| Age | Commit message (Collapse) | Author |
|---|
|
Since sector_ptr structure is now only containing a single paddr, there
is no need to use that structure.
Instead use phys_addr_t array for bio and stripe pointers.
This means several helpers are also needed to accept a paddr instead of
a sector_ptr pointer.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The uptodate boolean member can be extracted into a bitmap, which will
save us some space (1 bit in a byte vs 8 bits in a byte).
Furthermore we do not need to record the uptodate bitmap for bio
sectors, as if bio_sectors[].paddr is valid it means there is a bio and
will be uptodate.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
We can use paddr -1 as an indicator for unset/uninitialized paddr.
We can not use 0 paddr, unlike virtual address 0 which is never mapped
thus will always trigger a page fault, physical address 0 may be a valid
page.
So here we follow swiotlb to use (paddr)-1 as a special indicator for
invalid/unset physical address.
Even if the PFN may still be valid, our usage of the physical address
should always be aligned to fs block size (or page size for bs > ps
cases), thus such -1 paddr should never be a valid one.
With this special -1 paddr, we can get rid of has_paddr member and save
1 byte for sector_ptr structure.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
In btrfs_compr_pool_scan(), use LIST_HEAD() to declare and initialize
the 'remove' list_head in one step instead of using INIT_LIST_HEAD()
separately.
Signed-off-by: Baolin Liu <liubaolin@kylinos.cn>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The function scrub_raid56_parity_stripe() is handling the parity stripe
by the following steps:
- Scrub each data stripes
And make sure everything is fine in each data stripe
- Cache the data stripe into the raid bio
- Use the cached raid bio to scrub the target parity stripe
Extract the last two steps into a new helper,
scrub_raid56_cached_parity(), as a cleanup and make the error handling
more straightforward.
With the following minor cleanups:
- Use on-stack bio structure
The bio is always empty thus we do not need any bio vector nor the
block device. Thus there is no need to allocate a bio, the on-stack
one is more than enough to cut it.
- Remove the unnecessary btrfs_put_bioc() call if btrfs_map_block()
failed
If btrfs_map_block() is failed, @bioc_ret will not be touched thus
there is no need to call btrfs_put_bioc() in this case.
- Use a proper out: tag to do the cleanup
Now the error cleanup is much shorter and simpler, just
btrfs_bio_counter_dec() and bio_uninit().
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
scrub_raid56_parity_stripe()
Unlike queue_scrub_stripe() which uses the global sctx->extent_path and
sctx->csum_path which are always released at the end of scrub_stripe(),
scrub_raid56_parity_stripe() uses local extent_path and csum_path, as
that function is going to handle the full stripe, whose bytenr may be
smaller than the bytenr in the global sctx paths.
However the cleanup of local extent/csum paths is only happening after
we have successfully submitted an rbio.
There are several error routes that we didn't release those two paths:
- scrub_find_fill_first_stripe() errored out at csum tree search
In that case extent_path is still valid, and that function itself will
not release the extent_path passed in.
And the function returns directly without releasing both paths.
- The full stripe is empty
- Some blocks failed to be recovered
- btrfs_map_block() failed
- raid56_parity_alloc_scrub_rbio() failed
The function returns directly without releasing both paths.
Fix it by covering btrfs_release_path() calls inside the out: tag.
This is just a hot fix, in the long run we will go scoped based auto
freeing for both local paths.
Fixes: 1dc4888e725d ("btrfs: scrub: avoid unnecessary extent tree search preparing stripes")
Fixes: 3c771c194402 ("btrfs: scrub: avoid unnecessary csum tree search preparing stripes")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
[BUG]
There is a report that memory allocation failed for btrfs_bio::csum
during a large read:
b2sum: page allocation failure: order:4, mode:0x40c40(GFP_NOFS|__GFP_COMP), nodemask=(null),cpuset=/,mems_allowed=0
CPU: 0 UID: 0 PID: 416120 Comm: b2sum Tainted: G W 6.17.0 #1 NONE
Tainted: [W]=WARN
Hardware name: Raspberry Pi 4 Model B Rev 1.5 (DT)
Call trace:
show_stack+0x18/0x30 (C)
dump_stack_lvl+0x5c/0x7c
dump_stack+0x18/0x24
warn_alloc+0xec/0x184
__alloc_pages_slowpath.constprop.0+0x21c/0x730
__alloc_frozen_pages_noprof+0x230/0x260
___kmalloc_large_node+0xd4/0xf0
__kmalloc_noprof+0x1c8/0x260
btrfs_lookup_bio_sums+0x214/0x278
btrfs_submit_chunk+0xf0/0x3c0
btrfs_submit_bbio+0x2c/0x4c
submit_one_bio+0x50/0xac
submit_extent_folio+0x13c/0x340
btrfs_do_readpage+0x4b0/0x7a0
btrfs_readahead+0x184/0x254
read_pages+0x58/0x260
page_cache_ra_unbounded+0x170/0x24c
page_cache_ra_order+0x360/0x3bc
page_cache_async_ra+0x1a4/0x1d4
filemap_readahead.isra.0+0x44/0x74
filemap_get_pages+0x2b4/0x3b4
filemap_read+0xc4/0x3bc
btrfs_file_read_iter+0x70/0x7c
vfs_read+0x1ec/0x2c0
ksys_read+0x4c/0xe0
__arm64_sys_read+0x18/0x24
el0_svc_common.constprop.0+0x5c/0x130
do_el0_svc+0x1c/0x30
el0_svc+0x30/0xa0
el0t_64_sync_handler+0xa0/0xe4
el0t_64_sync+0x198/0x19c
[CAUSE]
Btrfs needs to allocate memory for btrfs_bio::csum for large reads, so
that we can later verify the contents of the read.
However nowadays a read bio can easily go beyond BIO_MAX_VECS *
PAGE_SIZE (which is 1M for 4K page sizes), due to the multi-page bvec
that one bvec can have more than one pages, as long as the pages are
physically adjacent.
This will become more common when the large folio support is moved out
of experimental features.
In the above case, a read larger than 4MiB with SHA256 checksum (32
bytes for each 4K block) will be able to trigger a order 4 allocation.
The order 4 is larger than PAGE_ALLOC_COSTLY_ORDER (3), thus without
extra flags such allocation will not retry.
And if the system has very small amount of memory (e.g. RPI4 with low
memory spec) or VMs with small vRAM, or the memory is heavily
fragmented, such allocation will fail and cause the above warning.
[FIX]
Although btrfs is handling the memory allocation failure correctly, we
do not really need the physically contiguous memory just to restore
our checksum.
In fact btrfs_csum_one_bio() is already using kvzalloc() to reduce the
memory pressure.
So follow the step to use kvcalloc() for btrfs_bio::csum.
Reported-by: Calvin Owens <calvin@wbinvd.org>
Link: https://lore.kernel.org/linux-btrfs/20251105180054.511528-1-calvin@wbinvd.org/
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The current definition of ASSERT(cond) as (void)(cond) is redundant,
since these checks have no side effects and don't affect code logic.
However, some checks contain READ_ONCE() or other compiler-unfriendly
constructs. For example, ASSERT(list_empty) in btrfs_add_dealloc_inode()
was compiled to a redundant mov instruction due to this issue.
Define ASSERT as BUILD_BUG_ON_INVALID for !CONFIG_BTRFS_ASSERT builds
which uses sizeof(cond) trick. Also mark full_page_sectors_uptodate()
as __maybe_unused to suppress "unneeded declaration" warning (it's
needed in compile time)
Signed-off-by: Gladyshev Ilya <foxido@foxido.dev>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
[ENHANCEMENT]
Btrfs currently calculates data checksums then submits the bio.
But after commit 968f19c5b1b7 ("btrfs: always fallback to buffered write
if the inode requires checksum"), any writes with data checksum will
fallback to buffered IO, meaning the content will not change during
writeback.
This means we're safe to calculate the data checksum and submit the bio
in parallel, and only need the following new behavior:
- Wait the csum generation to finish before calling btrfs_bio::end_io()
Or this can lead to use-after-free for the csum generation worker.
- Save the current bi_iter for csum_one_bio()
As the submission part can advance btrfs_bio::bio.bi_iter, if not
saved csum_one_bio() may got an empty bi_iter and do not generate any
checksum.
Unfortunately this means we have to increase the size of btrfs_bio for
16 bytes, but this is still acceptable.
As usual, such new feature is hidden behind the experimental flag.
[THEORETIC ANALYZE]
Consider the following theoretic hardware performance, which should be
more or less close to modern mainstream hardware:
Memory bandwidth: 50GiB/s
CRC32C bandwidth: 45GiB/s
SSD bandwidth: 8GiB/s
Then write bandwidth with data checksum before the patch is:
1 / ( 1 / 50 + 1 / 45 + 1 / 8) = 5.98 GiB/s
After the patch, the bandwidth is:
1 / ( 1 / 50 + max( 1 / 45 + 1 / 8)) = 6.90 GiB/s
The difference is 15.32% improvement.
[REAL WORLD BENCHMARK]
I'm using a Zen5 (HX 370) as the host, the VM has 4GiB memory, 10 vCPUs, the
storage is backed by a PCIe gen3 x4 NVMe.
The test is a direct IO write, with 1MiB block size, write 7GiB data
into a btrfs mount with data checksum. Thus the direct write will
fallback to buffered one:
Vanilla Datasum: 1619.97 GiB/s
Patched Datasum: 1792.26 GiB/s
Diff +10.6 %
In my case, the bottleneck is the storage, thus the improvement is not
reaching the theoretic one, but still some observable improvement.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
We used IRQ version of spinlock for ordered_tree_lock, as
btrfs_finish_ordered_extent() can be called in end_bbio_data_write()
which was in IRQ context.
However since we're moving all the btrfs_bio::end_io() calls into task
context, there is no more need to support IRQ context thus we can relax
to regular spin_lock()/spin_unlock() for btrfs_inode::ordered_tree_lock.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The reason why end_bbio_compressed_write() queues a work into
compressed_write_workers wq is for end_compressed_writeback() call, as
it will grab all the involved folios and clear the writeback flags,
which may sleep.
However now we always run btrfs_bio::end_io() in task context, there is
no need to queue the work anymore.
Just remove btrfs_fs_info::compressed_write_workers and
compressed_bio::write_end_work.
There is a comment about the works queued into
compressed_write_workers, now change to flush endio wq instead, which is
responsible to handle all data endio functions.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
[BACKGROUND]
Btrfs has a lot of different bi_end_io functions, to handle different
raid profiles. But they introduced a lot of different contexts for
btrfs_bio::end_io() calls:
- Simple read bios
Run in task context, backed by either endio_meta_workers or
endio_workers.
- Simple write bios
Run in IRQ context.
- RAID56 write or rebuild bios
Run in task context, backed by rmw_workers.
- Mirrored write bios
Run in irq context.
This is inconsistent, and contributes to the number of workqueues used
in btrfs.
[ENHANCEMENT]
Make all the above bios call their btrfs_bio::end_io() in task context,
backed by either endio_meta_workers for metadata, or endio_workers for
data.
For simple write bios, merge the handling into simple_end_io_work().
For mirrored write bios, it will be a little more complex, since both
the original or the cloned bios can run the final btrfs_bio::end_io().
Here we make sure the cloned bios are using btrfs_bioset, to reuse the
end_io_work, and run both original and cloned work inside the workqueue.
Add extra ASSERT()s to make sure btrfs_bio_end_io() is running in task
context.
This not only unifies the context for btrfs_bio::end_io() functions, but
also opens a new door for further btrfs_bio::end_io() related cleanups.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Currently there is only one caller which doesn't populate
btrfs_bio::inode, and that's scrub.
The idea is scrub doesn't want any automatic csum verification nor
read-repair, as everything will be handled by scrub itself.
However that behavior is really no different than metadata inode, thus
we can reuse btree_inode as btrfs_bio::inode for scrub.
The only exception is in btrfs_submit_chunk() where if a bbio is from
scrub or data reloc inode, we set rst_search_commit_root to true.
This means we still need a way to distinguish scrub from metadata, but
that can be done by a new flag inside btrfs_bio.
Now btrfs_bio::inode is a mandatory parameter, we can extract fs_info
from that inode thus can remove btrfs_bio::fs_info to save 8 bytes from
btrfs_bio structure.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
[BUG]
When I tried to remove btrfs_bio::fs_info and use btrfs_bio::inode to
grab the fs_info, the header "btrfs_inode.h" is needed to access the
full btrfs_inode structure.
Then btrfs will fail to compile.
[CAUSE]
There is a recursive including chain:
"bio.h" -> "btrfs_inode.h" -> "extent_map.h" -> "compression.h" ->
"bio.h"
That recursive including is causing problems for btrfs.
[ENHANCEMENT]
To reduce the risk of recursive including:
- Remove unnecessary local includes from btrfs headers
Either the included header is pulled in by other headers, or is
completely unnecessary.
- Remove btrfs local includes if the header only requires a pointer
In that case let the implementing C file to pull the required header.
This is especially important for headers like "btrfs_inode.h" which
pulls in a lot of other btrfs headers, thus it's a mine field of
recursive including.
- Remove unnecessary temporary structure definition
Either if we have included the header defining the structure, or
completely unused.
Now including "btrfs_inode.h" inside "bio.h" is completely fine,
although "btrfs_inode.h" still includes "extent_map.h", but that header
only includes "fs.h", no more chain back to "bio.h".
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
It's impossible to have a btrfs bio with more than BIO_MAX_VECS vectors
anyway. And there is only one location utilizing that macro, just
replace it with BIO_MAX_VECS. Both have the same value.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
const_ilog2() was a workaround of some sparse issue, which has never
appeared in the C functions. Replace it with ilog2().
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Provide statistics for zoned filesystems. These statistics include, the
number of active block-groups, how many of them are reclaimable or unused,
if the filesystem needs to be reclaimed, the currently assigned relocation
and treelog block-groups if they're present and a list of active zones.
Example:
active block-groups: 4
reclaimable: 0
unused: 2
need reclaim: false
data relocation block-group: 4294967296
active zones:
start: 1610612736, wp: 344064 used: 16384, reserved: 0, unusable: 327680
start: 1879048192, wp: 34963456 used: 131072, reserved: 0, unusable: 34832384
start: 4026531840, wp: 0 used: 0, reserved: 0, unusable: 0
start: 4294967296, wp: 0 used: 0, reserved: 0, unusable: 0
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The prealloc variable in these functions is always initialized to
NULL. Whenever we allocate memory for it, if it fails then NULL is
preserved, otherwise we delegate the ownership of the pointer to
add_qgroup_rb() and set it right after to NULL.
Since in any case the pointer ends up being NULL at the end of its
usage, we can safely remove calls to kfree() for it, while adding an
ASSERT as an extra check.
Signed-off-by: Miquel Sabaté Solà <mssola@mssola.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Apply the AUTO_KFREE and AUTO_KVFREE macros wherever it makes
sense. Since this macro is expected to improve code readability, it has
been avoided in places where the lifetime of objects wasn't easy to
follow and a cleanup attribute would've made things worse; or when the
cleanup section of a function involved many other things and thus there
was no readability impact anyways. This change has also not been applied
in extremely short functions where readability was clearly not an issue.
Signed-off-by: Miquel Sabaté Solà <mssola@mssola.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
These are two simple macros which ensure that a pointer is initialized
to NULL and with the proper cleanup attribute for it.
Signed-off-by: Miquel Sabaté Solà <mssola@mssola.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The free_ipath() function was being used as a cleanup function
everywhere. Declare it via DEFINE_FREE() so we can use this function
with the __free() helper.
The name has also been adjusted so it's closer to the type's name.
Signed-off-by: Miquel Sabaté Solà <mssola@mssola.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Unlike relocation, scrub never checks pending signals, and even for
relocation is only explicitly checking for fatal signal (SIGKILL), not
for regular ones.
Thankfully relocation can still be interrupted by regular signals by
the usage of wait_on_bit(), which is called with TASK_INTERRUPTIBLE.
Do the same for scrub/dev-replace, so that regular signals can also
cancel the scrub/replace run, and more importantly handle v2 cgroup
freezing which is based on signal handling code inside the kernel, and
freezing() function will not return true for v2 cgroup freezing.
This will address the problem that systemd slice freezing will timeout
on long running scrub/dev-replace.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
It's a known bug that btrfs scrub/dev-replace can prevent the system
from suspending.
There are at least two factors involved:
- Holding super_block::s_writers for the whole scrub/dev-replace duration
We hold that percpu rw semaphore through mnt_want_write_file() for the
whole scrub/dev-replace duration.
That will prevent the fs being frozen, which can be initiated by
either the user (e.g. fsfreeze) or power management suspend/hibernate.
- Stuck in the kernel space for a long time
During suspend all user processes (and some kernel threads) will
be frozen.
But if a user space progress has fallen into kernel (scrub ioctl) and
do not return for a long time, it will make process freezing time out.
Unfortunately scrub/dev-replace is a long running ioctl, and it will
prevent the btrfs process from returning to the user space, thus make PM
suspend/hibernate time out.
Address them in one go:
- Introduce a new helper should_cancel_scrub()
Which includes the existing cancel request and new fs/process freezing
checks.
Here we have to check both fs and process freezing for PM
suspend/hibernate.
PM can be configured to freeze filesystems before processes.
(The current default is not to freeze filesystems, but planned to
freeze the filesystems as the new default.)
Checking only fs freezing will fail PM without fs freezing, as the
process freezing will time out.
Checking only process freezing will fail PM with fs freezing since the
fs freezing happens before process freezing.
And the return value will indicate the reason, -ECANCLED for the
explicitly canceled runs, and -EINTR for fs freeze or PM reasons.
- Cancel the run if should_cancel_scrub() is true
Unfortunately canceling is the only feasible solution here, pausing is
not possible as we will still stay in the kernel space thus will still
prevent the process from being frozen.
This will cause a user impacting behavior change:
Dev-replace can be interrupted by PM, and there is no way to resume
but start from the beginning again.
This means dev-replace may fail on newer kernels, and end users will
need extra steps like using systemd-inhibit to prevent
suspend/hibernate, to get back the old uninterrupted behavior.
This behavior change will need extra documentation updates and
communication with projects involving scrub/dev-replace including
btrfs-progs.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Link: https://lore.kernel.org/linux-btrfs/d93b2a2d-6ad9-4c49-809f-11d769a6f30a@app.fastmail.com/
Reported-by: Chris Murphy <lists@colorremedies.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
For raid56, data and parity stripes are handled differently.
For data stripes they are handled just like regular RAID1/RAID10 stripes,
going through the regular scrub_simple_mirror().
But for parity stripes we have to read out all involved data stripes and
do any needed verification and repair, then scrub the parity stripe.
This process will take a much longer time than a regular stripe, but
unlike scrub_simple_mirror(), we do not check if we should cancel/pause
or the block group is already removed.
Aligned the behavior of scrub_raid56_parity_stripe() to
scrub_simple_mirror(), by adding:
- Cancel check
- Pause check
- Removed block group check
Since those checks are the same from the scrub_simple_mirror(), also
update the comments of scrub_simple_mirror() by:
- Remove too obvious comments
We do not need extra comments on what we're checking, it's really too
obvious.
- Remove a stale comment about pausing
Now the scrub is always queuing all involved stripes, and submit them
in one go, there is no more submission part during pausing.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
It's not expected to have the fs in an aborted state, so surround the
abortion checks with unlikely to make it clear it's unexpected and to
hint the compiler to generate better code.
Also at maybe_fail_all_tickets() untangle all repeated checks for the
abortion into a single if-then-else. This makes things more readable
and makes the compiler generate less code. On x86_64 with gcc 14.2.0-19
from Debian I got the following object size differences.
Before this change:
$ size fs/btrfs/btrfs.ko
text data bss dec hex filename
2021606 179704 25088 2226398 21f8de fs/btrfs/btrfs.ko
After this change:
$ size fs/btrfs/btrfs.ko
text data bss dec hex filename
2021458 179704 25088 2226250 21f84a fs/btrfs/btrfs.ko
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
When checking if a ticket was served, we take the space_info's spinlock.
If the ticket was served (its ->bytes is 0) or had an error (its ->error
it not 0) then we just unlock the space_info and return.
This however causes contention on the space_info's spinlock, which is
heavily used (space reservation, space flushing, allocating and
deallocating an extent from a block group (btrfs_update_block_group()),
etc).
Instead of using the space_info's spinlock to check if a ticket was
served, use a per ticket spinlock which isn't used by anyone other than
the task that created the ticket (stack allocated) and the task that
serves the ticket (a reclaim task or any task deallocating space that
ends up at btrfs_try_granting_tickets()).
After applying this patch and all previous patches from the same patchset
(many attempt to reduce space_info critical sections), lockstat showed
some improvements for a fs_mark test regarding the space_info's spinlock
'lock'. The lockstat results:
Before patchset:
con-bounces: 13733858
contentions: 15902322
waittime-total: 264902529.72
acq-bounces: 28161791
acquisitions: 38679282
After patchset:
con-bounces: 12032220
contentions: 13598034
waittime-total: 221806127.28
acq-bounces: 24717947
acquisitions: 34103281
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Instead of repeating the wakeup and setup of the ->bytes or ->error field,
move those steps to remove_ticket() to avoid duplication. This is also
needed for the next patch in the series, so that we avoid duplicating more
logic.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Surround the intentional empty list check with the data_race() annotation
so that tools like KCSAN don't report a data race. The race is intentional
as it's harmless and we want to avoid lock contention of the space_info
since its lock is heavily used (space reservation, space flushing, extent
allocation and deallocation, etc).
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
There's no need to have an 'out' label and jump there in case we can
not find a block group. We can simply return directly since there are no
resources to release, removing the need for the label and the 'ret'
variable.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
There's no need to update the bytes_pinned, bytes_readonly and
max_extent_size fields of the space_info while inside the critical section
delimited by the block group's lock. So move that out of the block group's
critical section, but sill inside the space_info's critical section.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
It's used as a boolean, so convert it from int type to bool type.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
All callers pass a value of 1 (true) to it, so remove it.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Instead of dereferencing the block group multiple times to access its
space_info, use a local variable to shorten the code horizontal wise and
make it easier to read. Also, while at it, also rename the block group
argument from 'cache' to 'bg', as the cache name is confusing and it's
from the old days where the block group structure was named as
'btrfs_block_group_cache'.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
There's no need to update the bytes_reserved and bytes_may_use fields of
the space_info while holding the block group's spinlock. We are only
making the critical section longer than necessary. So move the space_info
updates outside of the block group's critical section.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
There's no need to update the bytes_reserved and bytes_readonly fields of
the space_info while holding the block group's spinlock. We are only
making the critical section longer than necessary. So move the space_info
updates outside of the block group's critical section.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
We are doing some things inside the block group's critical section that
are relevant only to the space_info: updating the space_info counters
bytes_reserved and bytes_may_use as well as trying to grant tickets
(calling btrfs_try_granting_tickets()), and this later can take some
time. So move all those updates to outside the block group's critical
section and still inside the space_info's critical section. Like this
we keep the block group's critical section only for block group updates
and can help reduce contention on a block group's lock.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
There's no need to update the space_info fields (bytes_reserved,
max_extent_size, bytes_readonly, bytes_zone_unusable) while holding the
block group's spinlock. So move those updates to happen after we unlock
the block group (and while holding the space_info locked of course), so
that all we do under the block group's critical section is to update the
block group itself.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
There's no need to update local variables while holding the space_info's
spinlock, since the update isn't using anything from the space_info. So
move these updates outside the critical section to shorten it.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The use of a double underscore prefix is discouraged and we have no
justification at all for it all since there's no reserved_bytes() counter
part. So remove the prefix.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
In steal_from_global_rsv() there's no need to process the ticket inside
the critical section of the global reserve. Move the ticket processing to
happen after the critical section. This helps reduce contention on the
global reserve's spinlock.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
We have a couple places that are assigning 0 and 1 to the full field of
the global reserve. This is harmless since 0 is converted to false and
1 converted to true, but for better readability, replace these with true
and false since the field is of type bool.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The caller is supposed to have locked the space_info, so assert that.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
priority_reclaim_metadata_space()
If the given ticket was already served (its ->bytes is 0), then we wasted
time calculating the metadata reclaim size. So calculate it only after we
checked the ticket was not yet served.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
We are doing a lot of small calculations and assignments while holding the
space_info's spinlock, which is a heavily used lock for space reservation
and flushing. There's no point in holding the lock for so long when all we
want is to call need_preemptive_reclaim() and get a consistent value for a
couple of counters from the space_info. Instead, grab the counters into
local variables, release the lock and then use the local variables.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
In btrfs_preempt_reclaim_metadata_space() there's no need to increment the
local variable that tracks the number of iterations of the while loop
while inside the critical section delimited by the space_info's spinlock.
That spinlock is heavily used by space reservation and flushing code, so
it's desirable to have its critical sections as short as possible.
So move the loop count incremented outside the critical section.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Instead of doing some calculations and then return false if it turns out
we have queued tickets, check first if we have tickets and return false
immediately if we have tickets, without wasting time on doing those
computations.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The function is simple enough to be inlined and in fact doing it even
reduces the object code. In x86_64 with gcc 14.2.0-19 from Debian the
results were the following:
Before this change
$ size fs/btrfs/btrfs.ko
text data bss dec hex filename
1919410 161703 15592 2096705 1ffe41 fs/btrfs/btrfs.ko
After this change
$ size fs/btrfs/btrfs.ko
text data bss dec hex filename
1918991 161675 15592 2096258 1ffc82 fs/btrfs/btrfs.ko
Also remove the ASSERT() that checks the space_info argument is not NULL,
as it's odd to be there since it can never be NULL and in case that ever
happens during development, a stack trace from a NULL pointer dereference
will be obvious. It was originally added when btrfs_space_info_used() was
introduced in commit 4136135b080f ("Btrfs: use helper to get used bytes
of space_info").
Also add a lockdep assertion to check the space_info's lock is being held
by the calling task.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
In __reserve_bytes() we have 3 repeated calls to btrfs_space_info_used(),
one early on as soon as take the space_info's spinlock, another one when
we call btrfs_can_overcommit(), which calls btrfs_space_info_used() again,
and a final one when we are reserving for a flush emergency.
During all these calls we are holding the space_info's spinlock, which is
heavily used by the space reservation and flushing code, so it's desirable
to make the critical sections as short as possible.
So make this more efficient by:
1) Instead of calling btrfs_can_overcommit() call the new variant
can_overcommit() which takes the space_info's used space as an argument
and pass the value we already computed and have in the 'used' variable;
2) Instead of calling btrfs_space_info_used() with its second argument as
false when we are doing a flush emergency, decrement the space_info's
bytes_may_use counter from the 'used' variable, as the difference
between passing true or false as the second argument to
btrfs_space_info_used() is whether or not to include the space_info's
bytes_may_use counter in the computation.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
In btrfs_try_granting_tickets(), we call btrfs_can_overcommit() and that
calls btrfs_space_info_used(). But we already keep track, in the 'used'
local variable, of the used space in the space_info, so we are just
repeating the same computation and doing an extra function call while we
are holding the space_info's spinlock, which is heavily used by the space
reservation and flushing code.
So add a local variant of btrfs_can_overcommit() that takes in the used
space as an argument and therefore does not call btrfs_space_info_used(),
and use it in btrfs_try_granting_tickets().
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
It's a boolean function, so switch its return type to bool.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
