prop_encode_decode_roundtrip failed on "\175" in C locale.
This may be a new problem after the switch to RawFilePath, but it
already had filtering for high chars, so changed to only test ascii
chars.
eg, `git-annex get . ..` used to order the files strangly, because it
did not realize that when git ls-files output eg "foo", that should be
grouped with the first set of files and not the second set.
Fixed by making dirContains "." "./foo" = True
which makes sense, because dirContains ".." "../foo" = True
the encode' and decode' functions on Windows should not apply the
filesystem encoding, which does not work there. Instead, convert to and
from UTF-8.
Also, avoid exporting encodeW8 and decodeW8. Both use the filesystem
encoding, so won't work as expected on windows.
git-annex find is now RawFilePath end to end, no string conversions.
So is git-annex get when it does not need to get anything.
So this is a major milestone on optimisation.
Benchmarks indicate around 30% speedup in both commands.
Probably many other performance improvements. All or nearly all places
where a file is statted use RawFilePath now.
Since the sqlite branch uses blobs extensively, there are some
performance benefits, ByteStrings now get stored and retrieved w/o
conversion in some cases like in Database.Export.
Only done on those calls to getFileStatus that had a RawFilePath, not a
FilePath. The others would probably be just as fast if converted to use
it with toRawFilePath, but I'm not 100% sure.
Note that genInodeCache' uses fromRawFilePath, but that value only gets
used on Windows, so on unix the thunk will never be evaluated.
File mode is octal not decimal. This broke in the conversion to
attoparsec.
(I've submitted the content of Utility.Attoparsec to the attoparsec
developers.)
Test suite passes 100% now.
Finally builds (oh the agoncy of making it build), but still very
unmergable, only Command.Find is included and lots of stuff is badly
hacked to make it compile.
Benchmarking vs master, this git-annex find is significantly faster!
Specifically:
num files old new speedup
48500 4.77 3.73 28%
12500 1.36 1.02 66%
20 0.075 0.074 0% (so startup time is unchanged)
That's without really finishing the optimization. Things still to do:
* Eliminate all the fromRawFilePath, toRawFilePath, encodeBS,
decodeBS conversions.
* Use versions of IO actions like getFileStatus that take a RawFilePath.
* Eliminate some Data.ByteString.Lazy.toStrict, which is a slow copy.
* Use ByteString for parsing git config to speed up startup.
It's likely several of those will speed up git-annex find further.
And other commands will certianly benefit even more.
Goal is to make git-annex faster by using ByteString for all the
worktree traversal. For now, this is focusing on Command.Find,
in order to benchmark how much it helps. (All other commands are
temporarily disabled)
Currently in a very bad unbuildable in-between state.
This will speed up the common case where a Key is deserialized from
disk, but is then serialized to build eg, the path to the annex object.
Previously attempted in 4536c93bb2
and reverted in 96aba8eff7.
The problems mentioned in the latter commit are addressed now:
Read/Show of KeyData is backwards-compatible with Read/Show of Key from before
this change, so Types.Distribution will keep working.
The Eq instance is fixed.
Also, Key has smart constructors, avoiding needing to remember to update
the cached serialization.
Used git-annex benchmark:
find is 7% faster
whereis is 3% faster
get when all files are already present is 5% faster
Generally, the benchmarks are running 0.1 seconds faster per 2000 files,
on a ram disk in my laptop.
Used to work but was broken in version 7.20181031, specifically commit
5ab0f48ffb.
That this was not noticed over at least 1 daylight savings time zone
changes makes me wonder if the TSDelta stuff is still needed.
Perhaps the mtime on Windows no longer changes when the time zone is changed?
(cherry picked from commit 09ee6b0ccb)
Eliminated some dead code. In other cases, exported a currently unused
function, since it was a logical part of the API.
Of course this improves the API documentation. It may also sometimes
let ghc optimize code better, since it can know a function is internal
to a module.
364 modules still to go, according to
git grep -E 'module [A-Za-z.]+ where'
Convert Utility.Url to return Either String so the error message can be
displated in the annex monad and so captured.
(When curl is used, its errors are still not caught.)
Used to work but was broken in version 7.20181031, specifically commit
5ab0f48ffb.
That this was not noticed over at least 1 daylight savings time zone
changes makes me wonder if the TSDelta stuff is still needed.
Perhaps the mtime on Windows no longer changes when the time zone is changed?
The only good thing about it is it does not require a major version bump
to improve the database. That will need to happen at some point though.
Potentially very very slow in a large repository.
Ugly use of raw sql.
gksu is no longer in debian, even stable
kdesu in debian is not installed in PATH any longer, though the executable
is still present under /usr/lib
pkexec is packagekit's replacement for those older commands.
That git fixed a memory leak that could cause an OOM during the upgrade.
Most git-annex builds have a new enough git already.
OSX git was upgraded with brew.
Linux i386ancient build's git was too old. Upgrading it to a fixed
git didn't work (due to the newer git not working with the old ssh,
https://bugs.chromium.org/p/git/issues/detail?id=7 )
Choices to deal with that were:
* Somehow make direct mode upgrade work with the old git, avoiding its
OOM problem. One way would be to switch the repo to indirect mode
first, and so upgrade to a repo with locked files. Not good when
the filesystem does not support symlinks.
* backport the OOM fix from git 2.22
(And do what about the version number so git-annex knows it's fixed?)
* backport openssh (and possibly more stuff)
* move the i386ancient build to at least Debian stretch (still backporting git)
But this will make it no longer work with some of the ancient kernels it
targets.
Of those, backporting the OOM fix seemed the best approach. Put "oomfix"
in the git version number to indicate it.
I have not automated building the git backport, so here's the patch I
used:
diff -ur orig/git-2.1.4/convert.c git-2.1.4/convert.c
--- orig/git-2.1.4/convert.c 2014-12-18 18:42:18.000000000 +0000
+++ git-2.1.4/convert.c 2019-08-29 20:05:04.371872338 +0100
@@ -404,7 +404,7 @@
if (start_async(&async))
return 0; /* error was already reported */
- if (strbuf_read(&nbuf, async.out, len) < 0) {
+ if (strbuf_read(&nbuf, async.out, 0) < 0) {
error("read from external filter %s failed", cmd);
ret = 0;
}
diff -ur orig/git-2.1.4/GIT-VERSION-GEN git-2.1.4/GIT-VERSION-GEN
--- orig/git-2.1.4/GIT-VERSION-GEN 2014-12-18 18:42:18.000000000 +0000
+++ git-2.1.4/GIT-VERSION-GEN 2019-08-29 20:06:39.132743228 +0100
@@ -1,7 +1,7 @@
#!/bin/sh
GVF=GIT-VERSION-FILE
-DEF_VER=v2.1.4
+DEF_VER=v2.1.4.oomfix
LF='
'
diff -ur orig/git-2.1.4/configure git-2.1.4/configure
--- orig/git-2.1.4/configure 2014-12-18 18:42:19.000000000 +0000
+++ git-2.1.4/configure 2019-08-29 20:27:45.896380015 +0100
@@ -580,8 +580,8 @@
# Identity of this package.
PACKAGE_NAME='git'
PACKAGE_TARNAME='git'
-PACKAGE_VERSION='2.1.4'
-PACKAGE_STRING='git 2.1.4'
+PACKAGE_VERSION='2.1.4.oomfix'
+PACKAGE_STRING='git 2.1.4.oomfix'
PACKAGE_BUGREPORT='git@vger.kernel.org'
PACKAGE_URL=''
diff -ur orig/git-2.1.4/version git-2.1.4/version
--- orig/git-2.1.4/version 2014-12-18 18:42:19.000000000 +0000
+++ git-2.1.4/version 2019-08-29 20:06:17.572545210 +0100
@@ -1 +1 @@
-2.1.4
+2.1.4.oomfix
I'm seeing the github lfs server request an upload of an object that has
already been uploaded to it before. Probably because they offload
storage to S3 and so skipped the overhead of checking for an unncessary
upload.
Let's keep this entirely pure.
git-annex has its own facilities for running a ssh command, that make it
respect various config settings, and cache connections, etc. So better
not to have the library run ssh itself.
In 40ecf58d4b I changed the license of code I
wrote from GPL to AGPL. But, two files containing code I wrote combined
with code by others were updated to say their license is AGPL, while in
fact part of it was (the code I wrote) but part remained under the original
license (the code written by others).
Remote/Ddar.hs is now changed entirely back to GPL 3.
Annex/DirHashes.hs stays AGPL, but I broke out Utility/MD5.hs with the code
not written by me, and corrected its license statement to GPL-2, which
is the actual version of the GPL included with the code in its original
distribution at http://www.cs.ox.ac.uk/people/ian.lynagh/md5/
I made some improvements to its API after splitting it out of git-annex,
so merge those back in.
This is groundwork for removing the embedded copy of it and depending on
it.
Also moved the managerResponseTimeout disabling to Annex.Url as it's
git-annex specific.
This commit was sponsored by Ethan Aubin on Patreon.
Avoid statting file, just try to remove it.
Also a comment to explain why it tries to remove it, which was puzzling
me when I revisited this code until I saw that cp fails to overwrite a
mode 444 file, including perhaps one left by a previous interrupted cp.
This commit was sponsored by Fernando Jimenez on Patreon.
Improved probing when CoW copies can be made between files on the same
drive. Now supports CoW between BTRFS subvolumes. And, falls back to rsync
instead of using cp when CoW won't work, eg copies between repos on the
same EXT4 filesystem.
Rather than trying cp --reflink=always for each file copied to a remote,
it's tried once and if it fails it falls back to using rsync thereafter
for the lifetime of the Remote object. That avoids overhead of calling cp
which while small, will add up over a large number of files.
This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
using a blake2 variant optimised for 4-way CPUs
This had been deferred because the Debian package of cryptonite, and
possibly other builds, was broken for blake2bp, but I've confirmed #892855
is fixed.
This commit was sponsored by Brett Eisenberg on Patreon.
Drop support for building with ghc older than 8.4.4, and with older
versions of serveral haskell libraries than will be included in Debian 10.
The only remaining version ifdefs in the entire code base are now a couple
for aws!
This commit should only be merged after the Debian 10 release.
And perhaps it will need to wait longer than that; it would make
backporting new versions of git-annex to Debian 9 (stretch) which
has been actively happening as recently as this year.
This commit was sponsored by Ilya Shlyakhter.
When downloading an url and the destination file exists but is empty,
avoid using http range to resume, since a range "bytes=0-" is an unusual
edge case that it's best to avoid relying on working.
This is known to fix a case where importfeed downloaded a partial feed from
such a server. Since importfeed uses withTmpFile, the destination always exists
empty, so it would particularly tickle such problem servers. Resuming from 0
is otherwise possible, but unlikely.
Add back support for ftp urls, which was disabled as part of the fix for
security hole CVE-2018-10857 (except for configurations which enabled curl
and bypassed public IP address restrictions). Now it will work if allowed
by annex.security.allowed-ip-addresses.
The NonEmpty instance was moved out of QuickCheck and into a package
with more deps than I want to drag in, so I'm providing my own instance,
but with older QuickCheck, use theirs to avoid overlapping.
This does not change the overall license of the git-annex program, which
was already AGPL due to a number of sources files being AGPL already.
Legally speaking, I'm adding a new license under which these files are
now available; I already released their current contents under the GPL
license. Now they're dual licensed GPL and AGPL. However, I intend
for all my future changes to these files to only be released under the
AGPL license, and I won't be tracking the dual licensing status, so I'm
simply changing the license statement to say it's AGPL.
(In some cases, others wrote parts of the code of a file and released it
under the GPL; but in all cases I have contributed a significant portion
of the code in each file and it's that code that is getting the AGPL
license; the GPL license of other contributors allows combining with
AGPL code.)
An empty list of [ContentIdenfier] serialized to the same thing
as a single ContentIdentifier "". Avoid this ambiguity by requiring the
list be non-empty.
It got removed from network-3.0.0.0 and nothing in the haskell ecosystem
currently provides it (which seems it ought to be fixed).
Tested new code on both little-endian and big-endian with:
ghci> hostAddressToTuple $ fromJust $ embeddedIpv4 (0,0,0,0,0,0xffff,0x7f00,1)
(127,0,0,1)
The gitAnnexTmpOtherDir cleanup made it be deleted too early sometimes,
and so the test suite failed. Also there was a report of a similar
failure which likely had a similar cause and hopwfully this fixes that
too.
This reverts commit 4536c93bb2.
That broke Read/Show of a Key, and unfortunately Key is read in at least
one place; the GitAnnexDistribution data type.
It would be worth bringing this optimisation back, but it would need
either a custom Read/Show instance that preserves back-compat, or
wrapping Key in a data type that contains the serialization, or changing
how GitAnnexDistribution is serialized.
Also, the Eq instance would need to compare keys with and without a
cached seralization the same.
The builder produces a lazy ByteString, and L.toStrict has to copy it,
but needing to use the builder is no longer to common case; the
serialization will normally be cached already as a strict ByteString,
and this avoids keyFile' needing to use L.toStrict . serializeKey'
Now there's a ByteString used all the way from disk to Key.
The main complication in this conversion was the use of fromInternalGitPath
in several places to munge things on Windows. The things that used that
were changed to parse the ByteString using either path separator.
Also some code that had read from files to a String lazily was changed
to read a minimal strict ByteString.
MetaField was already limited to alphanumerics, so it makes sense to use
Text for it.
Note that technically a UUID can contain invalid UTF-8, and so
remoteMetaDataPrefix's use of T.pack . fromUUID could replace non-UTF8
values with '?' or whatever. In practice, a UUID is usually also text,
I only kept open the possibility of it containing invalid UTF-8 to avoid
breaking parsing of strange UUIDs in git-annex branch files. So, I
decided to let this edge case slip by.
Have not updated the rest of the code base yet for this change, as the
change took 2.5 hours longer than I expected to get working properly.
This is not as efficient as using ByteStrings throughout, but converting
the String to ByteString is actually significantly faster than the old
parser.
benchmarking parse/old
time 9.657 μs (9.600 μs .. 9.732 μs)
1.000 R² (0.999 R² .. 1.000 R²)
mean 9.703 μs (9.645 μs .. 9.785 μs)
std dev 231.6 ns (161.5 ns .. 323.7 ns)
variance introduced by outliers: 25% (moderately inflated)
benchmarking parse/new
time 834.6 ns (797.1 ns .. 886.9 ns)
0.987 R² (0.976 R² .. 0.999 R²)
mean 816.4 ns (802.7 ns .. 845.1 ns)
std dev 62.39 ns (37.66 ns .. 108.4 ns)
variance introduced by outliers: 82% (severely inflated)
There is a small behavior change from the old parsePOSIXTime,
which accepted any amount of trailing whitespace after the timestamp.
That behavior was not documented, and it doesn't seem anything relied on it.
ghc warned that the guards did not cover all values of h, but they
clearly do, and when rewritten as a case statement the warning goes
away.
Probably a ghc bug, but I kind of prefer the case statement over the
guards anyway.
Deleting directories is one of the great unsolved problems of CS, thanks to
abominations like NFS lock files and Windows and races with other processes
cleaning up after themselves in the background. The gpg test harness
sometimes failed to delete its temp directory on NFS. Avoid the problem
class by not deleting it at all, and putting it inside the tmp repo being
tested. The test suite's more robust (and/or nonsensical) workarounds for
deleting its test dir will thus be used, hopefully avoiding the problem
until an OS finds a new way to violate POSIX and the laws of nature.
Note that this means that the .gnupg directory will be on whatever
filesystem the test suite is being run on, which may be a lesser quality
filesystem than gpg is really expecting. Gpg does not seem to need to
write sockets etc to there so this seems ok. The only known problem is
that if the filesystem forces a directory mode like 777, gpg will warn
about unsafe home directory perms, but it still works.
Don't much like that there's no way to distinguish between having the whole
content and having an old version of the file that's bigger, but of course
resuming a http transfer can always yield the wrong result if the file on
the http server is changing, and git-annex will detect that when it
verifies the downloaded content.
This work is supported by the NIH-funded NICEMAN (ReproNim TR&D3) project.
Cache high-resolution mtimes for improved detection of modified files in v7
(and direct mode).
Including on Windows.
With back-compat support so old low-res mtimes won't break anything, and
so the new information also won't break old versions of git-annex.
Luckily, this did not affect any git-annex log files, since they all
include the trailing 's' for backwards compatability reasons.
But, if I later want to drop that, this is the first commit where
git-annex can be trusted to parse that right.
The misparse caused it to be off by up to 10 seconds.
I've seen intermittent failures of the test suite with v6 for a long time,
it seems to have possibly gotten worse with the changes around v7. Or just
being unlucky; all tests failed today.
Seen on amd64 and i386 builders, repeatedly but intermittently:
unused: FAIL (4.86s)
Test.hs:928:
git diff did not show changes to unlocked file
And I think other such failures, all involving v7/v6 mode tests.
I managed to reproduce the unused failure with --keep-failures,
and inside the repo, git diff was indeed not showing any changes for
the modified unlocked file.
The two stats will be the same other than mtime; the old and new files have
the same size and inode, since the test case writes to the file and then
overwrites it.
Indeed, notice the identical timestamps:
builder@orca:~/gitbuilder/build/.t/tmprepo335$ echo 1 > foo; stat foo; echo 2 > foo; stat foo
File: foo
Size: 2 Blocks: 8 IO Block: 4096 regular file
Device: 801h/2049d Inode: 3546179 Links: 1
Access: (0644/-rw-r--r--) Uid: ( 1000/ builder) Gid: ( 1000/ builder)
Access: 2018-10-29 22:14:10.894942036 +0000
Modify: 2018-10-29 22:14:10.894942036 +0000
Change: 2018-10-29 22:14:10.894942036 +0000
Birth: -
File: foo
Size: 2 Blocks: 8 IO Block: 4096 regular file
Device: 801h/2049d Inode: 3546179 Links: 1
Access: (0644/-rw-r--r--) Uid: ( 1000/ builder) Gid: ( 1000/ builder)
Access: 2018-10-29 22:14:10.894942036 +0000
Modify: 2018-10-29 22:14:10.898942036 +0000
Change: 2018-10-29 22:14:10.898942036 +0000
Birth: -
I'm seeing this in Linux VMs; it doesn't happen on my laptop. I've also
not experienced the intermittent test suite failures on my laptop.
So, I hope that this small delay will avoid the problem.
Update: I didn't, indeed I then reproduced the same failure on my
laptop, so it must be due to something else. But keeping this change anyway
since not needing to worry about lowish-resolution mtime in the test suite seems
worthwhile.
Install new git hooks in this version.
This does beg the question of what to do if git later gets eg a
post-smudge hook, that could run git-annex smudge --update. I think the
thing to do in that case would be to make git-annex smudge --update
install the new hooks. That way, as the user uses git-annex, the hook
would be created pretty quickly and without needing any extra syscalls
except for when git-annex smudge --update is called.
I considered doing something like that for installation of the
post-checkout and post-merge hooks, which would have avoided the need
for v7. But the only place it was cheap to do it would be in git-annex smudge
which could cheaply notice that smudge.log didn't exist yet and so know
the hooks needed to be installed. But since smudge used to populate pointer
files, it would be quite surprising if a single git checkout/merge failed
to update the work tree, and so that idea didn't work out.
The other reason for v7 is psychological -- users don't need to worry
about whether they might be running an old version of git-annex that
doesn't support their v7 repository very well. And bug reports about
"v6" have gotten a bit of a bad association in my head since they often
hit one of the known limitations and didn't realize it was experimental.
newtyped RepoVersion Int to avoid needing 2 comparisons in
versionSupportsUnlockedPointers etc. Also it's just nicer.
This commit was sponsored by John Pellman on Patreon.
Running git-annex linux builds in termux seems to work well enough that the
only reason to keep the Android app would be to support Android 4-5, which
the old Android app supported, and which I don't know if the termux method
works on (although I see no reason why it would not).
According to [1], Android 4-5 remains on around 29% of devices, down from
51% one year ago.
[1] https://www.statista.com/statistics/271774/share-of-android-platforms-on-mobile-devices-with-android-os/
This is a rather large commit, but mostly very straightfoward removal of
android ifdefs and patches and associated cruft.
Also, removed support for building with very old ghc < 8.0.1, and with
yesod < 1.4.3, and without concurrent-output, which were only being used
by the cross build.
Some documentation specific to the Android app (screenshots etc) needs
to be updated still.
This commit was sponsored by Brett Eisenberg on Patreon.
The error message displayed used to only come from curl/wget and perhaps
was clearer than the one displayed now that http-client is used. In any
case, it does make sense to hide it because git-annex prints its own
warning message.
This commit was sponsored by Jake Vosloo on Patreon.
download is documented as displaying an error when download fails, but
it didn't when the url was not valid at all. That leads to confusing
behavior.
Also, display the url with --debug
Untested, on FreeBSD but enough to fix the listed build errors.
Seems that System.Posix.Files must have used to export this stuff and it
was split.
This commit was sponsored by Peter on Patreon.
Work around git cat-file --batch's protocol not supporting newlines by
running git cat-file not batched and passing the filename as a
parameter.
Of course this is quite a lot less efficient, especially because it
currently runs it multiple times to query for different pieces of
information.
Also, it has subtly different behavior when the batch process was
started and then some changes were made, in which case the batch process
sees the old index but this workaround sees the current index. Since
that batch behavior is mostly a problem that affects the assistant and has
to be worked around in it, I think I can get away with this difference.
I don't know of any other problems with newlines in filenames, everything
else in git I can think of supports -z. And git-annex's json output
supports newlines in filenames so downstream parsers from git-annex will be ok.
git-annex commands that use --batch themselves don't support newlines
in input filenames; using --json --batch is currently a way around that
problem.
This commit was sponsored by Ewen McNeill on Patreon.
When git-annex used wget and curl, --debug would show urls. So there can't
be any new security problem with doing so.
This commit was sponsored by John Pellman on Patreon.
Avoids "git-annex-shell: <stdin>: hGetChar: end of file"
being displayed by the test suite, due to the way it
runs git-annex-shell without using ssh.
git-annex-shell over ssh was not affected because git-annex hangs up the
ssh connection and so never sees the error message that git-annnex-shell
probably did emit.
This commit was sponsored by Ryan Newton on Patreon.
In 2013, I wrote "Cryptohash benchmarks 90 to 101% faster than external
hashers". Re-benchmarking today, I found cryptonite's sha256 consistently
outperformed coreutils by 10% for large files. Tested 10 mb, 100 mb, 1 gb
files with both sha256 and sha512. And for smaller files, the external
process startup time swamps the hash time.
Perhaps cryptonite has improved. Or it could just do better on my
current CPU Intel(R) Pentium(R) CPU 4410Y @ 1.50GHz). Anyway, even if cryptonite
is slower in some situations, seems likely it would only be marginally slower;
it's got the same class of highly optimised C code under the hood as coreutils.
The main difference between the two sha256 implementations seems to be
how much of the inner loop they unroll..
This commit was sponsored by Henrik Riomar on Patreon.
Switched code to use a for loop to avoid a filterM that would have
doubled the memory used.
This commit was supported by the NSF-funded DataLad project.
Switch to using http-client for large file downloads caused the reversion;
the code for displaying a 404 response was instead displaying the raw html
document, which is not useful.
This commit was sponsored by Ryan Newton on Patreon.
Send User-Agent and any configured annex.http-headers when downloading with
http, fixes reversion introduced when switching to http-client.
This commit was sponsored by mo on Patreon.
This avoid a build problem when different versions of posix and
posixcompat are used. Does not normally happen as cabal prevents that,
but this is sometimes used with ghc --make which can get into that
situation.
p2p --pair: Fix interception of the magic-wormhole pairing code, which
since 0.8.2 it has sent to stderr rather than stdout.
This is highly annoying because I had asked the magic wormhole developers
for a machine-readable way to get the data, and instead they changed how
the data was output, and didn't even mention this in my issue, or in the
changelog.
Seems this needs to be tested periodically to make sure it's still working.
This commit was sponsored by Ethan Aubin.
They're no worse than http certianly. And, the backport of these
security fixes has to deal with wget, which supports http https and ftp
and has no way to turn off individual schemes, so this will make that
easier.
A local http proxy would bypass the security configuration. So,
the security configuration has to be applied when choosing whether to
use the proxy.
While http rebinding attacks against the dns lookup of the proxy IP
address seem very unlikely, this implementation does prevent them, since
it resolves the IP address once, checks it, and then reconfigures
http-client's proxy using the resolved address.
This commit was sponsored by Ole-Morten Duesund on Patreon.
Security fix!
* git-annex will refuse to download content from http servers on
localhost, or any private IP addresses, to prevent accidental
exposure of internal data. This can be overridden with the
annex.security.allowed-http-addresses setting.
* Since curl's interface does not have a way to prevent it from accessing
localhost or private IP addresses, curl defaults to not being used
for url downloads, even if annex.web-options enabled it before.
Only when annex.security.allowed-http-addresses=all will curl be used.
Since S3 and WebDav use the Manager, the same policies apply to them too.
youtube-dl is not handled yet, and a http proxy configuration can bypass
these checks too. Those cases are still TBD.
This commit was sponsored by Jeff Goeke-Smith on Patreon.
Would be nice to add CIDR notation to this, but this is the minimal
thing needed for the security fix.
This commit was sponsored by Ewen McNeill on Patreon.
For use in a security boundary enforcement.
Based on https://en.wikipedia.org/wiki/Reserved_IP_addresses
Including supporting IPv4 addresses embedded in IPv6 addresses. Because
while RFC6052 3.1 says "Address translators MUST NOT translate packets
in which an address is composed of the Well-Known Prefix and a non-
global IPv4 address; they MUST drop these packets", I don't want to
trust that implementations get that right when enforcing a security
boundary.
This commit was sponsored by John Pellman on Patreon.
This is a clean way to add IP address restrictions to http-client, and
any library using it.
See https://github.com/snoyberg/http-client/issues/354#issuecomment-397830259
Some code from http-client and http-client-tls was copied in and
modified. Credited its author accordingly, and used the same MIT license.
The restrictions don't apply to http proxies. If using http proxies is a
problem, http-client already has a way to disable them.
SOCKS support is not included. As far as I can tell, http-client-tls
does not support SOCKS by default, and so git-annex never has.
The additional dependencies are free; git-annex already transitively
depended on them via http-conduit.
This commit was sponsored by Eric Drechsel on Patreon.
Security fix! Allowing any schemes, particularly file: and
possibly others like scp: allowed file exfiltration by anyone who had
write access to the git repository, since they could add an annexed file
using such an url, or using an url that redirected to such an url,
and wait for the victim to get it into their repository and send them a copy.
* Added annex.security.allowed-url-schemes setting, which defaults
to only allowing http and https URLs. Note especially that file:/
is no longer enabled by default.
* Removed annex.web-download-command, since its interface does not allow
supporting annex.security.allowed-url-schemes across redirects.
If you used this setting, you may want to instead use annex.web-options
to pass options to curl.
With annex.web-download-command removed, nearly all url accesses in
git-annex are made via Utility.Url via http-client or curl. http-client
only supports http and https, so no problem there.
(Disabling one and not the other is not implemented.)
Used curl --proto to limit the allowed url schemes.
Note that this will cause git annex fsck --from web to mark files using
a disallowed url scheme as not being present in the web. That seems
acceptable; fsck --from web also does that when a web server is not available.
youtube-dl already disabled file: itself (probably for similar
reasons). The scheme check was also added to youtube-dl urls for
completeness, although that check won't catch any redirects it might
follow. But youtube-dl goes off and does its own thing with other
protocols anyway, so that's fine.
Special remotes that support other domain-specific url schemes are not
affected by this change. In the bittorrent remote, aria2c can still
download magnet: links. The download of the .torrent file is
otherwise now limited by annex.security.allowed-url-schemes.
This does not address any external special remotes that might download
an url themselves. Current thinking is all external special remotes will
need to be audited for this problem, although many of them will use
http libraries that only support http and not curl's menagarie.
The related problem of accessing private localhost and LAN urls is not
addressed by this commit.
This commit was sponsored by Brett Eisenberg on Patreon.
Prevent haskell http-client from decompressing gzip files, so downloads of
such files works the same as it used to with wget and curl.
Explicitly setting accept-encoding to "identity" is probably not needed,
but that's what wget sends (curl does not send the header), and since
http-client is trying to be excessively smart, it seems we need to set
hAcceptEncoding to something to prevent it from inserting its own,
and this seems better than some hack like "".
This commit was sponsored by Ole-Morten Duesund on Patreon.
* Display error message when http download fails.
There's nothing in the http-client library to nicely format a http
exception, so in some cases it has to fall back to using show on it.
Seems better than just saying "it failed" or only showing the http
status code.
* Avoid forward retry when 0 bytes were received.
forwardRetry was comparing Nothing to Just 0, and so thought there had
been progress made when 0 bytes were received.
This commit was supported by the NSF-funded DataLad project.
Switch to Data.Map.Strict everywhere that used it.
There are still lots of lazy maps in git-annex. I think switching these
is safe. The risk is that there might be a map that is used in a way
that relies on the values not being evaluated to WHNF, and switching to
strict might result in bad performance or memory use. So, I have not
switched everything.
As long as all code imports Utility.Aeson rather than Data.Aeson,
and no Strings that may contain utf-8 characters are used for eg, object
keys via T.pack, this is guaranteed to fix the problem everywhere that
git-annex generates json.
It's kind of annoying to need to wrap ToJSON with a ToJSON', especially
since every data type that has a ToJSON instance has to be ported over.
However, that only took 50 lines of code, which is worth it to ensure full
coverage. I initially tried an alternative approach of a newtype FileEncoded,
which had to be used everywhere a String was fed into aeson, and chasing
down all the sites would have been far too hard. Did consider creating an
intentionally overlapping instance ToJSON String, and letting ghc fail
to build anything that passed in a String, but am not sure that wouldn't
pollute some library that git-annex depends on that happens to use ToJSON
String internally.
This commit was supported by the NSF-funded DataLad project.
* For url downloads, git-annex now defaults to using a http library,
rather than wget or curl. But, if annex.web-options is set, it will
use curl. To use the .netrc file, run:
git config annex.web-options --netrc
* git-annex no longer uses wget (and wget is no longer shipped with
git-annex builds).
Note that curl is always run in silent mode, since the new API for
download has a MeterUpdate and doesn't make way for curl progress
output. It might be worth writing a parser for curl's progress output
to update the meter when using it, but I didn't bother with this edge
case for now.
This commit was supported by the NSF-funded DataLad project.
Remote.S3 and Remote.Helper.Http both had similar code to sink a
http-conduit Response to a file; refactor out sinkResponseFile.
downloadC downloads an url to a file using http-conduit, and supports
resuming. Falls back to curl to handle urls that http-conduit does not
support. This is not used yet, but the goal is to replace download with
it.
git-annex.cabal: conduit-extra was not actually used for a long time,
remove the dep. conduit moves into the main dependency list, but since
http-conduit was already in there, and it depends on conduit, that's not
really adding a new build dep.
This commit was supported by the NSF-funded DataLad project.
Enable HTTP connection reuse across multiple files, when git-annex
uses http-conduit. Before, a new Manager was created each time
Utility.Url used it. Now, a single Manager gets created the first time,
so connections are reused.
Doesn't help when external programs are used for url download,
but does speed up addurl --fast, fsck --from web, etc.
Testing fsck --fast --from web with 3 files, over high-latency
satellite internet, it sped up from 19.37s to 14.96s.
This commit was supported by the NSF-funded DataLad project.
The pipe's FDs got inherited by ssh and it did something that kept them
open even once it exited. Probably involving passing them on to the ssh
mux daemon.
Set close on exec, and all is well.
Kept Annex.Ssh not using processTranscript even though it no longer
hangs when it does use it, just because processTranscript is overkill
there.
This commit was supported by the NSF-funded DataLad project.
This is much clearer to follow.
I've tested this, and it still has the problem described in
doc/bugs/occasional_hang_with_p2pstdio.mdwn
Which I think indicates that problem is not with my code, but something
else. ghc runtime? Something crazy ssh does in this situation? Unsure..
Some blake hash varieties were not yet available in that version.
Rather than tracking exact details of what cryptonite supported when,
disable blake unless using a current cryptonite.
There are a lot of different variants and sizes, I suppose we might as well
export all the common ones.
Bump dep to cryptonite to 0.16, earlier versions lacked BLAKE2 support.
Even android has 0.16 or newer.
On Debian, Blake2bp_512 is buggy, so I have omitted it for now.
http://bugs.debian.org/892855
This commit was sponsored by andrea rota.
Noticed that getting a key whose size is not known resulted in a
progress display that didn't include the percent complete.
Fixed for P2P by making the size sent with DATA be used to update the
meter's total size.
In order for rateLimitMeterUpdate to also learn the total size,
had to make it be passed the Meter, and some other reorg in
Utility.Metered was also done so that --json-progress can construct a
Meter to pass to rateLimitMeterUpdate.
When the fallback rsync is done, the progress display still doesn't
include the percent complete. Only way to fix that seems to be to let rsync
display its output again, but that would conflict with git-annex's
own progress meter, which is also being displayed.
This commit was sponsored by Henrik Riomar on Patreon.
Much like Remote.P2P, there's a pool of connections to a peer, in order
to support concurrent operations.
Deals with old git-annex-ssh on the remote that does not support p2pstdio,
by only trying once to use it, and remembering if it's not supported.
Made p2pstdio send an AUTH_SUCCESS with its uuid, which serves the dual
purposes of something to detect to see that the connection is working,
and a way to verify that it's connected to the right uuid.
(There's a redundant uuid check since the uuid field is sent
by git_annex_shell, but I anticipate that being removed later when
the legacy git-annex-shell stuff gets removed.)
Not entirely happy with Remote.Git.runSsh's behavior
when the proto action fails. Running the fallback will work ok, but what
will we do when the fallbacks later get removed? It might be better to
try to reconnect, in case the connection got closed.
This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
Fourth or fifth try at this and finally found a way to make it work.
Absurd amount of busy-work forced on me by change in cabal's behavior.
Split up Utility modules that need posix stuff out of ones used by
Setup. Various other hacks around inability for Setup to use anything
that ifdefs a use of unix.
Probably lost a full day of my life to this.
This is how build systems make their users hate them. Just saying.
This reverts commit 51228c2306.
No, still doesn't work when built with cabal. It did with stack; stack
must somehow make the unix package implicitly available.
With cabal, System.Posix.Process and System.Posix.Env are both missing.
Seems I had all the work in past commits to make this build, at least on
linux. I'm actually surprised it does, without a unix dep, Utility.Env
still builds ok somehow despite using System.Posix.Env.
This commit was sponsored by Fernando Jimenez on Patreon.
This avoids warnings from stack about the module not being listed in the
cabal file. So, the generated file is also renamed to Build/SysConfig.
Note that the setup program seems to be cached despite these changes; I
had to cabal clean to get cabal to update it so that Build/SysConfig was
written.
This commit was sponsored by Jochen Bartl on Patreon.
Now youtubeDlCheck downloads the beginning of the url's content and
checks if it's html, only when it is does it pass it off the youtube-dl
to check if it supports it.
This means more work is done for urls that youtube-dl does support,
but is probably more efficient for other urls, since it only downloads
the first chunk of content, while youtube-dl probably downloads more.
As well as the reported bug, this also fixes behavior when an url
was added with youtube-dl, but the url content has now changed from
a html page to something else. Remote.Web.checkKey used to wrongly
succeed in that situation, since youtube-dl said sure it can download
that something else.
This commit was supported by the NSF-funded DataLad project.
initremote, enableremote: Really support gpg subkeys suffixed with an
exclamation mark, which forces gpg to use a specific subkey. (Previous try
had a bug.)
This commit was sponsored by Jake Vosloo on Patreon.
Fully working, including --fast/--relaxed.
Note that, while git-annex addurl --relaxed is not going to check
youtube-dl, I kept git annex importfeed --relaxed checking it.
Thinking is that, let's not break people's importfeed cron jobs, and
importfeed does not typically have to check a large number of new items,
so it's ok if it's a little bit slower when used with youtube playlist
feeds.
importfeed's behavior is also improved (?) when a feed has links in it
to non-media files. Before, those were skipped. Now, the content of the
link is downloaded. This had to be done, because trying to use
youtube-dl is slow, and if those were skipped, it would have to check
every time importfeed was run. While this behavior change may not be
desirable for some feeds, that intersperse links to web pages with
enclosures, it will be desirable for other feeds, that have
non-enclosure directy links to media files.
Remove old quvi modules.
This commit was sponsored by Øyvind Andersen Holm.
This will be used in youtube-dl integration, to tell when a html page has
been downloaded by addurl, in which case it is worth running youtube-dl
to see if it can extract media from it.
tagsoup is an almost free dependency, because yesod depends on it.
So, this only really adds a dep when git-annex is built without the
webapp.
I'd like this to as closely as possible match how browsers decide if a
page is html or not. Unfortunately, that is fairly heuristic, in order
to support malformed html. And, we don't want to falsely detect
something as html just because it has something that looks like a html
tag embedded somewhere in it. Probably any major video hosting site is
going to be serving html documents that at least start with a <html>
tag, so requiring that or a DOCTYPE should be good enough.
This commit was sponsored by Jeff Goeke-Smith on Patreon.
They need unix on non-windows, for Utility.Env, which Build.Configure uses,
but cabal can't express that in a custom-setup stanza.
To avoid this problem, Utility.Env would need to be moved into
unix-compat..
Windows needs the setenv package in custom-setup, but I don't want to
pull it in on unix, which would probably break some builds and need more
work. Instead, split out setEnv to a separate module.
Quite likely, unix-compat will get a portable environment layer, and
then both modules can be removed from here.
This commit was sponsored by Øyvind Andersen Holm.
That version has my patches for the problems that Utility.PosixFiles
was working around, so am able to get rid of that module now.
This will later allow bringing back the custom-setup stanza in the cabal
file. It will need to depend on unix-compat 0.5 on all OS's, which I'm
not ready to do yet.
This commit was sponsored by Nick Daly on Patreon.
Windows: Fix reversion that caused the path used to link to annexed
content include the drive letter and full path, rather than being
relative. (`git annex fix` will fix up after this problem).
I've not identified the commit that brought the reversion (probably it
happened this spring when I was removing MisingH and last touched
Utility.Path). Likely commit 18b9a4b8024115db67ae309fdaf54e1553037529?
The problem is that relPathDirToFile got called two paths that had the
slashes different ways around. Since takeDrive includes the first slash,
this made two paths on the same drive seem different and it bailed.
(ifdefs around this to avoid doing extra work on non-windows)
This commit was sponsored by Jack Hill on Patreon.
Get ugly reversion out of CHANGELOG.
Also, relocated the windows stack.yaml to top, and updated windows build
instructions.
This commit was sponsored by Henrik Riomar on Patreon.
Code for terminating processes on Windows is not linking anymore;
made a warning be displayed instead. This breaks restarting the
assistant and git annex assistant --stop.
I hope to see the code added to the Win32 library, where it should fit
better and should avoid whatever problem is making the linker not like it
when included in git-annex. I opened an issue requesting its addition,
here: https://github.com/haskell/win32/issues/91
This commit was sponsored by Thomas Hochstein on Patreon.
This reverts commit 839ec7e26c.
Neither way is working.. The other way failed:
.stack-work\dist\5f9bc736\build\git-annex\git-annex-tmp\Assistant.o:fake:(.text+0x6bb3): undefined reference to `terminatepid'
Seems that winprocess.c is not getting linked in.
Building with stack, it failed:
`_TerminateProcess' referenced in section `.text' of .stack-work\dist\5f9bc736\build\git-annex\git-annex-tmp\Utility\WinProcess.o: defined in discarded section `.text' of C:/Users/jenkins/AppData/Local/Programs/stack/i386-windows/ghc-8.0.2/mingw/bin/../lib/gcc/i686-w64-mingw32/5.2.0/../../../../i686-w64-mingw32/lib/../lib/libkernel32.a(dacgs01154.o)
This is a reversion of 86e638567a,
to try the other way to implement it, which will hopefully avoid the problem.
webdav: Checking if a non-existent file is present on Box.com triggered a
bug in its webdav support that generates an infinite series of redirects.
It seems to redirect foo to foo/ to foo/index.php to
foo/index.php/index.php ... Why a webdav endpoint would behave this way
who knows.
Deal with such problems by assuming such behavior means the file is not
present.
Can't simply disable following redirects, because the webdav endpoint could
legitimately be redirected to a new endpoint. So, when this happens
10 redirects have to be followed, before it gives up and assumes this means
the file does not exist.
This commit was supported by the NSF-funded DataLad project.
This avoids needing to deal with the complexity of partially transferred
files in the export. We'd not be able to resume uploading to such a file
anyway, so just avoid them.
The implementation in Remote.Directory is not completely ideal, because
it could leave the temp file hanging around in the export directory.
This only happens if it's killed with -9, or there's a power failure;
normally viaTmp cleans up after itself, even when interrupted. I could
not see a better way to do it though, since the export directory might
be the root of a filesystem.
Also some design thoughts on resuming, which depend on storeExport being
atomic.
This commit was sponsored by Fernando Jimenez on Partreon.
Security fix: Disallow hostname starting with a dash, which would get
passed to ssh and be treated an option. This could be used by an attacker
who provides a crafted ssh url (for eg a git remote) to execute arbitrary
code via ssh -oProxyCommand.
No CVE has yet been assigned for this hole.
The same class of security hole recently affected git itself,
CVE-2017-1000117.
Method: Identified all places where ssh is run, by git grep '"ssh"'
Converted them all to use a SshHost, if they did not already, for
specifying the hostname.
SshHost was made a data type with a smart constructor, which rejects
hostnames starting with '-'.
Note that git-annex already contains extensive use of Utility.SafeCommand,
which fixes a similar class of problem where a filename starting with a
dash gets passed to a program which treats it as an option.
This commit was sponsored by Jochen Bartl on Patreon.
By forking a worker process and only deleting the test directory once it exits.
This way, if a test leaves files open, they'll get closed when the worker
exits, so avoiding failure to delete open files on Windows, and failure to
delete directories due to NFS lock files.
If a test leaves a git worker process running, the closed pipes should
cause the worker to exit too, also avoiding the problem there. The 10
second sleep ought to give plenty of time for such worker processes to
exit, although this is of course a race.
Finally, even if test directory fails to be deleted still,
it won't appear as if the last test in the test suite failed; the error
will be displayed at the very end.
This commit was supported by the NSF-funded DataLad project.
QuickCheck 2.10 found a counterexample eg "\929184" broke the property.
As far as I can tell, Git.Filename is matching how git handles encoding
of strange high unicode characters in filenames for display. Git does
not display high unicode characters, and instead displays the C-style
escaped form of each byte. This is ambiguous, but since git is not
unicode aware, it doesn't need to roundtrip parse it.
So, making Git.FileName's roundtrip test only chars < 256 seems fine.
Utility.Format.format uses encode_c, in order to mimic git, so that's
ok.
Utility.Format.gen uses decode_c, but only so that stuff like "\n"
in the format string is handled. If the format string contains C-style
octal escapes, they will be converted to ascii characters, and not
combined into unicode characters, but that should not be a problem.
If the user wants unicode characters, they can include them in the
format string, without escaping them.
Finally, decode_c is used by Utility.Gpg.secretKeys, because gpg
--with-colons hex-escapes some characters in particular ':' and '\\'.
gpg passes unicode through, so this use of decode_c is not a problem.
This commit was sponsored by Henrik Riomar on Patreon.
QuickCheck added an Arbitrary instance for CTime aka EpochTime. However,
while git-annex's instance disallowed times before the epoch, QuickCheck's
does not. So, rather than using its instance, convert from an Integer.
This commit was sponsored by Thomas Hochstein on Patreon.
Don't trust OSX FSEvents's eventFlagItemModified to be called when the last
writer of a file closes it; apparently that sometimes does not happen,
which prevented files from being quickly added.
This commit was sponsored by John Peloquin on Patreon.
orElse is great, but was not the right thing to use here because
waitTakeLock could retry for other reasons than the lock being held,
which made tryTakeLock fail when it shouldn't.
Instead, move the code to tryTakeLock and implement waitTakeLock using
tryTakeLock and retry.
(Also, in runTransfer, when checkSaneLock fails, dropLock to avoid leaking a
lock handle.)
This commit was supported by the NSF-funded DataLad project.
Removed dependency on MissingH, instead depending on the split
library.
After laying groundwork for this since 2015, it
was mostly straightforward. Added Utility.Tuple and
Utility.Split. Eyeballed System.Path.WildMatch while implementing
the same thing.
Since MissingH's progress meter display was being used, I re-implemented
my own. Bonus: Now progress is displayed for transfers of files of
unknown size.
This commit was sponsored by Shane-o on Patreon.
Cryptonite is faster and allocates less, and I want to get rid of
MissingH use.
Note that the new dependency on memory is free; it's a dependency of
cryptonite.
This commit was supported by the NSF-funded DataLad project.
Moving toward dropping MissingH dep.
I think I've addressed the problem identified earlier in
09a66f702d. On Windows,
absPathFrom "/tmp/repo/xxx" "y/bar" would be "/tmp/repo/xxx\\y/bar",
which then confuses relPathDirToFile. Fixed by converting to unix (git)
style paths.
Also, relPathDirToFile was splitting only on \\ on windows and not /
which broke the example in 09a66f702d of
relPathDirToFile (absPathFrom "/tmp/repo/xxx" "y/bar") "/tmp/repo/.git/annex/objects/xxx"
Now, on windows, that will yield "..\\..\\..\\.git/annex/objects/xxx"
which once converted to unix style paths is what we want.
The bug was that withFile closes the handle afterwards, but the content
of the file was not read due to laziness. Using readFile avoids it.
This commit was sponsored by Nick Daly on Patreon.
The COPYRIGHT had Utility/DirWatcher* listed as GPL, but they were
actually BSD licensed.
No idea why I put the GPL on Utility/GPG.hs file originally.
I wrote all of it, except for guilhem's small changes to it in
00fc21bfec, which seem too small to be
independently copyrightable. I'm relicencing it BSD.
findShellCommand needs a full path to a file in order to check it for a
shebang on Windows. It was being run with only the base name of the external
special remote program, which would only work when it was in the current
directory.
This is why users in
https://github.com/DanielDent/git-annex-remote-rclone/pull/10 and elsewhere
were complaining that the previous improvements to git-annex didn't make
git-remote-rclone work on Windows.
Also, reworked checkearlytermination, which while it worked, seemed
to rely on a race condition. And, improved its error messages.
This commit was sponsored by Shane-o on Patreon.
* Run curl with -S, so HTTP errors are displayed, even when
it's otherwise silent.
* When downloading in --json or --quiet mode, use curl in preference
to wget, since curl is able to display only errors to stderr, unlike
wget.
This does mean that downloadQuiet is only silent on stdout, not necessarily
on stderr, which affects a couple other calls of it. For example,
downloading the .git/config of a http remote may show an error message now,
perhaps with slightly suboptimal formatting due to other output.
This adds one extra line of output when a download is successful,
after the progress bar. I don't much like that, but wget does not provide a
way to show HTTP errors without it.
This allows using functions that generate CreateProcess and passing the
result to processTranscript', which is more flexible, and also simpler
than the old interface.
This commit was sponsored by Riku Voipio.
Refactored some common code into initDb.
This only deals with the problem when creating new databases. If a repo
got bad permissions into it, it's up to the user to deal with it.
This commit was sponsored by Ole-Morten Duesund on Patreon.
Probing for hard link support in the pid locking code is redundant since
git-annex init already probes that. But, it didn't seem worth threading
that data through; the pid locking code runs at most once per git-annex
process, and only on unusual filesystems. Optimising a single hard link
and unlink isn't worth it.
This commit was sponsored by Francois Marier on Patreon.
Wormhole pairing will start to provide an appid to wormhole on 2021-12-31.
An appid can't be provided now because Debian stable is going to ship a
older version of git-annex that does not provide an appid. Assumption is
that by 2021-12-31, this version of git-annex will be shipped in a Debian
stable release. If that turns out to not be the case, this change will need
to be cherry-picked into the git-annex in Debian stable, or its wormhole
pairing will break.
This commit was sponsored by Thomas Hochstein on Patreon.
Turns out that Data.List.Utils.split is slow and makes a lot of
allocations. Here's a much simpler single character splitter that behaves
the same (even in wacky corner cases) while running in half the time and
75% the allocations.
As well as being an optimisation, this helps move toward eliminating use of
missingh.
(Data.List.Split.splitOn is nearly as slow as Data.List.Utils.split and
allocates even more.)
I have not benchmarked the effect on git-annex, but would not be surprised
to see some parsing of eg, large streams from git commands run twice as
fast, and possibly in less memory.
This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
hSetEncoding of a closed handle segfaults.
https://ghc.haskell.org/trac/ghc/ticket/71618484c0c197 introduced the crash.
In particular, stdin may get closed (by eg, getContents) and then trying
to set its encoding will crash. We didn't need to adjust stdin's
encoding anyway, but only stderr, to work around
https://github.com/yesodweb/persistent/issues/474
Thanks to Mesar Hameed for assistance related to reproducing this bug.
Since the user does not know whether it will run su or sudo, indicate
whether the password prompt will be for root or the user's password,
when possible.
I assume that programs like gksu that can prompt for either depending on
system setup will make clear in their prompt what they're asking for.
weasel explained that apparmor limits on what files tor can read do not
apply to sockets (because they're not files). And apparently the
problems I was seeing with hidden services not being accessible had to
do with onion address propigation and not the location of the socket
file.
remotedaemon looks up the HiddenServicePort in torrc, so if it was
previously configured with the socket in /etc, that will still work.
This commit was sponsored by Denis Dzyubenko on Patreon.
This interacts with it using stdio, which is surprisingly hard.
sendFile does not currently work, due to
https://github.com/warner/magic-wormhole/issues/108
Parsing the output to find the magic code is done as robustly as
possible, and should continue to work unless wormhole radically changes
the format of its codes. Presumably it will never output something that
looks like a wormhole code before the actual wormhole code; that would
also break this. It would be better if there was a way to make
wormhole not mix the code with other output, as requested in
https://github.com/warner/magic-wormhole/issues/104
Only exchange of files/directories is supported. To exchange messages,
https://github.com/warner/magic-wormhole/issues/99 would need to be resolved.
I don't need message exchange however.
The attacker could just send a very lot of data, with no \n and it would
all be buffered in memory until the kernel killed git-annex or perhaps OOM
killed some other more valuable process.
This is a low impact security hole, only affecting communication between
local git-annex and git-annex-shell on the remote system. (With either
able to be the attacker). Only those with the right ssh key can do it. And,
there are probably lots of ways to construct git repositories that make git
use a lot of memory in various ways, which would have similar impact as
this attack.
The fix in P2P/IO.hs would have been higher impact, if it had made it to a
released version, since it would have allowed DOSing the tor hidden
service without needing to authenticate.
(The LockContent and NotifyChanges instances may not be really
exploitable; since the line is read and ignored, it probably gets read
lazily and does not end up staying buffered in memory.)
Display progress meter on send and receive from remote.
Added a new hGetMetered that can read an exact number of bytes (or
less), updating a meter as it goes.
This commit was sponsored by Andreas on Patreon.
On Debian, apparmor prevents tor from reading from most locations. And,
it silently fails if it is prevented from reading the hidden service
socket. I filed #846275 about this; violating the FHS is the least bad of a
bad set of choices until that bug is fixed.
Still a couple bugs:
* Closing the connection to the server leaves git upload-pack /
receive-pack running, which could be used to DOS.
* Sometimes the data is transferred, but it fails at the end, sometimes
with:
git-remote-tor-annex: <socket: 10>: commitBuffer: resource vanished (Broken pipe)
Must be a race condition around shutdown.
Almost working, but there's a bug in the relaying.
Also, made tor hidden service setup pick a random port, to make it harder
to port scan.
This commit was sponsored by Boyd Stephen Smith Jr. on Patreon.
A bit tricky since Proto doesn't support threads. Rather than adding
threading support to it, ended up using a callback that waits for both
data on a Handle, and incoming messages at the same time.
This commit was sponsored by Denis Dzyubenko on Patreon.
For use with tor hidden services, and perhaps other transports later.
Based on Utility.SimpleProtocol, it's a line-based protocol,
interspersed with transfers of bytestrings of a specified size.
Implementation of the local and remote sides of the protocol is done
using a free monad. This lets monadic code be included here, without
tying it to any particular way to get bytes peer-to-peer.
This adds a dependency on the haskell package "free", although that
was probably pulled in transitively from other dependencies already.
This commit was sponsored by Jeff Goeke-Smith on Patreon.
ghc 8 added backtraces on uncaught errors. This is great, but git-annex was
using error in many places for a error message targeted at the user, in
some known problem case. A backtrace only confuses such a message, so omit it.
Notably, commands like git annex drop that failed due to eg, numcopies,
used to use error, so had a backtrace.
This commit was sponsored by Ethan Aubin.
This avoids needing to bind to the right port before something else
does.
The socket is in /var/run/user/$uid/ which ought to be writable by only
that uid. At least it is on linux systems using systemd.
For Windows, may need to revisit this and use ports or something.
The first version of tor to support sockets for hidden services
was 0.2.6.3. That is not in Debian stable, but is available in
backports.
This commit was sponsored by andrea rota.
Tor unfortunately does not come out of the box configured to let hidden
services register themselves on the fly via the ControlPort.
And, changing the config to enable the ControlPort and a particular type
of auth for it may break something already using the ControlPort, or
lessen the security of the system.
So, this leaves only one option to us: Add a hidden service to the
torrc. git-annex enable-tor does so, and picks an unused high port for
tor to listen on for connections to the hidden service.
It's up to the caller to somehow pick a local port to listen on
that won't be used by something else. That may be difficult to do..
This commit was sponsored by Jochen Bartl on Patreon.
Yesod didn't used to do auth checks for that, but this may have changed.
I don't have a way to reproduce the reported problem yet, but this change
certianly won't hurt anything.
This commit was sponsored by Thom May on Patreon.
Restarting a crashing git process could result in filename encoding issues
when not in a unicode locale, as the restarted processes's handles were not
read in raw mode.
Since rawMode is always used when starting a coprocess, didn't bother
to parameterise it and just always enable it for simplicity.
This commit was sponsored by Jake Vosloo on Patreon.
gpg-agent started deleting its socket file on shutdown, and this tickled an
ugly behavior in removeDirectoryRecursive,
https://github.com/haskell/directory/issues/60
Running removeDirectoryRecursive again on exception avoids the problem.
gpg 2.1.15 (or so) seems to have added some new fields to the --with-colons
--list-secret-keys output. These include "fpr" and "grp", and come before
the "uid" line. So, the parser was giving up before it saw the name. Fix by
continuing to look for the uid line until the next "sec" line.
This commit was sponsored by Ole-Morten,Duesund on Patreon.
This gets rid of quite a lot of ugly hacks around json generation.
I doubt that any real-world json parsers can parse incomplete objects, so
while it's not as nice to need to wait for the complete object, especially
for commands like `git annex info` that take a while, it doesn't seem worth
the added complexity.
This also causes the order of fields within the json objects to be
reordered. Since any real json parser shouldn't care, the only possible
problem would be with ad-hoc parsers of the old json output.
Keeping Text.JSON use for now, because it seems a better fit for most of
the commands, which don't use very structured JSON objects, but just output
whatever fields suites them. But this lets Aeson be used when a more
structured data type is available to serialize to JSON.
Mostly the username is only used for the git committer or other display
purposes, and we can just fall back to a dummy value in these cases.
The only remaining place where an error is thrown is when starting local
pairing, which needs the username to be known.
Sadly my bug report about this is not going to get fixed it seems, so
I have to drag around a whole added module file just to deal with it.
https://github.com/haskell/directory/issues/52
It started exporting a isSymbolicLink which supports windows. But,
git-annex does no use symlinks on windows yet and this conflicts with the
function by the same name from unix-compat, so hide it.
This fixes behavior in this situation:
l1 <- lockShared Nothing "lck"
l2 <- lockShared Nothing "lck"
dropLock l1
dropLock l2
Before, the lock was dropped upon the second dropLock call, but the fd
remained open, and would never be closed while the program was running.
Fixed by a rather round-about method, but it should work well enough.
It would have been simpler to open open the shared lock once, and not open
it again in the second call to lockShared. But, that's difficult to do
atomically.
This also affects Windows and PID locks, not just posix locks.
In the case of pid locks, multiple calls to waitLock within the same
process are allowed because the side lock is locked using a posix lock,
and so multiple exclusive locks can be taken in the same process. So,
this change fixes a similar problem with pid locks.
l1 <- waitLock (Seconds 1) "lck"
l2 <- waitLock (Seconds 1) "lck"
dropLock l1
dropLock l2
Here the l2 side lock fd remained open but not locked,
although the pid lock file was removed. After this change, the second
dropLock will close both fds to the side lock, and delete the pidlock.
According to https://github.com/redneb/disk-free-space/issues/3 ,
disk-free-space should be at least as portable as my homegrown code was.
One change I noticed is, getDiskSize was not implemented for windows
in the old code, and should work now.
This lets readonly repos be used. If a repo is readonly, we can ignore the
keys database, because nothing that we can do will change the state of the
repo anyway.
* Removed the webapp-secure build flag, rolling it into the webapp build
flag.
* Removed the quvi and tahoe build flags, which only adds aeson to
the core dependencies.
* Removed the feed build flag, which only adds feed to the core
dependencies.
Build flags have cost in both code complexity and also make Setup configure
have to work harder to find a usable set of build flags when some
dependencies are missing.
This reverts commit d14770ca9c.
That changed the type of error from an IOError to something else, so broke
stuff that was catching IOErrors.
So back to a UserError, but be explicit this time that's what it's
throwing.
Using fail here causes a "user error" exception to be thrown, which implies
the user is at fault in its wording, which is incorrect.
Also audited for other uses of fail in git-annex; the others are in monadic
contexts where fail may not throw an exception, and involve user input, so
kept them as-is.
The repo path is typically relative, not absolute, so
providing it to absPathFrom doesn't yield an absolute path.
This is not a bug, just unclear documentation.
Indeed, there seem to be no reason to simplifyPath here, which absPathFrom
does, so instead just combine the repo path and the TopFilePath.
Also, removed an export of the TopFilePath constructor; asTopFilePath
is provided to construct one as-is.
Several tricky parts:
* When the conflict is just between the same key being locked and unlocked,
the unlocked version wins, and the file is not renamed in this case.
* Need to update associated file map when conflict resolution renames
an unlocked file.
* git merge runs the smudge filter on the conflicting file, and actually
overwrites the file with the same content it had before, and so
invalidates its inode cache. This makes it difficult to know when it's
safe to remove such files as conflict cruft, without going so far as to
compare their entire contents.
Dealt with this by preventing the smudge filter from populating the file
when a merge is run. However, that also prevents the smudge filter being
run for non-conflicting files, so eg moving a file won't put its new
content into place.
* Ideally, if a merge or a merge conflict resolution renames an unlocked
file, the file in the work tree can just be moved, rather than copying
the content to a new worktree file.
This is attempted to be done in merge conflict resolution, but
due to git merge's behavior of running smudge filters, what actually
seems to happen is the old worktree file with the content is deleted and
rewritten as a pointer file, so doesn't get reused.
So, this is probably not as efficient as it optimally could be.
If that becomes a problem, could look into running the merge in a separate
worktree and updating the real worktree more efficiently, similarly to the
direct mode merge. However, the direct mode merge had a lot of bugs, and
I'd rather not use that more error-prone method unless really needed.
Writes are optimised by queueing up multiple writes when possible.
The queue is flushed after the Annex monad action finishes. That makes it
happen on program termination, and also whenever a nested Annex monad action
finishes.
Reads are optimised by checking once (per AnnexState) if the database
exists. If the database doesn't exist yet, all reads return mempty.
Reads also cause queued writes to be flushed, so reads will always be
consistent with writes (as long as they're made inside the same Annex monad).
A future optimisation path would be to determine when that's not necessary,
which is probably most of the time, and avoid flushing unncessarily.
Design notes for this commit:
- separate reads from writes
- reuse a handle which is left open until program
exit or until the MVar goes out of scope (and autoclosed then)
- writes are queued
- queue is flushed periodically
- immediate queue flush before any read
- auto-flush queue when database handle is garbage collected
- flush queue on exit from Annex monad
(Note that this may happen repeatedly for a single database connection;
or a connection may be reused for multiple Annex monad actions,
possibly even concurrent ones.)
- if database does not exist (or is empty) the handle
is not opened by reads; reads instead return empty results
- writes open the handle if it was not open previously
Caused by AMP.. Since I've finally upgraded my dev laptop to 7.10,
I may start missing imports that are not needed with it but are with older
versions..
http://bugs.debian.org/807341
* Fix insecure temporary permissions when git-annex repair is used in
in a corrupted git repository.
Other calls to withTmpDir didn't leak any potentially private data,
but repair clones the git repository to a temp directory which is made
using the user's umask. Thus, it might expose a git repo that is
otherwise locked down.
* Fix potential denial of service attack when creating temp dirs.
Since withTmpDir used easily predictable temporary directory names,
an attacker could create foo.0, foo.1, etc and as long as it managed to
keep ahead of it, could prevent it from ever returning.
I'd rate this as a low utility DOS attack. Most attackers in a position
to do this could just fill up the disk /tmp is on to prevent anything
from writing temp files. And few parts of git-annex use withTmpDir
anyway, so DOS potential is quite low.
Examined all callers of withTmpDir and satisfied myself that
switching to mkdtmp and so getting a mode 700 temp dir wouldn't break any
of them.
Note that withTmpDirIn continues to not force temp dir to 700.
But it's only used for temp directories inside .git/annex/wherever/
so that is not a problem.
Also re-audited all other uses of temp files and dirs in git-annex.
The Keys database can hold multiple inode caches for a given key. One for
the annex object, and one for each pointer file, which may not be hard
linked to it.
Inode caches for a key are recorded when its content is added to the annex,
but only if it has known pointer files. This is to avoid the overhead of
maintaining the database when not needed.
When the smudge filter outputs a file's content, the inode cache is not
updated, because git's smudge interface doesn't let us write the file. So,
dropping will fall back to doing an expensive verification then. Ideally,
git's interface would be improved, and then the inode cache could be
updated then too.
Had everything available, just didn't combine the progress meter with the
other places progress is sent to update it. (And to a remote repo already
did show progress.)
Most special remotes should already display progress meters with -J,
same as without it. One exception to this is the web, since it relies on
wget/curl progress display without -J. Still todo..
There's a potential race, but it's detected and just results in the other
process failing to take the side lock, so possibly retrying one second
later on. The race window is quite narrow so the extra delay is minor.
Left the side lock files mode 666 because an interruption can leave a side
lock file created by another user for a shared repository. When this
happens, the non-owning user can't delete it (+t) but can still lock it,
and so the code falls back to acting as it did before this commit.
This is less portable, since currently sidelocks rely on /dev/shm.
But, I've seen crazy lustre inconsistencies that make me not trust the
link() method at all, so what can you do.
I have a strace taken on a lustre filesystem on which link() returned 0,
but didn't actually succeed, since the file already existed.
One of the linux man pages recommended using link followed by checking like
this. I was reading it yesterday, but cannot find it now.
