On Sat, Sep 09, 2017 at 05:10:26PM +0000, Soccer Liu wrote:
> Hi:
> ? I was surprised to see a few write operations called into
> ext4_dax_huge_fault with FAULT_FLAG_WRITE set on vmf->flags
> Are those writes expected at all?
> static int ext4_dax_huge_fault(struct vm_fault *vmf,
> enum page_entry_size pe_size)
> {
> ...
> bool write = vmf->flags & FAULT_FLAG_WRITE;
> Based on the following Mount documentation below, I did mount this ext fs
> with ro and noloadAre there any other options that I could use to totally
> eliminate those unexpected writes?
>
> - -r, --read-only
> - Mount the filesystem read-only. A synonym is -o ro.Note that,
> depending on the filesystem type, state and kernel behavior, the
> system may still write to the device. For example, Ext3 or ext4 will
> replay itsjournal if the filesystem is dirty. To prevent this kind of
> write access, you may want to mount ext3 or ext4 filesystem with
> "ro,noload" mount options or setthe block device to read-only mode,
> see command blockdev(8). -
>
> Related DMESG output...
>
> [??? 0.891653] EXT4-fs (pmem0): DAX enabled. Warning: EXPERIMENTAL, use at your own risk
> [??? 0.892736] EXT4-fs (pmem0): mounted filesystem without journal. Opts: noload,dax
> [??? 0.898293] EXT4-fs (sda): mounted filesystem without journal. Opts: (null)
> [??? 0.902574] random: fast init done
> [??? 0.926157] IPVS: ftp: loaded support on port[0] = 21
> // My private debug messages
> [??? 0.997224] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2 <--read-only device?
> [??? 0.998052] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
> [??? 0.998787] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
> [??? 0.999720] EXT4-fs (pmem0): write is 0 handle = 0x0
> [??? 1.004626] EXT4-fs (pmem0): write is 0 handle = 0x0
> [??? 1.005359] EXT4-fs (pmem0): write is 0 handle = 0x0
> [??? 1.006128] EXT4-fs (pmem0): write is 0 handle = 0x0
> [??? 1.006824] EXT4-fs (pmem0): write is 0 handle = 0x0
> [??? 1.007499] EXT4-fs (pmem0): write is 0 handle = 0x0
> [??? 1.008149] EXT4-fs (pmem0): write is 0 handle = 0x0
> [??? 7.320626] random: crng init done
> [?? 18.780447] EXT4-fs: 53 callbacks suppressed
> [?? 18.781900] EXT4-fs (pmem0): write is 0 handle = 0x0
> [?? 18.783392] EXT4-fs (pmem0): write is 0 handle = 0x0
> [?? 18.790661] EXT4-fs (pmem0): write is 0 handle = 0x0
> [?? 18.792042] EXT4-fs (pmem0): write is 0 handle = 0x0
> [?? 18.793321] EXT4-fs (pmem0): write is 0 handle = 0x0
> [?? 18.795005] EXT4-fs (pmem0): write is 0 handle = 0x0
> [?? 18.795204] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
> [?? 18.795209] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
> [?? 18.795213] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
> [?? 18.795257] EXT4-fs (pmem0): write is 0 handle = 0x0
Adding the ext4 list.
I'm pretty sure that this is ext4/mm behavior that is independent of DAX. DAX
doesn't really have any impact on whether or not write page faults happen on a
RO ext4 filesystem, though this fix from Randy Dodgen:
https://www.spinics.net/lists/linux-ext4/msg58075.html
shows that FAULT_FLAG_WRITE can be set on DAX read-only ext4 mounts when
executing binaries, though in this case the write is going to a COW page.
- Ross
On Sep 11, 2017, at 12:21 PM, Ross Zwisler <[email protected]> wrote:
>
> On Sat, Sep 09, 2017 at 05:10:26PM +0000, Soccer Liu wrote:
>> Hi:
>> I was surprised to see a few write operations called into
>> ext4_dax_huge_fault with FAULT_FLAG_WRITE set on vmf->flags
>> Are those writes expected at all?
>> static int ext4_dax_huge_fault(struct vm_fault *vmf,
>> enum page_entry_size pe_size)
>> {
>> ...
>> bool write = vmf->flags & FAULT_FLAG_WRITE;
>> Based on the following Mount documentation below, I did mount this ext fs
>> with ro and noloadAre there any other options that I could use to totally
>> eliminate those unexpected writes?
>>
>> - -r, --read-only
>> - Mount the filesystem read-only. A synonym is -o ro.Note that,
>> depending on the filesystem type, state and kernel behavior, the
>> system may still write to the device. For example, Ext3 or ext4 will
>> replay itsjournal if the filesystem is dirty. To prevent this kind of
>> write access, you may want to mount ext3 or ext4 filesystem with
>> "ro,noload" mount options or setthe block device to read-only mode,
>> see command blockdev(8). -
>>
>> Related DMESG output...
>>
>> [ 0.891653] EXT4-fs (pmem0): DAX enabled. Warning: EXPERIMENTAL, use at your own risk
>> [ 0.892736] EXT4-fs (pmem0): mounted filesystem without journal. Opts: noload,dax
>> [ 0.898293] EXT4-fs (sda): mounted filesystem without journal. Opts: (null)
>> [ 0.902574] random: fast init done
>> [ 0.926157] IPVS: ftp: loaded support on port[0] = 21
>> // My private debug messages
>> [ 0.997224] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2 <--read-only device
>> [ 0.998052] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
>> [ 0.998787] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
>> [ 0.999720] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 1.004626] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 1.005359] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 1.006128] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 1.006824] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 1.007499] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 1.008149] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 7.320626] random: crng init done
>> [ 18.780447] EXT4-fs: 53 callbacks suppressed
>> [ 18.781900] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 18.783392] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 18.790661] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 18.792042] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 18.793321] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 18.795005] EXT4-fs (pmem0): write is 0 handle = 0x0
>> [ 18.795204] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
>> [ 18.795209] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
>> [ 18.795213] EXT4-fs (pmem0): write is 1 handle = 0xffffffffffffffe2
>> [ 18.795257] EXT4-fs (pmem0): write is 0 handle = 0x0
>
> Adding the ext4 list.
>
> I'm pretty sure that this is ext4/mm behavior that is independent of DAX. DAX
> doesn't really have any impact on whether or not write page faults happen on a
> RO ext4 filesystem, though this fix from Randy Dodgen:
>
> https://www.spinics.net/lists/linux-ext4/msg58075.html
>
> shows that FAULT_FLAG_WRITE can be set on DAX read-only ext4 mounts when
> executing binaries, though in this case the write is going to a COW page.
Also, ext4 journal recovery will be done on the filesystem regardless of
whether it is mounted read-only or not.
Cheers, Andreas
On Mon, Sep 11, 2017 at 02:31:41PM -0600, Andreas Dilger wrote:
> On Sep 11, 2017, at 12:21 PM, Ross Zwisler wrote:
>>
>> On Sat, Sep 09, 2017 at 05:10:26PM +0000, Soccer Liu wrote:
>>> [snip]
>>>
>>> Based on the following Mount documentation below, I did mount this ext fs
>>> with ro and noloadAre there any other options that I could use to totally
>>> eliminate those unexpected writes?
>
> [snip]
>
> Also, ext4 journal recovery will be done on the filesystem regardless of
> whether it is mounted read-only or not.
I thought the noload option should prevent journal recovery and
thus be truly read-only. At least it worked on a read-only device
for me.
Regards
Simon
--
+ privacy is necessary
+ using gnupg http://gnupg.org
+ public key id: 0x92FEFDB7E44C32F9
On Mon 11-09-17 12:21:12, Ross Zwisler wrote:
> On Sat, Sep 09, 2017 at 05:10:26PM +0000, Soccer Liu wrote:
> > Hi:
> > ? I was surprised to see a few write operations called into
> > ext4_dax_huge_fault with FAULT_FLAG_WRITE set on vmf->flags
> > Are those writes expected at all?
> > static int ext4_dax_huge_fault(struct vm_fault *vmf,
> > enum page_entry_size pe_size)
> > {
> > ...
> > bool write = vmf->flags & FAULT_FLAG_WRITE;
> > Based on the following Mount documentation below, I did mount this ext fs
> > with ro and noloadAre there any other options that I could use to totally
> > eliminate those unexpected writes?
Well, I expect these are actually 4KB faults into private mapping (they
would end up using ext4_dax_huge_fault() as well) and the fix Ross
references below should fix that... But please tell if it does not (along
with a reproducer). Thanks!
> Adding the ext4 list.
>
> I'm pretty sure that this is ext4/mm behavior that is independent of DAX. DAX
> doesn't really have any impact on whether or not write page faults happen on a
> RO ext4 filesystem, though this fix from Randy Dodgen:
>
> https://www.spinics.net/lists/linux-ext4/msg58075.html
>
> shows that FAULT_FLAG_WRITE can be set on DAX read-only ext4 mounts when
> executing binaries, though in this case the write is going to a COW page.
Honza
--
Jan Kara <[email protected]>
SUSE Labs, CR