d5be03126

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449258
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2019-03-28 03:45:29 +00:00
Jim Harris
f9ac7dcf11 reduce: move code into _read_complete_req helper function
This preps for some future changes.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I380258938e2412062b43c3d460a7322c5afdf819

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449096
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
2019-03-28 03:45:29 +00:00
Jim Harris
09aff2d0fd reduce: add intermediate buffers for compression operations
We will actually use these in future patches.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I730ef90da6d93922dd17a5ac16594e66132a5f11

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449086
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
2019-03-28 03:45:29 +00:00
Jim Harris
20145fd714 reduce: prepend rw to request buf and buf_iov
We will need separate temporary buffers for compress/decompress.
A single temporary buffer won't do because a user's read/write
operations may not cover a full chunk - so we'll need one buffer
to read/write the compressed data, and another buffer for
the uncompressed data.

So for now, just prepend rw to the existing fields - this
will signify these fields are used for the read/write operations
to the backing storage device.  We'll add additional ones in future
patches for the buffers that will hold the uncompressed data.

Note: the vol->buf_mem and ->buf_iov_mem are not prepending "rw"
because they will be used for the compress/decompress temporary
buffers as well.  We'll just double the size of these buffers
when allocating them for the volume.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I7df56ab72769f4689f3abac3354446e9cf78d423

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449085
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
2019-03-27 14:20:26 +00:00
Jim Harris
502ab5b66d reduce: rename vol->reqbufspace to buf_mem
This matches better with the existing buf_iov_mem.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I3b85708672553c281666becc8eaf9f9028da8782

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449084
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: wuzhouhui <wuzhouhui@kingsoft.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
2019-03-27 14:20:26 +00:00
Jim Harris
5a04099b01 reduce: save compressed data size in chunk map
Upcoming patches will try to compress the data before
writing it to backing I/O units, and we'll need to save
the exact number of compressed bytes.

For now, since we're not actually compressing data yet,
just save the uncompressed data size to this field.  There
will be cases when we cannot realize any compression
savings and will just store the uncompressed data, so
handling this now is one less path we'll need to add later.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I499a89590349e4816fe0fef77cc1bfabde6871bf

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449078
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
2019-03-27 14:20:26 +00:00
Jim Harris
3bb1f95570 reduce: add struct for chunk maps
Previously we were just passing around uint64_t pointers,
since the chunk map was literally only an array of uint64_t.
But we need to store the compressed data size in the chunk
as well.  Before adding the compressed data size, let's do
the prep work here to just create the chunk map structure
which for now only includes an array of uint64_t.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Icbfa6e509463d4f560444d3db8acd9e87917835f
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/449076
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: wuzhouhui <wuzhouhui@kingsoft.com>
Reviewed-by: Paul Luse <paul.e.luse@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
2019-03-27 14:20:26 +00:00
Jim Harris
4fb2260117 lib/reduce: queue overlapping I/O to same logical chunk
Write operations are not in-place, meaning that once it
is completed, it's chunk and backing io units are freed.
Reads in progress to these backing io units, from other
I/O to the same chunk, could cause data corruption.
Simultaneous writes to the same chunk can cause similar
issues.

There are cases where we can relax this in the future -
for example, overlapped reads to the same chunk poses
no problems.  We'll leave that as a future exercise.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I3a4d9e8e3ab920f823084cd1bae91f179a4386ff
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/448333
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-03-25 04:39:26 +00:00
Jim Harris
b1c5a962d4 lib/reduce: create helper functions for starting read/write ops
Upcoming changes will require queuing an I/O if it overlaps
with another I/O to the same logical chunk.  So split out
code (especially from the writev path) into separate functions,
so that we can execute an spdk_reduce_vol_request object
whether it's just been created via readv/writev, or after
another I/O to the same chunk has been completed.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I2a216c92e4999313ecc9f7cf157dccd65c9bb30f

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/448332
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: wuzhouhui <wuzhouhui@kingsoft.com>
2019-03-22 18:28:43 +00:00
Jim Harris
d7c80bfb44 lib/reduce: save logical_map_index in the request object
This will help optimize some upcoming patches which
guard against overlapped I/O to the same chunk in the logical
map.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ia1370fffd3bf3d55dc9ce88a993ed157ecee272a

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/448331
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: wuzhouhui <wuzhouhui@kingsoft.com>
2019-03-22 18:28:43 +00:00
Jim Harris
1036358ecf lib/reduce: rename variable to old_chunk_map_index
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I39479f44684e2c5e6652d07237ac40949dcf0aca

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/448330
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: wuzhouhui <wuzhouhui@kingsoft.com>
2019-03-22 18:28:43 +00:00
paul luse
cf467578f8 bdev/reduce: fix issue with vol_load
Discovered in bdev compression test.  Vol load was using the index
of the logical map to set the allocated chunk map bits as well as
retrieve the chunk. Exposed with overwrite situations. Also clarrified
one var name that I missed in previous patch.

Also added UT that fails without the fix, passes with the fix.

Change-Id: I776a5af9364cc3d4e8b7dafec22acf23bd463630
Signed-off-by: paul luse <paul.e.luse@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/447969
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2019-03-18 17:41:32 +00:00
paul luse
228396f898 lib/reduce: change some var names
Changed a few cases where var name chunk was used where a better
name would be logical_map_index to avoid confusion with 'chunk'
used as a data pointer.

Change-Id: Ib82a783725cad4609c31bfa8997debb3e71e75b9
Signed-off-by: paul luse <paul.e.luse@intel.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/447962
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2019-03-18 17:41:32 +00:00
paul luse
5abe0ec852 reduce: fix ordering bug
_allocate_bit_arrays() needs vol->backing_dev set which was being done
after the call.

Change-Id: Ic8c36c98aee94fbd8230273638011b948cd95675
Signed-off-by: paul luse <paul.e.luse@intel.com>
Reviewed-on: https://review.gerrithub.io/c/443048
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: wuzhouhui <wuzhouhui@kingsoft.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
2019-02-04 19:23:35 +00:00
Jim Harris
1fa0283f31 reduce: mark correct number of backing pages for md
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Iebdfe79dd8e4240f2333db6a758193e048f9b5f2

Reviewed-on: https://review.gerrithub.io/c/440574
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2019-01-17 19:43:28 +00:00
Jim Harris
42aa5a262b reduce: remove close callback
Let the application figure out how to close the
underlying block device after an init fails or
an initialized/loaded volume is closed/destroyed.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Id0b9b99c85ed13166ec9a229eaaa0b0eb04289fd

Reviewed-on: https://review.gerrithub.io/c/440573
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: wuzhouhui <wuzhouhui@kingsoft.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-01-16 22:25:13 +00:00
Jim Harris
c0f3805153 reduce: add spdk_reduce_vol_destroy
This will remove the metadata from the associated backing
device and delete the pm_file associated with the reduce
volume.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I30ffd68e5ba69b31ee1be85418c5b8c592d82e70

Reviewed-on: https://review.gerrithub.io/c/437995
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: wuzhouhui <wuzhouhui@kingsoft.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
2019-01-16 22:25:13 +00:00
Jim Harris
63ba545f54 reduce: rebuild bit arrays during load
This requires fixing up the unit test pmem emulation
code - we need to copy the 'persistent' buffer into
the newly 'mmaped' buffer in the pmem_map_file stub.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I4e474f437922e652a57d0b45b6ef92fc713eb587

Reviewed-on: https://review.gerrithub.io/c/437888
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
2019-01-16 22:25:13 +00:00
wuzhouhui
8db5ff2bdd reduce: _allocate_vol_requests: set pointer to NULL after freed
If _allocate_vol_requests() failed, the caller will call
_init_load_cleanup() to free memory that _allocate_vol_requests()
already freed, and cause segment fault. Setting pointer to NULL keeps
_init_load_cleanup() is safe to free it again.

Change-Id: I9ad09eabf6b65350f174bd5a4faf5ee643cbd23f
Signed-off-by: wuzhouhui <wuzhouhui@kingsoft.com>
Reviewed-on: https://review.gerrithub.io/437292
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2018-12-18 23:38:51 +00:00
wuzhouhui
3a4185be19 reduce: check strlen(SPDK_REDUCE_SIGNATURE) in buildtime
Change-Id: I4d12f2a693a4a34c840d972a4f60cbd0fc35011b
Signed-off-by: wuzhouhui <wuzhouhui@kingsoft.com>
Reviewed-on: https://review.gerrithub.io/437236
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2018-12-18 23:34:17 +00:00
wuzhouhui
7fbc5106e4 reduce: put pmem_unmap() in _init_load_cleanup()
In effect, this commit fixes error path of spdk_reduce_vol_init()
and spdk_reduce_vol_load(): forget to unmap pmem file after mapped.

Change-Id: I797f15e315fff3ff42c17509a73dc46e7f6bdb24
Signed-off-by: wuzhouhui <wuzhouhui@kingsoft.com>
Reviewed-on: https://review.gerrithub.io/437270
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
2018-12-18 23:33:50 +00:00
Konrad Sztyber
62db4ac2cf util: added spdk_divide_round_up()
Replaced divide_round_up() from blobstore.c, lvol.c and reduce.c with
new spdk_divide_round_up() from util.h.

Change-Id: I013383ac286ca52b5c15c7fab4fb40ad97b92656
Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-on: https://review.gerrithub.io/437649
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
2018-12-18 17:26:49 +00:00
Jim Harris
3d1c1f945c reduce: don't try to clear io units for empty chunk maps
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Reported-by: Shuhei Matsumoto
Change-Id: Idcf427e32962236acfeb7dab793e27c44cdc6151

Reviewed-on: https://review.gerrithub.io/436878
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
2018-12-13 14:55:39 +00:00
Jim Harris
0b0af8f30b reduce: calculate vol size from backing dev size
This more closely matches the sequence of events
a user will take in creating a compressed volume.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I2311aed49e2a623feb9c5fbdbd25fb3201aeeeb5

Reviewed-on: https://review.gerrithub.io/436065
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2018-12-13 14:55:39 +00:00
Jim Harris
8c2cf577b3 reduce: remove extra size check
This was the result of a previous merge conflict.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Ic657e7f9e16b6a073669b078b4323f7abfef95b9

Reviewed-on: https://review.gerrithub.io/436063
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2018-12-13 14:55:39 +00:00
Jim Harris
1c68d71107 reduce: change some member names to make them more descriptive
Based on feedback from Paul Luse in some earlier reviews.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I1930a3de05ada15d3ac4530c081c2f951b96fe97

Reviewed-on: https://review.gerrithub.io/435368
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Chandler-Test-Pool: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
2018-12-13 14:55:39 +00:00