Re: [RFC PATCH] xfs: drop experimental warning for fsdax

From: Shiyang Ruan
Date: Sun Oct 16 2022 - 10:05:43 EST




在 2022/10/14 23:50, Darrick J. Wong 写道:
On Fri, Oct 14, 2022 at 10:24:29AM +0800, Shiyang Ruan wrote:


在 2022/10/14 2:30, Darrick J. Wong 写道:
On Thu, Sep 29, 2022 at 12:05:14PM -0700, Darrick J. Wong wrote:
On Wed, Sep 28, 2022 at 10:46:17PM +0800, Shiyang Ruan wrote:

...

FWIW I saw dmesg failures in xfs/517 and xfs/013 starting with 6.0-rc5,
and I haven't even turned on reflink yet:

run fstests xfs/517 at 2022-09-26 19:53:34
XFS (pmem1): EXPERIMENTAL Large extent counts feature in use. Use at your own risk!
XFS (pmem1): Mounting V5 Filesystem
XFS (pmem1): Ending clean mount
XFS (pmem1): Quotacheck needed: Please wait.
XFS (pmem1): Quotacheck: Done.
XFS (pmem1): Unmounting Filesystem
XFS (pmem0): EXPERIMENTAL online scrub feature in use. Use at your own risk!
XFS (pmem1): EXPERIMENTAL Large extent counts feature in use. Use at your own risk!
XFS (pmem1): Mounting V5 Filesystem
XFS (pmem1): Ending clean mount
XFS (pmem1): Quotacheck needed: Please wait.
XFS (pmem1): Quotacheck: Done.
------------[ cut here ]------------
WARNING: CPU: 1 PID: 415317 at fs/dax.c:380 dax_insert_entry+0x22d/0x320

Ping?

This time around I replaced the WARN_ON with this:

if (page->mapping)
printk(KERN_ERR "%s:%d ino 0x%lx index 0x%lx page 0x%llx mapping 0x%llx <- 0x%llx\n", __func__, __LINE__, mapping->host->i_ino, index + i, (unsigned long long)page, (unsigned long long)page->mapping, (unsigned long long)mapping);

and promptly started seeing scary things like this:

[ 37.576598] dax_associate_entry:381 ino 0x1807870 index 0x370 page 0xffffea00133f1480 mapping 0x1 <- 0xffff888042fbb528
[ 37.577570] dax_associate_entry:381 ino 0x1807870 index 0x371 page 0xffffea00133f1500 mapping 0x1 <- 0xffff888042fbb528
[ 37.698657] dax_associate_entry:381 ino 0x180044a index 0x5f8 page 0xffffea0013244900 mapping 0xffff888042eaf128 <- 0xffff888042dda128
[ 37.699349] dax_associate_entry:381 ino 0x800808 index 0x136 page 0xffffea0013245640 mapping 0xffff888042eaf128 <- 0xffff888042d3ce28
[ 37.699680] dax_associate_entry:381 ino 0x180044a index 0x5f9 page 0xffffea0013245680 mapping 0xffff888042eaf128 <- 0xffff888042dda128
[ 37.700684] dax_associate_entry:381 ino 0x800808 index 0x137 page 0xffffea00132456c0 mapping 0xffff888042eaf128 <- 0xffff888042d3ce28
[ 37.701611] dax_associate_entry:381 ino 0x180044a index 0x5fa page 0xffffea0013245700 mapping 0xffff888042eaf128 <- 0xffff888042dda128
[ 37.764126] dax_associate_entry:381 ino 0x103c52c index 0x28a page 0xffffea001345afc0 mapping 0x1 <- 0xffff888019c14928
[ 37.765078] dax_associate_entry:381 ino 0x103c52c index 0x28b page 0xffffea001345b000 mapping 0x1 <- 0xffff888019c14928
[ 39.193523] dax_associate_entry:381 ino 0x184657f index 0x124 page 0xffffea000e2a4440 mapping 0xffff8880120d7628 <- 0xffff888019ca3528
[ 39.194692] dax_associate_entry:381 ino 0x184657f index 0x125 page 0xffffea000e2a4480 mapping 0xffff8880120d7628 <- 0xffff888019ca3528
[ 39.195716] dax_associate_entry:381 ino 0x184657f index 0x126 page 0xffffea000e2a44c0 mapping 0xffff8880120d7628 <- 0xffff888019ca3528
[ 39.196736] dax_associate_entry:381 ino 0x184657f index 0x127 page 0xffffea000e2a4500 mapping 0xffff8880120d7628 <- 0xffff888019ca3528
[ 39.197906] dax_associate_entry:381 ino 0x184657f index 0x128 page 0xffffea000e2a5040 mapping 0xffff8880120d7628 <- 0xffff888019ca3528
[ 39.198924] dax_associate_entry:381 ino 0x184657f index 0x129 page 0xffffea000e2a5080 mapping 0xffff8880120d7628 <- 0xffff888019ca3528
[ 39.247053] dax_associate_entry:381 ino 0x5dd1e index 0x2d page 0xffffea0015a0e640 mapping 0x1 <- 0xffff88804af88828
[ 39.248006] dax_associate_entry:381 ino 0x5dd1e index 0x2e page 0xffffea0015a0e680 mapping 0x1 <- 0xffff88804af88828
[ 39.490880] dax_associate_entry:381 ino 0x1a9dc index 0x7d page 0xffffea000e7012c0 mapping 0xffff888042fd1728 <- 0xffff88804afaec28
[ 39.492038] dax_associate_entry:381 ino 0x1a9dc index 0x7e page 0xffffea000e701300 mapping 0xffff888042fd1728 <- 0xffff88804afaec28
[ 39.493099] dax_associate_entry:381 ino 0x1a9dc index 0x7f page 0xffffea000e701340 mapping 0xffff888042fd1728 <- 0xffff88804afaec28
[ 40.926247] dax_associate_entry:381 ino 0x182e265 index 0x54c page 0xffffea0015da0840 mapping 0x1 <- 0xffff888019c0dd28
[ 41.675459] dax_associate_entry:381 ino 0x15e5d index 0x29 page 0xffffea000e4350c0 mapping 0x1 <- 0xffff888019c05828
[ 41.676418] dax_associate_entry:381 ino 0x15e5d index 0x2a page 0xffffea000e435100 mapping 0x1 <- 0xffff888019c05828
[ 41.677352] dax_associate_entry:381 ino 0x15e5d index 0x2b page 0xffffea000e435180 mapping 0x1 <- 0xffff888019c05828
[ 41.678372] dax_associate_entry:381 ino 0x15e5d index 0x2c page 0xffffea000e4351c0 mapping 0x1 <- 0xffff888019c05828
[ 41.965026] dax_associate_entry:381 ino 0x185adb4 index 0x87 page 0xffffea000e616d00 mapping 0x1 <- 0xffff88801a83b528
[ 41.966065] dax_associate_entry:381 ino 0x185adb4 index 0x88 page 0xffffea000e616d40 mapping 0x1 <- 0xffff88801a83b528
[ 43.565384] dax_associate_entry:381 ino 0x804d9d index 0x229 page 0xffffea0013653fc0 mapping 0x1 <- 0xffff88804bd97128
[ 43.566399] dax_associate_entry:381 ino 0x804d9d index 0x22a page 0xffffea0013654000 mapping 0x1 <- 0xffff88804bd97128
[ 43.567343] dax_associate_entry:381 ino 0x804d9d index 0x22b page 0xffffea0013654040 mapping 0x1 <- 0xffff88804bd97128
[ 45.512017] dax_associate_entry:381 ino 0x18192bb index 0x1f page 0xffffea00133f1300 mapping 0x1 <- 0xffff88804bcdb528
[ 45.512974] dax_associate_entry:381 ino 0x18192bb index 0x20 page 0xffffea00133f1340 mapping 0x1 <- 0xffff88804bcdb528
[ 45.513942] dax_associate_entry:381 ino 0x18192bb index 0x21 page 0xffffea00133f1380 mapping 0x1 <- 0xffff88804bcdb528
[ 45.514857] dax_associate_entry:381 ino 0x18192bb index 0x22 page 0xffffea00133f13c0 mapping 0x1 <- 0xffff88804bcdb528
[ 45.515760] dax_associate_entry:381 ino 0x18192bb index 0x23 page 0xffffea00133f1400 mapping 0x1 <- 0xffff88804bcdb528
[ 45.516673] dax_associate_entry:381 ino 0x18192bb index 0x24 page 0xffffea00133f1440 mapping 0x1 <- 0xffff88804bcdb528

I'm not sure what's going on here, but we're clearly turning COW daxpages
back into single-mapping daxpages. I'm not sure what's going on for the
cases where we're replacing one mapping with another. My dimwitted
guess is that dax_fault_is_cow() is incorrectly returning false in some
cases.

Replacing the contents of that function with:

if (iter->srcmap.type != IOMAP_HOLE)
return true;
if (iter->iomap.flags & IOMAP_F_SHARED)
return true;
return false;

Doesn't make the errors go away. Curiously, replacing the entire
function body with "return true;" fixes /that/ problem though...

I am looking into this error by adding debug message too. I found that
testcases which execute fsstress will randomly occur this error. I'm
guessing some concurrent operations caused the cow flag (returned by
dax_fault_is_cow()) to be incorrectly judged. But still haven't catch the
exactly operation yet.

I have an offhand guess that the following sequence might reproduce it:

<write data to a fsdax file1>

<process 1 maps file1, which sets up the pagecache mapping to pmem, and
goes to sleep>

cp --reflink=always file1 file2

<process 2 maps file2, which tries to map the same pmem "page" into
file2's pagecache and trips over page->mapping already being set to
file1's pagecache>

But I dunno, I haven't had much time for digging into this one.

I tried this sequence on v6.0 but it didn't reproduce the warning.


...but generic/649 still fails with things like:

[ 571.224285] run fstests generic/649 at 2022-10-13 11:26:59
[ 571.796353] XFS (pmem0): Mounting V5 Filesystem
[ 571.799059] XFS (pmem0): Ending clean mount
[ 572.378624] ------------[ cut here ]------------
[ 572.379598] WARNING: CPU: 1 PID: 48538 at fs/dax.c:930 dax_writeback_mapping_range+0x2f1/0x600

Which comes from this warning in dax_writeback_one:

/*
* A page got tagged dirty in DAX mapping? Something is seriously
* wrong.
*/
if (WARN_ON(!xa_is_value(entry)))
return -EIO;

Help?

Sorry, no time for this yet...

BTW, are these errors still occur when reflink is turned off? (dax on,
reflink off)

Hmm I'll try that later today.

Forgot to tell you that I have tested dax_on&reflink_off with the Kernel Config which you sent to me: no failure case caused by the dmesg warning.


--
Thanks,
Ruan.


--D


--
Thanks,
Ruan.


--D

Modules linked in: xfs nft_chain_nat xt_REDIRECT nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 ipt_REJECT nf_reject_ipv4 xt_tcpudp ip_set_hash_ip ip_set_hash_net xt_set nft_compat ip_set_hash_mac ip_set nf_tables libcrc32c bfq nfnetlink pvpanic_mmio pvpanic nd_pmem dax_pmem nd_btt sch_fq_codel fuse configfs ip_tables x_tables overlay nfsv4 af_packet [last unloaded: scsi_d

CPU: 1 PID: 415317 Comm: fsstress Tainted: G W 6.0.0-rc7-xfsx #rc7 727341edbd0773a36b78b09dab448fa1896eb3a5
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.15.0-1 04/01/2014
RIP: 0010:dax_insert_entry+0x22d/0x320
Code: e0 48 83 c4 20 5b 5d 41 5c 41 5d 41 5e 41 5f c3 48 8b 58 20 48 8d 53 01 e9 62 ff ff ff 48 8b 58 20 48 8d 53 01 e9 4d ff ff ff <0f> 0b e9 6d ff ff ff 31 f6 48 89 ef e8 72 74 12 00 eb a1 83 e0 02
RSP: 0000:ffffc90004693b28 EFLAGS: 00010002
RAX: ffffea0010a20480 RBX: 0000000000000001 RCX: 0000000000000001
RDX: ffffea0000000000 RSI: 0000000000000033 RDI: ffffea0010a204c0
RBP: ffffc90004693c08 R08: 0000000000000000 R09: 0000000000000000
R10: ffff88800c226228 R11: 0000000000000001 R12: 0000000000000011
R13: ffff88800c226228 R14: ffffc90004693e08 R15: 0000000000000000
FS: 00007f3aad8db740(0000) GS:ffff88803ed00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007f3aad8d1000 CR3: 0000000043104003 CR4: 00000000001706e0
Call Trace:
<TASK>
dax_fault_iter+0x26e/0x670
dax_iomap_pte_fault+0x1ab/0x3e0
__xfs_filemap_fault+0x32f/0x5a0 [xfs c617487f99e14abfa5deb24e923415b927df3d4b]
__do_fault+0x30/0x1e0
do_fault+0x316/0x6d0
? mmap_region+0x2a5/0x620
__handle_mm_fault+0x649/0x1250
handle_mm_fault+0xc1/0x220
do_user_addr_fault+0x1ac/0x610
? _copy_to_user+0x63/0x80
exc_page_fault+0x63/0x130
asm_exc_page_fault+0x22/0x30
RIP: 0033:0x7f3aada7f1ca
Code: c5 fe 7f 07 c5 fe 7f 47 20 c5 fe 7f 47 40 c5 fe 7f 47 60 c5 f8 77 c3 66 0f 1f 84 00 00 00 00 00 40 0f b6 c6 48 89 d1 48 89 fa <f3> aa 48 89 d0 c5 f8 77 c3 66 66 2e 0f 1f 84 00 00 00 00 00 66 90
RSP: 002b:00007ffe47afa688 EFLAGS: 00010206
RAX: 000000000000002e RBX: 0000000000033000 RCX: 000000000000999c
RDX: 00007f3aad8d1000 RSI: 000000000000002e RDI: 00007f3aad8d1000
RBP: 0000558851e13240 R08: 0000000000000000 R09: 0000000000033000
R10: 0000000000000008 R11: 0000000000000246 R12: 028f5c28f5c28f5c
R13: 8f5c28f5c28f5c29 R14: 000000000000999c R15: 0000000000001c81
</TASK>
---[ end trace 0000000000000000 ]---
XFS (pmem0): Unmounting Filesystem
XFS (pmem1): EXPERIMENTAL online scrub feature in use. Use at your own risk!
XFS (pmem1): *** REPAIR SUCCESS ino 0x80 type probe agno 0x0 inum 0x0 gen 0x0 flags 0x80000001 error 0
XFS (pmem1): Unmounting Filesystem
XFS (pmem1): EXPERIMENTAL Large extent counts feature in use. Use at your own risk!
XFS (pmem1): Mounting V5 Filesystem
XFS (pmem1): Ending clean mount
XFS (pmem1): Unmounting Filesystem

Yup, that's the same as what I'm seeing.

Could you send me your kernel config (or other configs needed for the test)?
I still cannot reproduce this warning when reflink is off, even without this
drop patch. Maybe something different in config file?


PS: I specifically tried the two cases Darrick mentioned (on v6.0-rc6):

[root@f33 xfstests-dev]# mkfs.xfs -m reflink=0,rmapbt=1 /dev/pmem0.1 -f
meta-data=/dev/pmem0.1 isize=512 agcount=4, agsize=257920 blks
= sectsz=4096 attr=2, projid32bit=1
= crc=1 finobt=1, sparse=1, rmapbt=1
= reflink=0 bigtime=1 inobtcount=1
nrext64=0
data = bsize=4096 blocks=1031680, imaxpct=25
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0, ftype=1
log =internal log bsize=4096 blocks=16384, version=2
= sectsz=4096 sunit=1 blks, lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0
[root@f33 xfstests-dev]# mkfs.xfs -m reflink=0,rmapbt=1 /dev/pmem0 -f
meta-data=/dev/pmem0 isize=512 agcount=4, agsize=257920 blks
= sectsz=4096 attr=2, projid32bit=1
= crc=1 finobt=1, sparse=1, rmapbt=1
= reflink=0 bigtime=1 inobtcount=1
nrext64=0
data = bsize=4096 blocks=1031680, imaxpct=25
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0, ftype=1
log =internal log bsize=4096 blocks=16384, version=2
= sectsz=4096 sunit=1 blks, lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0
[root@f33 xfstests-dev]# ./check xfs/013 xfs/517
FSTYP -- xfs (debug)
PLATFORM -- Linux/x86_64 f33 6.0.0-rc6 #84 SMP PREEMPT_DYNAMIC Wed Sep
28 18:27:33 CST 2022
MKFS_OPTIONS -- -f -m reflink=0,rmapbt=1 /dev/pmem0.1
MOUNT_OPTIONS -- -o dax -o context=system_u:object_r:root_t:s0 /dev/pmem0.1
/mnt/scratch

xfs/013 127s ... 166s
xfs/517 66s ... 66s
Ran: xfs/013 xfs/517
Passed all 2 tests

I'm not sure what exactly is going weird here -- I tried it on my dev
machine just now and it passed, but the similarly configured testcloud
failed it last night.

FSTYP -- xfs (debug)
PLATFORM -- Linux/x86_64 ca-nfsdev6-mtr03 6.0.0-rc7-xfsx #rc7 SMP
PREEMPT_DYNAMIC Wed Sep 28 15:35:58 PDT 2022
MKFS_OPTIONS -- -f -m reflink=0, -d daxinherit=1, /dev/pmem1
MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/pmem1 /opt

Note that I use libvirt to configure pmem in the VMs. This is an
excerpt of the end of domain xml file:

<memory model='nvdimm' access='shared'>
<source>
<path>/run/mtrdisk/g.mem</path>
</source>
<target>
<size unit='KiB'>21104640</size>
<node>0</node>
</target>
<address type='dimm' slot='0'/>
</memory>
<memory model='nvdimm' access='shared'>
<source>
<path>/run/mtrdisk/h.mem</path>
</source>
<target>
<size unit='KiB'>21104640</size>
<node>1</node>
</target>
<address type='dimm' slot='1'/>
</memory>
</devices>
</domain>

--D