The lock test creates a lock and checks that it is not stale. However,
it is possible that the lock is refreshed concurrently, which updates
the lock timestamp. Checking the timestamp in `Stale()` without
synchronization results in a data race. Thus add a lock to prevent
concurrent accesses.
The lock test creates a lock and checks that it is not stale. This also
tests whether the corresponding process still exists. However, it is
possible that the lock is refreshed concurrently, which updates the lock
timestamp. Calling `processExists()` with a value receiver, however,
creates an unsynchronized copy of this field. Thus call the method using
a pointer receiver.
In some rare cases files could be created which contain null IDs (all
zero) in their content list. This was caused by a race condition between
growing the `Content` slice and inserting the blob IDs into it. In some
cases the blob ID was written to the old slice, which a short time
afterwards was replaced with a larger copy, that did not yet contain the
blob ID.
Automatically fall back to hiding files if not authorized to permanently
delete files. This allows using restic with an append-only application
key with B2. Thus, an attacker cannot directly delete backups with the
API key used by restic.
To use this feature create an application key without the deleteFiles
capability. It is recommended to restrict the key to just one bucket.
For example using the b2 command line tool:
b2 create-key --bucket <bucketName> <keyName> listBuckets,readFiles,writeFiles,listFiles
Suggested-by: Daniel Gröber <dxld@darkboxed.org>
We previously checked whether the set of snapshots might have changed
based only on their number, which fails when as many snapshots are
forgotten as are added. Check for the SHA-256 of their id's instead.
The status bar got stuck once the first error was reported, the scanner
completed or some file was backed up. Either case sets a flag that the
scanner has started.
This flag is used to hide the progress bar until the flag is set. Due to
an inverted condition, the opposite happened and the status stopped
refreshing once the flag was set.
In addition, the scannerStarted flag was not set when the scanner just
reported progress information.
As the FileSaver is asynchronously waiting for all blobs of a file to be
stored, the number of active files is higher than the number of files
from which restic is reading concurrently. Thus to not confuse users,
only display files in the status from which restic is currently reading.
After reading and chunking all data in a file, the FutureFile still has
to wait until the FutureBlobs are completed. This was done synchronously
which results in blocking the file saver and prevents the next file from
being read.
By replacing the FutureBlob with a callback, it becomes possible to
complete the FutureFile asynchronously.
We always need both values, except in a test, so we don't need to lock
twice and risk scheduling in between.
Also, removed the resetting in Done. This copied a mutex, which isn't
allowed. Static analyzers tend to trip over that.
The channel-based algorithm had grown quite complicated. This is easier
to reason about and likely to be more performant with very many
CompleteBlob calls.
As long as only a small fraction of the data in a repository is
rewritten, the keepBlobs set will be rather small after cleaning it up.
As golang maps do not shrink their memory usage, just copy the contents
over to a new map. However, only copy the map if the cleanup removed at
least half the entries.
The set covers necessary, existing and duplicate blobs. This removes the
duplicate sets used to track whether all necessary blobs also exist.
This reduces the memory usage of prune by about 20-30%.
The RetryBackend tests depend on the mock backend. When the Backend
interface is eventually split from the restic package, this will lead to
a dependency cycle between backend and backend/mock. Thus split the
RetryBackend into a separate package to avoid this problem.
Archiver.Save queries the current time multiple times. This commit
removes one of these calls as they showed up while profiling a backup of
a nearly unchanged dataset containing 3 million files.
The string form was presumably useful before the introduction of
layouts, but right now it just makes call sequences and garbage
collection more expensive (the latter because every string contains
a pointer to be scanned).
if x { return true } return false => return x
fmt.Sprintf("%v", x) => fmt.Sprint(x) or x.String()
The fmt.Sprintf idiom is still used in the SecretString tests, where it
serves security hardening.
ID.UnmarshalJSON accepted non-JSON input with ' as the string delimiter.
Also, the error message for non-hex input was less informative than it
could be and it performed too many checks.
Changed ParseID to keep the error messages consistent.
FindFilteredSnapshots no longer prints errors during snapshot loading on
stderr, but instead passes the error to the callback to allow the caller
to decide on what to do.
In addition, it moves the logic to handle an explicit snapshot list from
the main package to restic.
The helper function uidGidInt used strconv.ParseInt instead of
ParseUint, so it silently ignored some invalid user/group IDs.
Also, improve the error message. "Invalid UID" is more informative than
having "ParseInt" twice (*strconv.NumError displays the function name).
Finally, the user.User struct can be passed by pointer to get reduce
code size.
This package is no longer needed, since we can use the stdlib's
http.NewRequestWithContext.
backend/rclone already did, but it needed a different error check due to
a difference between net/http and ctxhttp.
Also, store the http.Client by value in the REST backend (changed to a
pointer when ctxhttp was introduced) and use errors.WithStack instead
of errors.Wrap where the message was no longer accurate. Errors from
http.NewRequestWithContext will start with "net/http" or "net/url", so
they're easy to identify.
The backup command failed if a directory contains duplicate entries.
Downgrade the severity of this problem from fatal error to a warning.
This allows users to still create a backup.
SaveTree did not use the TreeSaver but rather managed the tree
collection and upload itself. This prevents using the parallelism
offered by the TreeSaver and duplicates all related code. Using the
TreeSaver can provide some speed-ups as all steps within the backup tree
now rely on FutureNodes. This can be especially relevant for backups
with large amounts of explicitly specified files.
The main difference between SaveTree and SaveDir is, that only the
former can save tree blobs in which nodes have a different name than the
actual file on disk. This is the result of resolving name conflicts
between multiple files with the same name. The filename that must be
used within the snapshot is now passed directly to
restic.NodeFromFileInfo. This ensures that a FutureNode already contains
the correct filename.
According to the documentation of exec.Cmd Wait() must not be called
before completing all reads from the pipe returned by StdErrPipe(). Thus
return a context that is canceled once rclone has exited and use that as
a precondition to calling Wait(). This should ensure that all errors
printed to stderr have been copied first.
When rclone fails during the connection setup this currently often
results in a context canceled error. Replace this error with the exit
code from rclone.
When backing up many small files, the unbuffered channels frequently
cause the FileSaver to block when reporting progress information. Thus,
add buffers to these channels to avoid unnecessary scheduling.
As the status information is purely informational, it doesn't matter
that the status reporting shutdown is somewhat racy and could miss a few
final updates.
The only use cases in the code were in errors.IsFatal, backend/b2,
which needs a workaround, and backend.ParseLayout. The last of these
requires all backends to implement error unwrapping in IsNotExist.
All backends except gs already did that.
Repositories with mixed packs are probably quite rare by now. When
loading data blobs from a mixed pack file, this will no longer trigger
caching that file. However, usually tree blobs are accessed first such
that this shouldn't make much of a difference.
The checker gets a simpler replacement.
While searching for lock file from concurrently running restic
instances, restic ignored unreadable lock files. These can either be
in fact invalid or just be temporarily unreadable. As it is not really
possible to differentiate between both cases, just err on the side of
caution and consider the repository as already locked.
The code retries searching for other locks up to three times to smooth
out temporarily unreadable lock files.
Restic continued e.g. a backup task even when it failed to renew the
lock or failed to do so in time. For example if a backup client enters
standby during the backup this can allow other operations like `prune`
to run in the meantime (after calling `unlock`). After leaving standby
the backup client will continue its backup and upload indexes which
refer pack files that were removed in the meantime.
This commit introduces a goroutine explicitly monitoring for locks that
are not refreshed in time. To simplify the implementation there's now a
separate goroutine to refresh the lock and monitor for timeouts for each
lock. The monitoring goroutine would now cause the backup to fail as the
client has lost it's lock in the meantime.
The lock refresh goroutines are bound to the context used to lock the
repository initially. The context returned by `lockRepo` is also
cancelled when any of the goroutines exits. This ensures that the
context is cancelled whenever for any reason the lock is no longer
refreshed.
Some backends generate additional files for each existing file, e.g.
1234567890abcdef1234567890abcdef1234567890abcdef1234567890abcdef
1234567890abcdef1234567890abcdef1234567890abcdef1234567890abcdef.sha256
For some commands this leads to an "multiple IDs with prefix" error when
trying to reference a snapshot.
Failing to process data requested from the cache usually indicates a
problem with the returned data. Assume that the cache entry is somehow
damaged and retry downloading it once.
Sparse files contain large regions containing only zero bytes. Checking
that a blob only contains zeros is possible with over 100GB/s for modern
x86 CPUs. Calculating sha256 hashes is only possible with 500MB/s (or
2GB/s using hardware acceleration). Thus we can speed up the hash
calculation for all zero blobs (which always have length
chunker.MinSize) by checking for zero bytes and then using the
precomputed hash.
The all zeros check is only performed for blobs with the minimal chunk
size, and thus should add no overhead most of the time. For chunks which
are not all zero but have the minimal chunks size, the overhead will be
below 2% based on the above performance numbers.
This allows reading sparse sections of files as fast as the kernel can
return data to us. On my system using BTRFS this resulted in about
4GB/s.
The restorer can issue multiple calls to WriteAt in parallel. This can
result in unexpected orderings of the Truncate and WriteAt calls and
sometimes too short restored files.
We can either preallocate storage for a file or sparsify it. This
detects a pack file as sparse if it contains an all zero block or
consists of only one block. As the file sparsification is just an
approximation, hide it behind a `--sparse` parameter.
This writes files by using (*os.File).Truncate, which resolves to the
truncate system call on Unix.
Compared to the naive loop,
for _, b := range p {
if b != 0 {
return false
}
}
the optimized allZero is about 10× faster:
name old time/op new time/op delta
AllZero-8 1.09ms ± 1% 0.09ms ± 1% -92.10% (p=0.000 n=10+10)
name old speed new speed delta
AllZero-8 3.84GB/s ± 1% 48.59GB/s ± 1% +1166.51% (p=0.000 n=10+10)
`restic unlock` now only shows `successfully removed locks` if there were locks to be removed.
In addition, it also reports the number of the removed lock files.
Sending data through a channel at very high frequency is extremely
inefficient. Thus use simple callbacks instead of channels.
> name old time/op new time/op delta
> MasterIndexEach-16 6.68s ±24% 0.96s ± 2% -85.64% (p=0.008 n=5+5)
This is especially useful if ssh asks for a password or if closing the
initial connection could return an error due to a problematic server
implementation.
bazil/fuse expects us to return context.Canceled to signal that a
syscall was successfully interrupted. Returning a wrapped version of
that error however causes the fuse library to signal an EIO (input/output
error). Thus unwrap context.Canceled errors before returning them.
rclone can exit early for example when the connection to rclone is
relayed for example via ssh: `-o rclone.program='ssh user@example.org
forced-command'`
For backends which are able to atomically replace files, we just can
overwrite the old copy, if it is necessary to retry an upload. This has
the benefit of issuing one operation less and might be beneficial if a
backend storage, due to bugs or similar, could mix up the order of the
upload and delete calls.
When hard deleting the latest file version on B2, this uncovers earlier
versions. If an upload required retries, multiple version might exist
for a file. Thus to reliably delete a file, we have to remove all
versions of it.
Ignored packs were reported as an empty pack by EachByPack. The most
immediate effect of this is that the progress bar for rebuilding the
index reports processing more packs than actually exist.
Previously the buffer was grown incrementally inside `repo.LoadUnpacked`.
But we can do better as we already know how large the index will be.
Allocate a bit more memory to increase the chance that the buffer can be
reused in the future.
Instead of first checking whether a file is in the repository cache and
then opening it, we just can open the file. This saves one stat call. If
the file is in the cache, everything is fine and otherwise the code
follows its normal fallback path.
sort.Sort is not guaranteed to be stable. Go 1.19 has changed the
sorting algorithm which resulted in changes of the sort order. When
comparing snapshots with identical timestamp but different paths and
tags lists, there is not meaningful order among them. So just keep their
order stable.
Cleanly separate the directory presentation and the snapshot directory
structure. SnapshotsDir now translates the dirStruct into a format
usable by the fuse library and contains only minimal special case rules.
All decisions have moved into SnapshotsDirStructure which now creates a
fully preassembled tree data structure.
For large pack sizes we might be only interested in the first and last
blob of a pack file. Thus stream a pack file in multiple parts if the
gaps between requested blobs grow too large.
Also make the errors a bit less verbose by not prepending the operation,
since pkg/xattr already does that. Old errors looked like
Listxattr: xattr.list /myfiles/.zfs/snapshot: invalid argument
pkg/sftp.Client.MkdirAll(d) does a Stat to determine if d exists and is
a directory, then a recursive call to create the parent, so the calls
for data/?? each take three round trips. Doing a Mkdir first should
eliminate two round trips for 255/256 data directories as well as all
but one of the top-level directories.
Also, we can do all of the calls concurrently. This may reintroduce some
of the Stat calls when multiple goroutines try to create the same
parent, but at the default number of connections, that should not be
much of a problem.
FutureBlob now uses a Take() method as a more memory-efficient way to
retrieve the futures result. In addition, futures are now collected
while saving the file. As only a limited number of blobs can be queued
for uploading, for a large file nearly all FutureBlobs already have
their result ready, such that the FutureBlob object just consumes
memory.
There is no real difference between the FutureTree and FutureFile
structs. However, differentiating both increases the size of the
FutureNode struct.
The FutureNode struct is now only 16 bytes large on 64bit platforms.
That way is has a very low overhead if the corresponding file/directory
was not processed yet.
There is a special case for nodes that were reused from the parent
snapshot, as a go channel seems to have 96 bytes overhead which would
result in a memory usage regression.
