Reusing http connection when operating on chunks is not done yet,
I had to submit some patches to DAV to support that. However, this is no
slower than old-style chunking was.
Note that it's a fileRetriever and a fileStorer, despite DAV using
bytestrings that would allow streaming. As a result, upload/download of
encrypted files is made a bit more expensive, since it spools them to temp
files. This was needed to get the progress meters to work.
There are probably ways to avoid that.. But it turns out that the current
DAV interface buffers the whole file content in memory, and I have
sent in a patch to DAV to improve its interfaces. Using the new interfaces,
it's certainly going to need to be a fileStorer, in order to read the file
size from the file (getting the size of a bytestring would destroy
laziness). It should be possible to use the new interface to make it be a
byteRetriever, so I'll change that when I get to it.
This commit was sponsored by Andreas Olsson.
This will allow special remotes to eg, open a http connection and reuse it,
while checking if chunks are present, or removing chunks.
S3 and WebDAV both need this to support chunks with reasonable speed.
Note that a special remote might want to cache a http connection across
multiple requests. A simple case of this is that CheckPresent is typically
called before Store or Remove. A remote using this interface can certianly
use a Preparer that eg, uses a MVar to cache a http connection.
However, it's up to the remote to then deal with things like stale or
stalled http connections when eg, doing a series of downloads from a remote
and other places. There could be long delays between calls to a remote,
which could lead to eg, http connection stalls; the machine might even
move to a new network, etc.
It might be nice to improve this interface later to allow
the simple case without needing to handle the full complex case.
One way to do it would be to have a `Transaction SpecialRemote cache`,
where SpecialRemote contains methods for Storer, Retriever, Remover, and
CheckPresent, that all expect to be passed a `cache`.
I tend to prefer moving toward explicit exception handling, not away from
it, but in this case, I think there are good reasons to let checkPresent
throw exceptions:
1. They can all be caught in one place (Remote.hasKey), and we know
every possible exception is caught there now, which we didn't before.
2. It simplified the code of the Remotes. I think it makes sense for
Remotes to be able to be implemented without needing to worry about
catching exceptions inside them. (Mostly.)
3. Types.StoreRetrieve.Preparer can only work on things that return a
Bool, which all the other relevant remote methods already did.
I do not see a good way to generalize that type; my previous attempts
failed miserably.
This reaping of any processes came to cause me problems when redoing the
rsync special remote -- a gpg process that was running gets waited on and
the place that then checks its return code fails.
I cannot reproduce any zombies when using the rsync special remote.
But I still can when using a normal git remote, accessed over ssh.
There is 1 zombie per file downloaded without this horrible hack enabled.
So, move the hack to only be used in that case.
Make the byteRetriever be passed the callback that consumes the bytestring.
This way, there's no worries about the lazy bytestring not all being read
when the resource that's creating it is closed.
Which in turn lets bup, ddar, and S3 each switch from using an unncessary
fileRetriver to a byteRetriever. So, more efficient on chunks and encrypted
files.
The only remaining fileRetrievers are hook and external, which really do
retrieve to files.
Chunking would complicate the assistant's code that checks when a pending
retrieval of a key from glacier is done. It would perhaps be nice to
support it to allow resuming, but not right now.
Converting to the new API still simplifies the code.
The forall a. in Preparer made resourcePrepare not seem to be usable, so
I specialized a to Bool. Which works for both Preparer Storer and
Preparer Retriever, but wouldn't let the Preparer be used for hasKey
as it currently stands.
And fixed a bug found by these tests; retrieveKeyFile would fail
when the dest file was already complete.
This commit was sponsored by Bradley Unterrheiner.
The content of unstable keys can potentially be different in different
repos, so eg, resuming a chunked upload started by another repo would
corrupt data.
This way, when the remote implementation neglects to update progress,
there will still be a somewhat useful progress display, as long as chunks
are used.
No need to read whole FileContent only to write it back out to a file in
this case. Can just rename! Yay.
Also indidentially, fixed an attempt to open a file for write that was
already opened for write, which caused a crash and deadlock.
Putting a callback in the Retriever type allows for the callback to
remove the retrieved file when it's done with it.
I did not really want to make Retriever be fixed to Annex Bool,
but when I tried to use Annex a, I got into some type of type mess.
Needed for eg, Remote.External.
Generally, any Retriever that stores content in a file is responsible for
updating the meter, while ones that procude a lazy bytestring cannot update
the meter, so are not asked to.
Some remotes like External need to run store and retrieve actions in Annex,
not IO. In order to do that lift, I had to dive pretty deep into the
utilities, making Utility.Gpg and Utility.Tmp be partly converted to using
MonadIO, and Control.Monad.Catch for exception handling.
There should be no behavior changes in this commit.
This commit was sponsored by Michael Barabanov.
Leverage the new chunked remotes to automatically resume uploads.
Sort of like rsync, although of course not as efficient since this
needs to start at a chunk boundry.
But, unlike rsync, this method will work for S3, WebDAV, external
special remotes, etc, etc. Only directory special remotes so far,
but many more soon!
This implementation will also allow starting an upload from one repository,
interrupting it, and then resuming the upload to the same remote from
an entirely different repository.
Note that I added a comment that storeKey should atomically move the content
into place once it's all received. This was already an undocumented
requirement -- it's necessary for hasKey to work reliably. This resume code
just uses hasKey to find the first chunk that's missing.
Note that if there are two uploads of the same key to the same chunked remote,
one might resume at the point the other had gotten to, but both will then
redundantly upload. As before.
In the non-resume case, this adds one hasKey call per storeKey, and only
if the remote is configured to use chunks. Future work: Try to eliminate that
hasKey. Notice that eg, `git annex copy --to` checks if the key is present
before sending it, so is already running hasKey.. which could perhaps
be cached and reused.
However, this additional overhead is not very large compared with
transferring an entire large file, and the ability to resume
is certianly worth it. There is an optimisation in place for small files,
that avoids trying to resume if the whole file fits within one chunk.
This commit was sponsored by Georg Bauer.
Leverage the new chunked remotes to automatically resume downloads.
Sort of like rsync, although of course not as efficient since this
needs to start at a chunk boundry.
But, unlike rsync, this method will work for S3, WebDAV, external
special remotes, etc, etc. Only directory special remotes so far,
but many more soon!
This implementation will also properly handle starting a download
from one remote, interrupting, and resuming from another one, and so on.
(Resuming interrupted chunked uploads is similarly doable, although
slightly more expensive.)
This commit was sponsored by Thomas Djärv.
When chunk=0, always try the unchunked key first. This avoids the overhead
of needing to read the git-annex branch to find the chunkcount.
However, if the unchunked key is not present, go on and try the chunks.
Also, when removing a chunked key, update the chunkcounts even when
chunk=0.
No need to process each L.ByteString chunk, instead ask it to split.
Doesn't seem to have really sped things up much, but it also made the code
simpler.
Note that this does (and already did) buffer in memory. It seems that only
the directory special remote could take advantage of streaming chunks to
files w/o buffering, so probably won't add an interface to allow for that.
This will allow things like WebDAV to opean a single persistent connection
and reuse it for all the chunked data.
The crazy types allow for some nice code reuse.
Push it down from needing to be done in every Storer,
to being checked once inside ChunkedEncryptable.
Also, catch exceptions from PrepareStorer and PrepareRetriever,
just in case..
I'd have liked to keep these two concepts entirely separate,
but that are entagled: Storing a key in an encrypted and chunked remote
need to generate chunk keys, encrypt the keys, chunk the data, encrypt the
chunks, and send them to the remote. Similar for retrieval, etc.
So, here's an implemnetation of all of that.
The total win here is that every remote was implementing encrypted storage
and retrival, and now it can move into this single place. I expect this
to result in several hundred lines of code being removed from git-annex
eventually!
This commit was sponsored by Henrik Ahlgren.
Not yet used by any special remotes, but should not be too hard to add it
to most of them.
storeChunks is the hairy bit! It's loosely based on
Remote.Directory.storeLegacyChunked. The object is read in using a lazy
bytestring, which is streamed though, creating chunks as needed, without
ever buffering more than 1 chunk in memory.
Getting the progress meter update to work right was also fun, since
progress meter values are absolute. Finessed by constructing an offset
meter.
This commit was sponsored by Richard Collins.
Slightly tricky as they are not normal UUIDBased logs, but are instead maps
from (uuid, chunksize) to chunkcount.
This commit was sponsored by Frank Thomas.
Moved old legacy chunking code, and cleaned up the directory and webdav
remotes use of it, so when no chunking is configured, that code is not
used.
The config for new style chunking will be chunk=1M instead of chunksize=1M.
There should be no behavior changes from this commit.
This commit was sponsored by Andreas Laas.
It is useful to be able to specify an alternative git-annex-shell
program to execute on the remote, e.g., to run a version not on the
PATH. Use remote.<name>.annex-shell if specified, instead of the
default "git-annex-shell" i.e., first so-named executable on the
PATH.
recvkey was told it was receiving a HMAC key from a direct mode repo,
and that confused it into rejecting the transfer, since it has no way to
verify a key using that backend, since there is no HMAC backend.
I considered making recvkey skip verification in the case of an unknown
backend. However, that could lead to bad results; a key can legitimately be
in the annex with a backend that the remote git-annex-shell doesn't know
about. Better to keep it rejecting if it cannot verify.
Instead, made the gcrypt special remote not set the direct mode flag when
sending (and receiving) files.
Also, added some recvkey messages when its checks fail, since otherwise
all that is shown is a confusing error message from rsync when the remote
git-annex-shell exits nonzero.
This is a git-remote-gcrypt encrypted special remote. Only sending files
in to the remote works, and only for local repositories.
Most of the work so far has involved making initremote work. A particular
problem is that remote setup in this case needs to generate its own uuid,
derivied from the gcrypt-id. That required some larger changes in the code
to support.
For ssh remotes, this will probably just reuse Remote.Rsync's code, so
should be easy enough. And for downloading from a web remote, I will need
to factor out the part of Remote.Git that does that.
One particular thing that will need work is supporting hot-swapping a local
gcrypt remote. I think it needs to store the gcrypt-id in the git config of the
local remote, so that it can check it every time, and compare with the
cached annex-uuid for the remote. If there is a mismatch, it can change
both the cached annex-uuid and the gcrypt-id. That should work, and I laid
some groundwork for it by already reading the remote's config when it's
local. (Also needed for other reasons.)
This commit was sponsored by Daniel Callahan.
Cipher is now a datatype
data Cipher = Cipher String | MacOnlyCipher String
which makes more precise its interpretation MAC-only vs. MAC + used to
derive a key for symmetric crypto.
With the initremote parameters "encryption=pubkey keyid=788A3F4C".
/!\ Adding or removing a key has NO effect on files that have already
been copied to the remote. Hence using keyid+= and keyid-= with such
remotes should be used with care, and make little sense unless the point
is to replace a (sub-)key by another. /!\
Also, a test case has been added to ensure that the cipher and file
contents are encrypted as specified by the chosen encryption scheme.
/!\ It is to be noted that revoking a key does NOT necessarily prevent
the owner of its private part from accessing data on the remote /!\
The only sound use of `keyid-=` is probably to replace a (sub-)key by
another, where the private part of both is owned by the same
person/entity:
git annex enableremote myremote keyid-=2512E3C7 keyid+=788A3F4C
Reference: http://git-annex.branchable.com/bugs/Using_a_revoked_GPG_key/
* Other change introduced by this patch:
New keys now need to be added with option `keyid+=`, and the scheme
specified (upon initremote only) with `encryption=`. The motivation for
this change is to open for new schemes, e.g., strict asymmetric
encryption.
git annex initremote myremote encryption=hybrid keyid=2512E3C7
git annex enableremote myremote keyid+=788A3F4C
Introduced a new per-remote option 'annex-rsync-transport' to specify
the remote shell that it to be used with rsync. In case the value is
'ssh', connections are cached unless 'sshcaching' is unset.
Most remotes have meters in their implementations of retrieveKeyFile
already. Simply hooking these up to the transfer log makes that information
available. Easy peasy.
This is particularly valuable information for encrypted remotes, which
otherwise bypass the assistant's polling of temp files, and so don't have
good progress bars yet.
Still some work to do here (see progressbars.mdwn changes), but this
is entirely an improvement from the lack of progress bars for encrypted
downloads.
Unless highRandomQuality=false (or --fast) is set, use Libgcypt's
'GCRY_VERY_STRONG_RANDOM' level by default for cipher generation, like
it's done for OpenPGP key generation.
On the assistant side, the random quality is left to the old (lower)
level, in order not to scare the user with an enless page load due to
the blocking PRNG waiting for IO actions.
There was confusion in different parts of the progress bar code about
whether an update contained the total number of bytes transferred, or the
number of bytes transferred since the last update. One way this bug
showed up was progress bars that seemed to stick at zero for a long time.
In order to fix it comprehensively, I add a new BytesProcessed data type,
that is explicitly a total quantity of bytes, not a delta.
Note that this doesn't necessarily fix every problem with progress bars.
Particularly, buffering can now cause progress bars to seem to run ahead
of transfers, reaching 100% when data is still being uploaded.
Pass subcommand as a regular param, which allows passing git parameters
like -c before it. This was already done in the pipeing set of functions,
but not the command running set.
Files are now written to a tmp directory in the remote, and once all
chunks are written, etc, it's moved into the final place atomically.
For now, checkpresent still checks every single chunk of a file, because
the old method could leave partially transferred files with some chunks
present and others not.
Both the directory and webdav special remotes used to have to buffer
the whole file contents before it could be decrypted, as they read
from chunks. Now the chunks are streamed through gpg with no buffering.
Transfer info files are updated when the callback is called, updating
the number of bytes transferred.
Left unused p variables at every place the callback should be used.
Which is rather a lot..
This *almost* works.
Along the way, I noticed that the --uuid parameter was being accidentially
passed after the --, so that has never been actually used by
git-annex-shell to verify it's running in the expected repository. Oops. Fixed.
In order to record a semi-useful filename associated with the key,
this required plumbing the filename all the way through to the remotes'
storeKey and retrieveKeyFile.
Note that there is potential for deadlock here, narrowly avoided.
Suppose the repos are A and B. A sends file foo to B, and at the same
time, B gets file foo from A. So, A locks its upload transfer info file,
and then locks B's download transfer info file. At the same time,
B is taking the two locks in the opposite order. This is only not a
deadlock because the lock code does not wait, and aborts. So one of A or
B's transfers will be aborted and the other transfer will continue.
Whew!
Baked into the code was an assumption that a repository's git directory
could be determined by adding ".git" to its work tree (or nothing for bare
repos). That fails when core.worktree, or GIT_DIR and GIT_WORK_TREE are
used to separate the two.
This was attacked at the type level, by storing the gitdir and worktree
separately, so Nothing for the worktree means a bare repo.
A complication arose because we don't learn where a repository is bare
until its configuration is read. So another Location type handles
repositories that have not had their config read yet. I am not entirely
happy with this being a Location type, rather than representing them
entirely separate from the Git type. The new code is not worse than the
old, but better types could enforce more safety.
Added support for core.worktree. Overriding it with -c isn't supported
because it's not really clear what to do if a git repo's config is read, is
not bare, and is then overridden to bare. What is the right git directory
in this case? I will worry about this if/when someone has a use case for
overriding core.worktree with -c. (See Git.Config.updateLocation)
Also removed and renamed some functions like gitDir and workTree that
misused git's terminology.
One minor regression is known: git annex add in a bare repository does not
print a nice error message, but runs git ls-files in a way that fails
earlier with a less nice error message. This is because before --work-tree
was always passed to git commands, even in a bare repo, while now it's not.
This option avoids gpg key distribution, at the expense of flexability, and
with the requirement that all clones of the git repository be equally
trusted.
getConfig got a remote-specific config, and this confusing name caused it
to be used a couple of places that only were interested in global configs.
Rename to getRemoteConfig and make getConfig only get global configs.
There are no behavior changes here, but remote.<name>.annex-web-options
never actually worked (and per-remote web options is a very unlikely to be
useful case so I didn't make it work), so fix the documentation for it.
Locking is used, so that, if there are multiple git-annex processes
using a remote concurrently, the stop hook is only run by the last
process that uses it.
Ssh connection caching is now enabled automatically by git-annex. Only one
ssh connection is made to each host per git-annex run, which can speed some
things up a lot, as well as avoiding repeated password prompts. Concurrent
git-annex processes also share ssh connections. Cached ssh connections are
shut down when git-annex exits.
Note: The rsync special remote does not yet participate in the ssh
connection caching.
Many functions took the repo as their first parameter. Changing it
consistently to be the last parameter allows doing some useful things with
currying, that reduce boilerplate.
In particular, g <- gitRepo is almost never needed now, instead
use inRepo to run an IO action in the repo, and fromRepo to get
a value from the repo.
This also provides more opportunities to use monadic and applicative
combinators.
Avoid ever using read to parse a non-haskell formatted input string.
show :: Key is arguably still show abuse, but displaying Keys as filenames
is just too useful to give up.
