diff mbox series

[v2,5/5] btrfs: fiemap: return extent physical size

Message ID 93686d5c4467befe12f76e4921bfc20a13a74e2d.1711588701.git.sweettea-kernel@dorminy.me (mailing list archive)
State New, archived
Headers show
Series fiemap extension for more physical information | expand

Commit Message

Sweet Tea Dorminy March 28, 2024, 1:22 a.m. UTC
Now that fiemap allows returning extent physical size, make btrfs return
the appropriate extent's actual disk size.

Signed-off-by: Sweet Tea Dorminy <sweettea-kernel@dorminy.me>
---
 fs/btrfs/extent_io.c | 70 ++++++++++++++++++++++++++++----------------
 1 file changed, 45 insertions(+), 25 deletions(-)

Comments

Qu Wenruo March 31, 2024, 9:03 a.m. UTC | #1
在 2024/3/28 11:52, Sweet Tea Dorminy 写道:
> Now that fiemap allows returning extent physical size, make btrfs return
> the appropriate extent's actual disk size.
> 
> Signed-off-by: Sweet Tea Dorminy <sweettea-kernel@dorminy.me>
[...]
> @@ -3221,7 +3239,9 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
>   
>   			ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
>   						 disk_bytenr + extent_offset,
> -						 extent_len, flags);
> +						 extent_len,
> +						 disk_size - extent_offset,

This means, we will emit a entry that uses the end to the physical 
extent end.

Considering a file layout like this:

	item 6 key (257 EXTENT_DATA 0) itemoff 15816 itemsize 53
		generation 7 type 1 (regular)
		extent data disk byte 13631488 nr 65536
		extent data offset 0 nr 4096 ram 65536
		extent compression 0 (none)
	item 7 key (257 EXTENT_DATA 4096) itemoff 15763 itemsize 53
		generation 8 type 1 (regular)
		extent data disk byte 13697024 nr 4096
		extent data offset 0 nr 4096 ram 4096
		extent compression 0 (none)
	item 8 key (257 EXTENT_DATA 8192) itemoff 15710 itemsize 53
		generation 7 type 1 (regular)
		extent data disk byte 13631488 nr 65536
		extent data offset 8192 nr 57344 ram 65536
		extent compression 0 (none)

For fiemap, we would got something like this:

fileoff 0, logical len 4k, phy 13631488, phy len 64K
fileoff 4k, logical len 4k, phy 13697024, phy len 4k
fileoff 8k, logical len 56k, phy 13631488 + 8k, phylen 56k

[HOW TO CALCULATE WASTED SPACE IN USER SPACE]
My concern is on the first entry. It indicates that we have wasted 60K 
(phy len is 64K, while logical len is only 4K)

But that information is not correct, as in reality we only wasted 4K, 
the remaining 56K is still referred by file range [8K, 64K).

Do you mean that user space program should maintain a mapping of each 
utilized physical range, and when handling the reported file range [8K, 
64K), the user space program should find that the physical range covers 
with one existing extent, and do calculation correctly?

[COMPRESSION REPRESENTATION]
The biggest problem other than the complexity in user space is the 
handling of compressed extents.

Should we return the physical bytenr (disk_bytenr of file extent item) 
directly or with the extent offset added?
Either way it doesn't look consistent to me, compared to non-compressed 
extents.

[ALTERNATIVE FORMAT]
The other alternative would be following the btrfs ondisk format, 
providing a unique physical bytenr for any file extent, then the 
offset/referred length inside the uncompressed extent.

That would handle compressed and regular extents more consistent, and a 
little easier for user space tool to handle (really just a tiny bit 
easier, no range overlap check needed), but more complex to represent, 
and I'm not sure if any other filesystem would be happy to accept the 
extra members they don't care.

Thanks,
Qu

> +						 flags);
>   		}
>   
>   		if (ret < 0) {
> @@ -3259,7 +3279,7 @@ int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
>   		prev_extent_end = range_end;
>   	}
>   
> -	if (cache.cached && cache.offset + cache.len >= last_extent_end) {
> +	if (cache.cached && cache.offset + cache.log_len >= last_extent_end) {
>   		const u64 i_size = i_size_read(&inode->vfs_inode);
>   
>   		if (prev_extent_end < i_size) {
Sweet Tea Dorminy April 3, 2024, 5:32 a.m. UTC | #2
On 3/31/24 05:03, Qu Wenruo wrote:
> 
> 
> 在 2024/3/28 11:52, Sweet Tea Dorminy 写道:
>> Now that fiemap allows returning extent physical size, make btrfs return
>> the appropriate extent's actual disk size.
>>
>> Signed-off-by: Sweet Tea Dorminy <sweettea-kernel@dorminy.me>
> [...]
>> @@ -3221,7 +3239,9 @@ int extent_fiemap(struct btrfs_inode *inode, 
>> struct fiemap_extent_info *fieinfo,
>>               ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
>>                            disk_bytenr + extent_offset,
>> -                         extent_len, flags);
>> +                         extent_len,
>> +                         disk_size - extent_offset,
> 
> This means, we will emit a entry that uses the end to the physical 
> extent end.
> 
> Considering a file layout like this:
> 
>      item 6 key (257 EXTENT_DATA 0) itemoff 15816 itemsize 53
>          generation 7 type 1 (regular)
>          extent data disk byte 13631488 nr 65536
>          extent data offset 0 nr 4096 ram 65536
>          extent compression 0 (none)
>      item 7 key (257 EXTENT_DATA 4096) itemoff 15763 itemsize 53
>          generation 8 type 1 (regular)
>          extent data disk byte 13697024 nr 4096
>          extent data offset 0 nr 4096 ram 4096
>          extent compression 0 (none)
>      item 8 key (257 EXTENT_DATA 8192) itemoff 15710 itemsize 53
>          generation 7 type 1 (regular)
>          extent data disk byte 13631488 nr 65536
>          extent data offset 8192 nr 57344 ram 65536
>          extent compression 0 (none)
> 
> For fiemap, we would got something like this:
> 
> fileoff 0, logical len 4k, phy 13631488, phy len 64K
> fileoff 4k, logical len 4k, phy 13697024, phy len 4k
> fileoff 8k, logical len 56k, phy 13631488 + 8k, phylen 56k
> 
> [HOW TO CALCULATE WASTED SPACE IN USER SPACE]
> My concern is on the first entry. It indicates that we have wasted 60K 
> (phy len is 64K, while logical len is only 4K)
> 
> But that information is not correct, as in reality we only wasted 4K, 
> the remaining 56K is still referred by file range [8K, 64K).
> 
> Do you mean that user space program should maintain a mapping of each 
> utilized physical range, and when handling the reported file range [8K, 
> 64K), the user space program should find that the physical range covers 
> with one existing extent, and do calculation correctly?

My goal is to give an unprivileged interface for tools like compsize to 
figure out how much space is used by a particular set of files. They 
report the total disk space referenced by the provided list of files, 
currently by doing a tree search (CAP_SYS_ADMIN) for all the extents 
pertaining to the requested files and deduplicating extents based on 
disk_bytenr.

It seems simplest to me for userspace for the kernel to emit the entire 
extent for each part of it referenced in a file, and let userspace deal 
with deduplicating extents. This is also most similar to the existing 
tree-search based interface. Reporting whole extents gives more 
flexibility for userspace to figure out how to report bookend extents, 
or shared extents, or ...

It does seem a little weird where if you request with fiemap only e.g. 
4k-16k range in that example file you'll get reported all 68k involved, 
but I can't figure out a way to fix that without having the kernel keep 
track of used parts of the extents as part of reporting, which sounds 
expensive.

You're right that I'm being inconsistent, taking off extent_offset from 
the reported disk size when that isn't what I should be doing, so I 
fixed that in v3.

> 
> [COMPRESSION REPRESENTATION]
> The biggest problem other than the complexity in user space is the 
> handling of compressed extents.
> 
> Should we return the physical bytenr (disk_bytenr of file extent item) 
> directly or with the extent offset added?
> Either way it doesn't look consistent to me, compared to non-compressed 
> extents.
> 

As I understand it, the goal of reporting physical bytenr is to provide 
a number which we could theoretically then resolve into a disk location 
or few if we cared, but which doesn't necessarily have any physical 
meaning. To quote the fiemap documentation page: "It is always undefined 
to try to update the data in-place by writing to the indicated location 
without the assistance of the filesystem". So I think I'd prefer to 
always report the entire size of the entire extent being referenced.

> [ALTERNATIVE FORMAT]
> The other alternative would be following the btrfs ondisk format, 
> providing a unique physical bytenr for any file extent, then the 
> offset/referred length inside the uncompressed extent.
> 
> That would handle compressed and regular extents more consistent, and a 
> little easier for user space tool to handle (really just a tiny bit 
> easier, no range overlap check needed), but more complex to represent, 
> and I'm not sure if any other filesystem would be happy to accept the 
> extra members they don't care.

I really want to make sure that this interface reports the unused space 
in e.g bookend extents well -- compsize has been an important tool for 
me in this respect, e.g. a time when a 10g file was taking up 110g of 
actual disk space. If we report the entire length of the entire extent, 
then when used on whole files one can establish the space referenced by 
that file but not used; similarly on multiple files. So while I like the 
simplicity of just reporting the used length, I don't think there's a 
way to make compsize unprivileged with that approach.

Thank you!!
Qu Wenruo April 3, 2024, 5:52 a.m. UTC | #3
在 2024/4/3 16:02, Sweet Tea Dorminy 写道:
[...]
>>
>> fileoff 0, logical len 4k, phy 13631488, phy len 64K
>> fileoff 4k, logical len 4k, phy 13697024, phy len 4k
>> fileoff 8k, logical len 56k, phy 13631488 + 8k, phylen 56k
>>
>> [HOW TO CALCULATE WASTED SPACE IN USER SPACE]
>> My concern is on the first entry. It indicates that we have wasted 60K 
>> (phy len is 64K, while logical len is only 4K)
>>
>> But that information is not correct, as in reality we only wasted 4K, 
>> the remaining 56K is still referred by file range [8K, 64K).
>>
>> Do you mean that user space program should maintain a mapping of each 
>> utilized physical range, and when handling the reported file range 
>> [8K, 64K), the user space program should find that the physical range 
>> covers with one existing extent, and do calculation correctly?
> 
> My goal is to give an unprivileged interface for tools like compsize to 
> figure out how much space is used by a particular set of files. They 
> report the total disk space referenced by the provided list of files, 
> currently by doing a tree search (CAP_SYS_ADMIN) for all the extents 
> pertaining to the requested files and deduplicating extents based on 
> disk_bytenr.
> 
> It seems simplest to me for userspace for the kernel to emit the entire 
> extent for each part of it referenced in a file, and let userspace deal 
> with deduplicating extents. This is also most similar to the existing 
> tree-search based interface. Reporting whole extents gives more 
> flexibility for userspace to figure out how to report bookend extents, 
> or shared extents, or ...

That's totally fine, no matter what solution you go, (reporting exactly 
as the on-disk file extent, or with offset into consideration), user 
space always need to maintain some type of mapping to calculate the 
wasted space by bookend extents.

> 
> It does seem a little weird where if you request with fiemap only e.g. 
> 4k-16k range in that example file you'll get reported all 68k involved, 
> but I can't figure out a way to fix that without having the kernel keep 
> track of used parts of the extents as part of reporting, which sounds 
> expensive.

I do not think mapping 4k-16K is a common scenario either, but since you 
mentioned, at least we need a consistent way to emit a filemap entry.

The tracking part can be done in the user space.

> 
> You're right that I'm being inconsistent, taking off extent_offset from 
> the reported disk size when that isn't what I should be doing, so I 
> fixed that in v3.
> 
>>
>> [COMPRESSION REPRESENTATION]
>> The biggest problem other than the complexity in user space is the 
>> handling of compressed extents.
>>
>> Should we return the physical bytenr (disk_bytenr of file extent item) 
>> directly or with the extent offset added?
>> Either way it doesn't look consistent to me, compared to 
>> non-compressed extents.
>>
> 
> As I understand it, the goal of reporting physical bytenr is to provide 
> a number which we could theoretically then resolve into a disk location 
> or few if we cared, but which doesn't necessarily have any physical 
> meaning. To quote the fiemap documentation page: "It is always undefined 
> to try to update the data in-place by writing to the indicated location 
> without the assistance of the filesystem". So I think I'd prefer to 
> always report the entire size of the entire extent being referenced.

The concern is, if we have a compressed file extent, reflinked to 
different part of the file.

Then the fiemap returns all different physical bytenr (since offset is 
added), user space tool have no idea they are the same extent on-disk.
Furthermore, if we emit the physical + offset directly to user space 
(which can be beyond the compressed extent), then we also have another 
uncompressed extent at previous physical + offset.

Would that lead to bad calculation in user space to determine how many 
bytes are really used?

> 
>> [ALTERNATIVE FORMAT]
>> The other alternative would be following the btrfs ondisk format, 
>> providing a unique physical bytenr for any file extent, then the 
>> offset/referred length inside the uncompressed extent.
>>
>> That would handle compressed and regular extents more consistent, and 
>> a little easier for user space tool to handle (really just a tiny bit 
>> easier, no range overlap check needed), but more complex to represent, 
>> and I'm not sure if any other filesystem would be happy to accept the 
>> extra members they don't care.
> 
> I really want to make sure that this interface reports the unused space 
> in e.g bookend extents well -- compsize has been an important tool for 
> me in this respect, e.g. a time when a 10g file was taking up 110g of 
> actual disk space. If we report the entire length of the entire extent, 
> then when used on whole files one can establish the space referenced by 
> that file but not used; similarly on multiple files. So while I like the 
> simplicity of just reporting the used length, I don't think there's a 
> way to make compsize unprivileged with that approach.

Why not? In user space we just need to maintain a mapping of each 
referred range.

Then we get the real actual disk space, meanwhile the fiemap report is 
no different than "btrfs ins dump-tree" for file extents (we have all 
the things we need, filepos, length (num_bytes), disk_bytenr, 
disk_num_bytes, offset, and ram_bytes.

For unused space, since we have the mapping, we can iterate through the 
mapping, finding out all the sectors not referred by any file extents.

It should really just be a fiemap based (and unprivilleged) compsize.
Or did I miss some important things?

Thanks,
Qu

> 
> Thank you!!
Sweet Tea Dorminy April 3, 2024, 6:02 a.m. UTC | #4
>> This means, we will emit a entry that uses the end to the physical 
>> extent end.
>>
>> Considering a file layout like this:
>>
>>      item 6 key (257 EXTENT_DATA 0) itemoff 15816 itemsize 53
>>          generation 7 type 1 (regular)
>>          extent data disk byte 13631488 nr 65536
>>          extent data offset 0 nr 4096 ram 65536
>>          extent compression 0 (none)
>>      item 7 key (257 EXTENT_DATA 4096) itemoff 15763 itemsize 53
>>          generation 8 type 1 (regular)
>>          extent data disk byte 13697024 nr 4096
>>          extent data offset 0 nr 4096 ram 4096
>>          extent compression 0 (none)
>>      item 8 key (257 EXTENT_DATA 8192) itemoff 15710 itemsize 53
>>          generation 7 type 1 (regular)
>>          extent data disk byte 13631488 nr 65536
>>          extent data offset 8192 nr 57344 ram 65536
>>          extent compression 0 (none)
>>
>> For fiemap, we would got something like this:
>>
>> fileoff 0, logical len 4k, phy 13631488, phy len 64K
>> fileoff 4k, logical len 4k, phy 13697024, phy len 4k
>> fileoff 8k, logical len 56k, phy 13631488 + 8k, phylen 56k
>>
>> [HOW TO CALCULATE WASTED SPACE IN USER SPACE]
>> My concern is on the first entry. It indicates that we have wasted 60K 
>> (phy len is 64K, while logical len is only 4K)
>>
>> But that information is not correct, as in reality we only wasted 4K, 
>> the remaining 56K is still referred by file range [8K, 64K).
>>
>> Do you mean that user space program should maintain a mapping of each 
>> utilized physical range, and when handling the reported file range 
>> [8K, 64K), the user space program should find that the physical range 
>> covers with one existing extent, and do calculation correctly?
> 
> My goal is to give an unprivileged interface for tools like compsize to 
> figure out how much space is used by a particular set of files. They 
> report the total disk space referenced by the provided list of files, 
> currently by doing a tree search (CAP_SYS_ADMIN) for all the extents 
> pertaining to the requested files and deduplicating extents based on 
> disk_bytenr.
> 
> It seems simplest to me for userspace for the kernel to emit the entire 
> extent for each part of it referenced in a file, and let userspace deal 
> with deduplicating extents. This is also most similar to the existing 
> tree-search based interface. Reporting whole extents gives more 
> flexibility for userspace to figure out how to report bookend extents, 
> or shared extents, or ...
> 
> It does seem a little weird where if you request with fiemap only e.g. 
> 4k-16k range in that example file you'll get reported all 68k involved, 
> but I can't figure out a way to fix that without having the kernel keep 
> track of used parts of the extents as part of reporting, which sounds 
> expensive.
> 
> You're right that I'm being inconsistent, taking off extent_offset from 
> the reported disk size when that isn't what I should be doing, so I 
> fixed that in v3.

Ah, I think I grasp a point I'd missed before.
- Without setting disk_bytenr to the actual start of the data on disk, 
there's no way to find the location of the actual data on disk within 
the extent from fiemap alone
- But reporting disk_bytenr + offset, to get actual start of data on 
disk, means we need to report a physical size to figure out the end of 
the extent and we can't know the beginning.

We can't convey both actual location, start, and end of the extent in 
just two pieces of information.

On the other hand, if someone really needs to know the actual location 
on disk of their data, they could use the tree_search ioctl as root to 
do so?

So I still think we should be reporting entire extents but am less 
confident that it doesn't break existing users. I am not sure how common 
it is to take fiemap output on btrfs and use it to try to get to 
physical data on disk - do you know of a tool that does so?

Thank you!
Sweet Tea Dorminy April 3, 2024, 7:18 a.m. UTC | #5
On 4/3/24 01:52, Qu Wenruo wrote:
> 
> 
> 在 2024/4/3 16:02, Sweet Tea Dorminy 写道:
> [...]
>>>
>>> fileoff 0, logical len 4k, phy 13631488, phy len 64K
>>> fileoff 4k, logical len 4k, phy 13697024, phy len 4k
>>> fileoff 8k, logical len 56k, phy 13631488 + 8k, phylen 56k
>>>
>>> [HOW TO CALCULATE WASTED SPACE IN USER SPACE]
>>> My concern is on the first entry. It indicates that we have wasted 
>>> 60K (phy len is 64K, while logical len is only 4K)
>>>
>>> But that information is not correct, as in reality we only wasted 4K, 
>>> the remaining 56K is still referred by file range [8K, 64K).
>>>
>>> Do you mean that user space program should maintain a mapping of each 
>>> utilized physical range, and when handling the reported file range 
>>> [8K, 64K), the user space program should find that the physical range 
>>> covers with one existing extent, and do calculation correctly?
>>
>> My goal is to give an unprivileged interface for tools like compsize 
>> to figure out how much space is used by a particular set of files. 
>> They report the total disk space referenced by the provided list of 
>> files, currently by doing a tree search (CAP_SYS_ADMIN) for all the 
>> extents pertaining to the requested files and deduplicating extents 
>> based on disk_bytenr.
>>
>> It seems simplest to me for userspace for the kernel to emit the 
>> entire extent for each part of it referenced in a file, and let 
>> userspace deal with deduplicating extents. This is also most similar 
>> to the existing tree-search based interface. Reporting whole extents 
>> gives more flexibility for userspace to figure out how to report 
>> bookend extents, or shared extents, or ...
> 
> That's totally fine, no matter what solution you go, (reporting exactly 
> as the on-disk file extent, or with offset into consideration), user 
> space always need to maintain some type of mapping to calculate the 
> wasted space by bookend extents.
> 
>>
>> It does seem a little weird where if you request with fiemap only e.g. 
>> 4k-16k range in that example file you'll get reported all 68k 
>> involved, but I can't figure out a way to fix that without having the 
>> kernel keep track of used parts of the extents as part of reporting, 
>> which sounds expensive.
> 
> I do not think mapping 4k-16K is a common scenario either, but since you 
> mentioned, at least we need a consistent way to emit a filemap entry.
> 
> The tracking part can be done in the user space.
> 
>>
>> You're right that I'm being inconsistent, taking off extent_offset 
>> from the reported disk size when that isn't what I should be doing, so 
>> I fixed that in v3.
>>
>>>
>>> [COMPRESSION REPRESENTATION]
>>> The biggest problem other than the complexity in user space is the 
>>> handling of compressed extents.
>>>
>>> Should we return the physical bytenr (disk_bytenr of file extent 
>>> item) directly or with the extent offset added?
>>> Either way it doesn't look consistent to me, compared to 
>>> non-compressed extents.
>>>
>>
>> As I understand it, the goal of reporting physical bytenr is to 
>> provide a number which we could theoretically then resolve into a disk 
>> location or few if we cared, but which doesn't necessarily have any 
>> physical meaning. To quote the fiemap documentation page: "It is 
>> always undefined to try to update the data in-place by writing to the 
>> indicated location without the assistance of the filesystem". So I 
>> think I'd prefer to always report the entire size of the entire extent 
>> being referenced.
> 
> The concern is, if we have a compressed file extent, reflinked to 
> different part of the file.
> 
> Then the fiemap returns all different physical bytenr (since offset is 
> added), user space tool have no idea they are the same extent on-disk.
> Furthermore, if we emit the physical + offset directly to user space 
> (which can be beyond the compressed extent), then we also have another 
> uncompressed extent at previous physical + offset.
> 
> Would that lead to bad calculation in user space to determine how many 
> bytes are really used?
> 
>>
>>> [ALTERNATIVE FORMAT]
>>> The other alternative would be following the btrfs ondisk format, 
>>> providing a unique physical bytenr for any file extent, then the 
>>> offset/referred length inside the uncompressed extent.
>>>
>>> That would handle compressed and regular extents more consistent, and 
>>> a little easier for user space tool to handle (really just a tiny bit 
>>> easier, no range overlap check needed), but more complex to 
>>> represent, and I'm not sure if any other filesystem would be happy to 
>>> accept the extra members they don't care.
>>
>> I really want to make sure that this interface reports the unused 
>> space in e.g bookend extents well -- compsize has been an important 
>> tool for me in this respect, e.g. a time when a 10g file was taking up 
>> 110g of actual disk space. If we report the entire length of the 
>> entire extent, then when used on whole files one can establish the 
>> space referenced by that file but not used; similarly on multiple 
>> files. So while I like the simplicity of just reporting the used 
>> length, I don't think there's a way to make compsize unprivileged with 
>> that approach.
> 
> Why not? In user space we just need to maintain a mapping of each 
> referred range.
> 
> Then we get the real actual disk space, meanwhile the fiemap report is 
> no different than "btrfs ins dump-tree" for file extents (we have all 
> the things we need, filepos, length (num_bytes), disk_bytenr, 
> disk_num_bytes, offset, and ram_bytes

The fiemap output (in this changeset) has equivalents of filepos, 
length; disk_bytenr + offset, disk_num_bytes - offset -- we don't get 
ram_bytes and we get two computed values from the three dump-tree outputs.
If it were reporting whole extents, it'd be disk_bytenr, disk_num_bytes, 
and we'd be missing offset.
I think we'd need a third piece of information about physical space in 
order to convey all three equivalents of disk_bytenr, disk_num_bytes, 
offset. And without that third piece of information, we can't both match 
up disk extents and also know exactly what disk bytenr data is stored 
at, I think? But maybe you're proposing exactly that, having a third number?
Qu Wenruo April 3, 2024, 7:19 a.m. UTC | #6
在 2024/4/3 16:32, Sweet Tea Dorminy 写道:
>>> This means, we will emit a entry that uses the end to the physical
>>> extent end.
>>>
>>> Considering a file layout like this:
>>>
>>>      item 6 key (257 EXTENT_DATA 0) itemoff 15816 itemsize 53
>>>          generation 7 type 1 (regular)
>>>          extent data disk byte 13631488 nr 65536
>>>          extent data offset 0 nr 4096 ram 65536
>>>          extent compression 0 (none)
>>>      item 7 key (257 EXTENT_DATA 4096) itemoff 15763 itemsize 53
>>>          generation 8 type 1 (regular)
>>>          extent data disk byte 13697024 nr 4096
>>>          extent data offset 0 nr 4096 ram 4096
>>>          extent compression 0 (none)
>>>      item 8 key (257 EXTENT_DATA 8192) itemoff 15710 itemsize 53
>>>          generation 7 type 1 (regular)
>>>          extent data disk byte 13631488 nr 65536
>>>          extent data offset 8192 nr 57344 ram 65536
>>>          extent compression 0 (none)
>>>
>>> For fiemap, we would got something like this:
>>>
>>> fileoff 0, logical len 4k, phy 13631488, phy len 64K
>>> fileoff 4k, logical len 4k, phy 13697024, phy len 4k
>>> fileoff 8k, logical len 56k, phy 13631488 + 8k, phylen 56k
>>>
>>> [HOW TO CALCULATE WASTED SPACE IN USER SPACE]
>>> My concern is on the first entry. It indicates that we have wasted
>>> 60K (phy len is 64K, while logical len is only 4K)
>>>
>>> But that information is not correct, as in reality we only wasted 4K,
>>> the remaining 56K is still referred by file range [8K, 64K).
>>>
>>> Do you mean that user space program should maintain a mapping of each
>>> utilized physical range, and when handling the reported file range
>>> [8K, 64K), the user space program should find that the physical range
>>> covers with one existing extent, and do calculation correctly?
>>
>> My goal is to give an unprivileged interface for tools like compsize
>> to figure out how much space is used by a particular set of files.
>> They report the total disk space referenced by the provided list of
>> files, currently by doing a tree search (CAP_SYS_ADMIN) for all the
>> extents pertaining to the requested files and deduplicating extents
>> based on disk_bytenr.
>>
>> It seems simplest to me for userspace for the kernel to emit the
>> entire extent for each part of it referenced in a file, and let
>> userspace deal with deduplicating extents. This is also most similar
>> to the existing tree-search based interface. Reporting whole extents
>> gives more flexibility for userspace to figure out how to report
>> bookend extents, or shared extents, or ...
>>
>> It does seem a little weird where if you request with fiemap only e.g.
>> 4k-16k range in that example file you'll get reported all 68k
>> involved, but I can't figure out a way to fix that without having the
>> kernel keep track of used parts of the extents as part of reporting,
>> which sounds expensive.
>>
>> You're right that I'm being inconsistent, taking off extent_offset
>> from the reported disk size when that isn't what I should be doing, so
>> I fixed that in v3.
>
> Ah, I think I grasp a point I'd missed before.
> - Without setting disk_bytenr to the actual start of the data on disk,
> there's no way to find the location of the actual data on disk within
> the extent from fiemap alone

Yes, that's my point.

> - But reporting disk_bytenr + offset, to get actual start of data on
> disk, means we need to report a physical size to figure out the end of
> the extent and we can't know the beginning.

disk_bytenr + offset + disk_num_bytes, and with the existing things like
length (aka, num_bytes), filepos (aka, key.offset) flags
(compression/hole/preallocated etc), we have everything we need to know
for regular extents.

For compressed extents, we also need ram_bytes.

If you ask me, I'd say put all the extra members into fiemap entry if we
have the space...

It would be u64 * 4 if we go 1:1 on the file extent items, otherwise we
may cheap on offset and ram_bytes (u32 is enough for btrfs at least), in
that case it would be u64 * 2 + u32 * 2.

But I'm also 100% sure, the extra members would not be welcomed by other
filesystems either.

Thanks,
Qu

>
> We can't convey both actual location, start, and end of the extent in
> just two pieces of information.
>
> On the other hand, if someone really needs to know the actual location
> on disk of their data, they could use the tree_search ioctl as root to
> do so?
>
> So I still think we should be reporting entire extents but am less
> confident that it doesn't break existing users. I am not sure how common
> it is to take fiemap output on btrfs and use it to try to get to
> physical data on disk - do you know of a tool that does so?
>
> Thank you!
>
David Sterba April 9, 2024, 6:52 p.m. UTC | #7
On Wed, Apr 03, 2024 at 05:49:42PM +1030, Qu Wenruo wrote:
> 
> 
> 在 2024/4/3 16:32, Sweet Tea Dorminy 写道:
> >>> This means, we will emit a entry that uses the end to the physical
> >>> extent end.
> >>>
> >>> Considering a file layout like this:
> >>>
> >>>      item 6 key (257 EXTENT_DATA 0) itemoff 15816 itemsize 53
> >>>          generation 7 type 1 (regular)
> >>>          extent data disk byte 13631488 nr 65536
> >>>          extent data offset 0 nr 4096 ram 65536
> >>>          extent compression 0 (none)
> >>>      item 7 key (257 EXTENT_DATA 4096) itemoff 15763 itemsize 53
> >>>          generation 8 type 1 (regular)
> >>>          extent data disk byte 13697024 nr 4096
> >>>          extent data offset 0 nr 4096 ram 4096
> >>>          extent compression 0 (none)
> >>>      item 8 key (257 EXTENT_DATA 8192) itemoff 15710 itemsize 53
> >>>          generation 7 type 1 (regular)
> >>>          extent data disk byte 13631488 nr 65536
> >>>          extent data offset 8192 nr 57344 ram 65536
> >>>          extent compression 0 (none)
> >>>
> >>> For fiemap, we would got something like this:
> >>>
> >>> fileoff 0, logical len 4k, phy 13631488, phy len 64K
> >>> fileoff 4k, logical len 4k, phy 13697024, phy len 4k
> >>> fileoff 8k, logical len 56k, phy 13631488 + 8k, phylen 56k
> >>>
> >>> [HOW TO CALCULATE WASTED SPACE IN USER SPACE]
> >>> My concern is on the first entry. It indicates that we have wasted
> >>> 60K (phy len is 64K, while logical len is only 4K)
> >>>
> >>> But that information is not correct, as in reality we only wasted 4K,
> >>> the remaining 56K is still referred by file range [8K, 64K).
> >>>
> >>> Do you mean that user space program should maintain a mapping of each
> >>> utilized physical range, and when handling the reported file range
> >>> [8K, 64K), the user space program should find that the physical range
> >>> covers with one existing extent, and do calculation correctly?
> >>
> >> My goal is to give an unprivileged interface for tools like compsize
> >> to figure out how much space is used by a particular set of files.
> >> They report the total disk space referenced by the provided list of
> >> files, currently by doing a tree search (CAP_SYS_ADMIN) for all the
> >> extents pertaining to the requested files and deduplicating extents
> >> based on disk_bytenr.
> >>
> >> It seems simplest to me for userspace for the kernel to emit the
> >> entire extent for each part of it referenced in a file, and let
> >> userspace deal with deduplicating extents. This is also most similar
> >> to the existing tree-search based interface. Reporting whole extents
> >> gives more flexibility for userspace to figure out how to report
> >> bookend extents, or shared extents, or ...
> >>
> >> It does seem a little weird where if you request with fiemap only e.g.
> >> 4k-16k range in that example file you'll get reported all 68k
> >> involved, but I can't figure out a way to fix that without having the
> >> kernel keep track of used parts of the extents as part of reporting,
> >> which sounds expensive.
> >>
> >> You're right that I'm being inconsistent, taking off extent_offset
> >> from the reported disk size when that isn't what I should be doing, so
> >> I fixed that in v3.
> >
> > Ah, I think I grasp a point I'd missed before.
> > - Without setting disk_bytenr to the actual start of the data on disk,
> > there's no way to find the location of the actual data on disk within
> > the extent from fiemap alone
> 
> Yes, that's my point.
> 
> > - But reporting disk_bytenr + offset, to get actual start of data on
> > disk, means we need to report a physical size to figure out the end of
> > the extent and we can't know the beginning.
> 
> disk_bytenr + offset + disk_num_bytes, and with the existing things like
> length (aka, num_bytes), filepos (aka, key.offset) flags
> (compression/hole/preallocated etc), we have everything we need to know
> for regular extents.
> 
> For compressed extents, we also need ram_bytes.
> 
> If you ask me, I'd say put all the extra members into fiemap entry if we
> have the space...
> 
> It would be u64 * 4 if we go 1:1 on the file extent items, otherwise we
> may cheap on offset and ram_bytes (u32 is enough for btrfs at least), in
> that case it would be u64 * 2 + u32 * 2.
> 
> But I'm also 100% sure, the extra members would not be welcomed by other
> filesystems either.

That's probably right, too many btrfs-specific information in the
generic FIEMAP, but we may also do our own enhanced fiemap ioctl that
would provide all the information you suggest and we'd be free to put
the compression information there too.
Andreas Dilger April 9, 2024, 7:31 p.m. UTC | #8
On Apr 9, 2024, at 12:52 PM, David Sterba <dsterba@suse.cz> wrote:
> 
> On Wed, Apr 03, 2024 at 05:49:42PM +1030, Qu Wenruo wrote:
>> 
>> 
>> 在 2024/4/3 16:32, Sweet Tea Dorminy 写道:
>>>>> This means, we will emit a entry that uses the end to the physical
>>>>> extent end.
>>>>> 
>>>>> Considering a file layout like this:
>>>>> 
>>>>>      item 6 key (257 EXTENT_DATA 0) itemoff 15816 itemsize 53
>>>>>          generation 7 type 1 (regular)
>>>>>          extent data disk byte 13631488 nr 65536
>>>>>          extent data offset 0 nr 4096 ram 65536
>>>>>          extent compression 0 (none)
>>>>>      item 7 key (257 EXTENT_DATA 4096) itemoff 15763 itemsize 53
>>>>>          generation 8 type 1 (regular)
>>>>>          extent data disk byte 13697024 nr 4096
>>>>>          extent data offset 0 nr 4096 ram 4096
>>>>>          extent compression 0 (none)
>>>>>      item 8 key (257 EXTENT_DATA 8192) itemoff 15710 itemsize 53
>>>>>          generation 7 type 1 (regular)
>>>>>          extent data disk byte 13631488 nr 65536
>>>>>          extent data offset 8192 nr 57344 ram 65536
>>>>>          extent compression 0 (none)
>>>>> 
>>>>> For fiemap, we would got something like this:
>>>>> 
>>>>> fileoff 0, logical len 4k, phy 13631488, phy len 64K
>>>>> fileoff 4k, logical len 4k, phy 13697024, phy len 4k
>>>>> fileoff 8k, logical len 56k, phy 13631488 + 8k, phylen 56k
>>>>> 
>>>>> [HOW TO CALCULATE WASTED SPACE IN USER SPACE]
>>>>> My concern is on the first entry. It indicates that we have wasted
>>>>> 60K (phy len is 64K, while logical len is only 4K)
>>>>> 
>>>>> But that information is not correct, as in reality we only wasted 4K,
>>>>> the remaining 56K is still referred by file range [8K, 64K).
>>>>> 
>>>>> Do you mean that user space program should maintain a mapping of each
>>>>> utilized physical range, and when handling the reported file range
>>>>> [8K, 64K), the user space program should find that the physical range
>>>>> covers with one existing extent, and do calculation correctly?
>>>> 
>>>> My goal is to give an unprivileged interface for tools like compsize
>>>> to figure out how much space is used by a particular set of files.
>>>> They report the total disk space referenced by the provided list of
>>>> files, currently by doing a tree search (CAP_SYS_ADMIN) for all the
>>>> extents pertaining to the requested files and deduplicating extents
>>>> based on disk_bytenr.
>>>> 
>>>> It seems simplest to me for userspace for the kernel to emit the
>>>> entire extent for each part of it referenced in a file, and let
>>>> userspace deal with deduplicating extents. This is also most similar
>>>> to the existing tree-search based interface. Reporting whole extents
>>>> gives more flexibility for userspace to figure out how to report
>>>> bookend extents, or shared extents, or ...
>>>> 
>>>> It does seem a little weird where if you request with fiemap only e.g.
>>>> 4k-16k range in that example file you'll get reported all 68k
>>>> involved, but I can't figure out a way to fix that without having the
>>>> kernel keep track of used parts of the extents as part of reporting,
>>>> which sounds expensive.
>>>> 
>>>> You're right that I'm being inconsistent, taking off extent_offset
>>>> from the reported disk size when that isn't what I should be doing, so
>>>> I fixed that in v3.
>>> 
>>> Ah, I think I grasp a point I'd missed before.
>>> - Without setting disk_bytenr to the actual start of the data on disk,
>>> there's no way to find the location of the actual data on disk within
>>> the extent from fiemap alone
>> 
>> Yes, that's my point.
>> 
>>> - But reporting disk_bytenr + offset, to get actual start of data on
>>> disk, means we need to report a physical size to figure out the end of
>>> the extent and we can't know the beginning.
>> 
>> disk_bytenr + offset + disk_num_bytes, and with the existing things like
>> length (aka, num_bytes), filepos (aka, key.offset) flags
>> (compression/hole/preallocated etc), we have everything we need to know
>> for regular extents.
>> 
>> For compressed extents, we also need ram_bytes.
>> 
>> If you ask me, I'd say put all the extra members into fiemap entry if we
>> have the space...
>> 
>> It would be u64 * 4 if we go 1:1 on the file extent items, otherwise we
>> may cheap on offset and ram_bytes (u32 is enough for btrfs at least), in
>> that case it would be u64 * 2 + u32 * 2.
>> 
>> But I'm also 100% sure, the extra members would not be welcomed by other
>> filesystems either.
> 
> That's probably right, too many btrfs-specific information in the
> generic FIEMAP, but we may also do our own enhanced fiemap ioctl that
> would provide all the information you suggest and we'd be free to put
> the compression information there too.

I read this thread when it was first posted, but I don't understand what
these extra fields actually mean?  Definitely adding the logical/physical
length makes sense for compressed extents, but I didn't see any clear
explanation of what these other fields actually mean?

I'm extrapolating something like btrfs has aggregated compressed chunks
that have multiple independent/disjoint blocks within a chunk, and you
are trying to get the exact offset within the compression byte stream
for the start of each block in the chunk?

Cheers, Andreas
diff mbox series

Patch

diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 30fcbb9393fe..9921dc1567d6 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -2456,7 +2456,8 @@  int try_release_extent_mapping(struct page *page, gfp_t mask)
 struct btrfs_fiemap_entry {
 	u64 offset;
 	u64 phys;
-	u64 len;
+	u64 log_len;
+	u64 phys_len;
 	u32 flags;
 };
 
@@ -2514,7 +2515,8 @@  struct fiemap_cache {
 	/* Fields for the cached extent (unsubmitted, not ready, extent). */
 	u64 offset;
 	u64 phys;
-	u64 len;
+	u64 log_len;
+	u64 phys_len;
 	u32 flags;
 	bool cached;
 };
@@ -2527,8 +2529,8 @@  static int flush_fiemap_cache(struct fiemap_extent_info *fieinfo,
 		int ret;
 
 		ret = fiemap_fill_next_extent(fieinfo, entry->offset,
-					      entry->phys, entry->len, 0,
-					      entry->flags);
+					      entry->phys, entry->log_len,
+					      entry->phys_len, entry->flags);
 		/*
 		 * Ignore 1 (reached max entries) because we keep track of that
 		 * ourselves in emit_fiemap_extent().
@@ -2553,7 +2555,8 @@  static int flush_fiemap_cache(struct fiemap_extent_info *fieinfo,
  */
 static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
 				struct fiemap_cache *cache,
-				u64 offset, u64 phys, u64 len, u32 flags)
+				u64 offset, u64 phys, u64 log_len,
+				u64 phys_len, u32 flags)
 {
 	struct btrfs_fiemap_entry *entry;
 	u64 cache_end;
@@ -2561,6 +2564,9 @@  static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
 	/* Set at the end of extent_fiemap(). */
 	ASSERT((flags & FIEMAP_EXTENT_LAST) == 0);
 
+	/* We always set the correct physical length. */
+	flags |= FIEMAP_EXTENT_HAS_PHYS_LEN;
+
 	if (!cache->cached)
 		goto assign;
 
@@ -2596,7 +2602,7 @@  static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
 	 * or equals to what we have in cache->offset. We deal with this as
 	 * described below.
 	 */
-	cache_end = cache->offset + cache->len;
+	cache_end = cache->offset + cache->log_len;
 	if (cache_end > offset) {
 		if (offset == cache->offset) {
 			/*
@@ -2620,10 +2626,10 @@  static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
 			 * where a previously found file extent item was split
 			 * due to an ordered extent completing.
 			 */
-			cache->len = offset - cache->offset;
+			cache->log_len = offset - cache->offset;
 			goto emit;
 		} else {
-			const u64 range_end = offset + len;
+			const u64 range_end = offset + log_len;
 
 			/*
 			 * The offset of the file extent item we have just found
@@ -2656,11 +2662,13 @@  static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
 			if (range_end <= cache_end)
 				return 0;
 
-			if (!(flags & (FIEMAP_EXTENT_DATA_COMPRESSED | FIEMAP_EXTENT_DELALLOC)))
+			if (!(flags & (FIEMAP_EXTENT_DATA_COMPRESSED | FIEMAP_EXTENT_DELALLOC))) {
 				phys += cache_end - offset;
+				phys_len -= cache_end - offset;
+			}
 
 			offset = cache_end;
-			len = range_end - cache_end;
+			log_len = range_end - cache_end;
 			goto emit;
 		}
 	}
@@ -2670,15 +2678,17 @@  static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
 	 * 1) Their logical addresses are continuous
 	 *
 	 * 2) Their physical addresses are continuous
-	 *    So truly compressed (physical size smaller than logical size)
-	 *    extents won't get merged with each other
 	 *
 	 * 3) Share same flags
+	 *
+	 * 4) Not compressed
 	 */
-	if (cache->offset + cache->len  == offset &&
-	    cache->phys + cache->len == phys  &&
-	    cache->flags == flags) {
-		cache->len += len;
+	if (cache->offset + cache->log_len  == offset &&
+	    cache->phys + cache->log_len == phys  &&
+	    cache->flags == flags &&
+	    !(flags & FIEMAP_EXTENT_DATA_COMPRESSED)) {
+		cache->log_len += log_len;
+		cache->phys_len += phys_len;
 		return 0;
 	}
 
@@ -2695,7 +2705,7 @@  static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
 		 * to miss it.
 		 */
 		entry = &cache->entries[cache->entries_size - 1];
-		cache->next_search_offset = entry->offset + entry->len;
+		cache->next_search_offset = entry->offset + entry->log_len;
 		cache->cached = false;
 
 		return BTRFS_FIEMAP_FLUSH_CACHE;
@@ -2704,7 +2714,8 @@  static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
 	entry = &cache->entries[cache->entries_pos];
 	entry->offset = cache->offset;
 	entry->phys = cache->phys;
-	entry->len = cache->len;
+	entry->log_len = cache->log_len;
+	entry->phys_len = cache->phys_len;
 	entry->flags = cache->flags;
 	cache->entries_pos++;
 	cache->extents_mapped++;
@@ -2717,7 +2728,8 @@  static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo,
 	cache->cached = true;
 	cache->offset = offset;
 	cache->phys = phys;
-	cache->len = len;
+	cache->log_len = log_len;
+	cache->phys_len = phys_len;
 	cache->flags = flags;
 
 	return 0;
@@ -2743,7 +2755,8 @@  static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo,
 		return 0;
 
 	ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys,
-				      cache->len, 0, cache->flags);
+				      cache->log_len, cache->phys_len,
+				      cache->flags);
 	cache->cached = false;
 	if (ret > 0)
 		ret = 0;
@@ -2937,13 +2950,15 @@  static int fiemap_process_hole(struct btrfs_inode *inode,
 			}
 			ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
 						 disk_bytenr + extent_offset,
-						 prealloc_len, prealloc_flags);
+						 prealloc_len, prealloc_len,
+						 prealloc_flags);
 			if (ret)
 				return ret;
 			extent_offset += prealloc_len;
 		}
 
 		ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0,
+					 delalloc_end + 1 - delalloc_start,
 					 delalloc_end + 1 - delalloc_start,
 					 FIEMAP_EXTENT_DELALLOC |
 					 FIEMAP_EXTENT_UNKNOWN);
@@ -2984,7 +2999,8 @@  static int fiemap_process_hole(struct btrfs_inode *inode,
 		}
 		ret = emit_fiemap_extent(fieinfo, cache, prealloc_start,
 					 disk_bytenr + extent_offset,
-					 prealloc_len, prealloc_flags);
+					 prealloc_len, prealloc_len,
+					 prealloc_flags);
 		if (ret)
 			return ret;
 	}
@@ -3130,6 +3146,7 @@  int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
 		u64 extent_offset = 0;
 		u64 extent_gen;
 		u64 disk_bytenr = 0;
+		u64 disk_size = 0;
 		u64 flags = 0;
 		int extent_type;
 		u8 compression;
@@ -3192,7 +3209,7 @@  int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
 			flags |= FIEMAP_EXTENT_DATA_INLINE;
 			flags |= FIEMAP_EXTENT_NOT_ALIGNED;
 			ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0,
-						 extent_len, flags);
+						 extent_len, extent_len, flags);
 		} else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 			ret = fiemap_process_hole(inode, fieinfo, &cache,
 						  &delalloc_cached_state,
@@ -3207,6 +3224,7 @@  int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
 						  backref_ctx, 0, 0, 0,
 						  key.offset, extent_end - 1);
 		} else {
+			disk_size = btrfs_file_extent_disk_num_bytes(leaf, ei);
 			/* We have a regular extent. */
 			if (fieinfo->fi_extents_max) {
 				ret = btrfs_is_data_extent_shared(inode,
@@ -3221,7 +3239,9 @@  int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
 
 			ret = emit_fiemap_extent(fieinfo, &cache, key.offset,
 						 disk_bytenr + extent_offset,
-						 extent_len, flags);
+						 extent_len,
+						 disk_size - extent_offset,
+						 flags);
 		}
 
 		if (ret < 0) {
@@ -3259,7 +3279,7 @@  int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo,
 		prev_extent_end = range_end;
 	}
 
-	if (cache.cached && cache.offset + cache.len >= last_extent_end) {
+	if (cache.cached && cache.offset + cache.log_len >= last_extent_end) {
 		const u64 i_size = i_size_read(&inode->vfs_inode);
 
 		if (prev_extent_end < i_size) {