The main thing to consider was how lfs_dir_fetchwith reacts to
corruption it finds and to make sure falling back to old values works
correctly.
Some of the tricky bits involved making sure we could fall back to both old
commits and old metadata blocks while still handling things like
synthetic moves correctly.
The main issue here was that the old orphan test relied on deleting the
block that contained the most recent update. In the new design this
doesn't really work since updates get appended to metadata-pairs
incrementally.
This is fixed by instead using the truncate command on the appropriate
block. We're now passing orphan tests.
Now that littlefs has been rebuilt almost from the ground up with the
intention to support custom attributes, adding in custom attribute
support is relatively easy.
The highest bit in the 9-bit type structure indicates that an attribute
is a user-specified custom attribute. The user then has a full 8-bits to
specify the attribute type. Other than that, custom attributes are
treated the same as system-level attributes.
Also made some tweaks to custom attributes:
- Adopted the opencfg for file-level attributes provided by dpgeorge
- Changed setattrs/getattrs to the simpler setattr/getattr functions
users will probably be more familiar with. Note that multiple
attributes can still be committed atomically with files, though not
with directories.
- Changed LFS_ATTRS_MAX -> LFS_ATTR_MAX since there's no longer a global
limit on the sum of attribute sizes, which was rather confusing.
Though they are still limited by what can fit in a metadata-pair.
Updated to account for changes as a result of commits/compacts. And
changed instances of iteration over both files and dirs to use a single
nested loop.
This does rely implicitly on the structure layout of dirs/files and
their location in lfs_t, which isn't great. But it gets the job done
with less code duplication.
This was a surprisingly tricky issue. One of the subtle requirements for
the new move handling to work is that the block containing the move does
not change until the move is resolved. Initially, this seemed easy to
implement, given that a move is always immediately followed by its
resolution.
However, the extra metadata-pair operations needed to maintain integrity
present a challenge. At any commit, a directory block may end up moved
as a side effect of relocation due to a bad block.
The fix here is to move the move resolution directly into the commit
logic. This means that any commit to a block containing a move will be
implicitly resolved, leaving the later attempt at move resolution as a
noop.
This fix required quite a bit of restructuring, but as a nice
side-effect some of the complexity around moves actually went away.
Additionally, the new move handling is surprisingly powerful at
combining moves with nearby commits. And we now get same-metadata-pair
renames for free! A win for procrasination on that minor feature.
Restrctured function organization to make a bit more sense, and made
some small refactoring tweaks, specifically around the commit logic and
global related functions.
The biggest change here is to make littlefs less obsessed with the
lfs_mattr_t struct. It was limiting our flexibility and can be entirely
replaced by passing the tag + data explicitly. The remaining use of
lfs_mattr_t is specific to the commit logic, where it replaces the
lfs_mattrlist_t struct.
Other changes:
- Added global lfs_diskoff struct for embedding disk references inside
the lfs_mattr_t.
- Reordered lfs_mattrlist_t to squeeze out some code savings
- Added commit_get for explicit access to entries from unfinished
metadata-pairs
- Parameterized the "stop_at_commit" flag instead of hackily storing it
in the lfs_mdir_t temporarily
- Changed return value of lfs_pred to error-only with ENOENT representing
a missing predecessor
- Adopted const where possible
One neat (if gimmicky) trick, is that each tag has a valid bit in the
highest bit position of the 32-bit word. This is used to determine when
to stop a fetch operation, but after fetch, the bit is free to use in
the driver. This means we can create a typed-union of sorts with error
codes and tags, returning both as the return value from a function.
Say what you will about this trick, it does have a significant impact on
code size. I suspect this is primarily due to the compiler having a hard
time optimizing around pointer access.
While it makes sense to reuse as many code paths as possible, it turns
out that the logic behind the traversal of littlefs's metadata-pairs is
so simple that it's actually cheaper to duplicate the traversal code
where needed.
This means instead of the code path move -> traverse -> movescan -> get
-> traverse -> getscan, we can use the relatively flatter code path of
move -> get.
Tags offer all of the necessary info from the find functions (which
makes sense, this is the structure that stores the info on disk).
Passing around a single tag instead of separate id and type fields
simplifies the internal functions while leverages the tag's compactness.
Unfortunately, the three different sets of get functions were not
contributing very much, and having three different get functions means
we may be wasting code on redundant code paths.
By dropping the user of the lfs_mattr_t struct in favor of a buffer, we
can combine the three code paths with a bit of tweaking.
Now that littlefs's fetchwith operations have stabilized a bit, there's
actually only a single fetchwith operation, the findscan function.
Given that there's no need for the internal functions to be a forward
compatible API, we can integrate the findscan behaviour directly into
fetchwith and avoid the (annoyingly) costly generalization overhead.
As an added benefit, we can easily add additional tag modifications
during fetch, such as the synthetic moves needed to resolve in-flight
move operations without disk modifications.
An interesting observation about the find and parent scanning functions
is that at their core, they're both actually doing the same operation.
They search a metadata-pair during fetch for an entry uses the entry data
instead of the entry tag. This means we can combine these functions and
get a decent code savings.
It's a little bit trickier because pair ordering isn't guaranteed. But
to work around that we can simply search for both pair orderings. It's a
bit more expensive but may be worth the code savings. A fancier
implementation in the future can avoid the 2x lfs_parent scans.
I've been trying to keep tag types organized with an encoding that hints
if a tag uses its id field for file ids. However this seem to have been
a mistake. Using a null id of 0x3ff greatly simplified quite a bit of
the logic around managing file related tags.
The downside is one less id we can use, but if we look at the encoding
cost, donating one full bit costs us 2^9 id permutations vs 1 id
permutation. So even if we had a perfect encoding it's in our favor to
use a null id. The cost of null ids is code size, but with the
complexity around figuring out if a type used it's id or not it just
works out better to use a null id.
Recall that the 32-bit tag structure contains a 9-bit type. The type
structure then decomposes into a bit more information:
[--- 9 ---]
[1|- 4 -|- 4 -]
^ ^ ^- specific type
| \------- subtype
\----------- user bit
The main change is an observation from moving type info to the name tag
from the struct tag. Since we don't need the type info in the struct
tag, we can significantly simplify the type structure.
Originally, I had type info encoded in the struct tag. This initially
made sense because the type info only directly impacts the struct tag.
However this was a case of focusing too much on the details instead of
the bigger picture.
A more file operations need to figure out the type of a file, but it's
only actually a small number of file operations that need to interact
with the file's structure. For the common case, providing the type of
the file early shortens operations by a full tag access.
Additionally, but storing the type in the file name tag, this opens up
the struct tag to use those bits for storing more struct descriptions.
The introduction of xored-globals required quite a bit of work to
integrate. But now that that is working, we can strip out the old move
logic.
It's worth noting that the xored-globals integration with commits is
relatively complex and subtle.
The issue lies in the reuse of the id field for globals. Before globals,
the only tags with a non-null (0x3ff) id field were names, structs, and
other file-specific metadata. But globals are also using this field for
the indirect delete, since otherwise the globals structure would be very
unaligned (74-bits long).
To make matters worse, the id field for globals contains the delta used
to reconstruct the globals at mount time. Which means the id field could
take on very absurd values and break the dir fetch logic if we're not
careful.
Solution is to use the scope portion of the type field where necessary,
although unforunately this does add some code cost.
Unfortunately, the behaviour needed of lfs_dir_fetchwith is as subtle as
it is important. When fetching from a block corrupted by power-loss,
lfs_dir_fetch must be able to rewind any state it picks up to before the
corruption. This is not limited to the directory state, but includes
find results and other side-effects.
This gets a bit complicated when trying to generalize littlefs's
fetchwith mechanics. Being able to scan a directory block during a fetch
greatly impacts the runtime of littlefs operations, but if the state is
generic how do we know what to rollback to?
The fix here is to leave the management of rolling back state to the
fetchwith match functions, and transparently pass a CRC tag to indicate
the temporary state can be saved.
Originally I tried to reuse the indirect delete to accomplish truely
atomic directory removes, however this fell apart when it came to
implementing directory removes as a side-effect of renames.
A single indirect-delete simply can't handle renames with removes as
a side effects. When copying an entry to its destination, we need to
atomically delete both the old entry, and the source of our copy. We
can't delete both with only a single indirect-delete. It is possible to
accomplish this with two indirect-deletes, but this is such an uncommon
case that it's really not worth supporting efficiently due to how
expensive globals are.
I also dropped indirect-deletes for normal directory removes. I may add
it back later, but at the moment it's extra code cost for that's not
traveled very often.
As a result, restructured the indirect delete handling to be a bit more
generic, now with a multipurpose lfs_globals_t struct instead of the
delete specific lfs_entry_t struct.
Also worked on integrating xored-globals, now with several primitive
global operations to manage fetching/updating globals on disk.
32-bit tag structure:
[--- 32 ---]
[1|- 9 -|- 10 -|-- 12 --]
^ ^ ^ ^- entry length
| | \--------- file id
| \--------------- tag type
\------------------- valid
In this tag, the type decomposes into some more information:
[--- 9 ---]
[1|- 2 -|- 3 -|- 3 -]
^ ^ ^ ^- struct
| | \------- type
| \------------- scope
\----------------- user
The change in this encoding is the addition of a global scope:
LFS_SCOPE_STRUCT = 0 00 xxx xxx
LFS_SCOPE_ENTRY = 0 01 xxx xxx
LFS_SCOPE_DIR = 0 10 xxx xxx
LFS_SCOPE_FS = 0 11 xxx xxx
LFS_SCOPE_USER = 1 xx xxx xxx
This was a big roadblock for a while: with the new feature of inlined
files, the existing move logic was fundamentally flawed.
To pull off atomic moves between two different metadata-pairs, littlefs
uses a simple, if a bit clumsy trick.
1. Marks entry as "moving"
2. Copies entry to new metadata-pair
3. Deletes old entry
If power is lost before the move operation is completed, we will find the
"moving" tag. This means there may or may not be an incomplete move on
the filesystem. In this case, we simply search for the moved entry, if
we find it, we remove the old entry, otherwise we just remove the
"moving" tag.
This worked perfectly, until we introduced inlined files. See, unlike
the existing directory and ctz entries, inlined files have no guarantee
they are unique. There is nothing we can search for that will allow us
to find a moved file unless we assign entries globally-unique ids. (note
that moves are fundamentally rename operations, so searching for names
does not make sense).
---
Solving this problem required completely restructuring how littlefs
handled moves and pulled out a really old idea that had been left in the
cutting room floor back when littlefs was going through many
designs: xored-globals.
The problem xored-globals solves is the need to maintain some global state
via commits to these distributed, independent metadata-pairs. The idea
is that we can use some sort of symmetric operation, such as xor, to
introduces deltas of the global state that can be committed atomically
along with any other info to these metadata-pairs.
This means that to figure out our global state, we xor together the global
delta stored in every metadata-pair.
Which means any commit can update the global state atomically, opening
up a whole new set atomic possibilities.
There is a couple of downsides. These globals may end up with deltas on
every single metadata-pair, effectively duplicating the data for each
block. Additionally, these globals need to have multiple copies in RAM.
This means and globals need to be a bounded size and very small, since even
small globals will have a large footprint.
---
On top of xored-globals, it's trivial to fix our move logic. Here we've
added an indirect delete tag which allows us to atomically specify a
delete of any entry on the filesystem.
Our move operation is now:
1. Copy entry to new metadata-pair and atomically xor globals to
indirectly delete our original entry.
2. Delete the original entry and xor globals to remove the indirect
delete.
Extra exciting is that this now takes our relatively clumsy move
operation into a sexy guaranteed O(1) move operation with no searching
necessary (though we do need to xor globals during mount).
Also reintroduced entry struct, now with a specific purpose to describe
the metadata-pair + id combo needed by indirect deletes to locate an
entry.
Unfortunately, it's hard to make directory lookups for root not a
special case, and we don't like special cases when we're trying to keep
code size small.
Since there are a handful of code paths where opening root should return
EISDIR (such as lfs_file_open("/")), using EISDIR to note that the
argument is in fact a path to the root.
This is needed because we no longer look up an entries contents in
lfs_dir_find for free, since entries are more ephemeral.
Similarly to the internal "meta-attributes", I was finding quite a bit
of need for an internal structure that mirrors the user-facing directory
structure for when I need to do an operation on a metadata-pair, but
don't need all of the state associated with a fully iterable directory
chain.
lfs_mdir_t - meta-directory, describes a single metadata-pair
lfs_dir_t - directory, describes an iterable directory chain
While it may seem complex to have all these structures lying around,
they only complicate the code at compile time. To the machine, any
number of nested structures all looks the same.
Attributes are used to describe more than just entries, so calling these
list of attributes "entries" was inaccurate. However, the name
"attributes" would conflict with "user attributes", user-facing
attributes with a very similar purpose. "user attributes" must be kept
distinct due to differences in binary layout (internal attributes can
use a more compact tag+buffer representation, but expecting users to
jump through hoops to get their data to look like that isn't very
user-friendly).
Decided to go with "mattr" as shorthand for "meta-attributes", similar
to "metadata".
I've found using temporary names to duplicate functions temporarily
(lfs_dir_commit + lfs_dir_commit_) is a great way to introduce sweeping
changes while keeping the code base functional and (mostly) passing
tests.
It does mean at some point I need to deduplicate all these functions.
lfs_dir_fetchwith did not recover from failed dir fetches correctly,
added a temporary dir variable to hold dir contents while being
populated, allowing us to fall back to a known good dir state if a
commit is corrupted.
There is a RAM cost, but the upside is that our lfs_dir_fetchwith
actually works.
Also added better handling of move ids during some get functions.
Integration with the new journaling metadata has now progressed to the
point where all of the basic functionality tests are passing. This
includes:
- test_format
- test_dirs
- test_files
- test_seek
- test_truncate
- test_interspersed
- test_paths
Some of the fixes:
- Modified move to correctly change entry ids
- Called lfs_commit_move directly from compact, avoiding commit parsing
logic during a compact
- Opened up commit filters to be passed down from compact for moves
- Added correct drop logic to lfs_dir_delete
- Updated lfs_dir_seek to use ids instead of offsets
- Caught id updates manually where possible (this needs to be fixed)
Now with some tweaks to commit/compact, and a committers for entrylists and
moves specifically. No longer relying on a commitwith callback, the
types of commits are now infered from their tags.
This means we can now commit things atomically with special commits,
such as moves. Now lfs_rename can move entries to new names correctly.
Passing more tests now with the journalling change, but still have more
work to do.
The most humorous bug was a bug where during the three step move
process, the entry move logic would dumbly copy over any tags associated
with the moving entry, including the tag used to temporarily mark the
entry as "moving".
Also combined dir and commit traversal using a "stop_at_commit" flag in
directory struct as a short-term hack to combine the code paths.
Also refactored lfs_dir_compact a bit, adding begin and end as arguments
since they simplify a bit of the logic and can be found out much easier
earlier in the commit logic.
Also changed add -> append and drop -> delete and cleaned up some of the
logic around there.
This was the simpler part of transitioning since file operations only
interact with metadata at sync time.
Also switched from array to linked-list of entries.
- Integrated into lfs_file_t_, duplicating functions where necessary
- Added lfs_dir_fetchwith_ as common parent to both lfs_dir_fetch_ and
lfs_dir_find_
- Added similar parent with lfs_dir_commitwith_
- Made matching find/get operations with getbuffer/getentry and
findbuffer/findentry
- lfs_dir_alloc now populates tail, since almost all directory block
allocations need to populate tail
- Integrated journaling into lfs_dir_t_ struct and operations,
duplicating functions where necessary
- Added internal lfs_tag_t and lfs_stag_t
- Consolidated lfs_region and lfs_entry structures
This is a big change stemming from the fact that resizable entries
were surprisingly complicated to implement and came in with a sizable
code cost.
The theory is that the journalling has a comparable cost to resizable
entries. Both need to handle overflowing blocks, and managing offsets is
comparable to managing attribute IDs. But by jumping all the way to full
journaling, we can statically wear-level the metadata written to
metadata pairs.
The idea of journaling littlefs's metadata has been mentioned several times in
discussions and fits well into how littlefs works. You could even view the
existing metadata log as a log of size 2.
The downside of this approach is that changing the metadata in this way
would break compatibility from the existing layout on disk. Something
that resizable entries does not do.
That being said, adopting journaling at the metadata layer offers a big
improvement to littlefs's performance and wear-leveling, with very
little cost (maybe even none or negative after resizable entries?).
This has existed for some time in the form of the lfs_traverse
function, through which a user could provide a simple callback that
would just count the number of blocks lfs_traverse finds. However,
this approach is relatively unconventional and has proven to be confusing
for most users.
In the form of lfs_file_setattr, lfs_file_getattr, lfs_fs_setattr,
lfs_fs_getattr.
This enables atomic updates of custom attributes as described in
6c754c8, and provides a custom attribute API that allows custom attributes
to be stored on the filesystem itself.
Mostly just removed LFS_FROM_DROP and changed the DSL grammar a bit to
allow drops to occur naturally through oldsize -> newsize diff expressed
in the region struct. This prevents us from having to add a drop every
time we want to update an entry in-place.
Although it's simple and probably what most users expect, the previous
custom attributes API suffered from one problem: the inability to update
attributes atomically.
If we consider our timestamp use case, updating a file would require:
1. Update the file
2. Update the timestamp
If a power loss occurs during this sequence of updates, we could end up
with a file with an incorrect timestamp.
Is this a big deal? Probably not, but it could be a surprise only found
after a power-loss. And littlefs was developed with the _specifically_
to avoid suprises during power-loss.
The littlefs is perfectly capable of bundling multiple attribute updates
in a single directory commit. That's kind of what it was designed to do.
So all we need is a new committer opcode for list of attributes, and
then poking that list of attributes through the API.
We could provide the single-attribute functions, but don't, because the
fewer functions makes for a smaller codebase, and these are already the
more advanced functions so we can expect more from users. This also
changes semantics about what happens when we don't find an attribute,
since erroring would throw away all of the other attributes we're
processing.
To atomically commit both custom attributes and file updates, we need a
new API, lfs_file_setattr. Unfortunately the semantics are a bit more
confusing than lfs_setattr, since the attributes aren't written out
immediately.
A much requested feature (mostly because of littlefs's notable lack of
timestamps), this commits adds support for user-specified custom
attributes.
Planned (though underestimated) since v1, custom attributes provide a
route for OSs and applications to provide their own metadata in
littlefs, without limiting portability.
However, unlike custom attributes that can be found on much more
powerful PC filesystems, these custom attributes are very limited,
intended for only a handful of bytes for very important metadata. Each
attribute has only a single byte to identify the attribute, and the
size of all attributes attached to a file is limited to 64 bytes.
Custom attributes can be accessed through the lfs_getattr, lfs_setattr,
and lfs_removeattr functions.
One of the big benefits of inline files is that small files no longer need to
take up a full block. This opens up an opportunity to provide much better
support for storage devices with only a handful of very large blocks. Such as
the internal flash found on most microcontrollers.
After investigating some use cases for a filesystem on internal flash,
it has become apparent that the 255-byte limit is going to be too
restrictive to be useful in many cases. Most uses I found needed files
~4-64 bytes in size, but it wasn't uncommon to find files ~512 bytes in
length.
To try to remedy this, I've pushed the 255 byte limit up to 1023 bytes,
by stealing some bits from the previously-unused attributes's size.
Unfortunately this limits attributes to 63 bytes in total and has a
minor code cost, but I'm not sure even 1023 bytes will be sufficient for
a lot of cases.
The littlefs will probably never be as efficient with internal flash as
other filesystems such as SPIFFS, it just wasn't designed for this sort of
limited geometry. However, this feature has been heavily requested, even
with limitations, because of the opportunity for code reuse on
microcontrollers with both internal and external flash.
Being a portable, microcontroller-scale embedded filesystem, littlefs is
presented with a relatively unique challenge. The amount of RAM
available is on completely different scales from machine to machine, and
what is normally a reasonable RAM assumption may break completely on an
embedded system.
A great example of this is file names. On almost every PC these days, the limit
for a file name is 255 bytes. It's a very convenient limit for a number
of reasons. However, on microcontrollers, allocating 255 bytes of RAM to
do a file search can be unreasonable.
The simplest solution (and one that has existing in littlefs for a
while), is to let this limit be redefined to a smaller value on devices
that need to save RAM. However, this presents an interesting portability
issue. If these devices are plugged into a PC with relatively infinite
RAM, nothing stops the PC from writing files with full 255-byte file
names, which can't be read on the small device.
One solution here is to store this limit on the superblock during format
time. When mounting a disk, the filesystem implementation is responsible for
checking this limit in the superblock. If it's larger than what can be
read, raise an error. If it's smaller, respect the limit on the
superblock and raise an error if the user attempts to exceed it.
In this commit, this strategy is adopted for file names, inline files,
and the size of all attributes, since these could impact the memory
consumption of the filesystem. (Recording the attribute's limit is
iffy, but is the only other arbitrary limit and could be used for disabling
support of custom attributes).
Note! This changes makes it very important to configure littlefs
correctly at format time. If littlefs is formatted on a PC without
changing the limits appropriately, it will be rejected by a smaller
device.
This move was surprisingly complex, but offers the ultimate opportunity for
code reuse in terms of resizable entries. Instead of needing to provide
separate functions for adding and removing entries, adding and removing
entries can just be viewed as changing an entry's size to-and-from zero.
Unfortunately, it's not _quite_ that simple, since append and remove
hide some relatively complex operations for when directory blocks
overflow or need to be cleaned up.
However, with enough shoehorning, and a new committer type that allows
specifying recursive commit lists (is this now a push-down automata?),
it does seem to be possible to shove all of the entry update logic into
a single function.
Sidenote, I switched back to an enum-based DSL, since the addition of a
recursive region opcode breaks the consistency of what needs to be
passed to the DSL callback functions. It's much simpler to handle each
opcode explicitly inside a recursive lfs_commit_region function.
