| Message ID | 1274b3b2-d196-ee9e-2261-d2f5cb906477@cn.fujitsu.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 09/26/2016 10:10 PM, Qu Wenruo wrote: > > > At 09/26/2016 10:31 PM, Goldwyn Rodrigues wrote: >> >> >> On 09/25/2016 09:33 PM, Qu Wenruo wrote: >>> >>> >>> At 09/23/2016 09:43 PM, Goldwyn Rodrigues wrote: >>>> >>>> > [snipped] >>> >>> Sorry I still don't get the point. >>> Would you please give a call flow of the timing dirtying page and >>> calling btrfs_qgroup_reserve/free/release_data()? >>> >>> Like: >>> __btrfs_buffered_write() >>> |- btrfs_check_data_free_space() >>> | |- btrfs_qgroup_reserve_data() <- Mark QGROUP_RESERVED bit >>> |- btrfs_dirty_pages() <- Mark page dirty >>> >>> >>> [[Timing of btrfs_invalidatepage()]] >>> About your commit message "once while invalidating the page, and the >>> next time while free'ing delalloc." >>> "Free'ing delalloc" did you mean btrfs_qgroup_free_delayed_ref(). >>> >>> If so, it means one extent goes through full write back, and long before >>> calling btrfs_qgroup_free_delayed_ref(), it will call >>> btrfs_qgroup_release_data() to clear QGROUP_RESERVED. >>> >>> So the call will be: >>> __btrfs_buffered_write() >>> |- btrfs_check_data_free_space() >>> | |- btrfs_qgroup_reserve_data() <- Mark QGROUP_RESERVED bit >>> |- btrfs_dirty_pages() <- Mark page dirty >>> >>> <data write back happens> >>> run_delalloc_range() >>> |- cow_file_range() >>> |- extent_clear_unlock_delalloc() <- Clear page dirty >>> >>> <modifying metadata> >>> >>> btrfs_finish_ordered_io() >>> |- insert_reserved_file_extent() >>> |- btrfs_qgroup_release_data() <- Clear QGROUP_RESERVED bit >>> but not decrease reserved space >>> >>> <run delayed refs, normally happens in commit_trans> >>> run_one_delyaed_refs() >>> |- btrfs_qgroup_free_delayed_ref() <- Directly decrease reserved space >>> >>> >>> So the problem seems to be, btrfs_invalidatepage() is called after >>> run_delalloc_range() but before btrfs_finish_ordered_io(). >>> >>> Did you mean that? >> >> This happens event before a writeback happens. So, here is what is >> happening. This is in reference, and specific to the test case in the >> description. > > The flowchart helps a lot, thanks. > > But still some flow is still strange. >> >> Process: dd - first time >> __btrfs_buffered_write() >> |- btrfs_check_data_free_space() >> | |- btrfs_qgroup_reserve_data() <- Mark QGROUP_RESERVED bit >> |- btrfs_dirty_pages() <- Mark page dirty >> >> Please note data writeback does _not_ happen/complete. > > This part is completely fine. > >> >> Process: dd - next time, new process >> sys_open O_TRUNC >> . >> |-btrfs_setattr() >> |-truncate_pagecache() >> |-truncate_inode_pages_range() >> |-truncate_inode_page() - Page is cleared of Dirty flag. >> |-btrfs_invalidatepage(page) >> |-__btrfs_qgroup_release_data() >> |-btrfs_qgroup_free_refroot() - qgroup->reserved is >> reduced by PAGESIZE. > > That's OK too. No problem. > Although the __btrfs_qgroup_release_data() is called by > btrfs_qgroup_free_data(). > Which cleared QGROUP_RESERVED flag. > > > So the call flow should be > > |-btrfs_setattr() > |-truncate_pagecache() > |-truncate_inode_pages_range() > |-truncate_inode_page() <- Clear page dirty > |-btrfs_invalidatepage(page) > |-btrfs_qgroup_free_data() <- reduce qgroup->reserved > and clear QGROUP_RESERVED > > After btrfs_setattr(), the new dd write data. > So __btrfs_buffered_write() happens again, > dirtying pages and mark QGROUP_RESERVED and increase qgroup->reserved. > > So here I don't see any problem > buffered_write: > Mark dirty > Mark QGROUP_RESERVED > Increase qgroup->reserved > > btrfs_setattr: > Clear dirty > Clear QGROUP_RESERVED > Decrease qgroup->reserved > > All pairs with each other, no leak no underflow. Yes, but the problem is run_one_delayed_ref() |-btrfs_qgroup_free_delayed_ref uses delayed_ref_head->qgroup_reserved to reduce the count. Which function is responsible for decreasing delayed_ref_head->qgroup_reserved when btrfs_invalidatepage() is reducing the count? > > Until the last buffered_write(), which doesn't got truncated. > >> >> >> Process: sync >> btrfs_sync_fs() >> |-btrfs_commit_transaction() >> |-btrfs_run_delayed_refs() >> |- qgroup_free_refroot() - Reduces reserved by the size of the >> extent (in this case, the filesize of dd (first time)), even though some >> of the PAGESIZEs have been reduced while performing the truncate on the >> file. > > The delayed_ref belongs to the last extent, which doesn't got truncated. > > And in that case, that last extent got into disk. > run_dealloc_range() <- Write data into disk > finish_ordered_io() <- Update metadata > |- insert_reserved_file_extents() > |- btrfs_alloc_reserved_file_extent() > | |- btrfs_add_delayed_data_ref > | <This will cause a new delayed_data_ref> > |- btrfs_qgroup_release_data() > <This will clear QGROURP_RESERVED> > > Then goes to your btrfs_sync_fs() => qgroup_free_refroot() > > > > [[I think I find the problem now]] > > Between btrfs_alloc_reserved_file_extent() and > btrfs_qgroup_release_data(), there is a window that delayed_refs can be > executed. > Since that d*mn delayed_refs can be run at any time asynchronously. > > In that case, it will call qgroup_free_refroot() first at another > process context, and later btrfs_qgroup_release_data() get schedualed, > which will clear QGROUP_RESERVED again, causing the underflow. > > Although it's not the original cause you reported, would you mind to try > the following quick-dirty fix without your fix, to see if it fixes the > problem? > > ------- > diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c > index e6811c4..70efa14 100644 > --- a/fs/btrfs/inode.c > +++ b/fs/btrfs/inode.c > @@ -2168,15 +2168,15 @@ static int insert_reserved_file_extent(struct > btrfs_trans_handle *trans, > ins.objectid = disk_bytenr; > ins.offset = disk_num_bytes; > ins.type = BTRFS_EXTENT_ITEM_KEY; > - ret = btrfs_alloc_reserved_file_extent(trans, root, > - root->root_key.objectid, > - btrfs_ino(inode), file_pos, > - ram_bytes, &ins); > /* > * Release the reserved range from inode dirty range map, as it is > * already moved into delayed_ref_head > */ > btrfs_qgroup_release_data(inode, file_pos, ram_bytes); > + ret = btrfs_alloc_reserved_file_extent(trans, root, > + root->root_key.objectid, > + btrfs_ino(inode), file_pos, > + ram_bytes, &ins); > out: > btrfs_free_path(path); > > ------- No, it does not work. > >> >> I hope that makes it clear. >> >>> >>> [[About test case]] >>> And for the test case, I can't reproduce the problem no matter if I >>> apply the fix or not. >>> >>> Either way it just fails after 3 loops of dd, and later dd will all >>> fail. >>> But I can still remove the file and write new data into the fs. >>> >> >> Strange? I can reproduce at every instance of running it. Even on >> 4.8.0-rc7 > > Maybe related to the memory size. > > Since you're also using VM, maybe your VM RAM is too small, so dirty > pages pressure triggers a commit halfway and cause the race? Not quite, this problem happens on a physical machine as well. Besides, this VM is not in memory pressure. I tried it again on 4.8.0-rc8.
At 09/27/2016 10:04 PM, Goldwyn Rodrigues wrote: > > > On 09/26/2016 10:10 PM, Qu Wenruo wrote: >> >> >> At 09/26/2016 10:31 PM, Goldwyn Rodrigues wrote: >>> >>> >>> On 09/25/2016 09:33 PM, Qu Wenruo wrote: >>>> >>>> >>>> At 09/23/2016 09:43 PM, Goldwyn Rodrigues wrote: >>>>> >>>>> >> [snipped] >>>> >>>> Sorry I still don't get the point. >>>> Would you please give a call flow of the timing dirtying page and >>>> calling btrfs_qgroup_reserve/free/release_data()? >>>> >>>> Like: >>>> __btrfs_buffered_write() >>>> |- btrfs_check_data_free_space() >>>> | |- btrfs_qgroup_reserve_data() <- Mark QGROUP_RESERVED bit >>>> |- btrfs_dirty_pages() <- Mark page dirty >>>> >>>> >>>> [[Timing of btrfs_invalidatepage()]] >>>> About your commit message "once while invalidating the page, and the >>>> next time while free'ing delalloc." >>>> "Free'ing delalloc" did you mean btrfs_qgroup_free_delayed_ref(). >>>> >>>> If so, it means one extent goes through full write back, and long before >>>> calling btrfs_qgroup_free_delayed_ref(), it will call >>>> btrfs_qgroup_release_data() to clear QGROUP_RESERVED. >>>> >>>> So the call will be: >>>> __btrfs_buffered_write() >>>> |- btrfs_check_data_free_space() >>>> | |- btrfs_qgroup_reserve_data() <- Mark QGROUP_RESERVED bit >>>> |- btrfs_dirty_pages() <- Mark page dirty >>>> >>>> <data write back happens> >>>> run_delalloc_range() >>>> |- cow_file_range() >>>> |- extent_clear_unlock_delalloc() <- Clear page dirty >>>> >>>> <modifying metadata> >>>> >>>> btrfs_finish_ordered_io() >>>> |- insert_reserved_file_extent() >>>> |- btrfs_qgroup_release_data() <- Clear QGROUP_RESERVED bit >>>> but not decrease reserved space >>>> >>>> <run delayed refs, normally happens in commit_trans> >>>> run_one_delyaed_refs() >>>> |- btrfs_qgroup_free_delayed_ref() <- Directly decrease reserved space >>>> >>>> >>>> So the problem seems to be, btrfs_invalidatepage() is called after >>>> run_delalloc_range() but before btrfs_finish_ordered_io(). >>>> >>>> Did you mean that? >>> >>> This happens event before a writeback happens. So, here is what is >>> happening. This is in reference, and specific to the test case in the >>> description. >> >> The flowchart helps a lot, thanks. >> >> But still some flow is still strange. >>> >>> Process: dd - first time >>> __btrfs_buffered_write() >>> |- btrfs_check_data_free_space() >>> | |- btrfs_qgroup_reserve_data() <- Mark QGROUP_RESERVED bit >>> |- btrfs_dirty_pages() <- Mark page dirty >>> >>> Please note data writeback does _not_ happen/complete. >> >> This part is completely fine. >> >>> >>> Process: dd - next time, new process >>> sys_open O_TRUNC >>> . >>> |-btrfs_setattr() >>> |-truncate_pagecache() >>> |-truncate_inode_pages_range() >>> |-truncate_inode_page() - Page is cleared of Dirty flag. >>> |-btrfs_invalidatepage(page) >>> |-__btrfs_qgroup_release_data() >>> |-btrfs_qgroup_free_refroot() - qgroup->reserved is >>> reduced by PAGESIZE. >> >> That's OK too. No problem. >> Although the __btrfs_qgroup_release_data() is called by >> btrfs_qgroup_free_data(). >> Which cleared QGROUP_RESERVED flag. >> >> >> So the call flow should be >> >> |-btrfs_setattr() >> |-truncate_pagecache() >> |-truncate_inode_pages_range() >> |-truncate_inode_page() <- Clear page dirty >> |-btrfs_invalidatepage(page) >> |-btrfs_qgroup_free_data() <- reduce qgroup->reserved >> and clear QGROUP_RESERVED >> >> After btrfs_setattr(), the new dd write data. >> So __btrfs_buffered_write() happens again, >> dirtying pages and mark QGROUP_RESERVED and increase qgroup->reserved. >> >> So here I don't see any problem >> buffered_write: >> Mark dirty >> Mark QGROUP_RESERVED >> Increase qgroup->reserved >> >> btrfs_setattr: >> Clear dirty >> Clear QGROUP_RESERVED >> Decrease qgroup->reserved >> >> All pairs with each other, no leak no underflow. > > Yes, but the problem is > run_one_delayed_ref() > |-btrfs_qgroup_free_delayed_ref > uses delayed_ref_head->qgroup_reserved to reduce the count. Which > function is responsible for decreasing delayed_ref_head->qgroup_reserved > when btrfs_invalidatepage() is reducing the count? Decreasing qgroup_reserved can happen in two ways: 1) Through qgroup_release_data() + qgroup_free_delayed_ref() combination That's used for case like writing pages into disk. The work flow is: __btrfs_buffered_write() |- btrfs_check_free_space() |- btrfs_qgroup_reserve_data() <- Mark QGROUP_RESERVED flag Increase qgroup->reserved. run_dealloc_range() <- Write data into disk finish_ordered_io() <- Update metadata |- add_delayed_data_ref() <- Create a new data_ref, record | how much qgroup->reserved needs to | free |- btrfs_qgroup_release_data() <- Only clearing QGROUP_RESERVED flag But qgroup->reserved is not decreased commit_trans() |- run_delayed_refs() |- run_one_delayed_ref() |- btrfs_qgroup_free_delayed_ref() <- Decrease qgroup->reserved So if we hit btrfs_qgroup_free_delayed_ref(), it means one extent hits disk. The complexity is because qgroup is done at commit_trans() time, so qgroup excl/rfer is not updated at finish_ordered_io() time. If we decrease qgroup->reserved here, then we can easily exceed qgroup limit. Maybe the problem is about the delayed_ref. IIRC some case we can create delayed_ref whose refs is still 0. Which means the delayed ref although hit disk, but got freed in the same trans. If that's the case, we shouldn't call qgroup_free_delayed_ref() for such delayed_ref. 2) Through qgroup_free_data() That's for case where extent doesn't reach disk. The flow is much more straightforward: __btrfs_buffered_write() |- btrfs_check_free_space() |- btrfs_qgroup_reserve_data() <- Mark QGROUP_RESERVED flag Increase qgroup->reserved. btrfs_invalidatepage() |- btrfs_qgroup_free_data() <- Clear QGROUP_RESERVED flag Decrease qgroup->reserved. > > >> >> Until the last buffered_write(), which doesn't got truncated. >> >>> >>> >>> Process: sync >>> btrfs_sync_fs() >>> |-btrfs_commit_transaction() >>> |-btrfs_run_delayed_refs() >>> |- qgroup_free_refroot() - Reduces reserved by the size of the >>> extent (in this case, the filesize of dd (first time)), even though some >>> of the PAGESIZEs have been reduced while performing the truncate on the >>> file. >> >> The delayed_ref belongs to the last extent, which doesn't got truncated. >> >> And in that case, that last extent got into disk. >> run_dealloc_range() <- Write data into disk >> finish_ordered_io() <- Update metadata >> |- insert_reserved_file_extents() >> |- btrfs_alloc_reserved_file_extent() >> | |- btrfs_add_delayed_data_ref >> | <This will cause a new delayed_data_ref> >> |- btrfs_qgroup_release_data() >> <This will clear QGROURP_RESERVED> >> >> Then goes to your btrfs_sync_fs() => qgroup_free_refroot() >> >> >> >> [[I think I find the problem now]] >> >> Between btrfs_alloc_reserved_file_extent() and >> btrfs_qgroup_release_data(), there is a window that delayed_refs can be >> executed. >> Since that d*mn delayed_refs can be run at any time asynchronously. >> >> In that case, it will call qgroup_free_refroot() first at another >> process context, and later btrfs_qgroup_release_data() get schedualed, >> which will clear QGROUP_RESERVED again, causing the underflow. >> >> Although it's not the original cause you reported, would you mind to try >> the following quick-dirty fix without your fix, to see if it fixes the >> problem? >> >> ------- >> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c >> index e6811c4..70efa14 100644 >> --- a/fs/btrfs/inode.c >> +++ b/fs/btrfs/inode.c >> @@ -2168,15 +2168,15 @@ static int insert_reserved_file_extent(struct >> btrfs_trans_handle *trans, >> ins.objectid = disk_bytenr; >> ins.offset = disk_num_bytes; >> ins.type = BTRFS_EXTENT_ITEM_KEY; >> - ret = btrfs_alloc_reserved_file_extent(trans, root, >> - root->root_key.objectid, >> - btrfs_ino(inode), file_pos, >> - ram_bytes, &ins); >> /* >> * Release the reserved range from inode dirty range map, as it is >> * already moved into delayed_ref_head >> */ >> btrfs_qgroup_release_data(inode, file_pos, ram_bytes); >> + ret = btrfs_alloc_reserved_file_extent(trans, root, >> + root->root_key.objectid, >> + btrfs_ino(inode), file_pos, >> + ram_bytes, &ins); >> out: >> btrfs_free_path(path); >> >> ------- > > No, it does not work. > >> >>> >>> I hope that makes it clear. >>> >>>> >>>> [[About test case]] >>>> And for the test case, I can't reproduce the problem no matter if I >>>> apply the fix or not. >>>> >>>> Either way it just fails after 3 loops of dd, and later dd will all >>>> fail. >>>> But I can still remove the file and write new data into the fs. >>>> >>> >>> Strange? I can reproduce at every instance of running it. Even on >>> 4.8.0-rc7 >> >> Maybe related to the memory size. >> >> Since you're also using VM, maybe your VM RAM is too small, so dirty >> pages pressure triggers a commit halfway and cause the race? > > Not quite, this problem happens on a physical machine as well. Besides, > this VM is not in memory pressure. I tried it again on 4.8.0-rc8. > Finally reproduced it. Although in a low chance, about 1/10. Under most case, the final remove gets executed without error. ------- sudo ./qgroup_underflow.sh btrfs-progs v4.7.3 See http://btrfs.wiki.kernel.org for more information. Label: (null) UUID: Node size: 16384 Sector size: 4096 Filesystem size: 10.00GiB Block group profiles: Data: single 8.00MiB Metadata: DUP 1.00GiB System: DUP 8.00MiB SSD detected: no Incompat features: extref, skinny-metadata Number of devices: 1 Devices: ID SIZE PATH 1 10.00GiB /dev/vdb5 Create subvolume './a' 40+0 records in 40+0 records out 41943040 bytes (42 MB, 40 MiB) copied, 0.0450651 s, 931 MB/s 40+0 records in 40+0 records out 41943040 bytes (42 MB, 40 MiB) copied, 0.0771046 s, 544 MB/s dd: error writing '/mnt/scratch/a/file': Disk quota exceeded 10+0 records in 9+0 records out 9961472 bytes (10 MB, 9.5 MiB) copied, 0.0179116 s, 556 MB/s dd: error writing '/mnt/scratch/a/file': Disk quota exceeded 1+0 records in 0+0 records out 0 bytes copied, 0.00101051 s, 0.0 kB/s dd: error writing '/mnt/scratch/a/file': Disk quota exceeded 1+0 records in 0+0 records out 0 bytes copied, 0.000875674 s, 0.0 kB/s dd: error writing '/mnt/scratch/a/file': Disk quota exceeded 1+0 records in 0+0 records out 0 bytes copied, 0.000889771 s, 0.0 kB/s dd: failed to open '/mnt/scratch/a/file': Disk quota exceeded dd: failed to open '/mnt/scratch/a/file': Disk quota exceeded dd: failed to open '/mnt/scratch/a/file': Disk quota exceeded dd: failed to open '/mnt/scratch/a/file': Disk quota exceeded dd: failed to open '/mnt/scratch/a/file': Disk quota exceeded dd: failed to open '/mnt/scratch/a/file': Disk quota exceeded dd: failed to open '/mnt/scratch/a/file': Disk quota exceeded dd: failed to open '/mnt/scratch/a/file': Disk quota exceeded dd: failed to open '/mnt/scratch/a/file': Disk quota exceeded Removing file ------- I'd try to tweak the script to improve the possibility and then use ftrace to track qgroup reserve and delayed_ref related event. Thanks, Qu -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 09/27/2016 08:44 PM, Qu Wenruo wrote: <snipped> > Finally reproduced it. > > Although in a low chance, about 1/10. > > Under most case, the final remove gets executed without error. Change 50m to 500m while setting the qgroup limit, the probability will increase.
Thanks for your test script.
I succeeded in pinning down the problem.
The problem is, btrfs is invalidate pages that belongs to ordered
extent(goes through writeback)
The ftrace(with more tracepoints to trace qgroup->reserved change) shows:
------
btrfs_qgroup_release_data: root=257, ino=257, start=0, len=4096,
reserved=4096, op=free
qgroup_update_reserve: qgid = 257, cur_reserved = 16171008, diff = -4096
btrfs_qgroup_release_data: root=257, ino=257, start=4096, len=4096,
reserved=4096, op=free
qgroup_update_reserve: qgid = 257, cur_reserved = 16166912, diff = -4096
btrfs_qgroup_release_data: root=257, ino=257, start=8192, len=4096,
reserved=4096, op=free
qgroup_update_reserve: qgid = 257, cur_reserved = 16162816, diff = -4096
btrfs_qgroup_release_data: root=257, ino=257, start=12288, len=4096,
reserved=4096, op=free
qgroup_update_reserve: qgid = 257, cur_reserved = 16158720, diff = -4096
btrfs_qgroup_release_data: root=257, ino=257, start=16384, len=4096,
reserved=4096, op=free
qgroup_update_reserve: qgid = 257, cur_reserved = 16154624, diff = -4096
^^^^^ Btrfs invalidates 5 pages of ino 257 qg 257, range 0 ~ 20K.
btrfs_qgroup_release_data: root=257, ino=257, start=0, len=5242880,
reserved=5222400, op=release
^^^^^ Here ordered_io finishes, write the whole 5M data, while
QGROUP_RESERVED only returned 5M - 20K.
------
The problem is at btrfs_add_delayed_data_ref(), we use the assumption
that, qgroup reserved space is the same as extent size(5M)
But in fact, btrfs_qgroup_release_data() only release (5M - 20K).
Leaking the 20K.
Calltrace:
btrfs_finish_ordered_io()
|- inserted_reserved_file_extent()
|- btrfs_alloc_reserved_file_extent()
| ^^ Here we tell delayed_ref to free 5M later
|- btrfs_qgroup_release_data()
^^ We only released (5M - 20K), as the first 5 pages
are freed by btrfs_invalidatepage()
If btrfs is designed to freeing pages which belongs to ordered_extent,
then it totally breaks the qgroup assumption.
Then we have several ways to fix it:
1) Fix race between ordered extent(writeback) with invalidatepage()
Make btrfs_invalidatepage() skips pages that are already in
ordered_extent, then we're OK.
This is much like your original fix, but I'm not sure if DIRTY page
flag is bond to ordered_extent.
And more comment will help.
2) Make btrfs_qgroup_release_data() return accurate num bytes
And then pass it to delayed_ref head.
This is quite anti-intuition. If we're adding a new extent, how
could it happen that some pages are already invalidated?
Anyway, awesome work, exposed a quite strange race and makes us re-think
the base of qgroup reserve space framework.
Thanks,
Qu
At 09/28/2016 10:19 AM, Goldwyn Rodrigues wrote:
>
>
> On 09/27/2016 08:44 PM, Qu Wenruo wrote:
>
> <snipped>
>
>> Finally reproduced it.
>>
>> Although in a low chance, about 1/10.
>>
>> Under most case, the final remove gets executed without error.
>
> Change 50m to 500m while setting the qgroup limit, the probability will
> increase.
>
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
On 09/29/2016 03:57 AM, Qu Wenruo wrote: > Thanks for your test script. > > I succeeded in pinning down the problem. > > The problem is, btrfs is invalidate pages that belongs to ordered > extent(goes through writeback) No, I don't think invalidated pages are going through writeback. The problem is that the space for extent allocation is done before the writeback and if pages are invalidated before writeback, it is not accounted for. > > The ftrace(with more tracepoints to trace qgroup->reserved change) shows: > ------ > btrfs_qgroup_release_data: root=257, ino=257, start=0, len=4096, > reserved=4096, op=free > qgroup_update_reserve: qgid = 257, cur_reserved = 16171008, diff = -4096 > btrfs_qgroup_release_data: root=257, ino=257, start=4096, len=4096, > reserved=4096, op=free > qgroup_update_reserve: qgid = 257, cur_reserved = 16166912, diff = -4096 > btrfs_qgroup_release_data: root=257, ino=257, start=8192, len=4096, > reserved=4096, op=free > qgroup_update_reserve: qgid = 257, cur_reserved = 16162816, diff = -4096 > btrfs_qgroup_release_data: root=257, ino=257, start=12288, len=4096, > reserved=4096, op=free > qgroup_update_reserve: qgid = 257, cur_reserved = 16158720, diff = -4096 > btrfs_qgroup_release_data: root=257, ino=257, start=16384, len=4096, > reserved=4096, op=free > qgroup_update_reserve: qgid = 257, cur_reserved = 16154624, diff = -4096 This is a little confusing. There is no qgroup_update_reserve() function in the source. Where did you add this? > > > ^^^^^ Btrfs invalidates 5 pages of ino 257 qg 257, range 0 ~ 20K. > > btrfs_qgroup_release_data: root=257, ino=257, start=0, len=5242880, > reserved=5222400, op=release > > ^^^^^ Here ordered_io finishes, write the whole 5M data, while > QGROUP_RESERVED only returned 5M - 20K. > ------ > > The problem is at btrfs_add_delayed_data_ref(), we use the assumption > that, qgroup reserved space is the same as extent size(5M) > But in fact, btrfs_qgroup_release_data() only release (5M - 20K). > Leaking the 20K. Yes, this is more like it. However, I would put it the other way. It releases 5M when it should have released 5M-20K, because 20k was released when invalidating page. > > Calltrace: > btrfs_finish_ordered_io() > |- inserted_reserved_file_extent() > |- btrfs_alloc_reserved_file_extent() > | ^^ Here we tell delayed_ref to free 5M later > |- btrfs_qgroup_release_data() > ^^ We only released (5M - 20K), as the first 5 pages > are freed by btrfs_invalidatepage() > > If btrfs is designed to freeing pages which belongs to ordered_extent, > then it totally breaks the qgroup assumption. > > Then we have several ways to fix it: > 1) Fix race between ordered extent(writeback) with invalidatepage() > Make btrfs_invalidatepage() skips pages that are already in > ordered_extent, then we're OK. > > This is much like your original fix, but I'm not sure if DIRTY page > flag is bond to ordered_extent. > And more comment will help. So, what you mean is that fix is correct, I just need to reword the patch header. > > 2) Make btrfs_qgroup_release_data() return accurate num bytes > And then pass it to delayed_ref head. > > This is quite anti-intuition. If we're adding a new extent, how > could it happen that some pages are already invalidated? > I agree. It is counter-intutive and will increase complexity. > Anyway, awesome work, exposed a quite strange race and makes us re-think > the base of qgroup reserve space framework. Thanks.
At 09/29/2016 07:13 PM, Goldwyn Rodrigues wrote: > > > On 09/29/2016 03:57 AM, Qu Wenruo wrote: >> Thanks for your test script. >> >> I succeeded in pinning down the problem. >> >> The problem is, btrfs is invalidate pages that belongs to ordered >> extent(goes through writeback) > > > No, I don't think invalidated pages are going through writeback. The > problem is that the space for extent allocation is done before the > writeback and if pages are invalidated before writeback, it is not > accounted for. > >> >> The ftrace(with more tracepoints to trace qgroup->reserved change) shows: >> ------ >> btrfs_qgroup_release_data: root=257, ino=257, start=0, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16171008, diff = -4096 >> btrfs_qgroup_release_data: root=257, ino=257, start=4096, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16166912, diff = -4096 >> btrfs_qgroup_release_data: root=257, ino=257, start=8192, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16162816, diff = -4096 >> btrfs_qgroup_release_data: root=257, ino=257, start=12288, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16158720, diff = -4096 >> btrfs_qgroup_release_data: root=257, ino=257, start=16384, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16154624, diff = -4096 > > This is a little confusing. There is no qgroup_update_reserve() function > in the source. Where did you add this? Sorry, I just submitted them few minutes ago. Have CCed you. > >> >> >> ^^^^^ Btrfs invalidates 5 pages of ino 257 qg 257, range 0 ~ 20K. >> >> btrfs_qgroup_release_data: root=257, ino=257, start=0, len=5242880, >> reserved=5222400, op=release >> >> ^^^^^ Here ordered_io finishes, write the whole 5M data, while >> QGROUP_RESERVED only returned 5M - 20K. >> ------ >> >> The problem is at btrfs_add_delayed_data_ref(), we use the assumption >> that, qgroup reserved space is the same as extent size(5M) >> But in fact, btrfs_qgroup_release_data() only release (5M - 20K). >> Leaking the 20K. > > Yes, this is more like it. However, I would put it the other way. It > releases 5M when it should have released 5M-20K, because 20k was > released when invalidating page. > >> >> Calltrace: >> btrfs_finish_ordered_io() >> |- inserted_reserved_file_extent() >> |- btrfs_alloc_reserved_file_extent() >> | ^^ Here we tell delayed_ref to free 5M later >> |- btrfs_qgroup_release_data() >> ^^ We only released (5M - 20K), as the first 5 pages >> are freed by btrfs_invalidatepage() >> >> If btrfs is designed to freeing pages which belongs to ordered_extent, >> then it totally breaks the qgroup assumption. >> >> Then we have several ways to fix it: >> 1) Fix race between ordered extent(writeback) with invalidatepage() >> Make btrfs_invalidatepage() skips pages that are already in >> ordered_extent, then we're OK. >> >> This is much like your original fix, but I'm not sure if DIRTY page >> flag is bond to ordered_extent. >> And more comment will help. > > > So, what you mean is that fix is correct, I just need to reword the > patch header. Not patch header, but comment before PageDirty() check. I'm still digging the code to see if page DIRTY check is safe and if there is more elegant fix. Thanks, Qu > >> >> 2) Make btrfs_qgroup_release_data() return accurate num bytes >> And then pass it to delayed_ref head. >> >> This is quite anti-intuition. If we're adding a new extent, how >> could it happen that some pages are already invalidated? >> > > I agree. It is counter-intutive and will increase complexity. > >> Anyway, awesome work, exposed a quite strange race and makes us re-think >> the base of qgroup reserve space framework. > > Thanks. > -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Your original fix is good. Feel free to my tag after enriching the comment in btrfs_invalidatepage(). Reviewed-by: Qu Wenruo <quwenruo@cn.fujitsu.com> Thanks again for your founding and sorry for the extra time reviewing. Thanks Qu At 09/29/2016 07:13 PM, Goldwyn Rodrigues wrote: > > > On 09/29/2016 03:57 AM, Qu Wenruo wrote: >> Thanks for your test script. >> >> I succeeded in pinning down the problem. >> >> The problem is, btrfs is invalidate pages that belongs to ordered >> extent(goes through writeback) > > > No, I don't think invalidated pages are going through writeback. The > problem is that the space for extent allocation is done before the > writeback and if pages are invalidated before writeback, it is not > accounted for. > >> >> The ftrace(with more tracepoints to trace qgroup->reserved change) shows: >> ------ >> btrfs_qgroup_release_data: root=257, ino=257, start=0, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16171008, diff = -4096 >> btrfs_qgroup_release_data: root=257, ino=257, start=4096, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16166912, diff = -4096 >> btrfs_qgroup_release_data: root=257, ino=257, start=8192, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16162816, diff = -4096 >> btrfs_qgroup_release_data: root=257, ino=257, start=12288, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16158720, diff = -4096 >> btrfs_qgroup_release_data: root=257, ino=257, start=16384, len=4096, >> reserved=4096, op=free >> qgroup_update_reserve: qgid = 257, cur_reserved = 16154624, diff = -4096 > > This is a little confusing. There is no qgroup_update_reserve() function > in the source. Where did you add this? > >> >> >> ^^^^^ Btrfs invalidates 5 pages of ino 257 qg 257, range 0 ~ 20K. >> >> btrfs_qgroup_release_data: root=257, ino=257, start=0, len=5242880, >> reserved=5222400, op=release >> >> ^^^^^ Here ordered_io finishes, write the whole 5M data, while >> QGROUP_RESERVED only returned 5M - 20K. >> ------ >> >> The problem is at btrfs_add_delayed_data_ref(), we use the assumption >> that, qgroup reserved space is the same as extent size(5M) >> But in fact, btrfs_qgroup_release_data() only release (5M - 20K). >> Leaking the 20K. > > Yes, this is more like it. However, I would put it the other way. It > releases 5M when it should have released 5M-20K, because 20k was > released when invalidating page. > >> >> Calltrace: >> btrfs_finish_ordered_io() >> |- inserted_reserved_file_extent() >> |- btrfs_alloc_reserved_file_extent() >> | ^^ Here we tell delayed_ref to free 5M later >> |- btrfs_qgroup_release_data() >> ^^ We only released (5M - 20K), as the first 5 pages >> are freed by btrfs_invalidatepage() >> >> If btrfs is designed to freeing pages which belongs to ordered_extent, >> then it totally breaks the qgroup assumption. >> >> Then we have several ways to fix it: >> 1) Fix race between ordered extent(writeback) with invalidatepage() >> Make btrfs_invalidatepage() skips pages that are already in >> ordered_extent, then we're OK. >> >> This is much like your original fix, but I'm not sure if DIRTY page >> flag is bond to ordered_extent. >> And more comment will help. > > > So, what you mean is that fix is correct, I just need to reword the > patch header. > >> >> 2) Make btrfs_qgroup_release_data() return accurate num bytes >> And then pass it to delayed_ref head. >> >> This is quite anti-intuition. If we're adding a new extent, how >> could it happen that some pages are already invalidated? >> > > I agree. It is counter-intutive and will increase complexity. > >> Anyway, awesome work, exposed a quite strange race and makes us re-think >> the base of qgroup reserve space framework. > > Thanks. > -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
At 09/30/2016 10:18 AM, Qu Wenruo wrote: > Your original fix is good. > > Feel free to my tag after enriching the comment in btrfs_invalidatepage(). Feel free to "add" my tag... > > Reviewed-by: Qu Wenruo <quwenruo@cn.fujitsu.com> > > Thanks again for your founding and sorry for the extra time reviewing. for your "finding"... My English just sucks.... > > Thanks > Qu > > > At 09/29/2016 07:13 PM, Goldwyn Rodrigues wrote: >> >> >> On 09/29/2016 03:57 AM, Qu Wenruo wrote: >>> Thanks for your test script. >>> >>> I succeeded in pinning down the problem. >>> >>> The problem is, btrfs is invalidate pages that belongs to ordered >>> extent(goes through writeback) >> >> >> No, I don't think invalidated pages are going through writeback. The >> problem is that the space for extent allocation is done before the >> writeback and if pages are invalidated before writeback, it is not >> accounted for. >> >>> >>> The ftrace(with more tracepoints to trace qgroup->reserved change) >>> shows: >>> ------ >>> btrfs_qgroup_release_data: root=257, ino=257, start=0, len=4096, >>> reserved=4096, op=free >>> qgroup_update_reserve: qgid = 257, cur_reserved = 16171008, diff = -4096 >>> btrfs_qgroup_release_data: root=257, ino=257, start=4096, len=4096, >>> reserved=4096, op=free >>> qgroup_update_reserve: qgid = 257, cur_reserved = 16166912, diff = -4096 >>> btrfs_qgroup_release_data: root=257, ino=257, start=8192, len=4096, >>> reserved=4096, op=free >>> qgroup_update_reserve: qgid = 257, cur_reserved = 16162816, diff = -4096 >>> btrfs_qgroup_release_data: root=257, ino=257, start=12288, len=4096, >>> reserved=4096, op=free >>> qgroup_update_reserve: qgid = 257, cur_reserved = 16158720, diff = -4096 >>> btrfs_qgroup_release_data: root=257, ino=257, start=16384, len=4096, >>> reserved=4096, op=free >>> qgroup_update_reserve: qgid = 257, cur_reserved = 16154624, diff = -4096 >> >> This is a little confusing. There is no qgroup_update_reserve() function >> in the source. Where did you add this? >> >>> >>> >>> ^^^^^ Btrfs invalidates 5 pages of ino 257 qg 257, range 0 ~ 20K. >>> >>> btrfs_qgroup_release_data: root=257, ino=257, start=0, len=5242880, >>> reserved=5222400, op=release >>> >>> ^^^^^ Here ordered_io finishes, write the whole 5M data, while >>> QGROUP_RESERVED only returned 5M - 20K. >>> ------ >>> >>> The problem is at btrfs_add_delayed_data_ref(), we use the assumption >>> that, qgroup reserved space is the same as extent size(5M) >>> But in fact, btrfs_qgroup_release_data() only release (5M - 20K). >>> Leaking the 20K. >> >> Yes, this is more like it. However, I would put it the other way. It >> releases 5M when it should have released 5M-20K, because 20k was >> released when invalidating page. >> >>> >>> Calltrace: >>> btrfs_finish_ordered_io() >>> |- inserted_reserved_file_extent() >>> |- btrfs_alloc_reserved_file_extent() >>> | ^^ Here we tell delayed_ref to free 5M later >>> |- btrfs_qgroup_release_data() >>> ^^ We only released (5M - 20K), as the first 5 pages >>> are freed by btrfs_invalidatepage() >>> >>> If btrfs is designed to freeing pages which belongs to ordered_extent, >>> then it totally breaks the qgroup assumption. >>> >>> Then we have several ways to fix it: >>> 1) Fix race between ordered extent(writeback) with invalidatepage() >>> Make btrfs_invalidatepage() skips pages that are already in >>> ordered_extent, then we're OK. >>> >>> This is much like your original fix, but I'm not sure if DIRTY page >>> flag is bond to ordered_extent. >>> And more comment will help. >> >> >> So, what you mean is that fix is correct, I just need to reword the >> patch header. >> >>> >>> 2) Make btrfs_qgroup_release_data() return accurate num bytes >>> And then pass it to delayed_ref head. >>> >>> This is quite anti-intuition. If we're adding a new extent, how >>> could it happen that some pages are already invalidated? >>> >> >> I agree. It is counter-intutive and will increase complexity. >> >>> Anyway, awesome work, exposed a quite strange race and makes us re-think >>> the base of qgroup reserve space framework. >> >> Thanks. >> > > > -- > To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c index e6811c4..70efa14 100644 --- a/fs/btrfs/inode.c +++ b/fs/btrfs/inode.c @@ -2168,15 +2168,15 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, ins.objectid = disk_bytenr; ins.offset = disk_num_bytes; ins.type = BTRFS_EXTENT_ITEM_KEY; - ret = btrfs_alloc_reserved_file_extent(trans, root, - root->root_key.objectid, - btrfs_ino(inode), file_pos, - ram_bytes, &ins); /* * Release the reserved range from inode dirty range map, as it is * already moved into delayed_ref_head */ btrfs_qgroup_release_data(inode, file_pos, ram_bytes); + ret = btrfs_alloc_reserved_file_extent(trans, root, + root->root_key.objectid, + btrfs_ino(inode), file_pos, + ram_bytes, &ins); out: btrfs_free_path(path);