Justification
-------------
This feature adds support for variable length dnodes. Our motivation is
to eliminate the overhead associated with using spill blocks. Spill
blocks are used to store system attribute data (i.e. file metadata) that
does not fit in the dnode's bonus buffer. By allowing a larger bonus
buffer area the use of a spill block can be avoided. Spill blocks
potentially incur an additional read I/O for every dnode in a dnode
block. As a worst case example, reading 32 dnodes from a 16k dnode block
and all of the spill blocks could issue 33 separate reads. Now suppose
those dnodes have size 1024 and therefore don't need spill blocks. Then
the worst case number of blocks read is reduced to from 33 to two--one
per dnode block. In practice spill blocks may tend to be co-located on
disk with the dnode blocks so the reduction in I/O would not be this
drastic. In a badly fragmented pool, however, the improvement could be
significant.
ZFS-on-Linux systems that make heavy use of extended attributes would
benefit from this feature. In particular, ZFS-on-Linux supports the
xattr=sa dataset property which allows file extended attribute data
to be stored in the dnode bonus buffer as an alternative to the
traditional directory-based format. Workloads such as SELinux and the
Lustre distributed filesystem often store enough xattr data to force
spill bocks when xattr=sa is in effect. Large dnodes may therefore
provide a performance benefit to such systems.
Other use cases that may benefit from this feature include files with
large ACLs and symbolic links with long target names. Furthermore,
this feature may be desirable on other platforms in case future
applications or features are developed that could make use of a
larger bonus buffer area.
Implementation
--------------
The size of a dnode may be a multiple of 512 bytes up to the size of
a dnode block (currently 16384 bytes). A dn_extra_slots field was
added to the current on-disk dnode_phys_t structure to describe the
size of the physical dnode on disk. The 8 bits for this field were
taken from the zero filled dn_pad2 field. The field represents how
many "extra" dnode_phys_t slots a dnode consumes in its dnode block.
This convention results in a value of 0 for 512 byte dnodes which
preserves on-disk format compatibility with older software.
Similarly, the in-memory dnode_t structure has a new dn_num_slots field
to represent the total number of dnode_phys_t slots consumed on disk.
Thus dn->dn_num_slots is 1 greater than the corresponding
dnp->dn_extra_slots. This difference in convention was adopted
because, unlike on-disk structures, backward compatibility is not a
concern for in-memory objects, so we used a more natural way to
represent size for a dnode_t.
The default size for newly created dnodes is determined by the value of
a new "dnodesize" dataset property. By default the property is set to
"legacy" which is compatible with older software. Setting the property
to "auto" will allow the filesystem to choose the most suitable dnode
size. Currently this just sets the default dnode size to 1k, but future
code improvements could dynamically choose a size based on observed
workload patterns. Dnodes of varying sizes can coexist within the same
dataset and even within the same dnode block. For example, to enable
automatically-sized dnodes, run
# zfs set dnodesize=auto tank/fish
The user can also specify literal values for the dnodesize property.
These are currently limited to powers of two from 1k to 16k. The
power-of-2 limitation is only for simplicity of the user interface.
Internally the implementation can handle any multiple of 512 up to 16k,
and consumers of the DMU API can specify any legal dnode value.
The size of a new dnode is determined at object allocation time and
stored as a new field in the znode in-memory structure. New DMU
interfaces are added to allow the consumer to specify the dnode size
that a newly allocated object should use. Existing interfaces are
unchanged to avoid having to update every call site and to preserve
compatibility with external consumers such as Lustre. The new
interfaces names are given below. The versions of these functions that
don't take a dnodesize parameter now just call the _dnsize() versions
with a dnodesize of 0, which means use the legacy dnode size.
New DMU interfaces:
dmu_object_alloc_dnsize()
dmu_object_claim_dnsize()
dmu_object_reclaim_dnsize()
New ZAP interfaces:
zap_create_dnsize()
zap_create_norm_dnsize()
zap_create_flags_dnsize()
zap_create_claim_norm_dnsize()
zap_create_link_dnsize()
The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The
spa_maxdnodesize() function should be used to determine the maximum
bonus length for a pool.
These are a few noteworthy changes to key functions:
* The prototype for dnode_hold_impl() now takes a "slots" parameter.
When the DNODE_MUST_BE_FREE flag is set, this parameter is used to
ensure the hole at the specified object offset is large enough to
hold the dnode being created. The slots parameter is also used
to ensure a dnode does not span multiple dnode blocks. In both of
these cases, if a failure occurs, ENOSPC is returned. Keep in mind,
these failure cases are only possible when using DNODE_MUST_BE_FREE.
If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0.
dnode_hold_impl() will check if the requested dnode is already
consumed as an extra dnode slot by an large dnode, in which case
it returns ENOENT.
* The function dmu_object_alloc() advances to the next dnode block
if dnode_hold_impl() returns an error for a requested object.
This is because the beginning of the next dnode block is the only
location it can safely assume to either be a hole or a valid
starting point for a dnode.
* dnode_next_offset_level() and other functions that iterate
through dnode blocks may no longer use a simple array indexing
scheme. These now use the current dnode's dn_num_slots field to
advance to the next dnode in the block. This is to ensure we
properly skip the current dnode's bonus area and don't interpret it
as a valid dnode.
zdb
---
The zdb command was updated to display a dnode's size under the
"dnsize" column when the object is dumped.
For ZIL create log records, zdb will now display the slot count for
the object.
ztest
-----
Ztest chooses a random dnodesize for every newly created object. The
random distribution is more heavily weighted toward small dnodes to
better simulate real-world datasets.
Unused bonus buffer space is filled with non-zero values computed from
the object number, dataset id, offset, and generation number. This
helps ensure that the dnode traversal code properly skips the interior
regions of large dnodes, and that these interior regions are not
overwritten by data belonging to other dnodes. A new test visits each
object in a dataset. It verifies that the actual dnode size matches what
was stored in the ztest block tag when it was created. It also verifies
that the unused bonus buffer space is filled with the expected data
patterns.
ZFS Test Suite
--------------
Added six new large dnode-specific tests, and integrated the dnodesize
property into existing tests for zfs allow and send/recv.
Send/Receive
------------
ZFS send streams for datasets containing large dnodes cannot be received
on pools that don't support the large_dnode feature. A send stream with
large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be
unrecognized by an incompatible receiving pool so that the zfs receive
will fail gracefully.
While not implemented here, it may be possible to generate a
backward-compatible send stream from a dataset containing large
dnodes. The implementation may be tricky, however, because the send
object record for a large dnode would need to be resized to a 512
byte dnode, possibly kicking in a spill block in the process. This
means we would need to construct a new SA layout and possibly
register it in the SA layout object. The SA layout is normally just
sent as an ordinary object record. But if we are constructing new
layouts while generating the send stream we'd have to build the SA
layout object dynamically and send it at the end of the stream.
For sending and receiving between pools that do support large dnodes,
the drr_object send record type is extended with a new field to store
the dnode slot count. This field was repurposed from unused padding
in the structure.
ZIL Replay
----------
The dnode slot count is stored in the uppermost 8 bits of the lr_foid
field. The bits were unused as the object id is currently capped at
48 bits.
Resizing Dnodes
---------------
It should be possible to resize a dnode when it is dirtied if the
current dnodesize dataset property differs from the dnode's size, but
this functionality is not currently implemented. Clearly a dnode can
only grow if there are sufficient contiguous unused slots in the
dnode block, but it should always be possible to shrink a dnode.
Growing dnodes may be useful to reduce fragmentation in a pool with
many spill blocks in use. Shrinking dnodes may be useful to allow
sending a dataset to a pool that doesn't support the large_dnode
feature.
Feature Reference Counting
--------------------------
The reference count for the large_dnode pool feature tracks the
number of datasets that have ever contained a dnode of size larger
than 512 bytes. The first time a large dnode is created in a dataset
the dataset is converted to an extensible dataset. This is a one-way
operation and the only way to decrement the feature count is to
destroy the dataset, even if the dataset no longer contains any large
dnodes. The complexity of reference counting on a per-dnode basis was
too high, so we chose to track it on a per-dataset basis similarly to
the large_block feature.
Signed-off-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#3542
When running a kernel with CONFIG_LOCKDEP=y, lockdep reports possible
recursive locking in some cases and possible circular locking dependency
in others, within the SPL and ZFS modules.
This patch uses a mutex type defined in SPL, MUTEX_NOLOCKDEP, to mark
such mutexes when they are initialized. This mutex type causes
attempts to take or release those locks to be wrapped in lockdep_off()
and lockdep_on() calls to silence the dependency checker and allow the
use of lock_stats to examine contention.
For RW locks, it uses an analogous lock type, RW_NOLOCKDEP.
The goal is that these locks are ultimately changed back to type
MUTEX_DEFAULT or RW_DEFAULT, after the locks are annotated to reflect
their relationship (e.g. z_name_lock below) or any real problem with the
lock dependencies are fixed.
Some of the affected locks are:
tc_open_lock:
=============
This is an array of locks, all with same name, which txg_quiesce must
take all of in order to move txg to next state. All default to the same
lockdep class, and so to lockdep appears recursive.
zp->z_name_lock:
================
In zfs_rmdir,
dzp = znode for the directory (input to zfs_dirent_lock)
zp = znode for the entry being removed (output of zfs_dirent_lock)
zfs_rmdir()->zfs_dirent_lock() takes z_name_lock in dzp
zfs_rmdir() takes z_name_lock in zp
Since both dzp and zp are type znode_t, the locks have the same default
class, and lockdep considers it a possible recursive lock attempt.
l->l_rwlock:
============
zap_expand_leaf() sometimes creates two new zap leaf structures, via
these call paths:
zap_deref_leaf()->zap_get_leaf_byblk()->zap_leaf_open()
zap_expand_leaf()->zap_create_leaf()->zap_expand_leaf()->zap_create_leaf()
Because both zap_leaf_open() and zap_create_leaf() initialize
l->l_rwlock in their (separate) leaf structures, the lockdep class is
the same, and the linux kernel believes these might both be the same
lock, and emits a possible recursive lock warning.
Signed-off-by: Olaf Faaland <faaland1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#3895
Author: Alex Reece <alex@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Paul Dagnelie <paul.dagnelie@delphix.com>
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
Reviewed by: Albert Lee <trisk@nexenta.com>
Approved by: Dan McDonald <danmcd@omniti.com>
References:
https://www.illumos.org/issues/5422https://github.com/illumos/illumos-gate/commit/a846f19
Ported-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#3381
5027 zfs large block support
Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
Reviewed by: Richard Elling <richard.elling@richardelling.com>
Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Approved by: Dan McDonald <danmcd@omniti.com>
References:
https://www.illumos.org/issues/5027https://github.com/illumos/illumos-gate/commit/b515258
Porting Notes:
* Included in this patch is a tiny ISP2() cleanup in zio_init() from
Illumos 5255.
* Unlike the upstream Illumos commit this patch does not impose an
arbitrary 128K block size limit on volumes. Volumes, like filesystems,
are limited by the zfs_max_recordsize=1M module option.
* By default the maximum record size is limited to 1M by the module
option zfs_max_recordsize. This value may be safely increased up to
16M which is the largest block size supported by the on-disk format.
At the moment, 1M blocks clearly offer a significant performance
improvement but the benefits of going beyond this for the majority
of workloads are less clear.
* The illumos version of this patch increased DMU_MAX_ACCESS to 32M.
This was determined not to be large enough when using 16M blocks
because the zfs_make_xattrdir() function will fail (EFBIG) when
assigning a TX. This was immediately observed under Linux because
all newly created files must have a security xattr created and
that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M.
* On 32-bit platforms a hard limit of 1M is set for blocks due
to the limited virtual address space. We should be able to relax
this one the ABD patches are merged.
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#354
5056 ZFS deadlock on db_mtx and dn_holds
Author: Justin Gibbs <justing@spectralogic.com>
Reviewed by: Will Andrews <willa@spectralogic.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.com>
References:
https://www.illumos.org/issues/5056https://github.com/illumos/illumos-gate/commit/bc9014e
Porting Notes:
sa_handle_get_from_db():
- the original patch includes an otherwise unmentioned fix for a
possible usage of an uninitialised variable
dmu_objset_open_impl():
- Under Illumos list_link_init() is the same as filling a list_node_t
with NULLs, so they don't notice if they miss doing list_link_init()
on a zero'd containing structure (e.g. allocated with kmem_zalloc as
here). Under Linux, not so much: an uninitialised list_node_t goes
"Boom!" some time later when it's used or destroyed.
dmu_objset_evict_dbufs():
- reduce stack usage using kmem_alloc()
Ported-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
5095 panic when adding a duplicate dbuf to dn_dbufs
Author: Alex Reece <alex@delphix.com>
Reviewed by: Adam Leventhal <adam.leventhal@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Mattew Ahrens <mahrens@delphix.com>
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Reviewed by: Josef Sipek <jeffpc@josefsipek.net>
Approved by: Robert Mustacchi <rm@joyent.com>
References:
https://www.illumos.org/issues/5095https://github.com/illumos/illumos-gate/commit/86bb58a
Ported-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
4873 zvol unmap calls can take a very long time for larger datasets
Author: Alex Reece <alex@delphix.com>
Reviewed by: George Wilson <george@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Paul Dagnelie <paul.dagnelie@delphix.com>
Reviewed by: Basil Crow <basil.crow@delphix.com>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Approved by: Robert Mustacchi <rm@joyent.com>
References:
https://www.illumos.org/issues/4873https://github.com/illumos/illumos-gate/commit/0f6d88a
Porting Notes:
dbuf_free_range():
- reduce stack usage using kmem_alloc()
- the sorted AVL tree will handle the spill block case correctly
without all the special handling in the for() loop
Ported-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
5630 stale bonus buffer in recycled dnode_t leads to data corruption
Author: Justin T. Gibbs <justing@spectralogic.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george@delphix.com>
Reviewed by: Will Andrews <will@freebsd.org>
Approved by: Robert Mustacchi <rm@joyent.com>
References:
https://www.illumos.org/issues/5630https://github.com/illumos/illumos-gate/commit/cd485b4
Ported-by: Chris Dunlop <chris@onthe.net.au>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Richard Yao <ryao@gentoo.org>
Issue #3172
By marking DMU transaction processing contexts with PF_FSTRANS
we can revert the KM_PUSHPAGE -> KM_SLEEP changes. This brings
us back in line with upstream. In some cases this means simply
swapping the flags back. For others fnvlist_alloc() was replaced
by nvlist_alloc(..., KM_PUSHPAGE) and must be reverted back to
fnvlist_alloc() which assumes KM_SLEEP.
The one place KM_PUSHPAGE is kept is when allocating ARC buffers
which allows us to dip in to reserved memory. This is again the
same as upstream.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Callers of kmem_alloc() which passed the KM_NODEBUG flag to suppress
the large allocation warning have been replaced by vmem_alloc() as
appropriate. The updated vmem_alloc() call will not print a warning
regardless of the size of the allocation.
A careful reader will notice that not all callers have been changed
to vmem_alloc(). Some have only had the KM_NODEBUG flag removed.
This was possible because the default warning threshold has been
increased to 32k. This is desirable because it minimizes the need
for Linux specific code changes.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Older versions of GCC (e.g. GCC 4.4.7 on RHEL6) do not allow duplicate
typedef declarations with the same type. The trace.h header contains
some typedefs to avoid 'unknown type' errors for C files that haven't
declared the type in question. But this causes build failures for C
files that have already declared the type. Newer versions of GCC (e.g.
v4.6) allow duplicate typedefs with the same type unless pedantic error
checking is in force. To support the older versions we need to remove
the duplicate typedefs.
Removal of the typedefs means we can't built tracepoints code using
those types unless the required headers have been included. To
facilitate this, all tracepoint event declarations have been moved out
of trace.h into separate headers. Each new header is explicitly included
from the C file that uses the events defined therein. The trace.h header
is still indirectly included form zfs_context.h and provides the
implementation of the dprintf(), dbgmsg(), and SET_ERROR() interfaces.
This makes those interfaces readily available throughout the code base.
The macros that redefine DTRACE_PROBE* to use Linux tracepoints are also
still provided by trace.h, so it is a prerequisite for the other
trace_*.h headers.
These new Linux implementation-specific headers do introduce a small
divergence from upstream ZFS in several core C files, but this should
not present a significant maintenance burden.
Signed-off-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #2953
5139 SEEK_HOLE failed to report a hole at end of file
Reviewed by: Adam Leventhal <adam.leventhal@delphix.com>
Reviewed by: Alex Reece <alex.reece@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Max Grossman <max.grossman@delphix.com>
Reviewed by: Peng Dai <peng.dai@delphix.com>
Reviewed by: Richard Elling <richard.elling@gmail.com>
Approved by: Dan McDonald <danmcd@omniti.com>
References:
https://www.illumos.org/issues/5139https://github.com/illumos/illumos-gate/commit/0fbc0cd
Ported by: Turbo Fredriksson <turbo@bayour.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2714
4757 ZFS embedded-data block pointers ("zero block compression")
4913 zfs release should not be subject to space checks
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Max Grossman <max.grossman@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Approved by: Dan McDonald <danmcd@omniti.com>
References:
https://www.illumos.org/issues/4757https://www.illumos.org/issues/4913https://github.com/illumos/illumos-gate/commit/5d7b4d4
Porting notes:
For compatibility with the fastpath code the zio_done() function
needed to be updated. Because embedded-data block pointers do
not require DVAs to be allocated the associated vdevs will not
be marked and therefore should not be unmarked.
Ported by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2544
4374 dn_free_ranges should use range_tree_t
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Max Grossman <max.grossman@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com
Reviewed by: Garrett D'Amore <garrett@damore.org>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Approved by: Dan McDonald <danmcd@omniti.com>
References:
https://www.illumos.org/issues/4374https://github.com/illumos/illumos-gate/commit/bf16b11
Ported by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2531
4370 avoid transmitting holes during zfs send
4371 DMU code clean up
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
Approved by: Garrett D'Amore <garrett@damore.org>a
References:
https://www.illumos.org/issues/4370https://www.illumos.org/issues/4371https://github.com/illumos/illumos-gate/commit/43466aa
Ported by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2529
4101 metaslab_debug should allow for fine-grained control
4102 space_maps should store more information about themselves
4103 space map object blocksize should be increased
4105 removing a mirrored log device results in a leaked object
4106 asynchronously load metaslab
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Sebastien Roy <seb@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>
Prior to this patch, space_maps were preferred solely based on the
amount of free space left in each. Unfortunately, this heuristic didn't
contain any information about the make-up of that free space, which
meant we could keep preferring and loading a highly fragmented space map
that wouldn't actually have enough contiguous space to satisfy the
allocation; then unloading that space_map and repeating the process.
This change modifies the space_map's to store additional information
about the contiguous space in the space_map, so that we can use this
information to make a better decision about which space_map to load.
This requires reallocating all space_map objects to increase their
bonus buffer size sizes enough to fit the new metadata.
The above feature can be enabled via a new feature flag introduced by
this change: com.delphix:spacemap_histogram
In addition to the above, this patch allows the space_map block size to
be increase. Currently the block size is set to be 4K in size, which has
certain implications including the following:
* 4K sector devices will not see any compression benefit
* large space_maps require more metadata on-disk
* large space_maps require more time to load (typically random reads)
Now the space_map block size can adjust as needed up to the maximum size
set via the space_map_max_blksz variable.
A bug was fixed which resulted in potentially leaking an object when
removing a mirrored log device. The previous logic for vdev_remove() did
not deal with removing top-level vdevs that are interior vdevs (i.e.
mirror) correctly. The problem would occur when removing a mirrored log
device, and result in the DTL space map object being leaked; because
top-level vdevs don't have DTL space map objects associated with them.
References:
https://www.illumos.org/issues/4101https://www.illumos.org/issues/4102https://www.illumos.org/issues/4103https://www.illumos.org/issues/4105https://www.illumos.org/issues/4106https://github.com/illumos/illumos-gate/commit/0713e23
Porting notes:
A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also,
the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary.
Ported-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#2488
We should not override the default memory type of the kmem cache. This
was done previously to force certain objects which were slightly over
object size limit cut off in to KMC_KMEM caches for better performance.
The zfsonlinux/spl#356 patch slightly increases the default cut off
from 511 bytes 1024 bytes for x86_64. This means there is long longer
a need to override the default for the caches. And since the default
values are now being used the new spl_kmem_cache_slab_limit and
spl_kmem_cache_kmem_limit tunables will apply to all kmem caches.
The following is a list of caches that will be impacted:
| object size | forced type | default type
----------------- | ------------- | ------------- | --------------
dnode_t | 936 bytes | KMC_KMEM | KMC_KMEM
zio_cache | 1104 bytes | *KMC_KMEM | *KMC_VMEM
zio_link_cache | 48 bytes | KMC_KMEM | KMC_KMEM
zio_vdev_cache | 131088 bytes | KMC_VMEM | KMC_VMEM
zio_buf_512 | 512 bytes | KMC_KMEM | KMC_KMEM
zio_data_buf_512 | 512 bytes | KMC_KMEM | KMC_KMEM
zio_buf_1024 | 1024 bytes | KMC_KMEM | KMC_KMEM
zio_data_buf_1024 | 1024 bytes | +KMC_VMEM | +KMC_KMEM
* Cache memory type will change from KMC_KMEM to KMC_VMEM.
+ Cache memory type will change from KMC_VMEM to KMC_KMEM.
This patch removes another slight point of divergence between ZoL
and Illumos.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Closes#2337
4045 zfs write throttle & i/o scheduler performance work
1. The ZFS i/o scheduler (vdev_queue.c) now divides i/os into 5 classes: sync
read, sync write, async read, async write, and scrub/resilver. The scheduler
issues a number of concurrent i/os from each class to the device. Once a class
has been selected, an i/o is selected from this class using either an elevator
algorithem (async, scrub classes) or FIFO (sync classes). The number of
concurrent async write i/os is tuned dynamically based on i/o load, to achieve
good sync i/o latency when there is not a high load of writes, and good write
throughput when there is. See the block comment in vdev_queue.c (reproduced
below) for more details.
2. The write throttle (dsl_pool_tempreserve_space() and
txg_constrain_throughput()) is rewritten to produce much more consistent delays
when under constant load. The new write throttle is based on the amount of
dirty data, rather than guesses about future performance of the system. When
there is a lot of dirty data, each transaction (e.g. write() syscall) will be
delayed by the same small amount. This eliminates the "brick wall of wait"
that the old write throttle could hit, causing all transactions to wait several
seconds until the next txg opens. One of the keys to the new write throttle is
decrementing the amount of dirty data as i/o completes, rather than at the end
of spa_sync(). Note that the write throttle is only applied once the i/o
scheduler is issuing the maximum number of outstanding async writes. See the
block comments in dsl_pool.c and above dmu_tx_delay() (reproduced below) for
more details.
This diff has several other effects, including:
* the commonly-tuned global variable zfs_vdev_max_pending has been removed;
use per-class zfs_vdev_*_max_active values or zfs_vdev_max_active instead.
* the size of each txg (meaning the amount of dirty data written, and thus the
time it takes to write out) is now controlled differently. There is no longer
an explicit time goal; the primary determinant is amount of dirty data.
Systems that are under light or medium load will now often see that a txg is
always syncing, but the impact to performance (e.g. read latency) is minimal.
Tune zfs_dirty_data_max and zfs_dirty_data_sync to control this.
* zio_taskq_batch_pct = 75 -- Only use 75% of all CPUs for compression,
checksum, etc. This improves latency by not allowing these CPU-intensive tasks
to consume all CPU (on machines with at least 4 CPU's; the percentage is
rounded up).
--matt
APPENDIX: problems with the current i/o scheduler
The current ZFS i/o scheduler (vdev_queue.c) is deadline based. The problem
with this is that if there are always i/os pending, then certain classes of
i/os can see very long delays.
For example, if there are always synchronous reads outstanding, then no async
writes will be serviced until they become "past due". One symptom of this
situation is that each pass of the txg sync takes at least several seconds
(typically 3 seconds).
If many i/os become "past due" (their deadline is in the past), then we must
service all of these overdue i/os before any new i/os. This happens when we
enqueue a batch of async writes for the txg sync, with deadlines 2.5 seconds in
the future. If we can't complete all the i/os in 2.5 seconds (e.g. because
there were always reads pending), then these i/os will become past due. Now we
must service all the "async" writes (which could be hundreds of megabytes)
before we service any reads, introducing considerable latency to synchronous
i/os (reads or ZIL writes).
Notes on porting to ZFS on Linux:
- zio_t gained new members io_physdone and io_phys_children. Because
object caches in the Linux port call the constructor only once at
allocation time, objects may contain residual data when retrieved
from the cache. Therefore zio_create() was updated to zero out the two
new fields.
- vdev_mirror_pending() relied on the depth of the per-vdev pending queue
(vq->vq_pending_tree) to select the least-busy leaf vdev to read from.
This tree has been replaced by vq->vq_active_tree which is now used
for the same purpose.
- vdev_queue_init() used the value of zfs_vdev_max_pending to determine
the number of vdev I/O buffers to pre-allocate. That global no longer
exists, so we instead use the sum of the *_max_active values for each of
the five I/O classes described above.
- The Illumos implementation of dmu_tx_delay() delays a transaction by
sleeping in condition variable embedded in the thread
(curthread->t_delay_cv). We do not have an equivalent CV to use in
Linux, so this change replaced the delay logic with a wrapper called
zfs_sleep_until(). This wrapper could be adopted upstream and in other
downstream ports to abstract away operating system-specific delay logic.
- These tunables are added as module parameters, and descriptions added
to the zfs-module-parameters.5 man page.
spa_asize_inflation
zfs_deadman_synctime_ms
zfs_vdev_max_active
zfs_vdev_async_write_active_min_dirty_percent
zfs_vdev_async_write_active_max_dirty_percent
zfs_vdev_async_read_max_active
zfs_vdev_async_read_min_active
zfs_vdev_async_write_max_active
zfs_vdev_async_write_min_active
zfs_vdev_scrub_max_active
zfs_vdev_scrub_min_active
zfs_vdev_sync_read_max_active
zfs_vdev_sync_read_min_active
zfs_vdev_sync_write_max_active
zfs_vdev_sync_write_min_active
zfs_dirty_data_max_percent
zfs_delay_min_dirty_percent
zfs_dirty_data_max_max_percent
zfs_dirty_data_max
zfs_dirty_data_max_max
zfs_dirty_data_sync
zfs_delay_scale
The latter four have type unsigned long, whereas they are uint64_t in
Illumos. This accommodates Linux's module_param() supported types, but
means they may overflow on 32-bit architectures.
The values zfs_dirty_data_max and zfs_dirty_data_max_max are the most
likely to overflow on 32-bit systems, since they express physical RAM
sizes in bytes. In fact, Illumos initializes zfs_dirty_data_max_max to
2^32 which does overflow. To resolve that, this port instead initializes
it in arc_init() to 25% of physical RAM, and adds the tunable
zfs_dirty_data_max_max_percent to override that percentage. While this
solution doesn't completely avoid the overflow issue, it should be a
reasonable default for most systems, and the minority of affected
systems can work around the issue by overriding the defaults.
- Fixed reversed logic in comment above zfs_delay_scale declaration.
- Clarified comments in vdev_queue.c regarding when per-queue minimums take
effect.
- Replaced dmu_tx_write_limit in the dmu_tx kstat file
with dmu_tx_dirty_delay and dmu_tx_dirty_over_max. The first counts
how many times a transaction has been delayed because the pool dirty
data has exceeded zfs_delay_min_dirty_percent. The latter counts how
many times the pool dirty data has exceeded zfs_dirty_data_max (which
we expect to never happen).
- The original patch would have regressed the bug fixed in
zfsonlinux/zfs@c418410, which prevented users from setting the
zfs_vdev_aggregation_limit tuning larger than SPA_MAXBLOCKSIZE.
A similar fix is added to vdev_queue_aggregate().
- In vdev_queue_io_to_issue(), dynamically allocate 'zio_t search' on the
heap instead of the stack. In Linux we can't afford such large
structures on the stack.
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Ned Bass <bass6@llnl.gov>
Reviewed by: Brendan Gregg <brendan.gregg@joyent.com>
Approved by: Robert Mustacchi <rm@joyent.com>
References:
http://www.illumos.org/issues/4045illumos/illumos-gate@69962b5647
Ported-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1913
4047 panic from dbuf_free_range() from dmu_free_object() while
doing zfs receive
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Dan McDonald <danmcd@nexenta.com>
References:
https://www.illumos.org/issues/4047illumos/illumos-gate@713d6c2088
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1775
Porting notes:
1. The exported symbol dmu_free_object() was renamed to
dmu_free_long_object() in Illumos.
3742 zfs comments need cleaner, more consistent style
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Eric Schrock <eric.schrock@delphix.com>
Approved by: Christopher Siden <christopher.siden@delphix.com>
References:
https://www.illumos.org/issues/3742illumos/illumos-gate@f717074149
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1775
Porting notes:
1. The change to zfs_vfsops.c was dropped because it involves
zfs_mount_label_policy, which does not exist in the Linux port.
3598 want to dtrace when errors are generated in zfs
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>
References:
https://www.illumos.org/issues/3598illumos/illumos-gate@be6fd75a69
Ported-by: Richard Yao <ryao@gentoo.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1775
Porting notes:
1. include/sys/zfs_context.h has been modified to render some new
macros inert until dtrace is available on Linux.
2. Linux-specific changes have been adapted to use SET_ERROR().
3. I'm NOT happy about this change. It does nothing but ugly
up the code under Linux. Unfortunately we need to take it to
avoid more merge conflicts in the future. -Brian
3006 VERIFY[S,U,P] and ASSERT[S,U,P] frequently check if first
argument is zero
Reviewed by Matt Ahrens <matthew.ahrens@delphix.com>
Reviewed by George Wilson <george.wilson@delphix.com>
Approved by Eric Schrock <eric.schrock@delphix.com>
References:
illumos/illumos-gate@fb09f5aad4https://illumos.org/issues/3006
Requires:
zfsonlinux/spl@1c6d149feb
Ported-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#1509
2619 asynchronous destruction of ZFS file systems
2747 SPA versioning with zfs feature flags
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <gwilson@delphix.com>
Reviewed by: Richard Lowe <richlowe@richlowe.net>
Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com>
Approved by: Eric Schrock <Eric.Schrock@delphix.com>
References:
illumos/illumos-gate@53089ab7c8illumos/illumos-gate@ad135b5d64
illumos changeset: 13700:2889e2596bd6
https://www.illumos.org/issues/2619https://www.illumos.org/issues/2747
NOTE: The grub specific changes were not ported. This change
must be made to the Linux grub packages.
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
Differences between how paging is done on Solaris and Linux can cause
deadlocks if KM_SLEEP is used in any the following contexts.
* The txg_sync thread
* The zvol write/discard threads
* The zpl_putpage() VFS callback
This is because KM_SLEEP will allow for direct reclaim which may result
in the VM calling back in to the filesystem or block layer to write out
pages. If a lock is held over this operation the potential exists to
deadlock the system. To ensure forward progress all memory allocations
in these contexts must us KM_PUSHPAGE which disables performing any I/O
to accomplish the memory allocation.
Previously, this behavior was acheived by setting PF_MEMALLOC on the
thread. However, that resulted in unexpected side effects such as the
exhaustion of pages in ZONE_DMA. This approach touchs more of the zfs
code, but it is more consistent with the right way to handle these cases
under Linux.
This is patch lays the ground work for being able to safely revert the
following commits which used PF_MEMALLOC:
21ade34 Disable direct reclaim for z_wr_* threads
cfc9a5c Fix zpl_writepage() deadlock
eec8164 Fix ASSERTION(!dsl_pool_sync_context(tx->tx_pool))
Signed-off-by: Richard Yao <ryao@cs.stonybrook.edu>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #726
Profiling the system during meta data intensive workloads such
as creating/removing millions of files, revealed that the system
was cpu bound. A large fraction of that cpu time was being spent
waiting on the virtual address space spin lock.
It turns out this was caused by certain heavily used kmem_caches
being backed by virtual memory. By default a kmem_cache will
dynamically determine the type of memory used based on the object
size. For large objects virtual memory is usually preferable
and for small object physical memory is a better choice. See
the spl_slab_alloc() function for a longer discussion on this.
However, there is a certain amount of gray area when defining a
'large' object. For the following caches it turns out they were
just over the line:
* dnode_cache
* zio_cache
* zio_link_cache
* zio_buf_512_cache
* zfs_data_buf_512_cache
Now because we know there will be a lot of churn in these caches,
and because we know the slabs will still be reasonably sized.
We can safely request with the KMC_KMEM flag that the caches be
backed with physical memory addresses. This entirely avoids the
need to serialize on the virtual address space lock.
As a bonus this also reduces our vmalloc usage which will be good
for 32-bit kernels which have a very small virtual address space.
It will also probably be good for interactive performance since
unrelated processes could also block of this same global lock.
Finally, we may see less cpu time being burned in the arc_reclaim
and txg_sync_threads.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #258
In my machine, dnode_hold_impl() allocates 9992 bytes in DEBUG mode and it
causes a large stream of stack traces in the logs. Instead, use KM_NODEBUG
to quiet down this known large alloc.
Signed-off-by: Ricardo M. Correia <ricardo.correia@oracle.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
The dnode_move() functionality is only used in the kernel build.
As such we should be careful to wrap all of the related code
with '#ifdef _KERNEL' to avoid gcc warnings about unused code.
