BatchRefUpdate: repro racy atomic update, and fix it
PackedBatchRefUpdate was creating a new packed-refs list that was
potentially unsorted. This would be papered over when the list was
read back from disk in parsePackedRef, which detects unsorted ref
lists on reading, and sorts them. However, the BatchRefUpdate also
installed the new (unsorted) list in-memory in
RefDirectory#packedRefs.
With the timestamp granularity code committed to stable-5.1, we can
more often accurately decide that the packed-refs file is clean, and
will return the erroneous unsorted data more often. Unluckily timed
delays also cause the file to be clean, hence this problem was
exacerbated under load.
The symptom is that refs added by a BatchRefUpdate would stop being
visible directly after they were added. In particular, the Gerrit
integration tests uses BatchRefUpdate in its setup for creating the
Admin group, and then tries to read it out directly afterward.
The tests recreates one failure case. A better approach would be to
revise RefList.Builder, so it detects out-of-order lists and
automatically sorts them.
Fixes https://bugs.eclipse.org/bugs/show_bug.cgi?id=548716 and
https://bugs.chromium.org/p/gerrit/issues/detail?id=11373.
Bug: 548716
Change-Id: I613c8059964513ce2370543620725b540b3cb6d1
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Keep track of the original cause for a packfile invalidation.
It is needed for the sysadmin to understand if there is a real
underlying filesystem problem and repository corruption or if it is
simply a consequence of a concurrency of Git operations (e.g. repack
or GC).
Change-Id: I06ddda9ec847844ec31616ab6d17f153a5a34e33
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Signed-off-by: David Pursehouse <david.pursehouse@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Fix pack files scan when filesnapshot isn't modified
Do not reload packfiles when their associated filesnapshot is not
modified on disk compared to the one currently stored in memory.
Fix the regression introduced by fef78212 which, in conjunction with
core.trustfolderstats = false, caused any lookup of objects inside
the packlist to loop forever when the object was not found in the pack
list.
Bug: 546190
Change-Id: I38d752ebe47cefc3299740aeba319a2641f19391
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Fix GC to delete empty fanout directories after repacking
The prune method did not delete empty fanout directories when loose
objects moved to a new pack file but only when loose unreferenced
objects were pruned.
Change-Id: Ia068f4914c54d9cf9f40b75e8ea50759402b5000
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
When reading from a packfile, make sure that is valid
and has a non-null file-descriptor.
Because of concurrency between a thread invalidating a packfile
and another trying to read it, the read() may result into a NPE
that won't be able to be automatically recovered.
Throwing a PackInvalidException would instead cause the packlist
to be refreshed and the read to eventually succeed.
Bug: 544199
Change-Id: I27788b3db759d93ec3212de35c0094ecaafc2434
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Move throw of PackInvalidException outside the catch
When a packfile is invalid, throw an exception explicitly
outside any catch scope, so that is not accidentally caught
by the generic catch-all cause, which would set the packfile
as valid again.
Flagging an invalid packfile as valid again would have
dangerous consequences such as the corruption of the in-memory
packlist.
Bug: 544199
Change-Id: If7a3188a68d7985776b509d636d5ddf432bec798
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Do not redundantly call File.lastModified() for extracting the
timestamp of the PackFile but rather use consistently the FileSnapshot
which reads all file attributes in a single bulk call.
Change-Id: I932675ae4fe56dcd3833dac249816f097303bb09
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Due to finite filesystem timestamp resolution the last modified
timestamp of files cannot detect file changes which happened in the
immediate past (less than one filesystem timer tick ago).
Read and consider file size also, so that differing file size can help
to more accurately detect file changes without reading the file content.
Use bulk read to avoid multiple stat calls to retrieve file attributes.
Change-Id: I974288fff78ac78c52245d9218b5639603f67a46
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
The pack reload mechanism from the filesystem works only by name
and does not check the actual last modified date of the packfile.
This lead to concurrency issues where multiple threads were loading
and removing from each other list of packfiles when one of those
was failing the checksum.
Rely on FileSnapshot rather than directly checking lastModified
timestamp so that more checks can be performed.
Bug: 544199
Change-Id: I173328f29d9914007fd5eae3b4c07296ab292390
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
In case of concurrent pack file access, threads may wait on the idx()
function even for already open files. This happens especially with a
slow file system.
Performance numbers are listed in the bug report.
Bug: 543739
Change-Id: Iff328d347fa65ae07ecce3267d44184161248978
Signed-off-by: Juergen Denner <j.denner@sap.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
PackFile: report correct message for checksum mismatch
When the packfile checksum does not match the expected one
report the correct checksum error instead of reporting that
the number of objects is incorrect.
Change-Id: I040f36dacc4152ae05453e7acbf8dfccceb46e0d
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
(cherry picked from commit 436c99ce59)
Externalize the message and log the pack file with absolute path.
Change-Id: I019052dfae8fd96ab67da08b3287d699287004cb
Signed-off-by: David Pursehouse <david.pursehouse@gmail.com>
(cherry picked from commit 9665d86ba1)
ObjectDirectory: extra logging on packfile exceptions
Display extra logging, including the exception with the associated
stacktrace, whenever a packFile can't be read and thus removed
from the packlist.
Change-Id: I97a4e31dc427bfcc0baae438dcbe2dcd4704b824
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
(cherry picked from commit 962babc4b2)
The AdvertiseRefsHook can be called twice if the following conditions
hold:
1. This AdvertiseRefsHook doesn't set this.refs.
2. getAdvertisedOrDefaultRefs is called after getFilteredRefs.
For example, this can happen when fetchV2 is called after lsRefsV2
when using a stateful bidirectional transport.
The second call does not accomplish anything useful. Guard it with
'if (!advertiseRefsHookCalled)' to avoid wasted work.
Reported-by: Jonathan Tan <jonathantanmy@google.com>
Change-Id: Ib746582e4ef645b767a5b3fb969596df99ac2ab5
Signed-off-by: Jonathan Nieder <jrn@google.com>
AdvertiseRefsHook is used to limit the visibility of the refs in Gerrit.
If this hook is not called, then all refs are treated as visible.
In protocol v2, the hook is not called, causing the server to advertise
all refs. This bug was introduced in v5.0.0.201805221745-rc1~1^2~9
(Execute AdvertiseRefsHook only for protocol v0 and v1, 2018-05-14).
Even before then, the hook was not called in requests after the
capability advertisement, so in transports like HTTP that do not retain
state between round-trips, the server would advertise all refs in
response to an ls-refs (ls-remote) request.
Fix both cases by using getAdvertisedOrDefaultRefs to retrieve the
advertised refs in lsRefs, ensuring the hook is called in all cases that
use its result.
[jn: backported to stable-5.0; split out from a larger patch that also
fixes protocol v0; avoided filtering this.refs by ref prefix]
Change-Id: I64bce0e72d15b90baccc235c067e57b6af21b55f
Signed-off-by: Masaya Suzuki <masayasuzuki@google.com>
Signed-off-by: Jonathan Nieder <jrn@google.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
AdvertiseRefsHook is used to limit the visibility of the refs in Gerrit.
If this hook is not called, then all refs are treated as visible,
causing the server to serve commits reachable from branches the client
should not be able to access, if asked to via a request naming a guessed
object id.
This bug was introduced in v2.0.0.201206130900-r~123 (Modify refs in
UploadPack/ReceivePack using a hook interface, 2012-02-08). Stateful
bidirectional transports are not affected.
Fix it by moving the AdvertiseRefsHook call to
getAdvertisedOrDefaultRefs, ensuring the hook is called in all cases.
[jn: backported to stable-4.5 by splitting out tests and the protocol v2
specific parts]
Change-Id: I159f396216354f2eda3968d17802e166d8c8ec2d
Signed-off-by: Masaya Suzuki <masayasuzuki@google.com>
Signed-off-by: Jonathan Nieder <jrn@google.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
when multiple match options are given in git describe the result must
not depend on the order of the match options. JGit wrongly picked the
first match using the match options in the order they were defined. Fix
this by concatenating the streams of matching tags for all match options
and then choosing the first match on the concatenated stream sorted in
tie break order.
See https://git-scm.com/docs/git-describe#git-describe---matchltpatterngt
Change-Id: Id01433d35fa16fb4c30526605bee041ac1d954b2
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Correct behaviour as git 1.7.1.1 is to resolve tie-breakers to choose
the most recent tag.
https://github.com/git/git/blob/master/Documentation/RelNotes/1.7.1.1.txt:
* "git describe" did not tie-break tags that point at the same commit
correctly; newer ones are preferred by paying attention to the
tagger date now.
Bug: 538610
Change-Id: Ib0b2a301997bb7f75935baf7005473f4de952a64
Signed-off-by: Håvard Wall <haavardw@gmail.com>
A .gitmodules file can include a submodule without a path to configure
the URL for a submodule that is only present on other branches.
A .gitmodules file can include a submodule with no URL and no path to
reserve the name for a submodule that existed in earlier history but
is not available from any URL any more.
"git fsck" permits both of these cases. Permit them in JGit as well
(instead of throwing NullPointerException).
Change-Id: I3b442639ad79ea7a59227f96406a12e62d3573ae
Reported-by: David Pursehouse <david.pursehouse@gmail.com>
Signed-off-by: Jonathan Nieder <jrn@google.com>
The text "<tree, blob>" with angle brackets should not be used in javadoc
since it is interpreted as an HTML tag and then rejected since it's not a
valid HTML tag. Wrap the text in a @literal tag.
Also add a missing space.
Change-Id: Ide045e8c04a39a916f5b2e964e58c151e4555830
Signed-off-by: David Pursehouse <david.pursehouse@gmail.com>
The main concern are submodule urls starting with '-' that could pass as
options to an unguarded tool.
Pass through the parser the ids of blobs identified as .gitmodules
files in the ObjectChecker. Load the blobs and parse/validate them
in SubmoduleValidator.
Change-Id: Ia0cc32ce020d288f995bf7bc68041fda36be1963
Signed-off-by: Ivan Frade <ifrade@google.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
ObjectChecker: Report .gitmodules files found in the pack
In order to validate .gitmodules files, we first need to find them
in the incoming pack.
Do it in the ObjectChecker stage. Check in the tree objects if they
point to a .gitmodules file and report the tree id and the .gitmodules
blob id.
This can be used later to check if the file is in the root of the
project and if the contents are good.
While we're here, make isMacHFSGit more accurate by detecting variants
of filenames that vary in case.
[jn: tweaked NTFS and HFS+ checking; added more tests]
Change-Id: I70802e7d2c1374116149de4f89836b9498f39582
Signed-off-by: Ivan Frade <ifrade@google.com>
Signed-off-by: Jonathan Nieder <jrn@google.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
SubmoduleAddCommand: Reject submodule URIs that look like cli options
In C git versions before 2.19.1, the submodule is fetched by running
"git clone <uri> <path>". A URI starting with "-" would be interpreted
as an option, causing security problems. See CVE-2018-17456.
Refuse to add submodules with URIs, names or paths starting with "-",
that could be confused with command line arguments.
[jn: backported to JGit 4.7.y, bringing portions of Masaya Suzuki's
dotdot check code in v5.1.0.201808281540-m3~57 (Add API to specify
the submodule name, 2018-07-12) along for the ride]
Change-Id: I2607c3acc480b75ab2b13386fe2cac435839f017
Signed-off-by: Ivan Frade <ifrade@google.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
This happened if the LockTokens hard link was already deleted earlier.
Bug: 531759
Change-Id: Idc84bd695fac1a763b3cbb797c9c4c636a16e329
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Externalize warning message in RefDirectory.delete()
Change-Id: Icec16c01853a3f5ea016d454b3d48624498efcce
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
(cherry picked from commit 5e68fe245f)
Signed-off-by: David Pursehouse <david.pursehouse@gmail.com>
Suppress warning for trying to delete non-empty directory
This is actually a fairly common occurrence; deleting the parent
directories can work only if the file deleted was the last one
in the directory.
Bug: 537872
Change-Id: I86d1d45e1e2631332025ff24af8dfd46c9725711
Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
(cherry picked from commit d9e767b431)
Signed-off-by: David Pursehouse <david.pursehouse@gmail.com>
FS_POSIX.createNewFile(File) failed to properly implement atomic file
creation on NFS using the algorithm [1]:
- name of the hard link must be unique to prevent that two processes
using different NFS clients try to create the same link. This would
render nlink useless to detect if there was a race.
- the hard link must be retained for the lifetime of the file since we
don't know when the state of the involved NFS clients will be
synchronized. This depends on NFS configuration options.
To fix these issues we need to change the signature of createNewFile
which would break API. Hence deprecate the old method
FS.createNewFile(File) and add a new method createNewFileAtomic(File).
The new method returns a LockToken which needs to be retained by the
caller (LockFile) until all involved NFS clients synchronized their
state. Since we don't know when the NFS caches are synchronized we need
to retain the token until the corresponding file is no longer needed.
The LockToken must be closed after the LockFile using it has been
committed or unlocked. On Posix, if core.supportsAtomicCreateNewFile =
false this will delete the hard link which guarded the atomic creation
of the file. When acquiring the lock fails ensure that the hard link is
removed.
[1] https://www.time-travellers.org/shane/papers/NFS_considered_harmful.html
also see file creation flag O_EXCL in
http://man7.org/linux/man-pages/man2/open.2.html
Change-Id: I84fcb16143a5f877e9b08c6ee0ff8fa4ea68a90d
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Fix handling of option core.supportsAtomicCreateNewFile
When core.supportsAtomicCreateNewFile was set to false and the
repository was located on a filesystem which doesn't support the file
attribute "unix:nlink" then FS_POSIX#createNewFile may report an error
even if everything was ok. Modify FS_POSIX#createNewFile to silently
ignore this situation. An example of such a filesystem is sshfs where
reading "unix:nlink" always returns 1 (instead of throwing a exception).
Bug: 537969
Change-Id: I6deda7672fa7945efa8706ea1cd652272604ff19
Also-by: Thomas Wolf <thomas.wolf@paranor.ch>
GC: Avoid logging errors when deleting non-empty folders
I88304d34c and Ia555bce00 modified the way errors are handled when
trying to delete non-empty reference folders. Before, this error was
silently ignored as it was considered an expected output. Now, every
failed folder delete is logged which can be noisy.
Ignore the DirectoryNotEmptyException but log any other error avoiding
deletion of an eligible folder.
Signed-off-by: Hector Oswaldo Caballero <hector.caballero@ericsson.com>
Change-Id: I194512f67885231d62c03976ae683e5cc450ec7c
Fix NoSuchFileException in GC.deleteTempPacksIdx()
This exception is thrown in GC.deleteTempPacksIdx() if the repository
has no packs.
Bug: 538286
Change-Id: Ieb482be751226baf0843068a0f847e0cdc6e0cb6
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Suppress warning for trying to delete non-empty directory
This is actually a fairly common occurrence; deleting the parent
directories can work only if the file deleted was the last one
in the directory.
Bug: 537872
Change-Id: I86d1d45e1e2631332025ff24af8dfd46c9725711
Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
If packed refs are used, duplicate updates result in an exception
because JGit tries to lock the same lock file twice. With non-atomic
ref updates, this used to work, since the same ref would simply be
locked and updated twice in succession.
Let's be more lenient in this case and remove duplicates before
trying to do the ref updates. Silently skip duplicate updates
for the same ref, if they both would update the ref to the same
object ID. (If they don't, behavior is undefined anyway, and we
still throw an exception.)
Add a test that results in a duplicate ref update for a tag.
Bug: 529400
Change-Id: Ide97f20b219646ac24c22e28de0c194a29cb62a5
Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
Fetch(Process): should tolerate duplicate refspecs
Bug: 529314
Change-Id: I91eaeda8a988d4786908fba6de00478cfc47a2a2
Signed-off-by: Marc Strapetz <marc.strapetz@syntevo.com>
Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
Since I3870cadb4, GC task was always delegated to an executor even when
background option was set to false. This was an issue because if more
than one GC object was instantiated and executed in parallel, only one GC
was actually running because of the single thread executor.
Change-Id: I8c587d22d63c1601b7d75914692644a385cd86d6
Signed-off-by: Hugo Arès <hugo.ares@ericsson.com>
Ensure that JSch knows HostKeyAlgorithms ssh-rsa and ssh-dss
Without these registrations, JSch's up-front checks which algorithms
are available at all fail if the ssh config explicitly sets only these
algorithms.
Bug: 537790
Change-Id: Idb0431190a7f101913363ee95af6c8fcbda6c923
Signed-off-by: Thomas Wolf <thomas.wolf@paranor.ch>
Remove completely the empty directories under refs/<namespace>
including the first level partition of the changes, when they are
completely empty.
Bug: 536777
Change-Id: I88304d34cc42435919c2d1480258684d993dfdca
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
Use java.nio to delete path to get detailed errors
Get the full IOException of the reason why a directory
cannot be removed during GC.
Change-Id: Ia555bce009fa48087a73d677f1ce3b9c0b685b57
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Signed-off-by: Matthias Sohn <matthias.sohn@sap.com>
After packaging references, the folders containing these references are
not deleted. In a busy repository, this causes operations to slow down
as traversing the references tree becomes longer.
Delete empty reference folders after the loose references have been
packed.
To avoid deleting a folder that was just created by another concurrent
operation, only delete folders that were not modified in the last 30
seconds.
Signed-off-by: Hector Oswaldo Caballero <hector.caballero@ericsson.com>
Change-Id: Ie79447d6121271cf5e25171be377ea396c7028e0
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
Signed-off-by: David Pursehouse <david.pursehouse@gmail.com>
Log as warning when an attempt to remove a directory
fails. This helps troubleshooting some bugs like the GC leaving
behind empty directories.
Change-Id: Idb94ce17f8be9668a970c7ecae31436bf434073c
Signed-off-by: Luca Milanesio <luca.milanesio@gmail.com>
ResolveMerger: Fix encoding with string; use bytes
This change fixes the issue [1]. Before this fix, a merge involving
the caching of consecutive yet similar filenames with Norwegian
characters [2] used to throw an IllegalStateException: Duplicate
stages not allowed. This was caused by inaccurate decoding of the
filenames, using string values assuming default encoding. In the
toString method of DirCacheEntry, used before through getPathString,
UTF-8 encoding is used, but the end result becomes default encoding,
through Object's default toString usage. The special characters in
those two consecutive (particular) filenames [2] were becoming the
very same decoded /single character, lending consecutive -but then
identical- filenames. Thus the perceived duplicate 0-staging of the
file(s).
Replace getPathString usage with getRawPath for this specific case,
or use byte array representations of cached entries instead of string.
Adding a test for this change is not possible, as there is no known
way to change the default encoding for filenames such as [2] (e.g.).
JGitTestUtil does write file contents through UTF-8, but encoding like
so does not apply to the actual file name. Hence there is no way to
create files with names properly made of special characters such as
[2]'s. And the test that is necessary for this case assumes such
Norwegian (or similar characters) filenames. Changing the default
locale programmatically in a test has no effect either. And changing
the LANG value passed to the JVM is only possible upon starting it.
[1] https://bugs.chromium.org/p/gerrit/issues/detail?id=9153
[2] <=>
(...)
"a/b/SíÒr-Norge.map",
"a/b/Sør-Norge.map",
(...)
Change-Id: Ib9f2f5297932337c9817064cc09d9f774dd168f4
Signed-off-by: Marco Miller <marco.miller@ericsson.com>
If I run
git config --global protocol.version 2
mkdir repo
cd repo
git init --bare
git remote add origin https://go.googlesource.com/proposal
git fetch --depth=1
git fetch --unshallow
then I expect to have a full history, just as though I had fetched
without --depth in the first place. Instead, it reports success
but does not fetch enough objects:
$ git fsck
notice: HEAD points to an unborn branch (master)
Checking object directories: 100% (256/256), done.
Checking objects: 100% (468/468), done.
broken link from commit 2c6bc83f23
to commit 878975cf2b
missing commit 878975cf2b
dangling commit 314be00dae
The false success indicates problems in the client and the server.
Git 2.18-rc2 (the client) ought to have been more defensive, noticing
the incomplete history. The greater error is in JGit (the server),
which neglects to send the objects requested.
When serving protocol v0 requests, JGit sends the correct objects by
taking unshallowCommits into account when generating the pack to send
to the client. Do the same in the protocol v2 code path. I forgot to
do this in v5.0.0.201806050710-rc3~6 (Teach UploadPack shallow fetch
in protocol v2, 2018-03-15).
Reported-by: Russ Cox <rsc@golang.org>
Change-Id: I282b45f47616a641b9e8d6210b4a070d3efdbb9b
Signed-off-by: Jonathan Nieder <jrn@google.com>
Han-Wen Nienhuys
Luca Milanesio
Matthias Sohn
Juergen Denner
David Pursehouse
Jonathan Nieder
Masaya Suzuki
Håvard Wall
Ivan Frade
Thomas Wolf
Christian Halstrick
Hector Caballero
Marc Strapetz
Hugo Arès
Marco Miller