The J2SE NIO APIs require that FileChannel close the underlying file
descriptor if a thread is interrupted while it is inside of a read or
write operation on that channel. This is insane, because it means we
cannot share the file descriptor between threads. If a thread is in
the middle of the FileChannel variant of IO.readFully() and it
receives an interrupt, the pack will be automatically closed on us.
This causes the other threads trying to use that same FileChannel to
receive IOExceptions, which leads to the pack getting marked as
invalid. Once the pack is marked invalid, JGit loses access to its
entire contents and starts to report MissingObjectExceptions.
Because PackWriter must ensure that the chosen pack file stays
available until the current object's data is fully copied to the
output, JGit cannot simply reopen the pack when its automatically
closed due to an interrupt being sent at the wrong time. The pack may
have been deleted by a concurrent `git gc` process, and that open file
descriptor might be the last reference to the inode on disk. Once its
closed, the PackWriter loses access to that object representation, and
it cannot complete sending the object the client.
Fortunately, RandomAccessFile's readFully method does not have this
problem. Interrupts during readFully() are ignored. However, it
requires us to first seek to the offset we need to read, then issue
the read call. This requires locking around the file descriptor to
prevent concurrent threads from moving the pointer before the read.
This reduces the concurrency level, as now only one window can be
paged in at a time from each pack. However, the WindowCache should
already be holding most of the pages required to handle the working
set for a process, and its own internal locking was already limiting
us on the number of concurrent loads possible. Provided that most
concurrent accesses are getting hits in the WindowCache, or are for
different repositories on the same server, we shouldn't see a major
performance hit due to the more serialized loading.
I would have preferred to use a pool of RandomAccessFiles for each
pack, with threads borrowing an instance dedicated to that thread
whenever they needed to page in a window. This would permit much
higher levels of concurrency by using multiple file descriptors (and
file pointers) for each pack. However the code became too complex to
develop in any reasonable period of time, so I've chosen to retrofit
the existing code with more serialization instead.
Bug: 308945
Change-Id: I2e6e11c6e5a105e5aef68871b66200fd725134c9
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Merges the current head with one other commit.
In this first iteration the merge command supports
only fast forward and already up-to-date.
Change-Id: I0db480f061e01b343570cf7da02cac13a0cbdf8f
Signed-off-by: Stefan Lay <stefan.lay@sap.com>
Signed-off-by: Christian Halstrick <christian.halstrick@sap.com>
Signed-off-by: Chris Aniszczyk <caniszczyk@gmail.com>
The CommitCommand should take care to create a merge commit if the file
$GIT_DIR/MERGE_HEAD exists. It should then read the parents for the merge
commit out of this file. It should also take care that when commiting
a merge and no commit message was specified to read the message from
$GIT_DIR/MERGE_MSG.
Finally the CommitCommand should remove these files if the commit
succeeded.
Change-Id: I4e292115085099d5b86546d2021680cb1454266c
Signed-off-by: Christian Halstrick <christian.halstrick@sap.com>
The strings are externalized into the root resource bundles.
The resource bundles are stored under the new "resources" source
folder to get proper maven build.
Strings from tests are, in general, not externalized. Only in
cases where it was necessary to make the test pass the strings
were externalized. This was typically necessary in cases where
e.getMessage() was used in assert and the exception message was
slightly changed due to reuse of the externalized strings.
Change-Id: Ic0f29c80b9a54fcec8320d8539a3e112852a1f7b
Signed-off-by: Sasa Zivkov <sasa.zivkov@sap.com>
In close() method of SshFetchConnection and SshPushConnection
errorThread.join() can wait forever if JSch will not close the
channel's error stream. Join with a timeout, and interrupt the
copy thread if its blocked on data that will never arrive.
Bug: 312863
Change-Id: I763081267653153eed9cd7763a015059338c2df8
Reported-by: Dmitry Neverov <dmitry.neverov@gmail.com>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we get an interrupt during an IO operation (src.read or dst.write)
caused by the flush() method incrementing the flush counter, ensure
we restart the proper section of code. Just ignore the interrupt
and continue running.
Bug: 313082
Change-Id: Ib2b37901af8141289bbac9807cacf42b4e2461bd
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The number of bytes to copy was truncated to an int, but the
pack's copyToStream() method expected to be passed a long here.
Pass through the long so we don't truncate a giant object.
Change-Id: I0786ad60a3a33f84d8746efe51f68d64e127c332
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Rather than holding onto the PackedObjectLoader, only hold the
PackFile and the object offset. During a reuse copy that is all
we should need to complete a reuse, and the other parts of the
PackedObjectLoader just waste memory.
This change reduces the per-object memory usage of a PackWriter by
32 bytes on a 32 bit JVM using only OFS_DELTA formatted objects.
The savings is even larger (by another 20 bytes) for REF_DELTAs.
This is close to a 50% reduction in the size of ObjectToPack,
making it rather worthwhile to do.
Beyond the memory reduction, this change will help to make future
refactoring work easier. We need to redo the API used to support
copying data, and disconnecting it from the PackedObjectLoader is
a good first step.
Change-Id: I24ba4e621e101f14e79a16463aec5379f447aa9b
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Rather than keep track of both the position of the object, and the
position of its data, just keep track of the number of bytes used
by the object's header in the pack. This shaves 4 bytes out of the
size of the PackedObjectLoader instances.
We also can defer the addition instruction to the materialize()
operation, avoiding it entirely if the caller never actually uses
the loader. This may be relevant for PackWriter invocations,
where only 1 loader gets chosen for a given object, even though
the object may appear on disk in more than one pack file.
Error reporting is now simplified, as we can rely on the object
offset rather than its data offset. This is the value displayed
by pack debugging tools like `git verify-pack -v`, so its better
to use that in our own errors.
Because nobody needs getDataOffset() now, we can drop that from
the public API.
Change-Id: Ic639c0d5a722315f4f5c8ffda6e26643d90e5f42
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Since we use this code twice, pull it into a private method. Let
the compiler/JIT worry about whether or not this logic should be
inlined into the call sites.
Change-Id: Ia44fb01e0328485bcdfd7af96835d62b227a0fb1
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Originally when I wrote this code I had hoped to use OffsetCache
to also implement the UnpackedObjectCache. But it turns out they
need rather different code, and it just wasn't worth trying to
reuse the OffsetCache base class.
Before doing any major refactoring or code cleanups here, squash the
two classes together and delete OffsetCache. As WindowCache is our
only subclass, this is pretty simple to do. We also get a minor
code reduction due to less duplication between the two classes,
and the JIT should be able to do a better job of optimization here
as we can define types up front rather than relying on generics
that erase back to java.lang.Object.
Change-Id: Icac8bda01260e405899efabfdd274928e98f3521
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
When we read the object header we copy 20 bytes from the pack data,
then start parsing out the type and the inflated size. For most
objects, this is only going to require 3 bytes, which is sufficient
to represent objects with inflated sizes of up to 2^16. The local
buffer however still has 17 bytes remaining in it, and that can be
used to satisfy the OBJ_OFS_DELTA header.
We shouldn't need to worry about walking off the end of the buffer
here, because delta offsets cannot be larger than 64 bits, and that
requires only 9 bytes in the OFS_DELTA encoding.
Assuming worst-case scenarios of 9 bytes for the OFS_DELTA encoding,
the pack file itself must be approaching 2^64 bytes, an infeasible
size to store on any current technology. However, even if this
were the case we still have 11 bytes for the type/size header.
In that encoding we can represent an object as large as 2^74 bytes,
which is also an infeasible size to process in JGit.
So drop the second read here.
The data offsets we pass into the ObjectLoaders being constructed
need to be computed individually now. This saves a local variable,
but pushes the addition operation into each branch of the switch.
Change-Id: I6cf64697a9878db87bbf31c7636c03392b47a062
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
JSch may hang or abort with the timeout if JGit connects before
its obtained the streams. Instead defer the connect() call until
after the streams have been configured.
Bug: 312383
Change-Id: I7c3a687ba4cb69a41a85e2b60d381d42b9090e3f
Reported-by: Dmitry Neverov <dmitry.neverov@gmail.com>
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If a flush() gets delivered at the same time that we are blocking
while writing to an interruptable stream, the copy thread will
abort assuming its a stream error. Instead ignore the interrupt,
and retry the write.
Change-Id: Icbf62d1b8abe0fabbb532dbee088020eecf4c6c2
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
It is possible to miss flush() invocation in StreamCopyThread.
In this case some data will not be sent to remote host and we will
wait forever (or until timeout) in src.read().
Use a counter to keep track of the flush requests.
Change-Id: Ia818be9b109a1674d9e2a9c78e125ab248cfb75b
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Added a new package org.eclipse.jgit.api and a builder-style API for
jgit. Added also the first implementation for two git commands: Commit
and Log.
This API is intended to be used by external components when
functionalities of the standard git commands are required. It will also
help to ease writing JGit tests.
For internal usages this API may often not be optimal because the git
commands are doing much more than required or they expect parameters of
an unappropriate type.
Change-Id: I71ac4839ab9d2f848307eba9252090c586b4146b
Signed-off-by: Christian Halstrick <christian.halstrick@sap.com>
Currently, if the Index contains a file in more than one stage, only
the last entry (containing the highest stage) will be registered in
GitIndex. For applications it can be useful to not only know about the
highest stage, but also which other stages are present, e.g. to detect
the type of conflict the file is in.
Change-Id: I2d4ff9f6023335d9ba6ea25d8e77c8e283ae53cb
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
The repository state tells in which state the repo is and also which actions
are currently allowed. The state MERGING is telling that a commit is not
possible. But this is only true in the case of unmerged paths in the index.
When we are merging but have resolved all conflicts then we are in a special
state: We are still merging (means the next commit should have multiple
parents) but a commit is now allowed.
Since the MERGING state "canCommit()" cannot be enhanced to return true/false
based on the index state (MERGING is an enum value which does not have a
reference to the repository its state it is representing) I had to introduce a new
state MERGING_RESOLVED. This new state will report that a commit is possible.
CAUTION: there might be the chance that users of jgit previously blindly did a
plain commit (with only one parent) when the RepositoryState allowed them to
do so. With this change these users will now be confronted with a RepositoryState
which says a commit is possible but before they can commit they'll have to
check the MERGE_MESSAGE and MERGE_HEAD files and use the info from these
files.
Change-Id: I0a885e2fe8c85049fb23722351ab89cf2c81a431
Signed-off-by: Christian Halstrick <christian.halstrick@sap.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
If two keys are the same length, but don't share the same sequence
of characters, we were incorrectly claiming they still matched due
to a bug in the for loop condition. I used the wrong variable and
the loop never executed, resulting in equality anytime the two keys
being compared were the same length.
Use the proper local variable to loop through the arrays, and add
a JUnit test to verify equality works as expected.
Change-Id: I4a02400e65a9b2e0da925b05a2cc4b579e1dd33a
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If a loose object was corrupted by truncation, JGit would hang.
Change-Id: I7e4c14f44183a5fcb37c1562e81682bddeba80ad
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
It is incorrect to use Eclipse.org as the providerName now,
we'll use Eclipse JGit.
Change-Id: I1621b93d4f401176704e7c43935a5ce0c8ee8419
Signed-off-by: Chris Aniszczyk <caniszczyk@gmail.com>
If a concurrent thread picks up a newly created PackFile and adds
it to the pack list before the IndexPack thread itself can insert
the item onto the front of the list, do nothing and use the item
that was picked up by that other concurrent scanning thread.
This avoids a potential condition where the same pack exists in
memory twice, which causes confusion later during a rescan of the
directory because we don't know exactly which PackFile instance
should be retained into the new list, and which should be discarded.
We can stop searching through the old pack list as soon as the
sort function declares that the item to insert should be before
the item already in the list. Because the list is always sorted
by modification time (in seconds), we should never encounter a
case where the pack is positioned at the wrong spot in the list.
This early break out still permits an efficient implementation of
the common case, inserting a new pack at the head of the list.
Change-Id: Ice4459bbd4ee9487078aff5257893883d04f05fb
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
There is a potential race condition during insertPack that can lead
to us having the same pack file open twice in the same directory.
A different thread can miss an object on disk, and trigger a scan
of the directory, and notice the pack that was put in by IndexPack.
So the pack winds up in the newly created PackList.
The IndexPack thread then wakes up and finishes its insertPack by
creating a new PackFile and inserting it into position 0 of the list.
We now have the same pack listed twice.
Readers will favor the earlier PackFile instance, because its the
first one they come across as they iterate through the list.
Keep that earlier one when we scan the pack directory again, as
this will avoid needing to purge out all of the windows that may
have been cached.
Of course we should also fix that race condition, but this block
was taking the wrong resolution if this error ever shows up, so
lets first fix the block to use a more sane resolution.
Change-Id: I0d339b9fd1dd8012e8fe5a564b893c0f69109e28
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This reuse line was identical between the two branches related to
reusing a delta, or reusing a whole object. Either way they reuse
the body of the object as-is. So just make that a common function
after the header is written.
Change-Id: I0e6673b8e813c8c08c594ea2ba546fd366339d5d
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If a corrupt loose object is read, UnpackedObjectLoader was disposing
of the Inflater, and then attempting to return the disposed Inflater
to the InflaterCache. Since the disposed Inflater had its native
libz resource deallocated and its reference cleared out, the Inflater
threw NullPointerException and refused to reset itself before being
put back into the cache.
Instead of disposing of the Inflater when corruption is found, do
nothing, and allow it to be returned to the cache. The instance
will get reset, and should be usable by a future caller.
Bug: 310291
Change-Id: I44f2247c08b6e04fa62f8399609341b07508c096
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This option was mis-named from day 1. Its not checking that the
objects provided by the client are reachable, its actually doing
a scan to prove that objects referenced by the client are already
reachable through another reference on the server, or were sent
as part of the pack from the client.
Rename it checkReferencedObjectsAreReachable, since we really are
trying to validate that objects referenced by the client's actions
are reachable to the client.
We also need to ensure we run checkConnectivity() anytime this is
enabled, even if the caller didn't turn on fsck for object formats.
Otherwise the check would be completely bypassed.
Change-Id: Ic352ddb0ca8464d407c6da5c83573093e018af19
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we are checking the visibility of everything referenced in the
pack that isn't already reachable by a reference, it needs to be
in the provided set. Since the provided set lists everything that
is in this pack, we can avoid checking to see if the blob exists
on disk, because we know it should be there, it was found in the
pack we just consumed.
Change-Id: Ie3c7746f734d13077242100a68e048f1ac18c34a
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If a tree was referenced but not provided in the pack, report it
as a missing tree and not as a missing blob.
Change-Id: Iab05705349cdf0d30cc3f8afc6698a8d2a941343
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The only current consumer of these collections is ReceivePack,
where it needs to test ObjectId equality between a RevObject and an
ObjectId. There we were copying from a traditional HashSet<ObjectId>
into an ObjectIdSubclassMap<ObjectId>, as the latter can perform
hashing using ObjectId's native value support, bypassing RevObject's
override on hashCode() and equals(). Instead of doing that copy,
directly create ObjectIdSubclassMap instances inside of ReceivePack.
We also only need to record the objects that do not appear in the
incoming pack, and were therefore copied from the local repositiory
in order to complete delta resolution. Instead of listing everything
that used an OBJ_REF_DELTA format, list only the objects that we
pulled from the destination repository via a normal ObjectLoader.
ReceivePack can now discard the IndexPack object, and all of its
other data, as soon as these collections are held by the check
connectivity method. This frees up memory for the ObjectWalk's
own RevObject pool.
Change-Id: I22ef71b45c2045a0202e7fd550a770ee1f6f38a6
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
These are more like internal implementation details of how IndexPack
works with ReceivePack to validate the incoming object stream.
Callers who are embedding the ReceivePack logic in their own
application don't really need to know the details of which objects
were used for delta bases in the incoming thin pack, or exactly
which objects were newly transmitted.
Hide these from the API, as exposing them through ReceivePack was
an early mistake.
Change-Id: I7ee44a314fa19e6a8520472ce05de92c324ad43e
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The IndexPack object carries a good bit of state within itself about
the objects received over the wire. The earlier we can discard it,
the sooner the GC is able to reclaim this chunk of memory for other
uses. So drop it as soon as we are certain the pack is valid and we
have no connectivity concerns.
Change-Id: I1e8bc87c2e9183733043622237a064e55957891f
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If ensureProvidedObjectsVisible is enabled we expected any trees or
blobs directly reachable from an advertised reference to be marked
with UNINTERESTING. Unfortunately ObjectWalk doesn't bother setting
this until the traversal is complete. Even then it won't necessarily
set it on every tree if the corresponding commit wasn't popped.
When we are going to check the base objects for the received pack,
ensure the UNINTERESTING flag gets carried into every immediately
reachable tree or blob, because these are the ones that the client
might try to use as delta bases in a thin pack.
Change-Id: I5d5fdcf07e25ac9fc360e79a25dff491925e4101
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The Iterator contract says next() shall throw NoSuchElementException
if there are no more items remaining in the iteration. We got this
wrong when I originally wrote the implementation, so fix it.
Change-Id: Iea25e6569ead5c8b3128b8a368c5b2caebec7ecc
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This class behaves like a cross between a Set and a Map, sometimes
we might expect to use the method isEmpty() to test for size() == 0.
So implement it, reducing the surprise folks get when they are given
one of these objects.
Change-Id: I0d68e1243da8e62edf79c6ba4fd925f643e80a88
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we need to append less than 20 bytes in order to fix a thin pack
and make it complete, we need to set the length of our file back to
the actual number of bytes used because the original SHA-1 footer was
not completely overwritten. That extra data will confuse the header
and footer fixup logic when it tries to read to the end of the file.
This isn't a very common case to occur, which is why we've never
seen it before. Getting a delta that requires a whole object which
uses less than 20 bytes in pack representation is really hard.
Generally a delta generator won't make these, because the delta
would be bigger than simply deflating the whole object. I only
managed to do this with a hand-crafted pack file where a 1 byte
delta was pointed to a 1 byte whole object.
Normally we try really hard to avoid truncating, because its
typically not safe across network filesystems. But the odds of
this occurring are very low. This truncation is done on a file
we have open for writing, will append more content onto, and is
a temporary file that we won't move into position for others to
see until we've validated its SHA-1 is sane. I don't think the
truncate on NFS issue is something we need to worry about here.
Change-Id: I102b9637dfd048dc833c050890d142f43c1e75ae
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
We can avoid one stat call by trying to perform a directory
listing without checking if the reference File is a directory.
Attempting a directory listing is defined to return. The other
case for null returns from list is when an I/O error occcurs.
Both cases are now intepreted as a possible plain reference. I/O
errors when reading plain references will be handled (ignored)
in scanRef().
Change-Id: I9906ed8c42eab4d6029c781aab87b3b07c1a1d2c
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
In the current implementation Repository reads user and repository
config only at creation point of time.
The new implementatiopn checks in Repository.getConfig if user or
repository config have changed on disk and reload the config if
required.
Change-Id: Ibd97515919ef66c6f8aa1a4fe8a11a6711335dad
Signed-off-by: Jens Baumgart <jens.baumgart@sap.com>
We only need to check file existense if some other stat returns
a value that may mean that the file does not exist. File.length() == 0
or File.lastModified() == 0 are two such properties. We use length
here.
Change-Id: If626b12e7bb4da994b5c086f6a5b7a12c187261c
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
The JSch bundle in Eclipse 3.4 does not export its packages with
version numbers. Use Require-Bundle on version 0.1.37 that comes
with Eclipse 3.4
There is no 0.1.37 in the maven repositories so the pom still refers
to 0.1.41 so the build can get the compile time dependencies right.
Bug: 308031
CQ: 3904 jsch Version: 0.1.37 (using Orbit CQ2014)
Change-Id: I12eba86bfbe584560c213882ebba58bf1f9fa0c1
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
When listing branches, EGit only reads the advertisement and
then disconnects. When it closes down the pack channel the remote
side is waiting for the client to send our list of commands, or a
flush-pkt to let it know there is nothing to do.
However if an error thread is open watching the SSH stderr stream,
we ask for it to finish before we send the flush-pkt. Unfortunately
the thread won't terminate until the main output stream closes,
which is waiting for the flush-pkt. A classic network deadlock.
If the output stream needs a flush-pkt we send it before we wait
for the error stream to close. If the flush-pkt is rejected, we
close down the output stream early, assuming that the remote side
is broken and we will get error information soon.
Change-Id: I8d078a339077756220c113f49d206b1bf295d434
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Since the API is changing relative to 0.7.0, we'll call our next
release 0.8.1. But until that gets released, builds from master
will be 0.8.0.qualifier.
Change-Id: I921e984f51ce498610c09e0db21be72a533fee88
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
By default a receive pack assumes that its user will only provide
references to objects that the user already has access to on their
local client. In certain cases, an additional check to verify the
references point only to reachable objects is necessary.
This additional checking is useful when the code doesn't trust
the client not to provide a forged SHA-1 reference to an object,
in an attempt to access parts of the DAG that they weren't allowed
to see by the configured RefFilter.
Change-Id: I3e4b8505cb2992e3e4be253abb14a1501e47b970
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Alex Blewitt
Shawn O. Pearce
Stefan Lay
Christian Halstrick
Sasa Zivkov
Dmitry Neverov
Robin Rosenberg
Chris Aniszczyk
Jens Baumgart
Marc Strapetz
Nico Sallembien